feat: add Broadcom FullMAC Wi-Fi driver

.gitignore

@@ -6,12 +6,21 @@
 archives/isos/
 montaukos.org/downloads/*.iso
 *.hdd
+*.img
+rpi4-firmware/
 qemu.log
 toolchain/local/
 toolchain/build/
 toolchain/src/
 toolchain/sysroot/
+toolchain/sysroot-*/
 programs/bin/
+# Non-x86 userspace builds (programs/mk/arch.mk)
+programs/bin-*/
+programs/obj-*/
+programs/src/*/obj-*/
+programs/lib/*/obj-*/
+programs/lib/*/build-*/
 programs/obj/
 programs/src/*/obj/
 programs/libs/*/obj/

GNUmakefile

@@ -10,6 +10,32 @@ QEMUFLAGS := -smp 4 -m 2G -d int -D qemu.log -no-reboot
 
 override IMAGE_NAME := montauk-$(ARCH)
 
+# KERNEL_ONLY=1 builds a ramdisk holding just the boot files (no
+# userspace), for bringing up a kernel on a new architecture.
+KERNEL_ONLY ?= 0
+
+# programs/ builds x86_64 into bin/, other architectures into bin-<arch>/.
+ifeq ($(ARCH),x86_64)
+    PROGRAMS_BIN := programs/bin
+else
+    PROGRAMS_BIN := programs/bin-$(ARCH)
+endif
+
+override EFI_LOADER_x86_64 := BOOTX64.EFI
+override EFI_LOADER_aarch64 := BOOTAA64.EFI
+override EFI_LOADER_riscv64 := BOOTRISCV64.EFI
+override EFI_LOADER_loongarch64 := BOOTLOONGARCH64.EFI
+
+# The host strip handles x86_64 images; other targets use their toolchain's.
+ifeq ($(ARCH),x86_64)
+    STRIP := strip
+else
+    STRIP := toolchain/local/bin/$(ARCH)-montauk-strip
+endif
+
+# AArch64 UEFI firmware for QEMU (Debian/Ubuntu: qemu-efi-aarch64).
+AARCH64_EFI_FIRMWARE ?= /usr/share/qemu-efi-aarch64/QEMU_EFI.fd
+
 # Toolchain for building the 'limine' executable for the host.
 HOST_CC := cc
 HOST_CFLAGS := -g -O2 -pipe
@@ -51,29 +77,72 @@ run-hdd-x86_64: $(IMAGE_NAME).hdd
 		$(QEMUFLAGS)
 
 .PHONY: run-aarch64
+# The machine mirrors the Raspberry Pi 4/400: Cortex-A72 cores, a GICv2, and
+# keyboard and mouse behind a hub (every USB 2 device on the Pi is: the
+# VL805's one USB 2 port carries an on-board hub). QEMU's hub is full-speed,
+# so the transaction-translator path is only exercised on the real board.
+# The kernel console is mirrored to the PL011 (-serial stdio).
 run-aarch64: $(IMAGE_NAME).iso
 	qemu-system-$(ARCH) \
-		-M virt \
+		-M virt,gic-version=2 \
 		-cpu cortex-a72 \
 		-device ramfb \
 		-device qemu-xhci,id=xhci \
-		-device usb-kbd,bus=xhci.0 \
+		-device usb-hub,bus=xhci.0,port=1 \
-		-device usb-mouse,bus=xhci.0 \
+		-device usb-kbd,bus=xhci.0,port=1.1 \
-		-bios /usr/share/ovmf/OVMF.fd \
+		-device usb-mouse,bus=xhci.0,port=1.2 \
+		-bios $(AARCH64_EFI_FIRMWARE) \
 		-cdrom $(IMAGE_NAME).iso \
+		-serial stdio \
 		$(QEMUFLAGS)
 
 .PHONY: run-hdd-aarch64
 run-hdd-aarch64: $(IMAGE_NAME).hdd
 	qemu-system-$(ARCH) \
-		-M virt \
+		-M virt,gic-version=2 \
 		-cpu cortex-a72 \
 		-device ramfb \
 		-device qemu-xhci,id=xhci \
-		-device usb-kbd,bus=xhci.0 \
+		-device usb-hub,bus=xhci.0,port=1 \
-		-device usb-mouse,bus=xhci.0 \
+		-device usb-kbd,bus=xhci.0,port=1.1 \
-		-bios /usr/share/ovmf/OVMF.fd \
+		-device usb-mouse,bus=xhci.0,port=1.2 \
-		-hda $(IMAGE_NAME).hdd \
+		-bios $(AARCH64_EFI_FIRMWARE) \
+		-drive if=none,id=hd0,format=raw,file=$(IMAGE_NAME).hdd \
+		-device virtio-blk-pci,drive=hd0 \
+		-serial stdio \
+		$(QEMUFLAGS)
+
+# The SD card image on QEMU, with the root read through the SDHCI driver:
+# the card goes on sdhci-pci as a padded copy (QEMU wants a power-of-two card;
+# RPI4_QEMU_CARD_MIB, default 4 GiB, leaves the first-boot grow room to work
+# with; every run starts from a fresh copy, so every run is a first boot),
+# and since QEMU's UEFI has no SD driver, the card's ESP is also
+# attached alone (the card's first partition, as its own disk) over virtio
+# for the firmware to boot Limine from.
+RPI4_QEMU_CARD_MIB ?= 4096
+.PHONY: run-rpi4-sd
+run-rpi4-sd: montauk-rpi4.img
+	cp --sparse=always montauk-rpi4.img rpi4-sd-qemu.img
+	size=$$(stat -c %s rpi4-sd-qemu.img); p=$$(($(RPI4_QEMU_CARD_MIB) * 1048576)); \
+		while [ $$p -lt $$size ]; do p=$$((p * 2)); done; truncate -s $$p rpi4-sd-qemu.img
+	end=$$(PATH=$$PATH:/usr/sbin:/sbin sfdisk -d montauk-rpi4.img | awk -F'[=,]' '/montauk-rpi4.img1/ {print $$2 + $$4}'); \
+		dd if=montauk-rpi4.img of=rpi4-esp-qemu.img bs=512 count=$$end status=none
+	PATH=$$PATH:/usr/sbin:/sbin sfdisk --quiet --delete rpi4-esp-qemu.img 2 2>/dev/null || true
+	qemu-system-aarch64 \
+		-M virt,gic-version=2 \
+		-cpu cortex-a72 \
+		-device ramfb \
+		-device qemu-xhci,id=xhci \
+		-device usb-hub,bus=xhci.0,port=1 \
+		-device usb-kbd,bus=xhci.0,port=1.1 \
+		-device usb-mouse,bus=xhci.0,port=1.2 \
+		-bios $(AARCH64_EFI_FIRMWARE) \
+		-drive if=none,id=esp,format=raw,file=rpi4-esp-qemu.img \
+		-device virtio-blk-pci,drive=esp \
+		-device sdhci-pci \
+		-drive if=none,id=sd,format=raw,file=rpi4-sd-qemu.img \
+		-device sd-card,drive=sd \
+		-serial stdio \
 		$(QEMUFLAGS)
 
 .PHONY: run-riscv64
@@ -196,13 +265,23 @@ ports: programs
 	fi
 
 .PHONY: ramdisk
+ifeq ($(KERNEL_ONLY),1)
+ramdisk: limine/limine kernel
+	rm -rf .kernel-only-root
+	mkdir -p .kernel-only-root/boot/limine
+	cp -v limine/$(EFI_LOADER_$(ARCH)) limine.conf .kernel-only-root/boot/limine/
+	cp -v kernel/bin-$(ARCH)/kernel .kernel-only-root/boot/
+	./scripts/mkramdisk.sh .kernel-only-root ramdisk.tar
+	rm -rf .kernel-only-root
+else
 ramdisk: limine/limine kernel programs
-	mkdir -p programs/bin/boot/limine
+	mkdir -p $(PROGRAMS_BIN)/boot/limine
-	cp -v limine/BOOTX64.EFI programs/bin/boot/limine/
+	cp -v limine/$(EFI_LOADER_$(ARCH)) $(PROGRAMS_BIN)/boot/limine/
-	cp -v limine.conf programs/bin/boot/limine/
+	cp -v limine.conf $(PROGRAMS_BIN)/boot/limine/
-	cp -v kernel/bin-$(ARCH)/kernel programs/bin/boot/
+	cp -v kernel/bin-$(ARCH)/kernel $(PROGRAMS_BIN)/boot/
-	rm -f programs/bin/boot/ramdisk.tar
+	rm -f $(PROGRAMS_BIN)/boot/ramdisk.tar
-	./scripts/mkramdisk.sh programs/bin ramdisk.tar
+	./scripts/mkramdisk.sh $(PROGRAMS_BIN) ramdisk.tar
+endif
 
 $(IMAGE_NAME).iso: limine/limine kernel ramdisk
 	rm -rf iso_root
@@ -252,13 +331,17 @@ ifeq ($(ARCH),loongarch64)
 endif
 	rm -rf iso_root
 
+# Example and regression-test programs that are built into 0:/os for
+# development but have no place on a release image.
+RELEASE_EXCLUDE_PROGRAMS := rusthello.elf futextest.elf memtest.elf
+
 .PHONY: release
 release: limine/limine kernel ramdisk
 	rm -rf iso_root .release-tmp
 	mkdir -p iso_root/boot
 	cp kernel/bin-$(ARCH)/kernel iso_root/boot/
-	strip iso_root/boot/kernel
+	$(STRIP) iso_root/boot/kernel
-	cp -r programs/bin .release-tmp
+	cp -r $(PROGRAMS_BIN) .release-tmp
 # A release ISO ships only in-tree apps, so drop any out-of-tree port bundles
 # that happen to be staged in programs/bin/apps (each carries a .port marker).
 # Pass RELEASE_WITH_PORTS=1 to deliberately cut a ports-included image.
@@ -281,11 +364,12 @@ ifeq ($(RELEASE_WITHOUT_SDK),1)
 	rm -rf .release-tmp/sdk/bin .release-tmp/sdk/lib .release-tmp/sdk/include \
 	       .release-tmp/sdk/libexec .release-tmp/sdk/x86_64-montauk
 endif
-	find .release-tmp -name '*.elf' -exec strip {} +
+	rm -f $(addprefix .release-tmp/os/,$(RELEASE_EXCLUDE_PROGRAMS))
+	find .release-tmp -name '*.elf' -exec $(STRIP) {} +
 # The ramdisk carries its own copy of the kernel at 0:/boot/kernel -- that is
 # the image the Installer lays down on the target disk -- and it is not named
 # *.elf, so the find above misses it. Unstripped it is ~6 MB against ~0.7 MB.
-	strip .release-tmp/boot/kernel
+	$(STRIP) .release-tmp/boot/kernel
 	./scripts/mkramdisk.sh .release-tmp iso_root/boot/ramdisk.tar
 	rm -rf .release-tmp
 	mkdir -p iso_root/boot/limine
@@ -331,6 +415,26 @@ ifeq ($(ARCH),loongarch64)
 endif
 	rm -rf iso_root
 
+# Raspberry Pi 4/400 SD card image (UEFI firmware + Limine + MontaukOS).
+# ROOTFS=ext2 (default) puts the system on an ext2 partition read through the
+# kernel's SD driver; ROOTFS=ramdisk has Limine load ramdisk.tar instead.
+# Phony so switching ROOTFS always rebuilds the image.
+ROOTFS ?= ext2
+.PHONY: rpi4-sd montauk-rpi4.img
+rpi4-sd: montauk-rpi4.img
+
+montauk-rpi4.img: limine/limine kernel ramdisk
+ifneq ($(ARCH),aarch64)
+	$(error rpi4-sd needs ARCH=aarch64)
+endif
+	./scripts/mk-rpi4-sd.sh kernel/bin-$(ARCH)/kernel ramdisk.tar $@ $(ROOTFS) $(PROGRAMS_BIN)-rpi
+
+# Graphical imager: write montauk-rpi4.img to an SD card, or update the
+# installation on one in place (scripts/montauk-imager.py, docs/raspberry-pi.md).
+.PHONY: rpi4-imager
+rpi4-imager:
+	./scripts/montauk-imager.py
+
 $(IMAGE_NAME).hdd: limine/limine kernel ramdisk
 	rm -f $(IMAGE_NAME).hdd
 	dd if=/dev/zero bs=1M count=0 seek=64 of=$(IMAGE_NAME).hdd
@@ -362,7 +466,7 @@ endif
 clean:
 	$(MAKE) -C kernel clean
 # 	$(MAKE) -C programs clean
-	rm -rf iso_root .release-tmp $(IMAGE_NAME).iso $(IMAGE_NAME).hdd ramdisk.tar
+	rm -rf iso_root .release-tmp $(IMAGE_NAME).iso $(IMAGE_NAME).hdd montauk-rpi4.img rpi4-sd-qemu.img rpi4-esp-qemu.img ramdisk.tar
 
 .PHONY: distclean
 distclean:

archives/releasenotes.txt

@@ -1,3 +1,78 @@
+MontaukOS 0.2.0 (October 14, 2026)
+    (278b20825815ec04a22fd1c2fcfedbd38e116e6d)
+
+New
+====
+    * Capability (application permissions) model added
+    * Consent prompts for missing app permissions
+    * New Capabilities app for reviewing and managing app permissions
+    * Unified system status codes; the shell and Files now print a readable message on failure
+    * Text is now UTF-8 throughout, with Unicode key events and dead keys
+    * Keyboard layout support, with English, Norwegian, Czech, Swedish, and Danish layouts
+    * NumLock support
+    * Battery percentage reporting via ACPI, with a battery indicator in the desktop panel
+    * Display brightness control, including on hybrid graphics laptops, with a slider in the desktop panel
+    * Discrete GPUs are powered down automatically on hybrid graphics laptops
+    * ext2 block cache with readahead, speeding up disk-heavy work such as on-OS compilation
+    * Userspace system logging: klog is now syslog, logs identify the user and program, and system services log through montauk::service_log
+    * The system log is written to disk (0:/state/logs) during shutdown
+    * Desktop sessions now own the processes started in them, and those processes are ended on logout
+    * Rust userspace support (no_std)
+    * Futex syscalls (SYS_FUTEX_WAIT/SYS_FUTEX_WAKE) as groundwork for Rust std and pthreads
+    * Process termination codes; abort() now ends the whole process from any thread
+    * isatty() now reports whether a terminal is actually attached
+    * Settings panel and a Pure Black theme in Terminal (the default for Console sessions)
+    * Syntax highlighting for C++, HTML, CSS, and JavaScript in edit and Text Editor
+    * Expanded SVG renderer
+    * New vol command for listing mounted volumes
+    * Revamped desktop application categories
+    * Fonts, icons, and manual pages moved to 0:/os; persistent state now lives in 0:/state
+    * RTL-SDR driver moved from the kernel to userspace
+
+Fixed
+=====
+    * Opening a directory as a file is now rejected on ext2 and FAT32, fixing on-OS gcc on installed systems
+    * Fixed several syscall result and status code comparisons
+    * Fixed a TLS layout bug for binaries whose TLS segment has an alignment below 16
+    * A fault in one thread now ends the whole process instead of leaving its sibling threads running
+    * Fixed an ACPI hardware ID endianness bug
+    * Fixed a race in Wi-Fi event processing and other Wi-Fi bugs
+    * DHCP no longer runs when no network device has an active link
+    * Reduced idle CPU wakeups and deferred-work polling
+    * Redirected consoles are now inherited by processes spawned with capabilities
+    * Fixed numpad navigation
+    * Fixed a timezone formatting bug
+    * Fixed the login wallpaper startup delay
+    * Fixed text overlap in file dialogs
+    * Persisted kernel logs are no longer double-spaced
+    * Files lists now scroll with the arrow keys, and volume labels are consistent
+    * mbrtowc added to libc
+
+Included in ISO
+================
+
+    Two x86_64 images are published for this release.
+    montauk-0.2.0-sdk-x86_64.iso additionally carries the
+    Montauk SDK (gcc, binutils, etc.)
+
+    * MontaukOS SMP kernel
+    * Userspace bootstrap (0:/os/init.elf) and other essential system utilities
+    * MontaukOS desktop environment and graphical apps
+    * NetSurf web browser (0:/apps/netsurf)
+    * Manual pages (0:/os/man), readable via man command
+    * Service and system configuration (0:/config: init.toml, capabilities.toml, ssh.toml, ntp.toml)
+    * System CA certificates (0:/os/certs/ca-certificates.crt)
+    * Intel Bluetooth controller and Wi-Fi (iwlwifi) firmware (0:/os/firmware/intel)
+    * Third-party license texts and notices (0:/os/licenses)
+    * Tiny C Compiler (tcc) with MontaukOS headers and libraries (0:/sdk/tcc)
+    * Lua 5.4 interpreter (lua) with headers and static library (0:/sdk/lua)
+    * Montauk SDK - native GCC/G++ and binutils (0:/sdk) - SDK image only
+    * Flat Remix icon pack
+    * JetBrains Mono, Noto Serif, Roboto, and C059 Roman (open Century Schoolbook) TrueType fonts
+    * DOOM engine with SDL_mixer audio support (game data not included)
+    * Experimental HTTP server (httpd) and default page (0:/www/index.html)
+    * Default wallpaper - blossoms by Nikhil Kumar, Unsplash (0:/os/wallpapers)
+
 MontaukOS 0.1.9 (August 13, 2026)
     (a28b8ad0b2932203b812e967f35f029f060fd6b5)
 

docs/building.md

@@ -70,10 +70,16 @@ sudo apt install build-essential nasm git curl xorriso mtools gdisk qemu-system-
 
 ### Cross-Architecture Builds
 
-To build for a different architecture, pass `ARCH=`:
+The kernel is split into architecture-neutral code and per-architecture
+ports under `kernel/src/Arch/` (see [porting.md](porting.md)). x86_64 is the
+only complete port. `aarch64` is a skeleton: the whole kernel compiles and
+links for it, but paging, exception vectors, the GIC and the timer are not
+written yet, so it does not boot. Userspace is x86_64-only.
+
+Non-x86 kernels are built with LLVM (`clang`, plus `ld.lld` for the link):
 
 ```bash
-make ARCH=aarch64
+make kernel ARCH=aarch64                    # kernel/bin-aarch64/kernel
-make ARCH=riscv64
+make kernel ARCH=aarch64 KERNEL_LD=/path/to/ld.lld
-make ARCH=loongarch64
+./scripts/check-arch-neutral.sh             # boundary check + aarch64 compile
 ```

docs/porting.md

@@ -0,0 +1,207 @@
+# Porting the Kernel to a New Architecture
+
+The kernel is written so that everything outside `kernel/src/Arch/` is
+architecture-neutral. A port is one directory, `kernel/src/Arch/<arch>/`,
+implementing the interfaces declared in `kernel/src/Arch/*.hpp` (plus a few
+interface headers elsewhere, listed below). The build compiles the generic
+tree plus `Arch/$(ARCH)/` only.
+
+`scripts/check-arch-neutral.sh` enforces the boundary and should pass before
+every commit that touches the kernel:
+
+- no inline assembly, assembly sources or architecture `#if`s outside
+  `src/Arch/` (the Limine adapter is the one exception: `limine.h` has
+  per-architecture SMP structures);
+- no generic file includes a port's private headers (`<Arch/x86_64/...>`);
+- with clang installed, the kernel compiles for `aarch64` and every symbol
+  the generic code references is defined by the port.
+
+## Layout
+
+```
+kernel/src/Arch/
+  Arch.hpp          target selection (MONTAUK_ARCH_*), MONTAUK_ARCH_NAME
+  Cpu.hpp           inline CPU primitives (interrupt state, halt, barriers,
+                    timestamp counter, TLB, page-table root)
+  Interrupts.hpp    IRQ slots, handler registration, legacy line routing,
+                    MSI message composition, IPIs
+  Smp.hpp           Smp::CpuData (generic fields + arch sub-structs),
+                    per-CPU base access, InitBsp / BootAPs
+  Context.hpp       thread context: FPU area, initial kernel stack,
+                    SwitchContext, ActivateThread, EnterUserMode
+  Abi.hpp           user ABI: ELF machine, relocation kinds, TLS variant,
+                    exit stub, initial user stack, EFI calling convention
+  Syscall.hpp       SyscallFrame contract and entry-point installation
+  Timer.hpp         per-CPU tick timer hardware
+  Platform.hpp      boot-sequence hooks, platform device list, reboot
+  CpuPower.hpp      frequency scaling / thermal governor (may be no-ops)
+  x86_64/           the complete port
+  aarch64/          the port in progress (see below)
+```
+
+Other interfaces a port implements:
+
+| Header | What the port provides |
+|---|---|
+| `Memory/Paging.hpp` | the whole `Memory::VMM::Paging` class: page-table format, kernel and user mappings, address-space creation and teardown |
+| `Io/IoPort.hpp` | port I/O (`Io::In8`...), or the absent-device fallback with `HasPortIo = false` |
+| `ACPI/AcpiSleep.hpp` | `Initialize` / `IsS3Available` / `Suspend` (S3 via the FACS waking vector is an x86 mechanism) |
+| `Common/Panic.hpp` | `Arch::PanicFrame` and `Arch::PrintPanicFrame` |
+
+Interface headers that need inline definitions include the port's header of
+the same name at their end (`Arch/Cpu.hpp` -> `Arch/<arch>/Cpu.hpp`); those
+are `Cpu.hpp`, `IoPort.hpp`, `PerCpu.hpp`, `Context.hpp`, `Abi.hpp` and
+`SyscallFrame.hpp`. Everything else is an ordinary out-of-line function.
+
+## Boot sequence
+
+`kmain` (`Boot/Main.cpp`) owns the order and calls the port at fixed points:
+
+1. `Arch::EarlyInitialize()` - descriptor tables, FPU/SIMD enable. No allocator.
+2. Frame allocator and heap come up (generic).
+3. `Arch::InitializeTraps()` - exception vectors.
+4. `Paging::Init()` switches to the kernel's own page tables, then
+   `Arch::InitializeMemoryAttributes()` (x86: PAT for write-combining).
+5. ACPI tables are parsed (generic). `Arch::InitializeInterrupts(xsdt)` brings
+   up interrupt controllers, calls `Smp::InitBsp()`, idle and power
+   management; kmain then calls `Arch::EnableInterrupts()`.
+6. PCI, AML, drivers (generic); `Timekeeping::TimerInitialize()` registers
+   `IRQ_TIMER` and calls `Arch::Timer::InitializeBsp(1)`.
+7. `Arch::InitializePlatformDevices()` (x86: PS/2).
+8. `Arch::PrepareUserMode()`, `montauk::abi::InitializeSyscalls()` (calls
+   `Arch::InitializeSyscallEntry()`), scheduler, IPC.
+9. `Smp::BootAPs()`, `Arch::LateInitialize()`, first process, idle loop.
+
+## Assumptions the generic kernel makes
+
+- 64-bit, little-endian, 4 KiB pages.
+- A higher-half direct map of physical memory (`Memory::HHDM`), as Limine
+  provides on every architecture.
+- User space in the lower half of each address space, kernel in the upper
+  half; an address space is identified by the physical address of its root
+  table (on ARM, the TTBR0 table; the kernel half lives in TTBR1).
+- Firmware tables via ACPI (`Hal::ACPI`); device-tree-only platforms would
+  need a discovery path first.
+- Interrupts are numbered IRQ slots (`Arch::IRQ_COUNT`); slot numbers for
+  MSI-capable drivers are fixed in the drivers.
+
+## The aarch64 port
+
+Real: CPU primitives (DAIF, WFI, `CNTVCT_EL0`, TLBI, TTBR0), per-CPU data via
+`TPIDR_EL1`, the context switch and EL0 entry (`Context.S`), TLS through
+`TPIDR_EL0`, the ELF/TLS/exit-stub ABI (syscall number in x8), the IRQ slot
+dispatch table (`ArchIrqDispatch`), panic-frame printing, and:
+
+- Exception vectors (`Vectors.S`, installed in `EarlyInitialize`). Every
+  exception saves one 288-byte frame that is both `Arch::PanicFrame` and
+  `Arch::SyscallFrame`. EL0 `svc` goes to `SyscallDispatch`; other
+  synchronous exceptions go to `ArchSyncException` (`Exceptions.cpp`), which
+  grows user stacks and lazy anonymous memory, kills faulting processes with
+  a crash report, and panics on kernel faults.
+- Translation tables (`Paging.cpp`): 4 KiB granule, 48-bit VAs. The kernel
+  half lives in its own TTBR1 table; `KernelRoot()` is the TTBR0 table of
+  kernel contexts (EFI identity mappings only, so null pointers fault). The
+  HHDM maps only the memory map's RAM and firmware regions, never the holes
+  between them (device space on ARM). MAIR: 0 Normal WB, 1 Normal NC
+  (`MapWC`), 2 Device-nGnRE (`MapMMIO`).
+- GICv2 (`Interrupts.cpp`, from the MADT): IRQ slots map onto INTIDs --
+  `IRQ_TIMER` onto the generic timer's PPI, the MSI slots (24-45) onto the
+  SPIs of a GICv2m MSI frame (`ComposeMsi`), the IPI slots onto SGIs 0/1,
+  and the platform lines (1-23) onto the SPIs `AssignPlatformLine` gives
+  them (see "Interrupt lines" below).
+  Interrupts are completed (EOI) before the handler runs, as on x86; a
+  level-triggered source drops its line first in an ack hook.
+- Generic timer tick (`Timer.cpp`): the EL1 virtual timer at
+  `CNTFRQ_EL0`, its PPI from the GTDT, re-armed from the ack hook.
+- PSCI `SYSTEM_RESET` / `SYSTEM_OFF` (`Arch::Reboot` / `Arch::PowerOff`)
+  through the conduit the FADT's ARM_BOOT_ARCH names. Hardware-reduced
+  ACPI has no PM1 block, so `AcpiShutdown` leaves power-off to the port.
+- `Arch::SyncInstructionCache` after the ELF loader writes code: ARM
+  instruction fetch does not snoop data writes (QEMU hides this; real
+  cores do not). EL0 may maintain caches itself (`SCTLR_EL1.UCI`).
+
+- SMP (`Smp.cpp`): Limine wakes each secondary CPU (PSCI `CPU_ON`
+  underneath) on the bootloader's tables; it programs its own system
+  registers and the kernel translation regime (`Bringup.hpp`), its banked
+  GIC CPU interface and timer PPI, then idles until a reschedule SGI.
+  TLB maintenance is broadcast (`tlbi ...is`).
+
+The generic kernel is compiled `-fsigned-char` for AArch64: it was written
+for x86's signed plain `char`.
+
+Building it needs `clang` and `ld.lld`; without a system `ld.lld` the
+kernel makefile uses the one rustup ships
+(`~/.rustup/toolchains/*/lib/rustlib/*/bin/gcc-ld/ld.lld`), and `KERNEL_LD=`
+overrides both. `make run ARCH=aarch64` boots the full image (desktop, apps,
+USB keyboard and mouse) on QEMU `virt` with a GICv2 and Cortex-A72 cores, the
+Raspberry Pi 4/400's configuration; it needs `qemu-system-aarch64` and
+`qemu-efi-aarch64`, and the userspace compiler
+(`./toolchain/build-montauk-toolchain.sh aarch64`). `KERNEL_ONLY=1` builds an
+image without userspace.
+
+The kernel console is mirrored to the serial port the ACPI SPCR table names
+(PL011, SBSA UART, 16550 or the BCM2835 mini UART; `Drivers/Serial/Uart.cpp`),
+including the log written before the port was found. On QEMU that is
+`-serial stdio`.
+
+## Interrupt lines
+
+Drivers for PCI functions without MSI read the Interrupt Line register and
+pass it to `Arch::EnableLegacyIrq`. On x86 the firmware fills that register
+with the function's I/O APIC input (`Arch::PlatformLinesAreFixed()`). On
+other platforms `Pci::AssignInterruptLines` (`Pci/PciAcpi.cpp`) fills it
+instead: it finds each pin's global system interrupt in the root bridge's
+`_PRT` (swizzling through PCI-to-PCI bridges, resolving PNP0C0F link
+devices through `_CRS`), binds the GSI to a platform line with
+`Hal::GsiRouting::LineForGsi` and writes the line number into the register.
+Devices that ACPI describes directly (platform devices) get their line from
+`LineForGsi` too.
+
+## DMA
+
+`Memory/Dma.hpp` serves devices whose DMA does not snoop the CPU caches
+(ACPI `_CCA = 0`): pages for rings and descriptors come uncached, and
+`SyncForDevice` / `SyncForCpu` clean and invalidate streaming buffers
+around a transfer (`Arch::CleanDataCache` / `InvalidateDataCache`). For
+coherent devices every call is the plain allocation or a no-op. The xHCI
+driver uses it; it also orders its TRB writes with `Arch::DmaWriteFence`
+(the cycle bit last, a control TD's Setup TRB after its Data and Status
+TRBs) and its event reads with `Arch::DmaReadFence`. Those fences cost
+nothing on x86 and are required on ARM, where stores to normal memory and
+device registers may otherwise be observed out of order. The other PCI
+drivers (AHCI, NVMe, e1000, HDA) do not have them yet: they work on QEMU,
+which does not reorder, but need them before they meet real ARM hardware.
+
+## Raspberry Pi 4 / 400
+
+See `docs/raspberry-pi.md`: `make ARCH=aarch64 rpi4-sd`, the firmware it
+expects and how the Pi's hardware is reached.
+
+## Userspace
+
+The programs are architecture-neutral C/C++ built by `<arch>-montauk-gcc`
+(see `toolchain/README.md`). `programs/mk/arch.mk` holds everything
+architecture-specific in the build: the toolchain prefix, FP/SIMD flag sets
+(`ARCH_FP_FLAGS` for GUI apps, `ARCH_NOFP_FLAGS` for system programs) and the
+output layout. x86_64 keeps `bin/`, `obj/` and libraries next to their
+sources; other architectures build beside them into `bin-<arch>/`,
+`obj-<arch>/` and `lib/<name>/build-<arch>/`, so both can be built from one
+tree. The few architecture-specific sources are the raw system-call
+wrappers (`#if defined(__x86_64__)` / `__aarch64__` blocks in
+`montauk/syscall.h`, `libc/montauk.h`, `lib/libc/libc.c` and
+`lib/libc/crt/crt1.c`), `lib/libc/arch/setjmp-<arch>.S` and the scanout
+store fence in `gui/framebuffer.hpp`.
+
+Not built for AArch64: tcc (it generates x86 code), the on-OS SDK and the
+Rust programs (their `x86_64-unknown-montauk` target).
+
+## Adding another architecture
+
+1. Create `kernel/src/Arch/<arch>/` with the six inline headers listed above
+   and stubs for the out-of-line functions (copy `aarch64/` as a template).
+2. Add the target to `Arch/Arch.hpp` and to the `#if` chains at the end of
+   the dispatching interface headers.
+3. Add `KernelStartSymbol` / `KernelEndSymbol` to `kernel/linker-<arch>.ld`,
+   and the `ld.lld` emulation to `LD_EMULATION_<arch>` in `kernel/GNUmakefile`.
+4. Teach `Boot/Protocols/LimineProtocol.cpp` the architecture's MP fields.
+5. Run `./scripts/check-arch-neutral.sh <arch>` until it passes.

docs/power.md

@@ -1,117 +0,0 @@
-# CPU Power and Thermal Management
-
-MontaukOS manages CPU frequency, idle states, and package temperature on
-Intel hardware. Without this, a modern many-core laptop part (the reference
-target is a TUXEDO Gemini Gen2 with an i9-13900HX, 24 cores / 32 threads)
-runs at whatever P-state the firmware left it, never drops below C1 when
-idle, and heats until the hardware throttles at TjMax - or until the
-embedded controller intervenes (on Clevo boards: forced full-volume beeping,
-then a hard power-off).
-
-Everything below is feature-gated by CPUID, so virtual machines (KVM masks
-the power bits in CPUID.06H) and non-Intel CPUs skip it safely.
-
-## Components
-
-### HWP frequency scaling (`Hal/CpuPower.cpp`)
-
-`Hal::CpuPower::InitializeBsp()` enables Hardware P-states
-(IA32_PM_ENABLE), reads the performance range from IA32_HWP_CAPABILITIES,
-and programs IA32_HWP_REQUEST with:
-
-- min = lowest performance level, max = highest
-- desired = 0 (fully autonomous hardware frequency selection)
-- EPP = 0x80 (balanced; same as Linux "balance_performance")
-
-This is the same mechanism intel_pstate uses on Linux and is the single
-biggest thermal win: the CPU idles near its minimum frequency and ramps
-only while there is work. APs apply the same request in `ApEntry` via
-`InitializeAp()` (HWP request MSRs are per-logical-CPU).
-
-CPUs with EPB but no HWP get IA32_ENERGY_PERF_BIAS = 6 (balanced) instead.
-
-### C1E promotion
-
-MSR_POWER_CTL bit 1 lets HLT-idle drop core voltage/frequency to minimum
-(C1E) instead of parking at the current ratio. Set per-core on the BSP and
-every AP. Gated on the DTS CPUID bit as a real-hardware heuristic, because
-MSR_POWER_CTL is not architectural and hypervisors may not emulate it.
-
-### Thermal governor (`Hal/CpuPower.cpp`)
-
-A closed-loop passive-cooling controller on the BSP, stepped from
-`Sched::RunBspMaintenance()` (reached from both the idle loop and the
-timer tick, so it keeps running under full load). Every 500 ms it reads
-the package digital thermal sensor (IA32_PACKAGE_THERM_STATUS readout
-relative to TjMax from MSR_TEMPERATURE_TARGET) and adjusts the HWP
-performance ceiling:
-
-- **>= 85 C (passive):** step the ceiling down 2 ratio units (~200 MHz)
-- **>= 94 C (hot):** step down 8 units per interval
-- **<= 78 C (resume):** step back up 1 unit per interval
-- between resume and passive: hold (hysteresis)
-
-The ceiling propagates through a policy epoch counter: each CPU compares
-the epoch in `Sched::Tick()` (and the AP idle loop) against the last one
-it applied and rewrites its own IA32_HWP_REQUEST when it changed, since
-MSR writes only affect the executing core. APs tick every 10 ms, so a
-throttle decision reaches all cores within one tick.
-
-Throttle engage/release transitions are logged (WARNING/OK, "CpuPower").
-
-### Deep AP idle (`ACPI/CpuIdle.cpp`)
-
-APs use deep MWAIT C-states while idle; C6 power-gates the core entirely,
-which is where most of the idle-heat win lives on many-core parts.
-
-The hint comes from ACPI `_CST` FFH entries (the GAS address low byte is
-the MWAIT hint, same convention as Linux), validated against the CPUID.05H
-sub-state enumeration and capped at the C6 family (0x2F). With no usable
-`_CST`, the deepest CPUID-enumerated hint (up to C6) is used.
-
-**The BSP stays in C1 (HLT).** On some laptops deep MWAIT states freeze
-the LAPIC timer even though ARAT is advertised, and the BSP's periodic
-1 ms tick is the system timekeeper (see the comment in
-`Timekeeping::IdleOnce`). APs have no timekeeping duty: an idle AP is
-woken by the reschedule IPI, and its coarse tick resumes on wake.
-
-### S3 wake
-
-IA32_PM_ENABLE, HWP requests, and MSR_POWER_CTL do not survive suspend.
-`Hal::CpuPower::ReapplyAfterWake()` runs in the S3 resume path, redoes the
-BSP setup, and bumps the policy epoch so every AP re-applies its HWP
-request (and re-checks C1E) on its next tick.
-
-## Diagnostics
-
-`SYS_POWERINFO` (149) fills a `PowerInfo` struct: package temperature,
-TjMax, HWP state and performance range, the governor's current ceiling,
-CPUID base/turbo frequency, the measured average active frequency
-(APERF/MPERF over the 500 ms governor window), and the AP idle MWAIT hint.
-Returns -1 when the CPU exposes no power features (e.g. under QEMU).
-
-The `power` program (`programs/src/power`) prints it:
-
-```
-power             one-shot snapshot
-power watch       1 Hz samples for 60 s (temp, avg freq, ceiling)
-power watch 300   same, for 300 s
-```
-
-Boot log lines to look for (Kernel Log app):
-
-```
-[CpuPower] HWP enabled: perf range 4-54, EPP=0x80
-[CpuPower] Thermal governor armed: TjMax=100C, passive=85C, hot=94C, ...
-[CpuIdle]  AP deep idle: MWAIT hint 0x20
-```
-
-## Tuning
-
-The governor constants live at the top of `Hal/CpuPower.cpp`
-(`PassiveTempC`, `HotTempC`, `ResumeTempC`, `GovernorIntervalMs`, step
-sizes) and were chosen to keep the package well below both TjMax (100 C
-on the i9-13900HX) and the Clevo EC's emergency thresholds. If the fans
-still spin up aggressively under sustained load, lower `PassiveTempC`
-first; if builds feel throttled while cool, raise the recovery rate
-instead of the trip points.

docs/raspberry-pi.md

@@ -0,0 +1,344 @@
+# Raspberry Pi 4 / 400
+
+MontaukOS runs on the Raspberry Pi 4 and 400 through the Pi Firmware Task
+Force's UEFI firmware ([pftf/RPi4](https://github.com/pftf/RPi4), an EDK2
+build) in ACPI mode, booting Limine like any other UEFI machine.
+
+## Building and writing the SD card
+
+```bash
+./toolchain/build-montauk-toolchain.sh aarch64   # once
+make ARCH=aarch64 rpi4-sd                        # -> montauk-rpi4.img
+sudo dd if=montauk-rpi4.img of=/dev/sdX bs=4M conv=fsync
+```
+
+Or use the imager, which also updates a card in place:
+
+```bash
+make rpi4-imager                                 # = ./scripts/montauk-imager.py
+```
+
+### The imager
+
+`scripts/montauk-imager.py` is a Tk window (`python3-tk`) that lists the SD
+cards and USB drives attached (never the host's own disks) and what is on
+them, and either writes a fresh image or updates the MontaukOS on a card. It
+runs as the user and does the writing by starting itself again through
+`pkexec` (it asks for the password each time). The same operations work from
+a terminal: `list`, and as root `flash`, `update` and `expand`
+(`--help` for the options). Images can be raw or `.gz`, `.xz` or `.zip`
+compressed; **Build** runs `make ARCH=aarch64 rpi4-sd`.
+
+- **Write a fresh image** erases the card, writes the image, clears the last
+  MiB of the card (a backup GPT from an earlier OS) and reads the card back to
+  verify it. *Expand* grows the root to fill the card right away instead of
+  leaving it to `firstboot.elf`.
+- **Update** leaves the partition table alone and finds both partitions by
+  label, so a root that has grown to fill the card (or was resized by hand)
+  stays as it is. On the ESP every file is written except `RPI_EFI.fd`, which
+  holds the firmware's own settings (*Update the UEFI firmware* replaces it);
+  a changed `config.txt` is kept as `config.txt.bak`. On the root, `os/`,
+  `apps/`, `sdk/`, `boot/` get the image's new and changed files; `config/`
+  and `www/` only get files the card lacks, `users/` and `state/` are not
+  touched. `init.toml` becomes the image's, keeping each service's `enabled`
+  from the card and appending services only the card has; `capabilities.toml`
+  becomes the image's plus grants only the card has (old copies: `.old`).
+  `os/firstboot.elf` is only written if the card still has it, so a grown
+  card does not grow again. *Remove system files the new image no longer
+  ships* prunes `os/`, `apps/`, `sdk/` and `boot/` (keeping `os/btkeys.bin`,
+  the Bluetooth bonds). Unchanged files are skipped and each file is
+  replaced by a rename; the volume is checked with `e2fsck` before and after,
+  and its features against what `Fs/Ext2.cpp` mounts.
+- **Expand** (a host-side first boot) grows the root partition to the end of
+  the card, at most 1 TiB, and the volume with `resize2fs`; `firstboot.elf`
+  then finds nothing to do and deletes itself.
+
+While it works, a runtime udev rule keeps the desktop from automounting the
+card's partitions (Linux writes to an ext2 volume as soon as it mounts it).
+
+`scripts/mk-rpi4-sd.sh` downloads the firmware release once into
+`rpi4-firmware/` (`PFTF_VERSION=v1.53` by default) and builds an MBR image:
+
+| Partition | Type | Contents |
+|---|---|---|
+| 1 | 0xEF, FAT32 | the firmware files, Limine (`EFI/BOOT/BOOTAA64.EFI`), `boot/kernel` |
+| 2 | 0x83, ext2, label `MONTAUKROOT` | the system tree (the contents of `ramdisk.tar`) |
+
+Limine loads only the kernel; the `limine.conf` on the card has no ramdisk
+module. Finding no ramdisk, the kernel mounts the partition labelled
+`MONTAUKROOT` as `0:/` (`Fs/Boot.cpp`), read through the SD driver
+(`Drivers/Storage/Sdhci`), and the ESP after it. Files load as they are
+used instead of the firmware reading the whole 80 MB tree before boot.
+The card can also boot from a USB card reader (or the image from a USB
+stick): the root then comes through `Drivers/USB/MassStorage.cpp`. A USB
+root can register after the boot scan (a SuperSpeed link still training, a
+reader still bringing its card up), so a boot that finds no `MONTAUKROOT`
+waits up to 10 s for one, servicing USB hot-plug meanwhile.
+Unlike the ramdisk, the root is persistent: changes to `0:/` survive a
+reboot, so shut down from the desktop rather than pulling power.
+The image is sized to its contents (`ROOT_FREE_MIB`, default 64, sets the
+free space on the root) and grows to fill the card on first boot.
+
+### First boot: growing the root
+
+Only the SD image ships `0:/os/firstboot.elf` (`programs/src/firstboot`,
+built into `programs/bin-aarch64-rpi/`, which the script lays over the
+system tree), and the script adds its init service, ahead of all others,
+and its capability grant (`storage_admin`, `system_image`) to the image's
+`init.toml` and `capabilities.toml`. On first boot it shows a status card
+in the login screen's style while it calls `SYS_FS_GROW` on drive 0 in a
+loop:
+
+1. The kernel grows the MBR entry of the root partition as far as the
+   filesystem can use (to the end of the card, or 1 TiB), on a 1 MiB
+   boundary (`Gpt::ResizeMbrPartition`).
+2. The ext2 driver grows the mounted volume (`Fs/Ext2.cpp`, "Online
+   grow"): each call writes the bitmaps and zeroed inode tables of a few
+   new block groups past the old end, then the last call commits the
+   bitmaps, descriptor table and superblock. The descriptor table grows
+   into the blocks `mke2fs -O resize_inode -E resize=2^28` reserved; the
+   leftover reservation is freed and `resize_inode` cleared, so the volume
+   grows once.
+
+The new groups' inode tables are the bulk of the writes (0.1% of the card,
+one inode per 256 KiB), written as long multi-block runs. Once the volume has grown, `firstboot.elf` deletes
+itself and init skips its (optional) service from then on. If it fails it
+stays and the next boot tries again; an interrupted grow is safe, since the
+new groups are only referenced once the superblock is written, and a
+partition larger than its filesystem is valid.
+
+`make ARCH=aarch64 rpi4-sd ROOTFS=ramdisk` builds the old layout instead:
+one FAT32 partition with `boot/ramdisk.tar`, which Limine loads through the
+firmware. It boots slowly but does not depend on the kernel's SD driver.
+
+`make ARCH=aarch64 run-rpi4-sd` boots the ext2 image on QEMU with the card
+on `sdhci-pci`, so the root comes through the same driver. QEMU's UEFI has
+no SD driver, so the card's ESP is also attached on its own over virtio for
+the firmware to load Limine from. The card is padded to
+`RPI4_QEMU_CARD_MIB` (default 4096), and every run starts from a fresh
+copy, so every run is a first boot.
+
+The image's `RPI_EFI.fd` comes with a **MontaukOS** boot entry first in the
+firmware's boot order (`scripts/rpi4-efi-bootorder.py` writes it into the
+release's empty variable store), so the card boots on its own instead of
+falling behind the PXE/HTTP entries the firmware creates. The entry names the
+ESP by the image's MBR disk ID and partition number, so it works from the SD
+slot or a USB drive. A card whose `RPI_EFI.fd` predates it keeps its old
+boot order until it is updated with *Update the UEFI firmware*
+(`--firmware`), which also copies the image's disk ID to the card.
+
+The firmware defaults are what MontaukOS needs; leave them:
+
+- **System Table Mode: ACPI.** The kernel reads ACPI only.
+- **Limit RAM to 3 GB: enabled.** The BCM2711's PCIe can only reach the
+  first 3 GB (its tables say so in `_DMA`). The kernel warns at boot when
+  RAM extends past a device's `_DMA` range.
+
+## Serial console
+
+The kernel console (the whole boot log, including what was printed before
+the port was found) is mirrored to the UART the firmware's SPCR table names.
+The firmware's `config.txt` applies the `miniuart-bt` overlay, which puts the
+PL011 on GPIO 14/15: header pins 8 (TX) and 10 (RX), ground on pin 6, 115200
+8N1, 3.3 V levels. On the Pi 400 the header is on the back.
+
+## How the hardware is reached
+
+| Part | Source | Driver |
+|---|---|---|
+| 4x Cortex-A72 | MADT, PSCI (TF-A, SMC) | `Arch/aarch64` SMP via Limine MP |
+| GIC-400 (GICv2) | MADT | `Arch/aarch64/Interrupts.cpp` |
+| Generic timer | GTDT | `Arch/aarch64/Timer.cpp` |
+| Display | UEFI GOP (VideoCore) | Limine framebuffer |
+| Keyboard (built in), USB ports | VL805 xHCI as ACPI `\_SB.SCB0.XHC0` (`PNP0D10`) | `Drivers/USB/Xhci.cpp` platform probe |
+| USB 2 devices (built-in keyboard too) | VIA hub on the VL805's USB 2 root port | `Drivers/USB/Hub.cpp` |
+| Reset, power-off | PSCI | `Arch::Reboot`, `Arch::PowerOff` |
+| Bluetooth (CYW43455 on the 4B, CYW43456 on the 400) | `BCM2EA6` (`\_SB.GDV0.BTH0`) on the mini UART (`BCM2836`) | `Drivers/Bluetooth/Broadcom.cpp`, `Drivers/Serial/BtUart.cpp` |
+| VideoCore firmware mailbox (radio power, clocks, board MAC) | `BCM2849` (`RPIQ`) | `Drivers/Platform/RpiFirmware.cpp` |
+| GPIO pin functions | `BCM2845` (`GPI0`) | `Drivers/Platform/RpiGpio.cpp` |
+| Ethernet (GENET v5 MAC, BCM54213PE PHY) | `BCM6E4E` (`\_SB.ETH0`) | `Drivers/Net/Genet.cpp` |
+| Wi-Fi (the same CYW4345x, SDIO) | Arasan SD controller `BCM2847`, GPIO 34-39 | `Drivers/Net/Wifi/Brcmf*.cpp` |
+
+The VL805 sits behind the BCM2711's PCIe, whose config space is not ECAM
+(the root bridge's `_DSD` says `linux-ecam-quirk-id = "bcm2711"`), so the
+kernel does not enumerate PCI there. The firmware has already brought up the
+link and the VL805 and describes the controller as a platform device: a
+register block at `0x600000000`, GSI 175, and `_CCA = 0`, i.e. its DMA does
+not snoop the CPU caches. `XHC0._INI` (run by the AML interpreter) enables
+the controller's PCI memory decoding and bus mastering. The xHCI driver then
+keeps its rings and contexts in uncached memory, bounces control-transfer
+data through an uncached page, and cleans/invalidates the caller buffers of
+bulk transfers (`Memory/Dma.hpp`). It also watches the event ring from the
+timer tick, so input keeps working if the INTx line never fires.
+
+The VL805 has one USB 2 root port (port 1) and four USB 3 ones. An on-board
+VIA hub (`2109:3431`) sits on the USB 2 port, and the USB 2 lines of every
+socket, and the Pi 400's internal keyboard (`04d9:0007`), hang off it: a
+keyboard or mouse is always a low/full-speed device behind a high-speed hub,
+reached through that hub's transaction translator.
+
+## Bluetooth
+
+The radio's Bluetooth half speaks HCI over a UART (H4) on GPIO 30-33 (CTS,
+RTS, TXD, RXD); its power switch, `BT_ON`, is line 0 of the VideoCore
+firmware's GPIO expander, reached through the mailbox. The firmware tables
+describe it as `BTH0`, whose `UARTSerialBus` names the UART it uses: with
+the `miniuart-bt` overlay the PL011 is the console, so Bluetooth has the mini
+UART. The kernel refuses to take a UART that is the SPCR console.
+
+At boot the driver probe routes the pins (`ALT5` for the mini UART, `ALT3`
+for the PL011), starts the UART (interrupt driven, 16 KiB receive and
+transmit rings) and the H4 transport under the shared HCI layer. Once the
+root is mounted the idle loop finishes, as for USB adapters:
+
+1. power-cycle the controller through `BT_ON`, HCI Reset at 115200 baud;
+2. Read Local Version: the ROM's LMP subversion picks the patch file,
+   `0:/os/firmware/brcm/BCM4345C0.hcd` (0x6119, 4B) or `BCM4345C5.hcd`
+   (0x6606, 400), shipped unmodified from Raspberry Pi OS's bluez-firmware;
+3. raise both ends to the operating speed (vendor `0xFC18`);
+4. Download Minidriver (`0xFC2E`), then every record of the `.hcd` file
+   (Write RAM, then Launch RAM), as Linux's `btbcm` does;
+5. the patched firmware restarts at 115200: reset, raise the speed again.
+
+The mini UART's FIFO is 8 bytes deep. Its hardware RTS is tried first; if
+the controller answers with it on, the link runs at 921600 baud, otherwise
+at 460800 without flow control (what Raspberry Pi OS uses), and at 115200
+if the UART has no interrupt. The baud rate derives from the VPU core
+clock, so `mk-rpi4-sd.sh` pins it (`core_freq=500`, `core_freq_min=500` in
+`config.txt`), as the overlay's documentation requires. A failed fast
+bring-up is retried entirely at 115200.
+
+The CYW43456 reports a chip-wide default address (`43:45:C5:00:1F:AC`);
+the driver replaces it with the board's Ethernet MAC from the firmware.
+`0:/config/bluetooth.toml` still overrides both (Broadcom `0xFC01`).
+
+## Ethernet
+
+The BCM2711's own Gigabit MAC, GENET v5 (a UniMAC behind a DMA engine),
+talks RGMII to a BCM54213PE PHY at MDIO address 1. The firmware describes
+it as `BCM6E4E`: a 64 KiB register block, two interrupts (the driver uses
+the first, which carries the default queue), `_CCA = 0`, and a `_DSD` with
+`phy-mode` (`rgmii-rxid`: the PHY delays the receive clock, the MAC the
+transmit clock) and `brcm,max-dma-burst-size`.
+
+The driver uses only the default queue (ring 16), with all 256 descriptors
+of each direction. The descriptors live in the MAC's own SRAM, so only the
+2 KiB packet buffers need cache maintenance. Received frames start two
+bytes into their buffer (`RBUF_ALIGN_2B`); neither direction uses the
+64-byte status blocks, so there is no checksum offload. The UniMAC accepts
+broadcast and its own address only (the MDF filter).
+
+The interrupt only acknowledges and masks; frames are dispatched from the
+idle bottom half, which also checks the ring itself, so receive works if
+the interrupt never arrives. The PHY is polled once a second (its interrupt
+pin is not described); a link change reprograms the RGMII block and the
+UniMAC's speed and duplex. A link the firmware already negotiated is kept,
+otherwise the PHY is reset and advertises 10/100 and 1000 full duplex.
+
+The MAC address is the board's, read from OTP through the firmware mailbox
+(the one the Bluetooth driver uses too); the address the firmware's own
+driver programmed is the fallback. `dhcp.elf` waits up to 10 s for a wired
+link, since autonegotiation takes about 3 s after the PHY reset.
+
+## Wi-Fi
+
+The radio's WLAN half is an SDIO card on the BCM2711's older SD controller,
+the Arasan-derived EMMC (`BCM2847`, `0x7E300000`); the card slot is on
+EMMC2, so the two never share a bus. Its pins are GPIO 34-39 (ALT3; clock
+without a pull, the rest pulled up, as Linux's `sdio_pins`) and its power
+switch, `WL_ON`, is line 1 of the firmware's GPIO expander. The probe
+claims the controller only while the slot is on EMMC2 (`BRCME88C` present);
+if the tables do not describe the Arasan, its address is taken as EMMC2's
+less 256 KiB.
+
+It is a FullMAC radio: the firmware does the 802.11 MLME itself and the host
+sees Ethernet frames. The driver follows Linux's brcmfmac (ISC license,
+Broadcom; its notice is in `NOTICES.txt`):
+
+| File | Part |
+|---|---|
+| `BrcmfSdio.cpp` | the Arasan as an SDIO host: CMD5/3/7, 4-bit at 25 MHz, CMD52/CMD53, polled PIO, 32-bit register access |
+| `BrcmfChip.cpp` | backplane window (function 1), clocks, the enumeration ROM's core list, firmware + NVRAM download, function 2, the firmware console |
+| `BrcmfProto.cpp` | SDPCM framing and transmit credit, CDC ioctls, BDC data, events, the receive pump, the transmit queue |
+| `BrcmfConnect.cpp` | preinit and CLM, escan, join, keys, link state |
+
+Once the root is mounted the idle loop:
+
+1. routes the pins, switches the EMMC clock on, pulses `WL_ON`, enumerates
+   the card (function 1 blocks of 64 bytes, function 2 of 512);
+2. requests the ALP clock, reads the chip ID: `0x4345` revision 9 is the
+   CYW43456 (Pi 400), any other revision the CYW43455 (Pi 4B), which picks
+   `0:/os/firmware/brcm/brcmfmac43456-sdio.*` or `brcmfmac43455-sdio.*`
+   (shipped unmodified from Raspberry Pi OS's firmware-nonfree);
+3. halts the ARM Cortex-R4, holds the 802.11 core in reset, sizes the RAM
+   from the CR4's banks, writes the firmware at `0x198000` and the NVRAM
+   (stripped to `key=value\0` pairs and a length token) at the top of RAM,
+   reading both back to verify, puts the reset vector at address 0 and
+   releases the CPU;
+4. waits for the HT clock, sends the SDPCM protocol version through the
+   mailbox, enables function 2, the F2 watermark and MES busy signalling of
+   Linux's CYW43455 settings, save/restore if the chip has it, and the card
+   interrupt;
+5. reads `cur_etheraddr` (the NVRAM's shared placeholder address
+   `b8:27:eb:74:f2:6c` is replaced with the board address + 1, as on
+   Raspberry Pi OS), loads the CLM blob (`clmload`, 1400-byte chunks: these
+   firmware builds carry no regulatory data of their own), turns off glom,
+   roaming, ARP/ND offload and power save, enables the events it handles and
+   brings the interface up (`WLC_UP`).
+
+Nothing uses the controller's interrupt: the bus is swept from the idle
+loop (and from the timer tick when the cores never idle), every 10 ms while
+a scan or a connection is active and every 250 ms otherwise, and at once when
+the controller shows the card interrupt. Every sweep reads the SDIO core's
+interrupt status, then every frame waiting in function 2 (a 4-byte header
+read, then the rest). Ioctls wait for transmit credit and their answer by
+sweeping the bus themselves. Frames are delivered with the bus unlocked, and
+transmitting never waits for the bus: data frames go into a 32-frame queue
+that whoever holds the bus, or the next sweep, sends.
+
+Scans are escans (all channels, active, firmware dwell times); each result
+event carries a `brcmf_bss_info_le` whose information elements go into the
+same scan table the Intel driver fills from beacons. A join sets `wpaie`
+(the RSN element our supplicant builds), `wpa_auth`, `auth` and `wsec`, then
+`join` with the BSSID and the chanspec escan reported. The 4-way handshake
+runs in the host's supplicant (`Wpa.cpp`) as for the Intel adapter: EAPOL
+frames are ordinary Ethernet frames here, and the keys go to the firmware
+with `wsec_key` once message 4 has left the transmit queue. The radio does
+AES-CCMP only: GCMP networks are refused, as are WPA3-only (SAE), WEP and
+TKIP ones.
+
+Unanswered ioctls (a leaky bucket of three), a halt reported through
+the mailbox, or repeated bus errors mark the radio failed: the firmware's
+console and any trap record are printed to the kernel log (`WiFi-FW`), and
+the facade's recovery powers the radio off and repeats the bring-up (at
+most three times).
+
+`tests/wifi/run.sh` drives `BrcmfProto.cpp` and `BrcmfConnect.cpp` with the
+real supplicant against a simulated firmware: preinit and the CLM chunks,
+escan parsing, open and WPA2 joins (an independent authenticator checks
+messages 2 and 4 and the installed keys), the wrong passphrase, refusals,
+join timeouts and a firmware that stops answering.
+
+## Not supported yet
+
+- SD high-speed mode: the driver transfers by polled PIO at 25 MHz on a
+  4-bit bus (multi-block CMD18/CMD25 with Auto CMD12), at most 12.5 MB/s;
+  switching the card to 50 MHz (CMD6) would double that.
+- Wi-Fi beyond 802.11 station mode on one interface: no access point mode,
+  no SDIO high speed (50 MHz), no glom, no firmware supplicant.
+- Audio (HDMI via VideoCore, the 3.5 mm jack via PWM).
+
+## Status
+
+Bluetooth works on a Pi 400 (CYW43456), including A2DP music to
+headphones. Ethernet has not run yet: QEMU has no GENET model. Wi-Fi has
+not run on hardware yet either (QEMU has no model of the radio); only its
+protocol and connection layers are tested, in the host harness. It is untested on a Pi 4B (CYW43455), and QEMU has no model of
+the radio.
+
+Everything else above is verified on QEMU only, including the Pi-specific path:
+the firmware's own XHC0 table, retargeted at QEMU's xHCI and loaded with the
+`bcm2711` quirk on QEMU's root bridge, boots to the desktop with non-coherent
+DMA. Real hardware has not been tried yet. The ext2 root on the SD card boots on QEMU (`run-rpi4-sd`); the
+first-boot grow has been tested only against the kernel's ext2 code in a
+host harness (images grown, then checked with `e2fsck`).

docs/ssh.md

@@ -1,100 +0,0 @@
-to-do: rewrite & convert to html for docs pages
-
-# SSH
-
-MontaukOS ships an SSH-2 server, `sshd`, that gives a remote client a shell on
-the machine. It speaks enough of the protocol to interoperate with stock
-OpenSSH clients:
-
-```
-ssh dan@montauk-box            interactive shell
-ssh dan@montauk-box 'ls 0:/'   run one command
-```
-
-Remote access is **off by default**. Turn it on from Settings > SSH Server,
-which flips `services.ssh.enabled` in `0:/config/init.toml`; `init` starts the
-daemon on the next boot.
-
-Two gates stand in front of a login: the account must exist in the MontaukOS
-user database with a valid password, and it must be listed in the `[allow]`
-table of `0:/config/ssh.toml`. A user absent from that table is refused even
-with the right password. The Settings applet maintains the table; `sshd` never
-writes it.
-
-## Layout
-
-```
-programs/src/sshd/
-  main.cpp        transport framing, key exchange, auth, session/channel loop
-  crypto.hpp      host key storage, DH group 14 arithmetic, cipher/MAC wrappers
-programs/src/sshserver/
-  main.cpp        Settings applet: enable the service, edit the allow table
-programs/include/montauk/ssh.h
-                  shared policy helpers used by both
-```
-
-The crypto primitives (SHA-256, HMAC, AES-CTR, RSA signing, the DRBG) come from
-BearSSL. The modular exponentiation for Diffie-Hellman is in `crypto.hpp`
-because BearSSL exposes no public bignum API.
-
-## What it implements
-
-| | |
-|---|---|
-| Key exchange | `diffie-hellman-group14-sha256` |
-| Host key | `rsa-sha2-256`, 2048-bit, generated on first start |
-| Cipher | `aes128-ctr` |
-| MAC | `hmac-sha2-256` |
-| Compression | none |
-| Auth | password only |
-
-Client-initiated rekeying works, so long sessions survive OpenSSH's 1 GiB /
-one-hour rekey threshold. Channel windows are tracked in both directions, so
-sessions that move more than a window's worth of data keep flowing.
-
-## What it does not
-
-No public key authentication, no port or X11 forwarding, no SFTP or SCP. One
-session channel per connection, and connections are served one at a time --
-there is no `fork`, so a second client waits for the first to finish. A
-pre-authentication timeout keeps an idle peer from holding the daemon shut.
-
-`exec` requests are run by typing the command into an interactive shell rather
-than executing it directly, so the prompt and the echoed command come back
-mixed into the output. This is why `scp` and friends will not work as-is.
-
-## The host key
-
-Generated on first start and written to `0:/config/ssh_hostkey.toml`, separate
-from the policy file so the unprivileged Settings applet -- which rewrites
-`ssh.toml` whenever the allow table changes -- never round-trips private key
-material.
-
-That separation is hygiene, not protection: **the filesystem has no permission
-model**, so any local process can read the host key. Closing that hole needs
-FS-level access control, at which point the key file should be restricted to
-the account `sshd` runs as. A key found in an older `ssh.toml` is migrated to
-the new file on first start so the host key does not change under existing
-clients.
-
-## Debugging
-
-`sshd` calls `montauk::print`, which reaches the boot console only. **It has no
-usable diagnostics once the desktop is up**, pending the unified syslog.
-
-Worth understanding, because it catches out anything written as a daemon:
-`montauk::print` is `SYS_PRINT`, which writes the kernel *terminal*, not the
-kernel *log*. Only in-kernel `KernelLogStream` writes raise `g_kernelLogDepth`,
-and only those append to the ring buffer `SYS_LOG` reads -- so daemon output
-never shows up in `klog`. On top of that, `Sys_Print` returns early once
-`g_suppressKernelLog` is set, which the desktop does at startup, so the output is
-discarded outright from then on. `init` spawns services with `spawn` rather than
-`spawn_redir`, so there is no stream to capture either.
-
-Until there is somewhere for it to go, debugging a handshake means watching the
-boot console, or temporarily pointing the print calls at a file.
-
-Failures print a numeric code -- `key exchange failed (N)`, `authentication
-failed (N)` -- where N identifies the step; see the `kex_error` assignments in
-`main.cpp`. Running the client with `ssh -vvv` and lining its trace up against
-those codes is usually the fastest way to find where a handshake diverged.

docs/wifi.md

@@ -1,395 +0,0 @@
-to-do: rewrite & convert to html for docs pages 
-
-# Wi-Fi
-
-MontaukOS drives Intel AX210-family adapters (the reference part is the AX211). 
-The driver scans, joins open and WPA2/WPA3-PSK networks, and presents itself to 
-the network stack as an ordinary Ethernet interface, so `dhcp`, `ping`, `nslookup` 
-and anything speaking sockets work over Wi-Fi exactly as they do over a cable.
-
-```
-wifi scan                        list nearby networks
-wifi connect <ssid> <passphrase> join one
-dhcp                             pick up an address
-wifi status                      what you are connected to
-wifi saved / wifi forget <ssid>  networks remembered for next time
-```
-
-There is a graphical path too: a Wi-Fi entry in the desktop panel and a
-Wi-Fi tab in the Network app. See "The desktop side" below.
-
-## Layout
-
-```
-kernel/src/Drivers/Net/Wifi/
-  IwxTrans.cpp    PCIe transport: MMIO, MSI-X, DMA rings, firmware boot,
-                  host commands, RX processing, frame TX, key installation
-  IwxFw.cpp       .ucode / .pnvm TLV parsing
-  IwxMvm.cpp      post-ALIVE init, NVM, UMAC scan, RX dispatch
-  IwxConnect.cpp  MLME: contexts, authenticate, associate, 802.11 <-> Ethernet
-  Wpa.cpp         WPA2/WPA3-PSK supplicant (EAPOL 4-way + group rekey)
-  Ieee80211.hpp   frame, element and RSN constants
-  Wifi.cpp        subsystem facade: probe, scan table, syscalls, netif hooks
-kernel/src/Libraries/Crypto.cpp   SHA-1/SHA-256, HMAC, PBKDF2, AES, CMAC
-kernel/src/Net/NetIf.cpp          interface registry the Ethernet layer uses
-```
-
-## Joining a network
-
-`SYS_WIFI_CONNECT` blocks until the link is up or the attempt fails, and
-returns a `WIFI_ERR_*` code the `wifi` tool turns into a specific message
-(wrong passphrase, unsupported security, AP out of range, and so on).
-
-The sequence:
-
-1. **Look up the BSS.** The SSID is matched against the scan table, strongest
-   signal first. If it is not there, one scan is run automatically and the
-   lookup retried, so `wifi connect` works without scanning first.
-2. **Negotiate ciphers.** The AP's RSN element (kept verbatim in the scan
-   table) picks the pairwise cipher and AKM. CCMP is preferred over GCMP,
-   plain PSK over PSK-SHA256.
-3. **Derive the PMK.** PBKDF2-HMAC-SHA1 over the passphrase with the SSID as
-   salt, 4096 iterations. A 64-character hex string is taken as a raw PSK
-   instead.
-4. **Bring up firmware contexts.** PHY, MAC, binding and station, then one TX
-   queue on the management TID.
-5. **Authenticate and associate.** Open-system authentication, then an
-   association request carrying the SSID, supported rates and, for encrypted
-   networks, the RSN element the supplicant built. Both are retransmitted up
-   to four times at 400 ms.
-6. **Run the 4-way handshake.** EAPOL-Key messages 1-4, then the pairwise key
-   and the group key go into the firmware with `ADD_STA_KEY`.
-7. **Report the link up.** Only now does `NetIf` see `wlan0` as usable.
-
-## The data path
-
-Once associated the driver translates between 802.11 and Ethernet II:
-
-- **TX** - an Ethernet frame becomes a to-DS 802.11 data header plus an
-  RFC 1042 LLC/SNAP shim carrying the EtherType. The protected bit is set
-  once keys are installed and the firmware does the CCMP encryption.
-- **RX** - the firmware decrypts and strips the MIC but leaves the 8-byte
-  CCMP header, which is skipped; the LLC/SNAP shim is replaced by an Ethernet
-  header built from addresses 1 and 3. EAPOL frames are diverted to the
-  supplicant instead.
-
-One TX queue carries management frames, EAPOL and non-QoS data. The firmware
-maps non-QoS data onto the management TID anyway, and the driver never
-negotiates block-ack sessions that would need per-TID queues.
-
-### Threading
-
-The RX path runs under `IwxProcessEvents()`'s reentrancy guard and **must not
-send a host command** - the command's completion is pumped by the very
-function it would be re-entering. Anything needing a command (post-association
-context updates, key installation, and therefore the whole EAPOL handshake) is
-queued and applied from `IwxConnectService()`, which every idling core calls
-and which is itself serialized. Transmitting is safe from either context: it
-only writes a descriptor and rings the doorbell.
-
-### Never wait on the clock under a spinlock
-
-`kcp::Spinlock::Acquire()` does `cli`, and `Timekeeping::GetMilliseconds()` is
-driven by the APIC timer interrupt. A wall-clock timeout inside a spinlock
-therefore cannot expire: the counter never advances, the loop never ends, and
-the machine locks solid with interrupts off - no mouse, no keyboard, no
-scheduler. This is the same trap `ApicTimer.cpp` documents for the idle path.
-
-`IwxSendCmd` holds a lock across a wait for the firmware's reply, so that lock
-is a `kcp::Mutex` (which keeps interrupts enabled), and the wait carries a spin
-cap as well as the clock check so it terminates even if the clock is somehow
-stuck. It also bails immediately once the firmware is known to have asserted,
-because nothing after that will ever be answered.
-
-A firmware command that goes unanswered now dumps the firmware's own error
-table on the first timeout - that names the command that asserted - and gives
-up on the adapter after three, rather than stalling for seconds per command.
-
-### Sample the clock after the work, not before it
-
-`IwxConnectService()` sends host commands, and every one of them is a round
-trip to the adapter that takes real milliseconds. Each step also stamps the
-timestamp its deadline is measured from - `EnterState()` sets
-`g_stateEnteredMs`, transmitting sets `g_lastTxMs`. A `now` read at the top of
-the pass is therefore *older* than the stamps it is about to be compared
-against, and because these are unsigned counters, `now - g_stateEnteredMs`
-wraps to about 2^64 instead of going negative. Every deadline in the pass then
-reads as long expired.
-
-The symptom was a join that failed the instant it succeeded:
-
-```
-WiFi: [OK] Associated, AID 3
-WiFi: [WARNING] Connection failed: timed out while joining the network
-WiFi: [INFO] EAPOL frame received (99 bytes)
-```
-
-The two post-association context commands moved the clock forward, the timeout
-check compared a stale `now` against the `g_stateEnteredMs` they had just set,
-and the contexts came down before the access point's first EAPOL frame could
-arrive - which is why message 1 shows up *after* the failure. So the clock is
-read only once the command-sending work in the pass is done, and the
-comparisons go through `Elapsed()`, which refuses to underflow.
-
-### Teardown unwinds contexts in the order they depend on each other
-
-Firmware 89 asserts when a context is taken away while another still points at
-it, and the assert names the command rather than the reason. Three of these
-have been hit so far:
-
-| UMAC error | Cause |
-|---|---|
-| 0x2010330F | PHY binding and link activation folded into one LINK_CONFIG |
-| 0x2010330E | link removed while its station still existed |
-| 0x2000320F | link deactivated while the MAC was still marked associated |
-
-The last one is the teardown side of the same rule. While
-`MAC_CONFIG.is_assoc` is set, the firmware's MAC context owns the link carrying
-the BSS, so `TearDown()` sends a `MAC_CONFIG` MODIFY clearing `is_assoc` first,
-and only then removes the station, deactivates the link, removes the link,
-removes the MAC and drops the PHY context - the order
-`iwl_mvm_mld_vif_cfg_changed_station` and the paths below it use on the way
-down. `tests/wifi/ap_mlme.py` pins that order.
-
-### A bad access point must not become a bad computer
-
-Three separate defects turned "the Wi-Fi connection went wrong" into "the whole
-machine went wrong". They are worth keeping straight because they have nothing
-to do with each other beyond sharing a trigger.
-
-**The command wait is a busy spin on a core the scheduler has been told not to
-touch.** `ServiceEvents()` runs from `ApicTimer::ServiceDeferredWork()`, which
-sets `cpu->reservedForKernelWork` so a bottom half holding a process-context
-mutex cannot be preempted into the process that would wait on it. The
-scheduler honours that by refusing to place any process on that core and by
-skipping it for the reschedule IPI, and on the BSP `RunBspMaintenance()` - the
-thing that wakes sleeping processes - is not reached until the pass returns.
-`IwxSendCmd` then waits for the firmware by spinning on `IwxDelayUs`, up to a
-second. One pass of `TearDown()` is eight commands. Against an adapter that has
-stopped answering, that pinned a core for the better part of ten seconds. The
-symptom is not a Wi-Fi symptom at all: the cursor crawls, windows stop
-repainting, everything stutters.
-
-So a service pass now carries a budget - `IwxBeginServicePass()` /
-`IwxEndServicePass()` - of roughly one command's worth of waiting in total.
-Commands that do not fit are not sent at all (an abandoned command still holds
-its ring slot, and a late answer would be misread as the *next* command's
-completion) and are retried on the next trip round the idle loop. The bring-up
-path does not bracket itself: it reserves its CPU deliberately and has nothing
-to starve, so it keeps the full per-command timeout.
-
-**Counting consecutive failures never fires on the failure that matters.** The
-give-up rule was three unanswered commands in a row, with any success resetting
-the count. An adapter that answers some commands and drops others - which is
-exactly what a marginal link leaves the firmware doing - therefore never
-reached the cutoff, while every drop still cost a full timeout. The stall was
-not a one-off; it repeated indefinitely, and the driver never concluded
-anything was wrong. It is a leaky bucket now: a failure adds one, a success
-drains one, so a firmware failing even a fraction of its commands trips the
-cutoff in bounded time.
-
-**`IwxFwState::Error` was a one-way door.** Nothing anywhere cleared it. A
-single firmware assert left `StartJoin()` returning `WIFI_ERR_NO_ADAPTER` -
-which the user sees as *no adapter is ready* - for the rest of the uptime, and
-the only way back was a reboot. Disconnecting and reconnecting could not help:
-there was nothing to reconnect with. Since `IwxReadFirmware()` is idempotent
-and the parsed image stays resident, the cure is to stop the device and run the
-same bring-up again, which `ServiceRecovery()` now does. It is rate-limited to
-one attempt per five seconds and capped at three, because the reset has its own
-multi-second handshakes and an adapter that will not come back after three
-tries is genuinely broken - repeating it forever would be its own kind of
-stall.
-
-### The link is up only while the access point says so
-
-`IwxLinkUp()` was nothing but `g_state == Connected`, and nothing moved it off
-that state unless the access point was polite enough to send a
-deauthentication frame. An access point that simply stops being there - a phone
-hotspot that sleeps, wanders off channel, or drops the station without saying
-so - left the link reported as up indefinitely. `NetIf::Active()` went on
-choosing `wlan0`, every packet went into the void, and the desktop showed a
-healthy connection while nothing resolved and nothing connected.
-
-Beacon loss cannot be used to notice this: once associated, `MacConfigCmd`
-stops asking for beacons and the firmware tracks them itself. What is
-observable is that our own frames stop being acknowledged, since `IwxTxComplete`
-gets a status per frame. A long enough run of failures - with any
-acknowledgement, or any frame received from the BSS, restarting the count -
-means the access point is gone, and the link comes down so the stack can fall
-back to a cable and the panel can report the truth.
-
-The threshold is counted in frames rather than seconds on purpose: a link that
-is merely idle has nothing to send and must not be torn down for it. And
-`IwxTxComplete` runs inside the RX pump, so it only sets a flag; the teardown
-happens in `IwxConnectService()`, where sending the commands it needs is
-allowed. `tests/wifi/ap_mlme.py` pins both halves of that.
-
-## The desktop side
-
-### Nothing in the GUI may block
-
-`SYS_WIFI_SCAN` sweeps for up to twenty seconds and `SYS_WIFI_CONNECT` waits
-out a whole handshake. Either one called from `desktop.elf` would freeze the
-compositor - and with it the mouse, the panel and every window - for seconds at
-a time. So the same work has a second, non-blocking entry point:
-
-| | |
-|---|---|
-| `SYS_WIFI_SCAN_START` | starts a sweep and returns immediately |
-| `SYS_WIFI_RESULTS` | copies the scan table out without touching the radio |
-| `SYS_WIFI_CONNECT_ASYNC` | starts a join and returns immediately |
-| `SYS_NETIFS` | lists the registered interfaces (see below) |
-
-`WifiInfo` grew the fields that make polling enough to follow along:
-`scanning` and `scanGeneration` for the sweep, `joining` and `connState` for
-the handshake, and `lastError` for how the last join ended. The GUI reads
-`SYS_WIFI_INFO` every 400 ms while something is in flight and every three
-seconds otherwise.
-
-The deadlines belong to the kernel, not the caller: `ServiceAsync()` runs from
-`ServiceEvents()` - after the RX pump has returned, so it is allowed to send
-commands - and aborts a scan that overruns, tears down a join that stalls, and
-records `lastError` when one fails. A failed join is unwound the moment it
-fails, so `connState` is back to idle by the time anyone looks; `lastError` is
-what survives to be reported.
-
-### Wired and wireless are separate on the panel
-
-The IP configuration is global to the stack, so "which interface does this
-address belong to?" is not a question `SYS_GETNETCFG` can answer. `SYS_NETIFS`
-reports each registered interface with its name, MAC, link state and an
-`active` flag - the one `NetIf::Active()` currently sends through. The Ethernet
-popup shows the address only while the wired interface is the active one, and
-says "Not in use" when the cable is up but Wi-Fi is carrying the traffic; the
-Wi-Fi popup does the mirror image. Each icon appears only when its hardware
-does: no wired interface, no Ethernet icon; no adapter, no Wi-Fi icon.
-
-The Wi-Fi icon stays white whatever the radio is doing. State belongs in the
-popup, and an icon that changes colour next to the clock is just noise.
-
-Picking a network that needs a passphrase opens a real window - created with
-`desktop_create_window()` and the four callbacks, exactly like the reboot and
-shutdown dialogs in `dialogs.cpp` - rather than something painted into the
-panel overlay. It therefore has a title bar, can be dragged and closed, appears
-in the window list, and gets its text field, checkboxes and buttons from the
-same `mtk` widgets the settings apps use.
-
-### What runs at startup
-
-`desktop.elf` keeps looking for an adapter until one appears (firmware loads
-well after login), starts one scan as soon as the firmware reports ready, and
-when the results land joins the strongest saved network that is in range. Once
-the link is up, and only if no address is configured, it spawns `dhcp.elf`.
-
-### Saved networks
-
-`0:/config/wifi.toml` holds them:
-
-```toml
-[wifi]
-autoconnect = true
-
-[network.0]
-ssid = "Home"
-psk = "passphrase"
-```
-
-**No default copy of this file ships in the image.** It is created on the first
-save, the way `bluetooth.toml`, `display.toml` and `session.toml` are. A shipped
-default looks harmless - it only documents the schema - but it is laid down
-again by anything that refreshes the system files, and it takes the user's saved
-networks with it when it lands. That is exactly what happened the first time
-this was written, and it is why the schema is documented here instead.
-
-`programs/include/montauk/wifi.h` is the one implementation of reading,
-writing and searching that file, shared by the panel, the Network app and the
-`wifi` command, so all three agree on the schema. The passphrase is stored as
-typed because that is what the join needs - the kernel derives the PMK from it,
-or takes a 64-character hex string as a raw PSK. There is no key store to hide
-it in: anyone who can read `0:/config` can read the passphrases.
-
-## The interface registry
-
-`Net::NetIf` replaced the Ethernet layer's direct calls into the E1000
-drivers. Drivers register a name, a kind, and three function pointers; the
-stack sends through `NetIf::Active()`, which prefers a wired interface with a
-link and otherwise takes the first interface reporting one. A Wi-Fi-only
-machine therefore has no link until it joins a network, and a machine with a
-cable plugged in keeps using it.
-
-`SYS_NETSTATUS` reports whichever interface is active, so `ifconfig` shows the
-wireless counters once Wi-Fi is carrying traffic.
-
-## What is supported
-
-| | |
-|---|---|
-| Open networks | yes |
-| WPA2-PSK, CCMP or GCMP | yes |
-| WPA2-PSK-SHA256 | yes |
-| WPA3 transition mode (PSK advertised alongside SAE) | yes, joins via PSK |
-| Group key rekeying | yes |
-| WPA3-only (SAE) | no |
-| Management frame protection required (MFPR) | no |
-| WEP, original WPA / TKIP | no |
-| 802.1X enterprise (EAP) | no |
-| Block-ack aggregation, HT/VHT/HE rates | no - legacy rates only |
-
-SAE needs finite-field or elliptic-curve arithmetic that does not belong in
-this kernel, and MFP needs BIP. Both are rejected up front with a specific
-log line rather than failing partway through a handshake. Mixed WPA/WPA2
-networks that still broadcast under TKIP are refused for the same reason: the
-pairwise key would install but every broadcast frame would be dropped, which
-looks like a working connection that cannot get a DHCP lease.
-
-## Crypto
-
-`kernel/src/Libraries/Crypto.cpp` exists because the supplicant runs in the
-kernel and BearSSL is a userspace library. It provides SHA-1, SHA-256, HMAC
-over both, PBKDF2-HMAC-SHA1, AES-128/256, RFC 3394 key wrap/unwrap and
-AES-CMAC. It is not a general-purpose crypto library and should not be used
-as one.
-
-## Testing
-
-`./tests/wifi/run.sh` compiles the shipping sources for the host against a
-small shim and drives them from Python. It is the real `Crypto.cpp`,
-`Wpa.cpp` and `IwxConnect.cpp`, not a copy, with only the transport stubbed.
-
-- **Crypto primitives** against the published vectors - FIPS-197 for AES,
-  RFC 2202/4231 for HMAC, RFC 3394 for key wrap, RFC 4493 for CMAC, and the
-  IEEE 802.11i Annex H.4 WPA passphrase vectors for PBKDF2.
-- **The supplicant** against an independent authenticator using `hashlib` and
-  `cryptography`: messages 2 and 4 carry MICs that verify under a PTK the AP
-  derived itself, the installed TK and GTK match the AP's, a wrong passphrase
-  produces a MIC the AP rejects, group rekeys and message-3 retransmissions
-  are answered, and RSN negotiation picks the right suites across eight
-  real-world information elements.
-- **The MLME and data path** against a simulated AP that decodes every frame
-  the driver emits: the authentication request, the association request and
-  its elements (SSID, rates, capabilities, RSN), the handshake carried inside
-  real 802.11 data frames, key installation arguments, and the encapsulation
-  both ways - to-DS addressing, the protected bit, LLC/SNAP, sequence numbers,
-  broadcast delivery, and the filtering of foreign-BSSID and null-data frames.
-  Also the branches: open networks, retransmission and give-up when the AP is
-  silent, authentication and association rejections, and an AP-initiated
-  deauthentication bringing the link down.
-- **Firmware context ordering** - the MLD command sizes and field offsets
-  against the decoded Linux trace, and the order the teardown unwinds the
-  contexts in, which is what the asserts above are about.
-
-The harness clock advances on every host command
-(`CMD_ROUND_TRIP_MS` in `mlme_harness.cpp`) rather than standing still. That
-detail matters: a frozen clock makes every elapsed-time comparison in the
-service loop trivially true or trivially false, and hid the underflow described
-under "Sample the clock after the work, not before it" - the host tests passed
-while the adapter could not join a network at all. Anything that reads
-`Timekeeping::GetMilliseconds()` should be tested with time actually moving.
-
-What is left needs the adapter, because it is the firmware's opinion rather
-than the driver's logic: whether the firmware accepts the TX command and TFD
-layout and actually radiates the frames, whether `ADD_STA_KEY` installs the
-keys the driver asks for, whether the RX MPDU descriptor is read correctly off
-real receptions, and whether association succeeds against a real AP's timing
-and rate expectations. None of that can be exercised in QEMU, which has no
-AX210-family device to emulate.

