Signals over IPC, one-shot timers, and the service harness (M17.4)

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.

README.md

@@ -2,12 +2,31 @@
 Codename: Shodan
 Version: 1
 
-A small operating system, written from scratch in Zig — a bootloader (`boot/`)
-and a microkernel (`system/kernel/`), sharing a neutral handoff contract (`system/boot-handoff.zig`).
-It boots x86-64 via UEFI, and so far has a framebuffer console, a physical frame
-allocator, its own paging with W^X permissions, interrupt/exception handling, a
-LAPIC timer, a kernel heap, a fixed-priority preemptive scheduler, and in-kernel IPC
-channels. See [`docs/`](docs/README.md) for how each piece works.
+A small resilient operating system, written from scratch in Zig.
+
+## Zen of DanOS:
+
+- Resilient Micro-Kernel Architecture.
+    - Every process run in an isolated user space not kernel space.
+    - Processes cannot take down the entire OS with it when they die or is killed
+- Stable public runtime library, private OS ABI.
+    - Keeps a stable runtime for user space processes between OS versions (great for backwards compatibility)
+    - Allows the underlying OS to be changed without effecting applications
+    - Provides a boundary to enable compatibility between OS's e.g. POSIX, MUSL etc
+- Drivers are just isolated processes in user space.
+    - Thin binaries that can be restarted like applications.
+    - Useful during driver development.
+    - Drivers can claim MMIO / ports
+    - Driver resources (e.g. IRQ/Port/MMIO) claims are automatically cleaned up if the driver dies or is killed
+    - Drivers can also hook into the process lifecyle to clean up or reset hardware
+- No legacy to deal with
+    - Zig code uses a clean coding style (Zen of Zig)
+    - Favor reading code over writing code.
+    - No magic numbers.
+    - No shortend names unless its for ABI compatibility or acronyms
+- Inter-Process Communication (IPC)
+  - Publish and subscribe to Asynchronous Messages
+  - Talk to services and processes synchronously
 
 ## Prerequisites
 
@@ -60,7 +79,7 @@ straight into CI.
 
 ## Documentation
 
-Design notes explaining the *why* behind the code live in
+Design notes explaining *why* behind the code live in
 [`docs/`](docs/README.md) — start with [`docs/README.md`](docs/README.md).
 
