New thread_self=42 syscall backs runtime.Thread.getCurrentId (the calling
thread's kernel task id). thread-test gains an id mode: two workers read
getCurrentId and the main thread confirms all three ids are non-zero and
distinct. Per-thread threadlocal TLS is deferred by design (no consumer; it would
need context-switched fs.base for an unused feature), as are RwLock/WaitGroup.
Marks the threading feature built (M1-M6): threading.md + docs/README.md status
updated, all thread-* cases wired into qemu_test.py.
Gate thread-id PASS; full suite green: 21/21 (thread-spawn/join/futex/mutex/id,
aspace-refcount, + 15 guardrail cases), zig build clean, zig build test green.
runtime.Thread.Mutex is the classic three-state futex mutex (unlocked/locked/
contended): the fast path is a single CAS and only a contended lock enters the
kernel. Condition is a futex sequence counter (wait/timedWait/signal/broadcast,
spurious wakeups allowed, use in a predicate loop); a signal racing the unlock
bumps the seq so it is never missed. Semaphore is permits guarded by
Mutex+Condition. All mirror std.Thread's shapes, ported onto runtime.Thread.Futex.
thread-test gains a mutex mode: 2 producers + 2 consumers move 2000 unique items
through an 8-slot ring (small enough that both sides block); the consumed
checksum and tally match exactly, proving the lock and condvars correct under
real cross-core contention.
Deferred with rationale (see docs/threading-plan.md): migrating join to a futex
completion word needs kernel clear-on-exit (else use-after-free munmapping a live
stack); host unit tests need a mockable Futex seam.
Gate thread-mutex PASS (3x); 17 guardrail/thread cases green; build + host tests
clean.
New private syscalls futex_wait(addr, expected, timeout_ns)=40 and
futex_wake(addr, count)=41. A waiter is a .blocked task tagged with
Task.futex_addr (no queue linkage); futex_wait reads the user word under the big
lock and parks only if it still equals expected, so a concurrent wake can't slip
between the check and the block. futex_wake scans the task table and readies up
to count waiters in the same address space. A timed wait also sets wake_at so the
existing wakeExpired times it out; futex_addr staying non-zero (only futex_wake
clears it) distinguishes timeout from a real wake. Waiters park in-kernel, so an
idle core still halts (no busy-wait).
runtime.Thread.Futex mirrors std.Thread.Futex (wait/timedWait/wake). thread-test
gains a futex mode: a waiter parks, the main thread wakes it (serial order
waiting/waking/woke, asserted by the case regex), and timedWait reports a
timeout.
Gate thread-futex PASS (3x); 18 guardrail cases green incl. sleep/event/ipc
blocking paths; build + host tests clean.
thread_spawn takes a 4th arg, an exit-endpoint handle: spawnThreadSupervised
resolves and refcounts it under the spawn lock (like spawnProcessSupervised), so
a thread's death posts a child-exit notification carrying its tid. runtime
Thread.join blocks in replyWait on that (private) endpoint for its tid, then
munmaps the stack; detach relinquishes the join (stack reclaimed at process
exit, for now). New current_core=39 syscall + Thread.currentCore() lets a worker
observe which core it ran on.
The closure now lives at the top of the thread's own (private) stack instead of
the heap, so spawn/join never touch the not-yet-thread-safe runtime heap.
thread-test gains a join mode: 4 workers x 100k atomic increments, joined, with
counter == N*K and >1 core stamped (real parallelism), plus a detached worker.
Gate thread-join PASS (4x, non-flaky); 17 guardrail/M1/M2 cases green; build +
host tests clean.
A thread is a task sharing the caller's address space. New private syscalls
thread_spawn(entry, stack_top, arg)=37 and thread_exit=38: thread_spawn goes
through scheduler.spawnThread (retains the shared aspace), thread_exit ends the
task like a process exit(0) (terminateCurrent -> releaseAspace, so the space
survives while siblings hold it). The closure pointer reaches the new thread in
rdi via a new jump_to_user_arg asm path and a per-task user_arg (0 for a normal
process, whose _start ignores it) - so the runtime trampoline is a plain C-ABI
Zig function, no naked asm.
runtime.Thread (library/runtime/thread.zig) mirrors std.Thread.spawn: mmap a
stack, heap-allocate the args closure, hand the kernel the trampoline + closure.
addThreadedUserBinary opts a binary into single_threaded=false; thread-test is
the first, and proves a worker runs in the shared address space via a shared
global the main thread polls.
Gate thread-spawn PASS; 16 guardrail cases green (incl. args/init/process on the
new jump_to_user_arg path) + aspace-refcount; build + host tests clean.
Route address-space lifetime through a refcount keyed by the page-table root
(scheduler.zig aspace_refs): retainAspace on the spawnUserLocked success path,
releaseAspace from both teardown paths (exitUserLocked, destroyTaskLocked),
destroying the space only when the last task on it exits. Behaviour is identical
today (every space has exactly one task); this is the foundation shared-address-
space threads (docs/threading.md) build on.
Test-observable liveAspaceCount/aspaceDestroyCount + a new aspace-refcount kernel
self-test and QEMU case: spawn and reap 5 ring-3 probes, assert live spaces return
to baseline and destructions advance by exactly 5 (destroyed once each, no leak,
no double-free). Gate passes; 13 guardrail cases green; build + host tests clean.
The kernel enabled SSE at boot and both kernel and userspace keep live values in
XMM (LLVM emits movdqu/movaps for >=16-byte struct copies, plus floats and SIMD),
yet the kernel never saved the SSE/FPU register file anywhere — not across
switch_context, not across the syscall boundary, not across interrupts. Any value
the compiler parked in an XMM register across a kernel entry could be silently
clobbered by kernel code, or by whatever the scheduler ran while the task blocked:
a whole-kernel, timing-dependent data-corruption bug.