kernel/GNUmakefile

@@ -27,6 +27,20 @@ ifeq ($(ARCH),x86_64)
     endif
 endif
 
+# The other architectures have no in-tree cross GCC: default to LLVM, which
+# cross-compiles to every target (linking needs ld.lld; see docs/porting.md).
+ifneq ($(ARCH),x86_64)
+    ifeq ($(filter command line environment,$(origin CC)),)
+        CC := clang
+    endif
+    ifeq ($(filter command line environment,$(origin CXX)),)
+        CXX := clang++
+    endif
+    ifeq ($(filter command line environment,$(origin AR)),)
+        AR := ar
+    endif
+endif
+
 # User controllable C compiler command.
 CC  ?= cc
 
@@ -139,8 +153,10 @@ ifeq ($(ARCH),aarch64)
         override CXX += \
             -target aarch64-unknown-none
     endif
+    # Plain char is signed on x86; the generic kernel is written for that.
     override SHARED_FLAGS += \
-        -mgeneral-regs-only
+        -mgeneral-regs-only \
+        -fsigned-char
     override LDFLAGS += \
         -Wl,-m,aarch64elf
 endif
@@ -197,8 +213,10 @@ override CXXFLAGS += \
     -std=gnu++20 \
     -fno-rtti \
     -fno-exceptions \
-    -fmodules-ts \
     $(SHARED_FLAGS)
+ifneq ($(CXX_IS_CLANG),1)
+    override CXXFLAGS += -fmodules-ts
+endif
 
 # Internal linker flags that should not be changed by the user.
 override LDFLAGS += \
@@ -212,6 +230,9 @@ override LDFLAGS += \
 # Use "find" to glob all *.c, *.cpp, *.S, and *.asm files in the tree and obtain the
 # object and header dependency file names.
 override SRCFILES := $(shell cd src && find -L * -type f | LC_ALL=C sort)
+# Architecture-specific code lives in src/Arch/<arch>/; only the target's
+# directory is built. Everything else in src/ is architecture-neutral.
+override SRCFILES := $(filter-out Arch/%,$(SRCFILES)) $(filter Arch/$(ARCH)/%,$(SRCFILES))
 override CFILES := $(filter %.c,$(SRCFILES))
 override CXXFILES := $(filter %.cpp,$(SRCFILES))
 override OBJCFILES := $(filter %.mm,$(SRCFILES))
@@ -253,6 +274,11 @@ $(BUILDNO_HDR): $(filter-out $(BUILDNO_OBJ),$(OBJ)) $(PROG_SRC)
 # on a clean build (where no .d dependency exists yet).
 $(BUILDNO_OBJ): $(BUILDNO_HDR)
 
+# Compile everything without linking (used by scripts/check-arch-neutral.sh
+# to verify a port's objects before a cross linker is available).
+.PHONY: objects
+objects: $(OBJ) cc-runtime-$(ARCH)/cc-runtime.a
+
 # Link rules for building the C compiler runtime.
 cc-runtime-$(ARCH)/cc-runtime.a: GNUmakefile cc-runtime/*
 	rm -rf cc-runtime-$(ARCH)
@@ -263,10 +289,35 @@ cc-runtime-$(ARCH)/cc-runtime.a: GNUmakefile cc-runtime/*
 		CFLAGS="$(CFLAGS)" \
 		CPPFLAGS='-isystem ../freestnd-c-hdrs/$(ARCH)/include -DCC_RUNTIME_NO_FLOAT'
 
+# Clang's generic-ELF driver links cross targets through the host's GCC,
+# so a clang-built non-x86 kernel is linked by ld.lld directly.
+ifneq ($(ARCH),x86_64)
+    ifeq ($(CXX_IS_CLANG),1)
+        # Without a system lld, fall back to the one rustup ships.
+        KERNEL_LD ?= $(firstword $(shell command -v ld.lld) \
+            $(wildcard $(HOME)/.rustup/toolchains/*/lib/rustlib/*/bin/gcc-ld/ld.lld))
+        override LD_EMULATION_aarch64 := aarch64elf
+        override LD_EMULATION_riscv64 := elf64lriscv
+        override LD_EMULATION_loongarch64 := elf64loongarch
+        override KERNEL_LDFLAGS := \
+            -m $(LD_EMULATION_$(ARCH)) \
+            --build-id=none \
+            --no-relax \
+            -static \
+            -z max-page-size=0x1000 \
+            --gc-sections \
+            -T linker-$(ARCH).ld
+    endif
+endif
+
 # Link rules for the final executable.
 bin-$(ARCH)/$(OUTPUT): GNUmakefile linker-$(ARCH).ld $(OBJ) cc-runtime-$(ARCH)/cc-runtime.a
 	mkdir -p "$$(dirname $@)"
+ifeq ($(KERNEL_LD),)
 	$(CXX) $(CXXFLAGS) $(LDFLAGS) $(OBJ) cc-runtime-$(ARCH)/cc-runtime.a -o $@
+else
+	$(KERNEL_LD) $(KERNEL_LDFLAGS) $(OBJ) cc-runtime-$(ARCH)/cc-runtime.a -o $@
+endif
 
 # Compilation rules for *.c files.
 obj-$(ARCH)/%.c.o: src/%.c GNUmakefile

kernel/linker-aarch64.ld

@@ -22,6 +22,7 @@ SECTIONS
     /* Any address in this region will do, but often 0xffffffff80000000 is chosen as */
     /* that is the beginning of the region. */
     . = 0xffffffff80000000;
+    KernelStartSymbol = .;
 
     /* Define a section to contain the Limine requests and assign it to its own PHDR */
     .limine_requests : {
@@ -74,4 +75,6 @@ SECTIONS
         *(.eh_frame*)
         *(.note .note.*)
     }
+
+    KernelEndSymbol = .;
 }

kernel/src/ACPI/AML/AmlInterpreter.hpp

@@ -7,6 +7,7 @@
 
 #pragma once
 #include "AmlNamespace.hpp"
+#include <CppLib/Spinlock.hpp>
 #include <cstdint>
 
 namespace Hal {
@@ -96,6 +97,35 @@ namespace Hal {
         static constexpr uint8_t  ConcatResOp   = 0x84;
         static constexpr uint8_t  ToHexStringOp = 0x98;
         static constexpr uint8_t  ToDecimalStringOp = 0x97;
+        static constexpr uint8_t  ExternalOp    = 0x15;
+        static constexpr uint8_t  RefOfOp       = 0x71;
+        static constexpr uint8_t  FindSetLeftBitOp  = 0x81;
+        static constexpr uint8_t  FindSetRightBitOp = 0x82;
+        static constexpr uint8_t  NotifyOp      = 0x86;
+        static constexpr uint8_t  MatchOp       = 0x89;
+        static constexpr uint8_t  ObjectTypeOp  = 0x8E;
+        static constexpr uint8_t  CreateQWordFieldOp = 0x8F;
+        static constexpr uint8_t  ToStringOp    = 0x9C;
+        static constexpr uint8_t  CopyObjectOp  = 0x9D;
+        static constexpr uint8_t  MidOp         = 0x9E;
+        static constexpr uint8_t  ContinueOp    = 0x9F;
+        static constexpr uint8_t  BreakPointOp  = 0xCC;
+
+        // More extended opcodes (after ExtOpPrefix)
+        static constexpr uint8_t  CondRefOfOp   = 0x12;
+        static constexpr uint8_t  CreateFieldOp = 0x13;
+        static constexpr uint8_t  LoadTableOp   = 0x1F;
+        static constexpr uint8_t  LoadOp        = 0x20;
+        static constexpr uint8_t  SignalOp      = 0x24;
+        static constexpr uint8_t  WaitOp        = 0x25;
+        static constexpr uint8_t  ResetOp       = 0x26;
+        static constexpr uint8_t  FromBCDOp     = 0x28;
+        static constexpr uint8_t  ToBCDOp       = 0x29;
+        static constexpr uint8_t  UnloadOp      = 0x2A;
+        static constexpr uint8_t  DebugOp       = 0x31;
+        static constexpr uint8_t  FatalOp       = 0x32;
+        static constexpr uint8_t  TimerOp       = 0x33;
+        static constexpr uint8_t  DataRegionOp  = 0x88;
 
         // ============================================================================
 
@@ -103,7 +133,29 @@ namespace Hal {
 
         // ============================================================================
         static constexpr int MaxCallDepth       = 16;
-        static constexpr int MaxLoopIterations  = 1024;
+
+        // ACPICA's ceiling. Firmware polls hardware in While loops with a
+        // Sleep or Stall in the body (link training, power-good, L2/L3 ready)
+        // and expects more than a token number of passes.
+        static constexpr int MaxLoopIterations  = 0xFFFF;
+
+        // Backstop cap on evaluation nesting. This counter is shared by
+        // expression evaluation AND block execution, and block nesting is
+        // legitimately deep -- every If/Else/While body is another level, so a
+        // long firmware method reaches well past a handful. The real protection
+        // against running off the stack is HaveStackHeadroom(), which measures
+        // actual usage; this only catches a decode that recurses without
+        // consuming input.
+        static constexpr int MaxExprDepth       = 48;
+
+        // How much stack the interpreter may consume before it gives up. Note
+        // this is an ALLOWANCE, not a reserve: frames here are expensive (a
+        // name-resolving EvalTerm frame carries a 256-byte path buffer plus
+        // 272-byte Object temporaries, so ~800 bytes each), and a real
+        // firmware _BIF nests far enough to need tens of them. Sized against
+        // Sched::StackPages -- keep well under it, since the syscall path below
+        // has already spent some of that stack.
+        static constexpr uint64_t StackBudget   = 32768;
 
         // ============================================================================
 
@@ -114,8 +166,10 @@ namespace Hal {
         public:
             Interpreter();
 
-            // Parse a DSDT or SSDT table into the namespace.
+            // Parse a DSDT or SSDT table into the namespace. Tables may be
-            // tableData points to the CommonSDTHeader (HHDM-mapped).
+            // loaded one after another; each method remembers which table
+            // its body lives in. tableData points to the CommonSDTHeader
+            // (HHDM-mapped) and must stay mapped.
             bool LoadTable(void* tableData);
 
             // Evaluate a named object, returning its value.
@@ -125,6 +179,28 @@ namespace Hal {
             // Evaluate a method with arguments.
             bool EvaluateMethod(const char* path, const Object* args, int argCount, Object& result);
 
+            // Evaluate a node: a method is run with `args`, a field is read,
+            // anything else hands back its value (non-data objects as a
+            // Reference to the node).
+            bool EvaluateNode(int32_t node, const Object* args, int argCount, Object& result);
+
+            // Find the child of `parent` named `seg` ("_PS3", "_ON"), -1 if
+            // there is none. Short names are padded with '_'.
+            int32_t FindChild(int32_t parent, const char* seg) const;
+
+            // EvaluateNode() on a named child. False when it does not exist.
+            bool EvaluateChild(int32_t parent, const char* seg, const Object* args,
+                               int argCount, Object& result);
+
+            // Turn a Reference, or a NameRef/String naming an object relative
+            // to `scopeNode`, into a node index. -1 when it names nothing.
+            int32_t ResolveReference(const Object& obj, int32_t scopeNode) const;
+
+            // The PCI function a Device node describes: its _ADR on the bus
+            // behind its parent (a root bridge's _BBN, or a PCI-to-PCI
+            // bridge's secondary bus). False when it cannot be placed.
+            bool GetPciAddress(int32_t deviceNode, uint8_t& bus, uint8_t& dev, uint8_t& fn);
+
             // Read a field value. Returns integer.
             bool ReadField(int32_t nodeIndex, uint64_t& value);
 
@@ -138,18 +214,67 @@ namespace Hal {
             // Check if the interpreter has been initialized
             bool IsInitialized() const { return m_initialized; }
 
+            // Nodes [0, DsdtNodeCount()) came from the DSDT. Nodes past it were
+            // added by SSDTs (or by methods run since).
+            int32_t DsdtNodeCount() const { return m_dsdtNodeCount; }
+
+            // Log every opcode executed, with its offset, for the next
+            // `budget` opcodes. The only way to see where a method that runs to
+            // completion without returning a value actually goes.
+            void SetTrace(int budget) { m_traceBudget = budget; }
+
+            // ============================================================================
+
+            // Packages
+
+            // ============================================================================
+            // Read one element of a Package object. Works for both a package
+            // whose elements were computed at run time (materialized in the
+            // package store) and one that is still just the constant AML bytes
+            // it was declared with. A name in a constant package comes back as
+            // a NameRef unless it resolves (from the root) to plain data, in
+            // which case its value is returned. Returns false when the object
+            // is not a package or the index is out of range.
+            bool GetPackageElement(const Object& package, int index, Object& result);
+
+            // Number of elements a Package declares.
+            int GetPackageCount(const Object& package);
+
+            // Give a package stored in a namespace node real per-element
+            // storage so Index() can be written through. Returns the store
+            // index, or -1 if the node does not hold a package.
+            int32_t MaterializeNodePackage(int32_t nodeIndex);
+
         private:
+            struct ExecContext;
+
             // ============================================================================
             // Parsing (table load)
             // ============================================================================
+            // ctx is the method being executed when a definition appears in
+            // a method body, nullptr at table load. Operands that are TermArgs
+            // (region offsets, buffer sizes, bank values) are evaluated in it.
             bool ParseBlock(const uint8_t* aml, uint32_t offset, uint32_t endOffset,
                             int32_t scopeNode);
 
             bool ParseNamedObject(const uint8_t* aml, uint32_t* pos, uint32_t endOffset,
-                                  int32_t scopeNode);
+                                  int32_t scopeNode, ExecContext* ctx = nullptr);
 
             bool ParseExtendedOp(const uint8_t* aml, uint32_t* pos, uint32_t endOffset,
-                                 int32_t scopeNode);
+                                 int32_t scopeNode, ExecContext* ctx = nullptr);
+
+            // The FieldList shared by Field, IndexField and BankField.
+            void ParseFieldList(const uint8_t* aml, uint32_t* pos, uint32_t fieldEnd,
+                                int32_t scopeNode, uint8_t fieldFlags, const Object& proto);
+
+            // Evaluate a TermArg operand of a definition: in the running
+            // method, or in the table-load context.
+            bool EvalOperand(ExecContext* ctx, const uint8_t* aml, uint32_t* pos,
+                             uint32_t endOffset, int32_t scopeNode, Object& result);
+
+            // CreateBitField .. CreateQWordField and CreateField.
+            bool CreateBufferField(ExecContext& ctx, uint32_t* pos, uint32_t endOffset,
+                                   uint8_t op);
 
             // ============================================================================
 
@@ -158,8 +283,21 @@ namespace Hal {
             // ============================================================================
             // Read a NameString from AML and produce an absolute path.
             // Advances *pos past the name.
+            // Produces the absolute path (rebased onto scopeNode for a
+            // relative name). outRelative, when given, receives the name as it
+            // actually appeared in the AML, and *outSearchable says whether the
+            // ACPI search rule applies to it -- true only when the name carried
+            // no root or parent prefix. Resolution must fall back to
+            // Namespace::SearchUpward() in that case: the absolute path is a
+            // guess at one scope, and the object usually lives in a parent.
             int ReadNameString(const uint8_t* aml, uint32_t* pos, int32_t scopeNode,
-                               char* outPath, int maxLen);
+                               char* outPath, int maxLen,
+                               char* outRelative = nullptr, int relLen = 0,
+                               bool* outSearchable = nullptr);
+
+            // Read a NameString and resolve it with the search rule. -1 when
+            // it names nothing.
+            int32_t ReadAndResolveName(const uint8_t* aml, uint32_t* pos, int32_t scopeNode);
 
             // Read a single 4-char NameSeg from AML. Advances *pos.
             void ReadNameSeg(const uint8_t* aml, uint32_t* pos, char* outSeg);
@@ -170,6 +308,12 @@ namespace Hal {
 
             // ============================================================================
             uint32_t DecodePkgLength(const uint8_t* aml, uint32_t* pos);
+
+            // Decode one constant package element out of raw AML bytes and
+            // advance *pos past it. Names come back as NameRef; locals and
+            // anything else that needs a running method decode to integer 0.
+            bool DecodeConstElement(const uint8_t* data, uint32_t length,
+                                    uint32_t* pos, Object& out) const;
             uint64_t DecodeInteger(const uint8_t* aml, uint32_t* pos);
 
             // ============================================================================
@@ -187,7 +331,28 @@ namespace Hal {
                 Object         ReturnValue;
                 bool           Returned;
                 bool           Broken;
-                int            Depth;
+                bool           Continued;
+                bool           Aborted;    // undecodable AML: unwind the method
+                int            Depth;      // method-call nesting
+                int            ExprDepth;  // expression/block nesting
+            };
+
+            // Increments a depth counter for the duration of a scope; pairs
+            // with the MaxExprDepth check at each recursive entry point.
+            struct DepthGuard {
+                int& Value;
+                explicit DepthGuard(int& v) : Value(v) { Value++; }
+                ~DepthGuard() { Value--; }
+            };
+
+            // Where a Store lands.
+            struct Target {
+                enum class Kind : uint8_t { None, Node, Local, Arg, PkgElement, BufByte };
+                Kind     What = Kind::None;
+                int32_t  Node = -1;
+                int      Index = 0;          // local/arg number, or element index
+                int32_t  PkgStore = -1;
+                Object*  Buf = nullptr;      // BufByte: the buffer being indexed
             };
 
             bool ExecuteBlock(ExecContext& ctx, uint32_t offset, uint32_t endOffset);
@@ -196,35 +361,140 @@ namespace Hal {
             // Evaluate a term (expression that produces a value) within an execution context.
             bool EvalTerm(ExecContext& ctx, uint32_t* pos, uint32_t endOffset, Object& result);
 
-            // Evaluate a "SuperName" target for Store operations.
+            // Extended (0x5B-prefixed) opcodes in expression or statement position.
-            // Returns the node index for named targets, or handles locals/args.
+            bool EvalExtOp(ExecContext& ctx, uint32_t* pos, uint32_t endOffset, Object& result);
-            // If isLocal/isArg is set, localIdx/argIdx contains the index.
-            bool EvalTarget(ExecContext& ctx, uint32_t* pos,
-                            int32_t& nodeIndex, bool& isLocal, int& localIdx,
-                            bool& isArg, int& argIdx);
 
-            // Store a value to a target (node, local, or arg).
+            // EvalTerm's opcode handlers. They live out of line so the
-            void StoreToTarget(ExecContext& ctx, const Object& value,
+            // recursive path only carries the locals of the one running.
-                               int32_t nodeIndex, bool isLocal, int localIdx,
+            bool EvalBufferOp(ExecContext& ctx, uint32_t* pos, uint32_t endOffset, Object& result);
-                               bool isArg, int argIdx);
+            bool EvalPackageOp(ExecContext& ctx, uint32_t* pos, uint32_t endOffset, Object& result);
+            bool EvalBinary(ExecContext& ctx, uint32_t* pos, uint32_t endOffset, Object& result);
+            bool EvalUnary(ExecContext& ctx, uint32_t* pos, uint32_t endOffset, Object& result);
+            bool EvalDivide(ExecContext& ctx, uint32_t* pos, uint32_t endOffset, Object& result);
+            bool EvalStoreOp(ExecContext& ctx, uint32_t* pos, uint32_t endOffset, Object& result);
+            bool EvalReference(ExecContext& ctx, uint32_t* pos, uint32_t endOffset, Object& result);
+            bool EvalIndexOp(ExecContext& ctx, uint32_t* pos, uint32_t endOffset, Object& result);
+            bool EvalMatchOp(ExecContext& ctx, uint32_t* pos, uint32_t endOffset, Object& result);
+            bool EvalMidOp(ExecContext& ctx, uint32_t* pos, uint32_t endOffset, Object& result);
+            bool EvalNameRef(ExecContext& ctx, uint32_t* pos, uint32_t endOffset, Object& result);
+            bool CallMethod(ExecContext& ctx, int32_t node, uint32_t* pos, uint32_t endOffset,
+                            Object& result);
+            bool IndexElement(const Object& source, uint64_t idx, Object& out);
+            void ReportUnresolved(ExecContext& ctx, uint32_t namePos);
+
+            // Target forms that need operand evaluation of their own.
+            bool EvalIndexTarget(ExecContext& ctx, uint32_t* pos, Target& target);
+            bool EvalDerefTarget(ExecContext& ctx, uint32_t* pos, Target& target);
+
+            // A package element: like EvalTerm, except that a name is a
+            // reference to the object rather than a call or a read.
+            bool EvalPackageElement(ExecContext& ctx, uint32_t* pos, uint32_t endOffset,
+                                    Object& result);
+
+            // Parse a SuperName/Target operand.
+            bool EvalTarget(ExecContext& ctx, uint32_t* pos, Target& target);
+
+            // Store a value to a target, with ACPI's implicit conversion for
+            // named data objects.
+            void StoreToTarget(ExecContext& ctx, const Object& value, const Target& target);
+
+            // Parse a Target operand and store `value` there.
+            void StoreResult(ExecContext& ctx, uint32_t* pos, const Object& value);
+
+            // The current value behind a target (for Increment and friends).
+            Object ReadTarget(ExecContext& ctx, const Target& target);
+
+            // Run a method node. Caller holds m_evalLock.
+            bool InvokeMethodNode(int32_t node, const Object* args, int argCount,
+                                  Object& result, int depth, int exprDepth);
+
+            // Run a method in a context already acquired and holding its
+            // arguments; releases the context.
+            bool RunMethodIn(ExecContext* ctx, int32_t node, int depth, int exprDepth,
+                             Object& result);
+
+            // The value of a node as an operand: fields read, methods run
+            // without arguments, non-data objects become References.
+            bool NodeValue(int32_t node, Object& result, int depth, int exprDepth);
+
+            // Log an opcode we cannot decode and stop the method: carrying on
+            // past it would execute operand bytes as code.
+            bool Abort(ExecContext& ctx, uint32_t pos, const char* why);
+
+            // Native methods (\_OSI).
+            bool RunNativeMethod(uint8_t which, const Object* args, int argCount, Object& result);
+
+            // Install the OS-defined objects (\_OSI, \_REV, \_OS_, \_GL_).
+            void InstallPredefined();
+
+            // Device identification (for PCI_Config regions).
+            bool IntegerChild(int32_t device, const char* seg, uint64_t fallback,
+                              uint64_t& value, int depth);
+            bool IsPciRootBridge(int32_t device, int depth);
+            bool ResolvePciAddress(int32_t node, uint8_t& bus, uint8_t& dev,
+                                   uint8_t& fn, int depth);
 
             // ============================================================================
 
             // Field I/O
 
             // ============================================================================
-            bool ReadRegion(RegionSpace space, uint64_t address, uint32_t bitWidth, uint64_t& value);
+            // Access `bits` (little-endian, bitLength long) of a field unit.
-            bool WriteRegion(RegionSpace space, uint64_t address, uint32_t bitWidth, uint64_t value);
+            bool AccessField(int32_t fieldNode, bool write, uint8_t* bits);
+            bool ReadFieldObject(int32_t fieldNode, Object& value);
+            bool WriteFieldObject(int32_t fieldNode, const Object& value);
+
+            // One aligned access unit of a field's backing store.
+            bool FieldUnitIo(const Object& field, uint64_t byteOffset, uint32_t width,
+                             bool write, uint64_t& value);
+
+            // Region access at byte `offset` into the region, `width` bytes.
+            bool ReadRegion(int32_t regionNode, uint64_t offset, uint32_t width, uint64_t& value);
+            bool WriteRegion(int32_t regionNode, uint64_t offset, uint32_t width, uint64_t value);
+            bool RegionIo(int32_t regionNode, uint64_t offset, uint32_t width,
+                          bool write, uint64_t& value);
+
+            bool ReadBufferField(const Object& field, Object& value);
+            bool WriteBufferField(const Object& field, const Object& value);
 
             // ============================================================================
 
             // State
 
             // ============================================================================
+            // ExecContext is ~4.4 KB (sixteen Objects, each carrying a
+            // 256-byte buffer inline). One per nested method call on a 16 KiB
+            // kernel stack overflows after three levels, so they come from a
+            // pool instead. The pool is only safe because evaluation is
+            // serialized -- see m_evalLock.
+            // Stack-exhaustion check; `frame` is the address of a local in
+            // the calling frame.
+            bool          HaveStackHeadroom(const void* frame) const;
+
+            // One-shot warning when the nesting backstop trips.
+            static void   ReportDepthLimit();
+
+            ExecContext*  AcquireContext();
+            void          ReleaseContext();
+
+            // Run `body` as a top-level evaluation: serialized, with the stack
+            // budget anchored here.
+            template<typename Fn> bool TopLevel(Fn body);
+
             Namespace     m_ns;
-            const uint8_t* m_dsdt;
-            uint32_t      m_dsdtLength;
             bool          m_initialized;
+            bool          m_predefined;
+            int32_t       m_dsdtNodeCount;
+
+            ExecContext*  m_ctxPool;      // MaxCallDepth + 1 entries, heap
+            int           m_ctxInUse;
+            int           m_evalNesting;  // top-level evaluations in progress
+            uint64_t      m_stackFloor;   // stack address the interpreter stops at
+            int           m_traceBudget;  // opcodes left to log, 0 = off
+            ExecContext*  m_loadCtx;      // table-load operand context
+            uint64_t      m_delayedUs;    // total Sleep/Stall time requested
+            int           m_methodScan;   // >0 while scanning a method body at load
+            kcp::Mutex    m_evalLock;
         };
 
         // ============================================================================

kernel/src/ACPI/AML/AmlNamespace.cpp

@@ -7,10 +7,118 @@
 #include "AmlNamespace.hpp"
 #include <Libraries/Memory.hpp>
 #include <Memory/Heap.hpp>
+#include <Terminal/Terminal.hpp>
+#include <CppLib/Stream.hpp>
+
+using namespace Kt;
 
 namespace Hal {
     namespace AML {
 
+        // ============================================================================
+
+        // Package element store
+
+        // ============================================================================
+        namespace PackageStore {
+
+            struct Entry {
+                Object* Elements;
+                int     Count;
+            };
+
+            static Entry s_persistent[MaxPersistent];
+            static Entry s_transient[MaxTransient];
+            static int   s_persistentUsed = 0;
+            static int   s_transientNext  = 0;
+
+            // Indices handed out are 1-based so that a zeroed Object (nodes are
+            // memset before Clear() runs, and callers zero-init freely) never
+            // looks like a reference to entry 0.
+            static constexpr int32_t TransientBase = MaxPersistent;
+
+            static Object* AllocElements(int count) {
+                auto* mem = (Object*)Memory::g_heap->Request(sizeof(Object) * count);
+                if (!mem) return nullptr;
+                for (int i = 0; i < count; i++)
+                    mem[i] = Object{};
+                return mem;
+            }
+
+            int32_t AllocPersistent(int count) {
+                if (count <= 0 || count > MaxPackageElements) return -1;
+                if (s_persistentUsed >= MaxPersistent) return -1;
+
+                Entry& e = s_persistent[s_persistentUsed];
+                e.Elements = AllocElements(count);
+                if (!e.Elements) return -1;
+                e.Count = count;
+                return (int32_t)(s_persistentUsed++) + 1;
+            }
+
+            int32_t AllocTransient(int count) {
+                if (count <= 0 || count > MaxPackageElements) return -1;
+
+                int slot = s_transientNext;
+                if (slot >= MaxTransient) {
+                    // The ring is only rewound between top-level evaluations, so
+                    // a method that builds more packages than this reuses the
+                    // oldest slots rather than failing outright.
+                    slot = 0;
+                    s_transientNext = 0;
+                }
+                s_transientNext++;
+
+                Entry& e = s_transient[slot];
+                // Element arrays are sized on first use and reused afterwards;
+                // a larger request reallocates.
+                if (e.Elements == nullptr || e.Count < count) {
+                    Object* mem = AllocElements(count);
+                    if (!mem) return -1;
+                    if (e.Elements) Memory::g_heap->Free(e.Elements);
+                    e.Elements = mem;
+                } else {
+                    for (int i = 0; i < count; i++)
+                        e.Elements[i] = Object{};
+                }
+                e.Count = count;
+                return TransientBase + slot + 1;
+            }
+
+            void ResetTransient() {
+                s_transientNext = 0;
+            }
+
+            bool IsPersistent(int32_t index) {
+                return index > 0 && index <= MaxPersistent;
+            }
+
+            static Entry* Lookup(int32_t index) {
+                if (index <= 0) return nullptr;
+                int32_t i = index - 1;
+
+                if (i < MaxPersistent) {
+                    if (i >= s_persistentUsed) return nullptr;
+                    return s_persistent[i].Elements ? &s_persistent[i] : nullptr;
+                }
+
+                i -= TransientBase;
+                if (i < 0 || i >= MaxTransient) return nullptr;
+                return s_transient[i].Elements ? &s_transient[i] : nullptr;
+            }
+
+            Object* Element(int32_t index, int element) {
+                Entry* e = Lookup(index);
+                if (!e || element < 0 || element >= e->Count) return nullptr;
+                return &e->Elements[element];
+            }
+
+            int Count(int32_t index) {
+                Entry* e = Lookup(index);
+                return e ? e->Count : 0;
+            }
+        };
+
         Namespace::Namespace() : m_chunkCount(0), m_nodeCount(0) {
             for (int i = 0; i < MaxChunks; i++)
                 m_chunks[i] = nullptr;
@@ -30,7 +138,17 @@ namespace Hal {
         }
 
         bool Namespace::AllocChunk() {
-            if (m_chunkCount >= MaxChunks) return false;
+            if (m_chunkCount >= MaxChunks) {
+                static bool warned = false;
+                if (!warned) {
+                    warned = true;
+                    KernelLogStream(WARNING, "AML")
+                        << "Namespace node pool full at " << base::dec
+                        << (uint64_t)(MaxChunks * NodesPerChunk)
+                        << " nodes; part of the namespace will be missing";
+                }
+                return false;
+            }
 
             auto* chunk = (NamespaceNode*)Memory::g_heap->Request(
                 sizeof(NamespaceNode) * NodesPerChunk);
@@ -45,6 +163,41 @@ namespace Hal {
             return true;
         }
 
+        bool Namespace::AppendChild(NamespaceNode* parent, int32_t child) {
+            if (parent->ChildCount >= parent->ChildCapacity) {
+                int32_t newCap = parent->ChildCapacity ? parent->ChildCapacity * 2
+                                                       : InitialChildren;
+
+                auto* grown = (int32_t*)Memory::g_heap->Request(sizeof(int32_t) * newCap);
+                if (!grown) {
+                    // Losing a child silently costs whole devices, so say it
+                    // once rather than leaving a hole nobody can see.
+                    static bool warned = false;
+                    if (!warned) {
+                        warned = true;
+                        KernelLogStream(WARNING, "AML")
+                            << "Out of heap growing a namespace node's child list;"
+                            << " part of the namespace will be missing";
+                    }
+                    return false;
+                }
+
+                for (int32_t i = 0; i < parent->ChildCount; i++)
+                    grown[i] = parent->ChildIndices[i];
+                for (int32_t i = parent->ChildCount; i < newCap; i++)
+                    grown[i] = -1;
+
+                if (parent->ChildIndices)
+                    Memory::g_heap->Free(parent->ChildIndices);
+
+                parent->ChildIndices = grown;
+                parent->ChildCapacity = newCap;
+            }
+
+            parent->ChildIndices[parent->ChildCount++] = child;
+            return true;
+        }
+
         int32_t Namespace::AllocNode() {
             // Check if we need a new chunk
             int32_t capacity = m_chunkCount * NodesPerChunk;
@@ -144,6 +297,12 @@ namespace Hal {
             return -1;
         }
 
+        int32_t Namespace::FindChild(int32_t parentIndex, const char* seg) const {
+            char padded[MaxNameSegLen + 1];
+            PadSegment(seg, padded);
+            return FindChildByName(parentIndex, padded);
+        }
+
         int32_t Namespace::CreateNode(const char* absolutePath) {
             EnsureRoot();
             char segments[MaxPathDepth][MaxNameSegLen + 1];
@@ -165,11 +324,7 @@ namespace Hal {
                     // Add to parent's children
                     auto* parent = GetNode(current);
                     if (!parent) return -1;
-                    if (parent->ChildCount < MaxChildren) {
+                    if (!AppendChild(parent, child)) return -1;
-                        parent->ChildIndices[parent->ChildCount++] = child;
-                    } else {
-                        return -1; // too many children
-                    }
                 }
                 current = child;
             }
@@ -188,6 +343,51 @@ namespace Hal {
             return current;
         }
 
+        int32_t Namespace::SearchUpward(const char* relative, int32_t scopeNodeIndex) const {
+            char segments[MaxPathDepth][MaxNameSegLen + 1];
+            int segCount = 0;
+
+            // Split on '.', padding each segment to four characters the way
+            // node names are stored.
+            int i = 0;
+            while (relative[i] && segCount < MaxPathDepth) {
+                char seg[MaxNameSegLen + 1];
+                int n = 0;
+                while (relative[i] && relative[i] != '.' && n < MaxNameSegLen)
+                    seg[n++] = relative[i++];
+                seg[n] = '\0';
+                while (relative[i] && relative[i] != '.') i++;   // overlong segment
+                if (relative[i] == '.') i++;
+
+                PadSegment(seg, segments[segCount]);
+                segCount++;
+            }
+
+            if (segCount == 0) return -1;
+
+            // Walk up from the current scope looking for the first segment.
+            int32_t scope = scopeNodeIndex;
+            while (scope >= 0) {
+                int32_t found = FindChildByName(scope, segments[0]);
+                if (found >= 0) {
+                    // Descend the rest. A scope that matches the first segment
+                    // but not the remainder does not win -- keep walking up.
+                    int32_t cur = found;
+                    bool ok = true;
+                    for (int k = 1; k < segCount; k++) {
+                        cur = FindChildByName(cur, segments[k]);
+                        if (cur < 0) { ok = false; break; }
+                    }
+                    if (ok) return cur;
+                }
+
+                auto* node = GetNode(scope);
+                scope = node ? node->ParentIndex : -1;
+            }
+
+            return -1;
+        }
+
         int32_t Namespace::ResolveName(const char* name, int32_t scopeNodeIndex) const {
             // If it starts with '\', it's absolute
             if (name[0] == '\\')

kernel/src/ACPI/AML/AmlNamespace.hpp

@@ -31,6 +31,16 @@ namespace Hal {
             ThermalZone,
             PowerResource,
             BufferField,
+            Event,
+            // A reference to a namespace node: what RefOf() yields, and what a
+            // name evaluates to when it names something that is not data (a
+            // device or power resource listed in a _PRx package, say).
+            Reference,
+            // A name inside a constant package, kept as written ("PG00",
+            // "\_SB.PCI0.PEG0") because its scope is only known to whoever
+            // reads the package. Interpreter::ResolveReference() turns it
+            // into a node.
+            NameRef,
         };
 
         // ============================================================================
@@ -55,10 +65,15 @@ namespace Hal {
         // ============================================================================
         static constexpr int MaxNameSegLen     = 4;
         static constexpr int MaxPathDepth      = 16;
-        static constexpr int MaxChildren       = 64;
+        // Children are held in a heap array that grows on demand, starting at
+        // this many entries. A fixed ceiling here is not survivable: an EC or
+        // GPE scope in a modern laptop DSDT carries well over a hundred named
+        // objects, and dropping the overflow loses whole devices (the battery
+        // among them) with no way to tell from the outside.
+        static constexpr int InitialChildren   = 8;
         static constexpr int MaxStringLen      = 64;
         static constexpr int MaxBufferLen      = 256;
-        static constexpr int MaxPackageElements = 16;
+        static constexpr int MaxPackageElements = 64;
         static constexpr int MaxMethodArgs     = 7;
         static constexpr int MaxMethodLocals   = 8;
 
@@ -76,6 +91,14 @@ namespace Hal {
         struct Object {
             ObjectType Type = ObjectType::None;
 
+            // Packages keep their raw AML bytes in Buffer (constant elements can
+            // be decoded straight from there), but a package whose elements are
+            // computed or written through Index() needs real per-element
+            // storage. PkgIndex names that storage once it is materialized;
+            // -1 means "not materialized yet". It lives outside the union so a
+            // package can carry both representations at once.
+            int32_t PkgIndex = -1;
+
             union {
                 uint64_t Integer;
 
@@ -89,10 +112,23 @@ namespace Hal {
                     uint32_t Length;
                 } Buffer;
 
+                // A package's constant form: its element bytes, left in the
+                // table they were declared in, and the declared element count. ACPI tables stay
+                // mapped for the life of the kernel, so pointing at them is
+                // both cheaper and free of the size cap a copy would carry --
+                // a GPIO community table runs well past MaxBufferLen.
+                struct {
+                    const uint8_t* Data;     // first element
+                    uint32_t       Length;   // bytes of element encodings
+                    uint32_t       Count;    // NumElements
+                } Package;
+
                 struct {
                     uint8_t  ArgCount;    // bits 0-2 of method flags
                     bool     Serialized;  // bit 3
-                    uint32_t AmlOffset;   // offset into DSDT AML where the method body starts
+                    uint8_t  Native;      // NativeMethod; 0 = AML
+                    const uint8_t* Table; // the table the body lives in
+                    uint32_t AmlOffset;   // offset into Table where the method body starts
                     uint32_t AmlLength;   // length of the method body
                 } Method;
 
@@ -103,12 +139,33 @@ namespace Hal {
                 } Region;
 
                 struct {
-                    uint32_t RegionNodeIndex; // index of the parent OperationRegion node
+                    // Field / BankField: the OperationRegion node.
+                    // IndexField: the index register field.
+                    uint32_t RegionNodeIndex;
                     uint32_t BitOffset;
                     uint32_t BitLength;
                     uint8_t  AccessType;      // 0=Any, 1=Byte, 2=Word, 3=DWord, 4=QWord, 5=Buffer
+                    uint8_t  UpdateRule;      // 0=Preserve, 1=WriteAsOnes, 2=WriteAsZeros
+                    uint8_t  Kind;            // FieldKind
+                    int32_t  DataNodeIndex;   // IndexField: the data register field
+                    int32_t  BankNodeIndex;   // BankField: the bank select field
+                    uint64_t BankValue;       // BankField: value selecting this bank
                 } Field;
 
+                // CreateXxxField over a buffer. Source points at the object
+                // holding the buffer -- a namespace node's, or a local/arg slot
+                // of the method that created the field -- so reads and writes
+                // go through to it.
+                struct {
+                    Object*  Source;
+                    uint32_t BitOffset;
+                    uint32_t BitLength;
+                } BufferField;
+
+                struct {
+                    int32_t  Node;
+                } Reference;
+
                 struct {
                     uint8_t  ProcId;
                     uint32_t PblkAddr;
@@ -116,7 +173,51 @@ namespace Hal {
                 } Processor;
             };
 
-            Object() : Type(ObjectType::None), Integer(0) {}
+            Object() : Type(ObjectType::None), PkgIndex(-1), Integer(0) {}
+        };
+
+        enum class FieldKind : uint8_t {
+            Normal = 0,
+            Index  = 1,
+            Bank   = 2,
+        };
+
+        // Methods the OS provides rather than the firmware.
+        enum class NativeMethod : uint8_t {
+            None = 0,
+            Osi  = 1,   // \_OSI
+        };
+
+        // ============================================================================
+
+        // Package element store
+
+        // ============================================================================
+        // Element arrays for materialized packages, kept outside Object so the
+        // tagged union stays small. Entries are split in two: a persistent
+        // region for packages declared with Name() (materialized once and kept
+        // for the life of the namespace) and a transient ring for packages
+        // built inside a method, which is rewound at each top-level evaluation
+        // so repeated calls -- polling _BST, say -- cannot exhaust it.
+        namespace PackageStore {
+            static constexpr int MaxPersistent = 32;
+            static constexpr int MaxTransient  = 32;
+
+            // Allocate an element array of `count` objects. Returns -1 when the
+            // store is full or the count is out of range.
+            int32_t AllocPersistent(int count);
+            int32_t AllocTransient(int count);
+
+            // Rewind the transient ring. Called once per top-level evaluation.
+            void ResetTransient();
+
+            // True for an entry that outlives the evaluation that made it.
+            bool IsPersistent(int32_t index);
+
+            // Element access. Returns nullptr for an unknown entry or an
+            // out-of-range element.
+            Object* Element(int32_t index, int element);
+            int     Count(int32_t index);
         };
 
         // ============================================================================
@@ -129,16 +230,19 @@ namespace Hal {
             char         Name[MaxNameSegLen + 1]; // null-terminated 4-char segment
             Object       Obj;
             int32_t      ParentIndex;             // -1 for root
-            int32_t      ChildIndices[MaxChildren];
+            int32_t*     ChildIndices;            // heap array, ChildCapacity long
             int32_t      ChildCount;
+            int32_t      ChildCapacity;
 
+            // Only ever called on a freshly allocated (zeroed) node, so there is
+            // no child array to release here.
             void Clear() {
                 Name[0] = 0;
                 Obj = Object{};
                 ParentIndex = -1;
+                ChildIndices = nullptr;
                 ChildCount = 0;
-                for (int i = 0; i < MaxChildren; i++)
+                ChildCapacity = 0;
-                    ChildIndices[i] = -1;
             }
         };
 
@@ -166,6 +270,18 @@ namespace Hal {
             //   3. Root scope
             int32_t ResolveName(const char* name, int32_t scopeNodeIndex) const;
 
+            // The ACPI name search rule: for a name written WITHOUT a root or
+            // parent prefix, search the current scope for its first segment,
+            // then each parent scope up to the root, and descend the remaining
+            // segments from wherever it matched. `relative` is the name as it
+            // appeared in the AML ("PBIF", "BAT0.PBIF"), never a path already
+            // rebased onto a scope. Returns -1 if no scope contains it.
+            int32_t SearchUpward(const char* relative, int32_t scopeNodeIndex) const;
+
+            // Find the child of `parent` named `seg`; a short name is padded
+            // with '_' the way node names are stored. Returns -1 if absent.
+            int32_t FindChild(int32_t parentIndex, const char* seg) const;
+
             // Get a node by index.
             NamespaceNode* GetNode(int32_t index);
             const NamespaceNode* GetNode(int32_t index) const;
@@ -217,6 +333,11 @@ namespace Hal {
         private:
             int32_t AllocNode();
             bool    AllocChunk();
+
+            // Append a child index to a parent, growing its array as needed.
+            // Returns false only when the heap is exhausted.
+            bool    AppendChild(NamespaceNode* parent, int32_t child);
+
             void    EnsureRoot();
             int32_t FindChildByName(int32_t parentIndex, const char* seg) const;
 

kernel/src/ACPI/AML/AmlParser.cpp

@@ -9,6 +9,8 @@
 #include <ACPI/ACPI.hpp>
 #include <Terminal/Terminal.hpp>
 #include <CppLib/Stream.hpp>
+#include <Memory/HHDM.hpp>
+#include <Memory/Heap.hpp>
 
 using namespace Kt;
 
@@ -239,15 +241,144 @@ namespace Hal {
         // Full interpreter initialization
 
         // ============================================================================
-        void InitializeInterpreter(void* dsdtData) {
+        void InitializeInterpreter(void* dsdtData, ACPI::CommonSDTHeader* xsdt) {
             auto& interp = GetInterpreter();
             if (!interp.LoadTable(dsdtData)) {
                 KernelLogStream(ERROR, "AML") << "Failed to load DSDT into AML interpreter";
             }
+
+            // SSDTs extend the DSDT's namespace and are loaded after it, in
+            // XSDT order. They are where most laptops keep CPU power tables,
+            // and -- the reason they matter here -- the discrete GPU's power
+            // control (_PR3 power resources, _PS3, NVIDIA/AMD _DSM).
+            int ssdts = 0;
+            if (xsdt) {
+                uint32_t entryCount = (xsdt->Length - sizeof(ACPI::CommonSDTHeader)) / 8;
+                auto* entries = (uint64_t*)((uint64_t)xsdt + sizeof(ACPI::CommonSDTHeader));
+                for (uint32_t i = 0; i < entryCount; i++) {
+                    if (entries[i] == 0) continue;
+                    auto* table = (ACPI::CommonSDTHeader*)Memory::HHDM(entries[i]);
+                    const char* sig = table->Signature;
+                    bool isSsdt = sig[0] == 'S' && sig[1] == 'S' && sig[2] == 'D' && sig[3] == 'T';
+                    bool isPsdt = sig[0] == 'P' && sig[1] == 'S' && sig[2] == 'D' && sig[3] == 'T';
+                    if (!isSsdt && !isPsdt) continue;
+                    if (table->Length <= sizeof(ACPI::CommonSDTHeader)) continue;
+                    if (interp.LoadTable(table)) ssdts++;
+                }
+            }
+            KernelLogStream(OK, "AML") << "Loaded DSDT + " << base::dec << (uint64_t)ssdts
+                << " SSDT(s)";
+
             auto& ns = interp.GetNamespace();
             KernelLogStream(OK, "AML") << "Namespace: " << base::dec
                 << (uint64_t)ns.NodeCount() << " / " << (uint64_t)ns.MaxCapacity() << " nodes (dynamic)";
         }
 
+
+        // ============================================================================
+
+        // Namespace initialization (_REG, _INI)
+
+        // ============================================================================
+        static bool SegIs(const char* name, const char* seg) {
+            return name[0] == seg[0] && name[1] == seg[1] &&
+                   name[2] == seg[2] && name[3] == seg[3];
+        }
+
+        void InitializeNamespaceObjects() {
+            auto& interp = GetInterpreter();
+            if (!interp.IsInitialized()) return;
+            auto& ns = interp.GetNamespace();
+
+            // Collect first, evaluate second: executing AML creates nodes,
+            // which reallocates a parent's child array under a live walk.
+            int32_t capacity = ns.NodeCount();
+            auto* regs = (int32_t*)Memory::g_heap->Request(sizeof(int32_t) * capacity);
+            auto* devices = (int32_t*)Memory::g_heap->Request(sizeof(int32_t) * capacity);
+            if (!regs || !devices) {
+                if (regs) Memory::g_heap->Free(regs);
+                if (devices) Memory::g_heap->Free(devices);
+                return;
+            }
+            int regCount = 0, deviceCount = 0;
+
+            ns.WalkDescendants(ns.RootIndex(), ObjectType::None,
+                [&](int32_t idx, const NamespaceNode* node) {
+                    if (node->Obj.Type == ObjectType::Method && SegIs(node->Name, "_REG") &&
+                        regCount < capacity)
+                        regs[regCount++] = idx;
+                    if (node->Obj.Type == ObjectType::Device && deviceCount < capacity)
+                        devices[deviceCount++] = idx;
+                });
+
+            // _REG(PCI_Config, connect). A _REG for another space checks Arg0
+            // and ignores the call, as with the EC's.
+            Object args[2]{};
+            args[0].Type = ObjectType::Integer;
+            args[0].Integer = (uint64_t)RegionSpace::PciConfig;
+            args[1].Type = ObjectType::Integer;
+            args[1].Integer = 1;
+            for (int i = 0; i < regCount; i++) {
+                Object result{};
+                interp.EvaluateNode(regs[i], args, 2, result);
+            }
+
+            // \_SB._INI runs before any device's.
+            int32_t sb = ns.FindNode("\\_SB_");
+            if (sb >= 0) {
+                Object result{};
+                if (interp.EvaluateChild(sb, "_INI", nullptr, 0, result))
+                    KernelLogStream(INFO, "AML") << "Ran \\_SB._INI";
+            }
+
+            // Then each device, parents first (WalkDescendants is pre-order).
+            // A device that is neither present nor functional takes its whole
+            // subtree with it; one that is functional but absent skips only
+            // its own _INI.
+            static constexpr int MaxSkipped = 256;
+            int32_t skipped[MaxSkipped];
+            int skippedCount = 0;
+            int ran = 0;
+
+            for (int i = 0; i < deviceCount; i++) {
+                int32_t dev = devices[i];
+
+                bool underSkipped = false;
+                for (int32_t p = ns.GetNode(dev) ? ns.GetNode(dev)->ParentIndex : -1;
+                     p > 0 && !underSkipped; p = ns.GetNode(p) ? ns.GetNode(p)->ParentIndex : -1) {
+                    for (int k = 0; k < skippedCount; k++)
+                        if (skipped[k] == p) { underSkipped = true; break; }
+                }
+                if (underSkipped) continue;
+
+                uint64_t sta = 0x0F;
+                Object staObj{};
+                if (interp.EvaluateChild(dev, "_STA", nullptr, 0, staObj) &&
+                    staObj.Type == ObjectType::Integer)
+                    sta = staObj.Integer;
+
+                if (!(sta & 0x01)) {
+                    if (!(sta & 0x08) && skippedCount < MaxSkipped)
+                        skipped[skippedCount++] = dev;
+                    continue;
+                }
+
+                if (interp.FindChild(dev, "_INI") < 0) continue;
+                Object result{};
+                bool ok = interp.EvaluateChild(dev, "_INI", nullptr, 0, result);
+                ran++;
+
+                char path[128];
+                ns.GetNodePath(dev, path, sizeof(path));
+                KernelLogStream(DEBUG, "AML") << "  _INI " << path << (ok ? "" : " FAILED");
+            }
+
+            KernelLogStream(OK, "AML") << "Namespace initialized: " << base::dec
+                << (uint64_t)regCount << " _REG, " << (uint64_t)ran << " device _INI";
+
+            Memory::g_heap->Free(regs);
+            Memory::g_heap->Free(devices);
+        }
+
     };
 };

kernel/src/ACPI/AML/AmlParser.hpp

@@ -5,6 +5,7 @@
 */
 
 #pragma once
+#include <ACPI/ACPI.hpp>
 #include <cstdint>
 
 namespace Hal {
@@ -27,11 +28,20 @@ namespace Hal {
         // interpreter or namespace.
         SleepObject FindSleepState(void* dsdtData, int state);
 
-        // Initialize the AML interpreter with the DSDT.
+        // Initialize the AML interpreter with the DSDT, then every SSDT the
-        // This loads the full table into the namespace and enables
+        // XSDT lists (xsdt may be null to load the DSDT alone).
+        // This loads the full tables into the namespace and enables
         // method evaluation, device enumeration, and field access.
         // Should be called during boot after ACPI table discovery.
-        void InitializeInterpreter(void* dsdtData);
+        void InitializeInterpreter(void* dsdtData, ACPI::CommonSDTHeader* xsdt);
+
+        // Run the namespace initialization the OS owes the firmware once its
+        // address-space handlers are up: _REG for PCI_Config, then \_SB._INI
+        // and every present device's _INI, parents first. Firmware works out
+        // which OS it is running under (OSYS, from _OSI) in these, and gates
+        // much of its later behaviour -- device power control included -- on
+        // the answer. Call once the Embedded Controller and PCI are up.
+        void InitializeNamespaceObjects();
 
     };
 };

kernel/src/ACPI/AcpiDevices.cpp

@@ -319,6 +319,58 @@ namespace Hal {
 
         // EnumerateAll
 
+        // ============================================================================
+        // Fill one DeviceInfo from a namespace node.
+        static void DescribeDevice(AML::Namespace& ns, int32_t idx, DeviceInfo& dev) {
+            dev.NodeIndex = idx;
+            ns.GetNodePath(idx, dev.Path, 128);
+
+            dev.Status = EvaluateSta(idx);
+            dev.IsPresent = (dev.Status & STA_PRESENT) && (dev.Status & STA_FUNCTIONAL);
+
+            if (!EvaluateHid(idx, dev.HardwareId, 16))
+                dev.HardwareId[0] = '\0';
+            if (!EvaluateUid(idx, dev.UniqueId, 16))
+                dev.UniqueId[0] = '\0';
+
+            dev.Address = EvaluateAdr(idx);
+        }
+
+        // ============================================================================
+
+        // FindDeviceByHid
+
+        // ============================================================================
+        bool FindDeviceByHid(const char* hid, DeviceInfo& out) {
+            auto& interp = AML::GetInterpreter();
+            if (!interp.IsInitialized()) return false;
+
+            auto& ns = interp.GetNamespace();
+            int32_t found = -1;
+
+            // Only _HID is evaluated per candidate; the full description is
+            // built once, for the match. WalkDescendants has no early exit, so
+            // the first match wins and later ones are ignored.
+            ns.WalkDescendants(ns.RootIndex(), AML::ObjectType::Device,
+                [&](int32_t idx, [[maybe_unused]] const AML::NamespaceNode* node) {
+                    if (found >= 0) return;
+
+                    char candidate[16];
+                    if (!EvaluateHid(idx, candidate, 16)) return;
+                    if (StrEqual(candidate, hid)) found = idx;
+                });
+
+            if (found < 0) return false;
+
+            out = DeviceInfo{};
+            DescribeDevice(ns, found, out);
+            return true;
+        }
+
+        // ============================================================================
+
+        // EnumerateAll
+
         // ============================================================================
         void EnumerateAll(DeviceList& result) {
             auto& interp = AML::GetInterpreter();
@@ -326,39 +378,32 @@ namespace Hal {
 
             auto& ns = interp.GetNamespace();
             result.Count = 0;
+            result.Truncated = false;
 
             // Walk all namespace nodes looking for Device objects
             ns.WalkDescendants(ns.RootIndex(), AML::ObjectType::Device,
                 [&](int32_t idx, [[maybe_unused]] const AML::NamespaceNode* node) {
-                    if (result.Count >= MaxDevices) return;
+                    if (result.Count >= MaxDevices) {
-
+                        result.Truncated = true;
-                    auto& dev = result.Devices[result.Count];
+                        return;
-                    dev.NodeIndex = idx;
+                    }
-
-                    // Get the path
-                    ns.GetNodePath(idx, dev.Path, 128);
-
-                    // Evaluate _STA
-                    dev.Status = EvaluateSta(idx);
-                    dev.IsPresent = (dev.Status & STA_PRESENT) && (dev.Status & STA_FUNCTIONAL);
-
-                    // Evaluate _HID
-                    if (!EvaluateHid(idx, dev.HardwareId, 16))
-                        dev.HardwareId[0] = '\0';
-
-                    // Evaluate _UID
-                    if (!EvaluateUid(idx, dev.UniqueId, 16))
-                        dev.UniqueId[0] = '\0';
-
-                    // Evaluate _ADR
-                    dev.Address = EvaluateAdr(idx);
 
+                    DescribeDevice(ns, idx, result.Devices[result.Count]);
                     result.Count++;
                 });
 
             KernelLogStream(OK, "ACPI") << "Enumerated " << base::dec
                 << (uint64_t)result.Count << " ACPI devices";
 
+            // The list is a fixed-size stack object, so a large namespace runs
+            // past it. Say so: a device missing from the log below may simply
+            // be one that did not fit, and looking it up by HID still works.
+            if (result.Truncated) {
+                KernelLogStream(WARNING, "ACPI") << "Device list truncated at "
+                    << base::dec << (uint64_t)MaxDevices
+                    << "; use FindDeviceByHid to reach the rest";
+            }
+
             // Log discovered devices
             for (int i = 0; i < result.Count; i++) {
                 auto& dev = result.Devices[i];
@@ -408,15 +453,14 @@ namespace Hal {
             auto* nsNode = ns.GetNode(node);
             if (!nsNode) return result;
 
-            if (nsNode->Obj.Type == AML::ObjectType::Package) {
+            if (nsNode->Obj.Type == AML::ObjectType::Package && nsNode->Obj.Package.Data) {
-                // Parse the raw package buffer to extract SLP_TYP values
+                // Parse the raw package elements to extract SLP_TYP values
-                const uint8_t* data = nsNode->Obj.Buffer.Data;
+                const uint8_t* data = nsNode->Obj.Package.Data;
-                uint32_t len = nsNode->Obj.Buffer.Length;
+                uint32_t len = nsNode->Obj.Package.Length;
-                if (len < 2) return result;
+                if (len < 1) return result;
 
-                // NumElements
+                uint32_t numElements = nsNode->Obj.Package.Count;
-                uint8_t numElements = data[0];
+                uint32_t pos = 0;
-                uint32_t pos = 1;
 
                 if (numElements >= 1 && pos < len) {
                     // Decode first element

kernel/src/ACPI/AcpiDevices.hpp

@@ -45,8 +45,9 @@ namespace Hal {
         struct DeviceList {
             DeviceInfo Devices[MaxDevices];
             int        Count;
+            bool       Truncated;   // the namespace held more than MaxDevices
 
-            DeviceList() : Count(0) {}
+            DeviceList() : Count(0), Truncated(false) {}
 
             // Find a device by HID. Returns nullptr if not found.
             const DeviceInfo* FindByHid(const char* hid) const;
@@ -64,6 +65,12 @@ namespace Hal {
         // The interpreter must be initialized first (LoadTable called).
         void EnumerateAll(DeviceList& result);
 
+        // Find the first device with the given _HID, walking the namespace
+        // directly. Unlike EnumerateAll this has no device-count ceiling and
+        // needs no multi-kilobyte list on the caller's stack, so it is the way
+        // to reach a specific device on a machine whose namespace is large.
+        bool FindDeviceByHid(const char* hid, DeviceInfo& out);
+
         // Evaluate _STA for a device node. Returns STA_DEFAULT if no _STA exists.
         uint32_t EvaluateSta(int32_t deviceNodeIndex);
 

kernel/src/ACPI/AcpiEvents.cpp

@@ -7,9 +7,7 @@
 #include "AcpiEvents.hpp"
 #include <ACPI/AcpiSleep.hpp>
 #include <Io/IoPort.hpp>
-#include <Hal/Apic/Apic.hpp>
+#include <Arch/Interrupts.hpp>
-#include <Hal/Apic/IoApic.hpp>
-#include <Hal/Apic/Interrupts.hpp>
 #include <Terminal/Terminal.hpp>
 #include <CppLib/Stream.hpp>
 
@@ -191,24 +189,21 @@ namespace Hal {
             // routed. Clear stale status only after the enables are down.
             DisableAndClearGpes();
 
-            // Route SCI to an IRQ vector. The SCI is level-triggered,
+            // Route the SCI to its IRQ slot. The SCI is level-triggered,
             // active-low (ACPI spec requirement). The MADT may have an
             // Interrupt Source Override for this IRQ that already sets
-            // these flags; RouteIrq applies overrides automatically.
+            // these flags; RouteLegacyIrq applies overrides automatically.
-            uint8_t sciVector = IRQ_VECTOR_BASE + (uint8_t)g_sciIrq;
-            uint8_t bspApicId = (uint8_t)LocalApic::GetId();
 
             // Register the handler before routing to avoid missing events
-            RegisterIrqHandler((uint8_t)g_sciIrq, SciHandler);
+            Arch::RegisterIrqHandler((uint8_t)g_sciIrq, SciHandler);
 
-            // Route via IoApic (applies MADT overrides for polarity/trigger)
+            // Route to the boot CPU (applies firmware polarity/trigger overrides)
-            IoApic::RouteIrq((uint8_t)g_sciIrq, sciVector, bspApicId);
+            Arch::RouteLegacyIrq((uint8_t)g_sciIrq);
 
             g_initialized = true;
 