 ## Logo

docs/device-manager.md

@@ -0,0 +1,144 @@
+# The device manager
+
+**Status: design.** The primitives this builds on are real ([process-management.md](process-management.md):
+spawn/supervise/kill/exit-notification; [driver-model.md](driver-model.md): the device
+table as a capability system; [drivers.md](drivers.md): claim/map/IRQ), and the first
+per-device driver spawn works (the device manager matches the xHCI controller by PCI
+class and spawns `usb-xhci-bus` with the device id as argv[1]). This document designs
+the rest: the device manager as **the tree, the matcher, and the supervisor** — the
+policy process that turns [resilience.md](resilience.md)'s restart goal into practice
+for drivers.
+
+How processes stop, reload, and report their deaths is deliberately **not** in this
+document: that is the universal lifecycle every danos process speaks —
+[process-lifecycle.md](process-lifecycle.md), signals over IPC and the stable
+`runtime.process` interface. The device manager is that design's first serious
+customer, not its owner. Its own protocol contains nothing lifecycle-shaped; a
+driver is stopped, health-checked, and buried exactly like any other process.
+
+## The tree: structure in the manager, authority in the kernel
+
+The device tree is two things fused: *information* (what exists, how it nests) and
+*authority* (a descriptor is a licence to map physical memory). They separate:
+
+- The **kernel keeps the capability system** — device, I/O-port, and interrupt
+  claims, resource containment on `device_register`, the
+  `mmio_map`/`irq_bind`/`msi_bind` gates — and **cleans all of it up when a process
+  dies** (settled; it is increment 1 of
+  [process-lifecycle.md](process-lifecycle.md)). The three invariants in
+  [driver-model.md](driver-model.md) stay exactly where they are. A device manager
+  that could mint MMIO mappings by its own say-so would be a second kernel, and a
+  buggy one would un-earn everything the microkernel bought.
+- The **device manager owns the tree as data** — identity, topology, naming, driver
+  matching, hotplug events, and being the one process everything else asks about
+  devices. Firmware discovery seeds it (today via the kernel's snapshot); **bus
+  drivers grow it** by reporting what they see; applications query and watch it.
+  `device_enumerate` fades to a manager-internal (then deleted) seam.
+
+Long-term, discovery itself leaves the kernel — but not *into* the manager. PCI
+enumeration is a **pci-bus driver**: the manager spawns it against the host bridge
+(already a device with the ECAM window as a resource), it scans, it reports functions
+like any bus reports children. ACPI becomes an **acpi service** that interprets the
+tables and reports the namespace. The manager only orchestrates and merges. Moving
+AML interpretation out of ring 0 is its own project on its own track; nothing here
+depends on when it lands.
+
+## The protocol
+
+A `device-manager-protocol` module (the vfs-protocol pattern): extern-struct
+messages, a version in the handshake, reserved fields everywhere. The manager is a
+well-known endpoint (`ipc.register(.device_manager)`); the badge tells it who is
+talking; the same endpoint receives its children's exit notifications — one loop,
+one world.
+
+| Direction | Message | Purpose |
+|---|---|---|
+| driver → manager | `hello { version, role, device_id }` | confirms the argv assignment, starts the deadline clock |
+| bus → manager | `child_added { parent, identity, resources }` | one node the bus discovered |
+| bus → manager | `child_removed { id }` | unplug, or the bus lost it |
+| app → manager | `enumerate` | snapshot of the tree (read-only) |
+| app → manager | `subscribe` | receive published add/remove events |
+
+`hello` is the one deadline the manager enforces itself: spawned and silent past the
+deadline means wrong binary, wrong protocol version, or wedged before main — apply
+the stop sequence and the restart policy. Everything else lifecycle-shaped
+(terminate, the common `ping` liveness call, exit reasons) arrives through
+[process-lifecycle.md](process-lifecycle.md)'s vocabulary, not this protocol.
+
+Assignment stays argv (`usb-xhci-bus <device id>`) for now — simple, and it works.
+The step after `hello` exists is delegation: the manager claims (or is granted) the
+devices and passes the claim to the driver over IPC (the M13 capability-transfer
+mechanism), replacing first-come-first-served `device_claim` with policy. Identity in
+`child_added` is per-bus: PCI children carry the class triple (`pci_class`, as the
+xHCI match already uses); USB children carry the (class, subclass, protocol) triple
+from usb-ids.zig — each bus's native language, decoded by the shared ids modules.
+
+## Supervision and restart
+
+Every driver is spawned with the manager's exit endpoint (`spawnSupervised` — built).
+On a death notification:
+
+1. **Read the reason** ([process-lifecycle.md](process-lifecycle.md) increment 2).
+   Clean exit → it meant to; don't restart. Fault or missed `hello` deadline →
+   restart with **backoff**, and a crash-loop cap (three fast deaths → mark failed,
+   stop respawning, log loudly; a later `reload` to the manager can retry).
+2. **Prune the subtree** the dead bus driver reported. Its children describe
+   protocol state (xHCI slot ids, transfer rings) that died with the process;
+   keeping the nodes would be keeping a lie. Watchers receive `child_removed` — the
+   input service losing, then regaining, a keyboard is the *honest* description of
+   what happened. The restarted instance rediscovers and re-reports.
+3. **The claim is already free** because the kernel released it at death — the
+   restarted instance claims the same controller and comes up.
+
+Who supervises the supervisor: **init** (PID 1), which already supervises the
+services it starts. If the manager dies, drivers keep running (they hold their
+claims; the kernel doesn't care who their supervisor was — though their exit
+notifications now dangle harmlessly). The restarted manager re-learns the world:
+kernel snapshot, then a re-`hello` round — drivers answer a broadcast or are stopped
+and respawned. Full state handoff is deliberately not attempted.
+
+## Thin drivers, class protocols
+
+The [driver-model.md](driver-model.md) three-shape split, restated as processes:
+
+- A **bus driver** (usb-xhci-bus) owns its controller — claim, MMIO, IRQ/MSI, DMA
+  rings — and offers a *transfer* protocol ("submit a control transfer to device N",
+  built from the usb-abi request constructors) plus tree reports to the manager.
+- A **class driver** (usb-hid, usb-storage) owns nothing: it is matched to a reported
+  child by its identity triple, speaks the bus's transfer protocol downward and its
+  service's protocol upward — HID reports to the input service, blocks to the block
+  service. It works unchanged over any controller.
+- **Services** (input, display, block) aggregate class drivers and face applications.
+
+Each arrow is a protocol module. The manager routes none of the data plane — it
+introduces the parties (matching), supervises them (lifecycle), and gets out of the
+way.
+
+## Increments
+
+Increments 1–4 are the lifecycle prerequisites and live in
+[process-lifecycle.md](process-lifecycle.md) (claim cleanup on death, exit reasons,
+published exit events, signals + `runtime.process`). On top of those:
+
+5. **device-manager-protocol**: `hello`, supervised spawn with restart policy;
+   usb-xhci-bus becomes the first conforming driver.
+6. **Tree reports**: `child_added`/`child_removed`; the manager mirrors; xHCI reports
+   the mouse and keyboard QEMU already hangs off it.
+7. **App surface**: `enumerate`/`subscribe` over IPC; `device_enumerate` retreats
+   to a manager-internal seam.
+8. **Discovery migration**: pci-bus driver first, acpi service second, kernel scan
+   retired last. (AML-in-user-space is its own track.)
+
+## Settled questions (2026-07-12)
+
+- **Stateful buses**: pruning the subtree on bus-driver death is right for USB. A
+  future storage bus with in-flight writes wants drain-before-terminate — which is
+  exactly the `deadline_ms` parameter `stop()` already has; a per-driver deadline
+  is one value in the manager's policy table when such a bus arrives. No design
+  change.
+- **Manager death**: drivers survive the manager; the restarted manager re-learns
+  the world (above). Checkpointing driver state with the manager is deferred until
+  something demonstrates the need.
+- **Matching stays code until the third bus.** `driverFor`/`pciDriverFor` are
+  honest at two bus types; the third triggers the manifest (a driver declares what
+  it binds: a PCI class triple, a USB class triple, an ACPI `_HID`).