It was the real root cause of the "device_enumerate corruption" Heisenbug: the
display service read garbage framebuffer geometry (height=0) because findDisplay
held the 16-byte .display field live in xmm0 across the claim/mmio_map syscalls,
and a timer preemption to the busy device-manager clobbered it. The tell that it
was register-only: memory always read correct, and the bug vanished whenever an
added syscall spilled the value to the stack.
Fix: fxsave/fxrstor the register file in the asm stubs (isr_common and
syscall_entry), right after pushing the GP trap frame — before any Zig kernel code
can touch XMM — and right before the pops. rbx bridges the exact rsp across the
call to interruptDispatch: it is callee-saved, so it survives even a blocking
dispatch that context-switches away and back, and `and $-16,%rsp; sub $512,%rsp`
gives fxsave its 16-byte-aligned scratch on the kernel stack. Context-switch-time
save/restore alone is not enough — kernel code between the interrupt and switchTo
already clobbers XMM.
The commit-58927ed Gop.init workaround (copy scalar geometry fields rather than the
whole descriptor by value) is now redundant but harmless; left in place. Deferred:
a fresh task inherits the previous task's XMM (minor info-leak / nondeterminism),
and the fxsave runs on every interrupt including ring0->ring0.
pciScanTest put three [64]DeviceDescriptor arrays on the stack; at 328
bytes each that is a 66,096-byte frame, larger than the 64 KiB bootstrap
stack the boot context runs on, so the prologue overflowed the stack and
triple-faulted on entry before the test could print anything — the CPU
reset with no exception line. Reuse one descriptor buffer (frame ~21 KiB).
With the fault gone, the restart-drill checks proved flaky. They polled
process.write_buffer (which holds only the latest write-syscall message)
for the transient restarting/re-scan lines, which the concurrent
class-driver output overwrote before the starved low-priority poller could
observe them. The kernel broker count can't witness the restart either:
register is idempotent and the table has no unregister, so the count is
invariant across kill/prune/respawn.
Follow the sibling restart drills (usbReportTest/driverRestartTest): assert
the ordered drill in the harness regex over the full serial log (a
backreference requires the respawn to re-scan the same count), and keep only
race-free broker checks in the kernel — empty before the scan, populated
once it settles, never growing past N — sampled via scheduler.sleep at a
fixed cadence instead of a busy-poll.
The compositor now survives the virtio-gpu driver dying and re-attaches when device-manager
restarts it — the last piece of display v2.
Three parts:
- The driver hellos the device manager (role: bus). It never did, so the manager — which
spawns it supervised and expects a hello — was stopping it at the 3s hello deadline every
run (the gate markers just printed first). Now it is properly supervised: not stopped for
silence, and restarted on death.
- A kernel IPC fix so a call to a dead service errors instead of hanging. An endpoint records
its owner; when that task dies, its registered endpoints are marked dead (and any parked
senders woken with -EPEER), so ipc_call returns -EPEER rather than blocking on a reply that
will never come. Without this the compositor's first present after the driver died blocked
forever. General robustness — any client of any service benefits.
- The compositor re-attaches. Its .scanout calls now fail cleanly (caught), freezing the last
frame; when the restarted driver re-announces, attach_scanout detects the backend is already
virtio and logs "scanout re-attached", mapping the fresh shared surface and re-looking-up
.scanout. (The previous shm mapping leaks — no shm_unmap syscall yet — but its frames are the
dead driver's, reclaimed on exit.)
- device-manager gains a "test-scanout-restart" mode (like test-usb-restart) that kills the
virtio-gpu driver once after it hellos; the displayReattachTest kernel scenario drives it.
Gate: python3 test/qemu_test.py display-reattach — "scanout upgraded to virtio-gpu" then
"scanout re-attached", no CPU exception, passing 3/3. host tests, ipc/ipc-call/ipc-cap,
supervision, shm, display-service, display-demo, virtio-gpu, display-native, and
display-modeset all pass; default zig build clean. v2 (V1-V6) complete.
The compositor now boots on the GOP framebuffer and upgrades to the virtio-gpu driver
the moment it announces itself — the pluggable-scanout payoff.
The shared surface. The scanout resource is an shm region the driver creates
(shm_physical, a new syscall, hands it the guest-physical for attach_backing) and passes
to the compositor as a capability. The compositor maps it and composites straight into
it: on x86 DMA is cache-coherent, so the cacheable shared pages the CPU paints are exactly
what the device transfers-and-flushes — no copy, no explicit flush.
The handshake. After bring-up the driver looks up .display and sends attach_scanout with
the geometry + the surface capability. The compositor maps the surface, looks up the
driver's .scanout endpoint itself (the driver registered it — no need to pass it), switches
to backend.VirtioGpu, and re-composites the current frame. present() over the native
backend is a present request on .scanout -> transfer-to-host + resource flush. The first
native present is deferred to a one-shot timer: presenting inline from the announce handler
would deadlock, since the driver is still blocked on our reply and not yet serving .scanout.
After it lands, the compositor reads a pixel back from the shared surface to confirm the
frame reached the device's backing.
- shm_physical (syscall 36) + runtime.shm.physical.
- scanout-protocol (the compositor->driver present channel), separate from the
client-facing display protocol; the display protocol gains attach_scanout.
- backend.VirtioGpu joins backend.Gop in the tagged union; select() still boots GOP.
- the virtio-gpu driver's scanout backing is now shm (was DMA); it announces + serves
.scanout present requests (transfer-to-host + flush of the shared surface).