             KernelLogStream(OK, "ACPI") << "ACPI events initialized (SCI IRQ "
-                << base::dec << (uint64_t)g_sciIrq << ", vector "
+                << base::dec << (uint64_t)g_sciIrq << ")";
-                << (uint64_t)sciVector << ")";
         }
 
         void Reinitialize() {

kernel/src/ACPI/AcpiPower.cpp

@@ -0,0 +1,336 @@
+/*
+    * AcpiPower.cpp
+    * ACPI device power management: _PSx methods, _PRx power resources,
+    * _DSM calls, and finding the namespace node behind a PCI function
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#include "AcpiPower.hpp"
+#include <ACPI/AML/AmlInterpreter.hpp>
+#include <Memory/Heap.hpp>
+#include <Libraries/Memory.hpp>
+#include <Terminal/Terminal.hpp>
+#include <CppLib/Stream.hpp>
+
+using namespace Kt;
+
+namespace Hal {
+    namespace AcpiPower {
+
+        // ============================================================================
+
+        // PCI <-> ACPI
+
+        // ============================================================================
+        struct PciMapEntry {
+            int32_t Node;
+            uint8_t Bus, Dev, Fn;
+            uint8_t Score;
+        };
+
+        static PciMapEntry* g_map = nullptr;
+        static int g_mapCount = 0;
+        static bool g_mapBuilt = false;
+
+        bool HasObject(int32_t device, const char* name) {
+            return AML::GetInterpreter().FindChild(device, name) >= 0;
+        }
+
+        // How much power management a node carries. Picks the real node out
+        // of duplicates at one address.
+        static uint8_t PowerScore(int32_t node) {
+            uint8_t score = 0;
+            if (HasObject(node, "_PR0") || HasObject(node, "_PR3")) score += 8;
+            if (HasObject(node, "_PS3")) score += 4;
+            if (HasObject(node, "_DSM")) score += 2;
+            if (HasObject(node, "_OFF") || HasObject(node, "_PS0")) score += 1;
+            return score;
+        }
+
+        static void BuildPciMap() {
+            auto& interp = AML::GetInterpreter();
+            auto& ns = interp.GetNamespace();
+            g_mapBuilt = true;
+            g_mapCount = 0;
+            if (!interp.IsInitialized()) return;
+
+            // Device nodes with an _ADR, collected before anything is
+            // evaluated (evaluation may add nodes).
+            int32_t capacity = ns.NodeCount();
+            auto* nodes = (int32_t*)Memory::g_heap->Request(sizeof(int32_t) * capacity);
+            if (!nodes) return;
+            int count = 0;
+            int32_t sb = ns.FindNode("\\_SB_");
+            if (sb < 0) sb = ns.RootIndex();
+            ns.WalkDescendants(sb, AML::ObjectType::Device,
+                [&](int32_t idx, const AML::NamespaceNode*) {
+                    if (count < capacity && ns.FindChild(idx, "_ADR") >= 0)
+                        nodes[count++] = idx;
+                });
+
+            if (g_map) Memory::g_heap->Free(g_map);
+            g_map = (PciMapEntry*)Memory::g_heap->Request(sizeof(PciMapEntry) * (count ? count : 1));
+            if (!g_map) { Memory::g_heap->Free(nodes); return; }
+
+            for (int i = 0; i < count; i++) {
+                uint8_t bus, dev, fn;
+                if (!interp.GetPciAddress(nodes[i], bus, dev, fn)) continue;
+                g_map[g_mapCount++] = { nodes[i], bus, dev, fn, PowerScore(nodes[i]) };
+            }
+
+            Memory::g_heap->Free(nodes);
+        }
+
+        void InvalidatePciMap() {
+            g_mapBuilt = false;
+        }
+
+        int32_t FindPciDevice(uint8_t bus, uint8_t dev, uint8_t fn) {
+            if (!g_mapBuilt) BuildPciMap();
+
+            int32_t best = -1;
+            int bestScore = -1;
+            for (int i = 0; i < g_mapCount; i++) {
+                const auto& e = g_map[i];
+                if (e.Bus != bus || e.Dev != dev || e.Fn != fn) continue;
+                if (e.Score > bestScore) {
+                    best = e.Node;
+                    bestScore = e.Score;
+                }
+            }
+            return best;
+        }
+
+        // ============================================================================
+
+        // Power resources
+
+        // ============================================================================
+        static int AddUnique(int32_t* out, int count, int max, int32_t node) {
+            if (node < 0) return count;
+            for (int i = 0; i < count; i++)
+                if (out[i] == node) return count;
+            if (count < max) out[count++] = node;
+            return count;
+        }
+
+        // The resources one _PRx object lists. The list is copied out before
+        // anything else is evaluated: a computed package lives in storage the
+        // next evaluation reuses.
+        static int CollectList(int32_t device, const char* name, int32_t* out, int count, int max) {
+            auto& interp = AML::GetInterpreter();
+            auto& ns = interp.GetNamespace();
+
+            AML::Object pkg{};
+            if (!interp.EvaluateChild(device, name, nullptr, 0, pkg)) return count;
+            if (pkg.Type != AML::ObjectType::Package) return count;
+
+            int n = interp.GetPackageCount(pkg);
+            int32_t found[MaxResources];
+            int foundCount = 0;
+            for (int i = 0; i < n && foundCount < MaxResources; i++) {
+                AML::Object elem{};
+                if (!interp.GetPackageElement(pkg, i, elem)) break;
+                int32_t node = interp.ResolveReference(elem, device);
+                auto* nn = ns.GetNode(node);
+                if (nn && nn->Obj.Type == AML::ObjectType::PowerResource)
+                    found[foundCount++] = node;
+            }
+
+            for (int i = 0; i < foundCount; i++)
+                count = AddUnique(out, count, max, found[i]);
+            return count;
+        }
+
+        int CollectResources(int32_t device, int32_t* out, int max) {
+            int count = 0;
+            count = CollectList(device, "_PR0", out, count, max);
+            count = CollectList(device, "_PR1", out, count, max);
+            count = CollectList(device, "_PR2", out, count, max);
+            count = CollectList(device, "_PR3", out, count, max);
+            return count;
+        }
+
+        int ResourceState(int32_t resource) {
+            AML::Object sta{};
+            if (!AML::GetInterpreter().EvaluateChild(resource, "_STA", nullptr, 0, sta))
+                return -1;
+            return (ResultInteger(sta) & 1) ? 1 : 0;
+        }
+
+        // Does a present device outside `owners` list this resource for D0?
+        // Turning off a rail another device is running on is how a GPU
+        // power-down would take out, say, an NVMe drive sharing a slot rail.
+        static bool NeededElsewhere(int32_t resource, const int32_t* owners, int ownerCount) {
+            auto& interp = AML::GetInterpreter();
+            auto& ns = interp.GetNamespace();
+
+            int32_t capacity = ns.NodeCount();
+            auto* users = (int32_t*)Memory::g_heap->Request(sizeof(int32_t) * capacity);
+            if (!users) return true;   // cannot tell: assume shared
+            int count = 0;
+            ns.WalkDescendants(ns.RootIndex(), AML::ObjectType::Device,
+                [&](int32_t idx, const AML::NamespaceNode*) {
+                    if (count < capacity && ns.FindChild(idx, "_PR0") >= 0)
+                        users[count++] = idx;
+                });
+
+            bool shared = false;
+            for (int i = 0; i < count && !shared; i++) {
+                int32_t dev = users[i];
+
+                // Everything at or below an owner goes down with it: the
+                // card's own child nodes, a duplicate placeholder node at the
+                // same address, a function the firmware hid from PCI.
+                bool own = false;
+                for (int32_t a = dev; a > 0 && !own; ) {
+                    for (int k = 0; k < ownerCount; k++)
+                        if (owners[k] == a) { own = true; break; }
+                    auto* an = ns.GetNode(a);
+                    a = an ? an->ParentIndex : -1;
+                }
+                if (own) continue;
+
+                int32_t list[MaxResources];
+                int n = CollectList(dev, "_PR0", list, 0, MaxResources);
+                bool listed = false;
+                for (int k = 0; k < n; k++)
+                    if (list[k] == resource) { listed = true; break; }
+                if (!listed) continue;
+
+                // Only a device that is actually there counts.
+                AML::Object sta{};
+                uint64_t status = 0x0F;
+                if (interp.EvaluateChild(dev, "_STA", nullptr, 0, sta))
+                    status = ResultInteger(sta);
+                if (status & 0x01) {
+                    char path[128];
+                    ns.GetNodePath(dev, path, sizeof(path));
+                    KernelLogStream(INFO, "AcpiPower") << "  power resource also used by "
+                        << path << "; leaving it on";
+                    shared = true;
+                }
+            }
+
+            Memory::g_heap->Free(users);
+            return shared;
+        }
+
+        // ============================================================================
+
+        // Device power transitions
+
+        // ============================================================================
+        void EnterD3Cold(int32_t device, const int32_t* owners, int ownerCount,
+                         D3ColdReport& report) {
+            auto& interp = AML::GetInterpreter();
+            auto& ns = interp.GetNamespace();
+            report = D3ColdReport{};
+            if (device < 0) return;
+
+            char path[128];
+            ns.GetNodePath(device, path, sizeof(path));
+
+            // Gather the resources first: _PS3 may change what _PRx reports.
+            int32_t resources[MaxResources];
+            int count = CollectResources(device, resources, MaxResources);
+
+            if (HasObject(device, "_PS3")) {
+                AML::Object result{};
+                bool ok = interp.EvaluateChild(device, "_PS3", nullptr, 0, result);
+                report.RanPs3 = true;
+                KernelLogStream(INFO, "AcpiPower") << "  " << path << "._PS3"
+                    << (ok ? "" : " (failed)");
+            }
+
+            // Off in reverse order of listing (ACPI turns resources off in
+            // reverse ResourceOrder; firmware lists them in that order).
+            for (int i = count - 1; i >= 0; i--) {
+                int32_t res = resources[i];
+                char rpath[128];
+                ns.GetNodePath(res, rpath, sizeof(rpath));
+
+                if (ResourceState(res) == 0) {
+                    KernelLogStream(INFO, "AcpiPower") << "  " << rpath << " already off";
+                    continue;
+                }
+                if (NeededElsewhere(res, owners, ownerCount)) {
+                    report.ResourcesShared++;
+                    continue;
+                }
+
+                AML::Object result{};
+                bool ok = interp.EvaluateChild(res, "_OFF", nullptr, 0, result);
+                int after = ResourceState(res);
+                if (ok) report.ResourcesOff++;
+                if (after == 1) report.ResourcesStillOn++;
+                KernelLogStream(INFO, "AcpiPower") << "  " << rpath << "._OFF"
+                    << (ok ? "" : " (failed)")
+                    << (after == 0 ? ", now off" : after == 1 ? ", still on" : "");
+            }
+        }
+
+        // ============================================================================
+
+        // _DSM
+
+        // ============================================================================
+        AML::Object IntegerArg(uint64_t value) {
+            AML::Object o{};
+            o.Type = AML::ObjectType::Integer;
+            o.Integer = value;
+            return o;
+        }
+
+        AML::Object BufferArg(const uint8_t* bytes, uint32_t length) {
+            AML::Object o{};
+            o.Type = AML::ObjectType::Buffer;
+            if (length > AML::MaxBufferLen) length = AML::MaxBufferLen;
+            memcpy(o.Buffer.Data, bytes, length);
+            o.Buffer.Length = length;
+            return o;
+        }
+
+        AML::Object EmptyPackageArg() {
+            AML::Object o{};
+            o.Type = AML::ObjectType::Package;
+            o.Package.Data = nullptr;
+            o.Package.Length = 0;
+            o.Package.Count = 0;
+            return o;
+        }
+
+        uint64_t ResultInteger(const AML::Object& result) {
+            if (result.Type == AML::ObjectType::Integer) return result.Integer;
+            if (result.Type == AML::ObjectType::Buffer) {
+                uint64_t v = 0;
+                for (uint32_t i = 0; i < result.Buffer.Length && i < 8; i++)
+                    v |= (uint64_t)result.Buffer.Data[i] << (8 * i);
+                return v;
+            }
+            return 0;
+        }
+
+        bool EvaluateDsm(int32_t device, const uint8_t uuid[16], uint64_t revision,
+                         uint64_t function, const AML::Object& arg3, AML::Object& result) {
+            auto& interp = AML::GetInterpreter();
+            if (device < 0 || interp.FindChild(device, "_DSM") < 0) return false;
+
+            AML::Object args[4];
+            args[0] = BufferArg(uuid, 16);
+            args[1] = IntegerArg(revision);
+            args[2] = IntegerArg(function);
+            args[3] = arg3;
+            return interp.EvaluateChild(device, "_DSM", args, 4, result);
+        }
+
+        uint64_t DsmFunctions(int32_t device, const uint8_t uuid[16], uint64_t revision,
+                              const AML::Object& arg3) {
+            AML::Object result{};
+            if (!EvaluateDsm(device, uuid, revision, 0, arg3, result)) return 0;
+            uint64_t mask = ResultInteger(result);
+            // Bit 0 clear means "no functions for this UUID/revision".
+            return (mask & 1) ? mask : 0;
+        }
+    };
+};

kernel/src/ACPI/AcpiPower.hpp

@@ -0,0 +1,92 @@
+/*
+    * AcpiPower.hpp
+    * ACPI device power management: _PSx methods, _PRx power resources,
+    * _DSM calls, and finding the namespace node behind a PCI function
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+#include <ACPI/AML/AmlNamespace.hpp>
+
+namespace Hal {
+    namespace AcpiPower {
+
+        // ============================================================================
+
+        // PCI <-> ACPI
+
+        // ============================================================================
+        // The Device node describing PCI function bus:dev.fn, or -1. When
+        // firmware declares more than one node at the same address (a generic
+        // placeholder plus a vendor SSDT's real one is common for GPUs), the
+        // one carrying power-management objects wins.
+        int32_t FindPciDevice(uint8_t bus, uint8_t dev, uint8_t fn);
+
+        // Forget the PCI address map built by FindPciDevice (bus numbers may
+        // change across a resume).
+        void InvalidatePciMap();
+
+        // True when `device` has a child object named `name` ("_PS3").
+        bool HasObject(int32_t device, const char* name);
+
+        // ============================================================================
+
+        // Power resources
+
+        // ============================================================================
+        static constexpr int MaxResources = 8;
+
+        // The power resources the device lists in _PR0.._PR3 (deduplicated).
+        int CollectResources(int32_t device, int32_t* out, int max);
+
+        // _STA of a power resource: 1 on, 0 off, -1 unknown.
+        int ResourceState(int32_t resource);
+
+        // ============================================================================
+
+        // Device power transitions
+
+        // ============================================================================
+        struct D3ColdReport {
+            bool RanPs3;
+            int  ResourcesOff;      // resources whose _OFF was run
+            int  ResourcesShared;   // left on: some other present device needs them
+            int  ResourcesStillOn;  // _OFF ran but _STA still says on
+        };
+
+        // Put an ACPI device into D3cold the way the spec orders it: _PS3,
+        // then every power resource it depends on off, last-listed first. A
+        // resource that a device outside `owners` also depends on is left
+        // on -- `owners` lists the nodes being powered down together (the
+        // functions of one card and the port above them), which may freely
+        // share, as may anything below them in the namespace. The PCI side
+        // (PMCSR to D3hot) is the caller's to do first.
+        void EnterD3Cold(int32_t device, const int32_t* owners, int ownerCount,
+                         D3ColdReport& report);
+
+        // ============================================================================
+
+        // _DSM
+
+        // ============================================================================
+        // Evaluate _DSM(uuid, revision, function, arg3) on a device. False
+        // when the device has no _DSM or it failed.
+        bool EvaluateDsm(int32_t device, const uint8_t uuid[16], uint64_t revision,
+                         uint64_t function, const AML::Object& arg3, AML::Object& result);
+
+        // The function bitmap _DSM function 0 reports for (uuid, revision),
+        // 0 when unsupported. Bit 0 set means "some functions exist".
+        uint64_t DsmFunctions(int32_t device, const uint8_t uuid[16], uint64_t revision,
+                              const AML::Object& arg3);
+
+        // Helpers for building _DSM / method arguments.
+        AML::Object IntegerArg(uint64_t value);
+        AML::Object BufferArg(const uint8_t* bytes, uint32_t length);
+        AML::Object EmptyPackageArg();
+
+        // Integer value of a method result: an integer as is, a buffer read
+        // little-endian (up to 8 bytes).
+        uint64_t ResultInteger(const AML::Object& result);
+    };
+};

kernel/src/ACPI/AcpiShutdown.cpp

@@ -5,6 +5,7 @@
 */
 
 #include "AcpiShutdown.hpp"
+#include <Arch/Cpu.hpp>
 #include "AcpiSleep.hpp"
 #include "AcpiEvents.hpp"
 #include <ACPI/FADT.hpp>
@@ -37,8 +38,24 @@ namespace Hal {
                 return;
             }
 
-            // Map and parse the DSDT to find \_S5_
             auto* dsdt = (void*)Memory::HHDM(fadt.DsdtAddress);
+
+            AML::InitializeInterpreter(dsdt, xsdt);
+
+            if (AML::GetInterpreter().IsInitialized()) {
+                AcpiDevices::DeviceList devices{};
+                AcpiDevices::EnumerateAll(devices);
+            }
+
+            AcpiSleep::Initialize(xsdt);
+
+            // Hardware-reduced ACPI (ARM) has no PM1 registers: the
+            // architecture powers off instead (Arch::PowerOff).
+            if ((fadt.Flags & FADT::FlagHardwareReduced) || fadt.PM1aControlBlock == 0) {
+                KernelLogStream(INFO, "ACPI") << "No PM1 control block - ACPI S5 shutdown unavailable";
+                return;
+            }
+
             AML::S5Object s5 = AML::FindS5(dsdt);
 
             if (!s5.Valid) {
@@ -54,23 +71,6 @@ namespace Hal {
 
             KernelLogStream(OK, "ACPI") << "ACPI shutdown initialized (PM1a=" << base::hex
                 << (uint64_t)g_pm1aControlBlock << " SLP_TYPa=" << (uint64_t)g_slpTypA << ")";
-
-            // Initialize the full AML interpreter with the DSDT.
-            // This loads the entire DSDT namespace, enabling device enumeration,
-            // method evaluation, and field I/O for the rest of the kernel.
-            AML::InitializeInterpreter(dsdt);
-
-            // Enumerate ACPI devices now that the namespace is loaded
-            if (AML::GetInterpreter().IsInitialized()) {
-                AcpiDevices::DeviceList devices{};
-                AcpiDevices::EnumerateAll(devices);
-            }
-
-            // Initialize S3 suspend support (reads FACS and \_S3_ from namespace)
-            AcpiSleep::Initialize(xsdt);
-
-            // Note: ACPI event handling (SCI, power button) is initialized
-            // from Main.cpp after the APIC subsystem is ready.
         }
 
         bool IsAvailable() {
@@ -95,7 +95,7 @@ namespace Hal {
                 }
             }
 
-            asm volatile("cli");
+            Arch::DisableInterrupts();
 
             // Phase 1: Write SLP_TYP to PM1x_CNT without SLP_EN,
             //          preserving existing register bits
@@ -110,7 +110,7 @@ namespace Hal {
             }
 
             // Flush all caches before entering sleep state
-            asm volatile("wbinvd");
+            Arch::WritebackInvalidateCaches();
 
             // Phase 2: Assert SLP_EN to trigger the transition
             Io::Out16(pm1a | SLP_EN, (uint16_t)g_pm1aControlBlock);
@@ -122,7 +122,7 @@ namespace Hal {
 
             // Halt and wait for the hardware to power off
             for (;;) {
-                asm volatile("hlt");
+                Arch::Halt();
             }
         }
     };

kernel/src/ACPI/AcpiSleep.hpp

@@ -75,21 +75,6 @@ namespace Hal {
 
         // CPU State (saved across S3)
 
-        // ============================================================================
-        // Layout must match S3Wake.asm offsets exactly.
-        // No FPU/SSE state — the kernel is compiled with -mno-sse.
-        struct CpuState {
-            uint64_t Rax, Rbx, Rcx, Rdx;   // 0x00 - 0x18
-            uint64_t Rsi, Rdi, Rbp, Rsp;   // 0x20 - 0x38
-            uint64_t R8, R9, R10, R11;     // 0x40 - 0x58
-            uint64_t R12, R13, R14, R15;   // 0x60 - 0x78
-            uint64_t Rflags;               // 0x80
-            uint64_t Rip;                  // 0x88
-            uint64_t Cr3;                  // 0x90
-            uint64_t Cr0, Cr4;             // 0x98, 0xA0
-            uint8_t  GdtPtr[10];           // 0xA8
-            uint8_t  IdtPtr[10];           // 0xB2
-        } __attribute__((aligned(16)));
 
         // ============================================================================
 
@@ -109,5 +94,9 @@ namespace Hal {
         // Returns 0 on successful resume, or -1 if S3 is unavailable.
         int Suspend();
 
+        // Initialize/IsS3Available/Suspend are implemented by the
+        // architecture (the FACS waking-vector path is an x86 platform
+        // mechanism); other targets report S3 as unavailable.
+
     };
 };

kernel/src/ACPI/AcpiVideo.cpp

@@ -0,0 +1,212 @@
+/*
+    * AcpiVideo.cpp
+    * ACPI video extensions (_BCL/_BCM/_BQC) panel brightness fallback
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#include "AcpiVideo.hpp"
+#include "AcpiPower.hpp"
+#include <ACPI/AML/AmlInterpreter.hpp>
+#include <Pci/Pci.hpp>
+#include <CppLib/Spinlock.hpp>
+#include <CppLib/Stream.hpp>
+#include <Terminal/Terminal.hpp>
+
+using namespace Kt;
+
+namespace Hal {
+    namespace AcpiVideo {
+
+        // ============================================================================
+
+        // State
+
+        // ============================================================================
+        static constexpr int MaxLevels = 101;
+
+        static kcp::Mutex g_lock;
+        static bool      g_probed = false;
+        static int32_t   g_output = -1;          // output device carrying _BCM
+        static bool      g_hasBqc = false;
+        static uint32_t  g_levels[MaxLevels];    // ascending, de-duplicated
+        static int       g_levelCount = 0;
+        static int       g_percent = -1;         // last level set, when no _BQC
+
+        // ============================================================================
+
+        // Discovery
+
+        // ============================================================================
+        static bool HasMethod(int32_t node, const char* name) {
+            return AML::GetInterpreter().FindChild(node, name) >= 0;
+        }
+
+        // _ADR display output IDs (ACPI 6.x, B.4.2): bits 11:8 are the type,
+        // 4 = internal flat panel. The panel is the output to drive.
+        static int32_t FindOutput() {
+            auto& interp = AML::GetInterpreter();
+            auto& ns = interp.GetNamespace();
+
+            int32_t fallback = -1;
+            auto& devices = Pci::GetDevices();
+            for (uint64_t i = 0; i < devices.size(); i++) {
+                const auto& d = devices[i];
+                if (d.ClassCode != 0x03) continue;
+                int32_t gpu = AcpiPower::FindPciDevice(d.Bus, d.Device, d.Function);
+                if (gpu < 0) continue;
+
+                int32_t panel = -1;
+                ns.ForEachChild(gpu, AML::ObjectType::Device,
+                    [&](int32_t idx, const AML::NamespaceNode*) {
+                        if (!HasMethod(idx, "_BCM") || !HasMethod(idx, "_BCL")) return true;
+                        AML::Object adr{};
+                        uint64_t id = interp.EvaluateChild(idx, "_ADR", nullptr, 0, adr)
+                                    ? AcpiPower::ResultInteger(adr) : 0;
+                        if (((id >> 8) & 0xF) == 0x4) {
+                            panel = idx;
+                            return false;
+                        }
+                        if (fallback < 0) fallback = idx;
+                        return true;
+                    });
+                if (panel >= 0) return panel;
+            }
+            if (fallback >= 0) return fallback;
+
+            // Some firmware hangs the output device off a bridge the PCI scan
+            // could not map; any _BCM in the namespace is the only one there is.
+            ns.WalkDescendants(ns.RootIndex(), AML::ObjectType::Method,
+                [&](int32_t idx, const AML::NamespaceNode* n) {
+                    if (fallback >= 0 || n->Name[0] != '_' || n->Name[1] != 'B'
+                        || n->Name[2] != 'C' || n->Name[3] != 'M')
+                        return;
+                    const AML::NamespaceNode* node = ns.GetNode(idx);
+                    if (node && node->ParentIndex >= 0 && HasMethod(node->ParentIndex, "_BCL"))
+                        fallback = node->ParentIndex;
+                });
+            return fallback;
+        }
+
+        // _BCL: { level on AC, level on battery, supported levels... }. Only
+        // the supported list counts; firmware often repeats or misorders it.
+        static bool ReadLevels() {
+            auto& interp = AML::GetInterpreter();
+            AML::Object bcl{};
+            if (!interp.EvaluateChild(g_output, "_BCL", nullptr, 0, bcl)
+                || bcl.Type != AML::ObjectType::Package)
+                return false;
+
+            g_levelCount = 0;
+            int count = interp.GetPackageCount(bcl);
+            for (int i = 2; i < count; i++) {
+                AML::Object elem{};
+                if (!interp.GetPackageElement(bcl, i, elem)
+                    || elem.Type != AML::ObjectType::Integer)
+                    continue;
+                uint32_t level = (uint32_t)elem.Integer;
+                int pos = 0;
+                while (pos < g_levelCount && g_levels[pos] < level) pos++;
+                if (pos < g_levelCount && g_levels[pos] == level) continue;
+                if (g_levelCount == MaxLevels) break;
+                for (int j = g_levelCount; j > pos; j--) g_levels[j] = g_levels[j - 1];
+                g_levels[pos] = level;
+                g_levelCount++;
+            }
+            return g_levelCount >= 2 && g_levels[g_levelCount - 1] > 0;
+        }
+
+        static void Probe() {
+            g_probed = true;
+            auto& interp = AML::GetInterpreter();
+            if (!interp.IsInitialized()) return;
+
+            g_output = FindOutput();
+            if (g_output < 0) return;
+            if (!ReadLevels()) {
+                g_output = -1;
+                return;
+            }
+            g_hasBqc = HasMethod(g_output, "_BQC");
+            // Without _BQC the level is only known once we set one; until
+            // then firmware's boot level is assumed to be full brightness.
+            if (!g_hasBqc) g_percent = 100;
+
+            char path[128];
+            interp.GetNamespace().GetNodePath(g_output, path, sizeof(path));
+            KernelLogStream(INFO, "AcpiVideo") << "Brightness via " << path
+                << "._BCM, " << base::dec << (uint64_t)g_levelCount << " levels"
+                << (g_hasBqc ? ", _BQC" : "");
+        }
+
+        // ============================================================================
+
+        // Levels
+
+        // ============================================================================
+        static int LevelToPercent(uint32_t level) {
+            uint32_t top = g_levels[g_levelCount - 1];
+            if (level > top) level = top;
+            return (int)(((uint64_t)level * 100 + top / 2) / top);
+        }
+
+        // The supported level closest to the requested share of the top one.
+        static uint32_t PercentToLevel(int percent) {
+            uint64_t target = (uint64_t)g_levels[g_levelCount - 1] * (uint32_t)percent / 100;
+            uint32_t best = g_levels[0];
+            uint64_t bestDistance = ~0ULL;
+            for (int i = 0; i < g_levelCount; i++) {
+                uint64_t distance = g_levels[i] > target ? g_levels[i] - target
+                                                         : target - g_levels[i];
+                if (distance < bestDistance) {
+                    bestDistance = distance;
+                    best = g_levels[i];
+                }
+            }
+            return best;
+        }
+
+        static int Query() {
+            if (g_hasBqc) {
+                AML::Object result{};
+                if (AML::GetInterpreter().EvaluateChild(g_output, "_BQC", nullptr, 0, result)
+                    && result.Type == AML::ObjectType::Integer)
+                    return LevelToPercent((uint32_t)result.Integer);
+            }
+            return g_percent;
+        }
+
+        // ============================================================================
+
+        // Public API
+
+        // ============================================================================
+        int Brightness(int percent) {
+            g_lock.Acquire();
+            if (!g_probed) Probe();
+            if (g_output < 0) {
+                g_lock.Release();
+                return -1;
+            }
+
+            int result;
+            if (percent < 0) {
+                result = Query();
+            } else {
+                if (percent > 100) percent = 100;
+                uint32_t level = PercentToLevel(percent);
+                AML::Object arg = AcpiPower::IntegerArg(level);
+                AML::Object ignored{};
+                if (AML::GetInterpreter().EvaluateChild(g_output, "_BCM", &arg, 1, ignored)) {
+                    g_percent = LevelToPercent(level);
+                    int reported = Query();
+                    result = reported >= 0 ? reported : g_percent;
+                } else {
+                    result = -2;
+                }
+            }
+            g_lock.Release();
+            return result;
+        }
+
+    };
+};

kernel/src/ACPI/AcpiVideo.hpp

@@ -0,0 +1,23 @@
+/*
+    * AcpiVideo.hpp
+    * ACPI video extensions (_BCL/_BCM/_BQC) panel brightness fallback
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+
+namespace Hal {
+    namespace AcpiVideo {
+
+        // Panel brightness through the firmware's ACPI video output device,
+        // for machines no native backlight driver covers (a panel wired to a
+        // discrete GPU, a GPU Montauk has no driver for). percent < 0 queries;
+        // otherwise the nearest level the firmware lists is selected. Returns
+        // the resulting 0..100, -1 when the firmware has no brightness
+        // methods, -2 when evaluating them failed. Discovery runs on first
+        // use. May sleep (AML).
+        int Brightness(int percent);
+
+    };
+};

kernel/src/ACPI/AcpiWmi.cpp

@@ -0,0 +1,127 @@
+/*
+    * AcpiWmi.cpp
+    * ACPI-WMI mapper (PNP0C14): find a WMI GUID's data block in _WDG and call
+    * its WMxx method
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#include "AcpiWmi.hpp"
+#include "AcpiDevices.hpp"
+#include <ACPI/AML/AmlInterpreter.hpp>
+#include <Libraries/Memory.hpp>
+#include <CppLib/Stream.hpp>
+#include <Terminal/Terminal.hpp>
+
+using namespace Kt;
+
+namespace Hal {
+    namespace AcpiWmi {
+
+        static constexpr const char* HidWmi = "PNP0C14";
+        static constexpr uint32_t WdgEntrySize = 20;   // GUID, id, count, flags
+
+        static int HexDigit(char c) {
+            if (c >= '0' && c <= '9') return c - '0';
+            if (c >= 'a' && c <= 'f') return c - 'a' + 10;
+            if (c >= 'A' && c <= 'F') return c - 'A' + 10;
+            return -1;
+        }
+
+        bool ParseGuid(const char* text, uint8_t out[16]) {
+            // Bytes in text order, then the first three fields byte-swapped.
+            uint8_t bytes[16];
+            int count = 0;
+            for (int i = 0; text[i]; i++) {
+                if (text[i] == '-') continue;
+                int hi = HexDigit(text[i]);
+                int lo = text[i + 1] ? HexDigit(text[i + 1]) : -1;
+                if (hi < 0 || lo < 0 || count == 16) return false;
+                bytes[count++] = (uint8_t)(hi << 4 | lo);
+                i++;
+            }
+            if (count != 16) return false;
+            static constexpr int order[16] = {3, 2, 1, 0, 5, 4, 7, 6,
+                                              8, 9, 10, 11, 12, 13, 14, 15};
+            for (int i = 0; i < 16; i++) out[i] = bytes[order[i]];
+            return true;
+        }
+
+        static bool SearchDevice(int32_t device, const uint8_t guid[16], Block& out) {
+            auto& interp = AML::GetInterpreter();
+            AML::Object wdg{};
+            if (!interp.EvaluateChild(device, "_WDG", nullptr, 0, wdg)
+                || wdg.Type != AML::ObjectType::Buffer)
+                return false;
+
+            // The interpreter caps buffers; a _WDG that fills the cap may have
+            // lost entries past it.
+            if (wdg.Buffer.Length >= (uint32_t)AML::MaxBufferLen) {
+                char path[128];
+                interp.GetNamespace().GetNodePath(device, path, sizeof(path));
+                KernelLogStream(WARNING, "WMI") << path << "._WDG may be truncated ("
+                    << base::dec << (uint64_t)wdg.Buffer.Length << " bytes read)";
+            }
+
+            for (uint32_t off = 0; off + WdgEntrySize <= wdg.Buffer.Length; off += WdgEntrySize) {
+                const uint8_t* entry = wdg.Buffer.Data + off;
+                if (memcmp(entry, guid, 16) != 0) continue;
+                out.Device = device;
+                out.ObjectId[0] = (char)entry[16];
+                out.ObjectId[1] = (char)entry[17];
+                out.InstanceCount = entry[18];
+                out.Flags = entry[19];
+                return true;
+            }
+            return false;
+        }
+
+        bool Find(const uint8_t guid[16], Block& out) {
+            auto& interp = AML::GetInterpreter();
+            if (!interp.IsInitialized()) return false;
+            auto& ns = interp.GetNamespace();
+
+            bool found = false;
+            ns.WalkDescendants(ns.RootIndex(), AML::ObjectType::Device,
+                [&](int32_t idx, const AML::NamespaceNode*) {
+                    if (found) return;
+                    char hid[16];
+                    if (!AcpiDevices::EvaluateHid(idx, hid, sizeof(hid))) return;
+                    if (memcmp(hid, HidWmi, sizeof("PNP0C14")) != 0) return;
+                    if (!(AcpiDevices::EvaluateSta(idx) & AcpiDevices::STA_PRESENT)) return;
+                    found = SearchDevice(idx, guid, out);
+                });
+            return found;
+        }
+
+        bool EvaluateMethod(const Block& block, uint8_t instance, uint32_t methodId,
+                            const uint8_t* input, uint32_t inputLength,
+                            AML::Object& result) {
+            if (!(block.Flags & FlagMethod) || instance >= block.InstanceCount)
+                return false;
+
+            AML::Object args[3];
+            args[0].Type = AML::ObjectType::Integer;
+            args[0].Integer = instance;
+            args[1].Type = AML::ObjectType::Integer;
+            args[1].Integer = methodId;
+            if (block.Flags & FlagString) {
+                uint32_t n = inputLength;
+                if (n >= (uint32_t)AML::MaxStringLen) n = AML::MaxStringLen - 1;
+                args[2].Type = AML::ObjectType::String;
+                if (n) memcpy(args[2].String.Data, input, n);
+                args[2].String.Data[n] = '\0';
+                args[2].String.Length = (uint16_t)n;
+            } else {
+                uint32_t n = inputLength;
+                if (n > (uint32_t)AML::MaxBufferLen) n = AML::MaxBufferLen;
+                args[2].Type = AML::ObjectType::Buffer;
+                if (n) memcpy(args[2].Buffer.Data, input, n);
+                args[2].Buffer.Length = n;
+            }
+
+            char method[5] = {'W', 'M', block.ObjectId[0], block.ObjectId[1], '\0'};
+            return AML::GetInterpreter().EvaluateChild(block.Device, method, args, 3, result);
+        }
+
+    };
+};

kernel/src/ACPI/AcpiWmi.hpp

@@ -0,0 +1,44 @@
+/*
+    * AcpiWmi.hpp
+    * ACPI-WMI mapper (PNP0C14): find a WMI GUID's data block in _WDG and call
+    * its WMxx method
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+#include <ACPI/AML/AmlNamespace.hpp>
+
+namespace Hal {
+    namespace AcpiWmi {
+
+        // _WDG flags (Microsoft "Windows Instrumentation: WMI and ACPI").
+        static constexpr uint8_t FlagExpensive = 0x1;
+        static constexpr uint8_t FlagMethod    = 0x2;
+        static constexpr uint8_t FlagString    = 0x4;
+        static constexpr uint8_t FlagEvent     = 0x8;
+
+        // One _WDG entry, bound to the PNP0C14 device that declared it.
+        struct Block {
+            int32_t Device;          // namespace node of the PNP0C14 device
+            char    ObjectId[2];     // names the WQxx / WSxx / WMxx methods
+            uint8_t InstanceCount;
+            uint8_t Flags;
+        };
+
+        // Parse "603E9613-EF25-4338-A3D0-C46177516DB7" into the byte order
+        // _WDG stores (first three fields little-endian). False if malformed.
+        bool ParseGuid(const char* text, uint8_t out[16]);
+
+        // Search every present PNP0C14 device's _WDG for `guid`. The first
+        // declaration wins, as on Linux.
+        bool Find(const uint8_t guid[16], Block& out);
+
+        // Evaluate WMxx(instance, methodId, input) on a method block. The input
+        // is passed as a String instead of a Buffer when the block says so.
+        bool EvaluateMethod(const Block& block, uint8_t instance, uint32_t methodId,
+                            const uint8_t* input, uint32_t inputLength,
+                            AML::Object& result);
+
+    };
+};

kernel/src/ACPI/Battery.cpp

@@ -0,0 +1,422 @@
+/*
+    * Battery.cpp
+    * ACPI Control Method Battery: _STA / _BIF / _BIX / _BST, plus _PSR on the
+    * AC adapter
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#include "Battery.hpp"
+#include "AcpiDevices.hpp"
+#include <ACPI/AML/AmlInterpreter.hpp>
+#include <CppLib/Spinlock.hpp>
+#include <Timekeeping/Timer.hpp>
+#include <Terminal/Terminal.hpp>
+#include <CppLib/Stream.hpp>
+
+using namespace Kt;
+
+namespace Hal {
+    namespace Battery {
+
+        // ============================================================================
+
+        // State
+
+        // ============================================================================
+        // ACPI reports an unreadable field as all-ones rather than zero.
+        static constexpr uint32_t Unknown32 = 0xFFFFFFFFu;
+
+        // The EC transactions behind _BST take a few milliseconds, and the
+        // value only moves on a scale of minutes, so a short cache keeps a
+        // one-second panel repaint from becoming a bus workload.
+        static constexpr uint64_t CacheMs = 2000;
+
+        static bool         g_initialized = false;
+        static bool         g_haveBattery = false;
+        static bool         g_haveAdapter = false;
+        static char         g_batteryPath[128];
+        static char         g_adapterPath[128];
+        static int32_t      g_batteryNode = -1;
+
+        static kcp::Mutex   g_lock;
+
+        // Set for the boot-path probe only, so the per-method tracing does not
+        // repeat on every poll once the system is up.
+        static bool         g_tracing = false;
+
+        static bool Refresh(Status& out);
+        static void Probe();
+        static Status       g_cached{};
+        static bool         g_cacheValid = false;
+        static uint64_t     g_cacheStamp = 0;
+
+        // ============================================================================
+
+        // Path helpers
+
+        // ============================================================================
+        static void AppendMethod(const char* devicePath, const char* method, char* out, int maxLen) {
+            int i = 0;
+            while (devicePath[i] && i < maxLen - 6) {
+                out[i] = devicePath[i];
+                i++;
+            }
+            // The root scope is written "\" with no separator before the first
+            // segment; every other path takes a dot.
+            if (!(i == 1 && out[0] == '\\'))
+                out[i++] = '.';
+            for (int j = 0; j < 4 && i < maxLen - 1; j++)
+                out[i++] = method[j];
+            out[i] = '\0';
+        }
+
+        static void CopyPath(char* dst, const char* src, int maxLen) {
+            int i = 0;
+            while (src[i] && i < maxLen - 1) {
+                dst[i] = src[i];
+                i++;
+            }
+            dst[i] = '\0';
+        }
+
+        // ============================================================================
+
+        // Initialize
+
+        // ============================================================================
+        void Initialize() {
+            if (g_initialized) return;
+            g_initialized = true;
+
+            auto& interp = AML::GetInterpreter();
+            if (!interp.IsInitialized()) return;
+
+            // Searched by HID rather than off an enumerated list: a laptop
+            // namespace runs well past the list's fixed size, and the battery
+            // sits deep under the EC where it would be cut off.
+            AcpiDevices::DeviceInfo bat{};
+            if (AcpiDevices::FindDeviceByHid(AcpiDevices::HID_BATTERY, bat)) {
+                CopyPath(g_batteryPath, bat.Path, sizeof(g_batteryPath));
+                g_batteryNode = bat.NodeIndex;
+                g_haveBattery = true;
+            }
+
+            AcpiDevices::DeviceInfo ac{};
+            if (AcpiDevices::FindDeviceByHid(AcpiDevices::HID_AC_ADAPTER, ac)) {
+                CopyPath(g_adapterPath, ac.Path, sizeof(g_adapterPath));
+                g_haveAdapter = true;
+            }
+
+            if (!g_haveBattery) {
+                // Distinguish "this machine has no battery" from "the namespace
+                // walk found nothing at all", which would point at the walk
+                // rather than the hardware.
+                auto& ns = interp.GetNamespace();
+                int deviceCount = 0;
+                ns.WalkDescendants(ns.RootIndex(), AML::ObjectType::Device,
+                    [&](int32_t, const AML::NamespaceNode*) { deviceCount++; });
+
+                KernelLogStream(INFO, "Battery") << "No ACPI battery device (PNP0C0A) among "
+                    << base::dec << (uint64_t)deviceCount << " ACPI devices";
+                return;
+            }
+
+            KernelLogStream(OK, "Battery") << "ACPI battery at " << g_batteryPath
+                << (g_haveAdapter ? " (AC adapter present)" : " (no AC adapter device)");
+
+            Probe();
+        }
+
+        // Evaluate the battery methods once, here on the boot path, announcing
+        // each before it runs. These are the first firmware methods this kernel
+        // executes that reach the embedded controller, and if one of them takes
+        // the machine down the last line printed says which -- far better than
+        // discovering it when the desktop makes its first query.
+        static void Probe() {
+            KernelLogStream(INFO, "Battery") << "Probing battery methods";
+
+            g_tracing = true;
+            Status s{};
+            bool ok = Refresh(s);
+            g_tracing = false;
+
+            if (!ok) {
+                KernelLogStream(WARNING, "Battery") << "Probe failed; battery status unavailable";
+                g_haveBattery = false;
+                return;
+            }
+
+            g_cached = s;
+            g_cacheStamp = 0;      // let the first real query refresh
+            g_cacheValid = true;
+
+            if (!s.Present) {
+                KernelLogStream(INFO, "Battery") << "Battery slot empty"
+                    << (s.AcOnline ? " (on AC)" : "");
+                return;
+            }
+
+            KernelLogStream(OK, "Battery") << "Battery at " << base::dec
+                << (uint64_t)s.Percent << "% ("
+                << (s.Charging ? "charging" : (s.Discharging ? "discharging" : "idle"))
+                << ", " << (uint64_t)s.RemainingCapacity << " of "
+                << (uint64_t)s.FullCapacity << (s.PowerUnitMilliAmps ? " mAh)" : " mWh)");
+        }
+
+        bool IsPresent() {
+            return g_haveBattery;
+        }
+
+        // ============================================================================
+
+        // Method evaluation
+
+        // ============================================================================
+        static const char* TypeName(AML::ObjectType t) {
+            switch (t) {
+                case AML::ObjectType::None:            return "none";
+                case AML::ObjectType::Integer:         return "integer";
+                case AML::ObjectType::String:          return "string";
+                case AML::ObjectType::Buffer:          return "buffer";
+                case AML::ObjectType::Package:         return "package";
+                case AML::ObjectType::Device:          return "device";
+                case AML::ObjectType::Method:          return "method";
+                case AML::ObjectType::OperationRegion: return "opregion";
+                case AML::ObjectType::Field:           return "field";
+                default:                               return "other";
+            }
+        }
+
+        static bool EvalPackage(const char* devicePath, const char* method, AML::Object& out) {
+            char path[160];
+            AppendMethod(devicePath, method, path, sizeof(path));
+
+            auto& interp = AML::GetInterpreter();
+
+            // Separate "the method is not there" from "it ran and gave us
+            // something that is not a package": those need opposite fixes, and
+            // the failure looks identical from the caller.
+            int32_t node = interp.GetNamespace().FindNode(path);
+            if (node < 0) {
+                if (g_tracing)
+                    KernelLogStream(INFO, "Battery") << "  " << path << ": absent";
+                return false;
+            }
+
+            // Trace the opcodes of the one method whose behaviour we cannot
+            // otherwise explain: it runs to completion, trips no limit, and
+            // still yields no value.
+            bool traceThis = g_tracing && method[1] == 'B';
+            if (traceThis) interp.SetTrace(120);
+
+            bool ok = interp.EvaluateObject(path, out);
+
+            if (traceThis) interp.SetTrace(0);
+
+            if (g_tracing) {
+                KernelLogStream(INFO, "Battery") << "  " << path << ": "
+                    << (ok ? "ok" : "FAILED") << ", returned " << TypeName(out.Type)
+                    << " with " << base::dec
+                    << (uint64_t)interp.GetPackageCount(out) << " element(s)";
+            }
+
+            return ok && out.Type == AML::ObjectType::Package;
+        }
+
+        static uint32_t ElementInteger(const AML::Object& pkg, int index) {
+            AML::Object elem{};
+            if (!AML::GetInterpreter().GetPackageElement(pkg, index, elem)) return Unknown32;
+            if (elem.Type != AML::ObjectType::Integer) return Unknown32;
+            return (uint32_t)elem.Integer;
+        }
+
+        // _BIX supersedes _BIF and prepends a Revision field, shifting every
+        // shared field by one. Both are read into the same three numbers.
+        struct BatteryInfo {
+            bool     Valid;
+            bool     PowerUnitMilliAmps;
+            uint32_t DesignCapacity;
+            uint32_t FullCapacity;
+        };
+
+        static bool ReadInfo(BatteryInfo& info) {
+            info = BatteryInfo{};
+
+            AML::Object pkg{};
+            int base = 0;
+
+            if (EvalPackage(g_batteryPath, "_BIX", pkg)) {
+                // _BIX revision 0 starts with Revision; later revisions keep it
+                // in the same place, so the offset is fixed at one.
+                base = 1;
+            } else if (EvalPackage(g_batteryPath, "_BIF", pkg)) {
+                base = 0;
+            } else {
+                return false;
+            }
+
+            uint32_t unit    = ElementInteger(pkg, base + 0);
+            uint32_t design  = ElementInteger(pkg, base + 1);
+            uint32_t full    = ElementInteger(pkg, base + 2);
+
+            info.PowerUnitMilliAmps = (unit != Unknown32) && (unit & 1);
+            info.DesignCapacity = (design == Unknown32) ? 0 : design;
+            info.FullCapacity   = (full   == Unknown32) ? 0 : full;
+            info.Valid = true;
+            return true;
+        }
+
+        static bool AcOnline() {
+            if (!g_haveAdapter) return false;
+
+            char path[160];
+            AppendMethod(g_adapterPath, "_PSR", path, sizeof(path));
+
+            if (g_tracing)
+                KernelLogStream(INFO, "Battery") << "  evaluating " << path;
+
+            AML::Object result{};
+            if (!AML::GetInterpreter().EvaluateObject(path, result)) return false;
+            return result.Type == AML::ObjectType::Integer && result.Integer != 0;
+        }
+
+        // ============================================================================
+
+        // GetStatus
+
+        // ============================================================================
+        static bool Refresh(Status& out) {
+            out = Status{};
+            out.MinutesRemaining = -1;
+            out.HasAcAdapter = g_haveAdapter;
+
+            // _STA bit 4 says whether a battery is actually in the slot. A
+            // laptop with the pack removed still has the device.
+            uint32_t sta = (g_batteryNode >= 0)
+                ? AcpiDevices::EvaluateSta(g_batteryNode)
+                : AcpiDevices::STA_DEFAULT;
+
+            // Bit 4 is the one that decides everything below, and firmware
+            // clears it for reasons that have nothing to do with the pack
+            // (an EC that has not been told its region handler is live, most
+            // often), so log the raw value rather than just the verdict.
+            if (g_tracing) {
+                KernelLogStream(INFO, "Battery") << "  _STA = " << base::hex
+                    << (uint64_t)sta << (((sta & AcpiDevices::STA_BATTERY) != 0)
+                        ? " (battery present)" : " (battery bit clear)");
+            }
+
+            out.AcOnline = AcOnline();
+
+            // _STA bit 4 is the spec's answer for "is a pack installed", but
+            // it is only as good as the firmware path behind it. Read the
+            // battery methods either way and let the capacity data have the
+            // final say -- see the staSaysAbsent handling below.
+            bool staSaysAbsent = (sta & AcpiDevices::STA_BATTERY) == 0;
+
+            BatteryInfo info{};
+            if (!ReadInfo(info)) {
+                if (g_tracing)
+                    KernelLogStream(INFO, "Battery") << "  no _BIX/_BIF package";
+                return true;
+            }
+
+            AML::Object bst{};
+            if (!EvalPackage(g_batteryPath, "_BST", bst)) {
+                if (g_tracing)
+                    KernelLogStream(INFO, "Battery") << "  no _BST package";
+                return true;
+            }
+
+            uint32_t state     = ElementInteger(bst, 0);
+            uint32_t rate      = ElementInteger(bst, 1);
+            uint32_t remaining = ElementInteger(bst, 2);
+            uint32_t voltage   = ElementInteger(bst, 3);
+
+            if (g_tracing) {
+                KernelLogStream(INFO, "Battery") << "  _BIF/_BIX: design="
+                    << base::dec << (uint64_t)info.DesignCapacity
+                    << " lastFull=" << (uint64_t)info.FullCapacity
+                    << (info.PowerUnitMilliAmps ? " (mAh)" : " (mWh)");
+                KernelLogStream(INFO, "Battery") << "  _BST: state=" << base::hex
+                    << (uint64_t)state << " rate=" << base::dec << (uint64_t)rate
+                    << " remaining=" << (uint64_t)remaining
+                    << " voltage=" << (uint64_t)voltage << "mV";
+            }
+
+            // Trust real capacity data over a status bit that says otherwise:
+            // firmware that reports a pack's charge is describing a pack that
+            // is there.
+            if (staSaysAbsent) {
+                bool looksReal = info.FullCapacity > 0 && info.FullCapacity != Unknown32
+                              && remaining != Unknown32;
+                if (!looksReal) {
+                    if (g_tracing)
+                        KernelLogStream(INFO, "Battery")
+                            << "  _STA says absent and the methods agree";
+                    return true;
+                }
+                if (g_tracing)
+                    KernelLogStream(WARNING, "Battery")
+                        << "  _STA says absent but _BST reports charge; trusting _BST";
+            }
+
+            if (state == Unknown32) state = 0;
+
+            out.Present     = true;
+            out.Charging    = (state & StateCharging) != 0;
+            out.Discharging = (state & StateDischarging) != 0;
+            out.Critical    = (state & StateCritical) != 0;
+
+            out.PowerUnitMilliAmps = info.PowerUnitMilliAmps;
+            out.DesignCapacity     = info.DesignCapacity;
+            out.FullCapacity       = info.FullCapacity;
+            out.RemainingCapacity  = (remaining == Unknown32) ? 0 : remaining;
+            out.PresentRate        = (rate == Unknown32) ? 0 : rate;
+            out.VoltageMv          = (voltage == Unknown32) ? 0 : voltage;
+
+            // Some firmware leaves last-full at zero or below the current
+            // reading after a swap; fall back to the design capacity, and
+            // clamp so the panel can never show 143%.
+            uint32_t full = out.FullCapacity;
+            if (full == 0) full = out.DesignCapacity;
+
+            if (full > 0 && out.RemainingCapacity > 0) {
+                uint64_t pct = ((uint64_t)out.RemainingCapacity * 100) / full;
+                out.Percent = (pct > 100) ? 100 : (uint8_t)pct;
+            }
+
+            if (out.PresentRate > 0 && out.Discharging) {
+                out.MinutesRemaining =
+                    (int32_t)(((uint64_t)out.RemainingCapacity * 60) / out.PresentRate);
+            }
+
+            return true;
+        }
+
+        bool GetStatus(Status& out) {
+            if (!g_haveBattery) return false;
+
+            g_lock.Acquire();
+
+            uint64_t now = Timekeeping::GetMilliseconds();
+            if (g_cacheValid && now - g_cacheStamp < CacheMs) {
+                out = g_cached;
+                g_lock.Release();
+                return true;
+            }
+
+            Status fresh{};
+            bool ok = Refresh(fresh);
+            if (ok) {
+                g_cached = fresh;
+                g_cacheStamp = now;
+                g_cacheValid = true;
+            }
+
+            out = ok ? fresh : g_cached;
+            g_lock.Release();
+            return ok || g_cacheValid;
+        }
+    };
+};

kernel/src/ACPI/Battery.hpp

@@ -0,0 +1,62 @@
+/*
+    * Battery.hpp
+    * ACPI Control Method Battery (PNP0C0A) and AC adapter (ACPI0003)
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+
+namespace Hal {
+    namespace Battery {
+
+        // ============================================================================
+
+        // Battery state
+
+        // ============================================================================
+        // _BST battery state bits.
+        static constexpr uint32_t StateDischarging = (1 << 0);
+        static constexpr uint32_t StateCharging    = (1 << 1);
+        static constexpr uint32_t StateCritical    = (1 << 2);
+
+        struct Status {
+            bool     Present;         // a battery slot exists and reports a battery in it
+            bool     Charging;
+            bool     Discharging;
+            bool     Critical;
+            bool     AcOnline;        // AC adapter connected (false when no adapter device)
+            bool     HasAcAdapter;    // an ACPI0003 device exists
+
+            uint8_t  Percent;         // 0..100, clamped; 0 when capacity is unknown
+            uint32_t RemainingCapacity;  // mWh or mAh, per PowerUnitMilliAmps
+            uint32_t FullCapacity;       // last full charge capacity, same unit
+            uint32_t DesignCapacity;
+            uint32_t PresentRate;        // mW or mA, 0 when unknown
+            uint32_t VoltageMv;          // present voltage in mV, 0 when unknown
+            bool     PowerUnitMilliAmps; // capacities are mAh/mA rather than mWh/mW
+
+            // Minutes of runtime left at the present rate, or -1 when the rate
+            // is unknown/zero (which includes every charging battery).
+            int32_t  MinutesRemaining;
+        };
+
+        // ============================================================================
+
+        // API
+
+        // ============================================================================
+        // Discover battery and AC adapter devices in the ACPI namespace. Safe to
+        // call more than once; only the first call does work.
+        void Initialize();
+
+        // True when at least one battery device was found, whether or not a
+        // battery is currently installed in it.
+        bool IsPresent();
+
+        // Read the current status. Results are cached briefly so that polling
+        // from a panel indicator does not hammer the embedded controller.
+        // Returns false when the machine has no battery device at all.
+        bool GetStatus(Status& out);
+    };
+};

kernel/src/ACPI/EmbeddedController.cpp

@@ -0,0 +1,326 @@
+/*
+    * EmbeddedController.cpp
+    * ACPI Embedded Controller transport (polled, no burst mode)
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#include "EmbeddedController.hpp"
+#include "ACPI.hpp"
+#include "AcpiDevices.hpp"
+#include <ACPI/AML/AmlInterpreter.hpp>
+#include <CppLib/Spinlock.hpp>
+#include <Io/IoPort.hpp>
+#include <Memory/HHDM.hpp>
+#include <Terminal/Terminal.hpp>
+#include <CppLib/Stream.hpp>
+
+using namespace Kt;
+
+namespace Hal {
+    namespace EmbeddedController {
+
+        // ============================================================================
+
+        // EC register protocol
+
+        // ============================================================================
+        // EC_SC (status, read from the command port)
+        static constexpr uint8_t SC_OBF  = (1 << 0); // output buffer full: data is readable
+        static constexpr uint8_t SC_IBF  = (1 << 1); // input buffer full: EC has not consumed our byte
+
+        // EC_SC commands (written to the command port)
+        static constexpr uint8_t CMD_READ  = 0x80;
+        static constexpr uint8_t CMD_WRITE = 0x81;
+
+        // The EC is a slow microcontroller; a byte transaction is typically a
+        // few hundred microseconds but firmware is allowed to take much longer.
+        // One "tick" is a port-wait (~1us), so this caps a wait near 500ms.
+        static constexpr uint32_t PollTicks = 500000;
+
+        // ============================================================================
+
+        // State
+
+        // ============================================================================
+        static uint16_t       g_dataPort = 0;
+        static uint16_t       g_cmdPort  = 0;
+        static bool           g_available = false;
+        static bool           g_initialized = false;
+        // A Mutex, not a Spinlock: EC transactions run only from process and
+        // syscall context, and can spin for milliseconds waiting on the
+        // controller. A Spinlock would hold interrupts off for all of it.
+        static kcp::Mutex     g_lock;
+
+        // ============================================================================
+
+        // ECDT
+
+        // ============================================================================
+        // The ECDT gives the EC's ports without needing the AML namespace. It is
+        // optional, and some firmware ships one with zeroed ports, so treat a
+        // zero port as "not described here" and fall through to the namespace.
+        struct EcdtTable {
+            ACPI::CommonSDTHeader Header;
+            uint8_t  ControlGas[12];
+            uint8_t  DataGas[12];
+            uint32_t Uid;
+            uint8_t  GpeBit;
+            // followed by a null-terminated namespace path
+        } __attribute__((packed));
+
+        static bool SignatureMatch(const char* sig, const char* target) {
+            for (int i = 0; i < 4; i++) {
+                if (sig[i] != target[i]) return false;
+            }
+            return true;
+        }
+
+        // Decode a Generic Address Structure that must describe a 16-bit I/O port.
+        static uint16_t GasIoPort(const uint8_t gas[12]) {
+            if (gas[0] != 1) return 0; // AddressSpaceId 1 == SystemIO
+
+            uint64_t address = 0;
+            for (int i = 0; i < 8; i++)
+                address |= (uint64_t)gas[4 + i] << (i * 8);
+            return address <= 0xFFFF ? (uint16_t)address : 0;
+        }
+
+        static bool FromEcdt(ACPI::CommonSDTHeader* xsdt) {
+            if (xsdt == nullptr) return false;
+
+            uint32_t entryCount = (xsdt->Length - sizeof(ACPI::CommonSDTHeader)) / 8;
+            uint64_t* entries = (uint64_t*)((uint64_t)xsdt + sizeof(ACPI::CommonSDTHeader));
+
+            for (uint32_t i = 0; i < entryCount; i++) {
+                auto* header = (ACPI::CommonSDTHeader*)Memory::HHDM(entries[i]);
+                if (!SignatureMatch(header->Signature, "ECDT")) continue;
+                if (header->Length < sizeof(EcdtTable)) continue;
+
+                auto* ecdt = (EcdtTable*)header;
+                uint16_t cmd  = GasIoPort(ecdt->ControlGas);
+                uint16_t data = GasIoPort(ecdt->DataGas);
+                if (cmd == 0 || data == 0) continue;
+
+                g_cmdPort  = cmd;
+                g_dataPort = data;
+                return true;
+            }
+
+            return false;
+        }
+
+        // ============================================================================
+
+        // Namespace fallback
+
+        // ============================================================================
+        // _CRS on an EC device lists two I/O resources in a fixed order: data
+        // register first, then command/status. Anything else is not an EC we
+        // know how to drive.
+        static bool FromNamespace() {
+            if (!AML::GetInterpreter().IsInitialized()) return false;
+
+            AcpiDevices::DeviceInfo ec{};
+            if (!AcpiDevices::FindDeviceByHid(AcpiDevices::HID_EC, ec)) return false;
+
+            AML::ResourceList crs{};
+            if (!AcpiDevices::EvaluateCrs(ec.NodeIndex, crs)) return false;
+
+            uint16_t ports[2] = {0, 0};
+            int found = 0;
+            for (int i = 0; i < crs.Count && found < 2; i++) {
+                if (crs.Resources[i].Type == AML::ResourceType::IoPort)
+                    ports[found++] = crs.Resources[i].IoPort.Base;
+                else if (crs.Resources[i].Type == AML::ResourceType::FixedIoPort)
+                    ports[found++] = crs.Resources[i].FixedIoPort.Base;
+            }
+
+            if (found < 2 || ports[0] == 0 || ports[1] == 0) return false;
+
+            g_dataPort = ports[0];
+            g_cmdPort  = ports[1];
+            return true;
+        }
+
+        // ============================================================================
+
+        // Initialize
+
+        // ============================================================================
+        void Initialize(void* xsdt) {
+            if (g_initialized) return;
+            g_initialized = true;
+
+            if (!FromEcdt((ACPI::CommonSDTHeader*)xsdt) && !FromNamespace()) {
+                KernelLogStream(INFO, "EC") << "No ACPI Embedded Controller present";
+                return;
+            }
+
+            g_available = true;
+            KernelLogStream(OK, "EC") << "Embedded Controller at cmd=" << base::hex
+                << (uint64_t)g_cmdPort << " data=" << (uint64_t)g_dataPort;
+
+            RunRegMethods();
+
+            KernelLogStream(INFO, "EC") << "EC space dump:";
+            DumpSpace(0x00, 32);
+        }
+
+        // ============================================================================
+
+        // _REG
+
+        // ============================================================================
+        void RunRegMethods() {
+            auto& interp = AML::GetInterpreter();
+            if (!interp.IsInitialized()) return;
+
+            auto& ns = interp.GetNamespace();
+
+            // Collect first, evaluate second. Executing AML can create namespace
+            // nodes, which reallocates a parent's child array -- walking the
+            // tree while methods run would be iterating a freed array.
+            static constexpr int MaxRegMethods = 64;
+            int32_t nodes[MaxRegMethods];
+            int count = 0;
+
+            ns.WalkDescendants(ns.RootIndex(), AML::ObjectType::Method,
+                [&](int32_t idx, const AML::NamespaceNode* node) {
+                    if (count >= MaxRegMethods) return;
+                    if (node->Name[0] == '_' && node->Name[1] == 'R' &&
+                        node->Name[2] == 'E' && node->Name[3] == 'G')
+                        nodes[count++] = idx;
+                });
+
+            // Arg0 = region space (3, EmbeddedControl), Arg1 = 1 (handler is
+            // connected). A _REG that handles some other space checks Arg0 and
+            // ignores the call, so this is safe to issue namespace-wide.
+            AML::Object args[2]{};
+            args[0].Type = AML::ObjectType::Integer;
+            args[0].Integer = (uint64_t)AML::RegionSpace::EmbeddedControl;
+            args[1].Type = AML::ObjectType::Integer;
+            args[1].Integer = 1;
+
+            int ran = 0;
+            for (int i = 0; i < count; i++) {
+                char path[256];
+                ns.GetNodePath(nodes[i], path, sizeof(path));
+
+                AML::Object result{};
+                bool ok = interp.EvaluateMethod(path, args, 2, result);
+                if (ok) ran++;
+
+                KernelLogStream(INFO, "EC") << "  _REG " << path
+                    << (ok ? " ok" : " FAILED");
+            }
+
+            KernelLogStream(OK, "EC") << "Ran " << base::dec << (uint64_t)ran
+                << " of " << (uint64_t)count << " _REG method(s)";
+        }
+
+        // ============================================================================
+
+        // Diagnostics
+
+        // ============================================================================
+        void DumpSpace(uint8_t base_, int bytes) {
+            if (!g_available) return;
+
+            // Whether the transport works at all is not otherwise observable:
+            // a failed read and a genuinely zero register look identical by the
+            // time firmware AML has turned them into a status value. All-00 or
+            // all-FF here means the handshake is wrong, not that the battery is
+            // missing.
+            for (int row = 0; row < bytes; row += 16) {
+                char line[64];
+                int n = 0;
+                bool anyFailed = false;
+
+                for (int i = 0; i < 16 && (row + i) < bytes; i++) {
+                    uint8_t v = 0;
+                    if (!Read((uint8_t)(base_ + row + i), v)) anyFailed = true;
+
+                    static const char hex[] = "0123456789ABCDEF";
+                    line[n++] = hex[(v >> 4) & 0xF];
+                    line[n++] = hex[v & 0xF];
+                    line[n++] = ' ';
+                }
+                line[n] = '\0';
+
+                KernelLogStream(INFO, "EC") << "  " << base::hex
+                    << (uint64_t)(base_ + row) << ": " << line
+                    << (anyFailed ? " (read timeout)" : "");
+            }
+        }
+
+        bool IsAvailable() {
+            return g_available;
+        }
+
+        // ============================================================================
+
+        // Transport
+
+        // ============================================================================
+        // Spin until the named status bit reaches the wanted state.
+        static bool WaitStatus(uint8_t mask, bool wantSet) {
+            for (uint32_t i = 0; i < PollTicks; i++) {
+                uint8_t sc = Io::In8(g_cmdPort);
+                if (((sc & mask) != 0) == wantSet) return true;
+                Io::IoPortWait();
+            }
+            return false;
+        }
+
+        // Drain a byte the EC left in the output buffer from an earlier aborted
+        // transaction. Without this the next read returns the stale value.
+        static void DrainOutput() {
+            if (Io::In8(g_cmdPort) & SC_OBF)
+                (void)Io::In8(g_dataPort);
+        }
+
+        bool Read(uint8_t address, uint8_t& value) {
+            if (!g_available) return false;
+
+            g_lock.Acquire();
+            DrainOutput();
+
+            bool ok = false;
+            if (WaitStatus(SC_IBF, false)) {
+                Io::Out8(CMD_READ, g_cmdPort);
+                if (WaitStatus(SC_IBF, false)) {
+                    Io::Out8(address, g_dataPort);
+                    if (WaitStatus(SC_OBF, true)) {
+                        value = Io::In8(g_dataPort);
+                        ok = true;
+                    }
+                }
+            }
+
+            g_lock.Release();
+            return ok;
+        }
+
+        bool Write(uint8_t address, uint8_t value) {
+            if (!g_available) return false;
+
+            g_lock.Acquire();
+            DrainOutput();
+
+            bool ok = false;
+            if (WaitStatus(SC_IBF, false)) {
+                Io::Out8(CMD_WRITE, g_cmdPort);
+                if (WaitStatus(SC_IBF, false)) {
+                    Io::Out8(address, g_dataPort);
+                    if (WaitStatus(SC_IBF, false)) {
+                        Io::Out8(value, g_dataPort);
+                        ok = WaitStatus(SC_IBF, false);
+                    }
+                }
+            }
+
+            g_lock.Release();
+            return ok;
+        }
+    };
+};

kernel/src/ACPI/EmbeddedController.hpp

@@ -0,0 +1,50 @@
+/*
+    * EmbeddedController.hpp
+    * ACPI Embedded Controller (PNP0C09) transport
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+
+namespace Hal {
+    namespace EmbeddedController {
+
+        // ============================================================================
+
+        // Bring-up
+
+        // ============================================================================
+        // Locate the EC in the ACPI namespace and latch its command/data ports.
+        // Prefers the ECDT table when present (it is valid before the namespace
+        // is loaded); otherwise walks the namespace for PNP0C09 and reads _CRS.
+        // Safe to call more than once; only the first call does work.
+        void Initialize(void* xsdt);
+
+        // True once a usable command/data port pair has been found.
+        bool IsAvailable();
+
+        // Run the namespace's _REG(EmbeddedControl, 1) methods, telling firmware
+        // that EC region access is live. ACPI requires this once a handler for
+        // an address space is installed, and firmware leans on it hard: EC
+        // methods are typically guarded by a flag that _REG sets, and until it
+        // is set they return safe do-nothing answers (a battery reports itself
+        // absent, a thermal zone reports no temperature). Called by Initialize.
+        void RunRegMethods();
+
+        // Log `bytes` bytes of EC address space starting at `base_`. Purely a
+        // diagnostic: it is the only way to tell a broken handshake from a
+        // register that genuinely reads zero.
+        void DumpSpace(uint8_t base_, int bytes);
+
+        // ============================================================================
+
+        // Transport
+
+        // ============================================================================
+        // Read/write one byte of the EC's 256-byte address space. Both return
+        // false on timeout or when no EC is present, leaving *value untouched.
+        bool Read(uint8_t address, uint8_t& value);
+        bool Write(uint8_t address, uint8_t value);
+    };
+};

kernel/src/ACPI/FADT.cpp

@@ -107,6 +107,8 @@ namespace Hal {
             result.AcpiEnable       = fadt->AcpiEnable;
             result.AcpiDisable      = fadt->AcpiDisable;
             result.Flags            = fadt->Flags;
+            result.ArmBootArchFlags = fadt->Header.Length >= offsetof(Table, ArmBootArchFlags) + 2
+                                    ? fadt->ArmBootArchFlags : 0;
             result.Valid            = true;
 
             KernelLogStream(INFO, "FADT") << "DSDT at " << base::hex << result.DsdtAddress;

kernel/src/ACPI/FADT.hpp

@@ -69,6 +69,13 @@ namespace Hal {
             uint8_t  X_GPE1Block[12];
         } __attribute__((packed));
 
+        // FADT Flags
+        constexpr uint32_t FlagHardwareReduced = 1u << 20;
+
+        // ARM_BOOT_ARCH flags (ACPI 5.1+)
+        constexpr uint16_t ArmBootPsciCompliant = 1u << 0;
+        constexpr uint16_t ArmBootPsciUseHvc    = 1u << 1;
+
         struct ParsedFADT {
             uint64_t DsdtAddress;
             uint64_t FacsAddress;
@@ -91,6 +98,7 @@ namespace Hal {
             uint8_t  AcpiEnable;
             uint8_t  AcpiDisable;
             uint32_t Flags;
+            uint16_t ArmBootArchFlags;  // 0 when the table predates the field
             bool     Valid;
         };
 

kernel/src/ACPI/GsiRouting.cpp

@@ -0,0 +1,41 @@
+/*
+    * GsiRouting.cpp
+    * Platform interrupt lines for ACPI global system interrupts
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#include "GsiRouting.hpp"
+#include <Arch/Interrupts.hpp>
+#include <CppLib/Spinlock.hpp>
+
+namespace Hal::GsiRouting {
+
+    static constexpr uint8_t FirstLine = 1;
+    static constexpr uint8_t LineCount = 24;
+
+    static uint32_t g_lineGsi[LineCount] = {};
+    static uint8_t g_nextLine = FirstLine;
+    static kcp::Mutex g_lock;
+
+    uint8_t LineForGsi(uint32_t gsi, bool levelTriggered, bool activeLow) {
+        if (Arch::PlatformLinesAreFixed()) return 0;
+
+        g_lock.Acquire();
+        for (uint8_t line = FirstLine; line < g_nextLine; line++) {
+            if (g_lineGsi[line] == gsi) {
+                g_lock.Release();
+                return line;
+            }
+        }
+
+        uint8_t line = 0;
+        if (g_nextLine < LineCount &&
+            Arch::AssignPlatformLine(g_nextLine, gsi, levelTriggered, activeLow)) {
+            line = g_nextLine++;
+            g_lineGsi[line] = gsi;
+        }
+        g_lock.Release();
+        return line;
+    }
+
+}

kernel/src/ACPI/GsiRouting.hpp

@@ -0,0 +1,19 @@
+/*
+    * GsiRouting.hpp
+    * Platform interrupt lines for ACPI global system interrupts
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+
+namespace Hal::GsiRouting {
+
+    // The platform line (1-23) that delivers global system interrupt `gsi`,
+    // assigning one on first use (Arch::AssignPlatformLine); every device on
+    // one GSI shares its line. Drivers register their handler on the line
+    // and call Arch::EnableLegacyIrq. Returns 0 where the platform's lines
+    // are wired by the hardware (x86) or when none is left.
+    uint8_t LineForGsi(uint32_t gsi, bool levelTriggered, bool activeLow);
+
+}

kernel/src/ACPI/WmiEcBacklight.cpp

@@ -0,0 +1,160 @@
+/*
+    * WmiEcBacklight.cpp
+    * Firmware (EC) panel backlight behind the WMI interface NVIDIA specified
+    * for display-mux laptops
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#include "WmiEcBacklight.hpp"
+#include "AcpiWmi.hpp"
+#include "AcpiPower.hpp"
+#include <ACPI/AML/AmlInterpreter.hpp>
+#include <CppLib/Spinlock.hpp>
+#include <CppLib/Stream.hpp>
+#include <Terminal/Terminal.hpp>
+
+using namespace Kt;
+
+namespace Hal {
+    namespace WmiEcBacklight {
+
+        // ============================================================================
+
+        // Interface
+
+        // ============================================================================
+        static constexpr const char* Guid = "603E9613-EF25-4338-A3D0-C46177516DB7";
+
+        // WMxx method IDs.
+        static constexpr uint32_t MethodLevel  = 1;
+        static constexpr uint32_t MethodSource = 2;
+
+        // `mode` values in the argument block.
+        static constexpr uint32_t ModeGet         = 0;
+        static constexpr uint32_t ModeSet         = 1;
+        static constexpr uint32_t ModeGetMaxLevel = 2;
+
+        static constexpr uint64_t SourceEc = 2;   // 1 = GPU, 3 = eDP AUX
+
+        // Lowest level driven, so the panel never goes fully dark.
+        static constexpr int MinPercent = 1;
+
+        // ============================================================================
+
+        // State
+
+        // ============================================================================
+        static kcp::Mutex      g_lock;
+        static bool            g_probed = false;
+        static bool            g_available = false;
+        static AcpiWmi::Block  g_block{};
+        static uint32_t        g_maxLevel = 0;
+
+        // Argument block: { u32 mode; u32 val; u32 ret; u32 reserved[3]; }.
+        // The method answers with an Integer. Linux reads it back as the
+        // `ret` field of the same block, which is where ACPICA's copy-out of
+        // an Integer puts the value; a firmware returning the block itself as
+        // a Buffer is read at that same offset.
+        static bool Call(uint32_t method, uint32_t mode, uint32_t value, uint32_t& out) {
+            uint8_t args[24] = {};
+            args[0] = (uint8_t)mode;
+            args[1] = (uint8_t)(mode >> 8);
+            args[2] = (uint8_t)(mode >> 16);
+            args[3] = (uint8_t)(mode >> 24);
+            args[4] = (uint8_t)value;
+            args[5] = (uint8_t)(value >> 8);
+            args[6] = (uint8_t)(value >> 16);
+            args[7] = (uint8_t)(value >> 24);
+
+            AML::Object result{};
+            if (!AcpiWmi::EvaluateMethod(g_block, 0, method, args, sizeof(args), result))
+                return false;
+            if (result.Type == AML::ObjectType::Integer) {
+                out = (uint32_t)result.Integer;
+                return true;
+            }
+            if (result.Type == AML::ObjectType::Buffer && result.Buffer.Length >= 12) {
+                const uint8_t* r = result.Buffer.Data + 8;
+                out = r[0] | (uint32_t)r[1] << 8 | (uint32_t)r[2] << 16 | (uint32_t)r[3] << 24;
+                return true;
+            }
+            // A set may legitimately return nothing.
+            out = 0;
+            return mode == ModeSet;
+        }
+
+        static void Probe() {
+            g_probed = true;
+            uint8_t guid[16];
+            if (!AcpiWmi::ParseGuid(Guid, guid) || !AcpiWmi::Find(guid, g_block)) return;
+            if (!(g_block.Flags & AcpiWmi::FlagMethod) || g_block.InstanceCount == 0) return;
+
+            char path[128];
+            AML::GetInterpreter().GetNamespace().GetNodePath(g_block.Device, path, sizeof(path));
+
+            // The interface also exists on machines where the GPU drives the
+            // backlight; there it must be left alone.
+            uint32_t source = 0;
+            if (!Call(MethodSource, ModeGet, 0, source)) {
+                KernelLogStream(INFO, "Backlight") << "WMI EC backlight at " << path
+                    << ": source query failed";
+                return;
+            }
+            if (source != SourceEc) {
+                KernelLogStream(INFO, "Backlight") << "WMI EC backlight at " << path
+                    << " reports source " << base::dec << (uint64_t)source << ", not EC";
+                return;
+            }
+
+            uint32_t maximum = 0;
+            if (!Call(MethodLevel, ModeGetMaxLevel, 0, maximum) || maximum == 0) {
+                KernelLogStream(WARNING, "Backlight") << "WMI EC backlight at " << path
+                    << ": no maximum level";
+                return;
+            }
+            g_maxLevel = maximum;
+            g_available = true;
+            char method[5] = {'W', 'M', g_block.ObjectId[0], g_block.ObjectId[1], '\0'};
+            KernelLogStream(OK, "Backlight") << "Panel brightness via EC (WMI "
+                << path << "." << method << ", " << base::dec << (uint64_t)maximum
+                << " levels)";
+        }
+
+        static int ToPercent(uint32_t level) {
+            if (level > g_maxLevel) level = g_maxLevel;
+            return (int)(((uint64_t)level * 100 + g_maxLevel / 2) / g_maxLevel);
+        }
+
+        // ============================================================================
+
+        // Public API
+
+        // ============================================================================
+        int Brightness(int percent) {
+            g_lock.Acquire();
+            if (!g_probed) Probe();
+            int result = -1;
+            if (g_available) {
+                uint32_t level = 0;
+                if (percent >= 0) {
+                    if (percent > 100) percent = 100;
+                    if (percent < MinPercent) percent = MinPercent;
+                    uint32_t target = (uint32_t)(((uint64_t)g_maxLevel * (uint32_t)percent + 50) / 100);
+                    if (target == 0) target = 1;
+                    uint32_t ignored = 0;
+                    if (!Call(MethodLevel, ModeSet, target, ignored)) {
+                        result = -2;
+                    } else {
+                        result = Call(MethodLevel, ModeGet, 0, level) ? ToPercent(level)
+                                                                      : ToPercent(target);
+                    }
+                } else {
+                    result = Call(MethodLevel, ModeGet, 0, level) ? ToPercent(level) : -2;
+                }
+            }
+            g_lock.Release();
+            return result;
+        }
+
+    };
+};

kernel/src/ACPI/WmiEcBacklight.hpp

@@ -0,0 +1,28 @@
+/*
+    * WmiEcBacklight.hpp
+    * Firmware (EC) panel backlight behind the WMI interface NVIDIA specified
+    * for display-mux laptops
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+
+namespace Hal {
+    namespace WmiEcBacklight {
+
+        // On a laptop whose panel can be switched between GPUs, firmware may
+        // wire the backlight to the embedded controller instead of either
+        // GPU and expose it through WMI GUID 603E9613-EF25-4338-A3D0-
+        // C46177516DB7. Despite the name the interface is pure firmware: it
+        // works with the discrete GPU powered off. It is used only when the
+        // firmware itself reports the EC as the brightness source, which is
+        // also when GPU PWM writes stop reaching the panel.
+        //
+        // percent < 0 queries. Returns 0..100, -1 when this interface is not
+        // the panel's brightness source, -2 when the firmware call failed.
+        // Discovery runs on first use. May sleep (AML).
+        int Brightness(int percent);
+
+    };
+};

kernel/src/Api/Audio.hpp

@@ -11,8 +11,8 @@
 #include <Drivers/Audio/IntelHda.hpp>
 #include <Drivers/Audio/Mixer.hpp>
 #include <Terminal/Terminal.hpp>
-#include <Drivers/USB/Bluetooth/Bluetooth.hpp>
+#include <Drivers/Bluetooth/Bluetooth.hpp>
-#include <Drivers/USB/Bluetooth/A2dp.hpp>
+#include <Drivers/Bluetooth/A2dp.hpp>
 #include <Sched/Scheduler.hpp>
 
 #include "Syscall.hpp"
@@ -36,12 +36,12 @@ namespace montauk::abi {
                                                        bitsPerSample, pid, name);
         }
 
-        return -1;
+        return NoDevice;
     }
 
     static int64_t Sys_AudioClose(int handle) {
         Drivers::Audio::Mixer::Close(handle);
-        return 0;
+        return Success;
     }
 
     static int64_t Sys_AudioWrite(int handle, const uint8_t* data, uint32_t size) {
@@ -55,10 +55,10 @@ namespace montauk::abi {
             return Drivers::Audio::Mixer::SetOutput(
                 (Drivers::Audio::Mixer::Output)value);
         if (cmd == AUDIO_CTL_BT_STATUS) {
-            if (!Drivers::USB::Bluetooth::IsInitialized()) return 0;
+            if (!Drivers::Bluetooth::IsInitialized()) return 0;
-            if (Drivers::USB::Bluetooth::A2dp::IsReady()) return 2;
+            if (Drivers::Bluetooth::A2dp::IsReady()) return 2;
-            return Drivers::USB::Bluetooth::A2dp::GetState()
+            return Drivers::Bluetooth::A2dp::GetState()
-                       == Drivers::USB::Bluetooth::A2dp::State::Idle
+                       == Drivers::Bluetooth::A2dp::State::Idle
                 ? 0 : 1;
         }
 

kernel/src/Api/BluetoothSyscall.hpp

@@ -6,8 +6,8 @@
 
 #pragma once
 #include <cstdint>
-#include <Drivers/USB/Bluetooth/Bluetooth.hpp>
+#include <Drivers/Bluetooth/Bluetooth.hpp>
-#include <Drivers/USB/Bluetooth/Hci.hpp>
+#include <Drivers/Bluetooth/Hci.hpp>
 #include <Libraries/Memory.hpp>
 #include <Sched/Scheduler.hpp>
 #include <Terminal/Terminal.hpp>
@@ -18,15 +18,15 @@
 namespace montauk::abi {
 
     static int64_t Sys_BtScan(BtScanResult* buf, int maxCount, uint32_t timeoutMs) {
-        if (!buf || maxCount <= 0) return -1;
+        if (!buf || maxCount <= 0) return InvalidArgument;
-        if (!Drivers::USB::Bluetooth::IsInitialized()) return -1;
+        if (!Drivers::Bluetooth::IsInitialized()) return NoDevice;
 
         // Use a stack buffer for kernel-side results (max 16)
-        Drivers::USB::Bluetooth::Hci::InquiryDevice tmpBuf[16];
+        Drivers::Bluetooth::Hci::InquiryDevice tmpBuf[16];
         int kMax = maxCount;
         if (kMax > 16) kMax = 16;
 
-        int found = Drivers::USB::Bluetooth::Scan(tmpBuf, kMax, timeoutMs);
+        int found = Drivers::Bluetooth::Scan(tmpBuf, kMax, timeoutMs);
         if (found < 0) return found;
 