docs/ipc.md

@@ -103,3 +103,22 @@ This is what makes a user-space driver possible at all, and it's the subject of
   the oldest (discrete messages, not a coalescing level like the notification ring).
 - **A bounded reply.** `MSG_MAX` is 256 bytes and the copy runs under the big kernel
   lock; a bulk transfer wants shared pages, not a copy.
+
+## Lifecycle conventions over IPC (M17)
+
+Three conventions from [process-lifecycle.md](process-lifecycle.md) ride the
+notification mechanism:
+
+- **Signals** arrive as notifications on the endpoint a process nominated with
+  `signal_bind` (`runtime.process.bindSignals`): badge = the signal bit plus the
+  coalesced pending mask (`runtime.process.signalsFrom` decodes). Statements,
+  never questions; no payload, no reply.
+- **One-shot timers** (`timer_bind`, `runtime.system.timerOnce`) land as a
+  timer-bit notification — the timed wait: a service arms a deadline and keeps
+  serving, instead of blocking in sleep.
+- **The universal ping**: a **zero-length request is the liveness probe**,
+  answered with a zero-length reply by the service harness itself
+  (`runtime.service.run`). No protocol's requests start at length zero, so the
+  encoding cannot collide, and a wedged service simply fails to answer — which
+  is the diagnosis. Deep health ("can I reach my hardware?") stays a per-service
+  protocol message.