Also fixes a latent framebuffer-geometry corruption the display service hit only when it
enumerated the device tree alongside a busy device-manager: Gop.init now captures the
geometry into a small value the instant device_enumerate returns (rather than re-reading
the 328-byte descriptor across the later claim/mmio_map syscalls) and retries on a zero
geometry. The underlying device-table clobber is a separate kernel bug, tracked apart.
Gate: python3 test/qemu_test.py display-native (QEMU -device virtio-gpu-pci) — "display:
scanout upgraded to virtio-gpu" + "display: native present verified" + "display-demo: ok",
passing 3/3. host tests, display-service, display-demo, shm, and virtio-gpu still pass.
The first native scanout backend's driver half. The device manager matches the
virtio-gpu PCI function by its Display/Other class triple and spawns the driver,
which claims the device, confirms vendor 0x1AF4/device 0x1050 from config space,
enables memory-space + bus mastering, and walks the virtio vendor capabilities to
find the common-config and notify structures in its BAR.
From there it is the standard modern-virtio bring-up: reset, negotiate VERSION_1,
stand up the control virtqueue in coherent DMA, then drive the GPU end to end —
RESOURCE_CREATE_2D, ATTACH_BACKING (a coherent DMA buffer; V4 swaps in the shm-
shared surface), SET_SCANOUT, paint a known pattern, TRANSFER_TO_HOST_2D,
RESOURCE_FLUSH, and wait for the device's used-ring ack. Reading the backing back
proves it is CPU-visible RAM; the ack proves the device consumed the frame —
together the automated stand-in for "it's on screen", no screenshot.
- system/drivers/virtio-gpu/: the driver, plus virtio-gpu-protocol.zig (control
commands) and virtio-pci.zig (the 1.0 PCI transport + split-virtqueue), both with
host-tested struct sizes.
- device-manager matches the display/other class triple to "virtio-gpu"; the driver
self-confirms the vendor/device id, since the class alone cannot distinguish it.
- ServiceId.scanout (11): the driver registers it so the compositor finds it in V4.
Gate: python3 test/qemu_test.py virtio-gpu (QEMU -device virtio-gpu-pci) — the
device-manager stack discovers the function, the driver brings up a 640x480 scanout
and flushes a test pattern: "virtio-gpu: scanout 640x480 online" + "flush acked,
pixel check ok". host tests, display-service, shm, and device-list still pass.
Note: the pre-existing pci-scan case triple-faults on main (verified at 88ad432,
before this change); it is unrelated and tracked separately.
Generalize capability passing from endpoints to memory objects. The per-task
handle table now holds kind-tagged entries (scheduler.HandleObject{kind, ptr});
closeHandles and shareCapability dispatch by kind, so a shared-memory object
rides an ipc_call send_cap exactly like an endpoint and is refcount-freed only
when its last capability drops.
- shm_create(len) -> vaddr, handle: contiguous, zeroed, cacheable frames wrapped
in a refcounted ShmObject, mapped into the caller's shm arena (PML4[230]).
- shm_map(cap) -> vaddr: map the same physical pages into a receiver that got the
capability. mapUserSharedInto maps WB-cacheable + device_grant, so a sharer's
teardown never frees the shared frames — the object owns them.
- runtime.shm: create(len) -> Region{ptr, handle, len}, map(handle) -> ptr.
Gate: qemu_test.py shm — shm-client creates a region, writes a pattern, passes
its capability to shm-server, which maps it and reads the same bytes back
(shm: shared 4096 bytes ok). ipc/ipc-call/ipc-cap/supervision/dma/usermem/
display-service and host tests all still pass — the handle change broke no IPC.
Two gaps a real-hardware crash exposed, both in onException's terminal path (and
the panic path):
- The report was anonymous. Add the faulting task's id + name and whether it
trapped in ring 3 (a user process) or ring 0 (the trusted base) — so a fatal
fault says WHAT crashed and WHERE, not just the vector. scheduler gains
currentIdSafe/currentNameSafe (early-boot-guarded, like currentCpuIndex, so the
reporter can't fault a second time).
- halt() never released the big kernel lock, so a core that died holding it
deadlocked every other core spinning in acquire() — the whole machine hangs,
not just the one core the design promises. The BKL now records its owner
(architecture.cpuLocal(), a unique per-core token); sync.releaseIfHeldHere()
frees the lock only if this core holds it, called before halt on both fatal
paths. Caveat: if we held it mid-mutation the shared state may be inconsistent,
but letting the other cores + the supervisor keep running is strictly more
recoverable than a guaranteed total hang.
Now that the user-space display service owns the framebuffer, the bootstrap
console shrinks to fatal-only. status/statusPrint go to the diagnostic log
alone, so routine boot output no longer scribbles on a screen the compositor is
about to paint — and a driver's recoverable fault report stays off it too. A new
fatal/fatalPrint keeps panics and kernel-mode faults on screen, forcing the
console back on so a dying machine's last words show even over a live display.
console.zig now documents its early-boot + fatal-fallback role.
Drive the compositor from a separate process, proving the pipeline end to end.
- runtime.display: a Layer handle (fill / blitTile / configure / damage /
destroy), createLayer, and a color(r,g,b) helper that caches the mode and
packs via protocol.pack. Coordinates are signed over the u32 wire fields
(@bitCast both ways), so a layer may sit or move partly off-screen.
- system/services/display-demo: the input-source analog for the compositor — a
full-screen wallpaper, a rectangle it slides back and forth (moved by
configure each frame, so the damage-driven present repaints old + new), and a
cursor. Presents in a loop paced by runtime.time. Wired into build + initrd.
Fix this surfaced: protocol.message_maximum was 4096, but the kernel caps every
IPC message at MESSAGE_MAXIMUM = 256, so replyWait rejected the oversized receive
buffer with -E2BIG and the serve loop had been spinning since D2 (invisibly, as
those gates matched init-time heartbeats). Set it to 256; blit_tile is now
explicitly a small-tile path (larger bitmaps are the deferred shm surface).