         // Convert kernel InquiryDevice to userspace BtScanResult
@@ -46,39 +46,39 @@ namespace montauk::abi {
     }
 
     static int64_t Sys_BtConnect(const uint8_t* bdAddr) {
-        if (!bdAddr) return -1;
+        if (!bdAddr) return InvalidArgument;
-        if (!Drivers::USB::Bluetooth::IsInitialized()) return -1;
+        if (!Drivers::Bluetooth::IsInitialized()) return NoDevice;
 
-        return (int64_t)Drivers::USB::Bluetooth::Connect(bdAddr);
+        return (int64_t)Drivers::Bluetooth::Connect(bdAddr);
     }
 
     static int64_t Sys_BtDisconnect(const uint8_t* bdAddr) {
-        if (!bdAddr) return -1;
+        if (!bdAddr) return InvalidArgument;
-        if (!Drivers::USB::Bluetooth::IsInitialized()) return -1;
+        if (!Drivers::Bluetooth::IsInitialized()) return NoDevice;
 
         Kt::KernelLogStream(Kt::INFO, "Power") << "pid " << kcp::dec
             << (uint64_t)Sched::GetCurrentPid()
             << " entered Bluetooth disconnect stage";
-        return (int64_t)Drivers::USB::Bluetooth::Disconnect(bdAddr);
+        return (int64_t)Drivers::Bluetooth::Disconnect(bdAddr);
     }
 
     static int64_t Sys_BtSetAddr(const uint8_t* bdAddr) {
-        if (!bdAddr) return -1;
+        if (!bdAddr) return InvalidArgument;
-        if (!Drivers::USB::Bluetooth::IsInitialized()) return -1;
+        if (!Drivers::Bluetooth::IsInitialized()) return NoDevice;
 
         // bdAddr is a 6-byte buffer, addr[0] = least-significant octet.
-        return Drivers::USB::Bluetooth::SetAddress(bdAddr) ? 0 : -1;
+        return Drivers::Bluetooth::SetAddress(bdAddr) ? Success : GenericFailure;
     }
 
     static int64_t Sys_BtBonds(BtBondInfo* buf, int maxCount) {
         if (!buf || maxCount <= 0) return 0;
-        if (!Drivers::USB::Bluetooth::IsInitialized()) return 0;
+        if (!Drivers::Bluetooth::IsInitialized()) return 0;
 
-        Drivers::USB::Bluetooth::Hci::BondInfo tmpBuf[8];
+        Drivers::Bluetooth::Hci::BondInfo tmpBuf[8];
         int kMax = maxCount;
         if (kMax > 8) kMax = 8;
 
-        int count = Drivers::USB::Bluetooth::ListBonded(tmpBuf, kMax);
+        int count = Drivers::Bluetooth::ListBonded(tmpBuf, kMax);
 
         for (int i = 0; i < count; i++) {
             memcpy(buf[i].bdAddr, tmpBuf[i].Addr, 6);
@@ -90,21 +90,21 @@ namespace montauk::abi {
     }
 
     static int64_t Sys_BtForget(const uint8_t* bdAddr) {
-        if (!bdAddr) return -1;
+        if (!bdAddr) return InvalidArgument;
-        if (!Drivers::USB::Bluetooth::IsInitialized()) return -1;
+        if (!Drivers::Bluetooth::IsInitialized()) return NoDevice;
 
-        return (int64_t)Drivers::USB::Bluetooth::ForgetDevice(bdAddr);
+        return (int64_t)Drivers::Bluetooth::ForgetDevice(bdAddr);
     }
 
     static int64_t Sys_BtList(BtDevInfo* buf, int maxCount) {
         if (!buf || maxCount <= 0) return 0;
-        if (!Drivers::USB::Bluetooth::IsInitialized()) return 0;
+        if (!Drivers::Bluetooth::IsInitialized()) return 0;
 
-        Drivers::USB::Bluetooth::Hci::ConnectionInfo tmpBuf[4];
+        Drivers::Bluetooth::Hci::ConnectionInfo tmpBuf[4];
         int kMax = maxCount;
         if (kMax > 4) kMax = 4;
 
-        int count = Drivers::USB::Bluetooth::ListConnected(tmpBuf, kMax);
+        int count = Drivers::Bluetooth::ListConnected(tmpBuf, kMax);
 
         for (int i = 0; i < count; i++) {
             memcpy(buf[i].bdAddr, tmpBuf[i].BdAddr, 6);
@@ -119,18 +119,18 @@ namespace montauk::abi {
     }
 
     static int64_t Sys_BtInfo(BtAdapterInfo* buf) {
-        if (!buf) return -1;
+        if (!buf) return InvalidArgument;
 
         memset(buf, 0, sizeof(BtAdapterInfo));
 
-        if (!Drivers::USB::Bluetooth::IsInitialized()) {
+        if (!Drivers::Bluetooth::IsInitialized()) {
             buf->initialized = 0;
-            return -1;
+            return NoDevice;
         }
 
         buf->initialized = 1;
-        memcpy(buf->bdAddr, Drivers::USB::Bluetooth::GetBdAddr(), 6);
+        memcpy(buf->bdAddr, Drivers::Bluetooth::GetBdAddr(), 6);
-        buf->scanning = Drivers::USB::Bluetooth::Hci::IsInquiryActive() ? 1 : 0;
+        buf->scanning = Drivers::Bluetooth::Hci::IsInquiryActive() ? 1 : 0;
 
         // Copy adapter name
         const char* name = "MontaukOS";
@@ -138,7 +138,7 @@ namespace montauk::abi {
         for (; i < 63 && name[i]; i++) buf->name[i] = name[i];
         buf->name[i] = '\0';
 
-        return 0;
+        return Success;
     }
 
 }

kernel/src/Api/BuildNo.hpp

@@ -12,4 +12,4 @@
 
 #pragma once
 
-#define MONTAUK_BUILD_NUMBER 184
+#define MONTAUK_BUILD_NUMBER 123

kernel/src/Api/Clipboard.hpp

@@ -1,6 +1,6 @@
 /*
     * Clipboard.hpp
-    * SYS_CLIPBOARD_* plain-text clipboard syscalls
+    * SYS_CLIPBOARD_* syscalls: per-user multi-format clipboard
     * Copyright (c) 2026 Daniel Hammer
 */
 
@@ -11,24 +11,25 @@
 #include <CppLib/Spinlock.hpp>
 
 #include "Syscall.hpp"
+#include "DataTransfer.hpp"
 
 namespace montauk::abi {
 
     static constexpr int ClipboardMaxSlots = 32;
 
+    // One clipboard per user. It holds a multi-format transfer; the text
+    // syscalls read and replace its TRANSFER_MIME_TEXT format, so text-only
+    // apps and multi-format apps share the same clipboard.
     struct ClipboardSlot {
         bool inUse;
         char user[32];
-        char* text;
+        DataTransfer::Transfer transfer;
-        uint32_t len;
         uint64_t serial;
     };
 
     static ClipboardSlot g_clipboardSlots[ClipboardMaxSlots];
     static kcp::Mutex g_clipboardLock;
 
-    static int Sys_ClipboardClear();
-
     static const char* ClipboardCurrentUser() {
         auto* proc = Sched::GetCurrentProcessPtr();
         if (proc == nullptr) return nullptr;
@@ -81,66 +82,84 @@ namespace montauk::abi {
 
         freeSlot->inUse = true;
         ClipboardCopyUser(freeSlot->user, user);
-        freeSlot->text = nullptr;
+        freeSlot->transfer.count = 0;
-        freeSlot->len = 0;
         freeSlot->serial = 0;
         return freeSlot;
     }
 
-    static int Sys_ClipboardSetText(const char* data, uint32_t len) {
+    // Swap `incoming` into the caller's clipboard (it becomes empty) and free
-        if (len == 0)
+    // what was there. An empty transfer clears the clipboard.
-            return Sys_ClipboardClear();
+    static int ClipboardReplace(DataTransfer::Transfer& incoming) {
-        if (data == nullptr || len > CLIPBOARD_MAX_TEXT_BYTES || Memory::g_heap == nullptr)
-            return -1;
-
-        for (uint32_t i = 0; i < len; i++) {
-            if (data[i] == '\0')
-                return -1;
-        }
-
         const char* currentUser = ClipboardCurrentUser();
-        if (currentUser == nullptr) return -1;
+        if (currentUser == nullptr) {
+            DataTransfer::Free(incoming);
+            return GenericFailure;
+        }
 
         char user[32];
         ClipboardCopyUser(user, currentUser);
 
-        char* newText = (char*)Memory::g_heap->Request(len + 1);
+        DataTransfer::Transfer old;
-        if (newText == nullptr) return -1;
+        old.count = 0;
-        for (uint32_t i = 0; i < len; i++)
-            newText[i] = data[i];
-        newText[len] = '\0';
-
-        char* oldText = nullptr;
 
         g_clipboardLock.Acquire();
         ClipboardSlot* slot = ClipboardEnsureSlotLocked(user);
         if (slot == nullptr) {
             g_clipboardLock.Release();
-            Memory::g_heap->Free(newText);
+            DataTransfer::Free(incoming);
-            return -1;
+            return Busy;
         }
 
-        oldText = slot->text;
+        old = slot->transfer;
-        slot->text = newText;
+        slot->transfer = incoming;
-        slot->len = len;
         slot->serial = ClipboardNextSerial(slot->serial);
         g_clipboardLock.Release();
 
-        if (oldText != nullptr)
+        incoming.count = 0;
-            Memory::g_heap->Free(oldText);
+        DataTransfer::Free(old);
+        return Success;
+    }
 
-        return 0;
+    static int Sys_ClipboardClear() {
+        DataTransfer::Transfer empty;
+        empty.count = 0;
+        return ClipboardReplace(empty);
+    }
+
+    static int Sys_ClipboardSetText(const char* data, uint32_t len) {
+        if (len == 0)
+            return Sys_ClipboardClear();
+        if (data == nullptr || len > CLIPBOARD_MAX_TEXT_BYTES)
+            return InvalidArgument;
+
+        for (uint32_t i = 0; i < len; i++) {
+            if (data[i] == '\0')
+                return InvalidArgument;
+        }
+
+        DataTransfer::Transfer transfer;
+        int rc = DataTransfer::BuildText(data, len, transfer);
+        if (rc != Success) return rc;
+        return ClipboardReplace(transfer);
+    }
+
+    static int Sys_ClipboardSetData(const TransferItem* items, uint32_t count) {
+        if (count == 0) return Sys_ClipboardClear();
+        DataTransfer::Transfer transfer;
+        int rc = DataTransfer::BuildFromUser(items, count, transfer);
+        if (rc != Success) return rc;
+        return ClipboardReplace(transfer);
     }
 
     static int Sys_ClipboardGetInfo(ClipboardInfo* out) {
-        if (out == nullptr) return -1;
+        if (out == nullptr) return InvalidArgument;
 
         out->textBytes = 0;
         out->_pad = 0;
         out->serial = 0;
 
         const char* currentUser = ClipboardCurrentUser();
-        if (currentUser == nullptr) return -1;
+        if (currentUser == nullptr) return GenericFailure;
 
         char user[32];
         ClipboardCopyUser(user, currentUser);
@@ -148,73 +167,79 @@ namespace montauk::abi {
         g_clipboardLock.Acquire();
         ClipboardSlot* slot = ClipboardFindSlotLocked(user);
         if (slot != nullptr) {
-            out->textBytes = slot->len;
+            const DataTransfer::Format* text = DataTransfer::Find(slot->transfer, TRANSFER_MIME_TEXT);
+            out->textBytes = text ? text->len : 0;
             out->serial = slot->serial;
         }
         g_clipboardLock.Release();
-        return 0;
+        return Success;
     }
 
     static int Sys_ClipboardGetText(char* buffer, uint32_t bufferLen, uint32_t* outLen, uint64_t* outSerial) {
-        if (outLen == nullptr) return -1;
+        if (outLen == nullptr) return InvalidArgument;
 
         *outLen = 0;
         if (outSerial != nullptr) *outSerial = 0;
 
         const char* currentUser = ClipboardCurrentUser();
-        if (currentUser == nullptr) return -1;
+        if (currentUser == nullptr) return GenericFailure;
 
         char user[32];
         ClipboardCopyUser(user, currentUser);
 
         g_clipboardLock.Acquire();
         ClipboardSlot* slot = ClipboardFindSlotLocked(user);
-        if (slot == nullptr || slot->len == 0) {
+        if (slot != nullptr && outSerial != nullptr)
-            if (slot != nullptr && outSerial != nullptr)
+            *outSerial = slot->serial;
-                *outSerial = slot->serial;
+        const DataTransfer::Format* text =
+            slot ? DataTransfer::Find(slot->transfer, TRANSFER_MIME_TEXT) : nullptr;
+        if (text == nullptr || text->len == 0) {
             g_clipboardLock.Release();
-            return 0;
+            return Success;
         }
 
-        *outLen = slot->len;
+        int rc = DataTransfer::CopyOut(*text, buffer, bufferLen, outLen);
-        if (outSerial != nullptr) *outSerial = slot->serial;
-        if (buffer == nullptr || bufferLen < slot->len) {
-            g_clipboardLock.Release();
-            return -1;
-        }
-
-        for (uint32_t i = 0; i < slot->len; i++)
-            buffer[i] = slot->text[i];
         g_clipboardLock.Release();
-        return 0;
+        return rc;
     }
 
-    static int Sys_ClipboardClear() {
+    static int Sys_ClipboardFormats(TransferFormatInfo* out, uint32_t max, uint64_t* outSerial) {
+        if (outSerial != nullptr) *outSerial = 0;
+
         const char* currentUser = ClipboardCurrentUser();
-        if (currentUser == nullptr) return -1;
+        if (currentUser == nullptr) return GenericFailure;
 
         char user[32];
         ClipboardCopyUser(user, currentUser);
 
-        char* oldText = nullptr;
+        g_clipboardLock.Acquire();
+        ClipboardSlot* slot = ClipboardFindSlotLocked(user);
+        uint32_t n = 0;
+        if (slot != nullptr) {
+            if (outSerial != nullptr) *outSerial = slot->serial;
+            n = out ? DataTransfer::ListFormats(slot->transfer, out, max) : 0;
+            if (out == nullptr || max == 0) n = slot->transfer.count;
+        }
+        g_clipboardLock.Release();
+        return (int)n;
+    }
+
+    static int Sys_ClipboardGetData(const char* mime, void* buffer, uint32_t bufferLen, uint32_t* outLen) {
+        if (outLen == nullptr) return InvalidArgument;
+        *outLen = 0;
+
+        const char* currentUser = ClipboardCurrentUser();
+        if (currentUser == nullptr) return GenericFailure;
+
+        char user[32];
+        ClipboardCopyUser(user, currentUser);
 
         g_clipboardLock.Acquire();
-        ClipboardSlot* slot = ClipboardEnsureSlotLocked(user);
+        ClipboardSlot* slot = ClipboardFindSlotLocked(user);
-        if (slot == nullptr) {
+        const DataTransfer::Format* format = slot ? DataTransfer::Find(slot->transfer, mime) : nullptr;
-            g_clipboardLock.Release();
+        int rc = format ? DataTransfer::CopyOut(*format, buffer, bufferLen, outLen) : NotFound;
-            return -1;
-        }
-
-        oldText = slot->text;
-        slot->text = nullptr;
-        slot->len = 0;
-        slot->serial = ClipboardNextSerial(slot->serial);
         g_clipboardLock.Release();
-
+        return rc;
-        if (oldText != nullptr)
-            Memory::g_heap->Free(oldText);
-
-        return 0;
     }
 
 }

kernel/src/Api/DataTransfer.cpp

@@ -0,0 +1,156 @@
+/*
+    * DataTransfer.cpp
+    * Multi-format payloads shared by the clipboard and drag and drop
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#include "DataTransfer.hpp"
+#include "UserMemory.hpp"
+#include <Memory/Heap.hpp>
+
+namespace DataTransfer {
+
+    using namespace montauk::abi;
+
+    static bool MimeEquals(const char* a, const char* b) {
+        int i = 0;
+        for (; a[i] && b[i]; i++) {
+            char ca = (a[i] >= 'A' && a[i] <= 'Z') ? a[i] + 32 : a[i];
+            char cb = (b[i] >= 'A' && b[i] <= 'Z') ? b[i] + 32 : b[i];
+            if (ca != cb) return false;
+        }
+        return a[i] == b[i];
+    }
+
+    // Length of the "type/subtype" part, before any ";parameters".
+    static int MimeBaseLen(const char* mime) {
+        int n = 0;
+        while (mime[n] && mime[n] != ';') n++;
+        while (n > 0 && mime[n - 1] == ' ') n--;
+        return n;
+    }
+
+    static bool MimeBaseEquals(const char* a, const char* b) {
+        int la = MimeBaseLen(a);
+        int lb = MimeBaseLen(b);
+        if (la != lb) return false;
+        for (int i = 0; i < la; i++) {
+            char ca = (a[i] >= 'A' && a[i] <= 'Z') ? a[i] + 32 : a[i];
+            char cb = (b[i] >= 'A' && b[i] <= 'Z') ? b[i] + 32 : b[i];
+            if (ca != cb) return false;
+        }
+        return true;
+    }
+
+    bool ValidMime(const char* mime) {
+        if (mime == nullptr) return false;
+        bool slash = false;
+        uint32_t i = 0;
+        for (; mime[i]; i++) {
+            if (i + 1 >= TRANSFER_MIME_MAX) return false;
+            char c = mime[i];
+            if (c < 0x20 || c > 0x7E) return false;
+            if (c == '/') slash = true;
+        }
+        return i > 0 && slash;
+    }
+
+    void Free(Transfer& transfer) {
+        for (uint32_t i = 0; i < transfer.count; i++) {
+            if (transfer.formats[i].data != nullptr) Memory::g_heap->Free(transfer.formats[i].data);
+            transfer.formats[i].data = nullptr;
+            transfer.formats[i].len = 0;
+        }
+        transfer.count = 0;
+    }
+
+    static int AddFormat(Transfer& transfer, const char* mime, const uint8_t* data, uint32_t len) {
+        if (transfer.count >= TRANSFER_MAX_FORMATS) return InvalidArgument;
+        Format& f = transfer.formats[transfer.count];
+        uint32_t m = 0;
+        for (; mime[m] && m + 1 < TRANSFER_MIME_MAX; m++) f.mime[m] = mime[m];
+        f.mime[m] = '\0';
+
+        // Never hand out a null pointer for an empty format; one byte keeps
+        // "present but empty" distinct from "absent".
+        f.data = (uint8_t*)Memory::g_heap->Request(len > 0 ? len : 1);
+        if (f.data == nullptr) return OutOfMemory;
+        for (uint32_t i = 0; i < len; i++) f.data[i] = data[i];
+        f.len = len;
+        transfer.count++;
+        return Success;
+    }
+
+    int BuildFromUser(const TransferItem* items, uint32_t count, Transfer& out) {
+        out.count = 0;
+        if (items == nullptr || count == 0 || count > TRANSFER_MAX_FORMATS) return InvalidArgument;
+        if (Memory::g_heap == nullptr) return OutOfMemory;
+
+        // Validate everything before allocating anything.
+        uint64_t total = 0;
+        for (uint32_t i = 0; i < count; i++) {
+            if (!UserMemory::String(items[i].mime, TRANSFER_MIME_MAX)) return BadPointer;
+            if (!ValidMime((const char*)items[i].mime)) return InvalidArgument;
+            if (items[i].len > 0 && !UserMemory::Range(items[i].data, items[i].len, false)) return BadPointer;
+            total += items[i].len;
+            for (uint32_t j = 0; j < i; j++) {
+                if (MimeEquals((const char*)items[i].mime, (const char*)items[j].mime))
+                    return InvalidArgument;
+            }
+        }
+        if (total > TRANSFER_MAX_BYTES) return InvalidArgument;
+
+        for (uint32_t i = 0; i < count; i++) {
+            int rc = AddFormat(out, (const char*)items[i].mime,
+                               (const uint8_t*)items[i].data, items[i].len);
+            if (rc != Success) {
+                Free(out);
+                return rc;
+            }
+        }
+        return Success;
+    }
+
+    int BuildText(const char* text, uint32_t len, Transfer& out) {
+        out.count = 0;
+        if (Memory::g_heap == nullptr) return OutOfMemory;
+        int rc = AddFormat(out, TRANSFER_MIME_TEXT, (const uint8_t*)text, len);
+        if (rc != Success) Free(out);
+        return rc;
+    }
+
+    const Format* Find(const Transfer& transfer, const char* mime) {
+        if (mime == nullptr) return nullptr;
+        for (uint32_t i = 0; i < transfer.count; i++) {
+            if (MimeEquals(transfer.formats[i].mime, mime)) return &transfer.formats[i];
+        }
+        bool has_params = false;
+        for (int i = 0; mime[i]; i++) {
+            if (mime[i] == ';') { has_params = true; break; }
+        }
+        if (has_params) return nullptr;
+        for (uint32_t i = 0; i < transfer.count; i++) {
+            if (MimeBaseEquals(transfer.formats[i].mime, mime)) return &transfer.formats[i];
+        }
+        return nullptr;
+    }
+
+    uint32_t ListFormats(const Transfer& transfer, TransferFormatInfo* out, uint32_t max) {
+        uint32_t n = 0;
+        for (; n < transfer.count && n < max; n++) {
+            for (uint32_t c = 0; c < TRANSFER_MIME_MAX; c++) out[n].mime[c] = transfer.formats[n].mime[c];
+            out[n].len = transfer.formats[n].len;
+            out[n]._reserved = 0;
+        }
+        return n;
+    }
+
+    int CopyOut(const Format& format, void* buffer, uint32_t bufferLen, uint32_t* outLen) {
+        if (outLen != nullptr) *outLen = format.len;
+        if (format.len == 0) return Success;
+        if (buffer == nullptr || bufferLen < format.len) return BufferTooSmall;
+        uint8_t* dst = (uint8_t*)buffer;
+        for (uint32_t i = 0; i < format.len; i++) dst[i] = format.data[i];
+        return Success;
+    }
+}

kernel/src/Api/DataTransfer.hpp

@@ -0,0 +1,52 @@
+/*
+    * DataTransfer.hpp
+    * Multi-format payloads shared by the clipboard and drag and drop
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+#include "Syscall.hpp"
+
+// A transfer is a small set of formats, each one MIME type plus bytes, all
+// owned by the kernel heap. Producers offer every representation they can
+// (a file list and its paths as text, say) and each consumer reads the one it
+// understands. The clipboard holds one transfer per user; the window server
+// holds the active drag's.
+namespace DataTransfer {
+
+    struct Format {
+        char     mime[montauk::abi::TRANSFER_MIME_MAX];
+        uint8_t* data;
+        uint32_t len;
+    };
+
+    struct Transfer {
+        Format   formats[montauk::abi::TRANSFER_MAX_FORMATS];
+        uint32_t count;
+    };
+
+    // Copy `count` formats in from user memory, validating every pointer the
+    // items carry (the item array itself must already be validated). On
+    // failure `out` is left empty. Returns a StatusCodes value.
+    int BuildFromUser(const montauk::abi::TransferItem* items, uint32_t count, Transfer& out);
+
+    // A transfer holding just `text` as TRANSFER_MIME_TEXT.
+    int BuildText(const char* text, uint32_t len, Transfer& out);
+
+    void Free(Transfer& transfer);
+
+    // Exact match first; a request without parameters ("text/plain") also
+    // matches a stored format that has them ("text/plain;charset=utf-8").
+    const Format* Find(const Transfer& transfer, const char* mime);
+
+    uint32_t ListFormats(const Transfer& transfer, montauk::abi::TransferFormatInfo* out, uint32_t max);
+
+    // Copy a format to a (validated) user buffer. *outLen always receives the
+    // full length; BufferTooSmall when it does not fit, nothing copied then.
+    int CopyOut(const Format& format, void* buffer, uint32_t bufferLen, uint32_t* outLen);
+
+    // MIME string checks shared by every entry point: "type/subtype" with
+    // optional parameters, printable ASCII, shorter than TRANSFER_MIME_MAX.
+    bool ValidMime(const char* mime);
+}

kernel/src/Api/Device.hpp

@@ -6,18 +6,20 @@
 
 #pragma once
 #include <cstdint>
-#include <Hal/Apic/ApicInit.hpp>
-#include <Drivers/PS2/PS2Controller.hpp>
 #include <Drivers/USB/Xhci.hpp>
 #include <Drivers/USB/UsbDevice.hpp>
-#include <Drivers/USB/Bluetooth/Bluetooth.hpp>
+#include <Drivers/Bluetooth/Bluetooth.hpp>
 #include <Drivers/Net/E1000.hpp>
 #include <Drivers/Net/E1000E.hpp>
+#include <Drivers/Net/Genet.hpp>
 #include <Drivers/Net/Wifi/Wifi.hpp>
 #include <Drivers/Graphics/IntelGPU.hpp>
 #include <Drivers/Storage/Ahci.hpp>
 #include <Drivers/Audio/IntelHda.hpp>
 #include <Pci/Pci.hpp>
+#include <Arch/Arch.hpp>
+#include <Arch/Platform.hpp>
+#include <Arch/Smp.hpp>
 
 #include "Syscall.hpp"
 #include <Drivers/Storage/BlockDevice.hpp>
@@ -69,28 +71,21 @@ namespace montauk::abi {
         };
 
         // CPU cores (category 0)
-        int cpuCount = Hal::GetDetectedCpuCount();
+        int cpuCount = Smp::GetDetectedCpuCount();
         if (cpuCount > 0) {
             char detail[48];
             int p = 0;
-            p = dl_append(detail, p, "x86_64, ", 48);
+            p = dl_append(detail, p, MONTAUK_ARCH_NAME ", ", 48);
             p = dl_append_dec(detail, p, cpuCount, 48);
             p = dl_append(detail, p, " core(s)", 48);
             add(0, "Processor", detail);
         }
 
-        // Interrupt controllers (category 1)
+        // Interrupt controllers (category 1), timer (2), platform input (3)
-        add(1, "Local APIC", "");
+        Arch::ListPlatformDevices([](void* ctx, uint8_t category,
-        add(1, "I/O APIC", "");
+                                     const char* name, const char* detail) {
-
+            (*(decltype(add)*)ctx)(category, name, detail);
-        // Timer (category 2)
+        }, &add);
-        add(2, "LAPIC Timer", "Local APIC periodic timer");
-
-        // PS/2 Input (category 3)
-        add(3, "PS/2 Keyboard", "IRQ 1");
-        if (Drivers::PS2::IsDualChannel()) {
-            add(3, "PS/2 Mouse", "IRQ 12");
-        }
 
         // USB devices (category 4)
         if (Drivers::USB::Xhci::IsInitialized()) {
@@ -128,11 +123,24 @@ namespace montauk::abi {
         if (Drivers::Net::E1000E::IsInitialized()) {
             add(5, "Intel E1000E", "Gigabit Ethernet (82574L)");
         }
+        if (Drivers::Net::Genet::IsInitialized()) {
+            uint32_t speed = Drivers::Net::Genet::GetLinkSpeed();
+            add(5, "Broadcom GENET", speed == 1000 ? "Gigabit Ethernet (1000 Mbit/s)"
+                                   : speed == 100  ? "Gigabit Ethernet (100 Mbit/s)"
+                                   : speed == 10   ? "Gigabit Ethernet (10 Mbit/s)"
+                                   : "Gigabit Ethernet (no link)");
+        }
         if (Drivers::Net::Wifi::IsPresent()) {
             add(5, "Intel Wi-Fi", Drivers::Net::Wifi::IsLinkUp()
                                   ? "802.11 wireless (connected)"
                                   : "802.11 wireless");
         }
+        // An on-board UART Bluetooth controller (Raspberry Pi) is no USB
+        // device, so it is listed here; USB adapters appear above.
+        if (Drivers::Bluetooth::IsInitialized() &&
+            Drivers::Bluetooth::Hci::GetTransport() == Drivers::Bluetooth::Hci::Transport::Uart) {
+            add(5, "Bluetooth Adapter", "Broadcom, HCI over UART");
+        }
 
         // Display (category 6)
         if (Drivers::Graphics::IntelGPU::IsInitialized()) {
@@ -154,7 +162,7 @@ namespace montauk::abi {
             add(9, "Intel HDA", detail);
         }
 
-        if (Drivers::USB::Bluetooth::IsInitialized()) {
+        if (Drivers::Bluetooth::IsInitialized()) {
             add(9, "Bluetooth Audio", "A2DP (SBC)");
         }
 
@@ -225,10 +233,10 @@ namespace montauk::abi {
     }
 
     static int Sys_DiskInfo(DiskInfo* buf, int blockDevIndex) {
-        if (buf == nullptr) return -1;
+        if (buf == nullptr) return InvalidArgument;
 
         auto* bdev = Drivers::Storage::GetBlockDevice(blockDevIndex);
-        if (!bdev) return -1;
+        if (!bdev) return NoDevice;
 
         // Zero the struct first so SATA-specific fields default to 0
         for (unsigned i = 0; i < sizeof(DiskInfo); i++)
@@ -254,6 +262,10 @@ namespace montauk::abi {
         case Drivers::Storage::BLOCK_KIND_USB_MSC:
             buf->type = 4;
             break;
+        case Drivers::Storage::BLOCK_KIND_SD:
+            buf->type = 5;
+            buf->rpm = 1;
+            break;
         default:
             break;
         }
@@ -303,8 +315,8 @@ namespace montauk::abi {
         }
 
         // If type is still 0, this block device is not recognized
-        if (buf->type == 0) return -1;
+        if (buf->type == 0) return UnsupportedOp;
 
-        return 0;
+        return Success;
     }
 };

kernel/src/Api/Filesystem.hpp

@@ -20,14 +20,33 @@
 #include <Fs/ProtectedPaths.hpp>
 
 namespace montauk::abi {
+    // Writes refused here return SystemIntegrityError: the path is protected
+    // by Fs/ProtectedPaths, not merely gated on an ordinary capability.
     static bool CanModifyFilePath(const char* resolved) {
         uint64_t required = Fs::RequiredFileWriteCapability(resolved);
         return required == 0 || Sched::HasCapability(required);
     }
 
+    // Drivers fail a create under a missing parent with a bare -1; report it
+    // as NoSuchPath so callers can tell it apart (Rust's create_dir_all only
+    // creates the missing ancestors when mkdir says NotFound).
+    static bool ParentDirExists(const char* resolved) {
+        char parent[256];
+        int len = Lib::strlen(resolved);
+        if (len >= (int)sizeof(parent)) return false;
+        memcpy(parent, resolved, len + 1);
+        while (len > 0 && parent[len - 1] == '/') len--;
+        while (len > 0 && parent[len - 1] != '/') len--;
+        while (len > 0 && parent[len - 1] == '/') len--;
+        if (len == 0 || parent[len - 1] == ':') return true;  // drive root
+        parent[len] = '\0';
+        Fs::Vfs::StatInfo info;
+        return Fs::Vfs::VfsStat(parent, info) == 0 && info.isDir;
+    }
+
     static int Sys_Open(const char* path) {
         char resolved[256];
-        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return -1;
+        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return InvalidArgument;
         return Ipc::OpenFileHandle(resolved);
     }
 
@@ -53,8 +72,8 @@ namespace montauk::abi {
     static int Sys_ReadDirAt(const char* path, const char** outNames, int maxEntries,
                              int startIndex) {
         char resolved[256];
-        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return -1;
+        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return InvalidArgument;
-        if (startIndex < 0) return -1;
+        if (startIndex < 0) return InvalidArgument;
 
         // Get entries from VFS into a kernel-local array
         const char* kernelNames[256];
@@ -65,7 +84,7 @@ namespace montauk::abi {
 
         // Allocate a user-accessible page for string data via process heap
         auto* proc = Sched::GetCurrentProcessPtr();
-        if (proc == nullptr) return -1;
+        if (proc == nullptr) return GenericFailure;
 
         // Use a rotating ring of scratch pages below the heap instead of
         // extending the heap high-water mark on every call. This keeps repeated
@@ -75,16 +94,16 @@ namespace montauk::abi {
         proc->readdirCursor = (slot + 1) % Sched::UserReadDirSlots;
 
         uint64_t userVa = Sched::UserReadDirBase + (uint64_t)slot * 0x1000ULL;
-        uint64_t physAddr = Memory::VMM::Paging::GetPhysAddr(proc->pml4Phys, userVa);
+        uint64_t physAddr = Memory::VMM::Paging::GetPhysAddr(proc->pageTableRoot, userVa);
         uint8_t* pageBuf = nullptr;
 
         if (physAddr == 0) {
             void* page = Memory::g_pfa->AllocateZeroed();
-            if (page == nullptr) return -1;
+            if (page == nullptr) return OutOfMemory;
             physAddr = Memory::SubHHDM((uint64_t)page);
-            if (!Memory::VMM::Paging::MapUserIn(proc->pml4Phys, physAddr, userVa)) {
+            if (!Memory::VMM::Paging::MapUserIn(proc->pageTableRoot, physAddr, userVa)) {
                 Memory::g_pfa->Free(page);
-                return -1;
+                return OutOfMemory;
             }
             pageBuf = (uint8_t*)page;
         } else {
@@ -113,25 +132,26 @@ namespace montauk::abi {
 
     static int Sys_FCreate(const char* path) {
         char resolved[256];
-        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return -1;
+        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return InvalidArgument;
-        if (!CanModifyFilePath(resolved)) return SYS_ERR_PERMISSION;
+        if (!CanModifyFilePath(resolved)) return SystemIntegrityError;
+        if (!ParentDirExists(resolved)) return NoSuchPath;
         return Ipc::CreateFileHandle(resolved);
     }
 
     static int Sys_FDelete(const char* path) {
         char resolved[256];
-        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return -1;
+        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return InvalidArgument;
-        if (!CanModifyFilePath(resolved)) return SYS_ERR_PERMISSION;
+        if (!CanModifyFilePath(resolved)) return SystemIntegrityError;
         return Fs::Vfs::VfsDelete(resolved);
     }
 
     static int Sys_Stat(const char* path, FileStat* out) {
-        if (out == nullptr) return -1;
+        if (out == nullptr) return InvalidArgument;
         char resolved[256];
-        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return -1;
+        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return InvalidArgument;
 
         Fs::Vfs::StatInfo info;
-        if (Fs::Vfs::VfsStat(resolved, info) != 0) return -1;
+        if (Fs::Vfs::VfsStat(resolved, info) != 0) return NoSuchPath;
 
         out->size  = info.size;
         out->mtime = info.mtime;
@@ -139,38 +159,47 @@ namespace montauk::abi {
         out->atime = info.atime;
         out->mode  = info.mode;
         out->isDir = info.isDir ? 1u : 0u;
-        return 0;
+        return Success;
     }
 
     static int Sys_Utime(const char* path, int64_t atime, int64_t mtime,
                          bool useCurrent) {
         char resolved[256];
-        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return -1;
+        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return InvalidArgument;
         // Timestamps are file state like any other: a protected path must not
         // be mutable through a side door that skips the write check.
-        if (!CanModifyFilePath(resolved)) return SYS_ERR_PERMISSION;
+        if (!CanModifyFilePath(resolved)) return SystemIntegrityError;
         if (useCurrent) {
             int64_t now = Timekeeping::GetUnixTimestamp();
             atime = now;
             mtime = now;
         }
-        return Fs::Vfs::VfsSetTimes(resolved, atime, mtime);
+        // VfsSetTimes has its own codes: -1 not found, -2 unsupported by the
+        // filesystem, -3 timestamp out of range, -4 I/O error.
+        switch (Fs::Vfs::VfsSetTimes(resolved, atime, mtime)) {
+            case 0:  return Success;
+            case -2: return UnsupportedOp;
+            case -3: return InvalidArgument;
+            case -4: return IoFailure;
+            default: return NoSuchPath;
+        }
     }
 
     static int Sys_FMkdir(const char* path) {
         char resolved[256];
-        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return -1;
+        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return InvalidArgument;
-        if (!CanModifyFilePath(resolved)) return SYS_ERR_PERMISSION;
+        if (!CanModifyFilePath(resolved)) return SystemIntegrityError;
+        if (!ParentDirExists(resolved)) return NoSuchPath;
         return Fs::Vfs::VfsMkdir(resolved);
     }
 
     static int Sys_FRename(const char* oldPath, const char* newPath) {
         char resolvedOld[256];
         char resolvedNew[256];
-        if (!ResolveProcessPath(oldPath, resolvedOld, sizeof(resolvedOld))) return -1;
+        if (!ResolveProcessPath(oldPath, resolvedOld, sizeof(resolvedOld))) return InvalidArgument;
-        if (!ResolveProcessPath(newPath, resolvedNew, sizeof(resolvedNew))) return -1;
+        if (!ResolveProcessPath(newPath, resolvedNew, sizeof(resolvedNew))) return InvalidArgument;
         if (!CanModifyFilePath(resolvedOld) || !CanModifyFilePath(resolvedNew))
-            return SYS_ERR_PERMISSION;
+            return SystemIntegrityError;
         return Fs::Vfs::VfsRename(resolvedOld, resolvedNew);
     }
 

kernel/src/Api/Graphics.hpp

@@ -11,6 +11,8 @@
 #include <Memory/HHDM.hpp>
 #include <Graphics/Framebuffer.hpp>
 #include <Drivers/Graphics/IntelGPU.hpp>
+#include <ACPI/AcpiVideo.hpp>
+#include <ACPI/WmiEcBacklight.hpp>
 #include <Terminal/Terminal.hpp>
 
 #include "Syscall.hpp"
@@ -81,7 +83,7 @@ namespace montauk::abi {
 
         for (uint64_t i = 0; i < numPages; i++) {
             if (!Memory::VMM::Paging::MapUserInWC(
-                proc->pml4Phys,
+                proc->pageTableRoot,
                 fbPhys + i * 0x1000,
                 userVa + i * 0x1000
             )) {
@@ -96,7 +98,7 @@ namespace montauk::abi {
             uint64_t buf1Phys = Drivers::Graphics::IntelGPU::GetBufferPhys(1);
             for (uint64_t i = 0; i < numPages; i++) {
                 if (!Memory::VMM::Paging::MapUserInWC(
-                    proc->pml4Phys,
+                    proc->pageTableRoot,
                     buf1Phys + i * 0x1000,
                     userVa + (numPages + i) * 0x1000
                 )) {
@@ -118,18 +120,49 @@ namespace montauk::abi {
         if ((int64_t)index == -1) {
             return GPU::FlipSupported() ? 1 : 0;
         }
-        if (!GPU::FlipSupported()) return -1;
+        if (!GPU::FlipSupported()) return UnsupportedOp;
 
         auto* proc = Sched::GetCurrentProcessPtr();
         if ((int64_t)index == -2) {
-            return proc != nullptr ? GPU::AcquireFlip(proc->pid) : -1;
+            return proc != nullptr ? GPU::AcquireFlip(proc->pid) : GenericFailure;
         }
-        if (proc == nullptr) return -1;
+        if (proc == nullptr) return GenericFailure;
         return GPU::Flip((int)index, (flags & 1) != 0, proc->pid);
     }
 
+    // Firmware brightness interfaces (EC via WMI, ACPI _BCM) drive the
+    // built-in panel, so they apply only while that panel is what is on
+    // screen (or no native driver knows what is): never for an external
+    // monitor, where they would silently dim a closed lid instead.
+    static bool PanelIsActiveOutput() {
+        if (!Drivers::Graphics::IntelGPU::IsInitialized()) return true;
+        uint8_t type = Drivers::Graphics::IntelGPU::ActiveConnectorType();
+        return type == DISPLAY_CONNECTOR_EDP || type == DISPLAY_CONNECTOR_LVDS;
+    }
+
+    // Order follows who actually owns the backlight. A firmware that names
+    // the EC as the source has taken it away from the GPU, whose PWM then
+    // still accepts writes that reach nothing; so the EC interface goes
+    // first, then the GPU's own controls, then ACPI _BCM as a last resort.
+    static int DisplayBrightness(int percent) {
+        bool panel = PanelIsActiveOutput();
+        if (panel) {
+            int rc = Hal::WmiEcBacklight::Brightness(percent);
+            if (rc != -1) return rc;
+        }
+        int rc = Drivers::Graphics::IntelGPU::Brightness(percent);
+        if (rc == -1 && panel) rc = Hal::AcpiVideo::Brightness(percent);
+        return rc;
+    }
+
     static int64_t Sys_DisplayInfo(DisplayInfo* out) {
-        return Drivers::Graphics::IntelGPU::GetDisplayInfo(out) ? 0 : -1;
+        if (!Drivers::Graphics::IntelGPU::GetDisplayInfo(out)) return NoDevice;
+        // The driver only knows its own backends; report what a set would use.
+        int level = DisplayBrightness(-1);
+        out->brightness = level >= 0 ? level : -1;
+        if (level >= 0) out->capabilities |= DISPLAY_CAP_BRIGHTNESS;
+        else out->capabilities &= ~DISPLAY_CAP_BRIGHTNESS;
+        return Success;
     }
 
     static int64_t Sys_DisplayModes(DisplayModeInfo* out, int maxCount) {
@@ -137,11 +170,20 @@ namespace montauk::abi {
     }
 
     static int64_t Sys_DisplaySetMode(int modeIndex) {
-        return Drivers::Graphics::IntelGPU::SetDisplayMode(modeIndex);
+        // The driver answers -2 for a mode that would need link retraining
+        // and -3 for a modeset that failed and was rolled back.
+        int64_t rc = Drivers::Graphics::IntelGPU::SetDisplayMode(modeIndex);
+        if (rc == -2) return UnsupportedOp;
+        if (rc == -3) return IoFailure;
+        return rc;
     }
 
+    // percent < 0 queries. Returns 0..100, NoDevice when the active output
+    // has no controllable backlight, IoFailure when the hardware refused.
     static int64_t Sys_DisplayBrightness(int percent) {
-        return Drivers::Graphics::IntelGPU::Brightness(percent);
+        int rc = DisplayBrightness(percent);
+        if (rc >= 0) return rc;
+        return rc == -1 ? NoDevice : IoFailure;
     }
 
     static uint64_t Sys_TermSize() {

kernel/src/Api/Heap.hpp

@@ -107,7 +107,7 @@ namespace montauk::abi {
                 if (page == nullptr) break;
                 uint64_t phys = Memory::SubHHDM((uint64_t)page);
                 if (!Memory::VMM::Paging::MapUserInPermissions(
-                        proc->pml4Phys, phys, pageVa, writable, executable)) {
+                        proc->pageTableRoot, phys, pageVa, writable, executable)) {
                     Memory::g_pfa->Free(page);
                     break;
                 }
@@ -131,7 +131,7 @@ namespace montauk::abi {
 
     // Reset heap allocation tracking for a process slot.
     // The actual physical pages are freed by Paging::FreeUserHalf() during process cleanup.
-    inline void CleanupHeapForSlot(int slot, uint64_t /*pml4Phys*/) {
+    inline void CleanupHeapForSlot(int slot, uint64_t /*pageTableRoot*/) {
         if (slot < 0 || slot >= Sched::MaxProcesses) return;
         g_heapLocks[slot].Acquire();
         HeapAlloc* area = g_heapAllocs[slot];
@@ -182,7 +182,7 @@ namespace montauk::abi {
             }
             for (uint64_t i = 0; i < area->numPages; i++)
                 if (Memory::VMM::Paging::GetPhysAddr(
-                        proc->pml4Phys, area->va + i * 0x1000ULL) != 0)
+                        proc->pageTableRoot, area->va + i * 0x1000ULL) != 0)
                     resident++;
             *link = area->next;
             area->next = released;
@@ -194,7 +194,7 @@ namespace montauk::abi {
         // user TLB entry can otherwise corrupt the frame's next owner.
         while (released != nullptr) {
             HeapAlloc* next = released->next;
-            Memory::UnmapAndFreeUserRange(proc->pml4Phys, released->va,
+            Memory::UnmapAndFreeUserRange(proc->pageTableRoot, released->va,
                                        released->numPages);
             Sched::ReleaseUserHeapRange(slot, released->va,
                                         released->numPages * 0x1000ULL);
@@ -211,9 +211,9 @@ namespace montauk::abi {
         auto* proc = Sched::GetCurrentProcessPtr();
         int slot = GetCurrentSlot();
         if (proc == nullptr || slot < 0 || size == 0 || (addr & 0xFFFULL) != 0 ||
-            size > UINT64_MAX - 0xFFFULL) return -1;
+            size > UINT64_MAX - 0xFFFULL) return InvalidArgument;
         size = (size + 0xFFFULL) & ~0xFFFULL;
-        if (addr > UINT64_MAX - size) return -1;
+        if (addr > UINT64_MAX - size) return InvalidArgument;
 
         g_heapLocks[slot].Acquire();
         uint64_t firstPage = 0;
@@ -229,7 +229,7 @@ namespace montauk::abi {
             }
             link = &area->next;
         }
-        if (*link == nullptr) { g_heapLocks[slot].Release(); return -1; }
+        if (*link == nullptr) { g_heapLocks[slot].Release(); return InvalidArgument; }
 
         HeapAlloc* area = *link;
         uint64_t oldPages = area->numPages;
@@ -253,16 +253,16 @@ namespace montauk::abi {
 
         uint64_t resident = 0;
         for (uint64_t i = 0; i < pages; i++)
-            if (Memory::VMM::Paging::GetPhysAddr(proc->pml4Phys, addr + i * 0x1000ULL) != 0)
+            if (Memory::VMM::Paging::GetPhysAddr(proc->pageTableRoot, addr + i * 0x1000ULL) != 0)
                 resident++;
         g_heapLocks[slot].Release();
 
-        Memory::UnmapAndFreeUserRange(proc->pml4Phys, addr, pages);
+        Memory::UnmapAndFreeUserRange(proc->pageTableRoot, addr, pages);
         Sched::ReleaseUserHeapRange(slot, addr, size);
         g_heapLocks[slot].Acquire();
         Sched::g_allocatedPages[slot] -= resident;
         g_heapLocks[slot].Release();
-        return 0;
+        return Success;
     }
 
     inline int64_t Sys_Protect(uint64_t addr, uint64_t size, uint64_t prot) {
@@ -272,9 +272,9 @@ namespace montauk::abi {
             size > UINT64_MAX - 0xFFFULL ||
             (prot & ~(VmProtRead | VmProtWrite | VmProtExec)) != 0 ||
             (prot & VmProtRead) == 0 ||
-            ((prot & VmProtWrite) != 0 && (prot & VmProtExec) != 0)) return -1;
+            ((prot & VmProtWrite) != 0 && (prot & VmProtExec) != 0)) return InvalidArgument;
         size = (size + 0xFFFULL) & ~0xFFFULL;
-        if (addr > UINT64_MAX - size) return -1;
+        if (addr > UINT64_MAX - size) return InvalidArgument;
 
         g_heapLocks[slot].Acquire();
         uint64_t pages = size / 0x1000ULL;
@@ -293,7 +293,7 @@ namespace montauk::abi {
                     break;
                 }
             }
-            if (!covered) { g_heapLocks[slot].Release(); return -1; }
+            if (!covered) { g_heapLocks[slot].Release(); return InvalidArgument; }
         }
 
         uint64_t end = addr + size;
@@ -336,28 +336,28 @@ namespace montauk::abi {
         bool writable = (prot & VmProtWrite) != 0;
         bool executable = (prot & VmProtExec) != 0;
         for (uint64_t i = 0; i < pages; i++) {
-            if (Memory::VMM::Paging::GetPhysAddr(proc->pml4Phys,
+            if (Memory::VMM::Paging::GetPhysAddr(proc->pageTableRoot,
                                                  addr + i * 0x1000ULL) == 0)
                 continue;
-            if (!Memory::VMM::Paging::ProtectUserIn(proc->pml4Phys,
+            if (!Memory::VMM::Paging::ProtectUserIn(proc->pageTableRoot,
                                                      addr + i * 0x1000ULL,
                                                      writable, executable)) {
                 g_heapLocks[slot].Release();
-                return -1;
+                return GenericFailure;
             }
         }
         g_heapLocks[slot].Release();
-        Memory::ShootdownUserRange(proc->pml4Phys, addr, (uint32_t)pages);
+        Memory::ShootdownUserRange(proc->pageTableRoot, addr, (uint32_t)pages);
-        return 0;
+        return Success;
     }
 
     // Resolve a non-present access inside an anonymous VMA. This is used for
     // both ring-3 faults and kernel copies to untouched userspace buffers.
-    inline bool TryHandleAnonymousPageFault(uint64_t faultAddr, uint64_t errorCode) {
+    inline bool TryHandleAnonymousPageFault(uint64_t faultAddr, const Memory::VMM::PageFault& fault) {
         auto* proc = Sched::GetCurrentProcessPtr();
         int slot = GetCurrentSlot();
         if (proc == nullptr || slot < 0 || faultAddr < Sched::UserHeapBase ||
-            faultAddr >= Sched::UserHeapLimit || (errorCode & 1) != 0) return false;
+            faultAddr >= Sched::UserHeapLimit || fault.present) return false;
 
         uint64_t pageVa = faultAddr & ~0xFFFULL;
         g_heapLocks[slot].Acquire();
@@ -370,14 +370,14 @@ namespace montauk::abi {
             }
         }
         if (found == nullptr ||
-            ((errorCode & (1ULL << 1)) != 0 && (found->prot & VmProtWrite) == 0) ||
+            (fault.write && (found->prot & VmProtWrite) == 0) ||
-            ((errorCode & (1ULL << 4)) != 0 && (found->prot & VmProtExec) == 0)) {
+            (fault.execute && (found->prot & VmProtExec) == 0)) {
             g_heapLocks[slot].Release();
             return false;
         }
 
         // A sibling may have resolved the same fault while this CPU waited.
-        if (Memory::VMM::Paging::GetPhysAddr(proc->pml4Phys, pageVa) != 0) {
+        if (Memory::VMM::Paging::GetPhysAddr(proc->pageTableRoot, pageVa) != 0) {
             g_heapLocks[slot].Release();
             return true;
         }
@@ -389,7 +389,7 @@ namespace montauk::abi {
         }
         uint64_t phys = Memory::SubHHDM((uint64_t)page);
         bool mapped = Memory::VMM::Paging::MapUserInPermissions(
-            proc->pml4Phys, phys, pageVa,
+            proc->pageTableRoot, phys, pageVa,
             (found->prot & VmProtWrite) != 0, (found->prot & VmProtExec) != 0);
         if (!mapped) Memory::g_pfa->Free(page);
         else Sched::g_allocatedPages[slot]++;

kernel/src/Api/Info.hpp

@@ -17,13 +17,13 @@ namespace montauk::abi {
 
         // Copy strings into fixed-size arrays (user-accessible)
         const char* name = "MontaukOS";
-        const char* ver = "0.2.0";
+        const char* ver = "0.2.1";
         for (int i = 0; name[i]; i++) outInfo->osName[i] = name[i];
         outInfo->osName[9] = '\0';
         for (int i = 0; ver[i]; i++) outInfo->osVersion[i] = ver[i];
         outInfo->osVersion[5] = '\0';
 
-        outInfo->apiVersion = 11;
+        outInfo->apiVersion = 12;
         outInfo->maxProcesses = Sched::MaxProcesses;
         outInfo->buildNumber = MONTAUK_BUILD_NUMBER;
     }

kernel/src/Api/IoRedir.hpp

@@ -19,18 +19,18 @@ namespace montauk::abi {
             const char* path, const char* args,
             const SpawnCapabilities* capabilities) {
         char resolved[256];
-        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return -1;
+        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return InvalidArgument;
 
         int parentSlot = Ipc::CurrentSlot();
         int childPid = Sched::Spawn(resolved, args, false, nullptr, 0,
                                     capabilities);
-        if (childPid < 0) return -1;
+        if (childPid < 0) return childPid;
 
         auto* child = Sched::GetProcessByPid(childPid);
         int childSlot = Ipc::SlotForPid(childPid);
         if (child == nullptr || childSlot < 0 || parentSlot < 0) {
             Sched::KillProcess(childPid);
-            return -1;
+            return GenericFailure;
         }
 
         Ipc::Stream* outStream = Ipc::CreateStream(RedirOutputStreamCapacity);
@@ -41,7 +41,7 @@ namespace montauk::abi {
             if (inStream != nullptr) Ipc::ReleaseStream(inStream, false, false);
             if (keyMailbox != nullptr) Ipc::ReleaseMailbox(keyMailbox, false, false);
             Sched::KillProcess(childPid);
-            return -1;
+            return OutOfMemory;
         }
 
         int parentOutHandle = Ipc::InstallHandleForSlot(
@@ -71,7 +71,7 @@ namespace montauk::abi {
             if (child->ioInHandle < 0) Ipc::ReleaseStream(inStream, false, false);
             if (child->ioKeyHandle < 0) Ipc::ReleaseMailbox(keyMailbox, false, false);
             Sched::KillProcess(childPid);
-            return -1;
+            return GenericFailure;
         }
 
         child->redirected = true;
@@ -80,7 +80,7 @@ namespace montauk::abi {
         child->termRows = 25;
         if (Sched::StartProcess(childPid) < 0) {
             Sched::KillProcess(childPid);
-            return -1;
+            return GenericFailure;
         }
 
         return childPid;
@@ -93,11 +93,11 @@ namespace montauk::abi {
     static int Sys_SpawnRedirCaps(const char* path, const char* args,
                                   const SpawnCapabilities* requested) {
         auto* parent = Sched::GetCurrentProcessPtr();
-        if (parent == nullptr || requested == nullptr) return -1;
+        if (parent == nullptr || requested == nullptr) return InvalidArgument;
 
         SpawnCapabilities copy = *requested;
         if (!ValidCapabilityDelegation(copy, parent->delegableCaps)) {
-            return SYS_ERR_PERMISSION;
+            return CapabilityDenied;
         }
         return Sys_SpawnRedirInternal(path, args, &copy);
     }
@@ -105,51 +105,68 @@ namespace montauk::abi {
     static int Sys_ChildIoRead(int childPid, char* buf, int maxLen) {
         auto* child = Sched::GetProcessByPid(childPid);
         if (child == nullptr || child->parentPid != Sched::GetCurrentPid())
-            return SYS_ERR_PERMISSION;
+            return PermissionDenied;
         Ipc::HandleSnapshot snapshot;
         Ipc::Stream* stream = GetRedirOutStream(child, snapshot);
-        if (!child->redirected || stream == nullptr) return -1;
+        if (!child->redirected || stream == nullptr) return InvalidArgument;
         return Ipc::StreamRead(stream, (uint8_t*)buf, maxLen, true);
     }
 
     static int Sys_ChildIoWrite(int childPid, const char* data, int len) {
         auto* child = Sched::GetProcessByPid(childPid);
         if (child == nullptr || child->parentPid != Sched::GetCurrentPid())
-            return SYS_ERR_PERMISSION;
+            return PermissionDenied;
         Ipc::HandleSnapshot snapshot;
         Ipc::Stream* stream = GetRedirInStream(child, snapshot);
-        if (!child->redirected || stream == nullptr) return -1;
+        if (!child->redirected || stream == nullptr) return InvalidArgument;
         return WriteAllToStream(stream, (const uint8_t*)data, len);
     }
 
-    static int Sys_ChildIoWriteKey(int childPid, const KeyEvent* key) {
+    // Queue a key event (raw bytes, `len` of them) on a redirected child's key
-        if (key == nullptr) return -1;
+    // mailbox. Readers accept both the 9-byte legacy and the full format.
+    static int ChildIoQueueKey(int childPid, const void* key, uint16_t len) {
         auto* child = Sched::GetProcessByPid(childPid);
         if (child == nullptr || child->parentPid != Sched::GetCurrentPid())
-            return SYS_ERR_PERMISSION;
+            return PermissionDenied;
         Ipc::HandleSnapshot snapshot;
         Ipc::Mailbox* mailbox = GetRedirKeyMailbox(child, snapshot);
-        if (!child->redirected || mailbox == nullptr) return -1;
+        if (!child->redirected || mailbox == nullptr) return InvalidArgument;
 
         for (;;) {
             uint64_t observedWake = Sched::ObserveObjectWake(mailbox);
-            int rc = Ipc::MailboxSend(mailbox, 0, key, sizeof(KeyEvent));
+            int rc = Ipc::MailboxSend(mailbox, 0, key, len);
-            if (rc < 0) return -1;
+            if (rc < 0) return GenericFailure;
             if (rc == 0) {
                 Sched::BlockOnObjectSince(mailbox, 0, observedWake);
                 continue;
             }
-            return 0;
+            return Success;
         }
     }
 
+    static int Sys_ChildIoWriteKey(int childPid, const KeyEventV1* key) {
+        if (key == nullptr) return InvalidArgument;
+        KeyEventV1 copy = *key;
+        return ChildIoQueueKey(childPid, &copy, sizeof(copy));
+    }
+
+    // Size-versioned: copies min(size, sizeof(KeyEvent)) bytes (validated
+    // readable by the dispatcher) and zero-fills the rest.
+    static int Sys_ChildIoWriteKeyEx(int childPid, const void* key, uint64_t size) {
+        if (key == nullptr || size < sizeof(KeyEventV1)) return InvalidArgument;
+        KeyEvent ev;
+        memset(&ev, 0, sizeof(ev));
+        memcpy(&ev, key, size < sizeof(ev) ? size : sizeof(ev));
+        return ChildIoQueueKey(childPid, &ev, sizeof(ev));
+    }
+
     static int Sys_ChildIoSetTermsz(int childPid, int cols, int rows) {
         auto* child = Sched::GetProcessByPid(childPid);
         if (child == nullptr || child->parentPid != Sched::GetCurrentPid())
-            return SYS_ERR_PERMISSION;
+            return PermissionDenied;
-        if (!child->redirected) return -1;
+        if (!child->redirected) return InvalidArgument;
         child->termCols = cols;
         child->termRows = rows;
-        return 0;
+        return Success;
     }
 };

kernel/src/Api/IpcSyscall.hpp

@@ -20,15 +20,15 @@ namespace montauk::abi {
     }
 
     static int Sys_StreamCreate(int* outReadHandle, int* outWriteHandle, uint32_t capacity) {
-        if (outReadHandle == nullptr || outWriteHandle == nullptr) return -1;
+        if (outReadHandle == nullptr || outWriteHandle == nullptr) return InvalidArgument;
 
         int readHandle = -1;
         int writeHandle = -1;
-        if (Ipc::CreateStreamHandlePair(capacity, readHandle, writeHandle) < 0) return -1;
+        if (Ipc::CreateStreamHandlePair(capacity, readHandle, writeHandle) < 0) return GenericFailure;
 
         *outReadHandle = readHandle;
         *outWriteHandle = writeHandle;
-        return 0;
+        return Success;
     }
 
     static int Sys_StreamRead(int handle, uint8_t* buffer, int maxLen) {
@@ -40,15 +40,15 @@ namespace montauk::abi {
     }
 
     static int Sys_MailboxCreate(int* outSendHandle, int* outRecvHandle) {
-        if (outSendHandle == nullptr || outRecvHandle == nullptr) return -1;
+        if (outSendHandle == nullptr || outRecvHandle == nullptr) return InvalidArgument;
 
         int sendHandle = -1;
         int recvHandle = -1;
-        if (Ipc::CreateMailboxHandlePair(sendHandle, recvHandle) < 0) return -1;
+        if (Ipc::CreateMailboxHandlePair(sendHandle, recvHandle) < 0) return GenericFailure;
 
         *outSendHandle = sendHandle;
         *outRecvHandle = recvHandle;
-        return 0;
+        return Success;
     }
 
     static int Sys_MailboxSend(int handle, uint32_t msgType, const void* data, uint16_t len, int attachHandle) {

kernel/src/Api/Keyboard.hpp

@@ -12,6 +12,107 @@
 #include "Common.hpp"
 
 namespace montauk::abi {
+    // ---- Key text ----------------------------------------------------------
+    // Keyboard layouts are applied in userspace (desktop, terminal), which send
+    // KeyEvents carrying the Unicode codepoint. Binaries built before that send
+    // 9-byte KeyEventV1s whose `ascii` is a Windows-1252 byte; both kinds share
+    // a redirected child's key mailbox, told apart by message length.
+
+    static uint32_t KeyCp1252ToUnicode(uint8_t b) {
+        static constexpr uint16_t high[32] = {
+            0x20AC, 0x81,   0x201A, 0x0192, 0x201E, 0x2026, 0x2020, 0x2021,
+            0x02C6, 0x2030, 0x0160, 0x2039, 0x0152, 0x8D,   0x017D, 0x8F,
+            0x90,   0x2018, 0x2019, 0x201C, 0x201D, 0x2022, 0x2013, 0x2014,
+            0x02DC, 0x2122, 0x0161, 0x203A, 0x0153, 0x9D,   0x017E, 0x0178,
+        };
+        return (b >= 0x80 && b <= 0x9F) ? high[b - 0x80] : b;
+    }
+
+    // The character a key event types, or 0 for control/navigation keys.
+    static uint32_t KeyTextCodepoint(const KeyEvent& ev) {
+        if (ev.codepoint) return ev.codepoint;
+        uint8_t b = (uint8_t)ev.ascii;
+        if (b < 0x20 || b == 0x7F) return 0;
+        return KeyCp1252ToUnicode(b);
+    }
+
+    static int KeyEncodeUtf8(uint32_t cp, uint8_t* out) {
+        if (cp > 0x10FFFF || (cp >= 0xD800 && cp <= 0xDFFF)) cp = 0xFFFD;
+        if (cp < 0x80) { out[0] = (uint8_t)cp; return 1; }
+        if (cp < 0x800) {
+            out[0] = (uint8_t)(0xC0 | (cp >> 6));
+            out[1] = (uint8_t)(0x80 | (cp & 0x3F));
+            return 2;
+        }
+        if (cp < 0x10000) {
+            out[0] = (uint8_t)(0xE0 | (cp >> 12));
+            out[1] = (uint8_t)(0x80 | ((cp >> 6) & 0x3F));
+            out[2] = (uint8_t)(0x80 | (cp & 0x3F));
+            return 3;
+        }
+        out[0] = (uint8_t)(0xF0 | (cp >> 18));
+        out[1] = (uint8_t)(0x80 | ((cp >> 12) & 0x3F));
+        out[2] = (uint8_t)(0x80 | ((cp >> 6) & 0x3F));
+        out[3] = (uint8_t)(0x80 | (cp & 0x3F));
+        return 4;
+    }
+
+    static void KeyFromDriver(const Drivers::PS2::Keyboard::KeyEvent& k, KeyEvent* out) {
+        memset(out, 0, sizeof(KeyEvent));
+        out->scancode = k.Scancode;
+        out->ascii    = k.Ascii;
+        out->pressed  = k.Pressed;
+        out->shift    = k.Shift;
+        out->ctrl     = k.Ctrl;
+        out->alt      = k.Alt;
+        out->capslock = k.CapsLock;
+        out->altgr    = k.AltGr;
+        out->extended = k.Extended;
+        // The kernel's own scancode table is US ASCII.
+        uint8_t a = (uint8_t)k.Ascii;
+        out->codepoint = (a >= 0x20 && a < 0x7F) ? a : 0;
+    }
+
+    // Receive one key from a redirected child's mailbox as a full KeyEvent,
+    // upgrading legacy 9-byte messages. Returns MailboxRecv's result.
+    static int KeyMailboxRecv(Ipc::Mailbox* mailbox, KeyEvent* out) {
+        memset(out, 0, sizeof(KeyEvent));
+        uint16_t len = sizeof(KeyEvent);
+        int rc = Ipc::MailboxRecv(mailbox, nullptr, out, &len, true);
+        if (rc > 0 && len < sizeof(KeyEvent)) {
+            memset((uint8_t*)out + len, 0, sizeof(KeyEvent) - len);
+            out->codepoint = KeyTextCodepoint(*out);
+        }
+        return rc;
+    }
+
+    // Turn a pressed key into the next getchar byte. Characters above ASCII are
+    // returned as UTF-8, one byte per call; the rest is parked in the process.
+    static bool KeyTakeChar(Sched::Process* proc, const KeyEvent& ev, char* out) {
+        if (!ev.pressed) return false;
+        uint32_t cp = KeyTextCodepoint(ev);
+        if (cp >= 0x80 && proc != nullptr) {
+            int n = KeyEncodeUtf8(cp, proc->getcharUtf8);
+            proc->getcharUtf8Len = (uint8_t)n;
+            proc->getcharUtf8Pos = 1;
+            *out = (char)proc->getcharUtf8[0];
+            return true;
+        }
+        if (ev.ascii != 0 && ev.ascii != KEY_ASCII_SUBSTITUTE) {
+            *out = ev.ascii;
+            return true;
+        }
+        return false;
+    }
+
+    static bool KeyPendingChar(Sched::Process* proc, char* out) {
+        if (proc == nullptr || proc->getcharUtf8Pos >= proc->getcharUtf8Len) return false;
+        *out = (char)proc->getcharUtf8[proc->getcharUtf8Pos++];
+        return true;
+    }
+
+    // ---- Syscalls ------------------------------------------------------------
+
     static bool Sys_IsKeyAvailable() {
         auto* proc = Sched::GetCurrentProcessPtr();
         if (proc && proc->redirected) {
@@ -27,8 +128,8 @@ namespace montauk::abi {
         return Drivers::PS2::Keyboard::IsKeyAvailable();
     }
 
-    static void Sys_GetKey(KeyEvent* outEvent) {
+    // Blocking read of the next key event, in the full format.
-        if (outEvent == nullptr) return;
+    static void ReadKeyEvent(KeyEvent* outEvent) {
         auto* proc = Sched::GetCurrentProcessPtr();
         if (proc && proc->redirected) {
             Ipc::HandleSnapshot snapshot;
@@ -36,8 +137,7 @@ namespace montauk::abi {
             if (mailbox != nullptr) {
                 for (;;) {
                     uint64_t observedWake = Sched::ObserveObjectWake(mailbox);
-                    uint16_t len = sizeof(KeyEvent);
+                    int rc = KeyMailboxRecv(mailbox, outEvent);
-                    int rc = Ipc::MailboxRecv(mailbox, nullptr, outEvent, &len, true);
                     if (rc > 0) return;
                     if (rc < 0) {
                         memset(outEvent, 0, sizeof(KeyEvent));
@@ -47,20 +147,33 @@ namespace montauk::abi {
                 }
             }
         }
-        auto k = Drivers::PS2::Keyboard::GetKey();
+        KeyFromDriver(Drivers::PS2::Keyboard::GetKey(), outEvent);
-        outEvent->scancode = k.Scancode;
+    }
-        outEvent->ascii    = k.Ascii;
+
-        outEvent->pressed  = k.Pressed;
+    // SYS_GETKEY: legacy 9-byte event. A new-format event's first 9 bytes are
-        outEvent->shift    = k.Shift;
+    // already a valid KeyEventV1 (ascii holds the 1252 byte or SUB).
-        outEvent->ctrl     = k.Ctrl;
+    static void Sys_GetKey(KeyEventV1* outEvent) {
-        outEvent->alt      = k.Alt;
+        if (outEvent == nullptr) return;
-        outEvent->capslock = k.CapsLock;
+        KeyEvent ev;
-        outEvent->altgr    = k.AltGr;
+        ReadKeyEvent(&ev);
-        outEvent->extended = k.Extended;
+        memcpy(outEvent, &ev, sizeof(KeyEventV1));
+    }
+
+    // SYS_GETKEY_EX: size-versioned. The caller's size was validated by the
+    // dispatcher (>= sizeof(KeyEventV1), writable).
+    static int Sys_GetKeyEx(void* outEvent, uint64_t size) {
+        KeyEvent ev;
+        ReadKeyEvent(&ev);
+        uint64_t n = size < sizeof(KeyEvent) ? size : sizeof(KeyEvent);
+        memcpy(outEvent, &ev, n);
+        if (size > n) memset((uint8_t*)outEvent + n, 0, size - n);
+        return Success;
     }
 
     static char Sys_GetChar() {
         auto* proc = Sched::GetCurrentProcessPtr();
+        char pending;
+        if (KeyPendingChar(proc, &pending)) return pending;
         if (proc && proc->redirected) {
             Ipc::HandleSnapshot inputSnapshot;
             Ipc::HandleSnapshot mailboxSnapshot;
@@ -78,11 +191,11 @@ namespace montauk::abi {
                     }
 
                     if (mailbox != nullptr) {
-                        KeyEvent ev{};
+                        KeyEvent ev;
-                        uint16_t len = sizeof(ev);
+                        int rc = KeyMailboxRecv(mailbox, &ev);
-                        int rc = Ipc::MailboxRecv(mailbox, nullptr, &ev, &len, true);
                         if (rc > 0) {
-                            if (ev.pressed && ev.ascii != 0) return ev.ascii;
+                            char c;
+                            if (KeyTakeChar(proc, ev, &c)) return c;
                             continue;
                         }
                         if (rc < 0 && input == nullptr) return 0;
@@ -115,6 +228,8 @@ namespace montauk::abi {
      */
     static char Sys_GetCharNb() {
         auto* proc = Sched::GetCurrentProcessPtr();
+        char pending;
+        if (KeyPendingChar(proc, &pending)) return pending;
         if (proc && proc->redirected) {
             Ipc::HandleSnapshot inputSnapshot;
             Ipc::HandleSnapshot mailboxSnapshot;
@@ -127,10 +242,10 @@ namespace montauk::abi {
                 }
 
                 while (mailbox != nullptr) {
-                    KeyEvent ev{};
+                    KeyEvent ev;
-                    uint16_t len = sizeof(ev);
+                    if (KeyMailboxRecv(mailbox, &ev) <= 0) break;
-                    if (Ipc::MailboxRecv(mailbox, nullptr, &ev, &len, true) <= 0) break;
+                    char c;
-                    if (ev.pressed && ev.ascii != 0) return ev.ascii;
+                    if (KeyTakeChar(proc, ev, &c)) return c;
                 }
                 return 0;
             }

kernel/src/Api/LibSyscall.hpp

@@ -91,7 +91,7 @@ namespace montauk::abi {
         }
 
         // Load the library into the process's address space
-        uint64_t loadBias = Sched::ElfLoadLib(path, proc->pml4Phys, libSlot);
+        uint64_t loadBias = Sched::ElfLoadLib(path, proc->pageTableRoot, libSlot);
         if (loadBias == 0) {
             Kt::KernelLogStream(Kt::ERROR, "Lib") << "Failed to load library: " << path;
             g_libLocks[slot].Release();
@@ -115,15 +115,15 @@ namespace montauk::abi {
     // Decrements refcount; actually unmaps when refcount reaches 0.
     inline int Sys_UnloadLib(uint64_t handle) {
         int slot = GetCurrentSlot();
-        if (slot < 0) return -1;
+        if (slot < 0) return GenericFailure;
 
         int libSlot = (int)handle - 1;
-        if (libSlot < 0 || libSlot >= MaxLibsPerProcess) return -1;
+        if (libSlot < 0 || libSlot >= MaxLibsPerProcess) return InvalidArgument;
 
         g_libLocks[slot].Acquire();
         if (!g_libTable[slot][libSlot].inUse) {
             g_libLocks[slot].Release();
-            return -1;
+            return InvalidArgument;
         }
 
         g_libTable[slot][libSlot].refcount--;
@@ -134,7 +134,7 @@ namespace montauk::abi {
 
             auto* proc = Sched::GetCurrentProcessPtr();
             if (proc != nullptr) {
-                Memory::UnmapAndFreeUserRange(proc->pml4Phys, libBase,
+                Memory::UnmapAndFreeUserRange(proc->pageTableRoot, libBase,
                                            (libEnd - libBase) / 0x1000ULL);
             }
 
@@ -144,7 +144,7 @@ namespace montauk::abi {
         }
 
         g_libLocks[slot].Release();
-        return 0;
+        return Success;
     }
 
     // Resolve a symbol in a loaded library.
@@ -203,7 +203,7 @@ namespace montauk::abi {
                 uint64_t libBase = GetLibSlotBase(i);
                 uint64_t libEnd = libBase + Sched::LIB_MAX_SIZE;
 
-                Memory::UnmapAndFreeUserRange(proc->pml4Phys, libBase,
+                Memory::UnmapAndFreeUserRange(proc->pageTableRoot, libBase,
                                            (libEnd - libBase) / 0x1000ULL);
 
                 g_libTable[slot][i].inUse = false;

kernel/src/Api/Net.hpp

@@ -9,7 +9,7 @@
 #pragma once
 #include <cstdint>
 #include <Sched/Scheduler.hpp>
-#include <Timekeeping/ApicTimer.hpp>
+#include <Timekeeping/Timer.hpp>
 #include <Net/Icmp.hpp>
 #include <Net/Dns.hpp>
 #include <Net/Ipv4.hpp>
@@ -18,6 +18,7 @@
 #include <Net/NetIf.hpp>
 #include <Drivers/Net/E1000.hpp>
 #include <Drivers/Net/E1000E.hpp>
+#include <Drivers/Net/Genet.hpp>
 #include <Drivers/Net/Wifi/Wifi.hpp>
 
 #include "Syscall.hpp"
@@ -157,6 +158,10 @@ namespace montauk::abi {
             out->rxPackets = Drivers::Net::E1000E::GetRxPacketCount();
             out->txPackets = Drivers::Net::E1000E::GetTxPacketCount();
             CopyNetStatusDriver(out->driver, "Intel e1000e");
+        } else if (Drivers::Net::Genet::IsInitialized()) {
+            out->rxPackets = Drivers::Net::Genet::GetRxPacketCount();
+            out->txPackets = Drivers::Net::Genet::GetTxPacketCount();
+            CopyNetStatusDriver(out->driver, "Broadcom GENET");
         }
     }
 
@@ -164,7 +169,7 @@ namespace montauk::abi {
     // wired and wireless status apart: the IP configuration is global, so the
     // only way to say which interface owns it is `active`.
     static int Sys_NetIfs(NetIfInfo* out, int maxCount) {
-        if (out == nullptr || maxCount <= 0) return -1;
+        if (out == nullptr || maxCount <= 0) return InvalidArgument;
 
         const auto* active = ::Net::NetIf::Active();
         int count = ::Net::NetIf::Count();
@@ -193,12 +198,12 @@ namespace montauk::abi {
     }
 
     static int Sys_SetNetCfg(const NetCfg* in) {
-        if (in == nullptr) return -1;
+        if (in == nullptr) return InvalidArgument;
         Net::SetIpAddress(in->ipAddress);
         Net::SetSubnetMask(in->subnetMask);
         Net::SetGateway(in->gateway);
         Net::SetDnsServer(in->dnsServer);
-        return 0;
+        return Success;
     }
 