docs/m17-m18-plan.md

@@ -0,0 +1,191 @@
+# M17–M18 execution plan: process lifecycle + device manager
+
+The operational plan for building [process-lifecycle.md](process-lifecycle.md)
+(M17) and [device-manager.md](device-manager.md) increments 5–7 (M18). Design is
+settled in those documents; this file is the build order — one phase at a time,
+each phase green before the next starts. Delete or archive this file when M18
+lands.
+
+**Definition of green, every phase:** `zig build` clean, `zig build test` clean,
+`python3 test/qemu_test.py` passes (existing scenarios plus the phase's new one),
+and the relevant design doc's "known gaps" / status lines updated. Commit per
+green phase (no co-author trailers).
+
+**Workflow (settled 2026-07-12):** work happens in a dedicated git worktree, on
+feature branches cut from `main` — `feat/process-lifecycle` (M17.1–17.4),
+`feat/device-manager` (M18.1), `feat/usb-xhci-bus` (M18.2–18.3). When a branch's
+phases are all green it is **auto-merged into `main`**; branches are kept after
+merge, not deleted. Merges and branches are pushed to origin. Phase 0 (once):
+commit the design docs, merge the outstanding `feat/usb` work into `main`, and
+run the existing QEMU suite green before any new work starts.
+
+**Numbering note:** continues the milestone sequence (driver track ended at M16).
+
+## Status
+
+The loop marks a phase `[x]` in the same commit that lands it. A phase is marked
+only when its definition of green holds.
+
+- [x] **Phase 0** — baseline: docs committed, feat/usb merged to main, pushed;
+      `usb-xhci-libary.zig` renamed to `usb-xhci-library.zig`; existing QEMU
+      suite green from the worktree (48/48, 2026-07-12).
+- [x] **M17.1** — kernel releases claims/MSI on death (claims: `releaseAllOwnedBy`
+      in the reap; MSI was already swept by `irq.releaseOwner`; `claim-release`
+      test; suite 49/49)
+- [x] **M17.2** — exit reasons (`ExitReason` recorded at exit/fault/kill before
+      the notification; `process_exit_reason` supervisor-gated;
+      `runtime.process.exitReason`; kernel + ring-3 assertions; suite 49/49)
+- [x] **M17.3** — published exit events + VFS subscriber (`process_subscribe`,
+      bounded ref-counted table, publish on every death;
+      `runtime.process.subscribeExits`; VFS handles carry owners and are swept on
+      the owner's death; `vfs-client-death` test; suite 50/50)
+- [x] **M17.4** — signals, timer notifications, `runtime.process`, the service
+      harness (signal_bind/process_signal + coalescing pending mask; timer_bind
+      on the tick; bindSignals/signalsFrom/sendSignal/stop + timerOnce;
+      runtime.service.run with the zero-length ping; VFS converted; `signals`
+      scenario; suite 51/51)
+- [ ] **merge** `feat/process-lifecycle` → main, push
+- [ ] **M18.1** — device-manager protocol: hello + restart policy (branch `feat/device-manager`)
+- [ ] **merge** `feat/device-manager` → main, push
+- [ ] **M18.2** — xHCI port scan + tree reports (branch `feat/usb-xhci-bus`)
+- [ ] **M18.3** — app surface: enumerate/subscribe + device-list
+- [ ] **merge** `feat/usb-xhci-bus` → main, push — **loop ends here**
+
+---
+
+## M17.1 — the kernel releases a dead process's claims
+
+The cleanup half of iron rule 1; the prerequisite for every restart story.
+
+- `system/kernel/devices-broker.zig`: `releaseAllOwnedBy(owner: u32)` — clear
+  every `claimed[]` slot holding this task id.
+- `system/kernel/process.zig`: call it from the reap path, alongside the existing
+  IRQ-binding release (the ordering comment there says why IRQs go first — claims
+  slot in after them, before the exit notification).
+- MSI vectors: find where `msi_bind` records per-device vectors (interrupts
+  module) and release those by owner in the same pass.
+- Docs: remove the claims bullet from process-management.md "Known gaps".
+
+**Test:** new QEMU scenario `claim-release` — a test child claims an unclaimed
+device, is killed, is respawned, and claims the same device again successfully;
+assert both claims in the serial log. Kernel-side unit coverage in
+`system/kernel/tests.zig` for `releaseAllOwnedBy` (claim two devices as two owners,
+release one owner, verify exactly its claims freed).
+
+## M17.2 — exit reasons
+