Gate: `python3 test/qemu_test.py display-demo` — the demo drives frames of
motion through the layer client API and logs `display-demo: ok`. Regression:
zig build test, display (D1), display-service (D2/D3), and default zig build.
Stand up /system/services/display: a ring-3 process that claims the framebuffer
D1 seeded, maps it write-combining as the front buffer, allocates a cacheable
back buffer, and presents composed frames. The GUI track's compositor, reached
by name over ServiceId.display (= 9).
- protocol.zig: the display wire protocol (info/create_layer/configure_layer/
destroy_layer/fill_rect/blit_tile/damage/present). `info` and a whole-screen
`present` are live; the layer ops fail-stub until D3.
- display.zig: enumerate -> claim -> mmio_map(WC) the LFB, mmap a cacheable back
buffer, clear it and present it (proving the double-buffer path), then serve.
- runtime.display + barrel exports (display, display_protocol): a cached
`.display` client with info()/present(), the runtime.block shape.
- init spawns "display" in boot_services; build.zig wires the protocol onto the
runtime, builds the exe, packs it into the initial-ramdisk, installs it.
mmap fix the back buffer forced: systemMmap was capped at 256 pages (1 MiB) by a
fixed kernel-stack scratch array. Rewrote it to map page-by-page with rollback
(no array) and raised the cap to 8192 pages (32 MiB) — enough for a 4K back
buffer. A real limitation met.
Gate: `python3 test/qemu_test.py display-service` matches the service's own
serial heartbeats (display: online WxH / presented frame 0), printed only after
the full claim -> WC-map -> back-buffer -> present chain. Regression-checked
usermem, heap, init, and D1's display.
Kick off the display service track (docs/display.md, docs/display-plan.md): a
user-space compositor that owns the framebuffer. GOP and the PCI display device
are two views of one controller; GOP dies at ExitBootServices, so the portable
base is the boot-handoff linear framebuffer.
D1 makes that framebuffer reachable from user space over the existing device
claim/mmio_map path rather than a bespoke syscall:
- device-abi: a `display` DeviceClass, a DisplayInfo{w,h,pitch,format} on the
descriptor, and a flags field on resources with a write-combining bit.
- devices-broker: seedDisplay() publishes the loader's framebuffer as a
root-level `display` node (one WC-flagged memory resource); kmain seeds it
after discovery. displayDevice()/displayClaimed() track the claim.
- paging/mmio_map: mapUserDeviceInto gains a write_combining bool — a WC-flagged
resource maps through PAT entry 4 instead of strong-uncacheable (an
uncacheable framebuffer blit is glacial).
- console: falls silent while a display service holds the framebuffer, and is
forced back on by the panic/exception paths.
Gate: the `display` kernel test asserts the seeded node's shape and that the
claim + mmio_map leaf is genuinely write-combining (PAT bit set, PCD/PWT clear).
Regression-checked discovery/ioport/claim-release/supervision/device-list/
device-manager with the +1 device in the table.
Kernel discovery interpreted the whole DSDT/SSDTs (~0.5 MB -> ~9700 nodes) solely to extract the _S5 sleep type for a kernel-side soft-off — ~1-2s of work on the single-core, pre-scheduler critical path, with nothing to overlap it. The ring-3 acpi service already parses the same blobs and owns S5 end-to-end (reads pm1a_cnt from the FADT, writes SLP_TYP itself; orderly-shutdown tests it). So drop the kernel parse entirely.
- all *static*-table parsing stays (MADT/HPET/FADT/MCFG): CPUs, timers, PCIe, and the power register map are still detected from ACPI, not legacy/compat addresses (timer still calibrates via HPET/PM-timer/CPUID, PIT only as last resort)
- kernel keeps reboot (FADT reset register + legacy fallbacks — no AML); soft-off is userspace-only now
- removed PowerInformation.s5/s3, the kernel AML namespace, aml_stats, amlDeviceCount, and the aml import; power.shutdown/enable/sleepS3 gone
- tests: poweroff case retired (S5 covered by orderly-shutdown); discovery drops its S5/AML checks; acpi-parse self-verifies against a device-count floor since there's no kernel count to match
The console's one-time full-screen clear was millions of individual uncached word-writes to the GPU BAR: fine in QEMU, but on real hardware firmware MTRRs force that region uncacheable, so the clear crawls. Program PAT entry 4 to write-combining (setupPat, on the BSP and every AP) and map the framebuffer window with the PAT bit, so those writes batch into bursts.
The clear ran on the *loader's* uncached mapping because console.init happened before enablePaging. Move it to just after paging, so it uses the kernel's write-combining mapping instead. The cost: on-screen output is absent before paging (an early panic still lands in the serial/RAM log). Verified: 11 QEMU cases (incl. SMP AP-PAT, USB/FAT/ACPI) plus a screendump showing the framebuffer cleared to black with the status text rendered correctly.
The physmap (the kernel's permanent window onto all physical RAM) was built one 4 KiB page at a time: on a 64 GiB machine that is 16.7M mapPage calls and ~128 MiB of page tables (the bulk of the 'Kernel footprint' line). Map the 2 MiB-aligned interior with huge PD leaves instead — one entry per 2 MiB, no PT beneath — and only the unaligned head/tail with 4 KiB. 512x fewer entries and table frames; footprint at 8 GiB drops from ~16 MiB of tables to ~1 MiB total, and it scales.
translateIn/isExecutable now stop at a 2 MiB leaf, and descend() panics rather than walking through one (a 4 KiB map inside a huge page would corrupt RAM; the physmap and 4 KiB regions live in disjoint PML4 slots, so it can't happen — the guard just makes a bug loud). Verified: 20 QEMU cases (vmm/heap/wx/usermem/dma/ipc/smp/usb/fat/acpi/faults) green.