     // ---- DNS resolve ----

kernel/src/Api/Power.hpp

@@ -8,12 +8,13 @@
 #include <cstdint>
 #include <Efi/UEFI.hpp>
 #include <Memory/Paging.hpp>
-#include <Io/IoPort.hpp>
 #include <Graphics/Framebuffer.hpp>
 #include <ACPI/AcpiShutdown.hpp>
 #include <ACPI/AcpiSleep.hpp>
-#include <ACPI/CpuIdle.hpp>
+#include <ACPI/Battery.hpp>
-#include <Hal/CpuPower.hpp>
+#include <Arch/Cpu.hpp>
+#include <Arch/CpuPower.hpp>
+#include <Arch/Platform.hpp>
 #include <Sched/Scheduler.hpp>
 #include <Terminal/Terminal.hpp>
 #include <CppLib/Stream.hpp>
@@ -48,22 +49,20 @@ namespace montauk::abi {
         Kt::KernelLogStream(Kt::INFO, "Power") << "pid " << kcp::dec
             << (uint64_t)Sched::GetCurrentPid()
             << " posted graceful power request " << (uint64_t)action;
-        return 0;
+        return Success;
     }
 
     static void Sys_Reset() {
         Kt::KernelLogStream(Kt::INFO, "Power") << "pid " << kcp::dec
             << (uint64_t)Sched::GetCurrentPid() << " entered final reboot stage";
         if (Efi::g_ResetSystem) {
-            /* Switch to kernel PML4 which has identity-mapped UEFI runtime regions */
+            /* Switch to the kernel address space, which has identity-mapped UEFI runtime regions */
-            Memory::VMM::LoadCR3(Memory::VMM::g_paging->PML4);
+            Memory::VMM::ActivateKernelAddressSpace();
             Efi::g_ResetSystem(Efi::EfiResetCold, 0, 0, nullptr);
         }
 
-        /* Fallback: triple fault via null IDT */
+        /* Fallback: architecture-specific reset */
-        struct [[gnu::packed]] { uint16_t limit; uint64_t base; } nullIdt = {0, 0};
+        Arch::Reboot();
-        asm volatile("lidt %0; int $0x03" :: "m"(nullIdt));
-        __builtin_unreachable();
     }
 
     static void Sys_Shutdown() {
@@ -74,20 +73,22 @@ namespace montauk::abi {
             Hal::AcpiShutdown::Shutdown();
         }
 
+        /* Hardware-reduced platforms (PSCI on ARM) */
+        Arch::PowerOff();
+
         /* Fallback: EFI ResetSystem */
         if (Efi::g_ResetSystem) {
-            Memory::VMM::LoadCR3(Memory::VMM::g_paging->PML4);
+            Memory::VMM::ActivateKernelAddressSpace();
             Efi::g_ResetSystem(Efi::EfiResetShutdown, 0, 0, nullptr);
         }
 
         /* Last resort: halt the CPU */
-        asm volatile("cli; hlt");
+        Arch::HaltForever();
-        __builtin_unreachable();
     }
 
     static int64_t Sys_Suspend() {
         if (!Hal::AcpiSleep::IsS3Available()) {
-            return -1; // S3 not supported
+            return UnsupportedOp; // S3 not supported
         }
         return (int64_t)Hal::AcpiSleep::Suspend();
     }
@@ -95,9 +96,9 @@ namespace montauk::abi {
     // SYS_POWERINFO. Fills a CPU power/thermal snapshot. Returns -1 when the
     // CPU exposes no power management features (e.g. under QEMU).
     static int64_t Sys_PowerInfo(PowerInfo* out) {
-        Hal::CpuPower::Snapshot s{};
+        Arch::CpuPower::Snapshot s{};
-        if (!Hal::CpuPower::GetSnapshot(s)) {
+        if (!Arch::CpuPower::GetSnapshot(s)) {
-            return -1;
+            return NoDevice;
         }
 
         out->hwpActive   = s.HwpActive ? 1 : 0;
@@ -111,7 +112,33 @@ namespace montauk::abi {
         out->baseMHz     = s.BaseMHz;
         out->maxMHz      = s.MaxMHz;
         out->effMHz      = s.EffMHz;
-        out->apIdleHint  = Hal::CpuIdle::GetApIdleHint();
+        out->apIdleHint  = s.ApIdleHint;
-        return 0;
+        return Success;
+    }
+
+    // SYS_BATTERYINFO. Fills a battery snapshot. Returns -1 on a machine with
+    // no ACPI battery device, which is how a desktop-class caller learns not to
+    // draw an indicator at all.
+    static int64_t Sys_BatteryInfo(BatteryInfo* out) {
+        Hal::Battery::Status s{};
+        if (!Hal::Battery::GetStatus(s)) {
+            return NoDevice;
+        }
+
+        out->present            = s.Present ? 1 : 0;
+        out->charging           = s.Charging ? 1 : 0;
+        out->discharging        = s.Discharging ? 1 : 0;
+        out->critical           = s.Critical ? 1 : 0;
+        out->acOnline           = s.AcOnline ? 1 : 0;
+        out->hasAcAdapter       = s.HasAcAdapter ? 1 : 0;
+        out->percent            = s.Percent;
+        out->powerUnitMilliAmps = s.PowerUnitMilliAmps ? 1 : 0;
+        out->remainingCapacity  = s.RemainingCapacity;
+        out->fullCapacity       = s.FullCapacity;
+        out->designCapacity     = s.DesignCapacity;
+        out->presentRate        = s.PresentRate;
+        out->voltageMv          = s.VoltageMv;
+        out->minutesRemaining   = s.MinutesRemaining;
+        return Success;
     }
 };

kernel/src/Api/Process.hpp

@@ -7,7 +7,7 @@
 
 #pragma once
 #include <Sched/Scheduler.hpp>
-#include <Timekeeping/ApicTimer.hpp>
+#include <Timekeeping/Timer.hpp>
 #include <Memory/Paging.hpp>
 #include <Memory/PageFrameAllocator.hpp>
 #include <Memory/HHDM.hpp>
@@ -49,13 +49,13 @@ namespace montauk::abi {
     // stream, whether or not it also carries a capability grant.  The child is
     // created suspended (startReady == false) so its first instruction cannot
     // run before the channels exist, and is started here once they do.
-    // Returns childPid, or kills the child and returns -1 on failure.
+    // Returns childPid, or kills the child and returns GenericFailure.
     static int InheritRedirection(int childPid, Sched::Process* parent, int parentSlot) {
         auto* child = Sched::GetProcessByPid(childPid);
         int childSlot = Ipc::SlotForPid(childPid);
         if (child == nullptr || childSlot < 0 || parentSlot < 0) {
             Sched::KillProcess(childPid);
-            return -1;
+            return GenericFailure;
         }
 
         child->ioOutHandle = DuplicateHandleBetweenSlots(parentSlot, parent->ioOutHandle, childSlot);
@@ -65,7 +65,7 @@ namespace montauk::abi {
         if (child->ioOutHandle < 0 || child->ioInHandle < 0 || child->ioKeyHandle < 0 ||
             !ConfigureRedirWaitsetForSlot(childSlot, child)) {
             Sched::KillProcess(childPid);
-            return -1;
+            return GenericFailure;
         }
 
         child->redirected = true;
@@ -74,7 +74,7 @@ namespace montauk::abi {
         child->termRows = parent->termRows;
         if (Sched::StartProcess(childPid) < 0) {
             Sched::KillProcess(childPid);
-            return -1;
+            return GenericFailure;
         }
 
         return childPid;
@@ -83,7 +83,7 @@ namespace montauk::abi {
     static int Sys_Spawn(const char* path, const char* args,
                          const char* environment = nullptr, uint32_t environmentLength = 0) {
         char resolved[256];
-        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return -1;
+        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return InvalidArgument;
 
         auto* parent = Sched::GetCurrentProcessPtr();
         int parentSlot = Ipc::CurrentSlot();
@@ -101,7 +101,7 @@ namespace montauk::abi {
     static int Sys_SpawnCaps(const char* path, const char* args,
                              const char* user, const SpawnCapabilities* requested) {
         auto* parent = Sched::GetCurrentProcessPtr();
-        if (parent == nullptr || requested == nullptr) return -1;
+        if (parent == nullptr || requested == nullptr) return InvalidArgument;
 
         // Snapshot all security-sensitive userspace inputs before evaluating
         // them.  This prevents another thread from changing a mask or owner
@@ -111,7 +111,7 @@ namespace montauk::abi {
         const char* userOverride = nullptr;
         if (user != nullptr) {
             if (!Sched::HasCapability(CAP_USER_ADMIN))
-                return SYS_ERR_PERMISSION;
+                return CapabilityDenied;
             int i = 0;
             for (; i < 31 && user[i]; i++) childUser[i] = user[i];
             childUser[i] = '\0';
@@ -122,11 +122,11 @@ namespace montauk::abi {
         // possessing a capability is insufficient to pass it: the parent must
         // also hold it in its delegable set.
         if (!ValidCapabilityDelegation(copy, parent->delegableCaps)) {
-            return SYS_ERR_PERMISSION;
+            return CapabilityDenied;
         }
 
         char resolved[256];
-        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return -1;
+        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return InvalidArgument;
 
         int parentSlot = Ipc::CurrentSlot();
         bool inheritRedirection = parent->redirected;
@@ -140,10 +140,19 @@ namespace montauk::abi {
         return childPid;
     }
 
+    // The calling process's own capability masks.
+    static int Sys_GetCaps(SpawnCapabilities* out) {
+        auto* proc = Sched::GetCurrentProcessPtr();
+        if (proc == nullptr || out == nullptr) return InvalidArgument;
+        out->effective = proc->effectiveCaps;
+        out->delegable = proc->delegableCaps;
+        return Success;
+    }
+
     // Copy the absolute path this process was spawned from (argv[0]).
     static int Sys_GetExecPath(char* buf, uint64_t maxLen) {
         auto* proc = Sched::GetCurrentProcessPtr();
-        if (proc == nullptr || buf == nullptr || maxLen == 0) return -1;
+        if (proc == nullptr || buf == nullptr || maxLen == 0) return InvalidArgument;
         int i = 0;
         for (; i < (int)maxLen - 1 && proc->name[i]; i++) {
             buf[i] = proc->name[i];
@@ -154,7 +163,7 @@ namespace montauk::abi {
 
     static int Sys_GetArgs(char* buf, uint64_t maxLen) {
         auto* proc = Sched::GetCurrentProcessPtr();
-        if (proc == nullptr || buf == nullptr || maxLen == 0) return -1;
+        if (proc == nullptr || buf == nullptr || maxLen == 0) return InvalidArgument;
         int i = 0;
         for (; i < (int)maxLen - 1 && proc->args[i]; i++) {
             buf[i] = proc->args[i];
@@ -165,8 +174,8 @@ namespace montauk::abi {
 
     static int Sys_GetEnviron(char* buf, uint64_t maxLen) {
         auto* proc = Sched::GetCurrentProcessPtr();
-        if (proc == nullptr || buf == nullptr || maxLen == 0) return -1;
+        if (proc == nullptr || buf == nullptr || maxLen == 0) return InvalidArgument;
-        if (maxLen < proc->environmentLength) return -2;
+        if (maxLen < proc->environmentLength) return BufferTooSmall;
         for (uint32_t i = 0; i < proc->environmentLength; i++)
             buf[i] = proc->environment[i];
         return (int)proc->environmentLength;
@@ -175,10 +184,10 @@ namespace montauk::abi {
     static int Sys_SetEnviron(const char* buf, uint64_t len) {
         auto* proc = Sched::GetCurrentProcessPtr();
         if (proc == nullptr || buf == nullptr || len == 0 ||
-            len > Sched::Process::EnvironmentBytes || buf[len - 1] != '\0') return -1;
+            len > Sched::Process::EnvironmentBytes || buf[len - 1] != '\0') return InvalidArgument;
         for (uint64_t i = 0; i < len; i++) proc->environment[i] = buf[i];
         proc->environmentLength = (uint32_t)len;
-        return 0;
+        return Success;
     }
 
     static int Sys_ProcList(ProcInfo* buf, int maxCount) {
@@ -204,7 +213,6 @@ namespace montauk::abi {
             }
             buf[count].heapUsed = Sched::g_allocatedPages[i] * 0x1000;
             buf[count].cpuTimeMs = proc->cpuTimeMs;
-            buf[count].permittedCaps = proc->permittedCaps;
             buf[count].effectiveCaps = proc->effectiveCaps;
             buf[count].delegableCaps = proc->delegableCaps;
             count++;
@@ -212,7 +220,26 @@ namespace montauk::abi {
         return count;
     }
 
+    // Flag every thread of the caller's process with TerminatedAborted, then
+    // exit this one. A sibling's ExitProcess ends only itself; the main
+    // thread and the other siblings follow via killPending.
+    static void Sys_Abort() {
+        Sched::KillProcess(Sched::GetCurrentPid(), TerminatedAborted);
+        Sched::ExitProcess();
+    }
+
+    // Like Sys_Abort, but a normal exit with the caller's code: every thread
+    // goes, and waiters see exitCode & 0xFF, the same as SYS_EXIT on the main
+    // thread.
+    static void Sys_ExitGroup(int exitCode) {
+        Sched::KillProcess(Sched::GetCurrentPid(), exitCode & 0xFF);
+        Sched::ExitProcess();
+    }
+
     static int Sys_Kill(int pid) {
+        // A process may always kill itself (kill(getpid(), ...)); waiters
+        // see TerminatedKilled. abort() uses SYS_ABORT instead.
+        if (pid == Sched::GetCurrentPid()) return Sched::KillProcess(pid);
         if (!Sched::HasCapability(CAP_PROCESS_ADMIN)) {
             int ancestor = pid;
             bool descendant = false;
@@ -225,7 +252,7 @@ namespace montauk::abi {
                 }
                 ancestor = target->parentPid;
             }
-            if (!descendant) return SYS_ERR_PERMISSION;
+            if (!descendant) return PermissionDenied;
         }
         return Sched::KillProcess(pid);
     }
@@ -239,19 +266,19 @@ namespace montauk::abi {
     }
 
     static int Sys_SetUser(int pid, const char* name) {
-        if (name == nullptr) return -1;
+        if (name == nullptr) return InvalidArgument;
         auto* target = Sched::GetProcessByPid(pid);
-        if (target == nullptr) return -1;
+        if (target == nullptr) return InvalidArgument;
         int i = 0;
         for (; i < 31 && name[i]; i++)
             target->user[i] = name[i];
         target->user[i] = '\0';
-        return 0;
+        return Success;
     }
 
     static int Sys_GetUser(char* buf, uint64_t maxLen) {
         auto* proc = Sched::GetCurrentProcessPtr();
-        if (proc == nullptr || buf == nullptr || maxLen == 0) return -1;
+        if (proc == nullptr || buf == nullptr || maxLen == 0) return InvalidArgument;
         int i = 0;
         for (; i < (int)maxLen - 1 && proc->user[i]; i++)
             buf[i] = proc->user[i];
@@ -261,15 +288,15 @@ namespace montauk::abi {
 
     static int Sys_GetCwd(char* buf, uint64_t maxLen) {
         auto* proc = Sched::GetCurrentProcessPtr();
-        if (proc == nullptr || buf == nullptr || maxLen == 0) return -1;
+        if (proc == nullptr || buf == nullptr || maxLen == 0) return InvalidArgument;
 
         const char* cwd = proc->cwd[0] ? proc->cwd : "0:/";
         uint64_t cwdLen = 0;
         while (cwd[cwdLen] != '\0') cwdLen++;
 
-        // Reserve -1 for non-size failures so libc can distinguish them
+        // BufferTooSmall tells libc the call can be retried with a larger
-        // from a buffer that can be retried with a larger allocation.
+        // allocation (ERANGE), unlike the other failures.
-        if (maxLen <= cwdLen) return -2;
+        if (maxLen <= cwdLen) return BufferTooSmall;
 
         for (uint64_t i = 0; i <= cwdLen; i++) buf[i] = cwd[i];
         return (int)cwdLen;
@@ -294,12 +321,20 @@ namespace montauk::abi {
         return Sched::GetCurrentTid();
     }
 
+    static int Sys_FutexWait(uint64_t addr, uint32_t expected, uint64_t timeoutNs) {
+        return Sched::FutexWait(addr, expected, timeoutNs);
+    }
+
+    static int Sys_FutexWake(uint64_t addr, uint32_t count) {
+        return Sched::FutexWake(addr, count);
+    }
+
     static int Sys_Chdir(const char* path) {
         auto* proc = Sched::GetCurrentProcessPtr();
-        if (proc == nullptr || path == nullptr) return -1;
+        if (proc == nullptr || path == nullptr) return InvalidArgument;
 
         char resolved[256];
-        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return -1;
+        if (!ResolveProcessPath(path, resolved, sizeof(resolved))) return InvalidArgument;
 
         // ReadDir doubles as the directory-existence probe: it fails
         // for nonexistent paths and for regular files (directories are
@@ -307,12 +342,12 @@ namespace montauk::abi {
         // would reject every valid target).
         {
             const char* entries[1];
-            if (Fs::Vfs::VfsReadDir(resolved, entries, 1) < 0) return -1;
+            if (Fs::Vfs::VfsReadDir(resolved, entries, 1) < 0) return NoSuchPath;
         }
 
         int i = 0;
         for (; i < 255 && resolved[i]; i++) proc->cwd[i] = resolved[i];
         proc->cwd[i] = '\0';
-        return 0;
+        return Success;
     }
 };

kernel/src/Api/Random.hpp

@@ -6,23 +6,23 @@
 
 #pragma once
 #include <cstdint>
+#include <Arch/Cpu.hpp>
 
 namespace montauk::abi {
 
     // ---- Random number generation ----
-    // Uses RDTSC mixed with xorshift64* PRNG for entropy.
+    // Uses the CPU timestamp counter mixed with xorshift64* PRNG for entropy.
-    // RDRAND is intentionally avoided: some firmware disables the RDRAND
+    // RDRAND is intentionally avoided on x86: some firmware disables the
-    // hardware unit while CPUID still advertises support (bit 30 of ECX),
+    // RDRAND hardware unit while CPUID still advertises support (bit 30 of
-    // causing #UD on real hardware. RDTSC-based entropy is sufficient for
+    // ECX), causing #UD on real hardware. Timestamp-based entropy is
-    // seeding BearSSL's PRNG for TLS session keys.
+    // sufficient for seeding BearSSL's PRNG for TLS session keys.
 
     static int64_t Sys_GetRandom(uint8_t* buf, uint64_t len) {
-        uint64_t tsc;
+        uint64_t tsc = Arch::ReadTimestamp();
-        asm volatile("rdtsc; shl $32, %%rdx; or %%rdx, %%rax" : "=a"(tsc) :: "rdx");
         uint64_t state = tsc;
 
         for (uint64_t i = 0; i < len; i += 8) {
-            asm volatile("rdtsc; shl $32, %%rdx; or %%rdx, %%rax" : "=a"(tsc) :: "rdx");
+            tsc = Arch::ReadTimestamp();
             state ^= tsc;
             state ^= state >> 12;
             state ^= state << 25;

kernel/src/Api/Storage.hpp

@@ -29,7 +29,7 @@ namespace montauk::abi {
     // Fill a user buffer with partition info. Returns the number of entries written.
     static int Sys_PartList(PartInfo* buf, int maxCount) {
         if (buf == nullptr || maxCount <= 0) return 0;
-        if (!UserMemory::Range((uint64_t)buf, (uint64_t)maxCount * sizeof(PartInfo), true)) return -1;
+        if (!UserMemory::Range((uint64_t)buf, (uint64_t)maxCount * sizeof(PartInfo), true)) return BadPointer;
 
         int total = Drivers::Storage::Gpt::GetPartitionCount();
         int count = total < maxCount ? total : maxCount;
@@ -57,47 +57,89 @@ namespace montauk::abi {
         return count;
     }
 
-    // Read sectors from a block device. Returns bytes read, or -1 on error.
+    // Fill a user buffer with one entry per mounted volume. Returns the number
+    // of entries written, or BadPointer for a bad buffer.
+    static int Sys_VolInfo(VolInfo* buf, int maxCount) {
+        if (buf == nullptr || maxCount <= 0) return 0;
+        if (!UserMemory::Range((uint64_t)buf, (uint64_t)maxCount * sizeof(VolInfo), true)) return BadPointer;
+
+        int drives[Fs::Vfs::MaxDrives];
+        int n = Fs::Vfs::VfsDriveList(drives, Fs::Vfs::MaxDrives);
+        if (n > maxCount) n = maxCount;
+
+        for (int i = 0; i < n; i++) {
+            VolInfo& v = buf[i];
+            v = VolInfo{};
+            v.drive = drives[i];
+            v.blockDev = -1;
+            v.partIndex = -1;
+
+            Fs::Vfs::VfsDriveLabel(v.drive, v.label, sizeof(v.label));
+            Fs::Vfs::VfsDriveFsType(v.drive, v.fsType, sizeof(v.fsType));
+
+            v.partIndex = Fs::FsProbe::GetPartitionForDrive(v.drive);
+            if (v.partIndex < 0) continue;   // ramdisk: no backing partition
+
+            auto* part = Drivers::Storage::Gpt::GetPartition(v.partIndex);
+            if (part == nullptr) continue;
+
+            v.blockDev = part->BlockDevIndex;
+            v.startLba = part->StartLba;
+            v.sectorCount = part->SectorCount;
+            dl_strcpy_s(v.partName, part->Name, sizeof(v.partName));
+
+            auto* dev = Drivers::Storage::GetBlockDevice(v.blockDev);
+            if (dev == nullptr) continue;
+
+            v.kind = dev->Kind;
+            v.sectorSize = dev->SectorSize;
+            dl_strcpy_s(v.model, dev->Model, sizeof(v.model));
+        }
+
+        return n;
+    }
+
+    // Read sectors from a block device. Returns bytes read, or a status code.
     static int64_t Sys_DiskRead(int blockDev, uint64_t lba, uint32_t count, void* buffer) {
-        if (buffer == nullptr || count == 0) return -1;
+        if (buffer == nullptr || count == 0) return InvalidArgument;
-        if (count > 128) return -1; // same limit as AHCI
+        if (count > 128) return InvalidArgument; // same limit as AHCI
 
         auto* dev = Drivers::Storage::GetBlockDevice(blockDev);
-        if (!dev) return -1;
+        if (!dev) return NoDevice;
 
-        if (lba >= dev->SectorCount || count > dev->SectorCount - lba) return -1;
+        if (lba >= dev->SectorCount || count > dev->SectorCount - lba) return InvalidArgument;
-        if (!UserMemory::Range((uint64_t)buffer, (uint64_t)count * dev->SectorSize, true)) return -1;
+        if (!UserMemory::Range((uint64_t)buffer, (uint64_t)count * dev->SectorSize, true)) return BadPointer;
 
-        if (!dev->ReadSectors(dev->Ctx, lba, count, buffer)) return -1;
+        if (!dev->ReadSectors(dev->Ctx, lba, count, buffer)) return IoFailure;
 
         return (int64_t)(count * dev->SectorSize);
     }
 
-    // Write sectors to a block device. Returns bytes written, or -1 on error.
+    // Write sectors to a block device. Returns bytes written, or a status code.
     static int64_t Sys_DiskWrite(int blockDev, uint64_t lba, uint32_t count, const void* buffer) {
-        if (buffer == nullptr || count == 0) return -1;
+        if (buffer == nullptr || count == 0) return InvalidArgument;
-        if (count > 128) return -1;
+        if (count > 128) return InvalidArgument;
 
         auto* dev = Drivers::Storage::GetBlockDevice(blockDev);
-        if (!dev) return -1;
+        if (!dev) return NoDevice;
 
-        if (lba >= dev->SectorCount || count > dev->SectorCount - lba) return -1;
+        if (lba >= dev->SectorCount || count > dev->SectorCount - lba) return InvalidArgument;
-        if (!UserMemory::Range((uint64_t)buffer, (uint64_t)count * dev->SectorSize, false)) return -1;
+        if (!UserMemory::Range((uint64_t)buffer, (uint64_t)count * dev->SectorSize, false)) return BadPointer;
 
-        if (!dev->WriteSectors(dev->Ctx, lba, count, buffer)) return -1;
+        if (!dev->WriteSectors(dev->Ctx, lba, count, buffer)) return IoFailure;
 
         return (int64_t)(count * dev->SectorSize);
     }
 
-    // Initialize a new GPT on a block device. Returns 0 on success, -1 on error.
+    // Initialize a new GPT on a block device. Returns 0 on success, GenericFailure on error.
     static int64_t Sys_GptInit(int blockDev) {
         Fs::FsProbe::UnmountPartitionsForBlockDevice(blockDev);
         return (int64_t)Drivers::Storage::Gpt::InitializeGpt(blockDev);
     }
 
-    // Add a partition to an existing GPT. Returns partition index on success, -1 on error.
+    // Add a partition to an existing GPT. Returns partition index on success, GenericFailure on error.
     static int64_t Sys_GptAdd(const GptAddParams* params) {
-        if (params == nullptr) return -1;
+        if (params == nullptr) return InvalidArgument;
 
         Drivers::Storage::Gpt::Guid typeGuid;
         __builtin_memcpy(&typeGuid, &params->typeGuid, sizeof(typeGuid));
@@ -107,7 +149,7 @@ namespace montauk::abi {
             typeGuid, params->name);
     }
 
-    // Mount a partition as a VFS drive. Returns 0 on success, -1 on error.
+    // Mount a partition as a VFS drive. Returns 0 on success, GenericFailure on error.
     static int64_t Sys_FsMount(int partIndex, int driveNum) {
         return (int64_t)Fs::FsProbe::MountPartition(partIndex, driveNum);
     }
@@ -121,12 +163,51 @@ namespace montauk::abi {
         return (int64_t)Fs::FsProbe::SyncAndUnmountAll();
     }
 
-    // Format a partition with a filesystem. Returns 0 on success, -1 on error.
+    // Grow a mounted volume to fill the free space after its partition: the
+    // MBR entry first, only as far as the filesystem can use, then the
+    // filesystem a slice per call. Interrupted grows resume from the next
+    // call (or boot), since a partition larger than its filesystem is valid.
+    static int64_t Sys_FsGrow(int drive, FsGrowStatus* out, bool apply) {
+        int partIndex = Fs::FsProbe::GetPartitionForDrive(drive);
+        auto* part = Drivers::Storage::Gpt::GetPartition(partIndex);
+        if (part == nullptr) return UnsupportedOp;   // ramdisk, or no such drive
+
+        uint64_t maxSectors = Drivers::Storage::Gpt::GetMbrPartitionMaxSectors(partIndex);
+        if (maxSectors < part->SectorCount) maxSectors = part->SectorCount;
+
+        Fs::Vfs::GrowStatus gs{};
+        int rc = Fs::Vfs::VfsGrow(drive, maxSectors, false, &gs);
+        if (rc == -2) return UnsupportedOp;
+        if (rc < 0) return IoFailure;
+
+        // Leave whatever the filesystem cannot use unpartitioned, with the end
+        // on a 1 MiB boundary like the partitions mk-rpi4-sd.sh makes.
+        uint64_t wanted = (part->StartLba + (gs.targetBytes + 511) / 512 + 2047) / 2048 * 2048
+                        - part->StartLba;
+        if (wanted > maxSectors) wanted = maxSectors;
+
+        if (apply && wanted > part->SectorCount) {
+            if (Drivers::Storage::Gpt::ResizeMbrPartition(partIndex, wanted) != 0) return IoFailure;
+        }
+
+        rc = apply ? Fs::Vfs::VfsGrow(drive, part->SectorCount, true, &gs) : 0;
+        if (rc == -2) return UnsupportedOp;
+        if (rc < 0) return IoFailure;
+
+        out->fsBytes = gs.fsBytes;
+        out->partBytes = part->SectorCount * 512;
+        out->targetBytes = gs.targetBytes;
+        out->stepsDone = gs.stepsDone;
+        out->stepsTotal = gs.stepsTotal;
+        return rc;
+    }
+
+    // Format a partition with a filesystem. Returns 0 on success, GenericFailure on error.
     static int64_t Sys_FsFormat(const FsFormatParams* params) {
-        if (params == nullptr) return -1;
+        if (params == nullptr) return InvalidArgument;
 
         auto* part = Drivers::Storage::Gpt::GetPartition(params->partIndex);
-        if (!part) return -1;
+        if (!part) return InvalidArgument;
 
         switch (params->fsType) {
         case FS_TYPE_FAT32:
@@ -138,7 +219,7 @@ namespace montauk::abi {
                 part->BlockDevIndex, part->StartLba, part->SectorCount,
                 params->label[0] ? params->label : nullptr);
         default:
-            return -1; // unknown filesystem type
+            return UnsupportedOp; // unknown filesystem type
         }
     }
 

kernel/src/Api/Syscall.cpp

@@ -6,18 +6,17 @@
 
 #include "Syscall.hpp"
 #include <Terminal/Terminal.hpp>
-#include <Hal/MSR.hpp>
-#include <Hal/GDT.hpp>
 
 /* For common functions used by multiple syscall implementations*/
 #include "Common.hpp"
+#include <Sched/Consent.hpp>
 
 /* Syscall impl. includes */
 #include "Process.hpp"    // SYS_EXIT, SYS_YIELD, SYS_SLEEP_MS, SYS_GETPID, SYS_WAITPID, SYS_SPAWN, SYS_GETARGS, SYS_PROCLIST, SYS_KILL
 #include "Terminal.hpp"   // SYS_PRINT, SYS_PUTCHAR
 #include "Filesystem.hpp" // SYS_OPEN, SYS_READ, SYS_GETSIZE, SYS_CLOSE, SYS_READDIR, SYS_FWRITE, SYS_FCREATE
 #include "Heap.hpp"       // SYS_ALLOC, SYS_FREE
-#include "Time.hpp"       // SYS_GETTICKS, SYS_GETMILLISECONDS, SYS_GETTIME
+#include "Time.hpp"       // SYS_GETTICKS, SYS_GETMILLISECONDS, SYS_GETTIME, SYS_GETUNIXTIME
 #include "Keyboard.hpp"   // SYS_ISKEYAVAILABLE, SYS_GETKEY, SYS_GETCHAR
 #include "Info.hpp"       // SYS_GETINFO
 #include "Graphics.hpp"   // SYS_FBINFO, SYS_FBMAP, SYS_TERMSIZE, SYS_TERMSCALE
@@ -29,8 +28,8 @@
 #include "MemInfo.hpp"    // SYS_MEMSTATS
 #include "Device.hpp"     // SYS_DEVLIST, SYS_DISKINFO
 #include "Input.hpp"      // SYS_INPUT_WAIT
-#include "Storage.hpp"    // SYS_PARTLIST, SYS_DISKREAD, SYS_DISKWRITE
+#include "Storage.hpp"    // SYS_PARTLIST, SYS_VOLINFO, SYS_DISKREAD, SYS_DISKWRITE
-#include "Window.hpp"     // SYS_WINCREATE, SYS_WINDESTROY, SYS_WINPRESENT, SYS_WINPOLL, SYS_WINENUM, SYS_WINMAP, SYS_WINSENDEVENT, SYS_WINRESIZE, SYS_WINSETCURSOR, SYS_WINSETFLAGS, SYS_WINSETSCALE, SYS_WINGETSCALE
+#include "Window.hpp"     // SYS_WIN_REQUEST*, SYS_WIN_DRAG_*, SYS_WINCREATE, SYS_WINDESTROY, SYS_WINPRESENT, SYS_WINPOLL, SYS_WINENUM, SYS_WINMAP, SYS_WINSENDEVENT, SYS_WINRESIZE, SYS_WINSETCURSOR, SYS_WINSETFLAGS, SYS_WINSETSCALE, SYS_WINGETSCALE
 #include "Audio.hpp"      // SYS_AUDIOOPEN, SYS_AUDIOCLOSE, SYS_AUDIOWRITE, SYS_AUDIOCTL
 #include "BluetoothSyscall.hpp" // SYS_BTSCAN, SYS_BTCONNECT, SYS_BTDISCONNECT, SYS_BTLIST, SYS_BTINFO
 #include "Usb.hpp"        // generic process-owned USB interface access
@@ -38,12 +37,9 @@
 #include "IpcSyscall.hpp" // SYS_DUPHANDLE, SYS_WAIT_HANDLE, SYS_STREAM_CREATE, SYS_STREAM_READ, SYS_STREAM_WRITE, SYS_MAILBOX_CREATE, SYS_MAILBOX_SEND, SYS_MAILBOX_RECV, SYS_WAITSET_CREATE, SYS_WAITSET_ADD, SYS_WAITSET_REMOVE, SYS_WAITSET_WAIT, SYS_PROC_OPEN, SYS_SURFACE_CREATE, SYS_SURFACE_MAP, SYS_SURFACE_RESIZE
 #include "LibSyscall.hpp" // SYS_LOAD_LIB, SYS_UNLOAD_LIB, SYS_DLSYM
 #include "CrashReportSyscall.hpp" // SYS_CRASH_REPORT
-#include "Clipboard.hpp" // SYS_CLIPBOARD_*
+#include "Clipboard.hpp" // SYS_CLIPBOARD_* (text and multi-format)
 #include "UserMemory.hpp"
 
-// Assembly entry point
-extern "C" void SyscallEntry();
-
 namespace montauk::abi {
 
     static constexpr uint64_t kMaxPrintableStringBytes = 4096;
@@ -61,54 +57,54 @@ namespace montauk::abi {
         switch (frame->syscall_nr) {
             case SYS_EXIT: {
                 Sys_Exit((int)frame->arg1);
-                return 0;
+                return Success;
             }
             case SYS_YIELD:
                 Sys_Yield();
-                return 0;
+                return Success;
             case SYS_SLEEP_MS:
                 Sys_SleepMs(frame->arg1);
-                return 0;
+                return Success;
             case SYS_GETPID:
                 return (int64_t)Sys_GetPid();
             case SYS_PRINT:
-                if (!UserMemory::String(frame->arg1, kMaxPrintableStringBytes)) return -1;
+                if (!UserMemory::String(frame->arg1, kMaxPrintableStringBytes)) return BadPointer;
                 Sys_Print((const char*)frame->arg1);
-                return 0;
+                return Success;
             case SYS_PUTCHAR:
                 Sys_Putchar((char)frame->arg1);
-                return 0;
+                return Success;
             case SYS_OPEN:
-                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return -1;
+                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return BadPointer;
                 return (int64_t)Sys_Open((const char*)frame->arg1);
             case SYS_READ:
-                if (!UserMemory::Range(frame->arg2, frame->arg4, true)) return -1;
+                if (!UserMemory::Range(frame->arg2, frame->arg4, true)) return BadPointer;
                 return (int64_t)Sys_Read((int)frame->arg1, (uint8_t*)frame->arg2,
                                          frame->arg3, frame->arg4);
             case SYS_GETSIZE:
                 return (int64_t)Sys_GetSize((int)frame->arg1);
             case SYS_CLOSE:
                 Sys_Close((int)frame->arg1);
-                return 0;
+                return Success;
             case SYS_READDIR:
-                if ((int64_t)frame->arg3 < 0) return -1;
+                if ((int64_t)frame->arg3 < 0) return InvalidArgument;
-                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return -1;
+                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return BadPointer;
                 if (!UserMemory::Range(frame->arg2,
                                        (uint64_t)((frame->arg3 > 256) ? 256 : frame->arg3) * sizeof(const char*),
                                        true)) {
-                    return -1;
+                    return BadPointer;
                 }
                 return (int64_t)Sys_ReadDir((const char*)frame->arg1,
                                             (const char**)frame->arg2,
                                             (int)frame->arg3);
             case SYS_READDIR_AT:
-                if ((int64_t)frame->arg3 < 0) return -1;
+                if ((int64_t)frame->arg3 < 0) return InvalidArgument;
-                if ((int64_t)frame->arg4 < 0) return -1;
+                if ((int64_t)frame->arg4 < 0) return InvalidArgument;
-                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return -1;
+                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return BadPointer;
                 if (!UserMemory::Range(frame->arg2,
                                        (uint64_t)((frame->arg3 > 256) ? 256 : frame->arg3) * sizeof(const char*),
                                        true)) {
-                    return -1;
+                    return BadPointer;
                 }
                 return (int64_t)Sys_ReadDirAt((const char*)frame->arg1,
                                               (const char**)frame->arg2,
@@ -120,7 +116,7 @@ namespace montauk::abi {
                 return (int64_t)Sys_AllocEager(frame->arg1);
             case SYS_FREE:
                 Sys_Free(frame->arg1);
-                return 0;
+                return Success;
             case SYS_MMAP_ANON:
                 return (int64_t)Sys_MapAnonymous(frame->arg1, frame->arg2);
             case SYS_MUNMAP:
@@ -132,15 +128,15 @@ namespace montauk::abi {
             case SYS_GETMILLISECONDS:
                 return (int64_t)Sys_GetMilliseconds();
             case SYS_GETINFO:
-                if (!UserMemory::Writable<SysInfo>(frame->arg1)) return -1;
+                if (!UserMemory::Writable<SysInfo>(frame->arg1)) return BadPointer;
                 Sys_GetInfo((SysInfo*)frame->arg1);
-                return 0;
+                return Success;
             case SYS_ISKEYAVAILABLE:
                 return (int64_t)Sys_IsKeyAvailable();
             case SYS_GETKEY:
-                if (!UserMemory::Writable<KeyEvent>(frame->arg1)) return -1;
+                if (!UserMemory::Writable<KeyEventV1>(frame->arg1)) return BadPointer;
-                Sys_GetKey((KeyEvent*)frame->arg1);
+                Sys_GetKey((KeyEventV1*)frame->arg1);
-                return 0;
+                return Success;
             case SYS_GETCHAR:
                 return (int64_t)Sys_GetChar();
             case SYS_GETCHAR_NB:
@@ -148,73 +144,75 @@ namespace montauk::abi {
             case SYS_PING:
                 return (int64_t)Sys_Ping((uint32_t)frame->arg1, (uint32_t)frame->arg2);
             case SYS_SPAWN:
-                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return -1;
+                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return BadPointer;
-                if (frame->arg2 != 0 && !UserMemory::String(frame->arg2, kMaxArgsBytes)) return -1;
+                if (frame->arg2 != 0 && !UserMemory::String(frame->arg2, kMaxArgsBytes)) return BadPointer;
                 return (int64_t)Sys_Spawn((const char*)frame->arg1,
                                           UserMemory::IsUserPtr(frame->arg2) ? (const char*)frame->arg2 : nullptr);
             case SYS_SPAWN_CAPS:
-                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return -1;
+                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return BadPointer;
-                if (frame->arg2 != 0 && !UserMemory::String(frame->arg2, kMaxArgsBytes)) return -1;
+                if (frame->arg2 != 0 && !UserMemory::String(frame->arg2, kMaxArgsBytes)) return BadPointer;
-                if (frame->arg3 != 0 && !UserMemory::String(frame->arg3, 32)) return -1;
+                if (frame->arg3 != 0 && !UserMemory::String(frame->arg3, 32)) return BadPointer;
-                if (!UserMemory::Readable<SpawnCapabilities>(frame->arg4)) return -1;
+                if (!UserMemory::Readable<SpawnCapabilities>(frame->arg4)) return BadPointer;
                 return (int64_t)Sys_SpawnCaps(
                     (const char*)frame->arg1,
                     frame->arg2 ? (const char*)frame->arg2 : nullptr,
                     frame->arg3 ? (const char*)frame->arg3 : nullptr,
                     (const SpawnCapabilities*)frame->arg4);
             case SYS_SPAWN_ENV:
-                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return -1;
+                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return BadPointer;
-                if (frame->arg2 != 0 && !UserMemory::String(frame->arg2, kMaxArgsBytes)) return -1;
+                if (frame->arg2 != 0 && !UserMemory::String(frame->arg2, kMaxArgsBytes)) return BadPointer;
-                if (frame->arg3 == 0 || frame->arg4 == 0 ||
+                if (frame->arg4 == 0 || frame->arg4 > Sched::Process::EnvironmentBytes)
-                    frame->arg4 > Sched::Process::EnvironmentBytes ||
+                    return InvalidArgument;
-                    !UserMemory::Range(frame->arg3, frame->arg4, false) ||
+                if (frame->arg3 == 0 || !UserMemory::Range(frame->arg3, frame->arg4, false))
-                    ((const char*)frame->arg3)[frame->arg4 - 1] != '\0') return -1;
+                    return BadPointer;
+                if (((const char*)frame->arg3)[frame->arg4 - 1] != '\0') return InvalidArgument;
                 return (int64_t)Sys_Spawn((const char*)frame->arg1,
                                           UserMemory::IsUserPtr(frame->arg2) ? (const char*)frame->arg2 : nullptr,
                                           (const char*)frame->arg3, (uint32_t)frame->arg4);
             case SYS_WAITPID:
                 return Sys_WaitPid((int)frame->arg1);
             case SYS_GETENVIRON:
-                if (!UserMemory::Range(frame->arg1, frame->arg2, true)) return -1;
+                if (!UserMemory::Range(frame->arg1, frame->arg2, true)) return BadPointer;
                 return Sys_GetEnviron((char*)frame->arg1, frame->arg2);
             case SYS_SETENVIRON:
-                if (frame->arg2 == 0 || frame->arg2 > Sched::Process::EnvironmentBytes ||
+                if (frame->arg2 == 0 || frame->arg2 > Sched::Process::EnvironmentBytes)
-                    !UserMemory::Range(frame->arg1, frame->arg2, false)) return -1;
+                    return InvalidArgument;
+                if (!UserMemory::Range(frame->arg1, frame->arg2, false)) return BadPointer;
                 return Sys_SetEnviron((const char*)frame->arg1, frame->arg2);
             case SYS_FBINFO:
-                if (!UserMemory::Writable<FbInfo>(frame->arg1)) return -1;
+                if (!UserMemory::Writable<FbInfo>(frame->arg1)) return BadPointer;
                 Sys_FbInfo((FbInfo*)frame->arg1);
-                return 0;
+                return Success;
             case SYS_FBMAP:
                 return (int64_t)Sys_FbMap();
             case SYS_FBFLIP:
                 return Sys_FbFlip(frame->arg1, frame->arg2);
             case SYS_DISPLAYINFO:
-                if (!UserMemory::Writable<DisplayInfo>(frame->arg1)) return -1;
+                if (!UserMemory::Writable<DisplayInfo>(frame->arg1)) return BadPointer;
                 return Sys_DisplayInfo((DisplayInfo*)frame->arg1);
             case SYS_DISPLAYMODES:
                 // The Intel driver publishes at most 32 modes. Bounding this
                 // here also keeps the userspace range calculation overflow-free.
-                if ((int64_t)frame->arg2 < 0 || frame->arg2 > 32) return -1;
+                if ((int64_t)frame->arg2 < 0 || frame->arg2 > 32) return InvalidArgument;
                 if (!UserMemory::Range(frame->arg1,
-                        (uint64_t)frame->arg2 * sizeof(DisplayModeInfo), true)) return -1;
+                        (uint64_t)frame->arg2 * sizeof(DisplayModeInfo), true)) return BadPointer;
                 return Sys_DisplayModes((DisplayModeInfo*)frame->arg1, (int)frame->arg2);
             case SYS_DISPLAYSETMODE:
-                if (!Sched::HasCapability(CAP_DISPLAY_ADMIN)) return SYS_ERR_PERMISSION;
+                if (!Sched::Consent::Require(CAP_DISPLAY_ADMIN)) return CapabilityDenied;
                 return Sys_DisplaySetMode((int)frame->arg1);
             case SYS_DISPLAYBRIGHTNESS:
                 if ((int64_t)frame->arg1 >= 0 &&
-                    !Sched::HasCapability(CAP_DISPLAY_ADMIN)) return SYS_ERR_PERMISSION;
+                    !Sched::Consent::Require(CAP_BRIGHTNESS)) return CapabilityDenied;
                 return Sys_DisplayBrightness((int)frame->arg1);
             case SYS_GETEXECPATH:
-                if (!UserMemory::Range(frame->arg2 ? frame->arg1 : frame->arg1, frame->arg2, true)) return -1;
+                if (!UserMemory::Range(frame->arg2 ? frame->arg1 : frame->arg1, frame->arg2, true)) return BadPointer;
                 return Sys_GetExecPath((char*)frame->arg1, frame->arg2);
             case SYS_STAT:
-                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return -1;
+                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return BadPointer;
-                if (!UserMemory::Writable<FileStat>(frame->arg2)) return -1;
+                if (!UserMemory::Writable<FileStat>(frame->arg2)) return BadPointer;
                 return (int64_t)Sys_Stat((const char*)frame->arg1, (FileStat*)frame->arg2);
             case SYS_UTIME:
-                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return -1;
+                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return BadPointer;
                 return (int64_t)Sys_Utime((const char*)frame->arg1,
                                           (int64_t)frame->arg2,
                                           (int64_t)frame->arg3,
@@ -222,33 +220,35 @@ namespace montauk::abi {
             case SYS_TERMSIZE:
                 return (int64_t)Sys_TermSize();
             case SYS_GETARGS:
-                if (!UserMemory::Range(frame->arg1, frame->arg2, true)) return -1;
+                if (!UserMemory::Range(frame->arg1, frame->arg2, true)) return BadPointer;
                 return (int64_t)Sys_GetArgs((char*)frame->arg1, frame->arg2);
             case SYS_RESET:
-                if (!Sched::HasCapability(CAP_POWER_CONTROL)) return SYS_ERR_PERMISSION;
+                if (!Sched::HasCapability(CAP_POWER_CONTROL)) return CapabilityDenied;
                 Sys_Reset();
-                return 0;
+                return Success;
             case SYS_SHUTDOWN:
-                if (!Sched::HasCapability(CAP_POWER_CONTROL)) return SYS_ERR_PERMISSION;
+                if (!Sched::HasCapability(CAP_POWER_CONTROL)) return CapabilityDenied;
                 Sys_Shutdown();
-                return 0;
+                return Success;
             case SYS_POWER_REQUEST:
                 if (frame->arg1 != POWER_REQ_QUERY &&
                     frame->arg1 != POWER_REQ_SHUTDOWN &&
                     frame->arg1 != POWER_REQ_REBOOT &&
-                    frame->arg1 != POWER_REQ_PEEK) return -1;
+                    frame->arg1 != POWER_REQ_PEEK) return InvalidArgument;
                 if (frame->arg1 == POWER_REQ_QUERY) {
-                    if (!Sched::HasCapability(CAP_POWER_CONTROL)) return SYS_ERR_PERMISSION;
+                    if (!Sched::HasCapability(CAP_POWER_CONTROL)) return CapabilityDenied;
                 } else if (!Sched::HasCapability(CAP_POWER_REQUEST)) {
-                    return SYS_ERR_PERMISSION;
+                    return CapabilityDenied;
                 }
                 return Sys_PowerRequest((int)frame->arg1);
             case SYS_GETTIME:
-                if (!UserMemory::Writable<DateTime>(frame->arg1)) return -1;
+                if (!UserMemory::Writable<DateTime>(frame->arg1)) return BadPointer;
                 Sys_GetTime((DateTime*)frame->arg1);
-                return 0;
+                return Success;
+            case SYS_GETUNIXTIME:
+                return Sys_GetUnixTime();
             case SYS_SETUNIXTIME:
-                if (!Sched::HasCapability(CAP_SET_TIME)) return SYS_ERR_PERMISSION;
+                if (!Sched::Consent::Require(CAP_SET_TIME)) return CapabilityDenied;
                 return Sys_SetUnixTime((int64_t)frame->arg1);
             case SYS_SOCKET:
                 return (int64_t)Sys_Socket((int)frame->arg1);
@@ -261,114 +261,114 @@ namespace montauk::abi {
             case SYS_ACCEPT:
                 return (int64_t)Sys_Accept((int)frame->arg1);
             case SYS_SEND:
-                if (!UserMemory::Range(frame->arg2, frame->arg3, false)) return -1;
+                if (!UserMemory::Range(frame->arg2, frame->arg3, false)) return BadPointer;
                 return (int64_t)Sys_Send((int)frame->arg1, (const uint8_t*)frame->arg2, (uint32_t)frame->arg3);
             case SYS_RECV:
-                if (!UserMemory::Range(frame->arg2, frame->arg3, true)) return -1;
+                if (!UserMemory::Range(frame->arg2, frame->arg3, true)) return BadPointer;
                 return (int64_t)Sys_Recv((int)frame->arg1, (uint8_t*)frame->arg2, (uint32_t)frame->arg3);
             case SYS_CLOSESOCK:
                 Sys_CloseSock((int)frame->arg1);
-                return 0;
+                return Success;
             case SYS_GETNETCFG:
-                if (!UserMemory::Writable<NetCfg>(frame->arg1)) return -1;
+                if (!UserMemory::Writable<NetCfg>(frame->arg1)) return BadPointer;
                 Sys_GetNetCfg((NetCfg*)frame->arg1);
-                return 0;
+                return Success;
             case SYS_SETNETCFG:
-                if (!Sched::HasCapability(CAP_NETWORK_ADMIN)) return SYS_ERR_PERMISSION;
+                if (!Sched::Consent::Require(CAP_NETWORK_ADMIN)) return CapabilityDenied;
-                if (!UserMemory::Readable<NetCfg>(frame->arg1)) return -1;
+                if (!UserMemory::Readable<NetCfg>(frame->arg1)) return BadPointer;
                 return (int64_t)Sys_SetNetCfg((const NetCfg*)frame->arg1);
             case SYS_NETSTATUS:
-                if (!UserMemory::Writable<NetStatus>(frame->arg1)) return -1;
+                if (!UserMemory::Writable<NetStatus>(frame->arg1)) return BadPointer;
                 Sys_NetStatus((NetStatus*)frame->arg1);
-                return 0;
+                return Success;
             case SYS_SENDTO:
-                if (!UserMemory::Range(frame->arg2, frame->arg3, false)) return -1;
+                if (!UserMemory::Range(frame->arg2, frame->arg3, false)) return BadPointer;
                 return (int64_t)Sys_SendTo((int)frame->arg1, (const uint8_t*)frame->arg2,
                                            (uint32_t)frame->arg3, (uint32_t)frame->arg4,
                                            (uint16_t)frame->arg5);
             case SYS_RECVFROM:
-                if (!UserMemory::Range(frame->arg2, frame->arg3, true)) return -1;
+                if (!UserMemory::Range(frame->arg2, frame->arg3, true)) return BadPointer;
-                if (!UserMemory::OptionalRange(frame->arg4, sizeof(uint32_t), true)) return -1;
+                if (!UserMemory::OptionalRange(frame->arg4, sizeof(uint32_t), true)) return BadPointer;
-                if (!UserMemory::OptionalRange(frame->arg5, sizeof(uint16_t), true)) return -1;
+                if (!UserMemory::OptionalRange(frame->arg5, sizeof(uint16_t), true)) return BadPointer;
                 return (int64_t)Sys_RecvFrom((int)frame->arg1, (uint8_t*)frame->arg2,
                                              (uint32_t)frame->arg3,
                                              UserMemory::IsUserPtr(frame->arg4) ? (uint32_t*)frame->arg4 : nullptr,
                                              UserMemory::IsUserPtr(frame->arg5) ? (uint16_t*)frame->arg5 : nullptr);
             case SYS_FWRITE:
-                if (!UserMemory::Range(frame->arg2, frame->arg4, false)) return -1;
+                if (!UserMemory::Range(frame->arg2, frame->arg4, false)) return BadPointer;
                 return (int64_t)Sys_FWrite((int)frame->arg1, (const uint8_t*)frame->arg2,
                                            frame->arg3, frame->arg4);
             case SYS_FCREATE:
-                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return -1;
+                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return BadPointer;
                 return (int64_t)Sys_FCreate((const char*)frame->arg1);
             case SYS_FDELETE:
-                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return -1;
+                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return BadPointer;
                 return (int64_t)Sys_FDelete((const char*)frame->arg1);
             case SYS_FMKDIR:
-                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return -1;
+                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return BadPointer;
                 return (int64_t)Sys_FMkdir((const char*)frame->arg1);
             case SYS_FRENAME:
-                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return -1;
+                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return BadPointer;
-                if (!UserMemory::String(frame->arg2, kMaxPathBytes)) return -1;
+                if (!UserMemory::String(frame->arg2, kMaxPathBytes)) return BadPointer;
                 return (int64_t)Sys_FRename((const char*)frame->arg1, (const char*)frame->arg2);
             case SYS_DRIVELIST:
-                if ((int64_t)frame->arg2 < 0) return -1;
+                if ((int64_t)frame->arg2 < 0) return InvalidArgument;
-                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(int), true)) return -1;
+                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(int), true)) return BadPointer;
                 return (int64_t)Sys_DriveList((int*)frame->arg1, (int)frame->arg2);
             case SYS_DRIVELABEL:
-                if ((int64_t)frame->arg3 <= 0) return -1;
+                if ((int64_t)frame->arg3 <= 0) return InvalidArgument;
-                if (!UserMemory::Range(frame->arg2, frame->arg3, true)) return -1;
+                if (!UserMemory::Range(frame->arg2, frame->arg3, true)) return BadPointer;
                 return (int64_t)Sys_DriveLabel((int)frame->arg1, (char*)frame->arg2, (int)frame->arg3);
             case SYS_DRIVEKIND:
                 return (int64_t)Sys_DriveKind((int)frame->arg1);
             case SYS_TERMSCALE:
                 return Sys_TermScale(frame->arg1, frame->arg2);
             case SYS_RESOLVE:
-                if (!UserMemory::String(frame->arg1, kMaxHostnameBytes)) return -1;
+                if (!UserMemory::String(frame->arg1, kMaxHostnameBytes)) return BadPointer;
                 return Sys_Resolve((const char*)frame->arg1);
             case SYS_GETRANDOM:
-                if (!UserMemory::Range(frame->arg1, frame->arg2, true)) return -1;
+                if (!UserMemory::Range(frame->arg1, frame->arg2, true)) return BadPointer;
                 return Sys_GetRandom((uint8_t*)frame->arg1, frame->arg2);
             case SYS_LOG:
-                if (!Sched::HasCapability(CAP_LOG_READ)) return SYS_ERR_PERMISSION;
+                if (!Sched::Consent::Require(CAP_LOG_READ)) return CapabilityDenied;
-                if (!UserMemory::Range(frame->arg1, frame->arg2, true)) return -1;
+                if (!UserMemory::Range(frame->arg1, frame->arg2, true)) return BadPointer;
                 return Kt::ReadKernelLogBuffer((char*)frame->arg1, frame->arg2);
             case SYS_MOUSESTATE:
-                if (!UserMemory::Writable<MouseState>(frame->arg1)) return -1;
+                if (!UserMemory::Writable<MouseState>(frame->arg1)) return BadPointer;
                 Sys_MouseState((MouseState*)frame->arg1);
-                return 0;
+                return Success;
             case SYS_SETMOUSEBOUNDS:
                 Sys_SetMouseBounds((int32_t)frame->arg1, (int32_t)frame->arg2);
-                return 0;
+                return Success;
             case SYS_SPAWN_REDIR:
-                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return -1;
+                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return BadPointer;
-                if (frame->arg2 != 0 && !UserMemory::String(frame->arg2, kMaxArgsBytes)) return -1;
+                if (frame->arg2 != 0 && !UserMemory::String(frame->arg2, kMaxArgsBytes)) return BadPointer;
                 return (int64_t)Sys_SpawnRedir((const char*)frame->arg1,
                                                UserMemory::IsUserPtr(frame->arg2) ? (const char*)frame->arg2 : nullptr);
             case SYS_SPAWN_REDIR_CAPS:
-                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return -1;
+                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return BadPointer;
-                if (frame->arg2 != 0 && !UserMemory::String(frame->arg2, kMaxArgsBytes)) return -1;
+                if (frame->arg2 != 0 && !UserMemory::String(frame->arg2, kMaxArgsBytes)) return BadPointer;
-                if (!UserMemory::Readable<SpawnCapabilities>(frame->arg3)) return -1;
+                if (!UserMemory::Readable<SpawnCapabilities>(frame->arg3)) return BadPointer;
                 return (int64_t)Sys_SpawnRedirCaps(
                     (const char*)frame->arg1,
                     frame->arg2 ? (const char*)frame->arg2 : nullptr,
                     (const SpawnCapabilities*)frame->arg3);
             case SYS_CHILDIO_READ:
-                if ((int64_t)frame->arg3 < 0) return -1;
+                if ((int64_t)frame->arg3 < 0) return InvalidArgument;
-                if (!UserMemory::Range(frame->arg2, (uint64_t)frame->arg3, true)) return -1;
+                if (!UserMemory::Range(frame->arg2, (uint64_t)frame->arg3, true)) return BadPointer;
                 return (int64_t)Sys_ChildIoRead((int)frame->arg1, (char*)frame->arg2, (int)frame->arg3);
             case SYS_CHILDIO_WRITE:
-                if ((int64_t)frame->arg3 < 0) return -1;
+                if ((int64_t)frame->arg3 < 0) return InvalidArgument;
-                if (!UserMemory::Range(frame->arg2, (uint64_t)frame->arg3, false)) return -1;
+                if (!UserMemory::Range(frame->arg2, (uint64_t)frame->arg3, false)) return BadPointer;
                 return (int64_t)Sys_ChildIoWrite((int)frame->arg1, (const char*)frame->arg2, (int)frame->arg3);
             case SYS_CHILDIO_WRITEKEY:
-                if (!UserMemory::Readable<KeyEvent>(frame->arg2)) return -1;
+                if (!UserMemory::Readable<KeyEventV1>(frame->arg2)) return BadPointer;
-                return (int64_t)Sys_ChildIoWriteKey((int)frame->arg1, (const KeyEvent*)frame->arg2);
+                return (int64_t)Sys_ChildIoWriteKey((int)frame->arg1, (const KeyEventV1*)frame->arg2);
             case SYS_CHILDIO_SETTERMSZ:
                 return (int64_t)Sys_ChildIoSetTermsz((int)frame->arg1, (int)frame->arg2, (int)frame->arg3);
             case SYS_WINCREATE:
-                if (!UserMemory::String(frame->arg1, kMaxWindowTitleBytes)) return -1;
+                if (!UserMemory::String(frame->arg1, kMaxWindowTitleBytes)) return BadPointer;
-                if (!UserMemory::Writable<WinCreateResult>(frame->arg4)) return -1;
+                if (!UserMemory::Writable<WinCreateResult>(frame->arg4)) return BadPointer;
                 return (int64_t)Sys_WinCreate((const char*)frame->arg1, (int)frame->arg2,
                                               (int)frame->arg3, (WinCreateResult*)frame->arg4);
             case SYS_WINDESTROY:
@@ -376,38 +376,44 @@ namespace montauk::abi {
             case SYS_WINPRESENT:
                 return (int64_t)Sys_WinPresent((int)frame->arg1);
             case SYS_WINPOLL:
-                if (!UserMemory::Writable<WinEvent>(frame->arg2)) return -1;
+                if (!UserMemory::Writable<WinEvent>(frame->arg2)) return BadPointer;
                 return (int64_t)Sys_WinPoll((int)frame->arg1, (WinEvent*)frame->arg2);
             case SYS_WINENUM:
-                if ((int64_t)frame->arg2 < 0) return -1;
+                if ((int64_t)frame->arg2 < 0) return InvalidArgument;
-                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(WinInfo), true)) return -1;
+                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(WinInfo), true)) return BadPointer;
                 return (int64_t)Sys_WinEnum((WinInfo*)frame->arg1, (int)frame->arg2);
             case SYS_WINMAP:
                 return (int64_t)Sys_WinMap((int)frame->arg1);
             case SYS_WINUNMAP:
                 return (int64_t)Sys_WinUnmap((int)frame->arg1);
             case SYS_WINSENDEVENT:
-                if (!UserMemory::Readable<WinEvent>(frame->arg2)) return -1;
+                if (!UserMemory::Readable<WinEvent>(frame->arg2)) return BadPointer;
                 return (int64_t)Sys_WinSendEvent((int)frame->arg1, (const WinEvent*)frame->arg2);
             case SYS_WINRESIZE:
                 return (int64_t)Sys_WinResize((int)frame->arg1, (int)frame->arg2, (int)frame->arg3);
             case SYS_PROCLIST:
-                if ((int64_t)frame->arg2 < 0) return -1;
+                if ((int64_t)frame->arg2 < 0) return InvalidArgument;
-                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(ProcInfo), true)) return -1;
+                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(ProcInfo), true)) return BadPointer;
                 return (int64_t)Sys_ProcList((ProcInfo*)frame->arg1, (int)frame->arg2);
             case SYS_KILL:
                 return (int64_t)Sys_Kill((int)frame->arg1);
+            case SYS_ABORT:
+                Sys_Abort();
+                return Success;
+            case SYS_EXIT_GROUP:
+                Sys_ExitGroup((int)frame->arg1);
+                return Success;
             case SYS_SETSESSION:
                 return (int64_t)Sys_SetSession();
             case SYS_KILLSESSION:
-                if (!Sched::HasCapability(CAP_PROCESS_ADMIN)) return SYS_ERR_PERMISSION;
+                if (!Sched::HasCapability(CAP_PROCESS_ADMIN)) return CapabilityDenied;
                 return (int64_t)Sys_KillSession((int)frame->arg1);
             case SYS_DEVLIST:
-                if ((int64_t)frame->arg2 < 0) return -1;
+                if ((int64_t)frame->arg2 < 0) return InvalidArgument;
-                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(DevInfo), true)) return -1;
+                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(DevInfo), true)) return BadPointer;
                 return (int64_t)Sys_DevList((DevInfo*)frame->arg1, (int)frame->arg2);
             case SYS_DISKINFO:
-                if (!UserMemory::Writable<DiskInfo>(frame->arg1)) return -1;
+                if (!UserMemory::Writable<DiskInfo>(frame->arg1)) return BadPointer;
                 return (int64_t)Sys_DiskInfo((DiskInfo*)frame->arg1, (int)frame->arg2);
             case SYS_WINSETSCALE:
                 return (int64_t)Sys_WinSetScale((int)frame->arg1);
@@ -418,50 +424,58 @@ namespace montauk::abi {
             case SYS_WINSETFLAGS:
                 return (int64_t)Sys_WinSetFlags((int)frame->arg1, (uint32_t)frame->arg2);
             case SYS_MEMSTATS:
-                if (!UserMemory::Writable<MemStats>(frame->arg1)) return -1;
+                if (!UserMemory::Writable<MemStats>(frame->arg1)) return BadPointer;
                 Sys_MemStats((MemStats*)frame->arg1);
-                return 0;
+                return Success;
             case SYS_PARTLIST:
-                if ((int64_t)frame->arg2 < 0) return -1;
+                if ((int64_t)frame->arg2 < 0) return InvalidArgument;
-                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(PartInfo), true)) return -1;
+                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(PartInfo), true)) return BadPointer;
                 return (int64_t)Sys_PartList((PartInfo*)frame->arg1, (int)frame->arg2);
+            case SYS_VOLINFO:
+                if ((int64_t)frame->arg2 < 0) return InvalidArgument;
+                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(VolInfo), true)) return BadPointer;
+                return (int64_t)Sys_VolInfo((VolInfo*)frame->arg1, (int)frame->arg2);
             case SYS_DISKREAD:
-                if (!Sched::HasCapability(CAP_RAW_STORAGE)) return SYS_ERR_PERMISSION;
+                if (!Sched::HasCapability(CAP_RAW_STORAGE)) return CapabilityDenied;
                 return (int64_t)Sys_DiskRead((int)frame->arg1, frame->arg2,
                                              (uint32_t)frame->arg3, (void*)frame->arg4);
             case SYS_DISKWRITE:
-                if (!Sched::HasCapability(CAP_RAW_STORAGE)) return SYS_ERR_PERMISSION;
+                if (!Sched::HasCapability(CAP_RAW_STORAGE)) return CapabilityDenied;
                 return (int64_t)Sys_DiskWrite((int)frame->arg1, frame->arg2,
                                               (uint32_t)frame->arg3, (const void*)frame->arg4);
             case SYS_GPTINIT:
-                if (!Sched::HasCapability(CAP_STORAGE_ADMIN)) return SYS_ERR_PERMISSION;
+                if (!Sched::Consent::Require(CAP_STORAGE_ADMIN)) return CapabilityDenied;
                 return (int64_t)Sys_GptInit((int)frame->arg1);
             case SYS_GPTADD:
-                if (!Sched::HasCapability(CAP_STORAGE_ADMIN)) return SYS_ERR_PERMISSION;
+                if (!Sched::Consent::Require(CAP_STORAGE_ADMIN)) return CapabilityDenied;
-                if (!UserMemory::Readable<GptAddParams>(frame->arg1)) return -1;
+                if (!UserMemory::Readable<GptAddParams>(frame->arg1)) return BadPointer;
                 return (int64_t)Sys_GptAdd((const GptAddParams*)frame->arg1);
             case SYS_FSMOUNT:
-                if (!Sched::HasCapability(CAP_STORAGE_ADMIN)) return SYS_ERR_PERMISSION;
+                if (!Sched::Consent::Require(CAP_STORAGE_ADMIN)) return CapabilityDenied;
                 return (int64_t)Sys_FsMount((int)frame->arg1, (int)frame->arg2);
             case SYS_FS_SYNC:
-                if (!Sched::HasCapability(CAP_STORAGE_ADMIN)) return SYS_ERR_PERMISSION;
+                if (!Sched::Consent::Require(CAP_STORAGE_ADMIN)) return CapabilityDenied;
                 return Sys_FsSync();
+            case SYS_FS_GROW:
+                if (!Sched::Consent::Require(CAP_STORAGE_ADMIN)) return CapabilityDenied;
+                if (!UserMemory::Range(frame->arg2, sizeof(FsGrowStatus), true)) return BadPointer;
+                return Sys_FsGrow((int)frame->arg1, (FsGrowStatus*)frame->arg2, frame->arg3 != 0);
             case SYS_FSFORMAT:
-                if (!Sched::HasCapability(CAP_STORAGE_ADMIN)) return SYS_ERR_PERMISSION;
+                if (!Sched::Consent::Require(CAP_STORAGE_ADMIN)) return CapabilityDenied;
-                if (!UserMemory::Readable<FsFormatParams>(frame->arg1)) return -1;
+                if (!UserMemory::Readable<FsFormatParams>(frame->arg1)) return BadPointer;
                 return (int64_t)Sys_FsFormat((const FsFormatParams*)frame->arg1);
             case SYS_AUDIOOPEN:
                 return Sys_AudioOpen((uint32_t)frame->arg1, (uint8_t)frame->arg2, (uint8_t)frame->arg3);
             case SYS_AUDIOCLOSE:
                 return Sys_AudioClose((int)frame->arg1);
             case SYS_AUDIOWRITE:
-                if (!UserMemory::Range(frame->arg2, frame->arg3, false)) return -1;
+                if (!UserMemory::Range(frame->arg2, frame->arg3, false)) return BadPointer;
                 return Sys_AudioWrite((int)frame->arg1, (const uint8_t*)frame->arg2, (uint32_t)frame->arg3);
             case SYS_AUDIOCTL:
                 return Sys_AudioCtl((int)frame->arg1, (int)frame->arg2, (int)frame->arg3);
             case SYS_AUDIOLIST:
-                if ((int64_t)frame->arg2 < 0) return -1;
+                if ((int64_t)frame->arg2 < 0) return InvalidArgument;
-                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(AudioStreamInfo), true)) return -1;
+                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(AudioStreamInfo), true)) return BadPointer;
                 return Sys_AudioList((AudioStreamInfo*)frame->arg1, (int)frame->arg2);
             case SYS_AUDIOWAIT:
                 return Sys_AudioWait(frame->arg1, frame->arg2);
@@ -502,52 +516,83 @@ namespace montauk::abi {
                 return Sys_UsbBulkInRead((int)frame->arg1, (uint8_t*)frame->arg2,
                                          (uint32_t)frame->arg3);
             case SYS_POWERINFO:
-                if (!UserMemory::Writable<PowerInfo>(frame->arg1)) return -1;
+                if (!UserMemory::Writable<PowerInfo>(frame->arg1)) return BadPointer;
                 return Sys_PowerInfo((PowerInfo*)frame->arg1);
+            case SYS_BATTERYINFO:
+                // Read-only power telemetry, ungated like SYS_POWERINFO: the
+                // panel indicator has to work in an unprivileged session, which
+                // holds no device capability to spend on it.
+                if (!UserMemory::Writable<BatteryInfo>(frame->arg1)) return BadPointer;
+                return Sys_BatteryInfo((BatteryInfo*)frame->arg1);
+            case SYS_GETCAPS:
+                if (!UserMemory::Writable<SpawnCapabilities>(frame->arg1)) return BadPointer;
+                return Sys_GetCaps((SpawnCapabilities*)frame->arg1);
+            case SYS_CONSENT_REGISTER:
+                return Sched::Consent::Register();
+            case SYS_CONSENT_NEXT:
+                if (!UserMemory::Writable<ConsentRequest>(frame->arg1)) return BadPointer;
+                return Sched::Consent::Next((ConsentRequest*)frame->arg1);
+            case SYS_CONSENT_REPLY:
+                return Sched::Consent::Reply(frame->arg1, (int)frame->arg2);
+            case SYS_GETKEY_EX:
+                if (frame->arg2 < sizeof(KeyEventV1) || frame->arg2 > 4096) return InvalidArgument;
+                if (!UserMemory::Range(frame->arg1, frame->arg2, true)) return BadPointer;
+                return Sys_GetKeyEx((void*)frame->arg1, frame->arg2);
+            case SYS_CHILDIO_WRITEKEY_EX:
+                if (frame->arg3 < sizeof(KeyEventV1) || frame->arg3 > 4096) return InvalidArgument;
+                if (!UserMemory::Range(frame->arg2, frame->arg3, false)) return BadPointer;
+                return (int64_t)Sys_ChildIoWriteKeyEx((int)frame->arg1, (const void*)frame->arg2, frame->arg3);
             case SYS_THREAD_SPAWN:
                 return Sys_ThreadSpawn(frame->arg1, frame->arg2, frame->arg3);
             case SYS_THREAD_EXIT:
                 Sys_ThreadExit((int)frame->arg1);
-                return 0;
+                return Success;
             case SYS_THREAD_JOIN:
-                if (frame->arg2 != 0 && !UserMemory::Writable<int>(frame->arg2)) return -1;
+                if (frame->arg2 != 0 && !UserMemory::Writable<int>(frame->arg2)) return BadPointer;
                 return Sys_ThreadJoin((int)frame->arg1, (int*)frame->arg2);
             case SYS_THREAD_SELF:
                 return Sys_ThreadSelf();
+            case SYS_FUTEX_WAIT:
+                // Validation demand-commits the page so the scheduler can
+                // read the word with its lock held without faulting.
+                if (!UserMemory::Readable<uint32_t>(frame->arg1)) return BadPointer;
+                return Sys_FutexWait(frame->arg1, (uint32_t)frame->arg2, frame->arg3);
+            case SYS_FUTEX_WAKE:
+                return Sys_FutexWake(frame->arg1, (uint32_t)frame->arg2);
             case SYS_BTSCAN:
-                if (!Sched::HasCapability(CAP_DEVICE_ADMIN)) return SYS_ERR_PERMISSION;
+                if (!Sched::HasCapability(CAP_DEVICE_ADMIN)) return CapabilityDenied;
-                if ((int64_t)frame->arg2 < 0) return -1;
+                if ((int64_t)frame->arg2 < 0) return InvalidArgument;
-                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(BtScanResult), true)) return -1;
+                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(BtScanResult), true)) return BadPointer;
                 return Sys_BtScan((BtScanResult*)frame->arg1, (int)frame->arg2, (uint32_t)frame->arg3);
             case SYS_BTCONNECT:
-                if (!Sched::HasCapability(CAP_DEVICE_ADMIN)) return SYS_ERR_PERMISSION;
+                if (!Sched::HasCapability(CAP_DEVICE_ADMIN)) return CapabilityDenied;
-                if (!UserMemory::Range(frame->arg1, 6, false)) return -1;
+                if (!UserMemory::Range(frame->arg1, 6, false)) return BadPointer;
                 return Sys_BtConnect((const uint8_t*)frame->arg1);
             case SYS_BTDISCONNECT:
-                if (!Sched::HasCapability(CAP_DEVICE_ADMIN)) return SYS_ERR_PERMISSION;
+                if (!Sched::HasCapability(CAP_DEVICE_ADMIN)) return CapabilityDenied;
-                if (!UserMemory::Range(frame->arg1, 6, false)) return -1;
+                if (!UserMemory::Range(frame->arg1, 6, false)) return BadPointer;
                 return Sys_BtDisconnect((const uint8_t*)frame->arg1);
             case SYS_BTSETADDR:
-                if (!Sched::HasCapability(CAP_DEVICE_ADMIN)) return SYS_ERR_PERMISSION;
+                if (!Sched::HasCapability(CAP_DEVICE_ADMIN)) return CapabilityDenied;
-                if (!UserMemory::Range(frame->arg1, 6, false)) return -1;
+                if (!UserMemory::Range(frame->arg1, 6, false)) return BadPointer;
                 return Sys_BtSetAddr((const uint8_t*)frame->arg1);
             case SYS_BTBONDS:
-                if ((int64_t)frame->arg2 < 0) return -1;
+                if ((int64_t)frame->arg2 < 0) return InvalidArgument;
-                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(BtBondInfo), true)) return -1;
+                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(BtBondInfo), true)) return BadPointer;
                 return Sys_BtBonds((BtBondInfo*)frame->arg1, (int)frame->arg2);
             case SYS_BTFORGET:
-                if (!Sched::HasCapability(CAP_DEVICE_ADMIN)) return SYS_ERR_PERMISSION;
+                if (!Sched::HasCapability(CAP_DEVICE_ADMIN)) return CapabilityDenied;
-                if (!UserMemory::Range(frame->arg1, 6, false)) return -1;
+                if (!UserMemory::Range(frame->arg1, 6, false)) return BadPointer;
                 return Sys_BtForget((const uint8_t*)frame->arg1);
             case SYS_BTLIST:
-                if ((int64_t)frame->arg2 < 0) return -1;
+                if ((int64_t)frame->arg2 < 0) return InvalidArgument;
-                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(BtDevInfo), true)) return -1;
+                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(BtDevInfo), true)) return BadPointer;
                 return Sys_BtList((BtDevInfo*)frame->arg1, (int)frame->arg2);
             case SYS_BTINFO:
-                if (!UserMemory::Writable<BtAdapterInfo>(frame->arg1)) return -1;
+                if (!UserMemory::Writable<BtAdapterInfo>(frame->arg1)) return BadPointer;
                 return Sys_BtInfo((BtAdapterInfo*)frame->arg1);
             case SYS_WIFI_SCAN:
-                if (!Sched::HasCapability(CAP_NETWORK_ADMIN)) return SYS_ERR_PERMISSION;
+                if (!Sched::Consent::Require(CAP_NETWORK_ADMIN)) return SYS_ERR_PERMISSION;
                 if ((int64_t)frame->arg2 < 0) return -1;
                 if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(WifiNetwork), true)) return -1;
                 return Sys_WifiScan((WifiNetwork*)frame->arg1, (int)frame->arg2, (uint32_t)frame->arg3);
@@ -555,81 +600,81 @@ namespace montauk::abi {
                 if (!UserMemory::Writable<WifiInfo>(frame->arg1)) return -1;
                 return Sys_WifiInfo((WifiInfo*)frame->arg1);
             case SYS_WIFI_CONNECT:
-                if (!Sched::HasCapability(CAP_NETWORK_ADMIN)) return SYS_ERR_PERMISSION;
+                if (!Sched::Consent::Require(CAP_NETWORK_ADMIN)) return SYS_ERR_PERMISSION;
                 if (!UserMemory::String(frame->arg1, 64)) return -1;
                 if (frame->arg2 != 0 && !UserMemory::String(frame->arg2, 128)) return -1;
                 return Sys_WifiConnect((const char*)frame->arg1, (const char*)frame->arg2);
             case SYS_WIFI_DISCONNECT:
-                if (!Sched::HasCapability(CAP_NETWORK_ADMIN)) return SYS_ERR_PERMISSION;
+                if (!Sched::Consent::Require(CAP_NETWORK_ADMIN)) return SYS_ERR_PERMISSION;
                 return Sys_WifiDisconnect();
             case SYS_WIFI_SCAN_START:
-                if (!Sched::HasCapability(CAP_NETWORK_ADMIN)) return SYS_ERR_PERMISSION;
+                if (!Sched::Consent::Require(CAP_NETWORK_ADMIN)) return SYS_ERR_PERMISSION;
                 return Sys_WifiScanStart((uint32_t)frame->arg1);
             case SYS_WIFI_RESULTS:
                 if ((int64_t)frame->arg2 < 0) return -1;
                 if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(WifiNetwork), true)) return -1;
                 return Sys_WifiResults((WifiNetwork*)frame->arg1, (int)frame->arg2);
             case SYS_WIFI_CONNECT_ASYNC:
-                if (!Sched::HasCapability(CAP_NETWORK_ADMIN)) return SYS_ERR_PERMISSION;
+                if (!Sched::Consent::Require(CAP_NETWORK_ADMIN)) return SYS_ERR_PERMISSION;
                 if (!UserMemory::String(frame->arg1, 64)) return -1;
                 if (frame->arg2 != 0 && !UserMemory::String(frame->arg2, 128)) return -1;
                 return Sys_WifiConnectAsync((const char*)frame->arg1, (const char*)frame->arg2);
             case SYS_NETIFS:
-                if ((int64_t)frame->arg2 < 0) return -1;
+                if ((int64_t)frame->arg2 < 0) return InvalidArgument;
-                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(NetIfInfo), true)) return -1;
+                if (!UserMemory::Range(frame->arg1, (uint64_t)frame->arg2 * sizeof(NetIfInfo), true)) return BadPointer;
                 return Sys_NetIfs((NetIfInfo*)frame->arg1, (int)frame->arg2);
             case SYS_SUSPEND:
-                if (!Sched::HasCapability(CAP_SUSPEND)) return SYS_ERR_PERMISSION;
+                if (!Sched::HasCapability(CAP_SUSPEND)) return CapabilityDenied;
                 return Sys_Suspend();
             case SYS_SETTZ:
-                if (!Sched::HasCapability(CAP_SET_TIME)) return SYS_ERR_PERMISSION;
+                if (!Sched::Consent::Require(CAP_SET_TIME)) return CapabilityDenied;
                 return Sys_SetTZ((int32_t)frame->arg1);
             case SYS_GETTZ:
                 return Sys_GetTZ();
             case SYS_SETUSER:
-                if (!Sched::HasCapability(CAP_USER_ADMIN)) return SYS_ERR_PERMISSION;
+                if (!Sched::HasCapability(CAP_USER_ADMIN)) return CapabilityDenied;
-                if (!UserMemory::String(frame->arg2, kMaxUserNameBytes)) return -1;
+                if (!UserMemory::String(frame->arg2, kMaxUserNameBytes)) return BadPointer;
                 return Sys_SetUser((int)frame->arg1, (const char*)frame->arg2);
             case SYS_GETUSER:
-                if (!UserMemory::Range(frame->arg1, frame->arg2, true)) return -1;
+                if (!UserMemory::Range(frame->arg1, frame->arg2, true)) return BadPointer;
                 return Sys_GetUser((char*)frame->arg1, frame->arg2);
             case SYS_GETCWD:
-                if (!UserMemory::Range(frame->arg1, frame->arg2, true)) return -1;
+                if (!UserMemory::Range(frame->arg1, frame->arg2, true)) return BadPointer;
                 return Sys_GetCwd((char*)frame->arg1, frame->arg2);
             case SYS_CHDIR:
-                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return -1;
+                if (!UserMemory::String(frame->arg1, kMaxPathBytes)) return BadPointer;
                 return Sys_Chdir((const char*)frame->arg1);
             case SYS_DUPHANDLE:
                 return Sys_DupHandle((int)frame->arg1);
             case SYS_WAIT_HANDLE:
                 return (int64_t)Sys_WaitHandle((int)frame->arg1, (uint32_t)frame->arg2, frame->arg3);
             case SYS_STREAM_CREATE:
-                if (!UserMemory::Writable<int>(frame->arg1) || !UserMemory::Writable<int>(frame->arg2)) return -1;
+                if (!UserMemory::Writable<int>(frame->arg1) || !UserMemory::Writable<int>(frame->arg2)) return BadPointer;
                 return Sys_StreamCreate((int*)frame->arg1, (int*)frame->arg2, (uint32_t)frame->arg3);
             case SYS_STREAM_READ:
-                if ((int64_t)frame->arg3 < 0) return -1;
+                if ((int64_t)frame->arg3 < 0) return InvalidArgument;
-                if (!UserMemory::Range(frame->arg2, (uint64_t)frame->arg3, true)) return -1;
+                if (!UserMemory::Range(frame->arg2, (uint64_t)frame->arg3, true)) return BadPointer;
                 return Sys_StreamRead((int)frame->arg1, (uint8_t*)frame->arg2, (int)frame->arg3);
             case SYS_STREAM_WRITE:
-                if ((int64_t)frame->arg3 < 0) return -1;
+                if ((int64_t)frame->arg3 < 0) return InvalidArgument;
-                if (!UserMemory::Range(frame->arg2, (uint64_t)frame->arg3, false)) return -1;
+                if (!UserMemory::Range(frame->arg2, (uint64_t)frame->arg3, false)) return BadPointer;
                 return Sys_StreamWrite((int)frame->arg1, (const uint8_t*)frame->arg2, (int)frame->arg3);
             case SYS_MAILBOX_CREATE:
-                if (!UserMemory::Writable<int>(frame->arg1) || !UserMemory::Writable<int>(frame->arg2)) return -1;
+                if (!UserMemory::Writable<int>(frame->arg1) || !UserMemory::Writable<int>(frame->arg2)) return BadPointer;
                 return Sys_MailboxCreate((int*)frame->arg1, (int*)frame->arg2);
             case SYS_MAILBOX_SEND:
-                if (frame->arg4 > 0 && !UserMemory::Range(frame->arg3, frame->arg4, false)) return -1;
+                if (frame->arg4 > 0 && !UserMemory::Range(frame->arg3, frame->arg4, false)) return BadPointer;
                 return Sys_MailboxSend((int)frame->arg1, (uint32_t)frame->arg2,
                                        (const void*)frame->arg3, (uint16_t)frame->arg4,
                                        (int)frame->arg5);
             case SYS_MAILBOX_RECV:
-                if (!UserMemory::Writable<uint16_t>(frame->arg4)) return -1;
+                if (!UserMemory::Writable<uint16_t>(frame->arg4)) return BadPointer;
-                if (frame->arg2 != 0 && !UserMemory::Writable<uint32_t>(frame->arg2)) return -1;
+                if (frame->arg2 != 0 && !UserMemory::Writable<uint32_t>(frame->arg2)) return BadPointer;
                 if (frame->arg3 != 0) {
                     uint16_t maxCopy = *(uint16_t*)frame->arg4;
-                    if (!UserMemory::Range(frame->arg3, maxCopy, true)) return -1;
+                    if (!UserMemory::Range(frame->arg3, maxCopy, true)) return BadPointer;
                 }
-                if (frame->arg5 != 0 && !UserMemory::Writable<int>(frame->arg5)) return -1;
+                if (frame->arg5 != 0 && !UserMemory::Writable<int>(frame->arg5)) return BadPointer;
                 return Sys_MailboxRecv((int)frame->arg1,
                                        UserMemory::IsUserPtr(frame->arg2) ? (uint32_t*)frame->arg2 : nullptr,
                                        UserMemory::IsUserPtr(frame->arg3) ? (void*)frame->arg3 : nullptr,
@@ -642,7 +687,7 @@ namespace montauk::abi {
             case SYS_WAITSET_REMOVE:
                 return Sys_WaitsetRemove((int)frame->arg1, (int)frame->arg2);
             case SYS_WAITSET_WAIT:
-                if (frame->arg2 != 0 && !UserMemory::Writable<IpcWaitResult>(frame->arg2)) return -1;
+                if (frame->arg2 != 0 && !UserMemory::Writable<IpcWaitResult>(frame->arg2)) return BadPointer;
                 return Sys_WaitsetWait((int)frame->arg1,
                                        UserMemory::IsUserPtr(frame->arg2) ? (IpcWaitResult*)frame->arg2 : nullptr,
                                        frame->arg3);
@@ -664,28 +709,73 @@ namespace montauk::abi {
             case SYS_GETLIBBASE:
                 return (int64_t)Sys_GetLibBase(frame->arg1);
             case SYS_CRASH_REPORT:
-                if (!UserMemory::Writable<CrashReportInfo>(frame->arg1)) return -1;
+                if (!UserMemory::Writable<CrashReportInfo>(frame->arg1)) return BadPointer;
                 return Sys_CrashReport((CrashReportInfo*)frame->arg1);
             case SYS_CLIPBOARD_SET_TEXT:
-                if (frame->arg2 > 0 && !UserMemory::Range(frame->arg1, frame->arg2, false)) return -1;
+                if (frame->arg2 > 0 && !UserMemory::Range(frame->arg1, frame->arg2, false)) return BadPointer;
                 return Sys_ClipboardSetText((const char*)frame->arg1, (uint32_t)frame->arg2);
             case SYS_CLIPBOARD_GET_INFO:
-                if (!UserMemory::Writable<ClipboardInfo>(frame->arg1)) return -1;
+                if (!UserMemory::Writable<ClipboardInfo>(frame->arg1)) return BadPointer;
                 return Sys_ClipboardGetInfo((ClipboardInfo*)frame->arg1);
             case SYS_CLIPBOARD_GET_TEXT:
-                if (frame->arg2 > 0 && !UserMemory::Range(frame->arg1, frame->arg2, true)) return -1;
+                if (frame->arg2 > 0 && !UserMemory::Range(frame->arg1, frame->arg2, true)) return BadPointer;
-                if (!UserMemory::Writable<uint32_t>(frame->arg3)) return -1;
+                if (!UserMemory::Writable<uint32_t>(frame->arg3)) return BadPointer;
-                if (frame->arg4 != 0 && !UserMemory::Writable<uint64_t>(frame->arg4)) return -1;
+                if (frame->arg4 != 0 && !UserMemory::Writable<uint64_t>(frame->arg4)) return BadPointer;
                 return Sys_ClipboardGetText(UserMemory::IsUserPtr(frame->arg1) ? (char*)frame->arg1 : nullptr,
                                            (uint32_t)frame->arg2,
                                            (uint32_t*)frame->arg3,
                                            UserMemory::IsUserPtr(frame->arg4) ? (uint64_t*)frame->arg4 : nullptr);
             case SYS_CLIPBOARD_CLEAR:
                 return Sys_ClipboardClear();
+            case SYS_CLIPBOARD_SET_DATA:
+                if (frame->arg2 > TRANSFER_MAX_FORMATS) return InvalidArgument;
+                if (!UserMemory::Range(frame->arg1, frame->arg2 * sizeof(TransferItem), false)) return BadPointer;
+                return Sys_ClipboardSetData((const TransferItem*)frame->arg1, (uint32_t)frame->arg2);
+            case SYS_CLIPBOARD_FORMATS: {
+                uint64_t max = frame->arg2 > TRANSFER_MAX_FORMATS ? TRANSFER_MAX_FORMATS : frame->arg2;
+                if (!UserMemory::OptionalRange(frame->arg1, max * sizeof(TransferFormatInfo), true)) return BadPointer;
+                if (!UserMemory::OptionalRange(frame->arg3, sizeof(uint64_t), true)) return BadPointer;
+                return Sys_ClipboardFormats(frame->arg1 ? (TransferFormatInfo*)frame->arg1 : nullptr,
+                                            (uint32_t)max,
+                                            frame->arg3 ? (uint64_t*)frame->arg3 : nullptr);
+            }
+            case SYS_CLIPBOARD_GET_DATA:
+                if (!UserMemory::String(frame->arg1, TRANSFER_MIME_MAX)) return BadPointer;
+                if (frame->arg3 > 0 && !UserMemory::Range(frame->arg2, frame->arg3, true)) return BadPointer;
+                if (!UserMemory::Writable<uint32_t>(frame->arg4)) return BadPointer;
+                return Sys_ClipboardGetData((const char*)frame->arg1,
+                                            frame->arg3 > 0 ? (void*)frame->arg2 : nullptr,
+                                            (uint32_t)frame->arg3, (uint32_t*)frame->arg4);
+            case SYS_WIN_REQUEST:
+                if (!UserMemory::Readable<WinRequest>(frame->arg2)) return BadPointer;
+                return Sys_WinRequest((int)frame->arg1, (const WinRequest*)frame->arg2);
+            case SYS_WIN_REQUEST_POLL:
+                if (!UserMemory::Writable<WinRequest>(frame->arg1)) return BadPointer;
+                return Sys_WinRequestPoll((WinRequest*)frame->arg1);
+            case SYS_WIN_DRAG_BEGIN:
+                if (!UserMemory::Readable<WinDragBegin>(frame->arg2)) return BadPointer;
+                if (frame->arg4 == 0 || frame->arg4 > TRANSFER_MAX_FORMATS) return InvalidArgument;
+                if (!UserMemory::Range(frame->arg3, frame->arg4 * sizeof(TransferItem), false)) return BadPointer;
+                return Sys_WinDragBegin((int)frame->arg1, (const WinDragBegin*)frame->arg2,
+                                        (const TransferItem*)frame->arg3, (uint32_t)frame->arg4);
+            case SYS_WIN_DRAG_FORMATS: {
+                uint64_t max = frame->arg3 > TRANSFER_MAX_FORMATS ? TRANSFER_MAX_FORMATS : frame->arg3;
+                if (!UserMemory::OptionalRange(frame->arg2, max * sizeof(TransferFormatInfo), true)) return BadPointer;
+                return Sys_WinDragFormats((uint32_t)frame->arg1,
+                                          frame->arg2 ? (TransferFormatInfo*)frame->arg2 : nullptr,
+                                          (uint32_t)max);
+            }
+            case SYS_WIN_DRAG_GET_DATA:
+                if (!UserMemory::String(frame->arg2, TRANSFER_MIME_MAX)) return BadPointer;
+                if (frame->arg4 > 0 && !UserMemory::Range(frame->arg3, frame->arg4, true)) return BadPointer;
+                if (!UserMemory::Writable<uint32_t>(frame->arg5)) return BadPointer;
+                return Sys_WinDragGetData((uint32_t)frame->arg1, (const char*)frame->arg2,
+                                          frame->arg4 > 0 ? (void*)frame->arg3 : nullptr,
+                                          (uint32_t)frame->arg4, (uint32_t*)frame->arg5);
             case SYS_INPUT_WAIT:
                 return (int64_t)Sys_InputWait(frame->arg1, frame->arg2);
             case SYS_LOG_WRITE: {
-                if (!UserMemory::String(frame->arg1, kMaxUserspaceLogEntryBytes)) return -1;
+                if (!UserMemory::String(frame->arg1, kMaxUserspaceLogEntryBytes)) return BadPointer;
 