Serial is now a QEMU/dev aid, not a real-hardware necessity: a legacy-free board often has no live COM1, and the boot log is kept in RAM (klog) and flushed to disk. So the serial sink is compiled in only under -Dserial (default false).
- kernel.zig gates serialInit + the log sink on build_options.serial
- boot/efi.zig gates its EFI: progress breadcrumbs (con_out) via progress(); fatal-error messages stay always-on so a failed boot still explains itself
- run-x86-64 boots a serial-enabled image variant (factored addKernel/addBootImage helpers) so a dev boot always captures serial0, without baking serial into the flashable image
- test/qemu_test.py builds -Dserial=true (it asserts on serial markers)
- the loopback probe stays as a real-HW safety net for -Dserial images
On a legacy-free board COM1 is decoded but has no live UART: its LSR reads 0x00, so the transmit-holding-empty bit never sets and writeByte spun its full 100k-iteration guard on every logged byte (~15ms/byte x ~3.7k bytes to 'initialised' ~= 57s). QEMU always has a working UART, so this only bit real hardware; the boot log survived via log.ramSink.
- probe() loopback-tests the UART in init(); write() is a no-op when absent, so a dead port costs nothing per byte
- guard cut 100k -> 5k (backstop for a live-but-stalled UART only)
- serialPresent() + a boot line making the absent-UART case visible
Adds real (calendar) time, the foundation for filesystem timestamps. Monotonic
time (`clock`) says how long since boot; this says what time it actually is.
- cpu.zig (x86_64): readRtcUnixSeconds() reads the CMOS real-time clock (ports
0x70/0x71) — waits out an update-in-progress, reads twice until stable, handles
BCD-vs-binary and 12-vs-24-hour per status register B — and converts to Unix
epoch seconds (UTC).
- kernel/wall-clock.zig: reads the RTC once at boot and anchors it to the monotonic
clock, so a query is a cheap arithmetic offset — no per-call CMOS poll, no lock,
no SMP hazard on the shared ports. kmain calls init() once the monotonic clock is
final and logs the epoch.
- wall_clock() syscall (33) -> Unix epoch seconds, wrapped by runtime.system
.wallClock(). Wall-clock *seconds* are mechanism the kernel owns like the
monotonic clock; calendars/timezones are user-space policy (the stale comment on
systemClock that called wall-clock a "user-space service" is updated in spirit by
the new handler's doc).
- A `wall-clock` kernel test asserts the boot RTC read is a plausible current epoch.
Verified against the host: the guest read epoch 1783971244 while `date -u +%s` gave
1783971245 (one second of boot lag) — the CMOS read + epoch conversion are correct to
the second. zig build, zig build test, and smoke/clock/init are green.
On a headless or real board nothing captures serial, so the boot log — the whole
diagnostic stream — is lost at power-off. This retains it in the kernel and copies
it to the boot USB volume as /mnt/usb/DANOS.LOG, the on-disk equivalent of QEMU's
`-serial file:`. Pull the stick, read DANOS.LOG on another machine.
How it fits together:
- Kernel RAM sink (log.zig): a fixed 256 KiB in-image buffer registered as a log
sink in kmain, right after serial. Because userspace debug_write funnels through
log.write, it captures the entire stream — kernel lines and every service's
output — from the first line. Fills linearly and stops when full (earliest boot
output, the most valuable, is kept); no allocation, so it is panic-safe.
- klog_read syscall (32): copies that buffer out to a user buffer, the mirror of
debug_write — same overflow-safe user-half bounds check, kernel -> user copy,
under the kernel lock so the snapshot can't grow mid-copy. Wrapped by
runtime.system.klogRead.
- log-flush (new one-shot, in the initial-ramdisk): waits for the fat server to
mount /mnt/usb, then copies the whole log to /mnt/usb/DANOS.LOG. init spawns it
once the boot services are up (fire-and-forget; it polls the mount itself). If
no volume is mounted — no stick, or the no-VFS ramdisk sweep — it exits silently.
- init shutdown flush: init repeats the copy inline at the top of shutDown(),
BEFORE it tears down the storage services (the fat server is stopped first), so
a clean poweroff captures the fullest log while /mnt/usb is still writable.
- unistd.writeAll: loops write() past the 224-byte VFS payload cap; both flush
paths use it.
The filename is 8.3 (DANOS.LOG) at the mount root — the FAT short-name rule, and
there is no mkdir on the FAT path yet. Extend-only writes mean the two same-session
flushes never leave stale bytes (the shutdown log is a superset of the boot log);
a shorter log on a later boot of the same stick can leave a stale tail — a noted,
cosmetic limitation, not worth pulling O_TRUNC into the FAT write path for now.
Verified end to end under QEMU: a new `log-flush` case (reusing the orderly-shutdown
build) asserts both markers then S5, and DANOS.LOG is read back out of the image
afterwards (11804 bytes, containing the kernel init line, the FAT mount line, and
the boot flush marker). Regression stays green: zig build, zig build test, zig
build check-fat-image, and a sequential QEMU sweep — smoke, init, initial-ramdisk
(log-flush silent in the bare sweep), orderly-shutdown, fat-mount, usb-storage,
usb-hid, vfs, input, device-manager, process, signals, dma, fault-pf.
Add a FAT12/16/32 filesystem the VFS mounts at /mnt/usb, reading and writing a
USB stick through the block device. Verified end to end under QEMU: the fat
server mounts the volume, the VFS routes /mnt/usb to it, and a client lists the
root and reads a file (the ELF magic of /mnt/usb/system/kernel).
- engine.zig: the FAT engine over a BlockDevice interface — mount (a bare FAT or,
as QEMU's VVFAT and most real sticks present it, an MBR-partitioned disk), FAT
chain walk (12/16/32), cluster allocation, directory traversal with long-name
read, and file read / write / create. Host-tested against a RAM-backed FAT16
image (create, cluster-spanning write, mid-file overwrite, read-back, list).
- on-disk.zig: the align(1) boot-sector / directory / long-name / FSInfo structs
and the cluster-count FAT-type detection.
- fat.zig: the server — wraps the .block device (a DMA bounce buffer) in a
BlockDevice, mounts the FAT, serves the vfs-protocol as a backend, and mounts
itself into the VFS at /mnt/usb. Spawned by init as a boot service.
- runtime.block: the block-device client (geometry / read / write by physical
address, so whole sectors never cross IPC).
- Raise the kernel service-name registry (maximum_services) 8 -> 16: it is
indexed directly by ServiceId, and fat = 8 was being rejected, so the fat
server exited before registering.
- VFS: an absolute path with no matching mount is now not-found rather than
silently created in the flat ramfs — so /mnt/usb fails cleanly until mounted.
Tests: fat-mount (the full stack: block -> FAT -> VFS mount -> list + file read)
passes; host units cover the engine and on-disk structs; the vfs, shutdown, and
USB regression suite stays green (10/10).
Flesh out the xHCI host-controller driver into a full transfer engine and build
the three USB class drivers on top, all verified end to end under QEMU.
- xHCI engine (usb-xhci-library.zig): controller reset, command/event rings with
cycle-bit bookkeeping (gated on a No-Op-command proof), device slots, Address
Device, control transfers, full chapter-9 enumeration, Configure Endpoint, and
interrupt/bulk transfers. Each interface is device_registered with its
(class,subclass,protocol) identity, unique per (port,interface).
- Bus<->class transfer protocol (usb-transfer-protocol.zig + runtime.usb): open /
control / interrupt-subscribe (async report pump on a poll timer) / bulk-by-
physical-address, so sector data never crosses the 256-byte IPC limit.
- USB HID keyboard + mouse (usb-hid/): decode boot-protocol reports and publish
to the input service. A USB usage is already the input protocol's keycode.
- USB mass storage (usb-storage/): Bulk-Only Transport + transparent SCSI,
serving a block device under the new .block service id (block-protocol).
- device-manager matches USB interfaces to class drivers (usbDriverForIdentity).
- usb-abi / usb-ids made importable modules; add HID and mass-storage class
requests, packTriple, and a usb_device DeviceClass.
- Fix test/qemu_test.py on macOS: the QMP unix-socket path was built from the
deep worktree path and exceeded the 104-byte sun_path limit, so QEMU exited
before booting. It now lives under a short temp path.
Tests: usb-report, usb-hid, usb-storage pass under python3 test/qemu_test.py;
host units (usb-abi, usb-ids, hid-report, bulk-only-transport, scsi) green.
Time is a kernel concern in danos: the kernel owns the scheduling timer and
already exposes monotonic time via the clock/sleep/timer_bind syscalls, so a
userspace time service would be a redundant, slower path. This adds the generic
runtime.time module over those syscalls, retires the two demonstration drivers,
reorganizes the milestone docs, and makes the monotonic clock correct on Intel,
AMD, and inside any VM.
runtime.time (library/runtime/time.zig)
- Instant/Duration interface: now, sleep, spin, after, monotonicNanos, available
- a thin layer over system.clock/sleep/timerOnce; unit-tested arithmetic
Remove the demo drivers hpet and bus (a teaching example belongs in the docs,
not shipped in the tree)
- system/drivers/ now holds only real drivers: pci-bus, ps2-bus, usb-xhci-bus
- device-manager end-to-end test repointed to pci-bus (asserts on kernel state:
the process table and the device tree, not a racy serial marker)
- device_register containment moved to a new in-kernel `containment` test
- the driver-model worked example moved inline into docs/drivers.md
Reorganize milestone docs into topic docs
- m17-m18 / m19-m20 / m21 plans dissolved into process-lifecycle, device-manager,
discovery, and acpi docs; new docs/power.md and docs/timers.md; ~20 citations
repointed; plan docs deleted
TSC reliability (apic.zig, smp.zig, cpu.zig, kernel.zig)
- check the invariant-TSC bit (CPUID 0x80000007 EDX[8]) on Intel and AMD
- cross-core "warp" check at SMP bring-up, pairwise BSP<->AP as each core comes up
- fall back to the HPET clocksource when the TSC is not invariant (a bare VM) or
not synchronized (a warp), switched continuously so time never jumps
- boot log reports the outcome; new tsc-sync test exercises the TSC + warp path
Verified: zig build; zig build test; 60/60 QEMU cases (incl. new containment and
tsc-sync).
The QMP harness channel, the SCI + power button published from ring 3,
Notify/GPE dispatch in the AML interpreter, and orderly shutdown — init's
M17 stop cascade into a ring-3 S5 write. Proven by injecting a real ACPI
power-button event; QEMU powers off through the whole chain.
# Conflicts:
# system/services/init/init.zig
The debug-message refactor prefixed each service/driver line with its
source path (system/drivers/hpet:, ...) for readability, but two things
left main red: a 'drviers' typo in hpet.zig and pci-bus.zig, and five
kernel tests (init, hpet, pci-scan, device-manager, vfs-client-death)
that starts-with-matched the old short markers, which no longer sit at
the front of the prefixed lines.
Fix the typo, and convert the fragile starts-with matchers to substring
matching via a bufferHas helper — 'hpet: ok' now matches inside
'system/drivers/hpet: ok' regardless of prefix. Future-proof against
further prefix changes and harmless for the tests that already passed.