                 auto* process = Sched::GetCurrentProcessPtr();
                 const char* username = process != nullptr && process->user[0] != '\0'
@@ -698,38 +788,22 @@ namespace montauk::abi {
                 Kt::UserspaceLogStream(imageName, username)
                     << (const char*)frame->arg1;
 
-                return 0;
+                return Success;
             }
             case SYS_TERMINAL_ATTACHED:
                 return Sys_TerminalAttached();
             default:
-                return -1;
+                return BadInvocation;
         }
     }
 
-    // ---- SYSCALL MSR initialization ----
+    // ---- Entry point initialization ----
 
     void InitializeSyscalls() {
-        // Enable SYSCALL/SYSRET and no-execute page permissions in EFER.
+        Arch::InitializeSyscallEntry();
-        uint64_t efer = Hal::ReadMSR(Hal::IA32_EFER);
-        efer |= 1 | (1ULL << 11);  // SCE | NXE
-        Hal::WriteMSR(Hal::IA32_EFER, efer);
 
-        // STAR: kernel CS in [47:32], sysret base in [63:48]
+        Kt::KernelLogStream(Kt::OK, "Syscall") << "System call entry installed ("
-        // SYSCALL: CS=0x08, SS=0x10
+            << kcp::dec << (SYS_SPAWN_REDIR_CAPS + 1) << " syscall slots)";
-        // SYSRET:  CS=0x10+16=0x20|RPL3=0x23, SS=0x10+8=0x18|RPL3=0x1B
-        uint64_t star = (0x0010ULL << 48) | (0x0008ULL << 32);
-        Hal::WriteMSR(Hal::IA32_STAR, star);
-
-        // LSTAR: SYSCALL entry point
-        Hal::WriteMSR(Hal::IA32_LSTAR, (uint64_t)SyscallEntry);
-
-        // FMASK: mask IF on SYSCALL entry (bit 9 = IF)
-        Hal::WriteMSR(Hal::IA32_FMASK, 0x200);
-
-        Kt::KernelLogStream(Kt::OK, "Syscall") << "SYSCALL/SYSRET initialized (LSTAR="
-            << kcp::hex << (uint64_t)SyscallEntry << kcp::dec << ", "
-            << (SYS_SPAWN_REDIR_CAPS + 1) << " syscall slots)";
     }
 
 }

kernel/src/Api/Syscall.hpp

@@ -7,6 +7,8 @@
 #pragma once
 #include <cstdint>
 #include <cstddef>
+#include <Common/StatusCodes.hpp>
+#include <Arch/Syscall.hpp>
 
 namespace montauk::abi {
 
@@ -343,6 +345,69 @@ namespace montauk::abi {
        touch in full (image decode, heap slabs). */
     static constexpr uint64_t SYS_ALLOC_EAGER          = 187;  // (bytes) -> va, 0 on failure
 
+    /* Battery.hpp -- ACPI battery status */
+    static constexpr uint64_t SYS_BATTERYINFO          = 188;  // (BatteryInfo*) -> 0, -1 no battery device
+
+    /* Process.hpp -- the caller's own capability masks. */
+    static constexpr uint64_t SYS_GETCAPS              = 189;  // (SpawnCapabilities*) -> 0, -1 bad pointer
+
+    /* Storage.hpp */
+    static constexpr uint64_t SYS_VOLINFO              = 190;  // (VolInfo*, maxCount) -> count, -1 bad pointer
+
+    /* Consent.hpp */
+    static constexpr uint64_t SYS_CONSENT_REGISTER     = 191;  // () -> 0, -13 lacks CAP_COMPOSITOR, -1 another agent is live
+    static constexpr uint64_t SYS_CONSENT_NEXT         = 192;  // (ConsentRequest*) -> 1 delivered, 0 none queued
+    static constexpr uint64_t SYS_CONSENT_REPLY        = 193;  // (requestId, CONSENT_ALLOW/DENY) -> 0, -1 stale id
+
+    /* Unicode key events. Size-versioned: the caller passes sizeof(its
+       KeyEvent); the kernel copies min(size, its own) bytes and zero-fills the
+       rest, so the struct can grow without another syscall. */
+    static constexpr uint64_t SYS_GETKEY_EX            = 194;  // (KeyEvent*, size) -> 0; blocks like SYS_GETKEY
+    static constexpr uint64_t SYS_CHILDIO_WRITEKEY_EX  = 195;  // (childPid, const KeyEvent*, size) -> status
+
+    /* Futexes: process-private wait queues keyed by a user address. WAIT
+       blocks only if *addr still equals `expected` (checked atomically with
+       enrollment). timeoutNs: FUTEX_WAIT_FOREVER = no deadline, 0 = check
+       only; rounded up to the 1 ms scheduler tick. WAIT returns 0 when woken,
+       WouldBlock on value mismatch, TimedOut on deadline. WAKE returns the
+       number of threads woken (count = FUTEX_WAKE_ALL wakes every waiter). */
+    static constexpr uint64_t SYS_FUTEX_WAIT           = 196;  // (uint32_t* addr, expected, timeoutNs) -> status
+    static constexpr uint64_t SYS_FUTEX_WAKE           = 197;  // (uint32_t* addr, count) -> woken, or status
+    static constexpr uint64_t FUTEX_WAIT_FOREVER       = ~0ULL;
+    static constexpr uint32_t FUTEX_WAKE_ALL           = 0xFFFFFFFFu;
+
+    /* Abnormal termination from any thread: ends the whole process (every
+       thread) with exit code TerminatedAborted (256+SIGABRT). SYS_EXIT from a
+       sibling thread ends only that thread, so abort() needs its own call. */
+    static constexpr uint64_t SYS_ABORT                = 198;  // () -> does not return
+
+    /* Normal termination of the whole process from any thread, with an exit
+       code (0..255) instead of SYS_ABORT's fixed TerminatedAborted. libc
+       exit() and Rust std::process::exit() use it, so exiting from a sibling
+       thread ends the process rather than only that thread. */
+    static constexpr uint64_t SYS_EXIT_GROUP           = 199;  // (exitCode) -> does not return
+
+    /* Multi-format clipboard (see "Window server messaging" below). The text
+       calls above operate on its TRANSFER_MIME_TEXT format. */
+    static constexpr uint64_t SYS_CLIPBOARD_SET_DATA   = 200;  // (const TransferItem*, count) -> status
+    static constexpr uint64_t SYS_CLIPBOARD_FORMATS    = 201;  // (TransferFormatInfo*, max, uint64_t* serial) -> count
+    static constexpr uint64_t SYS_CLIPBOARD_GET_DATA   = 202;  // (mime, buf, max, uint32_t* len) -> status
+
+    /* Window server requests and drag and drop. */
+    static constexpr uint64_t SYS_WIN_REQUEST          = 203;  // (windowId, const WinRequest*) -> status
+    static constexpr uint64_t SYS_WIN_REQUEST_POLL     = 204;  // (WinRequest*) -> 1 / 0; compositor only
+    static constexpr uint64_t SYS_WIN_DRAG_BEGIN       = 205;  // (windowId or -1 for the compositor's own surface, const WinDragBegin*, const TransferItem*, count) -> serial
+    static constexpr uint64_t SYS_WIN_DRAG_FORMATS     = 206;  // (serial, TransferFormatInfo*, max) -> count
+    static constexpr uint64_t SYS_WIN_DRAG_GET_DATA    = 207;  // (serial, mime, buf, max, uint32_t* len) -> status
+
+    /* Storage.hpp -- grow a mounted volume (and its MBR partition) to fill the
+       free space after it. apply=0 only reports; apply=1 does one bounded slice
+       and returns 1 while more remains, 0 once done. CAP_STORAGE_ADMIN. */
+    static constexpr uint64_t SYS_FS_GROW              = 208;  // (drive, FsGrowStatus*, apply) -> 1 / 0 / status
+
+    /* Time.hpp -- wall clock as UTC unix seconds (SYS_GETTIME is local time). */
+    static constexpr uint64_t SYS_GETUNIXTIME          = 209;  // () -> seconds
+
     /* Kernel-owned process capabilities. User identities may namespace
        per-user resources, but never participate in authorization decisions. */
     static constexpr uint64_t CAP_PROCESS_ADMIN = 1ULL << 0;
@@ -363,8 +428,21 @@ namespace montauk::abi {
        whatever that image is granted at its next launch. Formatting a data
        volume must not carry that authority with it. */
     static constexpr uint64_t CAP_SYSTEM_IMAGE   = 1ULL << 12;
-    static constexpr uint64_t CAP_ALL            = (1ULL << 13) - 1;
+    /* Map other processes' windows and deliver input events to them
-    static constexpr uint64_t CAP_STANDARD_SESSION = CAP_POWER_REQUEST | CAP_SUSPEND;
+       (SYS_WINMAP, SYS_WINSENDEVENT): the desktop compositor's authority.
+       Anything holding it can read every window's pixels and click on any
+       window, including a consent prompt it is itself waiting on, so it is
+       never part of a session mask, never part of "all", and never
+       delegated by the compositor to what it launches. */
+    static constexpr uint64_t CAP_COMPOSITOR     = 1ULL << 13;
+    /* Set the backlight of the active display (SYS_DISPLAYBRIGHTNESS).
+       A seat preference rather than display administration, so every
+       session holds it; CAP_DISPLAY_ADMIN keeps mode setting. Reading the
+       level needs nothing. */
+    static constexpr uint64_t CAP_BRIGHTNESS     = 1ULL << 14;
+    static constexpr uint64_t CAP_ALL            = (1ULL << 15) - 1;
+    static constexpr uint64_t CAP_STANDARD_SESSION =
+        CAP_POWER_REQUEST | CAP_SUSPEND | CAP_BRIGHTNESS;
     static constexpr uint64_t CAP_ADMIN_SESSION =
         CAP_STANDARD_SESSION | CAP_PROCESS_ADMIN | CAP_STORAGE_ADMIN |
         CAP_RAW_STORAGE | CAP_NETWORK_ADMIN | CAP_SET_TIME | CAP_USER_ADMIN |
@@ -375,28 +453,63 @@ namespace montauk::abi {
     static_assert((CAP_ADMIN_SESSION & CAP_SYSTEM_IMAGE) == 0,
                   "an admin session must not imply authority to rewrite the "
                   "programs it launches; grant CAP_SYSTEM_IMAGE per binary");
+    static_assert((CAP_ADMIN_SESSION & CAP_COMPOSITOR) == 0,
+                  "a session must not delegate the compositor's authority; "
+                  "login grants it to the desktop directly");
+    // Capability denial for the Wi-Fi and USB syscalls only, which keep their
+    // own WIFI_ERR_* / USB_ERR_* spaces; everything else returns
+    // CapabilityDenied from StatusCodes.
     static constexpr int SYS_ERR_PERMISSION = -13;
 
+    // A process holds two independent sets: what it may use (effective) and
+    // what it may pass to children (delegable). A launcher can delegate what
+    // it never uses itself, so neither set bounds the other.
     struct SpawnCapabilities {
-        uint64_t permitted;
         uint64_t effective;
         uint64_t delegable;
     };
 
+    // Both of the child's sets must come out of the parent's delegable set.
     constexpr bool ValidCapabilityDelegation(const SpawnCapabilities& child,
                                              uint64_t parentDelegable) {
-        return (child.permitted & ~CAP_ALL) == 0 &&
+        return ((child.effective | child.delegable) & ~CAP_ALL) == 0 &&
-               (child.effective & ~child.permitted) == 0 &&
+               (child.effective & ~parentDelegable) == 0 &&
-               (child.delegable & ~child.permitted) == 0 &&
-               (child.permitted & ~parentDelegable) == 0 &&
                (child.delegable & ~parentDelegable) == 0;
     }
     static_assert(ValidCapabilityDelegation(
-        {CAP_NETWORK_ADMIN, CAP_NETWORK_ADMIN, 0}, CAP_NETWORK_ADMIN));
+        {CAP_NETWORK_ADMIN, 0}, CAP_NETWORK_ADMIN));
+    static_assert(ValidCapabilityDelegation(
+        {0, CAP_NETWORK_ADMIN}, CAP_NETWORK_ADMIN));
     static_assert(!ValidCapabilityDelegation(
-        {CAP_NETWORK_ADMIN, CAP_NETWORK_ADMIN, CAP_NETWORK_ADMIN}, 0));
+        {CAP_NETWORK_ADMIN, CAP_NETWORK_ADMIN}, 0));
     static_assert(!ValidCapabilityDelegation(
-        {CAP_NETWORK_ADMIN, CAP_NETWORK_ADMIN | CAP_SET_TIME, 0}, CAP_ALL));
+        {CAP_NETWORK_ADMIN | CAP_SET_TIME, 0}, CAP_NETWORK_ADMIN));
+    static_assert(!ValidCapabilityDelegation(
+        {0, CAP_NETWORK_ADMIN | CAP_SET_TIME}, CAP_NETWORK_ADMIN));
+
+    /* Capabilities a user can grant at first use through a consent prompt.
+       A consent grant never exceeds what the process's parent could have
+       delegated to it at spawn (its consent ceiling), is never delegable
+       onwards, and is issued only by the session's consent agent -- so it is
+       deferred delegation, not amplification. */
+    static constexpr uint64_t CAP_CONSENT_ELIGIBLE =
+        CAP_STORAGE_ADMIN | CAP_NETWORK_ADMIN | CAP_SET_TIME |
+        CAP_DISPLAY_ADMIN | CAP_LOG_READ | CAP_BRIGHTNESS;
+    static_assert((CAP_CONSENT_ELIGIBLE & ~CAP_ADMIN_SESSION) == 0,
+                  "consent can only hand out authority an admin session holds");
+
+    static constexpr int CONSENT_DENY  = 0;
+    static constexpr int CONSENT_ALLOW = 1;
+    // A request left unanswered this long is denied (and not remembered).
+    static constexpr uint64_t CONSENT_TIMEOUT_MS = 120000;
+
+    struct ConsentRequest {
+        uint64_t id;
+        uint64_t capability;        // exactly one CAP_* bit
+        int32_t  pid;
+        uint32_t imageProtected;    // 1 when writing the image needs CAP_SYSTEM_IMAGE
+        char     imagePath[256];
+    };
 
 
     // Generic USB errors. Claims are restricted to interfaces without a
@@ -537,7 +650,10 @@ namespace montauk::abi {
         uint64_t txPackets;
     };
 
-    struct KeyEvent {
+    // Legacy 9-byte key event, read by SYS_GETKEY and written by
+    // SYS_CHILDIO_WRITEKEY. Kept so binaries built before codepoints still work.
+    // `ascii` is a Windows-1252 byte.
+    struct KeyEventV1 {
         uint8_t scancode;
         char    ascii;
         bool    pressed;
@@ -553,6 +669,32 @@ namespace montauk::abi {
         bool    extended;
     };
 
+    // Key event with the Unicode text it types (SYS_GETKEY_EX,
+    // SYS_CHILDIO_WRITEKEY_EX, WinEvent). Its first 9 bytes ARE a KeyEventV1, so
+    // a legacy reader sees a valid event. `codepoint` is the character the key
+    // produces (0 for control/navigation keys and dead keys). `ascii` is its
+    // Windows-1252 byte, or KEY_ASCII_SUBSTITUTE when 1252 has none.
+    struct KeyEvent {
+        uint8_t  scancode;
+        char     ascii;
+        bool     pressed;
+        bool     shift;
+        bool     ctrl;
+        bool     alt;
+        bool     capslock;
+        bool     altgr;
+        bool     extended;
+        uint8_t  _reserved[3];
+        uint32_t codepoint;
+    };
+    static_assert(sizeof(KeyEventV1) == 9, "KeyEventV1 is frozen ABI");
+    static_assert(sizeof(KeyEvent) == 16, "KeyEvent must fit the WinEvent union");
+
+    // ASCII SUB: `ascii` of a text key whose character has no Windows-1252 byte.
+    // A control character, so legacy code that only accepts ascii >= 0x20 skips
+    // it, while code testing `ascii == 0` for navigation keys is not confused.
+    static constexpr char KEY_ASCII_SUBSTITUTE = 0x1A;
+
     struct MouseState {
         int32_t  x;
         int32_t  y;
@@ -586,16 +728,55 @@ namespace montauk::abi {
     };
 
     // Window server shared types
+    enum WinEventType : uint8_t {
+        WIN_EVENT_KEY        = 0,
+        WIN_EVENT_MOUSE      = 1,
+        WIN_EVENT_RESIZE     = 2,
+        WIN_EVENT_CLOSE      = 3,
+        WIN_EVENT_SCALE      = 4,
+        // Drag and drop (WinEvent::drag). A window may read the drag's data
+        // (SYS_WIN_DRAG_*) once it has received one of these for that serial.
+        WIN_EVENT_DRAG_ENTER = 5,  // pointer entered the window during a drag
+        WIN_EVENT_DRAG_OVER  = 6,  // pointer moved; coalesced like mouse moves
+        WIN_EVENT_DRAG_LEAVE = 7,  // pointer left, or the drag was cancelled
+        WIN_EVENT_DROP       = 8,  // released here; drag.actions = chosen action
+        WIN_EVENT_DRAG_END   = 9,  // to the source only; drag.actions = action
+                                   // taken (0 = cancelled / not accepted)
+    };
+
+    // Modifier keys held at the time of a drag event (WinEvent::drag.modifiers).
+    static constexpr uint8_t WIN_MOD_SHIFT = 1u << 0;
+    static constexpr uint8_t WIN_MOD_CTRL  = 1u << 1;
+    static constexpr uint8_t WIN_MOD_ALT   = 1u << 2;
+
+    // Drag actions; a source offers a set, a drop carries exactly one.
+    static constexpr uint8_t WIN_DRAG_COPY = 1u << 0;
+    static constexpr uint8_t WIN_DRAG_MOVE = 1u << 1;
+    static constexpr uint8_t WIN_DRAG_LINK = 1u << 2;
+
+    struct WinDragEvent {
+        int32_t  x, y;       // window-local; -1,-1 when not over the window
+        uint32_t serial;     // identifies the drag
+        uint8_t  actions;    // WIN_DRAG_* (see WinEventType)
+        uint8_t  modifiers;  // WIN_MOD_*
+        uint8_t  _pad[2];
+    };
+    static_assert(sizeof(WinDragEvent) == 16, "WinDragEvent must fit the WinEvent union");
+
     struct WinEvent {
-        uint8_t type;     // 0=key, 1=mouse, 2=resize, 3=close, 4=scale
+        uint8_t type;     // WinEventType
         uint8_t _pad[3];
         union {
-            KeyEvent key;
+            KeyEvent key;   // 16 bytes: the union was already that size (mouse)
-            struct { int32_t x, y, scroll; uint8_t buttons, prev_buttons; } mouse;
+            // modifiers (WIN_MOD_*) was padding before; compositors that
+            // predate it send 0.
+            struct { int32_t x, y, scroll; uint8_t buttons, prev_buttons, modifiers; } mouse;
             struct { int32_t w, h; } resize;
             struct { int32_t scale; } scale;
+            WinDragEvent drag;   // WIN_EVENT_DRAG_* / WIN_EVENT_DROP
         };
     };
+    static_assert(sizeof(WinEvent) == 20, "WinEvent layout is shared with existing binaries");
 
     struct WinInfo {
         int32_t  id;
@@ -616,6 +797,116 @@ namespace montauk::abi {
         uint64_t pixelVa;  // VA of pixel buffer in caller's address space
     };
 
+    // ---- Window server messaging ----------------------------------------
+    //
+    // Three pieces, shared by the compositor (the desktop) and every app:
+    //   * WinEvent types 5-9 below: drag-and-drop events, compositor -> app.
+    //   * WinRequest: typed records an app posts about one of its own windows
+    //     (SYS_WIN_REQUEST); only the compositor can collect them
+    //     (SYS_WIN_REQUEST_POLL). Unknown types are ignored, so new requests
+    //     can be added without breaking older compositors or apps.
+    //   * Transfers: multi-format payloads (one MIME type per format) carried
+    //     by the clipboard (SYS_CLIPBOARD_*_DATA) and by drags
+    //     (SYS_WIN_DRAG_*). The older text clipboard calls read and write the
+    //     TRANSFER_MIME_TEXT format of the same clipboard.
+
+    // Transfers -----------------------------------------------------------
+
+    static constexpr uint32_t TRANSFER_MAX_FORMATS = 8;
+    static constexpr uint32_t TRANSFER_MIME_MAX    = 64;          // incl. NUL
+    static constexpr uint32_t TRANSFER_MAX_BYTES   = 256 * 1024;  // all formats together
+
+    // Well-known formats. Any other "type/subtype" string is allowed.
+    static constexpr const char* TRANSFER_MIME_TEXT      = "text/plain;charset=utf-8";
+    // UTF-8 absolute paths ("0:/..."), one per line.
+    static constexpr const char* TRANSFER_MIME_FILE_LIST = "application/x-montauk-file-list";
+    // Clipboard companion to a file list: "copy" or "cut".
+    static constexpr const char* TRANSFER_MIME_FILE_OP   = "application/x-montauk-file-op";
+
+    // One format to store. Pointers are carried as 64-bit values so the
+    // layout is the same for every caller.
+    struct TransferItem {
+        uint64_t mime;       // const char*, NUL-terminated, < TRANSFER_MIME_MAX
+        uint64_t data;       // const void*
+        uint32_t len;
+        uint32_t _reserved;
+    };
+    static_assert(sizeof(TransferItem) == 24, "TransferItem is ABI");
+
+    struct TransferFormatInfo {
+        char     mime[TRANSFER_MIME_MAX];
+        uint32_t len;
+        uint32_t _reserved;
+    };
+    static_assert(sizeof(TransferFormatInfo) == 72, "TransferFormatInfo is ABI");
+
+    // Requests --------------------------------------------------------------
+
+    static constexpr uint32_t WIN_REQUEST_PAYLOAD_MAX = 240;
+
+    struct WinRequest {
+        uint16_t type;       // WinRequestType
+        uint16_t size;       // payload bytes in use
+        int32_t  windowId;   // window the request is about (the sender must own it
+                             // when posting); -1 if it was destroyed since
+        int32_t  senderPid;  // filled in by the kernel
+        uint32_t serial;     // filled in by the kernel; increases per request
+        uint8_t  payload[WIN_REQUEST_PAYLOAD_MAX];
+    };
+    static_assert(sizeof(WinRequest) == 256, "WinRequest is ABI");
+
+    enum WinRequestType : uint16_t {
+        WIN_REQ_LAUNCH_ITEM  = 1,      // WinReqLaunchItem
+        WIN_REQ_SET_PARENT   = 2,      // WinReqSetParent
+        WIN_REQ_DRAG_STATUS  = 3,      // WinReqDragStatus
+        WIN_REQ_ACTIVATE     = 4,      // no payload: raise and focus the window
+        // Types from here on are generated by the kernel; SYS_WIN_REQUEST
+        // refuses them so an app cannot forge one.
+        WIN_REQ_KERNEL_BASE  = 0x8000,
+        WIN_REQ_DRAG_BEGIN   = 0x8001, // WinReqDragBegin (from SYS_WIN_DRAG_BEGIN)
+    };
+
+    // Open a desktop catalog item (an app manifest id, or a built-in such as
+    // "desktop-settings"), as if the user had picked it from the launcher, so
+    // it runs with the authority the desktop would give it. `args` reach only
+    // apps whose manifest declares files.types (they take a document path);
+    // anything else starts without them, so a request cannot steer a
+    // privileged app through its command line.
+    struct WinReqLaunchItem {
+        char itemId[48];
+        char args[192];
+    };
+
+    // Mark the window as a dialog of another window owned by the same
+    // process: the compositor centres it over the parent.
+    struct WinReqSetParent {
+        int32_t parentWindowId;
+    };
+
+    // Answer to DRAG_ENTER/DRAG_OVER: the action this window would perform
+    // on a drop at the current position (0 = would refuse). Advisory; it
+    // drives the compositor's drag feedback.
+    struct WinReqDragStatus {
+        uint32_t serial;
+        uint8_t  action;     // one WIN_DRAG_* or 0
+        uint8_t  _pad[3];
+    };
+
+    // Describes a drag for the compositor's feedback. Supplied by the app to
+    // SYS_WIN_DRAG_BEGIN and relayed (with the serial) as WIN_REQ_DRAG_BEGIN.
+    struct WinDragBegin {
+        uint8_t  actions;    // WIN_DRAG_* offered
+        uint8_t  _pad[3];
+        uint32_t itemCount;  // for a "3 items" badge; 0 = do not show
+        char     label[64];  // short description, e.g. the file name
+    };
+    static_assert(sizeof(WinDragBegin) == 72, "WinDragBegin is ABI");
+
+    struct WinReqDragBegin {
+        uint32_t     serial;
+        WinDragBegin info;
+    };
+
     struct MemStats {
         uint64_t totalBytes;
         uint64_t freeBytes;
@@ -632,7 +923,7 @@ namespace montauk::abi {
 
     struct DiskInfo {
         uint8_t  port;              // block device index
-        uint8_t  type;              // 0=none, 1=SATA, 2=SATAPI, 3=NVMe, 4=USB mass storage
+        uint8_t  type;              // 0=none, 1=SATA, 2=SATAPI, 3=NVMe, 4=USB mass storage, 5=SD card
         uint8_t  sataGen;           // SATA gen (1/2/3)
         uint8_t  _pad0;
         uint64_t sectorCount;       // Total user-addressable sectors
@@ -673,6 +964,28 @@ namespace montauk::abi {
         char     typeName[24];    // human-readable type name
     };
 
+    // One mounted volume (returned by SYS_VOLINFO). A volume is a single
+    // mounted filesystem, which is what a drive number names: one physical
+    // disk with three partitions occupies three drive numbers.
+    //
+    // A volume with no backing partition -- the boot ramdisk -- reports
+    // blockDev and partIndex as -1 and leaves the geometry fields zero.
+    struct VolInfo {
+        int32_t  drive;           // VFS drive number ("N:/")
+        int32_t  blockDev;        // backing block device index, -1 if none
+        int32_t  partIndex;       // global GPT partition index, -1 if none
+        uint32_t kind;            // BlockDeviceKind: 0=none/ramdisk, 1=SATA, 2=SATAPI, 3=NVMe, 4=USB MSC
+        uint64_t startLba;        // partition start LBA, 0 if none
+        uint64_t sectorCount;     // partition length in sectors, 0 if none
+        uint32_t sectorSize;      // logical sector size in bytes, 0 if none
+        uint32_t _pad0;
+        char     label[36];       // filesystem volume label, "" if unset
+        char     fsType[12];      // "ext2", "FAT32", "ramdisk", "" if unknown
+        char     partName[72];    // GPT partition name, "" if none
+        char     model[41];       // backing disk model, "" if none
+        char     _pad1[3];
+    };
+
     struct GptAddParams {
         int32_t  blockDev;
         uint32_t _pad0;
@@ -692,6 +1005,14 @@ namespace montauk::abi {
         char     label[32];       // volume label
     };
 
+    struct FsGrowStatus {
+        uint64_t fsBytes;         // filesystem size now
+        uint64_t partBytes;       // partition size now
+        uint64_t targetBytes;     // filesystem size once grown; == fsBytes if it cannot grow
+        uint32_t stepsDone;       // progress of the grow, out of stepsTotal
+        uint32_t stepsTotal;
+    };
+
     struct ProcInfo {
         int32_t  pid;
         int32_t  parentPid;
@@ -700,7 +1021,6 @@ namespace montauk::abi {
         char     name[64];
         uint64_t heapUsed;     // Distance from UserHeapBase to high-water mark
         uint64_t cpuTimeMs;    // accumulated scheduler runtime
-        uint64_t permittedCaps;
         uint64_t effectiveCaps;
         uint64_t delegableCaps;
     };
@@ -881,6 +1201,26 @@ namespace montauk::abi {
         uint32_t apIdleHint;     // MWAIT hint used for AP deep idle
     };
 
+    // Battery status (returned by SYS_BATTERYINFO). A machine with no battery
+    // device fails the call outright; `present` distinguishes a laptop with the
+    // pack removed from one that is running on it.
+    struct BatteryInfo {
+        uint8_t  present;            // a battery is installed
+        uint8_t  charging;
+        uint8_t  discharging;
+        uint8_t  critical;
+        uint8_t  acOnline;           // AC adapter connected
+        uint8_t  hasAcAdapter;       // an AC adapter device exists at all
+        uint8_t  percent;            // 0..100, 0 when capacity is unknown
+        uint8_t  powerUnitMilliAmps; // capacities are mAh/mA, not mWh/mW
+        uint32_t remainingCapacity;
+        uint32_t fullCapacity;       // last full charge capacity
+        uint32_t designCapacity;
+        uint32_t presentRate;        // mW or mA, 0 when unknown
+        uint32_t voltageMv;
+        int32_t  minutesRemaining;   // -1 when not estimable (e.g. charging)
+    };
+
     // Crash report (filled by kernel on process fault, returned via SYS_CRASH_REPORT)
     struct CrashReportInfo {
         int pid;
@@ -906,15 +1246,11 @@ namespace montauk::abi {
         uint64_t timestampTick;
     };
 
-    // Stack frame pushed by SyscallEntry.asm
+    // Saved user context of a system call; its layout is the
-    struct SyscallFrame {
+    // architecture's (see Arch/Syscall.hpp).
-        uint64_t r15, r14, r13, r12, rbp, rbx;   // callee-saved
+    using SyscallFrame = Arch::SyscallFrame;
-        uint64_t arg6, arg5, arg4, arg3, arg2, arg1;
-        uint64_t syscall_nr;
-        uint64_t user_rflags, user_rip, user_rsp;
-    };
 
-    // Kernel-only: set up SYSCALL MSRs and initialize dispatch
+    // Kernel-only: install the system-call entry point and initialize dispatch
     void InitializeSyscalls();
 
 }

kernel/src/Api/Time.hpp

@@ -1,11 +1,11 @@
 /*
     * Time.hpp
-    * SYS_GETTICKS, SYS_GETMILLISECONDS, SYS_GETTIME syscalls
+    * SYS_GETTICKS, SYS_GETMILLISECONDS, SYS_GETTIME, SYS_GETUNIXTIME syscalls
     * Copyright (c) 2026 Daniel Hammer
 */
 
 #pragma once
-#include <Timekeeping/ApicTimer.hpp>
+#include <Timekeeping/Timer.hpp>
 #include <Timekeeping/Time.hpp>
 
 #include "Syscall.hpp"
@@ -32,14 +32,18 @@ namespace montauk::abi {
 
     static int64_t Sys_SetTZ(int32_t offsetMinutes) {
         Timekeeping::SetTZOffset(offsetMinutes);
-        return 0;
+        return Success;
     }
 
     static int64_t Sys_GetTZ() {
         return (int64_t)Timekeeping::GetTZOffset();
     }
 
+    static int64_t Sys_GetUnixTime() {
+        return Timekeeping::GetUnixTimestamp();
+    }
+
     static int64_t Sys_SetUnixTime(int64_t unixSeconds) {
-        return Timekeeping::SetUnixTimestamp(unixSeconds) ? 0 : -1;
+        return Timekeeping::SetUnixTimestamp(unixSeconds) ? Success : InvalidArgument;
     }
 };

kernel/src/Api/UserMemory.hpp

@@ -19,9 +19,9 @@ namespace montauk::abi::UserMemory {
         return addr != 0 && addr < USER_SPACE_END;
     }
 
-    inline uint64_t CurrentPml4() {
+    inline uint64_t CurrentRoot() {
         auto* proc = Sched::GetCurrentProcessPtr();
-        return proc ? proc->pml4Phys : 0;
+        return proc ? proc->pageTableRoot : 0;
     }
 
     inline bool Range(uint64_t addr, uint64_t size, bool requireWrite = false) {
@@ -30,10 +30,10 @@ namespace montauk::abi::UserMemory {
         if (addr + size < addr) return false;
         if (addr + size > USER_SPACE_END) return false;
 
-        uint64_t pml4Phys = CurrentPml4();
+        uint64_t pageTableRoot = CurrentRoot();
-        if (pml4Phys == 0) return false;
+        if (pageTableRoot == 0) return false;
 
-        if (Memory::VMM::Paging::IsUserRangeAccessible(pml4Phys, addr, size, requireWrite))
+        if (Memory::VMM::Paging::IsUserRangeAccessible(pageTableRoot, addr, size, requireWrite))
             return true;
 
         // Anonymous VMAs are demand-committed. Validation precedes the actual
@@ -42,13 +42,13 @@ namespace montauk::abi::UserMemory {
         uint64_t page = addr & ~0xFFFULL;
         uint64_t last = (addr + size - 1) & ~0xFFFULL;
         for (;;) {
-            if (!Memory::VMM::Paging::IsUserRangeAccessible(pml4Phys, page, 1, requireWrite) &&
+            if (!Memory::VMM::Paging::IsUserRangeAccessible(pageTableRoot, page, 1, requireWrite) &&
-                !TryHandleAnonymousPageFault(page, requireWrite ? 2ULL : 0ULL))
+                !TryHandleAnonymousPageFault(page, {false, requireWrite, false}))
                 return false;
             if (page == last) break;
             page += 0x1000ULL;
         }
-        return Memory::VMM::Paging::IsUserRangeAccessible(pml4Phys, addr, size, requireWrite);
+        return Memory::VMM::Paging::IsUserRangeAccessible(pageTableRoot, addr, size, requireWrite);
     }
 
     inline bool OptionalRange(uint64_t addr, uint64_t size, bool requireWrite = false) {
@@ -68,13 +68,13 @@ namespace montauk::abi::UserMemory {
     inline bool String(uint64_t addr, uint64_t maxLen) {
         if (!ValidUserPtr(addr) || maxLen == 0) return false;
 
-        uint64_t pml4Phys = CurrentPml4();
+        uint64_t pageTableRoot = CurrentRoot();
-        if (pml4Phys == 0) return false;
+        if (pageTableRoot == 0) return false;
 
         const char* str = (const char*)addr;
         for (uint64_t i = 0; i < maxLen; i++) {
-            if (!Memory::VMM::Paging::IsUserRangeAccessible(pml4Phys, addr + i, 1, false)) {
+            if (!Memory::VMM::Paging::IsUserRangeAccessible(pageTableRoot, addr + i, 1, false)) {
-                if (!TryHandleAnonymousPageFault(addr + i, 0)) return false;
+                if (!TryHandleAnonymousPageFault(addr + i, {false, false, false})) return false;
             }
             if (str[i] == '\0') return true;
         }

kernel/src/Api/WinServer.cpp

@@ -54,14 +54,14 @@ namespace WinServer {
         Ipc::ReleaseSurface(surface);
     }
 
-    static int UnmapRetiredSnapshotsLocked(int windowId, int callerPid, uint64_t callerPml4) {
+    static int UnmapRetiredSnapshotsLocked(int windowId, int callerPid, uint64_t callerRoot) {
-        int result = -1;
+        int result = InvalidArgument;
         for (int i = 0; i < MaxRetiredSnapshots; i++) {
             if (!g_retiredSnapshots[i].used || g_retiredSnapshots[i].windowId != windowId) continue;
 
             if (g_retiredSnapshots[i].surface != nullptr) {
-                if (Ipc::UnmapSurfaceForPid(g_retiredSnapshots[i].surface, callerPid, callerPml4) == 0) {
+                if (Ipc::UnmapSurfaceForPid(g_retiredSnapshots[i].surface, callerPid, callerRoot) == 0) {
-                    result = 0;
+                    result = Success;
                 }
                 Ipc::ReleaseSurface(g_retiredSnapshots[i].surface);
             }
@@ -73,11 +73,18 @@ namespace WinServer {
         return result;
     }
 
+    static void OrphanRequestsLocked(int windowId);
+
     static void ReleaseSlotResourcesLocked(WindowSlot& slot, bool unmapOwner) {
+        // Queued requests stay deliverable (a dialog may post "open this
+        // app" and close at once), but no longer name a window that could be
+        // handed to someone else.
+        OrphanRequestsLocked((int)(&slot - g_slots));
+
         if (unmapOwner && slot.liveSurface != nullptr) {
             auto* owner = Sched::GetProcessByPid(slot.ownerPid);
             if (owner != nullptr) {
-                Ipc::UnmapSurfaceForPid(slot.liveSurface, slot.ownerPid, owner->pml4Phys);
+                Ipc::UnmapSurfaceForPid(slot.liveSurface, slot.ownerPid, owner->pageTableRoot);
             }
         }
 
@@ -94,22 +101,67 @@ namespace WinServer {
         ResetSlotLocked(slot);
     }
 
-    static int SendEventLocked(int windowId, const montauk::abi::WinEvent* event) {
+    // ---- Messaging state ----
-        if (windowId < 0 || windowId >= MaxWindows || event == nullptr) return -1;
-        WindowSlot& slot = g_slots[windowId];
-        if (!slot.used || slot.eventMailbox == nullptr) return -1;
 
-        int rc = (event->type == 1)
+    static constexpr int MaxRequests = 64;
-            ? Ipc::MailboxSendCoalescedMouse(slot.eventMailbox, 0, event, sizeof(*event))
+    // One app may not hold more than this many queued requests, so a single
-            : Ipc::MailboxSend(slot.eventMailbox, 0, event, sizeof(*event));
+    // misbehaving process cannot starve the others out of the queue.
-        return rc > 0 ? 0 : -1;
+    static constexpr int MaxRequestsPerSender = 16;
+    static constexpr int MaxDragReaders = 8;
+
+    static montauk::abi::WinRequest g_requests[MaxRequests];
+    static int g_requestHead = 0;   // oldest
+    static int g_requestCount = 0;
+    static uint32_t g_requestSerial = 0;
+
+    struct DragState {
+        bool active;
+        uint32_t serial;
+        int sourceWindow;
+        int sourcePid;
+        DataTransfer::Transfer transfer;
+        int readers[MaxDragReaders];
+        int readerCount;
+    };
+    static DragState g_drag;
+    static uint32_t g_dragSerial = 0;
+
+    static bool IsDragEvent(uint8_t type) {
+        return type >= montauk::abi::WIN_EVENT_DRAG_ENTER && type <= montauk::abi::WIN_EVENT_DRAG_END;
     }
 
-    int Create(int ownerPid, uint64_t ownerPml4, const char* title, int w, int h,
+    static void GrantDragReaderLocked(uint32_t serial, int pid) {
+        if (!g_drag.active || g_drag.serial != serial || pid <= 0) return;
+        if (pid == g_drag.sourcePid) return;
+        for (int i = 0; i < g_drag.readerCount; i++) {
+            if (g_drag.readers[i] == pid) return;
+        }
+        if (g_drag.readerCount < MaxDragReaders) g_drag.readers[g_drag.readerCount++] = pid;
+    }
+
+    static int SendEventLocked(int windowId, const montauk::abi::WinEvent* event) {
+        if (windowId < 0 || windowId >= MaxWindows || event == nullptr) return InvalidArgument;
+        WindowSlot& slot = g_slots[windowId];
+        if (!slot.used || slot.eventMailbox == nullptr) return InvalidArgument;
+
+        // Delivering a drag event is what lets the receiving app read the
+        // drag's data: it needs the formats to decide whether to accept.
+        if (IsDragEvent(event->type)) GrantDragReaderLocked(event->drag.serial, slot.ownerPid);
+
+        bool motion = event->type == montauk::abi::WIN_EVENT_MOUSE ||
+                      event->type == montauk::abi::WIN_EVENT_DRAG_OVER;
+        int rc = motion
+            ? Ipc::MailboxSendCoalescedMotion(slot.eventMailbox, 0, event, sizeof(*event))
+            : Ipc::MailboxSend(slot.eventMailbox, 0, event, sizeof(*event));
+        if (rc < 0) return GenericFailure;
+        return rc > 0 ? Success : Busy;
+    }
+
+    int Create(int ownerPid, uint64_t ownerRoot, const char* title, int w, int h,
                uint64_t& heapNext, uint64_t& outVa) {
         WsGuard guard;
 
-        if (w <= 0 || h <= 0 || w > 16384 || h > 16384) return -1;
+        if (w <= 0 || h <= 0 || w > 16384 || h > 16384) return InvalidArgument;
 
         int slotIdx = -1;
         for (int i = 0; i < MaxWindows; i++) {
@@ -118,7 +170,7 @@ namespace WinServer {
                 break;
             }
         }
-        if (slotIdx < 0) return -1;
+        if (slotIdx < 0) return GenericFailure;
 
         uint64_t bufSize = (uint64_t)w * (uint64_t)h * 4ULL;
         Ipc::Surface* live = Ipc::CreateSurface(bufSize);
@@ -128,7 +180,7 @@ namespace WinServer {
             if (live != nullptr) Ipc::ReleaseSurface(live);
             if (snapshot != nullptr) Ipc::ReleaseSurface(snapshot);
             if (mailbox != nullptr) Ipc::ReleaseMailbox(mailbox, false, false);
-            return -1;
+            return OutOfMemory;
         }
 
         Ipc::RetainSurface(live);
@@ -142,11 +194,11 @@ namespace WinServer {
         FillSurfaceLocked(snapshot, w, h, InitialWindowPixel);
 
         uint64_t userVa = 0;
-        if (Ipc::MapSurfaceForPid(live, ownerPid, ownerPml4, heapNext, userVa, false) != 0) {
+        if (Ipc::MapSurfaceForPid(live, ownerPid, ownerRoot, heapNext, userVa, false) != 0) {
             Ipc::ReleaseMailbox(mailbox, true, true);
             Ipc::ReleaseSurface(snapshot);
             Ipc::ReleaseSurface(live);
-            return -1;
+            return GenericFailure;
         }
 
         WindowSlot& slot = g_slots[slotIdx];
@@ -176,40 +228,41 @@ namespace WinServer {
 
     int Destroy(int windowId, int callerPid) {
         WsGuard guard;
-        if (windowId < 0 || windowId >= MaxWindows) return -1;
+        if (windowId < 0 || windowId >= MaxWindows) return InvalidArgument;
 
         WindowSlot& slot = g_slots[windowId];
-        if (!slot.used || slot.ownerPid != callerPid) return -1;
+        if (!slot.used) return InvalidArgument;
+        if (slot.ownerPid != callerPid) return PermissionDenied;
 
         ReleaseSlotResourcesLocked(slot, true);
-        return 0;
+        return Success;
     }
 
     int Present(int windowId, int callerPid) {
         WsGuard guard;
-        if (windowId < 0 || windowId >= MaxWindows) return -1;
+        if (windowId < 0 || windowId >= MaxWindows) return InvalidArgument;
 
         WindowSlot& slot = g_slots[windowId];
-        if (!slot.used || slot.ownerPid != callerPid ||
+        if (!slot.used || slot.liveSurface == nullptr || slot.snapshotSurface == nullptr)
-            slot.liveSurface == nullptr || slot.snapshotSurface == nullptr) {
+            return InvalidArgument;
-            return -1;
+        if (slot.ownerPid != callerPid) return PermissionDenied;
-        }
 
-        if (Ipc::CopySurface(slot.snapshotSurface, slot.liveSurface) != 0) return -1;
+        if (Ipc::CopySurface(slot.snapshotSurface, slot.liveSurface) != 0) return GenericFailure;
         slot.dirty = true;
-        return 0;
+        return Success;
     }
 
     int Poll(int windowId, int callerPid, montauk::abi::WinEvent* outEvent) {
         WsGuard guard;
-        if (windowId < 0 || windowId >= MaxWindows || outEvent == nullptr) return -1;
+        if (windowId < 0 || windowId >= MaxWindows || outEvent == nullptr) return InvalidArgument;
 
         WindowSlot& slot = g_slots[windowId];
-        if (!slot.used || slot.ownerPid != callerPid || slot.eventMailbox == nullptr) return -1;
+        if (!slot.used || slot.eventMailbox == nullptr) return InvalidArgument;
+        if (slot.ownerPid != callerPid) return PermissionDenied;
 
         uint16_t len = sizeof(*outEvent);
         int rc = Ipc::MailboxRecv(slot.eventMailbox, nullptr, outEvent, &len, true);
-        if (rc < 0) return -1;
+        if (rc < 0) return GenericFailure;
         return rc > 0 ? 1 : 0;
     }
 
@@ -234,7 +287,7 @@ namespace WinServer {
         return count;
     }
 
-    uint64_t Map(int windowId, int callerPid, uint64_t callerPml4, uint64_t& heapNext) {
+    uint64_t Map(int windowId, int callerPid, uint64_t callerRoot, uint64_t& heapNext) {
         WsGuard guard;
         if (windowId < 0 || windowId >= MaxWindows) return 0;
 
@@ -242,24 +295,24 @@ namespace WinServer {
         if (!slot.used || slot.snapshotSurface == nullptr) return 0;
 
         uint64_t outVa = 0;
-        if (Ipc::MapSurfaceForPid(slot.snapshotSurface, callerPid, callerPml4, heapNext, outVa, false) != 0) {
+        if (Ipc::MapSurfaceForPid(slot.snapshotSurface, callerPid, callerRoot, heapNext, outVa, false) != 0) {
             return 0;
         }
         return outVa;
     }
 
-    int Unmap(int windowId, int callerPid, uint64_t callerPml4) {
+    int Unmap(int windowId, int callerPid, uint64_t callerRoot) {
         WsGuard guard;
         if (windowId < 0 || windowId >= MaxWindows) {
-            return UnmapRetiredSnapshotsLocked(windowId, callerPid, callerPml4);
+            return UnmapRetiredSnapshotsLocked(windowId, callerPid, callerRoot);
         }
 
-        int result = UnmapRetiredSnapshotsLocked(windowId, callerPid, callerPml4);
+        int result = UnmapRetiredSnapshotsLocked(windowId, callerPid, callerRoot);
         WindowSlot& slot = g_slots[windowId];
         if (!slot.used || slot.snapshotSurface == nullptr) return result;
 
-        int current = Ipc::UnmapSurfaceForPid(slot.snapshotSurface, callerPid, callerPml4);
+        int current = Ipc::UnmapSurfaceForPid(slot.snapshotSurface, callerPid, callerRoot);
-        if (current == 0) return 0;
+        if (current == 0) return Success;
         return result;
     }
 
@@ -268,17 +321,18 @@ namespace WinServer {
         return SendEventLocked(windowId, event);
     }
 
-    int Resize(int windowId, int callerPid, uint64_t ownerPml4, int newW, int newH,
+    int Resize(int windowId, int callerPid, uint64_t ownerRoot, int newW, int newH,
                uint64_t& heapNext, uint64_t& outVa) {
         WsGuard guard;
-        if (windowId < 0 || windowId >= MaxWindows) return -1;
+        if (windowId < 0 || windowId >= MaxWindows) return InvalidArgument;
 
         WindowSlot& slot = g_slots[windowId];
-        if (!slot.used || slot.ownerPid != callerPid) return -1;
+        if (!slot.used) return InvalidArgument;
-        if (newW <= 0 || newH <= 0 || newW > 16384 || newH > 16384) return -1;
+        if (slot.ownerPid != callerPid) return PermissionDenied;
+        if (newW <= 0 || newH <= 0 || newW > 16384 || newH > 16384) return InvalidArgument;
         if (newW == slot.width && newH == slot.height) {
             outVa = slot.ownerVa;
-            return 0;
+            return Success;
         }
 
         uint64_t bufSize = (uint64_t)newW * (uint64_t)newH * 4ULL;
@@ -287,17 +341,17 @@ namespace WinServer {
         if (newLive == nullptr || newSnapshot == nullptr) {
             if (newLive != nullptr) Ipc::ReleaseSurface(newLive);
             if (newSnapshot != nullptr) Ipc::ReleaseSurface(newSnapshot);
-            return -1;
+            return OutOfMemory;
         }
 
         Ipc::RetainSurface(newLive);
         Ipc::RetainSurface(newSnapshot);
 
         uint64_t newOwnerVa = 0;
-        if (Ipc::MapSurfaceForPid(newLive, callerPid, ownerPml4, heapNext, newOwnerVa, false) != 0) {
+        if (Ipc::MapSurfaceForPid(newLive, callerPid, ownerRoot, heapNext, newOwnerVa, false) != 0) {
             Ipc::ReleaseSurface(newSnapshot);
             Ipc::ReleaseSurface(newLive);
-            return -1;
+            return GenericFailure;
         }
 
         if (slot.liveSurface != nullptr) {
@@ -312,7 +366,7 @@ namespace WinServer {
         }
 
         if (slot.liveSurface != nullptr) {
-            Ipc::UnmapSurfaceForPid(slot.liveSurface, callerPid, ownerPml4);
+            Ipc::UnmapSurfaceForPid(slot.liveSurface, callerPid, ownerRoot);
             Ipc::ReleaseSurface(slot.liveSurface);
         }
         if (slot.snapshotSurface != nullptr) {
@@ -327,27 +381,29 @@ namespace WinServer {
         slot.dirty = true;
 
         outVa = newOwnerVa;
-        return 0;
+        return Success;
     }
 
     int SetCursor(int windowId, int callerPid, int cursor) {
         WsGuard guard;
-        if (windowId < 0 || windowId >= MaxWindows) return -1;
+        if (windowId < 0 || windowId >= MaxWindows) return InvalidArgument;
 
         WindowSlot& slot = g_slots[windowId];
-        if (!slot.used || slot.ownerPid != callerPid) return -1;
+        if (!slot.used) return InvalidArgument;
+        if (slot.ownerPid != callerPid) return PermissionDenied;
         slot.cursor = (uint8_t)cursor;
-        return 0;
+        return Success;
     }
 
     int SetFlags(int windowId, int callerPid, uint32_t flags) {
         WsGuard guard;
-        if (windowId < 0 || windowId >= MaxWindows) return -1;
+        if (windowId < 0 || windowId >= MaxWindows) return InvalidArgument;
 
         WindowSlot& slot = g_slots[windowId];
-        if (!slot.used || slot.ownerPid != callerPid) return -1;
+        if (!slot.used) return InvalidArgument;
+        if (slot.ownerPid != callerPid) return PermissionDenied;
         slot.flags = flags & montauk::abi::WIN_FLAG_FULLSCREEN;
-        return 0;
+        return Success;
     }
 
     int SetScale(int scale) {
@@ -364,19 +420,160 @@ namespace WinServer {
                 SendEventLocked(i, &ev);
             }
         }
-        return 0;
+        return Success;
     }
 
     int GetScale() {
         return g_uiScale;
     }
 
+    static void OrphanRequestsLocked(int windowId) {
+        for (int i = 0; i < g_requestCount; i++) {
+            montauk::abi::WinRequest& r = g_requests[(g_requestHead + i) % MaxRequests];
+            if (r.windowId == windowId) r.windowId = -1;
+        }
+    }
+
+    static void EndDragLocked() {
+        DataTransfer::Free(g_drag.transfer);
+        g_drag.active = false;
+        g_drag.readerCount = 0;
+    }
+
     void CleanupProcess(int pid) {
         WsGuard guard;
         for (int i = 0; i < MaxWindows; i++) {
             if (!g_slots[i].used || g_slots[i].ownerPid != pid) continue;
             ReleaseSlotResourcesLocked(g_slots[i], false);
         }
+        // Requests the process already posted are still delivered: they are
+        // messages in flight, and a process may well exit right after its
+        // last one ("open this file" from a dialog-only Files).
+        if (g_drag.active && g_drag.sourcePid == pid) EndDragLocked();
+    }
+
+    static int EnqueueRequestLocked(montauk::abi::WinRequest& request) {
+        if (g_requestCount >= MaxRequests) return WouldBlock;
+        request.serial = ++g_requestSerial;
+        if (request.serial == 0) request.serial = ++g_requestSerial;
+        g_requests[(g_requestHead + g_requestCount) % MaxRequests] = request;
+        g_requestCount++;
+        return Success;
+    }
+
+    static bool OwnsWindowLocked(int windowId, int pid) {
+        return windowId >= 0 && windowId < MaxWindows && g_slots[windowId].used &&
+               g_slots[windowId].ownerPid == pid;
+    }
+
+    int PostRequest(int windowId, int callerPid, const montauk::abi::WinRequest& request) {
+        using namespace montauk::abi;
+        if (request.type == 0 || request.type >= WIN_REQ_KERNEL_BASE) return InvalidArgument;
+        if (request.size > WIN_REQUEST_PAYLOAD_MAX) return InvalidArgument;
+
+        WsGuard guard;
+        if (!OwnsWindowLocked(windowId, callerPid)) return PermissionDenied;
+
+        int mine = 0;
+        for (int i = 0; i < g_requestCount; i++) {
+            if (g_requests[(g_requestHead + i) % MaxRequests].senderPid == callerPid) mine++;
+        }
+        if (mine >= MaxRequestsPerSender) return WouldBlock;
+
+        WinRequest copy = request;
+        copy.windowId = windowId;
+        copy.senderPid = callerPid;
+        // Bytes past `size` are not the sender's business to pass along.
+        for (uint32_t i = copy.size; i < WIN_REQUEST_PAYLOAD_MAX; i++) copy.payload[i] = 0;
+        return EnqueueRequestLocked(copy);
+    }
+
+    int PollRequest(montauk::abi::WinRequest* out) {
+        WsGuard guard;
+        while (g_requestCount > 0) {
+            montauk::abi::WinRequest& r = g_requests[g_requestHead];
+            g_requestHead = (g_requestHead + 1) % MaxRequests;
+            g_requestCount--;
+            // A destroyed window's requests arrive with windowId -1 (see
+            // OrphanRequestsLocked); anything else must still match.
+            if (r.windowId >= 0 && !OwnsWindowLocked(r.windowId, r.senderPid)) continue;
+            *out = r;
+            return 1;
+        }
+        return 0;
+    }
+
+    int DragBegin(int windowId, int callerPid, bool compositor,
+                  const montauk::abi::WinDragBegin& info, DataTransfer::Transfer& transfer) {
+        using namespace montauk::abi;
+        WsGuard guard;
+        // The compositor drags from its own surface (desktop icons) with
+        // windowId -1; it already knows about such a drag, so no request.
+        bool fromSurface = compositor && windowId == -1;
+        if (!fromSurface && !OwnsWindowLocked(windowId, callerPid)) {
+            DataTransfer::Free(transfer);
+            return PermissionDenied;
+        }
+
+        uint32_t serial = ++g_dragSerial;
+        if (serial == 0) serial = ++g_dragSerial;
+
+        if (!fromSurface) {
+            WinRequest request = {};
+            request.type = WIN_REQ_DRAG_BEGIN;
+            request.size = sizeof(WinReqDragBegin);
+            request.windowId = windowId;
+            request.senderPid = callerPid;
+            WinReqDragBegin* payload = (WinReqDragBegin*)request.payload;
+            payload->serial = serial;
+            payload->info = info;
+            payload->info.label[sizeof(payload->info.label) - 1] = '\0';
+            int rc = EnqueueRequestLocked(request);
+            if (rc != Success) {
+                DataTransfer::Free(transfer);
+                return rc;
+            }
+        }
+
+        // A new drag replaces the previous one (only one pointer exists).
+        if (g_drag.active) EndDragLocked();
+        g_drag.active = true;
+        g_drag.serial = serial;
+        g_drag.sourceWindow = windowId;
+        g_drag.sourcePid = callerPid;
+        g_drag.transfer = transfer;
+        g_drag.readerCount = 0;
+        transfer.count = 0;
+        return (int)serial;
+    }
+
+    static bool DragReadableLocked(uint32_t serial, int callerPid, bool compositor) {
+        if (!g_drag.active || g_drag.serial != serial) return false;
+        if (compositor || callerPid == g_drag.sourcePid) return true;
+        for (int i = 0; i < g_drag.readerCount; i++) {
+            if (g_drag.readers[i] == callerPid) return true;
+        }
+        return false;
+    }
+
+    int DragFormats(uint32_t serial, int callerPid, bool compositor,
+                    montauk::abi::TransferFormatInfo* out, uint32_t max) {
+        WsGuard guard;
+        if (!g_drag.active || g_drag.serial != serial) return NotFound;
+        if (!DragReadableLocked(serial, callerPid, compositor)) return PermissionDenied;
+        if (out == nullptr || max == 0) return (int)g_drag.transfer.count;
+        return (int)DataTransfer::ListFormats(g_drag.transfer, out, max);
+    }
+
+    int DragGetData(uint32_t serial, int callerPid, bool compositor, const char* mime,
+                    void* buffer, uint32_t bufferLen, uint32_t* outLen) {
+        WsGuard guard;
+        if (outLen != nullptr) *outLen = 0;
+        if (!g_drag.active || g_drag.serial != serial) return NotFound;
+        if (!DragReadableLocked(serial, callerPid, compositor)) return PermissionDenied;
+        const DataTransfer::Format* format = DataTransfer::Find(g_drag.transfer, mime);
+        if (format == nullptr) return NotFound;
+        return DataTransfer::CopyOut(*format, buffer, bufferLen, outLen);
     }
 
 }

kernel/src/Api/WinServer.hpp

@@ -8,6 +8,7 @@
 #include "Syscall.hpp"
 #include <cstdint>
 #include <Ipc/Ipc.hpp>
+#include "DataTransfer.hpp"
 
 namespace WinServer {
 
@@ -27,16 +28,16 @@ namespace WinServer {
         uint32_t flags;
     };
 