Suite 58/58.
The capstone. init becomes a real supervisor: it spawns its boot
services supervised against one endpoint that also carries its signals, a
re-arming heartbeat timer, and the power events it subscribes to. On the
power button (or a terminate signal — same path) it logs the shutdown,
runs the M17 stop sequence over its children in reverse spawn order
(vfs last), then asks the power service for S5.
The acpi service honors a shutdown request from a power subscriber — init
is the one subscriber, a soft gate that stands in for 'only the system
supervisor may power off' and, unlike a PID-1 check, survives the test
harness where the kernel's idle tasks take the early ids. The power
service is mechanism (write S5); deciding when to shut down and stopping
everything else first is init's policy — the microkernel split applied to
poweroff.
The orderly-shutdown scenario injects a real QMP power-button event and
watches the whole chain compose: button pressed -> init shutting down ->
entering S5 -> QEMU powers off. That single scenario proves the M17
lifecycle and the M21 event side compose into a clean shutdown. Suite
60/60.
The kernel publishes the FADT as one more acpi-tables memory resource
(tagged by its intact FACP header — the AML blobs are header-stripped);
the acpi service reads the PM1 event/control and GPE register ports from
that copy, so the kernel's own FADT parse is untouched. A power-protocol
module (ServiceId.power = 5, domain-named so an ARM PSCI service can serve
the same id) carries subscribe / shutdown / events.
The acpi service converts to runtime.service.run — device discovery, the
.power protocol, and the SCI notification all fold into one loop. At
startup it enables ACPI mode if SCI_EN is clear (the SMI dance), binds the
SCI (found as the node's len-1 irq resource, distinct from the broad
window), and sets PWRBTN_EN. On the SCI it reads PM1_STS, clears
PWRBTN_STS write-1, logs the press, publishes power_button to
subscribers, and always acks. The power-button scenario proves it with a
real QMP system_powerdown injected mid-run through the M21.0 channel.
The kernel no longer folds AML Device objects into the device tree — the
ring-3 acpi service is the sole builder of _HID device nodes. The kernel
keeps building the namespace only for the \_S5 sleep type, and still
seeds the static tables (MADT, HPET, MCFG, FADT) and the acpi-tables node.
The device manager matches ps2-bus from the service's _HID reports
(PNP0303 / PNP0F13, singleton-deduped) instead of boot-snapshot nodes;
its dead boot-snapshot ps2 arm is gone. The service registers every
device before reporting any, so a driver the manager spawns on the first
report already sees the full set — no keyboard-before-mouse race. The
acpi-ps2 scenario proves the whole chain: report -> spawn -> ps2-bus
finds the controller and attaches its keyboard, entirely in ring 3. The
ioport test moved to the acpi-tables I/O window, since the kernel-built
PS/2 node it used to scan for no longer exists. The retired
device-building functions in acpi.zig are dead but retained (a botched
mechanical deletion is worse mid-migration than a follow-up sweep, which
is flagged as a task). Suite 58/58.
AML method evaluation now runs in userspace touching real hardware: the
service builds an interpreter with a ring-3 Hal (port I/O routed through
its claimed acpi-tables node; a scratch page backs SystemMemory maps so a
stray OperationRegion degrades to zeros instead of faulting a process
that cannot map arbitrary physical memory). It walks the namespace and,
for each present _HID device that is not a PCI root, evaluates _CRS,
registers it under acpi-tables, and reports it with its EISA-decoded hid.
Containment for this needed the broker's irq check to become range-based
— an interrupt line is still indivisible, but a parent may own a range,
so the acpi-tables node's broad irq window contains its children's legacy
lines (a length-1 range is exactly the old equality, so single-irq
parents are unaffected). ChildAdded gained a hid field for firmware
string identity. Matching those reports to drivers stays off until M20.3,
so ps2-bus still comes up via the kernel path — no regression. The
acpi-report scenario proves the PS/2 keyboard (io 0x60/0x64 + IRQ) and
mouse (IRQ) are reported with their resources. Suite 57/57.
The AML module becomes a build module compiled into both the kernel (for
the \_S5 sleep state it still needs) and the new acpi service — one
source, two builds, no fork. The kernel publishes a single acpi-tables
node: the DSDT/SSDT blobs as memory resources, a broad io_port grant (the
honest trust boundary — firmware AML names whatever ports it chose, known
only after parsing), and the SCI for the M21 event track. The acpi
service claims the node, maps each blob through the ordinary mmio grant
(which preserves the sub-page offset onto the bytecode), and runs the
same parser the kernel does. It self-verifies its namespace Device count
against the kernel's — 34 = 34 — deterministically via an argv the
acpi-parse test passes, so no racing the shared serial buffer. Parse-only
touches no hardware; OperationRegion evaluation waits for _CRS/_STA in
M20.2. The manager spawns 'discovery' (the neutral ramdisk name) at
startup. Suite 56/56.
enumeratePci, addBars, pciConfigurationPtr, and the PciHeader struct are
deleted; the kernel seeds only the host bridge, and the ring-3 pci-bus
driver's reports are the sole source of PCI function nodes. The manager
matches PCI drivers from reported identity, deduped by registered device
id so a bus restart never double-spawns.
The flip did its job by exposing a latent SMP race: ring-3
device_register made the broker table concurrent for the first time, and
mmio_map read it lock-free — under load a torn resource length mapped
hpet's window wrong (its user fault) and underflowed r.len-1 into a
kernel integer-overflow panic. Fixed: the broker read in mmio_map (and
claim) runs under the big kernel lock, the arithmetic rejects
zero-length and wrapping windows cleanly, and pci-bus no longer registers
unimplemented size-0 BARs. driver-restart hammered 6x, suite 55/55.