-    int Create(int ownerPid, uint64_t ownerPml4, const char* title, int w, int h,
+    int Create(int ownerPid, uint64_t ownerRoot, const char* title, int w, int h,
                uint64_t& heapNext, uint64_t& outVa);
     int Destroy(int windowId, int callerPid);
     int Present(int windowId, int callerPid);
     int Poll(int windowId, int callerPid, montauk::abi::WinEvent* outEvent);
     int Enumerate(montauk::abi::WinInfo* outArray, int maxCount);
-    uint64_t Map(int windowId, int callerPid, uint64_t callerPml4, uint64_t& heapNext);
+    uint64_t Map(int windowId, int callerPid, uint64_t callerRoot, uint64_t& heapNext);
-    int Unmap(int windowId, int callerPid, uint64_t callerPml4);
+    int Unmap(int windowId, int callerPid, uint64_t callerRoot);
     int SendEvent(int windowId, const montauk::abi::WinEvent* event);
-    int Resize(int windowId, int callerPid, uint64_t ownerPml4, int newW, int newH,
+    int Resize(int windowId, int callerPid, uint64_t ownerRoot, int newW, int newH,
                uint64_t& heapNext, uint64_t& outVa);
     void CleanupProcess(int pid);
     int SetCursor(int windowId, int callerPid, int cursor);
@@ -44,4 +45,26 @@ namespace WinServer {
     int SetScale(int scale);
     int GetScale();
 
+    // ---- Messaging (see "Window server messaging" in Api/Syscall.hpp) ----
+
+    // Queue an app request about one of the caller's own windows.
+    int PostRequest(int windowId, int callerPid, const montauk::abi::WinRequest& request);
+    // Pop the oldest request (compositor only; the caller checks the cap).
+    // Returns 1 with *out filled, or 0 when the queue is empty.
+    int PollRequest(montauk::abi::WinRequest* out);
+
+    // Start a drag from the caller's window, or (compositor only, windowId
+    // -1) from the desktop surface itself. Takes ownership of `transfer` (it
+    // is emptied either way) and returns the new drag serial (> 0) or a
+    // negative StatusCodes value.
+    int DragBegin(int windowId, int callerPid, bool compositor,
+                  const montauk::abi::WinDragBegin& info, DataTransfer::Transfer& transfer);
+    // Read the drag's formats or data. Allowed for the source process, the
+    // compositor, and processes the compositor has sent a drag event for this
+    // serial to.
+    int DragFormats(uint32_t serial, int callerPid, bool compositor,
+                    montauk::abi::TransferFormatInfo* out, uint32_t max);
+    int DragGetData(uint32_t serial, int callerPid, bool compositor, const char* mime,
+                    void* buffer, uint32_t bufferLen, uint32_t* outLen);
+
 }

kernel/src/Api/Window.hpp

@@ -2,7 +2,9 @@
     * Window.hpp
     * SYS_WINCREATE, SYS_WINDESTROY, SYS_WINPRESENT, SYS_WINPOLL,
     * SYS_WINENUM, SYS_WINMAP, SYS_WINSENDEVENT, SYS_WINRESIZE,
-    * SYS_WINSETCURSOR, SYS_WINSETFLAGS, SYS_WINSETSCALE, SYS_WINGETSCALE syscalls
+    * SYS_WINSETCURSOR, SYS_WINSETFLAGS, SYS_WINSETSCALE, SYS_WINGETSCALE syscalls,
+    * and window server messaging: SYS_WIN_REQUEST, SYS_WIN_REQUEST_POLL,
+    * SYS_WIN_DRAG_BEGIN, SYS_WIN_DRAG_FORMATS, SYS_WIN_DRAG_GET_DATA
     * Copyright (c) 2026 Daniel Hammer
 */
 
@@ -15,16 +17,16 @@
 namespace montauk::abi {
 
     static int Sys_WinCreate(const char* title, int w, int h, WinCreateResult* result) {
-        if (result == nullptr || title == nullptr) return -1;
+        if (result == nullptr || title == nullptr) return InvalidArgument;
         auto* proc = Sched::GetCurrentProcessPtr();
-        if (proc == nullptr) return -1;
+        if (proc == nullptr) return GenericFailure;
 
         uint64_t outVa = 0;
-        int id = WinServer::Create(proc->pid, proc->pml4Phys, title, w, h,
+        int id = WinServer::Create(proc->pid, proc->pageTableRoot, title, w, h,
                                    proc->heapNext, outVa);
         result->id = id;
         result->pixelVa = (id >= 0) ? outVa : 0;
-        return id >= 0 ? 0 : -1;
+        return id >= 0 ? Success : id;
     }
 
     static int Sys_WinDestroy(int windowId) {
@@ -36,7 +38,7 @@ namespace montauk::abi {
     }
 
     static int Sys_WinPoll(int windowId, WinEvent* outEvent) {
-        if (outEvent == nullptr) return -1;
+        if (outEvent == nullptr) return InvalidArgument;
         return WinServer::Poll(windowId, Sched::GetCurrentPid(), outEvent);
     }
 
@@ -45,20 +47,26 @@ namespace montauk::abi {
         return WinServer::Enumerate(outArray, maxCount);
     }
 
+    // Mapping another process's window and injecting input into it are the
+    // compositor's job alone. Left open, any app could read every window's
+    // pixels or click on any window -- including a prompt asking the user
+    // whether to grant that same app authority.
     static uint64_t Sys_WinMap(int windowId) {
+        if (!Sched::HasCapability(CAP_COMPOSITOR)) return 0;
         auto* proc = Sched::GetCurrentProcessPtr();
         if (proc == nullptr) return 0;
-        return WinServer::Map(windowId, proc->pid, proc->pml4Phys, proc->heapNext);
+        return WinServer::Map(windowId, proc->pid, proc->pageTableRoot, proc->heapNext);
     }
 
     static int Sys_WinUnmap(int windowId) {
         auto* proc = Sched::GetCurrentProcessPtr();
-        if (proc == nullptr) return -1;
+        if (proc == nullptr) return GenericFailure;
-        return WinServer::Unmap(windowId, proc->pid, proc->pml4Phys);
+        return WinServer::Unmap(windowId, proc->pid, proc->pageTableRoot);
     }
 
     static int Sys_WinSendEvent(int windowId, const WinEvent* event) {
-        if (event == nullptr) return -1;
+        if (!Sched::HasCapability(CAP_COMPOSITOR)) return CapabilityDenied;
+        if (event == nullptr) return InvalidArgument;
         return WinServer::SendEvent(windowId, event);
     }
 
@@ -66,7 +74,7 @@ namespace montauk::abi {
         auto* proc = Sched::GetCurrentProcessPtr();
         if (proc == nullptr) return 0;
         uint64_t outVa = 0;
-        int r = WinServer::Resize(windowId, proc->pid, proc->pml4Phys, newW, newH,
+        int r = WinServer::Resize(windowId, proc->pid, proc->pageTableRoot, newW, newH,
                                   proc->heapNext, outVa);
         return (r == 0) ? outVa : 0;
     }
@@ -86,4 +94,41 @@ namespace montauk::abi {
     static int Sys_WinGetScale() {
         return WinServer::GetScale();
     }
+
+    // Any app may post requests, but only about its own windows (checked in
+    // WinServer); only the compositor may read them back.
+    static int Sys_WinRequest(int windowId, const WinRequest* request) {
+        if (request == nullptr) return InvalidArgument;
+        WinRequest copy = *request;
+        return WinServer::PostRequest(windowId, Sched::GetCurrentPid(), copy);
+    }
+
+    static int Sys_WinRequestPoll(WinRequest* out) {
+        if (!Sched::HasCapability(CAP_COMPOSITOR)) return CapabilityDenied;
+        if (out == nullptr) return InvalidArgument;
+        return WinServer::PollRequest(out);
+    }
+
+    static int Sys_WinDragBegin(int windowId, const WinDragBegin* info,
+                                const TransferItem* items, uint32_t count) {
+        if (info == nullptr) return InvalidArgument;
+        WinDragBegin infoCopy = *info;
+        DataTransfer::Transfer transfer;
+        int rc = DataTransfer::BuildFromUser(items, count, transfer);
+        if (rc != Success) return rc;
+        return WinServer::DragBegin(windowId, Sched::GetCurrentPid(),
+                                    Sched::HasCapability(CAP_COMPOSITOR), infoCopy, transfer);
+    }
+
+    static int Sys_WinDragFormats(uint32_t serial, TransferFormatInfo* out, uint32_t max) {
+        return WinServer::DragFormats(serial, Sched::GetCurrentPid(),
+                                      Sched::HasCapability(CAP_COMPOSITOR), out, max);
+    }
+
+    static int Sys_WinDragGetData(uint32_t serial, const char* mime, void* buffer,
+                                  uint32_t bufferLen, uint32_t* outLen) {
+        return WinServer::DragGetData(serial, Sched::GetCurrentPid(),
+                                      Sched::HasCapability(CAP_COMPOSITOR), mime,
+                                      buffer, bufferLen, outLen);
+    }
 };

kernel/src/Arch/Abi.hpp

@@ -0,0 +1,64 @@
+/*
+    * Abi.hpp
+    * Architecture-specific parts of the user-mode ABI
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+#include <Arch/Arch.hpp>
+
+// What the ELF loader and the process/thread setup code need to know about
+// the user-mode binary interface of the target. Defined inline by
+// Arch/<arch>/Abi.hpp.
+
+namespace Arch {
+
+    // ELF e_machine value accepted for user programs and libraries, and a
+    // name for diagnostics.
+    // constexpr uint16_t ElfMachine;
+    // constexpr const char* ElfMachineName;
+
+    // Calling-convention attribute for UEFI runtime services:
+    // #define ARCH_EFIAPI
+
+    // How the loader must treat a dynamic relocation type.
+    enum class RelocationKind {
+        None,         // no-op (type 0)
+        Relative,     // *target = loadBias + addend
+        Symbol,       // *target = symbol address + addend (64-bit absolute)
+        Unsupported,
+    };
+    inline RelocationKind ClassifyRelocation(uint32_t type);
+
+    // Geometry of one thread's TLS block, as offsets from its page-aligned
+    // base. The variant (I: TLS above the thread pointer, II: below it) is
+    // fixed by the architecture's ELF TLS ABI.
+    struct TlsLayout {
+        uint64_t templateOffset;  // where .tdata/.tbss start
+        uint64_t tpOffset;        // thread pointer; [TP] holds the self-pointer
+        uint64_t totalSize;       // bytes to map, incl. the TCB
+    };
+    inline TlsLayout ComputeTlsLayout(uint64_t memSize, uint64_t pAlign);
+
+    // Emit the user-mode code that calls SYS_EXIT(0). A process's entry
+    // function returns into it.
+    inline void WriteExitStub(uint8_t* page);
+
+    // Prepare the main thread's initial user stack so that returning from
+    // the entry function lands at `returnAddress`. `topPage` is the kernel
+    // mapping of the page just below `stackTop`. Returns the initial user SP.
+    inline uint64_t SetupUserEntryStack(uint8_t* topPage, uint64_t stackTop,
+                                        uint64_t returnAddress);
+
+    // Initial user SP for a sibling thread whose stack ends at `stackTop`
+    // (the thread trampoline never returns, so no return address is set).
+    inline uint64_t ThreadEntryStackPointer(uint64_t stackTop);
+
+}
+
+#if MONTAUK_ARCH_X86_64
+#include <Arch/x86_64/Abi.hpp>
+#elif MONTAUK_ARCH_AARCH64
+#include <Arch/aarch64/Abi.hpp>
+#endif

kernel/src/Arch/Arch.hpp

@@ -0,0 +1,25 @@
+/*
+    * Arch.hpp
+    * Target architecture selection
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+
+// Everything outside src/Arch/ is architecture-neutral and talks to the CPU
+// only through the interfaces in src/Arch/*.hpp. Each supported architecture
+// implements them under src/Arch/<arch>/; the build compiles only the
+// directory matching ARCH. See docs/porting.md.
+//
+// Interface headers that need inline (per-arch) definitions include the
+// matching implementation header at their end, selected below.
+
+#if defined(__x86_64__)
+#define MONTAUK_ARCH_X86_64 1
+#define MONTAUK_ARCH_NAME "x86_64"
+#elif defined(__aarch64__)
+#define MONTAUK_ARCH_AARCH64 1
+#define MONTAUK_ARCH_NAME "aarch64"
+#else
+#error "MontaukOS: unsupported target architecture (add one under src/Arch/)"
+#endif

kernel/src/Arch/Context.hpp

@@ -0,0 +1,49 @@
+/*
+    * Context.hpp
+    * Architecture-neutral thread context interface
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+#include <Arch/Arch.hpp>
+#include <Arch/Smp.hpp>
+
+// The scheduler's view of a thread's CPU context: a kernel stack pointer
+// saved by SwitchContext, an FPU/SIMD save area, and a page-table root.
+
+#if MONTAUK_ARCH_X86_64
+#include <Arch/x86_64/Context.hpp>
+#elif MONTAUK_ARCH_AARCH64
+#include <Arch/aarch64/Context.hpp>
+#endif
+
+// The architecture provides (in Arch/<arch>/Context.hpp):
+//   constexpr uint64_t Arch::FpuStateSize, Arch::FpuStateAlign
+
+namespace Arch {
+
+    // Reset an FPU/SIMD save area to the state a new thread starts with.
+    void InitFpuState(uint8_t* area);
+
+    // Build the initial kernel stack of a new thread so that the first
+    // SwitchContext to it "returns" into `entry`. Returns the saved SP.
+    uint64_t PrepareThreadStack(uint64_t kernelStackTop, void (*entry)());
+
+    // Save the running context (stack pointer to *oldSp, FPU state to
+    // oldFpu when non-null), switch to page-table root `newRoot` if it is
+    // not already active, and resume the context saved at newSp (restoring
+    // newFpu when non-null). Returns when the old context is resumed.
+    void SwitchContext(uint64_t* oldSp, uint64_t newSp, uint64_t newRoot,
+                       uint8_t* oldFpu, uint8_t* newFpu);
+
+    // Point this CPU at the thread about to run: the kernel stack used when
+    // it enters the kernel from user mode, and its TLS thread pointer.
+    void ActivateThread(Smp::CpuData* cpu, uint64_t kernelStackTop,
+                        uint64_t threadPointer);
+
+    // Drop to user mode at `entry` with stack `userSp`; `arg` is the first
+    // argument register. Interrupts are enabled in user mode.
+    [[noreturn]] void EnterUserMode(uint64_t entry, uint64_t userSp, uint64_t arg);
+
+}

kernel/src/Arch/Cpu.hpp

@@ -0,0 +1,121 @@
+/*
+    * Cpu.hpp
+    * Architecture-neutral CPU primitives
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+#include <Arch/Arch.hpp>
+
+// Every function here is defined inline by the architecture
+// (Arch/<arch>/Cpu.hpp), so the primitives cost what the raw instruction
+// costs. Generic kernel code must use these instead of inline assembly.
+
+namespace Arch {
+
+    // Saved interrupt-enable state, as returned by SaveAndDisableInterrupts.
+    // Opaque to generic code: only hand it back to RestoreInterrupts.
+    using IrqFlags = uint64_t;
+
+    // ---- Spinning and barriers ----
+
+    // Spin-loop hint (x86 PAUSE, ARM YIELD).
+    inline void Pause();
+
+    // Stop the compiler from caching memory across this point. Emits no
+    // instruction; use it in polling loops on device-written memory.
+    inline void CompilerBarrier() {
+        asm volatile("" ::: "memory");
+    }
+
+    // Order all earlier stores before all later stores, including stores to
+    // write-combining memory (framebuffers) that must become visible.
+    inline void StoreFence();
+
+    // Full read/write memory barrier.
+    inline void MemoryFence();
+
+    // Order this CPU's earlier writes to DMA memory (descriptors, buffers)
+    // before a later write the device acts on: the descriptor's valid/cycle
+    // bit, or a doorbell register. x86 keeps stores in order already.
+    inline void DmaWriteFence();
+
+    // Order the read of a device-written valid/cycle/status field before the
+    // reads of the data it guards.
+    inline void DmaReadFence();
+
+    // ---- Interrupt state ----
+
+    inline void EnableInterrupts();
+    inline void DisableInterrupts();
+    inline bool InterruptsEnabled();
+
+    // Disable interrupts and return the previous state.
+    inline IrqFlags SaveAndDisableInterrupts();
+
+    // Restore an interrupt state returned by SaveAndDisableInterrupts.
+    inline void RestoreInterrupts(IrqFlags flags);
+
+    // ---- Halting ----
+
+    // Wait for the next interrupt. Interrupt state is left as it was; with
+    // interrupts disabled this only returns on an NMI-class event.
+    inline void Halt();
+
+    // Called with interrupts disabled: atomically enable them and wait for
+    // an interrupt, then disable them again before returning. The atomicity
+    // closes the race where a wakeup IRQ lands between the readiness check
+    // and the wait, leaving the CPU asleep until the next timer deadline.
+    inline void HaltWithInterruptsDisabled();
+
+    // Disable interrupts and stop this CPU for good.
+    [[noreturn]] inline void HaltForever();
+
+    // ---- Time and delays ----
+
+    // A free-running, monotonically increasing cycle counter (x86 TSC, ARM
+    // CNTVCT). Its rate is architecture-defined; use it for entropy and
+    // relative timing, not wall-clock time.
+    inline uint64_t ReadTimestamp();
+
+    // Delay for at least about one microsecond. Works with interrupts
+    // disabled and before any clock is calibrated; errs long.
+    inline void IoDelay();
+
+    // ---- Caches and translation ----
+
+    // Write back and invalidate all CPU caches (before power state changes).
+    inline void WritebackInvalidateCaches();
+
+    // Cache maintenance for devices whose DMA does not snoop the CPU caches
+    // (ACPI _CCA = 0). Clean: write dirty lines back so the device reads
+    // what the CPU wrote. Invalidate: write back and discard, so the CPU
+    // then reads what the device wrote. No-ops where DMA is coherent (x86).
+    inline void CleanDataCache(const void* address, uint64_t size);
+    inline void InvalidateDataCache(const void* address, uint64_t size);
+
+    // Make instructions just written to [address, address+size) (any
+    // mapping of them) visible to instruction fetch on every CPU. A no-op
+    // where instruction fetch snoops data writes (x86).
+    inline void SyncInstructionCache(const void* address, uint64_t size);
+
+    // Invalidate this CPU's translation for one virtual page.
+    inline void FlushTlbPage(uint64_t virtualAddress);
+
+    // Invalidate this CPU's non-global translations.
+    inline void FlushTlbAll();
+
+    // Physical address of the active page-table root (x86 CR3, ARM TTBR0).
+    inline uint64_t ReadPageTableRoot();
+
+    // Switch to the page-table root at the given physical address.
+    inline void WritePageTableRoot(uint64_t rootPhys);
+
+}
+
+#if MONTAUK_ARCH_X86_64
+#include <Arch/x86_64/Cpu.hpp>
+#elif MONTAUK_ARCH_AARCH64
+#include <Arch/aarch64/Cpu.hpp>
+#endif

kernel/src/Arch/CpuPower.hpp

@@ -1,14 +1,18 @@
 /*
     * CpuPower.hpp
-    * Intel CPU power management: HWP frequency scaling, C1E promotion,
+    * Architecture-neutral CPU power/thermal management interface
-    * and the package thermal governor.
     * Copyright (c) 2026 Daniel Hammer
 */
 
 #pragma once
 #include <cstdint>
 
-namespace Hal {
+// Frequency scaling, idle-state policy and the thermal governor are CPU
+// specific (x86_64: Intel HWP, C1E promotion and a package-temperature
+// governor driven through MSRs). An architecture without them implements
+// these as no-ops and returns false from GetSnapshot.
+
+namespace Arch {
     namespace CpuPower {
 
         // Snapshot of the power/thermal state for SYS_POWERINFO and logging.
@@ -24,6 +28,7 @@ namespace Hal {
             uint32_t BaseMHz;        // CPUID base frequency (0 = unknown)
             uint32_t MaxMHz;         // CPUID max turbo frequency
             uint32_t EffMHz;         // measured average active frequency (BSP)
+            uint8_t  ApIdleHint;     // idle state secondary CPUs enter (arch-defined)
         };
 
         // BSP: detect features, enable HWP with a balanced energy preference,

kernel/src/Arch/Interrupts.hpp

@@ -0,0 +1,81 @@
+/*
+    * Interrupts.hpp
+    * Architecture-neutral hardware interrupt interface
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+
+// Devices and kernel services see interrupts as numbered IRQ slots. The
+// architecture maps each slot onto its interrupt controller (x86: an IDT
+// vector fed by the I/O APIC, MSIs and local APIC IPIs; ARM: a GIC INTID)
+// and calls the registered handler.
+
+namespace Arch {
+
+    // IRQ handler function type. The parameter is the IRQ slot number.
+    // `fromUser` is derived directly from the interrupted context. Do not
+    // cache it in per-CPU data: an IRQ handler may context-switch and later
+    // resume on another CPU, making per-CPU saved-origin state stale.
+    using IrqHandler = void(*)(uint8_t irq, bool fromUser);
+
+    // Number of IRQ slots: 0-23 platform interrupt lines, 24-45 message-
+    // signalled interrupts (one slot per MSI-capable driver), 46-47 IPIs.
+    constexpr int IRQ_COUNT = 48;
+
+    // The architecture's periodic scheduler/timekeeping timer.
+    constexpr uint8_t IRQ_TIMER    = 0;
+
+    // Platform interrupt lines with PC/ISA meaning. Only drivers for PC
+    // devices (PS/2, legacy PCI INTx through the I/O APIC) use these.
+    constexpr uint8_t IRQ_KEYBOARD = 1;
+    constexpr uint8_t IRQ_CASCADE  = 2;
+    constexpr uint8_t IRQ_COM2     = 3;
+    constexpr uint8_t IRQ_COM1     = 4;
+    constexpr uint8_t IRQ_FLOPPY   = 6;
+    constexpr uint8_t IRQ_RTC      = 8;
+    constexpr uint8_t IRQ_MOUSE    = 12;
+    constexpr uint8_t IRQ_ATA1     = 14;
+    constexpr uint8_t IRQ_ATA2     = 15;
+
+    // Inter-processor interrupts.
+    constexpr uint8_t IRQ_TLB_SHOOTDOWN = 46;
+    constexpr uint8_t IRQ_RESCHEDULE    = 47;
+
+    // Register the handler for an IRQ slot.
+    void RegisterIrqHandler(uint8_t irq, IrqHandler handler);
+
+    // Route platform interrupt line `irq` to IRQ slot `irq` on the boot CPU,
+    // applying the firmware's polarity/trigger overrides, and unmask it.
+    void RouteLegacyIrq(uint8_t irq);
+
+    // Unmask platform interrupt line `irq` (already routed at boot).
+    void EnableLegacyIrq(uint8_t irq);
+
+    // Whether the platform interrupt lines are wired by the hardware (x86:
+    // ISA IRQs and I/O APIC pins, as the firmware programs them into PCI
+    // Interrupt Line registers). Where they are not (ARM), the PCI layer
+    // assigns lines to PCI INTx pins from ACPI _PRT with AssignPlatformLine.
+    bool PlatformLinesAreFixed();
+
+    // Make platform line `irq` (1-23) deliver global system interrupt `gsi`
+    // (ACPI numbering) with the given trigger mode and polarity, so that
+    // EnableLegacyIrq(irq) unmasks it. Only where !PlatformLinesAreFixed().
+    bool AssignPlatformLine(uint8_t irq, uint32_t gsi, bool levelTriggered, bool activeLow);
+
+    // A message-signalled interrupt targeting IRQ slot `irq` on the boot
+    // CPU: the device writes `data` to `address` to raise it. Program these
+    // into a PCI MSI capability or MSI-X table entry. Returns false when the
+    // platform cannot receive MSIs (drivers then fall back to INTx/polling).
+    struct MsiMessage {
+        uint64_t address;
+        uint32_t data;
+    };
+    bool ComposeMsi(uint8_t irq, MsiMessage& out);
+
+    // Raise IPI slot `irq` (IRQ_RESCHEDULE / IRQ_TLB_SHOOTDOWN) on the CPU
+    // with the given Smp index. Returns false if delivery did not complete.
+    bool SendIpi(int cpuIndex, uint8_t irq);
+
+}

kernel/src/Arch/Platform.hpp

@@ -0,0 +1,58 @@
+/*
+    * Platform.hpp
+    * Architecture hooks for kernel bring-up and power-off paths
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+#include <ACPI/ACPI.hpp>
+
+// kmain (Boot/Main.cpp) owns the boot sequence and calls into the
+// architecture at these points, in this order.
+
+namespace Arch {
+
+    // First thing after the boot console exists: CPU descriptor tables,
+    // FPU/SIMD enablement. No memory allocation is available yet.
+    void EarlyInitialize();
+
+    // The frame allocator and heap exist: install exception/trap vectors.
+    void InitializeTraps();
+
+    // The kernel page tables are live: program memory-type attributes that
+    // MapWC/MapMMIO rely on (x86 PAT).
+    void InitializeMemoryAttributes();
+
+    // ACPI tables are available: bring up interrupt controllers, the boot
+    // CPU's per-CPU data (Smp::InitBsp), idle-state and power management.
+    // Interrupts stay disabled; kmain enables them afterwards.
+    void InitializeInterrupts(Hal::ACPI::CommonSDTHeader* xsdt);
+
+    // Platform devices that are not discovered on PCI (x86: the PS/2
+    // controller, keyboard and mouse). Called once the timer is running.
+    void InitializePlatformDevices();
+
+    // Right before the first user process is created: anything user-mode
+    // entry needs (x86: load the task register for ring-3 interrupt stacks).
+    void PrepareUserMode();
+
+    // Secondary CPUs are up and the first process is about to be spawned.
+    void LateInitialize();
+
+    // Report devices that belong to the platform rather than a bus (for
+    // SYS_DEVLIST): interrupt controllers (category 1), timers (2) and
+    // platform input devices (3).
+    using PlatformDeviceSink = void (*)(void* context, uint8_t category,
+                                        const char* name, const char* detail);
+    void ListPlatformDevices(PlatformDeviceSink sink, void* context);
+
+    // Reset the machine without firmware help (last resort after EFI
+    // ResetSystem). Never returns.
+    [[noreturn]] void Reboot();
+
+    // Switch the machine off where ACPI S5 does not apply (hardware-reduced
+    // ACPI: ARM's PSCI SYSTEM_OFF). Returns if the platform cannot.
+    void PowerOff();
+
+}

kernel/src/Arch/Smp.hpp

@@ -0,0 +1,94 @@
+/*
+    * Smp.hpp
+    * Architecture-neutral per-CPU data and SMP bring-up interface
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+#include <Arch/Arch.hpp>
+#include <Boot/BootInfo.hpp>
+
+#if MONTAUK_ARCH_X86_64
+#include <Arch/x86_64/PerCpu.hpp>
+#elif MONTAUK_ARCH_AARCH64
+#include <Arch/aarch64/PerCpu.hpp>
+#endif
+
+// The architecture provides (in Arch/<arch>/PerCpu.hpp):
+//   struct Arch::CpuEntryState  -- fields kernel-entry assembly reads at a
+//                                  fixed offset (8) from the per-CPU base
+//   struct Arch::CpuState       -- everything else the port keeps per CPU
+//   void* Arch::CurrentCpuLocal()     -- this CPU's CpuData (fast path)
+//   void* Arch::TryCurrentCpuLocal()  -- same, or nullptr before InitBsp
+
+namespace Smp {
+
+    static constexpr int MaxCPUs = 64;
+
+    struct CpuData {
+        // === Fixed layout: read by low-level entry code ===
+        uint64_t selfPtr;            // offset 0: pointer to self
+        Arch::CpuEntryState entry;   // offset 8: arch kernel-entry fields
+
+        // === Generic fields ===
+        int32_t  currentSlot;        // process table slot, -1 = idle
+        int cpuIndex;                // 0-based CPU index (0 = boot CPU)
+        uint64_t idleSavedSp;        // idle context's stack pointer while a process runs
+        volatile bool started;       // set by an AP once its init is complete
+
+        // When set, the scheduler tick must NOT pull this (idle) CPU away to
+        // run a ready process: its idle context is busy with long-running
+        // kernel work whose wall-clock-bounded waits would spuriously expire
+        // if the context only ran when nothing else is ready (observed: the
+        // deferred Bluetooth firmware download racing login-screen startup).
+        // Other CPUs schedule normally while this is held.
+        volatile bool reservedForKernelWork = false;
+
+        // Set by the BSP timer when an IRQ bottom half needs process-safe
+        // execution even though runnable user processes keep this CPU busy.
+        // Schedule() consumes it by yielding once to the idle kernel context.
+        bool kernelWorkPending = false;
+
+        // === Architecture-private state ===
+        Arch::CpuState arch;
+    };
+
+    static_assert(__builtin_offsetof(CpuData, selfPtr) == 0,
+                  "CpuData: per-CPU base must point at selfPtr");
+    static_assert(__builtin_offsetof(CpuData, entry) == 8,
+                  "CpuData: entry-assembly fields must start at offset 8");
+
+    // The current CPU's data.
+    inline CpuData* GetCurrentCpuData() {
+        return (CpuData*)Arch::CurrentCpuLocal();
+    }
+
+    // Exception paths may run before InitBsp() has installed the per-CPU
+    // base. This variant returns nullptr in that state instead of faulting
+    // recursively on a bogus base.
+    inline CpuData* TryGetCurrentCpuData() {
+        return (CpuData*)Arch::TryCurrentCpuLocal();
+    }
+
+    // Get CPU data by index
+    CpuData* GetCpuData(int index);
+
+    // Number of online CPUs
+    int GetCpuCount();
+
+    // Number of CPUs the firmware describes (available once interrupts are
+    // initialized; may exceed GetCpuCount if some failed to start).
+    // Implemented by the architecture.
+    int GetDetectedCpuCount();
+
+    // Initialize the boot CPU's per-CPU data (call before interrupts are
+    // enabled). Implemented by the architecture.
+    void InitBsp();
+
+    // Boot all secondary CPUs (call after all subsystems are ready).
+    // `smp` is the boot contract's SMP block; if FeatureSmp was not
+    // advertised the caller may still pass it (cpuCount<=1 is a no-op).
+    // Implemented by the architecture.
+    void BootAPs(const montauk::boot::SmpInfo& smp);
+}

kernel/src/Arch/Syscall.hpp

@@ -0,0 +1,38 @@
+/*
+    * Syscall.hpp
+    * Architecture-neutral system-call entry interface
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+#include <Arch/Arch.hpp>
+
+#if MONTAUK_ARCH_X86_64
+#include <Arch/x86_64/SyscallFrame.hpp>
+#elif MONTAUK_ARCH_AARCH64
+#include <Arch/aarch64/SyscallFrame.hpp>
+#endif
+
+// The architecture's system-call entry code saves the user context into an
+// Arch::SyscallFrame on the thread's kernel stack (Arch/<arch>/
+// SyscallFrame.hpp defines it, exposing at least `syscall_nr` and `arg1` ..
+// `arg6`), enables interrupts, and calls
+//
+//     extern "C" int64_t SyscallDispatch(Arch::SyscallFrame* frame);
+//
+// whose result it returns to user mode. Before returning it calls
+//
+//     extern "C" void SchedExitIfKilled();
+//
+// so a kill that arrived during the syscall takes effect at a point where
+// the syscall holds no locks.
+
+namespace Arch {
+
+    // Install the system-call entry point on the calling CPU. Called on the
+    // boot CPU by montauk::abi::InitializeSyscalls and by every secondary
+    // CPU during its bring-up.
+    void InitializeSyscallEntry();
+
+}

kernel/src/Arch/Timer.hpp

@@ -0,0 +1,32 @@
+/*
+    * Timer.hpp
+    * Architecture-neutral per-CPU tick timer interface
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+
+// The hardware behind Timekeeping (Timekeeping/Timer.hpp): a per-CPU timer
+// that raises Arch::IRQ_TIMER periodically. Timekeeping registers the IRQ
+// handler and owns all tick accounting; the architecture only calibrates
+// and programs the hardware (x86: local APIC timer calibrated against the
+// PIT; ARM: the generic timer).
+
+namespace Arch::Timer {
+
+    // Calibrate the timer and start the calling (boot) CPU's periodic tick
+    // with the given period. Returns false if no usable timer was found.
+    bool InitializeBsp(uint32_t periodMs);
+
+    // (Re)start the calling CPU's periodic tick, reusing the boot CPU's
+    // calibration: secondary CPU bring-up, leaving idle, and the boot CPU
+    // after S3 resume (timer state does not survive suspend). Only called
+    // after InitializeBsp succeeded.
+    void StartPeriodic(uint32_t periodMs);
+
+    // Mask the calling CPU's tick (idle secondary CPUs rely on IPIs).
+    // Cheap when it is already masked.
+    void Stop();
+
+}

kernel/src/Arch/aarch64/Abi.hpp

@@ -0,0 +1,55 @@
+/*
+    * Abi.hpp
+    * aarch64 user-mode ABI details (implements Arch/Abi.hpp)
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+
+// UEFI on AArch64 uses the standard AAPCS64 calling convention.
+#define ARCH_EFIAPI
+
+namespace Arch {
+
+    constexpr uint16_t ElfMachine = 183;  // EM_AARCH64
+    constexpr const char* ElfMachineName = "aarch64";
+
+    inline RelocationKind ClassifyRelocation(uint32_t type) {
+        switch (type) {
+            case 0:     return RelocationKind::None;
+            case 1027:  return RelocationKind::Relative;  // R_AARCH64_RELATIVE
+            case 257:                                     // R_AARCH64_ABS64
+            case 1025:                                    // R_AARCH64_GLOB_DAT
+            case 1026:  return RelocationKind::Symbol;    // R_AARCH64_JUMP_SLOT
+            default:    return RelocationKind::Unsupported;
+        }
+    }
+
+    // Variant I: TP points at a 16-byte TCB and the TLS block follows it,
+    // starting at the next p_align boundary.
+    inline TlsLayout ComputeTlsLayout(uint64_t memSize, uint64_t pAlign) {
+        uint64_t align = pAlign ? pAlign : 1;
+        uint64_t templateOffset = (16 + align - 1) & ~(align - 1);
+        return {templateOffset, 0, templateOffset + memSize};
+    }
+
+    // System calls: number in x8, arguments in x0-x5, `svc #0`.
+    inline void WriteExitStub(uint8_t* page) {
+        uint32_t* code = (uint32_t*)page;
+        code[0] = 0xD2800000;  // mov x0, #0  (exit code 0)
+        code[1] = 0xD2800008;  // mov x8, #0  (SYS_EXIT = 0)
+        code[2] = 0xD4000001;  // svc #0
+    }
+
+    // The return address travels in x30, not on the stack: EnterUserMode
+    // points x30 at the exit stub. The stack only needs 16-byte alignment.
+    inline uint64_t SetupUserEntryStack(uint8_t*, uint64_t stackTop, uint64_t) {
+        return stackTop & ~0xFULL;
+    }
+
+    inline uint64_t ThreadEntryStackPointer(uint64_t stackTop) {
+        return stackTop & ~0xFULL;
+    }
+
+}

kernel/src/Arch/aarch64/AcpiSleep.cpp

@@ -0,0 +1,26 @@
+/*
+    * AcpiSleep.cpp
+    * aarch64: no ACPI S3 (implements ACPI/AcpiSleep.hpp)
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#include <ACPI/AcpiSleep.hpp>
+#include <Common/StatusCodes.hpp>
+
+// Suspend on ARM goes through PSCI (SYSTEM_SUSPEND), not the FACS waking
+// vector; until that exists, S3 is reported as unavailable.
+
+namespace Hal::AcpiSleep {
+
+    void Initialize(ACPI::CommonSDTHeader*) {
+    }
+
+    bool IsS3Available() {
+        return false;
+    }
+
+    int Suspend() {
+        return -1;
+    }
+
+}

kernel/src/Arch/aarch64/Bringup.hpp

@@ -0,0 +1,22 @@
+/*
+    * Bringup.hpp
+    * aarch64 per-CPU bring-up shared by the boot CPU and secondaries
+    * (private to the port)
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+
+namespace Arch {
+
+    // System registers every CPU programs for itself: SP_EL1 as the stack,
+    // the exception vectors, SCTLR_EL1 policy bits, FP/SIMD access.
+    // Platform.cpp.
+    void InitializeCpuRegisters();
+
+    // Load the kernel's translation regime (MAIR, TCR, TTBR1 kernel half,
+    // TTBR0 kernel-context table) on the calling CPU. Paging::Init builds
+    // the tables and calls it on the boot CPU. Paging.cpp.
+    void ActivateKernelTranslation();
+
+}

kernel/src/Arch/aarch64/Context.S

@@ -0,0 +1,144 @@
+/*
+    Context.S
+    aarch64 context switch and first entry to EL0 (see Arch/Context.hpp)
+    Copyright (c) 2026 Daniel Hammer
+*/
+
+    /* The kernel is built -mgeneral-regs-only; this file alone touches
+       the FP/SIMD registers, to save and restore user state. */
+    .arch_extension fp
+    .arch_extension simd
+
+    .text
+
+/*
+ * void ArchContextSwitch(uint64_t* oldSp, uint64_t newSp, uint64_t newRoot,
+ *                        uint8_t* oldFpu, uint8_t* newFpu)
+ *   x0 = where to save the old SP     x1 = SP to resume
+ *   x2 = TTBR0 root for the new side  x3/x4 = FP/SIMD save areas (may be 0)
+ *
+ * Frame on the kernel stack (Context.cpp's PrepareThreadStack builds the
+ * same shape): x19..x28, x29, x30 -- 96 bytes. The kernel is built with
+ * -mgeneral-regs-only, so q0-q31/FPCR/FPSR only ever hold user state.
+ */
+    .global ArchContextSwitch
+    .type ArchContextSwitch, %function
+ArchContextSwitch:
+    cbz     x3, 1f
+    stp     q0, q1, [x3, #0]
+    stp     q2, q3, [x3, #32]
+    stp     q4, q5, [x3, #64]
+    stp     q6, q7, [x3, #96]
+    stp     q8, q9, [x3, #128]
+    stp     q10, q11, [x3, #160]
+    stp     q12, q13, [x3, #192]
+    stp     q14, q15, [x3, #224]
+    stp     q16, q17, [x3, #256]
+    stp     q18, q19, [x3, #288]
+    stp     q20, q21, [x3, #320]
+    stp     q22, q23, [x3, #352]
+    stp     q24, q25, [x3, #384]
+    stp     q26, q27, [x3, #416]
+    stp     q28, q29, [x3, #448]
+    stp     q30, q31, [x3, #480]
+    mrs     x9, fpcr
+    mrs     x10, fpsr
+    add     x11, x3, #512
+    stp     x9, x10, [x11]
+1:
+    sub     sp, sp, #96
+    stp     x19, x20, [sp, #0]
+    stp     x21, x22, [sp, #16]
+    stp     x23, x24, [sp, #32]
+    stp     x25, x26, [sp, #48]
+    stp     x27, x28, [sp, #64]
+    stp     x29, x30, [sp, #80]
+    mov     x9, sp
+    str     x9, [x0]
+
+    mov     sp, x1
+
+    /* Switch the process half only when it differs (no ASIDs yet, so a
+       switch invalidates this CPU's TLB). */
+    mrs     x9, ttbr0_el1
+    and     x9, x9, #0x0000fffffffffffe
+    cmp     x9, x2
+    b.eq    2f
+    dsb     ishst
+    msr     ttbr0_el1, x2
+    isb
+    tlbi    vmalle1
+    dsb     nsh
+    isb
+2:
+    ldp     x19, x20, [sp, #0]
+    ldp     x21, x22, [sp, #16]
+    ldp     x23, x24, [sp, #32]
+    ldp     x25, x26, [sp, #48]
+    ldp     x27, x28, [sp, #64]
+    ldp     x29, x30, [sp, #80]
+    add     sp, sp, #96
+
+    cbz     x4, 3f
+    ldp     q0, q1, [x4, #0]
+    ldp     q2, q3, [x4, #32]
+    ldp     q4, q5, [x4, #64]
+    ldp     q6, q7, [x4, #96]
+    ldp     q8, q9, [x4, #128]
+    ldp     q10, q11, [x4, #160]
+    ldp     q12, q13, [x4, #192]
+    ldp     q14, q15, [x4, #224]
+    ldp     q16, q17, [x4, #256]
+    ldp     q18, q19, [x4, #288]
+    ldp     q20, q21, [x4, #320]
+    ldp     q22, q23, [x4, #352]
+    ldp     q24, q25, [x4, #384]
+    ldp     q26, q27, [x4, #416]
+    ldp     q28, q29, [x4, #448]
+    ldp     q30, q31, [x4, #480]
+    add     x11, x4, #512
+    ldp     x9, x10, [x11]
+    msr     fpcr, x9
+    msr     fpsr, x10
+3:
+    ret
+    .size ArchContextSwitch, . - ArchContextSwitch
+
+/*
+ * void ArchEnterUserMode(uint64_t entry, uint64_t userSp, uint64_t arg,
+ *                        uint64_t returnAddress, uint64_t kernelStackTop)
+ *
+ * Drops to EL0t with interrupts unmasked. The kernel stack is reset to its
+ * top: SP_EL1 is what the next exception from EL0 runs on.
+ */
+    .global ArchEnterUserMode
+    .type ArchEnterUserMode, %function
+ArchEnterUserMode:
+    msr     daifset, #0xf
+    mov     sp, x4
+    msr     elr_el1, x0
+    msr     sp_el0, x1
+    msr     spsr_el1, xzr          /* EL0t, DAIF clear */
+    mov     x0, x2
+    mov     x30, x3
+    mov     x1, xzr
+    mov     x2, xzr
+    mov     x3, xzr
+    mov     x4, xzr
+    mov     x5, xzr
+    mov     x6, xzr
+    mov     x7, xzr
+    mov     x8, xzr
+    mov     x9, xzr
+    mov     x10, xzr
+    mov     x11, xzr
+    mov     x12, xzr
+    mov     x13, xzr
+    mov     x14, xzr
+    mov     x15, xzr
+    mov     x16, xzr
+    mov     x17, xzr
+    mov     x18, xzr
+    mov     x29, xzr
+    eret
+    .size ArchEnterUserMode, . - ArchEnterUserMode

kernel/src/Arch/aarch64/Context.cpp

@@ -0,0 +1,53 @@
+/*
+    * Context.cpp
+    * aarch64 thread context (implements Arch/Context.hpp)
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#include <Arch/Context.hpp>
+#include <Libraries/Memory.hpp>
+#include <Sched/Scheduler.hpp>
+
+// Context.S
+extern "C" void ArchContextSwitch(uint64_t* oldSp, uint64_t newSp, uint64_t newRoot,
+                                  uint8_t* oldFpu, uint8_t* newFpu);
+extern "C" [[noreturn]] void ArchEnterUserMode(uint64_t entry, uint64_t userSp, uint64_t arg,
+                                               uint64_t returnAddress, uint64_t kernelStackTop);
+
+namespace Arch {
+
+    void InitFpuState(uint8_t* area) {
+        // All-zero is the architectural reset state: round-to-nearest, no
+        // exceptions trapped, no flags raised.
+        memset(area, 0, FpuStateSize);
+    }
+
+    uint64_t PrepareThreadStack(uint64_t kernelStackTop, void (*entry)()) {
+        // The frame Context.S pops: x19..x28, x29, x30 (= return address).
+        uint64_t* sp = (uint64_t*)kernelStackTop - 12;
+        for (int i = 0; i < 11; i++) {
+            sp[i] = 0;
+        }
+        sp[11] = (uint64_t)entry;
+        return (uint64_t)sp;
+    }
+
+    void SwitchContext(uint64_t* oldSp, uint64_t newSp, uint64_t newRoot,
+                       uint8_t* oldFpu, uint8_t* newFpu) {
+        ArchContextSwitch(oldSp, newSp, newRoot, oldFpu, newFpu);
+    }
+
+    void ActivateThread(Smp::CpuData* cpu, uint64_t kernelStackTop,
+                        uint64_t threadPointer) {
+        cpu->entry.kernelStackTop = kernelStackTop;
+        asm volatile("msr tpidr_el0, %0" :: "r"(threadPointer));
+    }
+
+    void EnterUserMode(uint64_t entry, uint64_t userSp, uint64_t arg) {
+        // Returning from the entry function lands in the exit stub (see
+        // SetupUserEntryStack in Abi.hpp).
+        ArchEnterUserMode(entry, userSp, arg, Sched::ExitStubAddr,
+                          Smp::GetCurrentCpuData()->entry.kernelStackTop);
+    }
+
+}

kernel/src/Arch/aarch64/Context.hpp

@@ -0,0 +1,14 @@
+/*
+    * Context.hpp
+    * aarch64 thread context constants (see Arch/Context.hpp)
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+
+namespace Arch {
+    // q0-q31 followed by FPCR and FPSR (Context.S).
+    constexpr uint64_t FpuStateSize = 32 * 16 + 16;
+    constexpr uint64_t FpuStateAlign = 16;
+}

kernel/src/Arch/aarch64/Cpu.hpp

@@ -0,0 +1,170 @@
+/*
+    * Cpu.hpp
+    * aarch64 implementation of the Arch/Cpu.hpp primitives
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+
+namespace Arch {
+
+    // PSTATE.I in the DAIF view
+    static constexpr uint64_t DaifIrqMask = 1ULL << 7;
+
+    inline void Pause() {
+        asm volatile("yield" ::: "memory");
+    }
+
+    // Device-nGnRE / Normal-NC (write-combining) stores complete in order
+    // once a DSB has been issued.
+    inline void StoreFence() {
+        asm volatile("dsb st" ::: "memory");
+    }
+
+    inline void MemoryFence() {
+        asm volatile("dsb sy" ::: "memory");
+    }
+
+    // Devices are outer-shareable observers.
+    inline void DmaWriteFence() {
+        asm volatile("dmb oshst" ::: "memory");
+    }
+
+    inline void DmaReadFence() {
+        asm volatile("dmb oshld" ::: "memory");
+    }
+
+    inline void EnableInterrupts() {
+        asm volatile("msr daifclr, #2" ::: "memory");
+    }
+
+    inline void DisableInterrupts() {
+        asm volatile("msr daifset, #2" ::: "memory");
+    }
+
+    inline bool InterruptsEnabled() {
+        uint64_t daif;
+        asm volatile("mrs %0, daif" : "=r"(daif));
+        return (daif & DaifIrqMask) == 0;
+    }
+
+    inline IrqFlags SaveAndDisableInterrupts() {
+        uint64_t daif;
+        asm volatile("mrs %0, daif\n\tmsr daifset, #2" : "=r"(daif) :: "memory");
+        return daif;
+    }
+
+    inline void RestoreInterrupts(IrqFlags flags) {
+        asm volatile("msr daif, %0" :: "r"(flags) : "memory");
+    }
+
+    inline void Halt() {
+        asm volatile("wfi" ::: "memory");
+    }
+
+    // WFI wakes on a pending interrupt even while PSTATE.I masks it, so
+    // checking for work and then waiting with IRQs masked cannot lose a
+    // wakeup. Unmask briefly afterwards so the pending IRQ is taken.
+    inline void HaltWithInterruptsDisabled() {
+        asm volatile("wfi\n\tmsr daifclr, #2\n\tisb\n\tmsr daifset, #2" ::: "memory");
+    }
+
+    [[noreturn]] inline void HaltForever() {
+        for (;;) {
+            asm volatile("msr daifset, #0xf\n\twfi" ::: "memory");
+        }
+    }
+
+    inline uint64_t ReadTimestamp() {
+        uint64_t value;
+        asm volatile("isb\n\tmrs %0, cntvct_el0" : "=r"(value) :: "memory");
+        return value;
+    }
+
+    // The generic timer runs from boot at CNTFRQ_EL0 Hz, so it times short
+    // delays without calibration.
+    inline void IoDelay() {
+        uint64_t frequency;
+        asm volatile("mrs %0, cntfrq_el0" : "=r"(frequency));
+        uint64_t ticks = frequency / 1000000 + 1;
+        uint64_t start = ReadTimestamp();
+        while (ReadTimestamp() - start < ticks) {
+            Pause();
+        }
+    }
+
+    // ARMv8 has no "clean the whole cache" instruction: set/way maintenance
+    // is only safe from firmware, which PSCI performs on power transitions.
+    // Order outstanding accesses so the firmware call sees them.
+    inline void WritebackInvalidateCaches() {
+        asm volatile("dsb sy\n\tisb" ::: "memory");
+    }
+
+    // The TLBI operand is VA[55:12] in bits [43:0]; the bits above are RES0.
+    inline uint64_t DataCacheLineSize() {
+        uint64_t ctr;
+        asm volatile("mrs %0, ctr_el0" : "=r"(ctr));
+        return 4ULL << ((ctr >> 16) & 0xF);  // DminLine
+    }
+
+    // To the point of coherency: where non-snooping device DMA looks.
+    inline void CleanDataCache(const void* address, uint64_t size) {
+        uint64_t line = DataCacheLineSize();
+        uint64_t end = (uint64_t)address + size;
+        for (uint64_t p = (uint64_t)address & ~(line - 1); p < end; p += line) {
+            asm volatile("dc cvac, %0" :: "r"(p) : "memory");
+        }
+        asm volatile("dsb sy" ::: "memory");
+    }
+
+    // Clean too: a partial line at either end may hold CPU data next to the
+    // buffer, which a plain invalidate would throw away.
+    inline void InvalidateDataCache(const void* address, uint64_t size) {
+        uint64_t line = DataCacheLineSize();
+        uint64_t end = (uint64_t)address + size;
+        for (uint64_t p = (uint64_t)address & ~(line - 1); p < end; p += line) {
+            asm volatile("dc civac, %0" :: "r"(p) : "memory");
+        }
+        asm volatile("dsb sy" ::: "memory");
+    }
+
+    // Clean the data cache to the point of unification, then invalidate
+    // the whole instruction cache: invalidating by address would only
+    // reach the alias it names on an aliasing (VIPT) instruction cache.
+    inline void SyncInstructionCache(const void* address, uint64_t size) {
+        uint64_t line = DataCacheLineSize();
+        uint64_t start = (uint64_t)address & ~(line - 1);
+        uint64_t end = (uint64_t)address + size;
+        for (uint64_t p = start; p < end; p += line) {
+            asm volatile("dc cvau, %0" :: "r"(p) : "memory");
+        }
+        asm volatile("dsb ish\n\tic ialluis\n\tdsb ish\n\tisb" ::: "memory");
+    }
+
+    inline void FlushTlbPage(uint64_t virtualAddress) {
+        asm volatile("dsb ishst\n\ttlbi vaae1is, %0\n\tdsb ish\n\tisb"
+                     :: "r"((virtualAddress >> 12) & ((1ULL << 44) - 1)) : "memory");
+    }
+
+    inline void FlushTlbAll() {
+        asm volatile("dsb ishst\n\ttlbi vmalle1is\n\tdsb ish\n\tisb" ::: "memory");
+    }
+
+    // The process (lower) half of the address space is translated through
+    // TTBR0_EL1; the kernel half through TTBR1_EL1, which never changes.
+    static constexpr uint64_t TtbrAddressMask = 0x0000FFFFFFFFFFFEULL;
+
+    inline uint64_t ReadPageTableRoot() {
+        uint64_t ttbr;
+        asm volatile("mrs %0, ttbr0_el1" : "=r"(ttbr));
+        return ttbr & TtbrAddressMask;
+    }
+
+    // No ASIDs yet: every switch invalidates the old process's entries.
+    inline void WritePageTableRoot(uint64_t rootPhys) {
+        asm volatile("dsb ishst\n\tmsr ttbr0_el1, %0\n\tisb\n\ttlbi vmalle1\n\tdsb nsh\n\tisb"
+                     :: "r"(rootPhys) : "memory");
+    }
+
+}

kernel/src/Arch/aarch64/CpuPower.cpp

@@ -0,0 +1,26 @@
+/*
+    * CpuPower.cpp
+    * aarch64 CPU power management (implements Arch/CpuPower.hpp)
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#include <Arch/CpuPower.hpp>
+
+// No frequency scaling or thermal governor yet (on ARM these come from
+// ACPI CPPC / SCMI rather than CPU registers).
+
+namespace Arch::CpuPower {
+
+    void InitializeBsp() {}
+    void InitializeAp() {}
+    void ApplyPolicyIfChanged() {}
+    void ThermalTick(uint64_t) {}
+    void ServiceDeferredDiagnostics() {}
+    void ReapplyAfterWake() {}
+
+    bool GetSnapshot(Snapshot& out) {
+        out = {};
+        return false;
+    }
+
+}

kernel/src/Arch/aarch64/Exceptions.cpp

@@ -0,0 +1,184 @@
+/*
+    * Exceptions.cpp
+    * aarch64 synchronous exceptions: faults, user crashes, kernel panics
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#include <Arch/aarch64/InterruptFrame.hpp>
+#include <Arch/Cpu.hpp>
+#include <Arch/Smp.hpp>
+#include <Common/Panic.hpp>
+#include <Common/TerminationCodes.hpp>
+#include <CppLib/Stream.hpp>
+#include <Memory/Paging.hpp>
+#include <Sched/Scheduler.hpp>
+#include <Sched/CrashReport.hpp>
+#include <Api/Heap.hpp>
+#include <Timekeeping/Timer.hpp>
+
+using namespace Kt;
+
+namespace {
+
+    // ESR_EL1.EC exception classes
+    constexpr uint64_t EcUnknown         = 0x00;
+    constexpr uint64_t EcIllegalState    = 0x0E;
+    constexpr uint64_t EcInstAbortLower  = 0x20;
+    constexpr uint64_t EcInstAbortSame   = 0x21;
+    constexpr uint64_t EcPcAlignment     = 0x22;
+    constexpr uint64_t EcDataAbortLower  = 0x24;
+    constexpr uint64_t EcDataAbortSame   = 0x25;
+    constexpr uint64_t EcSpAlignment     = 0x26;
+    constexpr uint64_t EcFpException     = 0x2C;
+    constexpr uint64_t EcBreakpointLower = 0x30;
+    constexpr uint64_t EcBrk             = 0x3C;
+
+    constexpr uint64_t UserSpaceEnd = 0x0000800000000000ULL;
+
+    uint64_t ExceptionClass(const Arch::PanicFrame* frame) {
+        return (frame->esr >> 26) & 0x3F;
+    }
+
+    bool IsAbort(uint64_t ec) {
+        return ec == EcInstAbortLower || ec == EcInstAbortSame
+            || ec == EcDataAbortLower || ec == EcDataAbortSame;
+    }
+
+    // Fault status code (ISS[5:0]) of an abort: translation faults mean
+    // the page is not mapped, permission faults that it is.
+    bool IsTranslationFault(uint64_t esr) {
+        uint64_t fsc = esr & 0x3F;
+        return fsc >= 0x04 && fsc <= 0x07;
+    }
+
+    bool IsPermissionFault(uint64_t esr) {
+        uint64_t fsc = esr & 0x3F;
+        return fsc >= 0x0C && fsc <= 0x0F;
+    }
+
+    const char* ExceptionName(uint64_t ec) {
+        switch (ec) {
+            case EcUnknown:         return "Undefined Instruction";
+            case EcIllegalState:    return "Illegal Execution State";
+            case EcInstAbortLower:
+            case EcInstAbortSame:   return "Instruction Abort";
+            case EcPcAlignment:     return "PC Alignment Fault";
+            case EcDataAbortLower:
+            case EcDataAbortSame:   return "Data Abort";
+            case EcSpAlignment:     return "SP Alignment Fault";
+            case EcFpException:     return "Floating-Point Exception";
+            case EcBrk:             return "Breakpoint (BRK)";
+            default:
+                if (ec >= EcBreakpointLower && ec <= 0x35) return "Debug Exception";
+                return "Synchronous Exception";
+        }
+    }
+
+    TerminationCodes TerminationFromClass(uint64_t ec) {
+        if (ec == EcUnknown || ec == EcIllegalState) return TerminatedIllegalInstruction;
+        if (ec == EcBrk || (ec >= EcBreakpointLower && ec <= 0x35)) return TerminatedBreakpoint;
+        if (ec == EcFpException) return TerminatedArithmeticError;
+        return TerminatedAccessViolation;
+    }
+
+    template<size_t N>
+    void CopyName(char (&dst)[N], const char* src) {
+        size_t i = 0;
+        for (; i + 1 < N && src[i]; i++) dst[i] = src[i];
+        dst[i] = '\0';
+    }
+
+    // Kill the faulting process with a crash report. Never returns.
+    [[noreturn]] void TerminateUserProcess(Arch::PanicFrame* frame, uint64_t ec) {
+        // Teardown can wait on devices and other CPUs; keep the clock running.
+        Arch::EnableInterrupts();
+
+        auto* proc = Sched::GetCurrentProcessPtr();
+        const char* name = ExceptionName(ec);
+        KernelLogStream(ERROR, "Exception") << name << " in process \"" << proc->name
+            << "\" (pid " << proc->pid << ", ESR 0x" << base::hex << frame->esr
+            << ", PC 0x" << frame->elr << ") - process terminated";
+
+        CrashReport::Report rep{};
+        rep.valid = true;
+        rep.pid = proc->pid;
+        CopyName(rep.processName, proc->name);
+        rep.exceptionVector = (uint8_t)ec;
+        CopyName(rep.exceptionName, name);
+        rep.instructionPointer = frame->elr;
+        rep.flags = frame->spsr;
+        rep.stackPointer = frame->sp;
+
+        constexpr uint64_t snapshotBytes = CrashReport::StackSnapshotWords * sizeof(uint64_t);
+        if (Memory::VMM::Paging::IsUserRangeAccessible(proc->pageTableRoot, frame->sp,
+                                                       snapshotBytes, false)) {
+            auto* userStack = (const volatile uint64_t*)frame->sp;
+            for (int word = 0; word < CrashReport::StackSnapshotWords; word++)
+                rep.stackWords[word] = userStack[word];
+            rep.stackWordCount = CrashReport::StackSnapshotWords;
+        }
+
+        if (IsAbort(ec)) {
+            rep.faultingAddress = frame->far;
+            rep.pfPresent = IsPermissionFault(frame->esr);
+            rep.pfWrite = ec == EcDataAbortLower && (frame->esr & (1ULL << 6)) != 0;
+            rep.pfUser = 1;
+            rep.pfInstructionFetch = ec == EcInstAbortLower;
+        }
+        rep.timestampTick = Timekeeping::GetTicks();
+
+        CrashReport::AddReport(rep);
+        Sched::SpawnCrashPad(proc->pid);
+
+        // As on x86: a sibling thread's fault must end the whole process.
+        int code = TerminationFromClass(ec);
+        if (Sched::GetCurrentThreadPtr() != proc) {
+            Sched::KillProcess(proc->pid, code);
+        } else {
+            Sched::SetProcessExitCode(proc->pid, code);
+        }
+
+        Sched::ExitProcess();
+        __builtin_unreachable();
+    }
+
+    // A not-present user page: anonymous memory mapped lazily or the user
+    // stack growing. Also reached from the kernel touching user buffers.
+    bool TryResolveUserFault(const Arch::PanicFrame* frame, uint64_t ec) {
+        if (!IsAbort(ec) || !IsTranslationFault(frame->esr)) return false;
+        if (frame->far >= UserSpaceEnd) return false;
+
+        auto* cpu = Smp::TryGetCurrentCpuData();
+        if (cpu == nullptr || cpu->currentSlot < 0 || Sched::GetCurrentPid() < 0) return false;
+
+        bool isData = ec == EcDataAbortLower || ec == EcDataAbortSame;
+        Memory::VMM::PageFault fault{false, isData && (frame->esr & (1ULL << 6)) != 0, !isData};
+        return montauk::abi::TryHandleAnonymousPageFault(frame->far, fault)
+            || Sched::TryGrowUserStack(frame->far);
+    }
+
+}
+
+extern "C" void ArchSyncException(Arch::PanicFrame* frame, bool fromUser) {
+    uint64_t ec = ExceptionClass(frame);
+
+    if (TryResolveUserFault(frame, ec)) {
+        return;
+    }
+
+    if (fromUser && Sched::GetCurrentPid() >= 0) {
+        TerminateUserProcess(frame, ec);
+    }
+
+    Panic(ExceptionName(ec), frame);
+}
+
+// FIQs are never routed, SErrors are asynchronous external aborts, and the
+// kernel neither runs on SP_EL0 nor supports AArch32 user code.
+extern "C" void ArchUnexpectedException(Arch::PanicFrame* frame, uint64_t kind) {
+    static const char* const names[] = {
+        "Exception on SP_EL0", "Unexpected FIQ", "SError (asynchronous abort)",
+        "AArch32 exception",
+    };
+    Panic(kind < 4 ? names[kind] : "Unexpected exception", frame);
+}

kernel/src/Arch/aarch64/Gic.hpp

@@ -0,0 +1,40 @@
+/*
+    * Gic.hpp
+    * aarch64 GICv2 interrupt controller (private to the port)
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+#include <ACPI/ACPI.hpp>
+
+namespace Arch::Gic {
+
+    // Bring up the distributor and the calling CPU's interface from the
+    // ACPI MADT. Returns false when no usable GIC is described.
+    bool Initialize(Hal::ACPI::CommonSDTHeader* xsdt);
+
+    bool IsInitialized();
+
+    // The XSDT Initialize was given (for the other ACPI tables the port reads).
+    Hal::ACPI::CommonSDTHeader* CurrentXsdt();
+
+    // Per-CPU part of Initialize, for secondary CPUs.
+    void InitializeCpuInterface();
+
+    // Map a private peripheral interrupt (16-31) onto an IRQ slot and enable
+    // it on the calling CPU.
+    void RoutePpi(uint32_t intid, uint8_t irq, bool edgeTriggered);
+
+    // Called for an IRQ slot after the interrupt is acknowledged and before
+    // it is completed (EOI): level-triggered sources deassert here.
+    using AckHook = void (*)();
+    void SetAckHook(uint8_t irq, AckHook hook);
+
+    // The calling CPU's target bit, as the distributor sees it.
+    uint8_t CurrentTargetMask();
+
+    // Find an ACPI table in the XSDT by signature.
+    Hal::ACPI::CommonSDTHeader* FindTable(Hal::ACPI::CommonSDTHeader* xsdt, const char* signature);
+
+}

kernel/src/Arch/aarch64/InterruptFrame.hpp

@@ -0,0 +1,24 @@
+/*
+    * InterruptFrame.hpp
+    * aarch64 exception frame
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+
+namespace Arch {
+
+    // What the exception vectors (Vectors.S) save before calling into C++.
+    struct PanicFrame {
+        uint64_t x[31];   // x0-x30
+        uint64_t sp;      // interrupted SP (SP_EL0 for EL0, else SP_EL1)
+        uint64_t elr;     // ELR_EL1: faulting/preferred-return PC
+        uint64_t spsr;    // SPSR_EL1
+        uint64_t esr;     // ESR_EL1: exception syndrome
+        uint64_t far;     // FAR_EL1: faulting address
+    };
+
+    static_assert(sizeof(PanicFrame) == 288, "must match Vectors.S");
+
+}

kernel/src/Arch/aarch64/Interrupts.cpp

@@ -0,0 +1,420 @@
+/*
+    * Interrupts.cpp
+    * aarch64 GICv2 driver and IRQ slot dispatch (implements Arch/Interrupts.hpp)
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#include <Arch/Interrupts.hpp>
+#include <Arch/Smp.hpp>
+#include <Arch/aarch64/Gic.hpp>
+#include <Memory/HHDM.hpp>
+#include <Memory/Paging.hpp>
+#include <Terminal/Terminal.hpp>
+#include <CppLib/Stream.hpp>
+
+// IRQ slots map onto GIC interrupt IDs (INTIDs):
+//   IRQ_TIMER            the generic timer's PPI (Timer.cpp routes it)
+//   24-45 (MSI slots)    SPIs of the GICv2m MSI frame, one per slot
+//   IRQ_TLB_SHOOTDOWN,
+//   IRQ_RESCHEDULE       SGIs 0 and 1
+//   1-23 (platform lines) SPIs the PCI layer assigns to INTx pins from
+//                        ACPI _PRT (AssignPlatformLine)
+
+using namespace Kt;
+
+namespace Arch {
+
+    // Distributor registers
+    static constexpr uint32_t GICD_CTLR       = 0x000;
+    static constexpr uint32_t GICD_TYPER      = 0x004;
+    static constexpr uint32_t GICD_ISENABLER  = 0x100;
+    static constexpr uint32_t GICD_ICENABLER  = 0x180;
+    static constexpr uint32_t GICD_ICPENDR    = 0x280;
+    static constexpr uint32_t GICD_ICACTIVER  = 0x380;
+    static constexpr uint32_t GICD_IPRIORITYR = 0x400;
+    static constexpr uint32_t GICD_ITARGETSR  = 0x800;
+    static constexpr uint32_t GICD_ICFGR      = 0xC00;
+    static constexpr uint32_t GICD_SGIR       = 0xF00;
+
+    // CPU interface registers
+    static constexpr uint32_t GICC_CTLR = 0x00;
+    static constexpr uint32_t GICC_PMR  = 0x04;
+    static constexpr uint32_t GICC_BPR  = 0x08;
+    static constexpr uint32_t GICC_IAR  = 0x0C;
+    static constexpr uint32_t GICC_EOIR = 0x10;
+
+    // GICv2m MSI frame registers
+    static constexpr uint32_t V2M_MSI_TYPER     = 0x008;
+    static constexpr uint32_t V2M_MSI_SETSPI_NS = 0x040;
+
+    static constexpr uint32_t SpuriousIntid = 1020;
+    static constexpr uint32_t MaxIntids = 1020;
+    static constexpr uint8_t DefaultPriority = 0xA0;
+    static constexpr uint8_t NoSlot = 0xFF;
+
+    static constexpr uint8_t FirstMsiSlot = 24;
+    static constexpr uint8_t LastMsiSlot = 45;
+
+    // MADT interrupt-controller structures (ACPI 6.x)
+    static constexpr uint8_t MadtGicc = 0x0B;
+    static constexpr uint8_t MadtGicd = 0x0C;
+    static constexpr uint8_t MadtGicMsiFrame = 0x0D;
+
+    struct MadtEntry {
+        uint8_t Type;
+        uint8_t Length;
+    } __attribute__((packed));
+
+    struct MadtGiccEntry {
+        MadtEntry Header;
+        uint16_t Reserved;
+        uint32_t CpuInterfaceNumber;
+        uint32_t AcpiProcessorUid;
+        uint32_t Flags;
+        uint32_t ParkingProtocolVersion;
+        uint32_t PerformanceInterruptGsiv;
+        uint64_t ParkedAddress;
+        uint64_t PhysicalBaseAddress;  // GICC
+        uint64_t Gicv;
+        uint64_t Gich;
+        uint32_t VgicMaintenanceInterrupt;
+        uint64_t GicrBaseAddress;
+        uint64_t Mpidr;
+    } __attribute__((packed));
+
+    struct MadtGicdEntry {
+        MadtEntry Header;
+        uint16_t Reserved;
+        uint32_t GicId;
+        uint64_t PhysicalBaseAddress;
+        uint32_t SystemVectorBase;
+        uint8_t  GicVersion;
+    } __attribute__((packed));
+
+    struct MadtMsiFrameEntry {
+        MadtEntry Header;
+        uint16_t Reserved;
+        uint32_t FrameId;
+        uint64_t PhysicalBaseAddress;
+        uint32_t Flags;                 // bit 0: SpiCount/SpiBase override TYPER
+        uint16_t SpiCount;
+        uint16_t SpiBase;
+    } __attribute__((packed));
+
+    static constexpr uint32_t MadtHeaderSize = 44;  // SDT header + LAPIC address + flags
+
+    static IrqHandler g_irqHandlers[IRQ_COUNT] = {};
+    static Gic::AckHook g_ackHooks[IRQ_COUNT] = {};
+    static uint8_t g_intidToSlot[MaxIntids];
+    static uint32_t g_slotToIntid[IRQ_COUNT];
+
+    static volatile uint8_t* g_gicd = nullptr;
+    static volatile uint8_t* g_gicc = nullptr;
+    static uint32_t g_intidCount = 0;
+
+    static Hal::ACPI::CommonSDTHeader* g_xsdt = nullptr;
+
+    struct PlatformLine {
+        uint32_t gsi;       // 0: unassigned
+        bool edgeTriggered;
+    };
+    static PlatformLine g_platformLines[24] = {};
+
+    static uint64_t g_msiFramePhys = 0;
+    static uint32_t g_msiSpiBase = 0;
+    static uint32_t g_msiSpiCount = 0;
+
+    static inline uint32_t GicdRead(uint32_t reg) {
+        return *(volatile uint32_t*)(g_gicd + reg);
+    }
+
+    static inline void GicdWrite(uint32_t reg, uint32_t value) {
+        *(volatile uint32_t*)(g_gicd + reg) = value;
+    }
+
+    static inline void GicdWrite8(uint32_t reg, uint8_t value) {
+        *(volatile uint8_t*)(g_gicd + reg) = value;
+    }
+
+    static inline uint8_t GicdRead8(uint32_t reg) {
+        return *(volatile uint8_t*)(g_gicd + reg);
+    }
+
+    static inline uint32_t GiccRead(uint32_t reg) {
+        return *(volatile uint32_t*)(g_gicc + reg);
+    }
+
+    static inline void GiccWrite(uint32_t reg, uint32_t value) {
+        *(volatile uint32_t*)(g_gicc + reg) = value;
+    }
+
+    static volatile uint8_t* MapRegisters(uint64_t phys, uint64_t size) {
+        for (uint64_t offset = 0; offset < size; offset += 0x1000) {
+            Memory::VMM::g_paging->MapMMIO(phys + offset, Memory::HHDM(phys + offset));
+        }
+        return (volatile uint8_t*)Memory::HHDM(phys);
+    }
+
+    static void SetConfig(uint32_t intid, bool edgeTriggered) {
+        uint32_t reg = GICD_ICFGR + (intid / 16) * 4;
+        uint32_t shift = (intid % 16) * 2 + 1;
+        uint32_t value = GicdRead(reg);
+        if (edgeTriggered) value |= 1u << shift;
+        else value &= ~(1u << shift);
+        GicdWrite(reg, value);
+    }
+
+    static void EnableIntid(uint32_t intid) {
+        GicdWrite(GICD_ISENABLER + (intid / 32) * 4, 1u << (intid % 32));
+    }
+
+    static void Bind(uint32_t intid, uint8_t irq) {
+        if (intid >= MaxIntids || irq >= IRQ_COUNT) return;
+        g_intidToSlot[intid] = irq;
+        g_slotToIntid[irq] = intid;
+    }
+
+    // Route a shared peripheral interrupt to the boot CPU and unmask it.
+    static void RouteSpi(uint32_t intid, uint8_t irq, bool edgeTriggered) {
+        if (intid < 32 || intid >= g_intidCount) return;
+        Bind(intid, irq);
+        GicdWrite8(GICD_IPRIORITYR + intid, DefaultPriority);
+        GicdWrite8(GICD_ITARGETSR + intid, Smp::GetCpuData(0)->arch.gicTargetMask);
+        SetConfig(intid, edgeTriggered);
+        EnableIntid(intid);
+    }
+
+    namespace Gic {
+
+        Hal::ACPI::CommonSDTHeader* FindTable(Hal::ACPI::CommonSDTHeader* xsdt, const char* signature) {
+            if (xsdt == nullptr) return nullptr;
+            uint32_t count = (xsdt->Length - sizeof(Hal::ACPI::CommonSDTHeader)) / 8;
+            auto* entries = (uint64_t*)((uint64_t)xsdt + sizeof(Hal::ACPI::CommonSDTHeader));
+            for (uint32_t i = 0; i < count; i++) {
+                auto* header = (Hal::ACPI::CommonSDTHeader*)Memory::HHDM(entries[i]);
+                if (header->Signature[0] == signature[0] && header->Signature[1] == signature[1]
+                 && header->Signature[2] == signature[2] && header->Signature[3] == signature[3]) {
+                    return header;
+                }
+            }
+            return nullptr;
+        }
+
+        // ITARGETSR0-7 are banked: each byte reads back the reading CPU's
+        // own interface bit.
+        uint8_t CurrentTargetMask() {
+            uint32_t value = GicdRead(GICD_ITARGETSR);
+            return (uint8_t)(value | value >> 8 | value >> 16 | value >> 24);
+        }
+
+        void InitializeCpuInterface() {
+            // SGIs and PPIs are banked per CPU: disable the PPIs until a
+            // driver routes one, keep the SGIs (IPIs) enabled.
+            GicdWrite(GICD_ICENABLER, 0xFFFF0000u);
+            GicdWrite(GICD_ISENABLER, 0x0000FFFFu);
+            GicdWrite(GICD_ICPENDR, 0xFFFFFFFFu);
+            for (uint32_t intid = 0; intid < 32; intid++) {
+                GicdWrite8(GICD_IPRIORITYR + intid, DefaultPriority);
+            }
+
+            GiccWrite(GICC_PMR, 0xF0);
+            GiccWrite(GICC_BPR, 0);
+            GiccWrite(GICC_CTLR, 1);
+
+            auto* cpu = Smp::GetCurrentCpuData();
+            cpu->arch.gicTargetMask = CurrentTargetMask();
+        }
+
+        void RoutePpi(uint32_t intid, uint8_t irq, bool edgeTriggered) {
+            if (intid < 16 || intid >= 32) return;
+            Bind(intid, irq);
+            SetConfig(intid, edgeTriggered);
+            EnableIntid(intid);
+        }
+
+        void SetAckHook(uint8_t irq, AckHook hook) {
+            if (irq < IRQ_COUNT) g_ackHooks[irq] = hook;
+        }
+
+        bool IsInitialized() {
+            return g_gicc != nullptr;
+        }
+
+        Hal::ACPI::CommonSDTHeader* CurrentXsdt() {
+            return g_xsdt;
+        }
+
+        bool Initialize(Hal::ACPI::CommonSDTHeader* xsdt) {
+            g_xsdt = xsdt;
+            auto* madt = FindTable(xsdt, "APIC");
+            if (madt == nullptr) {
+                KernelLogStream(ERROR, "GIC") << "No MADT";
+                return false;
+            }
+
+            uint64_t gicdPhys = 0, giccPhys = 0;
+            uint8_t version = 0;
+            auto* cursor = (uint8_t*)madt + MadtHeaderSize;
+            auto* end = (uint8_t*)madt + madt->Length;
+            while (cursor + sizeof(MadtEntry) <= end) {
+                auto* entry = (MadtEntry*)cursor;
+                if (entry->Length < sizeof(MadtEntry)) break;
+
+                if (entry->Type == MadtGicd) {
+                    auto* gicd = (MadtGicdEntry*)entry;
+                    gicdPhys = gicd->PhysicalBaseAddress;
+                    version = gicd->GicVersion;
+                } else if (entry->Type == MadtGicc && giccPhys == 0) {
+                    giccPhys = ((MadtGiccEntry*)entry)->PhysicalBaseAddress;
+                } else if (entry->Type == MadtGicMsiFrame && g_msiFramePhys == 0) {
+                    auto* frame = (MadtMsiFrameEntry*)entry;
+                    g_msiFramePhys = frame->PhysicalBaseAddress;
+                    if (frame->Flags & 1) {
+                        g_msiSpiBase = frame->SpiBase;
+                        g_msiSpiCount = frame->SpiCount;
+                    }
+                }
+                cursor += entry->Length;
+            }
+
+            // Version 0 means "infer from the hardware"; 1 and 2 share a
+            // programming model.
+            if (gicdPhys == 0 || giccPhys == 0 || version > 2) {
+                KernelLogStream(ERROR, "GIC") << "No GICv2 in the MADT (version "
+                    << base::dec << (uint64_t)version << ")";
+                return false;
+            }
+
+            for (uint32_t i = 0; i < MaxIntids; i++) g_intidToSlot[i] = NoSlot;
+            for (int i = 0; i < IRQ_COUNT; i++) g_slotToIntid[i] = SpuriousIntid;
+            Bind(0, IRQ_TLB_SHOOTDOWN);
+            Bind(1, IRQ_RESCHEDULE);
+
+            g_gicd = MapRegisters(gicdPhys, 0x1000);
+            g_gicc = MapRegisters(giccPhys, 0x2000);
+
+            g_intidCount = ((GicdRead(GICD_TYPER) & 0x1F) + 1) * 32;
+            if (g_intidCount > MaxIntids) g_intidCount = MaxIntids;
+
+            // Distributor: everything masked, idle, level-triggered and
+            // aimed at no CPU until RouteSpi claims it.
+            GicdWrite(GICD_CTLR, 0);
+            for (uint32_t intid = 32; intid < g_intidCount; intid += 32) {
+                GicdWrite(GICD_ICENABLER + (intid / 32) * 4, 0xFFFFFFFFu);
+                GicdWrite(GICD_ICPENDR + (intid / 32) * 4, 0xFFFFFFFFu);
+                GicdWrite(GICD_ICACTIVER + (intid / 32) * 4, 0xFFFFFFFFu);
+            }
+            for (uint32_t intid = 32; intid < g_intidCount; intid++) {
+                GicdWrite8(GICD_IPRIORITYR + intid, DefaultPriority);
+                GicdWrite8(GICD_ITARGETSR + intid, 0);
+            }
+            for (uint32_t intid = 32; intid < g_intidCount; intid += 16) {
+                GicdWrite(GICD_ICFGR + (intid / 16) * 4, 0);
+            }
+            GicdWrite(GICD_CTLR, 1);
+
+            InitializeCpuInterface();
+
+            if (g_msiFramePhys != 0) {
+                volatile uint8_t* frame = MapRegisters(g_msiFramePhys, 0x1000);
+                if (g_msiSpiCount == 0) {
+                    uint32_t typer = *(volatile uint32_t*)(frame + V2M_MSI_TYPER);
+                    g_msiSpiBase = (typer >> 16) & 0x3FF;
+                    g_msiSpiCount = typer & 0x3FF;
+                }
+            }
+
+            KernelLogStream(OK, "GIC") << "GICv2: distributor 0x" << base::hex << gicdPhys
+                << ", CPU interface 0x" << giccPhys << ", " << base::dec
+                << (uint64_t)g_intidCount << " interrupt IDs";
+            if (g_msiSpiCount != 0) {
+                KernelLogStream(OK, "GIC") << "GICv2m MSI frame 0x" << base::hex << g_msiFramePhys
+                    << ": SPIs " << base::dec << (uint64_t)g_msiSpiBase << "-"
+                    << (uint64_t)(g_msiSpiBase + g_msiSpiCount - 1);
+            }
+            return true;
+        }
+
+    }
+
+    void RegisterIrqHandler(uint8_t irq, IrqHandler handler) {
+        if (irq >= IRQ_COUNT) return;
+        g_irqHandlers[irq] = handler;
+        KernelLogStream(DEBUG, "IRQ") << "Registered handler for IRQ "
+            << base::dec << (uint64_t)irq;
+    }
+
+    bool PlatformLinesAreFixed() {
+        return false;
+    }
+
+    // ACPI GSIs are GIC INTIDs. The GIC knows only active-high levels and
+    // rising edges; an active-low line is inverted on its way in.
+    bool AssignPlatformLine(uint8_t irq, uint32_t gsi, bool levelTriggered, bool) {
+        if (irq == 0 || irq >= 24 || gsi < 32 || gsi >= g_intidCount) return false;
+        g_platformLines[irq] = PlatformLine{ gsi, !levelTriggered };
+        return true;
+    }
+
+    void RouteLegacyIrq(uint8_t irq) {
+        if (irq >= 24 || g_platformLines[irq].gsi == 0) {
+            KernelLogStream(WARNING, "IRQ") << "aarch64: platform line "
+                << base::dec << (uint64_t)irq << " is not assigned";
+            return;
+        }
+        RouteSpi(g_platformLines[irq].gsi, irq, g_platformLines[irq].edgeTriggered);
+    }
+
+    void EnableLegacyIrq(uint8_t irq) {
+        RouteLegacyIrq(irq);
+    }
+
+    bool ComposeMsi(uint8_t irq, MsiMessage& out) {
+        if (g_gicd == nullptr || irq < FirstMsiSlot || irq > LastMsiSlot) return false;
+        uint32_t index = irq - FirstMsiSlot;
+        if (index >= g_msiSpiCount) return false;
+
+        uint32_t spi = g_msiSpiBase + index;
+        if (g_slotToIntid[irq] != spi) {
+            RouteSpi(spi, irq, true);
+        }
+        out = MsiMessage{ g_msiFramePhys + V2M_MSI_SETSPI_NS, spi };
+        return true;
+    }
+
+    bool SendIpi(int cpuIndex, uint8_t irq) {
+        if (g_gicd == nullptr) return false;
+        auto* cpu = Smp::GetCpuData(cpuIndex);
+        if (cpu == nullptr || cpu->arch.gicTargetMask == 0) return false;
+        uint32_t sgi = g_slotToIntid[irq];
+        if (sgi >= 16) return false;
+
+        // Make the IPI's memory effects visible before it is raised.
+        asm volatile("dsb ishst" ::: "memory");
+        GicdWrite(GICD_SGIR, ((uint32_t)cpu->arch.gicTargetMask << 16) | sgi);
+        return true;
+    }
+
+}
+
+// IRQ exception vector (Vectors.S). As on x86, the interrupt is completed
+// BEFORE the handler runs: the handler may context-switch away and not
+// come back here for a long time, and an un-EOId interrupt would hold this
+// CPU's running priority up (no further ticks). Level-triggered sources
+// get the chance to drop their line first through an ack hook.
+extern "C" void ArchIrqException(void*, bool fromUser) {
+    using namespace Arch;
+    uint32_t iar = GiccRead(GICC_IAR);
+    uint32_t intid = iar & 0x3FF;
+    if (intid >= SpuriousIntid) return;
+
+    uint8_t irq = g_intidToSlot[intid];
+    if (irq != NoSlot && g_ackHooks[irq] != nullptr) {
+        g_ackHooks[irq]();
+    }
+    GiccWrite(GICC_EOIR, iar);
+
+    if (irq != NoSlot && g_irqHandlers[irq] != nullptr) {
+        g_irqHandlers[irq](irq, fromUser);
+    }
+}

kernel/src/Arch/aarch64/IoPort.hpp

@@ -0,0 +1,28 @@
+/*
+    * IoPort.hpp
+    * aarch64 port I/O (implements Io/IoPort.hpp)
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+#include <Arch/Cpu.hpp>
+
+// ARM has no I/O address space. Until a PCI host bridge driver maps its I/O
+// aperture, port accesses behave like accesses to an absent device.
+
+namespace Io {
+    static constexpr bool HasPortIo = false;
+
+    inline void Out8(uint8_t, uint16_t) {}
+    inline void Out16(uint16_t, uint16_t) {}
+    inline void Out32(uint32_t, uint16_t) {}
+
+    inline uint8_t In8(uint16_t) { return 0xFF; }
+    inline uint16_t In16(uint16_t) { return 0xFFFF; }
+    inline uint32_t In32(uint16_t) { return 0xFFFFFFFF; }
+
+    inline void IoPortWait() {
+        Arch::IoDelay();
+    }
+};

kernel/src/Arch/aarch64/Paging.cpp

@@ -0,0 +1,528 @@
+/*
+    * Paging.cpp
+    * aarch64 virtual memory (implements Memory/Paging.hpp)
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#include <Memory/Paging.hpp>
+#include <Memory/PageFrameAllocator.hpp>
+#include <Memory/HHDM.hpp>
+#include <Arch/Cpu.hpp>
+#include <Common/Panic.hpp>
+#include <CppLib/Spinlock.hpp>
+#include <Arch/aarch64/Bringup.hpp>
+
+// 4 KiB granule, 48-bit virtual addresses, four levels (0-3). The kernel
+// half is translated through TTBR1_EL1 (g_kernelHalfRoot), which never
+// changes after Init. The "address space roots" the generic kernel switches
+// between are TTBR0_EL1 tables: a process's root holds only its own lower
+// half, and KernelRoot() is the TTBR0 table of kernel contexts, empty apart
+// from the EFI runtime identity mappings (so null pointers fault).
+//
+// The HHDM covers only what the memory map lists as RAM or firmware data:
+// unlike on x86, the holes between are device space on ARM machines, and a
+// cacheable mapping of device registers invites speculative accesses.
+
+namespace Memory::VMM {
+
+    Paging* g_paging = nullptr;
+
+    static std::uint64_t g_hhdmLimit = 0;
+
+    static std::uint64_t g_kernelHalfRoot = 0;
+
+    std::uint64_t Paging::HhdmLimit() { return g_hhdmLimit; }
+
+    // Translation table descriptor bits (VMSAv8-64, stage 1)
+    static constexpr uint64_t DescValid      = 1ULL << 0;
+    static constexpr uint64_t DescTable      = 1ULL << 1;   // table (levels 0-2) or page (level 3)
+    static constexpr uint64_t DescAttrShift  = 2;           // AttrIndx[2:0], into MAIR_EL1
+    static constexpr uint64_t DescAttrMask   = 7ULL << DescAttrShift;
+    static constexpr uint64_t DescApUser     = 1ULL << 6;   // AP[1]: EL0 may access
+    static constexpr uint64_t DescApReadOnly = 1ULL << 7;   // AP[2]
+    static constexpr uint64_t DescShInner    = 3ULL << 8;
+    static constexpr uint64_t DescAf         = 1ULL << 10;  // access flag (no AF faults)
+    static constexpr uint64_t DescNotGlobal  = 1ULL << 11;
+    static constexpr uint64_t DescPxn        = 1ULL << 53;
+    static constexpr uint64_t DescUxn        = 1ULL << 54;
+    static constexpr uint64_t DescAddrMask   = 0x0000FFFFFFFFF000ULL;
+
+    // MAIR_EL1 attribute indices
+    static constexpr uint64_t AttrNormal   = 0;  // write-back, read/write-allocate
+    static constexpr uint64_t AttrNormalNc = 1;  // non-cacheable (write-combining)
+    static constexpr uint64_t AttrDevice   = 2;  // Device-nGnRE
+    static constexpr uint64_t MairValue = 0xFFULL << (8 * AttrNormal)
+                                        | 0x44ULL << (8 * AttrNormalNc)
+                                        | 0x04ULL << (8 * AttrDevice);
+
+    static constexpr uint64_t KernelHalfBase = 0xFFFF000000000000ULL;
+    static constexpr uint64_t UserSpaceEnd   = 0x0000800000000000ULL;
+
+    // EFI memory types whose runtime regions are device registers
+    static constexpr uint32_t EfiMemoryMappedIo          = 11;
+    static constexpr uint32_t EfiMemoryMappedIoPortSpace = 12;
+
+    static kcp::Mutex pagingLock;
+
+    void LockUserPaging()   { pagingLock.Acquire(); }
+    void UnlockUserPaging() { pagingLock.Release(); }
+
+    void ActivateKernelAddressSpace() {
+        Arch::WritePageTableRoot(g_paging->KernelRoot());
+    }
+
+    void FlushTLB() {
+        Arch::FlushTlbAll();
+    }
+
+    using Descriptor = uint64_t;
+
+    static inline volatile Descriptor* TableAt(uint64_t phys) {
+        return (volatile Descriptor*)Memory::HHDM(phys);
+    }
+
+    static inline uint64_t Index(uint64_t va, int level) {
+        return (va >> (39 - 9 * level)) & 0x1FF;
+    }
+
+    // Order descriptor writes before the table walker (and later
+    // instructions) can observe them.
+    static inline void PublishTables() {
+        asm volatile("dsb ishst\n\tisb" ::: "memory");
+    }
+
+    static inline uint64_t ReadTtbr1() {
+        uint64_t ttbr;
+        asm volatile("mrs %0, ttbr1_el1" : "=r"(ttbr));
+        return ttbr & Arch::TtbrAddressMask;
+    }
+
+    // The level-3 descriptor for `va`, creating intermediate tables.
+    // Returns nullptr when out of memory or when a block mapping covers
+    // `va` (only the bootloader's tables contain blocks).
+    static volatile Descriptor* WalkCreate(uint64_t root, uint64_t va) {
+        uint64_t table = root;
+        for (int level = 0; level < 3; level++) {
+            volatile Descriptor* entry = &TableAt(table)[Index(va, level)];
+            Descriptor desc = *entry;
+            if ((desc & DescValid) == 0) {
+                void* page = Memory::g_pfa->AllocateZeroed();
+                if (page == nullptr) return nullptr;
+                table = Memory::SubHHDM(page);
+                // The zeroed table must be visible before the walker can
+                // reach it.
+                asm volatile("dsb ishst" ::: "memory");
+                *entry = table | DescTable | DescValid;
+            } else if ((desc & DescTable) == 0) {
+                return nullptr;
+            } else {
+                table = desc & DescAddrMask;
+            }
+        }
+        return &TableAt(table)[Index(va, 3)];
+    }
+
+    // The level-3 descriptor for `va`, or nullptr if a table is missing.
+    static volatile Descriptor* WalkExisting(uint64_t root, uint64_t va) {
+        uint64_t table = root;
+        for (int level = 0; level < 3; level++) {
+            Descriptor desc = TableAt(table)[Index(va, level)];
+            if ((desc & (DescValid | DescTable)) != (DescValid | DescTable)) return nullptr;
+            table = desc & DescAddrMask;
+        }
+        return &TableAt(table)[Index(va, 3)];
+    }
+
+    // Install a leaf. Replacing a live translation (other address or
+    // attributes) needs break-before-make: invalidate, flush, then write.
+    static void SetLeaf(volatile Descriptor* entry, Descriptor desc, uint64_t va) {
+        Descriptor old = *entry;
+        if (old == desc) return;
+        if (old & DescValid) {
+            *entry = 0;
+            Arch::FlushTlbPage(va);
+        }
+        *entry = desc;
+        PublishTables();
+    }
+
+    static Descriptor KernelLeaf(uint64_t phys, uint64_t attr) {
+        Descriptor desc = (phys & DescAddrMask) | DescTable | DescValid | DescAf | DescShInner
+                        | (attr << DescAttrShift) | DescUxn;
+        // Kernel memory stays executable, as on x86; device memory never is.
+        if (attr != AttrNormal) desc |= DescPxn;
+        return desc;
+    }
+
+    static Descriptor UserLeaf(uint64_t phys, uint64_t attr, bool writable, bool executable) {
+        Descriptor desc = (phys & DescAddrMask) | DescTable | DescValid | DescAf | DescShInner
+                        | DescNotGlobal | (attr << DescAttrShift) | DescApUser | DescPxn;
+        if (!writable) desc |= DescApReadOnly;
+        if (!executable) desc |= DescUxn;
+        return desc;
+    }
+
+    static inline bool IsUserLeaf(Descriptor desc) {
+        return (desc & (DescValid | DescTable)) == (DescValid | DescTable) && (desc & DescApUser);
+    }
+
+    // Translate through a table tree that may contain block mappings (the
+    // bootloader's). Returns 0 if unmapped.
+    static uint64_t Translate(uint64_t root, uint64_t va) {
+        uint64_t table = root;
+        for (int level = 0; level <= 3; level++) {
+            Descriptor desc = TableAt(table)[Index(va, level)];
+            if ((desc & DescValid) == 0) return 0;
+
+            uint64_t span = 1ULL << (39 - 9 * level);
+            bool leaf = level == 3 || (desc & DescTable) == 0;
+            if (leaf) {
+                if (level == 0) return 0;  // no level-0 blocks with 4 KiB granules
+                return ((desc & DescAddrMask) & ~(span - 1)) | (va & (span - 1));
+            }
+            table = desc & DescAddrMask;
+        }
+        return 0;
+    }
+
+    void Paging::MapEfiRuntime(const montauk::boot::EfiInfo& efi) {
+        if (efi.memoryMap == nullptr || efi.descriptorSize == 0) return;
+
+        struct EfiMemDesc {
+            uint32_t Type;
+            uint64_t PhysicalStart;
+            uint64_t VirtualStart;
+            uint64_t NumberOfPages;
+            uint64_t Attribute;
+        };
+
+        static constexpr uint64_t EFI_MEMORY_RUNTIME = 0x8000000000000000ULL;
+
+        auto* base = (uint8_t*)efi.memoryMap;
+        uint64_t count = efi.memoryMapSize / efi.descriptorSize;
+
+        for (uint64_t i = 0; i < count; i++) {
+            auto* desc = (EfiMemDesc*)(base + i * efi.descriptorSize);
+            if ((desc->Attribute & EFI_MEMORY_RUNTIME) == 0) continue;
+
+            bool device = desc->Type == EfiMemoryMappedIo || desc->Type == EfiMemoryMappedIoPortSpace;
+            for (uint64_t p = 0; p < desc->NumberOfPages; p++) {
+                uint64_t addr = desc->PhysicalStart + p * 0x1000;
+                // Firmware code runs at its HHDM alias (Efi::Init calls the
+                // services there) and reaches its data at physical addresses.
+                // Runtime regions are reserved memory, outside the HHDM.
+                if (device) {
+                    MapMMIO(addr, addr);
+                    MapMMIO(addr, Memory::HHDM(addr));
+                } else {
+                    Map(addr, addr);
+                    Map(addr, Memory::HHDM(addr));
+                }
+            }
+        }
+
+        FlushTLB();
+    }
+
+}
+
+namespace Arch {
+
+    // The new tables map the code running here at the same address as the
+    // bootloader's did, so the switch is a register write followed by a
+    // full invalidate.
+    void ActivateKernelTranslation() {
+        using namespace Memory::VMM;
+
+        // 48-bit VAs in both halves, 4 KiB granules, write-back walks of
+        // inner-shareable tables; the physical size is what the CPU has.
+        uint64_t mmfr0;
+        asm volatile("mrs %0, id_aa64mmfr0_el1" : "=r"(mmfr0));
+        uint64_t ips = mmfr0 & 0xF;
+        if (ips > 5) ips = 5;  // 52-bit output needs LPA2 descriptors
+        uint64_t tcr = 16ULL                          // T0SZ
+                     | 1ULL << 8 | 1ULL << 10        // IRGN0, ORGN0: write-back
+                     | 3ULL << 12                    // SH0: inner shareable
+                     | 0ULL << 14                    // TG0: 4 KiB
+                     | 16ULL << 16                   // T1SZ
+                     | 1ULL << 24 | 1ULL << 26       // IRGN1, ORGN1
+                     | 3ULL << 28                    // SH1
+                     | 2ULL << 30                    // TG1: 4 KiB
+                     | ips << 32;                    // IPS
+
+        asm volatile(
+            "dsb ish\n\t"
+            "msr mair_el1, %0\n\t"
+            "msr tcr_el1, %1\n\t"
+            "isb\n\t"
+            "msr ttbr1_el1, %2\n\t"
+            "msr ttbr0_el1, %3\n\t"
+            "isb\n\t"
+            "tlbi vmalle1\n\t"
+            "dsb nsh\n\t"
+            "isb"
+            :: "r"(MairValue), "r"(tcr), "r"(g_kernelHalfRoot), "r"(g_paging->KernelRoot()) : "memory");
+    }
+
+}
+
+namespace Memory::VMM {
+
+    Paging::Paging() {
+        rootPhys = Memory::SubHHDM(Memory::g_pfa->AllocateZeroed());
+        g_kernelHalfRoot = Memory::SubHHDM(Memory::g_pfa->AllocateZeroed());
+    }
+
+    // Kernel mappings of lower-half addresses (EFI identity mappings) go to
+    // the kernel contexts' TTBR0 table.
+    static uint64_t KernelRootFor(uint64_t kernelContextRoot, uint64_t va) {
+        return va >= KernelHalfBase ? g_kernelHalfRoot : kernelContextRoot;
+    }
+
+    void Paging::Init(std::uint64_t kernelBaseVirt, std::uint64_t kernelSize, const montauk::boot::MemoryMap& memMap,
+                      const montauk::boot::Framebuffer& framebuffer) {
+        if (Memory::HHDMBase < KernelHalfBase) {
+            Panic("aarch64: HHDM outside the 48-bit kernel half", nullptr);
+        }
+
+        uint64_t bootTtbr1 = ReadTtbr1();
+        uint64_t bootTcr, bootMair;
+        asm volatile("mrs %0, tcr_el1" : "=r"(bootTcr));
+        asm volatile("mrs %0, mair_el1" : "=r"(bootMair));
+        Kt::KernelLogStream(Kt::DEBUG, "VMM") << "Bootloader TCR 0x" << base::hex << bootTcr
+            << ", MAIR 0x" << bootMair << ", TTBR1 0x" << bootTtbr1;
+
+        // The kernel image, at the physical pages the bootloader loaded it to.
+        for (uint64_t va = kernelBaseVirt; va < kernelBaseVirt + kernelSize; va += 0x1000) {
+            uint64_t phys = Translate(bootTtbr1, va);
+            if (phys == 0) Panic("aarch64: kernel page missing from the bootloader's tables", nullptr);
+            Map(phys, va);
+        }
+
+        // HHDM: every region that is real backing store (see the top of
+        // this file). The limit is published for the AML interpreter.
+        uint64_t maxPhysAddr = 0;
+        for (size_t i = 0; i < memMap.count; i++) {
+            const auto& region = memMap.regions[i];
+            if (region.kind == montauk::boot::MemoryKind::Reserved
+             || region.kind == montauk::boot::MemoryKind::Unknown) {
+                continue;
+            }
+
+            uint64_t start = region.base & ~0xFFFULL;
+            uint64_t end = (region.base + region.length + 0xFFF) & ~0xFFFULL;
+            bool framebufferRegion = region.kind == montauk::boot::MemoryKind::Framebuffer;
+            for (uint64_t phys = start; phys < end; phys += 0x1000) {
+                if (framebufferRegion) MapWC(phys, Memory::HHDM(phys));
+                else Map(phys, Memory::HHDM(phys));
+            }
+            if (end > maxPhysAddr) maxPhysAddr = end;
+        }
+        g_hhdmLimit = maxPhysAddr;
+
+        // The framebuffer may not be listed in the memory map at all.
+        if (framebuffer.address != 0) {
+            uint64_t fbPhys = Memory::SubHHDM(framebuffer.address) & ~0xFFFULL;
+            uint64_t fbPages = (framebuffer.height * framebuffer.pitch + 0xFFF) / 0x1000;
+            for (uint64_t p = 0; p < fbPages; p++) {
+                uint64_t phys = fbPhys + p * 0x1000;
+                MapWC(phys, Memory::HHDM(phys));
+            }
+        }
+
+        // The bootloader wrote the image through the data cache. Push it to
+        // the point of unification and drop every instruction-cache line
+        // before the kernel relies on fetching code it has not run yet: a
+        // line still dirty in L1D is invisible to instruction fetch, which
+        // then reads stale DRAM (an "unknown reason" exception on a valid
+        // instruction).
+        Arch::SyncInstructionCache((const void*)kernelBaseVirt, kernelSize);
+
+        Arch::ActivateKernelTranslation();
+
+        Kt::KernelLogStream(Kt::OK, "VMM") << "Switched to kernel translation tables (HHDM limit 0x"
+            << base::hex << g_hhdmLimit << ")";
+    }
+
+    static void MapKernelPage(uint64_t root, uint64_t phys, uint64_t va, uint64_t attr, const char* what) {
+        if (va % 0x1000 != 0 || phys % 0x1000 != 0) {
+            Panic(what, nullptr);
+        }
+        volatile Descriptor* entry = WalkCreate(KernelRootFor(root, va), va);
+        if (entry == nullptr) {
+            Panic("OOM in Paging (kernel page table allocation)", nullptr);
+        }
+        SetLeaf(entry, KernelLeaf(phys, attr), va);
+    }
+
+    void Paging::Map(std::uint64_t physicalAddress, std::uint64_t virtualAddress) {
+        MapKernelPage(rootPhys, physicalAddress, virtualAddress, AttrNormal,
+                      "Value that isn't page-aligned passed as address to Paging::Map!");
+    }
+
+    void Paging::MapMMIO(std::uint64_t physicalAddress, std::uint64_t virtualAddress) {
+        MapKernelPage(rootPhys, physicalAddress, virtualAddress, AttrDevice,
+                      "Value that isn't page-aligned passed as address to Paging::MapMMIO!");
+    }
+
+    void Paging::MapWC(std::uint64_t physicalAddress, std::uint64_t virtualAddress) {
+        MapKernelPage(rootPhys, physicalAddress, virtualAddress, AttrNormalNc,
+                      "Value that isn't page-aligned passed as address to Paging::MapWC!");
+    }
+
+    void Paging::MapUser(std::uint64_t physicalAddress, std::uint64_t virtualAddress) {
+        if (virtualAddress % 0x1000 != 0 || physicalAddress % 0x1000 != 0) {
+            Panic("Value that isn't page-aligned passed as address to Paging::MapUser!", nullptr);
+        }
+        volatile Descriptor* entry = WalkCreate(rootPhys, virtualAddress);
+        if (entry == nullptr) {
+            Panic("OOM in Paging::MapUser (page table allocation)", nullptr);
+        }
+        SetLeaf(entry, UserLeaf(physicalAddress, AttrNormal, true, true), virtualAddress);
+    }
+
+    std::uint64_t Paging::GetPhysAddr(std::uint64_t root, std::uint64_t virtualAddress) {
+        if (virtualAddress >= KernelHalfBase) {
+            root = g_kernelHalfRoot;
+        }
+        return Translate(root, virtualAddress);
+    }
+
+    std::uint64_t Paging::GetPhysAddr(std::uint64_t virtualAddress) {
+        return GetPhysAddr(rootPhys, virtualAddress);
+    }
+
+    // The kernel half lives in TTBR1, so a new address space starts empty.
+    std::uint64_t Paging::CreateUserAddressSpace() {
+        void* page = Memory::g_pfa->AllocateZeroed();
+        if (page == nullptr) return 0;
+        return Memory::SubHHDM(page);
+    }
+
+    bool Paging::MapUserIn(std::uint64_t root, std::uint64_t physicalAddress, std::uint64_t virtualAddress) {
+        return MapUserInPermissions(root, physicalAddress, virtualAddress, true, true);
+    }
+
+    static bool MapUserLeaf(uint64_t root, uint64_t phys, uint64_t va, Descriptor desc, const char* what) {
+        pagingLock.Acquire();
+        if (va % 0x1000 != 0 || phys % 0x1000 != 0) {
+            pagingLock.Release();
+            Panic(what, nullptr);
+        }
+        volatile Descriptor* entry = WalkCreate(root, va);
+        if (entry == nullptr) {
+            pagingLock.Release();
+            return false;
+        }
+        SetLeaf(entry, desc, va);
+        pagingLock.Release();
+        return true;
+    }
+
+    bool Paging::MapUserInPermissions(std::uint64_t root, std::uint64_t physicalAddress,
+                                      std::uint64_t virtualAddress, bool writable, bool executable) {
+        return MapUserLeaf(root, physicalAddress, virtualAddress,
+                           UserLeaf(physicalAddress, AttrNormal, writable, executable),
+                           "Non-aligned address in Paging::MapUserIn!");
+    }
+
+    bool Paging::MapUserInWC(std::uint64_t root, std::uint64_t physicalAddress, std::uint64_t virtualAddress) {
+        return MapUserLeaf(root, physicalAddress, virtualAddress,
+                           UserLeaf(physicalAddress, AttrNormalNc, true, false),
+                           "Non-aligned address in Paging::MapUserInWC!");
+    }
+
+    // Permission changes need no break-before-make, only the TLB flush.
+    bool Paging::ProtectUserIn(std::uint64_t root, std::uint64_t virtualAddress,
+                               bool writable, bool executable) {
+        if ((virtualAddress & 0xFFFULL) != 0) return false;
+        pagingLock.Acquire();
+        volatile Descriptor* entry = WalkExisting(root, virtualAddress);
+        if (entry == nullptr || !IsUserLeaf(*entry)) {
+            pagingLock.Release();
+            return false;
+        }
+
+        Descriptor desc = *entry;
+        if (writable) desc &= ~DescApReadOnly;
+        else desc |= DescApReadOnly;
+        if (executable) desc &= ~DescUxn;
+        else desc |= DescUxn;
+        *entry = desc;
+        Arch::FlushTlbPage(virtualAddress);
+        pagingLock.Release();
+        return true;
+    }
+
+    void Paging::UnmapUserIn(std::uint64_t root, std::uint64_t virtualAddress) {
+        pagingLock.Acquire();
+        volatile Descriptor* entry = WalkExisting(root, virtualAddress);
+        if (entry != nullptr && (*entry & DescValid)) {
+            *entry = 0;
+            Arch::FlushTlbPage(virtualAddress);
+        }
+        pagingLock.Release();
+    }
+
+    void Paging::FreeUserHalf(std::uint64_t root) {
+        // User space is the lower 128 TiB: level-0 entries 0-255.
+        for (int i0 = 0; i0 < 256; i0++) {
+            Descriptor d0 = TableAt(root)[i0];
+            if ((d0 & DescValid) == 0) continue;
+            uint64_t l1 = d0 & DescAddrMask;
+
+            for (int i1 = 0; i1 < 512; i1++) {
+                Descriptor d1 = TableAt(l1)[i1];
+                if ((d1 & DescValid) == 0) continue;
+                uint64_t l2 = d1 & DescAddrMask;
+
+                for (int i2 = 0; i2 < 512; i2++) {
+                    Descriptor d2 = TableAt(l2)[i2];
+                    if ((d2 & DescValid) == 0) continue;
+                    uint64_t l3 = d2 & DescAddrMask;
+
+                    for (int i3 = 0; i3 < 512; i3++) {
+                        Descriptor leaf = TableAt(l3)[i3];
+                        if ((leaf & DescValid) == 0) continue;
+
+                        // Write-combining/device pages are not the process's.
+                        if (((leaf & DescAttrMask) >> DescAttrShift) != AttrNormal) continue;
+
+                        uint64_t pagePhys = leaf & DescAddrMask;
+                        if (pagePhys != 0) {
+                            Memory::g_pfa->Free((void*)Memory::HHDM(pagePhys));
+                        }
+                    }
+
+                    Memory::g_pfa->Free((void*)Memory::HHDM(l3));
+                }
+
+                Memory::g_pfa->Free((void*)Memory::HHDM(l2));
+            }
+
+            Memory::g_pfa->Free((void*)Memory::HHDM(l1));
+        }
+    }
+
+    bool Paging::IsUserRangeAccessible(std::uint64_t root, std::uint64_t virtualAddress,
+                                       std::uint64_t size, bool requireWrite) {
+        if (size == 0) return true;
+        if (virtualAddress >= UserSpaceEnd) return false;
+        if (virtualAddress + size < virtualAddress) return false;
+        if (virtualAddress + size > UserSpaceEnd) return false;
+
+        uint64_t page = virtualAddress & ~0xFFFULL;
+        uint64_t lastPage = (virtualAddress + size - 1) & ~0xFFFULL;
+        for (;;) {
+            volatile Descriptor* entry = WalkExisting(root, page);
+            if (entry == nullptr) return false;
+            Descriptor desc = *entry;
+            if (!IsUserLeaf(desc)) return false;
+            if (requireWrite && (desc & DescApReadOnly)) return false;
+            if (page == lastPage) break;
+            page += 0x1000;
+        }
+        return true;
+    }
+
+}

kernel/src/Arch/aarch64/PanicFrame.cpp

@@ -0,0 +1,36 @@
+/*
+    * PanicFrame.cpp
+    * aarch64 CPU state for the panic screen
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#include <Common/Panic.hpp>
+#include <Arch/aarch64/InterruptFrame.hpp>
+#include <CppLib/BoxUI.hpp>
+
+static constexpr int BoxWidth = PanicBoxWidth;
+
+namespace Arch {
+    void PrintPanicFrame(PanicFrame* frame) {
+        PrintBoxedSeparator(kerr, BoxWidth);
+        PrintBoxedLine(kerr, "CPU State:", BoxWidth, true);
+        PrintBoxedHex(kerr, "Exception Class (ESR)", frame->esr >> 26, BoxWidth);
+        PrintBoxedHex(kerr, "Syndrome (ESR)",        frame->esr,       BoxWidth);
+        PrintBoxedHex(kerr, "Fault Address (FAR)",   frame->far,       BoxWidth);
+
+        PrintBoxedSeparator(kerr, BoxWidth);
+        PrintBoxedLine(kerr, "Registers:", BoxWidth, true);
+        PrintBoxedHex(kerr, "Program Counter (ELR)", frame->elr,       BoxWidth);
+        PrintBoxedHex(kerr, "PSTATE (SPSR)",         frame->spsr,      BoxWidth);
+        PrintBoxedHex(kerr, "Stack Pointer",         frame->sp,        BoxWidth);
+        PrintBoxedHex(kerr, "Link Register (x30)",   frame->x[30],     BoxWidth);
+
+        // An "unknown reason" exception on a kernel PC: show the word the
+        // data side sees there. A valid instruction means instruction fetch
+        // saw something else (cache maintenance); garbage means the code
+        // itself was overwritten. The fetch succeeded, so the page is mapped.
+        if ((frame->esr >> 26) == 0 && frame->elr >= 0xFFFF000000000000ULL && (frame->elr & 3) == 0) {
+            PrintBoxedHex(kerr, "Instruction at PC",     *(volatile uint32_t*)frame->elr, BoxWidth);
+        }
+    }
+}

kernel/src/Arch/aarch64/PerCpu.hpp

@@ -0,0 +1,36 @@
+/*
+    * PerCpu.hpp
+    * aarch64 per-CPU state (see Arch/Smp.hpp)
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+
+namespace Arch {
+
+    // Read by the exception vectors through TPIDR_EL1.
+    struct CpuEntryState {
+        uint64_t kernelStackTop;   // SP_EL1 to use on entry from EL0
+        uint64_t scratch;          // vector-private scratch slot
+    };
+
+    struct CpuState {
+        uint64_t mpidr;            // MPIDR_EL1 affinity of this CPU
+        uint8_t gicTargetMask;     // GICv2 CPU-interface bit (SGI/SPI targets)
+        uint64_t timerPeriod;      // tick period in counter ticks, 0 = stopped
+    };
+
+    // TPIDR_EL1 holds the CpuData pointer (zeroed in EarlyInitialize until
+    // Smp::InitBsp installs it).
+    inline void* CurrentCpuLocal() {
+        void* ptr;
+        asm volatile("mrs %0, tpidr_el1" : "=r"(ptr));
+        return ptr;
+    }
+
+    inline void* TryCurrentCpuLocal() {
+        return CurrentCpuLocal();
+    }
+
+}

kernel/src/Arch/aarch64/Platform.cpp

@@ -0,0 +1,140 @@
+/*
+    * Platform.cpp
+    * aarch64 bring-up hooks (implements Arch/Platform.hpp)
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#include <Arch/Platform.hpp>
+#include <Arch/Cpu.hpp>
+#include <Arch/Smp.hpp>
+#include <Arch/aarch64/Gic.hpp>
+#include <Arch/aarch64/Bringup.hpp>
+#include <ACPI/FADT.hpp>
+#include <Terminal/Terminal.hpp>
+#include <CppLib/Stream.hpp>
+
+using namespace Kt;
+
+extern "C" char ArchExceptionVectors[];  // Vectors.S
+
+namespace Arch {
+
+    // SCTLR_EL1.WXN: writable memory is execute-never
+    static constexpr uint64_t SctlrWxn = 1ULL << 19;
+    // SCTLR_EL1.UCT/UCI: EL0 may read CTR_EL0 and clean/invalidate caches
+    // by address (what __builtin___clear_cache needs for generated code).
+    static constexpr uint64_t SctlrUct = 1ULL << 15;
+    static constexpr uint64_t SctlrUci = 1ULL << 26;
+
+    // PSCI 0.2+ function IDs (SMC32 calling convention)
+    static constexpr uint64_t PsciSystemOff   = 0x84000008;
+    static constexpr uint64_t PsciSystemReset = 0x84000009;
+
+    // How the firmware takes PSCI calls, from the FADT's ARM_BOOT_ARCH.
+    enum class PsciConduit { None, Smc, Hvc };
+    static PsciConduit g_psci = PsciConduit::None;
+
+    static uint64_t PsciCall(uint64_t function) {
+        register uint64_t x0 asm("x0") = function;
+        register uint64_t x1 asm("x1") = 0;
+        register uint64_t x2 asm("x2") = 0;
+        register uint64_t x3 asm("x3") = 0;
+        if (g_psci == PsciConduit::Hvc) {
+            asm volatile("hvc #0" : "+r"(x0), "+r"(x1), "+r"(x2), "+r"(x3) :: "memory");
+        } else if (g_psci == PsciConduit::Smc) {
+            asm volatile("smc #0" : "+r"(x0), "+r"(x1), "+r"(x2), "+r"(x3) :: "memory");
+        }
+        return x0;
+    }
+
+    static void InitializePsci(Hal::ACPI::CommonSDTHeader* xsdt) {
+        Hal::FADT::ParsedFADT fadt{};
+        if (!Hal::FADT::Parse(xsdt, fadt) || !fadt.Valid) return;
+        if (!(fadt.ArmBootArchFlags & Hal::FADT::ArmBootPsciCompliant)) {
+            KernelLogStream(WARNING, "PSCI") << "Firmware is not PSCI compliant: no reset or power-off";
+            return;
+        }
+        g_psci = (fadt.ArmBootArchFlags & Hal::FADT::ArmBootPsciUseHvc) ? PsciConduit::Hvc : PsciConduit::Smc;
+        KernelLogStream(OK, "PSCI") << "PSCI through " << (g_psci == PsciConduit::Hvc ? "HVC" : "SMC");
+    }
+
+    void InitializeCpuRegisters() {
+        // Run on SP_EL1 so exceptions taken from the kernel arrive on the
+        // stack it is already using (Vectors.S has no SP_EL0 handlers).
+        asm volatile("mov x9, sp\n\tmsr spsel, #1\n\tmov sp, x9" ::: "x9", "memory");
+
+        // The vectors need no allocation, so take over from the
+        // bootloader's right away.
+        asm volatile("msr vbar_el1, %0\n\tisb" :: "r"(ArchExceptionVectors) : "memory");
+
+        // Kernel mappings are RWX (Paging.cpp), so writable must not
+        // imply execute-never; user code may maintain its own caches.
+        uint64_t sctlr;
+        asm volatile("mrs %0, sctlr_el1" : "=r"(sctlr));
+        sctlr &= ~SctlrWxn;
+        sctlr |= SctlrUct | SctlrUci;
+        asm volatile("msr sctlr_el1, %0\n\tisb" :: "r"(sctlr));
+
+        // CPACR_EL1.FPEN = 0b11: FP/SIMD usable at EL0 and EL1 (the kernel
+        // itself is built -mgeneral-regs-only; Context.S saves user state).
+        uint64_t cpacr;
+        asm volatile("mrs %0, cpacr_el1" : "=r"(cpacr));
+        cpacr |= 3ULL << 20;
+        asm volatile("msr cpacr_el1, %0\n\tisb" :: "r"(cpacr));
+    }
+
+    void EarlyInitialize() {
+        // No per-CPU data yet: Smp::TryGetCurrentCpuData must see null.
+        asm volatile("msr tpidr_el1, xzr");
+
+        InitializeCpuRegisters();
+    }
+
+    // Installed by EarlyInitialize.
+    void InitializeTraps() {
+        KernelLogStream(OK, "Arch") << "Installed exception vectors at 0x" << base::hex
+            << (uint64_t)ArchExceptionVectors;
+    }
+
+    // MAIR_EL1 attribute indices are chosen together with the page-table
+    // format (Paging.cpp), so there is nothing to program separately.
+    void InitializeMemoryAttributes() {
+    }
+
+    void InitializeInterrupts(Hal::ACPI::CommonSDTHeader* xsdt) {
+        Smp::InitBsp();
+        if (!Gic::Initialize(xsdt)) {
+            KernelLogStream(ERROR, "Arch") << "aarch64: no interrupt controller, running without IRQs";
+        }
+        InitializePsci(xsdt);
+    }
+
+    void InitializePlatformDevices() {
+    }
+
+    // EL0 entry needs no global state beyond Arch::ActivateThread.
+    void PrepareUserMode() {
+    }
+
+    void LateInitialize() {
+    }
+
+    void ListPlatformDevices(PlatformDeviceSink sink, void* context) {
+        if (Gic::IsInitialized()) {
+            sink(context, 1, "GICv2", "ARM Generic Interrupt Controller");
+            sink(context, 2, "Generic Timer", "ARM architected timer (EL1 virtual)");
+        }
+    }
+
+    void Reboot() {
+        PsciCall(PsciSystemReset);
+        HaltForever();
+    }
+
+    void PowerOff() {
+        if (g_psci != PsciConduit::None) {
+            PsciCall(PsciSystemOff);
+        }
+    }
+
+}

kernel/src/Arch/aarch64/Smp.cpp

@@ -0,0 +1,135 @@
+/*
+    * Smp.cpp
+    * aarch64 per-CPU data and secondary CPU start-up (implements the arch half of Arch/Smp.hpp)
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#include <Arch/Smp.hpp>
+#include <Arch/Cpu.hpp>
+#include <Arch/Syscall.hpp>
+#include <Arch/aarch64/Bringup.hpp>
+#include <Arch/aarch64/Gic.hpp>
+#include <Boot/BootInfo.hpp>
+#include <Libraries/Memory.hpp>
+#include <Terminal/Terminal.hpp>
+#include <CppLib/Stream.hpp>
+#include <Timekeeping/Timer.hpp>
+
+using namespace Kt;
+
+namespace Smp {
+
+    static CpuData g_cpus[MaxCPUs];
+    static int g_cpuCount = 0;
+    static int g_detectedCpuCount = 1;
+
+    CpuData* GetCpuData(int index) {
+        if (index < 0 || index >= g_cpuCount) return nullptr;
+        return &g_cpus[index];
+    }
+
+    int GetCpuCount() {
+        return g_cpuCount;
+    }
+
+    int GetDetectedCpuCount() {
+        return g_detectedCpuCount;
+    }
+
+    void InitBsp() {
+        memset(g_cpus, 0, sizeof(g_cpus));
+
+        CpuData& bsp = g_cpus[0];
+        bsp.selfPtr = (uint64_t)&bsp;
+        bsp.cpuIndex = 0;
+        bsp.currentSlot = -1;
+        bsp.started = true;
+        asm volatile("mrs %0, mpidr_el1" : "=r"(bsp.arch.mpidr));
+
+        asm volatile("msr tpidr_el1, %0" :: "r"(&bsp));
+        g_cpuCount = 1;
+
+        KernelLogStream(OK, "SMP") << "BSP initialized (MPIDR " << base::hex
+            << bsp.arch.mpidr << ")";
+    }
+
+    // A secondary CPU, woken by the bootloader on its own stack and still
+    // on the bootloader's translation tables. The kernel stashed this CPU's
+    // CpuData in extraArgument.
+    static void ApEntry(montauk::boot::BootCpu* bootCpu) {
+        CpuData* cpu = (CpuData*)bootCpu->extraArgument;
+
+        Arch::InitializeCpuRegisters();
+        Arch::ActivateKernelTranslation();
+        asm volatile("msr tpidr_el1, %0" :: "r"(cpu));
+
+        Arch::Gic::InitializeCpuInterface();
+        Arch::InitializeSyscallEntry();
+        Timekeeping::TimerInitializeAP();
+
+        __atomic_store_n(&cpu->started, true, __ATOMIC_RELEASE);
+        Arch::EnableInterrupts();
+
+        // Runnable work arrives as a reschedule IPI; keep the periodic tick
+        // masked while idle.
+        for (;;) {
+            Timekeeping::TimerEnterApIdle();
+            Timekeeping::ServiceDeferredWork();
+            Arch::DisableInterrupts();
+            Arch::HaltWithInterruptsDisabled();
+            Arch::EnableInterrupts();
+        }
+    }
+
+    void BootAPs(const montauk::boot::SmpInfo& smp) {
+        if (smp.cpus == nullptr || smp.startCpu == nullptr || smp.cpuCount <= 1) {
+            KernelLogStream(INFO, "SMP") << "Single CPU system - no secondary CPUs to start";
+            return;
+        }
+        g_detectedCpuCount = (int)smp.cpuCount;
+
+        if (!Arch::Gic::IsInitialized()) {
+            KernelLogStream(WARNING, "SMP") << "No interrupt controller: secondary CPUs left parked";
+            return;
+        }
+
+        int apIndex = 1;
+        for (uint64_t i = 0; i < smp.cpuCount && apIndex < MaxCPUs; i++) {
+            montauk::boot::BootCpu& info = smp.cpus[i];
+            if (info.hardwareId == smp.bspHardwareId) continue;
+
+            CpuData& ap = g_cpus[apIndex];
+            ap.selfPtr = (uint64_t)&ap;
+            ap.cpuIndex = apIndex;
+            ap.currentSlot = -1;
+            ap.started = false;
+            ap.arch.mpidr = info.hardwareId;
+
+            info.extraArgument = (uint64_t)&ap;
+            // Publish the CpuData before the CPU that reads it is woken.
+            asm volatile("dsb ish" ::: "memory");
+            smp.startCpu(&info, ApEntry);
+            apIndex++;
+        }
+        g_cpuCount = apIndex;
+
+        int online = 1;
+        for (int i = 1; i < g_cpuCount; i++) {
+            uint64_t start = Arch::ReadTimestamp();
+            uint64_t frequency;
+            asm volatile("mrs %0, cntfrq_el0" : "=r"(frequency));
+            while (!__atomic_load_n(&g_cpus[i].started, __ATOMIC_ACQUIRE) &&
+                   Arch::ReadTimestamp() - start < frequency) {
+                Arch::Pause();
+            }
+            if (g_cpus[i].started) {
+                online++;
+            } else {
+                KernelLogStream(ERROR, "SMP") << "CPU " << base::dec << (uint64_t)i
+                    << " (MPIDR " << base::hex << g_cpus[i].arch.mpidr << ") failed to start";
+            }
+        }
+
+        KernelLogStream(OK, "SMP") << base::dec << (uint64_t)online << " CPU(s) online";
+    }
+}

kernel/src/Arch/aarch64/Syscall.cpp

@@ -0,0 +1,16 @@
+/*
+    * Syscall.cpp
+    * aarch64 system-call entry (implements Arch/Syscall.hpp)
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#include <Arch/Syscall.hpp>
+
+namespace Arch {
+
+    // `svc #0` arrives through the EL0 synchronous vector installed by
+    // InitializeTraps; there is no per-CPU register to program.
+    void InitializeSyscallEntry() {
+    }
+
+}

kernel/src/Arch/aarch64/SyscallFrame.hpp

@@ -0,0 +1,25 @@
+/*
+    * SyscallFrame.hpp
+    * aarch64 system-call frame (see Arch/Syscall.hpp)
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+
+namespace Arch {
+
+    // The exception frame Vectors.S saves for `svc #0` (the same layout as
+    // Arch::PanicFrame): x0-x5 carry the arguments and x8 the system-call
+    // number; the result is written back to x0.
+    struct SyscallFrame {
+        uint64_t arg1, arg2, arg3, arg4, arg5, arg6;  // x0-x5
+        uint64_t x6, x7;
+        uint64_t syscall_nr;                           // x8
+        uint64_t x9_x30[22];
+        uint64_t sp_el0, elr, spsr, esr, far;
+    };
+
+    static_assert(sizeof(SyscallFrame) == 288, "must match Vectors.S");
+
+}

kernel/src/Arch/aarch64/Timer.cpp

@@ -0,0 +1,116 @@
+/*
+    * Timer.cpp
+    * aarch64 tick timer: the EL1 virtual generic timer (implements Arch/Timer.hpp)
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#include <Arch/Timer.hpp>
+#include <Arch/Interrupts.hpp>
+#include <Arch/Cpu.hpp>
+#include <Arch/Smp.hpp>
+#include <Arch/aarch64/Gic.hpp>
+#include <ACPI/ACPI.hpp>
+#include <Terminal/Terminal.hpp>
+#include <CppLib/Stream.hpp>
+
+// The generic timer counts at CNTFRQ_EL0 on every CPU, so it needs no
+// calibration. Each CPU's virtual timer raises a PPI (the GTDT names it,
+// normally INTID 27); the tick is re-armed from the interrupt's ack hook,
+// which also drops the level-triggered line before the EOI.
+
+using namespace Kt;
+
+namespace Arch::Timer {
+
+    // CNTV_CTL_EL0
+    static constexpr uint64_t CtlEnable = 1ULL << 0;
+    static constexpr uint64_t CtlMask   = 1ULL << 1;
+
+    // GTDT timer flags
+    static constexpr uint32_t GtdtEdgeTriggered = 1u << 0;
+
+    static constexpr uint32_t DefaultVirtualTimerIntid = 27;
+
+    struct Gtdt {
+        Hal::ACPI::CommonSDTHeader Header;
+        uint64_t CntControlBase;
+        uint32_t Reserved;
+        uint32_t SecureEl1Gsiv;
+        uint32_t SecureEl1Flags;
+        uint32_t NonSecureEl1Gsiv;
+        uint32_t NonSecureEl1Flags;
+        uint32_t VirtualTimerGsiv;
+        uint32_t VirtualTimerFlags;
+    } __attribute__((packed));
+
+    static uint64_t g_ticksPerMs = 0;
+    static uint32_t g_intid = DefaultVirtualTimerIntid;
+    static bool g_edgeTriggered = false;
+
+    static inline void WriteCtl(uint64_t value) {
+        asm volatile("msr cntv_ctl_el0, %0\n\tisb" :: "r"(value) : "memory");
+    }
+
+    static inline uint64_t ReadCval() {
+        uint64_t value;
+        asm volatile("mrs %0, cntv_cval_el0" : "=r"(value));
+        return value;
+    }
+
+    static inline void WriteCval(uint64_t value) {
+        asm volatile("msr cntv_cval_el0, %0\n\tisb" :: "r"(value) : "memory");
+    }
+
+    // Advance the deadline by one period from the previous one (no drift),
+    // skipping ticks the CPU missed rather than replaying them.
+    static void Rearm() {
+        uint64_t period = Smp::GetCurrentCpuData()->arch.timerPeriod;
+        if (period == 0) {
+            WriteCtl(CtlMask);
+            return;
+        }
+        uint64_t now = ReadTimestamp();
+        uint64_t next = ReadCval() + period;
+        if ((int64_t)(next - now) <= 0) next = now + period;
+        WriteCval(next);
+    }
+
+    static void ReadGtdt() {
+        auto* gtdt = (Gtdt*)Gic::FindTable(Gic::CurrentXsdt(), "GTDT");
+        if (gtdt == nullptr || gtdt->Header.Length < sizeof(Gtdt)) return;
+        if (gtdt->VirtualTimerGsiv >= 16 && gtdt->VirtualTimerGsiv < 32) {
+            g_intid = gtdt->VirtualTimerGsiv;
+            g_edgeTriggered = (gtdt->VirtualTimerFlags & GtdtEdgeTriggered) != 0;
+        }
+    }
+
+    bool InitializeBsp(uint32_t periodMs) {
+        uint64_t frequency;
+        asm volatile("mrs %0, cntfrq_el0" : "=r"(frequency));
+        if (frequency < 1000 || !Gic::IsInitialized()) return false;
+        g_ticksPerMs = frequency / 1000;
+
+        ReadGtdt();
+        Gic::SetAckHook(IRQ_TIMER, Rearm);
+        StartPeriodic(periodMs);
+
+        KernelLogStream(OK, "Timer") << "Generic timer: " << base::dec << frequency
+            << " Hz counter, virtual timer INTID " << (uint64_t)g_intid;
+        return true;
+    }
+
+    // PPIs are banked: each CPU enables its own timer interrupt.
+    void StartPeriodic(uint32_t periodMs) {
+        Gic::RoutePpi(g_intid, IRQ_TIMER, g_edgeTriggered);
+        uint64_t period = g_ticksPerMs * periodMs;
+        Smp::GetCurrentCpuData()->arch.timerPeriod = period;
+        WriteCval(ReadTimestamp() + period);
+        WriteCtl(CtlEnable);
+    }
+
+    void Stop() {
+        Smp::GetCurrentCpuData()->arch.timerPeriod = 0;
+        WriteCtl(CtlMask);
+    }
+
+}

kernel/src/Arch/aarch64/Vectors.S

@@ -0,0 +1,201 @@
+/*
+    Vectors.S
+    aarch64 exception vector table (VBAR_EL1)
+    Copyright (c) 2026 Daniel Hammer
+*/
+
+/*
+ * Every exception saves the same 288-byte frame on SP_EL1, laid out as
+ * Arch::PanicFrame (InterruptFrame.hpp) and Arch::SyscallFrame:
+ *
+ *     0   x0 .. x30       (x30 at 240)
+ *   248   interrupted SP  (SP_EL0 from EL0, the pre-exception SP_EL1 from EL1)
+ *   256   ELR_EL1
+ *   264   SPSR_EL1
+ *   272   ESR_EL1
+ *   280   FAR_EL1
+ *
+ * The kernel runs on SP_EL1 (SPSel = 1, see Platform.cpp). From EL0,
+ * SP_EL1 is the top of the thread's kernel stack: ArchEnterUserMode sets
+ * it before the first ERET and every return to EL0 restores it.
+ */
+
+#define FRAME_SIZE  288
+#define FRAME_SP    248
+#define FRAME_ELR   256
+#define FRAME_SPSR  264
+#define FRAME_ESR   272
+#define FRAME_FAR   280
+
+    .text
+
+.macro SAVE_FRAME el
+    sub     sp, sp, #FRAME_SIZE
+    stp     x0, x1, [sp, #0]
+    stp     x2, x3, [sp, #16]
+    stp     x4, x5, [sp, #32]
+    stp     x6, x7, [sp, #48]
+    stp     x8, x9, [sp, #64]
+    stp     x10, x11, [sp, #80]
+    stp     x12, x13, [sp, #96]
+    stp     x14, x15, [sp, #112]
+    stp     x16, x17, [sp, #128]
+    stp     x18, x19, [sp, #144]
+    stp     x20, x21, [sp, #160]
+    stp     x22, x23, [sp, #176]
+    stp     x24, x25, [sp, #192]
+    stp     x26, x27, [sp, #208]
+    stp     x28, x29, [sp, #224]
+    .if \el == 0
+    mrs     x21, sp_el0
+    .else
+    add     x21, sp, #FRAME_SIZE
+    .endif
+    stp     x30, x21, [sp, #240]
+    mrs     x22, elr_el1
+    mrs     x23, spsr_el1
+    stp     x22, x23, [sp, #FRAME_ELR]
+    mrs     x24, esr_el1
+    mrs     x25, far_el1
+    stp     x24, x25, [sp, #FRAME_ESR]
+.endm
+
+/* Interrupts are masked again before ELR/SPSR are reloaded: an IRQ taken
+   in between would overwrite them. */
+.macro RESTORE_FRAME el
+    msr     daifset, #2
+    ldp     x22, x23, [sp, #FRAME_ELR]
+    msr     elr_el1, x22
+    msr     spsr_el1, x23
+    .if \el == 0
+    ldr     x21, [sp, #FRAME_SP]
+    msr     sp_el0, x21
+    .endif
+    ldp     x0, x1, [sp, #0]
+    ldp     x2, x3, [sp, #16]
+    ldp     x4, x5, [sp, #32]
+    ldp     x6, x7, [sp, #48]
+    ldp     x8, x9, [sp, #64]
+    ldp     x10, x11, [sp, #80]
+    ldp     x12, x13, [sp, #96]
+    ldp     x14, x15, [sp, #112]
+    ldp     x16, x17, [sp, #128]
+    ldp     x18, x19, [sp, #144]
+    ldp     x20, x21, [sp, #160]
+    ldp     x22, x23, [sp, #176]
+    ldp     x24, x25, [sp, #192]
+    ldp     x26, x27, [sp, #208]
+    ldp     x28, x29, [sp, #224]
+    ldr     x30, [sp, #240]
+    add     sp, sp, #FRAME_SIZE
+    eret
+.endm
+
+/* Each vector slot is 128 bytes; the handlers live outside the table. */
+.macro VECTOR target
+    .balign 0x80
+    b       \target
+.endm
+
+    .balign 0x800
+    .global ArchExceptionVectors
+ArchExceptionVectors:
+    /* Current EL with SP_EL0: the kernel never runs on SP_EL0. */
+    VECTOR  el1t_unexpected
+    VECTOR  el1t_unexpected
+    VECTOR  el1t_unexpected
+    VECTOR  el1t_unexpected
+    /* Current EL with SP_ELx */
+    VECTOR  el1h_sync
+    VECTOR  el1h_irq
+    VECTOR  el1h_fiq
+    VECTOR  el1h_serror
+    /* Lower EL, AArch64 */
+    VECTOR  el0_sync
+    VECTOR  el0_irq
+    VECTOR  el0_fiq
+    VECTOR  el0_serror
+    /* Lower EL, AArch32: not supported */
+    VECTOR  el0_aarch32
+    VECTOR  el0_aarch32
+    VECTOR  el0_aarch32
+    VECTOR  el0_aarch32
+
+/* ---- Exceptions from the kernel ---- */
+
+el1t_unexpected:
+    SAVE_FRAME 1
+    mov     x0, sp
+    mov     x1, #0
+    bl      ArchUnexpectedException
+
+el1h_sync:
+    SAVE_FRAME 1
+    mov     x0, sp
+    mov     x1, #0
+    bl      ArchSyncException
+    RESTORE_FRAME 1
+
+el1h_irq:
+    SAVE_FRAME 1
+    mov     x0, sp
+    mov     x1, #0
+    bl      ArchIrqException
+    RESTORE_FRAME 1
+
+el1h_fiq:
+    SAVE_FRAME 1
+    mov     x0, sp
+    mov     x1, #1
+    bl      ArchUnexpectedException
+
+el1h_serror:
+    SAVE_FRAME 1
+    mov     x0, sp
+    mov     x1, #2
+    bl      ArchUnexpectedException
+
+/* ---- Exceptions from user mode ---- */
+
+el0_sync:
+    SAVE_FRAME 0
+    lsr     x9, x24, #26            /* ESR_EL1.EC */
+    cmp     x9, #0x15               /* SVC from AArch64 */
+    b.ne    1f
+
+    msr     daifclr, #2
+    mov     x0, sp
+    bl      SyscallDispatch
+    str     x0, [sp, #0]
+    bl      SchedExitIfKilled
+    RESTORE_FRAME 0
+
+1:  mov     x0, sp
+    mov     x1, #1
+    bl      ArchSyncException
+    RESTORE_FRAME 0
+
+el0_irq:
+    SAVE_FRAME 0
+    mov     x0, sp
+    mov     x1, #1
+    bl      ArchIrqException
+    RESTORE_FRAME 0
+
+el0_fiq:
+    SAVE_FRAME 0
+    mov     x0, sp
+    mov     x1, #1
+    bl      ArchUnexpectedException
+
+el0_serror:
+    SAVE_FRAME 0
+    mov     x0, sp
+    mov     x1, #2
+    bl      ArchUnexpectedException
+
+el0_aarch32:
+    SAVE_FRAME 0
+    mov     x0, sp
+    mov     x1, #3
+    bl      ArchUnexpectedException

kernel/src/Arch/x86_64/Abi.hpp

@@ -0,0 +1,62 @@
+/*
+    * Abi.hpp
+    * x86_64 user-mode ABI details (implements Arch/Abi.hpp)
+    * Copyright (c) 2026 Daniel Hammer
+*/
+
+#pragma once
+#include <cstdint>
+
+// UEFI uses the Microsoft x64 calling convention.
+#define ARCH_EFIAPI __attribute__((__ms_abi__))
+
+namespace Arch {
+
+    constexpr uint16_t ElfMachine = 62;  // EM_X86_64
+    constexpr const char* ElfMachineName = "x86_64";
+
+    inline RelocationKind ClassifyRelocation(uint32_t type) {
+        switch (type) {
+            case 0:  return RelocationKind::None;
+            case 8:  return RelocationKind::Relative;  // R_X86_64_RELATIVE
+            case 1:                                    // R_X86_64_64
+            case 6:                                    // R_X86_64_GLOB_DAT
+            case 7:  return RelocationKind::Symbol;    // R_X86_64_JUMP_SLOT
+            default: return RelocationKind::Unsupported;
+        }
+    }
+
+    // Variant II. The linker addresses TLS variables at
+    // TP - align_up(memSize, p_align) + offset, so the template must start
+    // exactly that far below TP; rounding p_align up first (as an older
+    // version did) shifts every variable away from its initial value.
+    // TP itself is kept at least 16-byte aligned, which the gap between
+    // base and the template absorbs. The 16 bytes above TP are the TCB.
+    inline TlsLayout ComputeTlsLayout(uint64_t memSize, uint64_t pAlign) {
+        uint64_t align = pAlign ? pAlign : 1;
+        uint64_t block = (memSize + align - 1) & ~(align - 1);
+        uint64_t tpAlign = align < 16 ? 16 : align;
+        uint64_t tp = (block + tpAlign - 1) & ~(tpAlign - 1);
+        return {tp - block, tp, tp + 16};
+    }
+
+    inline void WriteExitStub(uint8_t* stub) {
+        stub[0] = 0x31; stub[1] = 0xFF;  // xor edi, edi  (exit code 0)
+        stub[2] = 0x31; stub[3] = 0xC0;  // xor eax, eax  (SYS_EXIT = 0)
+        stub[4] = 0x0F; stub[5] = 0x05;  // syscall
+    }
+
+    // The entry function is entered as if CALLed: the return address sits
+    // at [RSP], with RSP = 16n - 8.
+    inline uint64_t SetupUserEntryStack(uint8_t* topPage, uint64_t stackTop,
+                                        uint64_t returnAddress) {
+        *(uint64_t*)(topPage + 0xFF8) = returnAddress;
+        return stackTop - 8;
+    }
+
+    // SystemV alignment as if CALLed (16n - 8); the slot is never read.
+    inline uint64_t ThreadEntryStackPointer(uint64_t stackTop) {
+        return (stackTop & ~0xFULL) - 8;
+    }
+
+}

kernel/src/Arch/x86_64/Acpi/AcpiSleep.cpp

@@ -4,8 +4,10 @@
     * Copyright (c) 2026 Daniel Hammer
 */
 
-#include "AcpiSleep.hpp"
+#include <ACPI/AcpiSleep.hpp>
-#include "AcpiEvents.hpp"
+#include <ACPI/AcpiEvents.hpp>
+#include <Arch/Cpu.hpp>
+#include <Arch/x86_64/Acpi/S3State.hpp>
 #include <ACPI/FADT.hpp>
 #include <ACPI/AML/AmlParser.hpp>
 #include <ACPI/AML/AmlInterpreter.hpp>
@@ -14,18 +16,19 @@
 #include <CppLib/Stream.hpp>
 #include <Memory/HHDM.hpp>
 #include <Memory/Paging.hpp>
-#include <Hal/GDT.hpp>
+#include <Arch/x86_64/GDT.hpp>
-#include <Hal/Apic/Apic.hpp>
+#include <Arch/x86_64/Apic/Apic.hpp>
-#include <Hal/Apic/IoApic.hpp>
+#include <Arch/x86_64/Apic/IoApic.hpp>
-#include <Hal/Apic/Pic.hpp>
+#include <Arch/x86_64/Apic/Pic.hpp>
-#include <Timekeeping/ApicTimer.hpp>
+#include <Timekeeping/Timer.hpp>
-#include <Hal/CpuPower.hpp>
+#include <Arch/CpuPower.hpp>
 #include <Drivers/Graphics/IntelGPU.hpp>
+#include <Drivers/Graphics/DgpuPower.hpp>
 #include <Drivers/PS2/PS2Controller.hpp>
 #include <Graphics/Framebuffer.hpp>
 #include <Libraries/Memory.hpp>
-#include <Hal/MSR.hpp>
+#include <Arch/x86_64/MSR.hpp>
-#include <Hal/SmpBoot.hpp>
+#include <Arch/Smp.hpp>
 #include <Api/Syscall.hpp>
 
 using namespace Kt;
@@ -128,9 +131,9 @@ namespace Hal {
                     const char* stepNames[] = {
                         "?", "GDT", "TSS", "Syscalls", "PAT", "PIC",
                         "LocalAPIC", "IoAPIC", "APICTimer", "PM1Clear",
-                        "WAK", "PS2", "IntelGPU", "sti"
+                        "WAK", "PS2", "IntelGPU", "dGPU", "sti"
                     };
-                    const char* stepName = reinitStep <= 0x0D ? stepNames[reinitStep] : "?";
+                    const char* stepName = reinitStep <= 0x0E ? stepNames[reinitStep] : "?";
                     KernelLogStream(WARNING, "S3") << "Reinit stopped after step 0x"
                         << base::hex << (uint64_t)reinitStep << " (" << stepName
                         << ") - next step crashed/hung";
@@ -185,7 +188,7 @@ namespace Hal {
                     << base::hex << (uint64_t)g_s3SlpTypA
                     << " SLP_TYPb=" << base::hex << (uint64_t)g_s3SlpTypB << ")";
                 KernelLogStream(INFO, "S3") << "Kernel PML4 phys = " << base::hex
-                    << (uint64_t)Memory::VMM::g_paging->PML4;
+                    << Memory::VMM::g_paging->KernelRoot();
                 KernelLogStream(INFO, "S3") << "FACS at phys " << base::hex
                     << (uint64_t)fadt.FacsAddress << " version=" << base::dec
                     << (uint64_t)g_facs->Version << " flags=" << base::hex
@@ -313,7 +316,7 @@ namespace Hal {
             // shadow copy of the PML4 at a fixed low-memory address.
             // Only the PML4 page itself needs to be below 4GB — its entries
             // contain full 52-bit physical addresses for sub-tables.
-            uint64_t kernelPml4Phys = (uint64_t)Memory::VMM::g_paging->PML4;
+            uint64_t kernelPml4Phys = Memory::VMM::g_paging->KernelRoot();
             uint8_t* kernelPml4Virt = (uint8_t*)Memory::HHDM(kernelPml4Phys);
             uint8_t* shadowPml4Virt = (uint8_t*)Memory::HHDM((uint64_t)SHADOW_PML4_PHYS);
             memcpy(shadowPml4Virt, kernelPml4Virt, 4096);
@@ -438,7 +441,7 @@ namespace Hal {
             Hal::InitializePAT();
             // HWP enable and POWER_CTL reset across S3; re-apply them and
             // bump the policy epoch so APs re-apply on their next tick.
-            Hal::CpuPower::ReapplyAfterWake();
+            Arch::CpuPower::ReapplyAfterWake();
             reinitProgress(0x05);  // PIC
             Hal::DisableLegacyPic();
             reinitProgress(0x06);  // Local APIC
@@ -446,7 +449,7 @@ namespace Hal {
             reinitProgress(0x07);  // IO APIC
             Hal::IoApic::Reinitialize();
             reinitProgress(0x08);  // APIC Timer
-            Timekeeping::ApicTimerReinitialize();
+            Timekeeping::TimerReinitialize();
             reinitProgress(0x09);  // PM1 clear
             ClearPM1Status();
             reinitProgress(0x0A);  // _WAK
@@ -459,8 +462,10 @@ namespace Hal {
             if (Drivers::Graphics::IntelGPU::IsInitialized()) {
                 Drivers::Graphics::IntelGPU::Reinitialize();
             }
-            reinitProgress(0x0D);  // sti
+            reinitProgress(0x0D);  // discrete GPU back off
-            asm volatile("sti");
+            Drivers::Graphics::DgpuPower::Resume();
+            reinitProgress(0x0E);  // sti
+            Arch::EnableInterrupts();
 
             KernelLogStream(OK, "S3") << "Resumed from S3 suspend";
 
@@ -514,7 +519,7 @@ namespace Hal {
             Io::Out8(0x00, 0x71);
 
             // 3. Disable interrupts
-            asm volatile("cli");
+            Arch::DisableInterrupts();
 
             // 4. Identity-map the trampoline page so the CPU can execute it
             //    after paging is re-enabled with our CR3 during wake.
@@ -522,7 +527,7 @@ namespace Hal {
             Memory::VMM::g_paging->Map(TRAMPOLINE_PHYS, TRAMPOLINE_PHYS);
 
             // 5. Flush all caches
-            asm volatile("wbinvd");
+            Arch::WritebackInvalidateCaches();
 
             // 6. Set the waking vector in FACS
             SetWakingVector();
@@ -545,7 +550,7 @@ namespace Hal {
             }
 
             // 10. Flush caches again
-            asm volatile("wbinvd");
+            Arch::WritebackInvalidateCaches();
 
             // 11. Assert SLP_EN — this triggers S3 entry
             Io::Out16(pm1a | PM1_SLP_EN, (uint16_t)g_pm1aControlBlock);
@@ -560,7 +565,7 @@ namespace Hal {
             // on QEMU/KVM the vCPU must reach HLT before QEMU finalizes S3.
             // On resume, firmware jumps to our waking vector.
             for (;;) {
-                asm volatile("hlt");
+                Arch::Halt();
                 // If we wake from HLT (spurious or real wake), check WAK_STS
                 if (g_pm1aEventBlock != 0) {
                     uint16_t sts = Io::In16((uint16_t)g_pm1aEventBlock);
@@ -571,7 +576,7 @@ namespace Hal {
             // If we get here, either S3 failed or we woke immediately.
             // The normal wake path goes through AcpiResumeLongMode instead.
             // Re-enable interrupts as a safety measure.
-            asm volatile("sti");
+            Arch::EnableInterrupts();
 
             KernelLogStream(WARNING, "S3") << "S3 entry may have failed (fell through)";
             return 0;

kernel/src/Arch/x86_64/Acpi/CpuIdle.cpp

@@ -5,10 +5,11 @@
 */
 
 #include "CpuIdle.hpp"
+#include <Arch/Cpu.hpp>
 #include <ACPI/FADT.hpp>
 #include <ACPI/AML/AmlInterpreter.hpp>
-#include <Hal/Cpu.hpp>
+#include <Arch/x86_64/CpuFeatures.hpp>
-#include <Hal/SmpBoot.hpp>
+#include <Arch/Smp.hpp>
 #include <Io/IoPort.hpp>
 #include <Terminal/Terminal.hpp>
 #include <CppLib/Stream.hpp>
@@ -249,13 +250,17 @@ namespace Hal {
         }
 
         static bool ParseCstObject(const char* path, const AML::Object& obj) {
-            if (obj.Type != AML::ObjectType::Package || obj.Buffer.Length < 2) {
+            if (obj.Type != AML::ObjectType::Package || !obj.Package.Data ||
+                obj.Package.Length < 1) {
                 return false;
             }
 
-            uint32_t pos = 1; // package element count byte
+            // Package.Data starts at the first element (the state count).
+            const uint8_t* data = obj.Package.Data;
+            uint32_t length = obj.Package.Length;
+            uint32_t pos = 0;
             uint64_t declaredStateCount = 0;
-            if (!DecodeInteger(obj.Buffer.Data, obj.Buffer.Length, pos, declaredStateCount)) {
+            if (!DecodeInteger(data, length, pos, declaredStateCount)) {
                 return false;
             }
 
@@ -263,7 +268,7 @@ namespace Hal {
             for (uint64_t i = 0; i < declaredStateCount && parsedCount < MaxIdleStates; i++) {
                 const uint8_t* nestedData = nullptr;
                 uint32_t nestedLength = 0;
-                if (!DecodePackageObject(obj.Buffer.Data, obj.Buffer.Length, pos, nestedData, nestedLength)) {
+                if (!DecodePackageObject(data, length, pos, nestedData, nestedLength)) {
                     break;
                 }
 
@@ -310,7 +315,10 @@ namespace Hal {
             auto& ns = interp.GetNamespace();
             char path[256];
 
-            for (int32_t nodeIndex = 0; nodeIndex < ns.NodeCount(); nodeIndex++) {
+            // DSDT processors only, for the same reason as the _CST scan.
+            int32_t limit = interp.DsdtNodeCount();
+            if (limit == 0) limit = ns.NodeCount();
+            for (int32_t nodeIndex = 0; nodeIndex < limit; nodeIndex++) {
                 auto* node = ns.GetNode(nodeIndex);
                 if (node == nullptr || node->Obj.Type != AML::ObjectType::Processor) {
                     continue;
@@ -473,7 +481,7 @@ namespace Hal {
             if (hasMwait) {
                 Hal::IdleWaitWithInterruptsDisabled(addr, CurrentIdleHint());
             } else {
-                Hal::HaltWithInterruptsDisabled();
+                Arch::HaltWithInterruptsDisabled();
             }
         }
 
@@ -522,7 +530,13 @@ namespace Hal {
             auto& ns = interp.GetNamespace();
             char path[256];
 
-            for (int32_t nodeIndex = 0; nodeIndex < ns.NodeCount(); nodeIndex++) {
+            // DSDT objects only. SSDTs are loaded too now (for device power
+            // methods), and many carry processor _CST tables; those C-state
+            // lists have never been exercised against the deep-idle timer
+            // behaviour on real laptops, so they stay out until they have.
+            int32_t limit = interp.DsdtNodeCount();
+            if (limit == 0) limit = ns.NodeCount();
+            for (int32_t nodeIndex = 0; nodeIndex < limit; nodeIndex++) {
                 auto* node = ns.GetNode(nodeIndex);
                 if (node == nullptr) continue;
                 if (!IsName(node->Name, "_CST")) continue;