Each function is registered under the bridge with the config-space slice
and BARs sized by the same all-ones probe the kernel uses — byte-for-byte
equal descriptors, so the idempotent register returns the kernel's
existing node ids during coexistence instead of duplicating the tree.
The bridge gained the 16-bit io_port aperture that functions' I/O BARs
need to pass containment. Reports carry the registered device_id, and
the pci-scan scenario drills a forced restart: kill the enumerator after
its reports, watch the respawn re-scan, and assert the broker's PCI node
count never grew. The usb-restart test trigger is pinned to the xHCI
reporter (pci-bus racing it to two reports used to steal the kill).
Harness hardening: failing cases preserve their serial logs; the heavy
scenarios run at 150s.
The manager matches the pci_host_bridge node and spawns pci-bus with the
bridge id as its assignment — hello, supervision, restart, all the M18
contract for free. The driver claims the bridge, maps the ECAM window
(resource 0) through the ordinary mmio grant, and repeats the kernel's
brute-force bus/device/function walk from user space. The pci-scan
scenario builds its expected marker from the kernel's own function count,
so the two enumerations must agree exactly — the equivalence that
licenses retiring the kernel walk in M19.3.
The host bridge now carries MMIO apertures derived from the boot memory
map's gaps below 4 GiB (largest three, sort-merged; a single after-the-
last-region hole dies on OVMF's flash at the top) plus one aperture above
the described space — so a user-space device_register of PCI functions
with BAR resources can pass containment. The discovery test asserts
every PCI memory resource lies inside a bridge window and names any
escapee. device_register is idempotent on exact (parent, class, identity,
resources) match — a restarted registering bus cannot duplicate its
children; proven directly against the broker in the bus test.
ChildAdded gains device_id so a report can carry the registered kernel
id a matched driver needs as its assignment.
Applications ask the device manager for the tree (enumerate: a header
plus ChildEntry records) and subscribe to published add/remove events by
handing their endpoint over as the call's capability — the input-service
pattern; events are the same ChildAdded/ChildRemoved structs the bus
drivers send, one encoding in both directions. device-list is the first
client: it prints the tree, subscribes, and narrates the events through
a driver restart. The protocol's message maximum is capped at the
kernel's IPC MESSAGE_MAXIMUM (256 bytes, ten entries per reply; paging
joins the protocol when a tree outgrows one message). The startUserTask
debug print is gone: it wrote to serial unserialized against user-space
lines and sheared concurrent log markers in half — the root cause of the
scenario flakes.
child_added/child_removed join the device-manager protocol. The driver
maps its register BAR (resource 0 is the ECAM config space; the walk
starts at 1), reads CAPLENGTH and HCSPARAMS1, and reads one PORTSC per
port: the connect bit and speed class come straight from hardware, no
rings needed to see the devices. The manager mirrors reported children
keyed by (parent, port), remembers which instance reported each, and
prunes a dead reporter's children before deciding the restart — the
children describe protocol state that died with the process. The
usb-report scenario drives the whole loop: two QEMU devices reported,
reporter killed, children pruned, driver respawned with backoff, and the
new instance re-claims, re-scans, and re-reports.
The manager is now a harness service on the well-known .device_manager
endpoint. Every driver spawns supervised; drivers with an assignment must
hello (device-manager-protocol, versioned) within a deadline enforced by
a timer sweep. Exit reasons drive the restart decision: clean exits stay
down, faults restart with 300/600/1200ms backoff, and three fast deaths
mark a driver failed instead of respawning forever. usb-xhci-bus is the
first conforming driver; the crash-test fixture claims a device, hellos,
and faults on purpose — each respawn re-proving claim release on death
through the manager's own path. maximum_tasks grows 16 -> 32: the
initial-ramdisk sweep (15 binaries at once) was intermittently
overflowing the static pool.
Signals are statements delivered as coalescing notifications to the
endpoint a process nominates with signal_bind — never a hijacked stack,
never a question (liveness is the zero-length ping the harness answers).
process_signal is supervisor-or-self gated, like kill; unbound targets
accumulate a pending mask delivered on bind. timer_bind is the missing
timed wait: a one-shot deadline landing in the same replyWait as
everything else — what stop(), hello deadlines, and restart backoff are
built from. runtime.service.run folds requests, signals, and
notifications into callbacks; the VFS conversion deletes its hand-rolled
loop and gains the whole lifecycle contract. The signals scenario drives
ping, reload, terminate->exited, the timer, and the deaf-child
deadline->killed path from ring 3. docs/process-lifecycle.md increments
1-4 are now as-built.
process_subscribe adds an endpoint to a bounded, ref-counted subscriber
table; every death posts the same badge encoding a supervisor's exit
notification uses, equally late, so subscribers observe a fully-released
child. A dying subscriber's own subscriptions are removed first — it never
hears about itself. The VFS is the first subscriber: open handles now
record their owner and are swept when the owner dies, because a service
must never depend on clients cleaning up after themselves
(docs/process-lifecycle.md). Proven by the vfs-client-death scenario.
The kernel records an ExitReason at all three death sites — clean exit,
fault (classified by vector), and process_kill — into a bounded ring
before the exit notification posts, so a supervisor's query never races
the notice. process_exit_reason is gated by the same supervisor check as
kill; runtime.process.exitReason is the stable interface. This is the
input restart policy reads (docs/process-lifecycle.md iron rule 2).
Every path out of a process (exit, fault, kill) now releases its device
claims alongside its IRQ and MSI bindings, so a restarted driver can claim
its hardware again — the cleanup half of process-lifecycle.md's iron rule 1.
MSI vectors were already swept by irq.releaseOwner; claims were the gap.
The claim-release test proves kill -> release -> re-claim, plus the broker
release in isolation.
Daniel Samson
Daniel Samson