+- `system/abi.zig`: `ExitReason` (exited, aborted, segmentation_fault,
+  illegal_instruction, arithmetic_fault, killed).
+- Kernel: record the reason at every death site — clean exit path, each fault
+  class in `onException`, the kill path. Bounded recent-exits table (ids are never
+  reused, so a small ring keyed by id is enough).
+- New system call `process_exit_reason(id)` — supervisor-gated, like kill; returns
+  the recorded reason or `-ESRCH` once evicted.
+- `library/runtime/process.zig`: `ExitReason` + `exitReason(id: u32)`.
+- Docs: remove the no-exit-status bullet from process-management.md.
+
+**Test:** extend the `supervision` scenario — three children: one exits cleanly,
+one faults (the fault-recovery pattern), one is killed; the supervisor asserts all
+three reasons.
+
+## M17.3 — published exit events
+
+- Kernel: bounded subscriber table (endpoints); new system call
+  `process_subscribe(endpoint)` (ungated, like `process_enumerate`); every death
+  posts `notify_exit_bit | id` to each subscriber — the same post the supervisor
+  path already uses.
+- `library/runtime/process.zig`: `subscribeExits(endpoint)`.
+- VFS becomes the first subscriber: on an exit event, release every handle keyed
+  by that task id (badges already are task ids). Log the release.
+- Docs: note the convention in ipc.md (exit events reuse the exit-notification
+  badge encoding).
+
+**Test:** new QEMU scenario `vfs-client-death` — a client opens a file and is
+killed without closing; assert the VFS logs the handle release and its open-handle
+count returns to baseline.
+
+## M17.4 — signals and the service harness
+
+- Kernel: per-task pending mask + bound endpoint; system calls
+  `signal_bind(endpoint)` and `process_signal(id, signal)` (supervisor-or-self
+  gated); delivery posts `notify_signal_bit | pending mask`, coalescing; pending
+  signals with no bound endpoint pend silently.
+- `library/runtime/process.zig`: `Signal`, `SignalSet`, `bindSignals`,
+  `signalsFrom`, `sendSignal`, `stop(id, deadline_ms)` (terminate → wait for exit
+  notification → kill). Implement `terminate`, `reload`, `user_1`, `user_2`;
+  `interrupt`/`quit` are enum members with no sender yet; `alarm` stays unbuilt.
+- Kernel: **one-shot timer notifications** — `timer_bind(endpoint, ms)` posts a
+  notification badge when the deadline lands (IRQ-as-IPC again, on the timer
+  wheel `sleep` already uses). This is the missing timed-wait primitive:
+  `replyWait` blocks forever and `sleep` blocks the whole process, but `stop()`'s
+  escalation, the device manager's `hello` deadline (M18.1), and restart backoff
+  all need a deadline while staying responsive. It is also the mechanism `alarm`
+  gets for free later.
+- New `library/runtime/service.zig`: the harness — `run(callbacks)` owning the
+  replyWait loop, folding protocol messages, signals, and child-exit notifications
+  into `init` / `on_message` / `on_reload` / `on_terminate`; answers the common
+  `ping` automatically. Define the reserved `ping` request encoding here and
+  document it in ipc.md (one obvious encoding; smallest that cannot collide with
+  existing protocols).
+- Convert one existing service (input-source or hpet) to the harness as proof it
+  subtracts code rather than adding it.
+
+**Test:** extend `supervision` — a harness-built child: `sendSignal(reload)`
+observed in its log, `ping` answered, `stop()` produces a clean exit with reason
+`exited`; a second child that ignores signals (no bind) is killed by `stop()`'s
+deadline with reason `killed`.
+
+## M18.1 — device-manager protocol: hello + restart policy
+
+- New `system/services/device-manager/device-manager-protocol.zig` module
+  (vfs-protocol pattern): `hello { version, role, device_id }`; version constant;
+  reserved fields.
+- Device manager: register the `.device_manager` endpoint; spawn drivers with its
+  exit endpoint; enforce the hello deadline; restart policy — backoff, crash-loop
+  cap (three fast deaths → mark failed, log, stop), reasons from M17.2 deciding
+  restart vs not.
+- usb-xhci-bus: adopt the harness + send hello. hpet/ps2-bus follow only if the
+  conversion is mechanical; otherwise they keep working unconverted (the manager
+  only enforces hello on drivers spawned with an assignment).
+- build.zig: test-loop entry for the protocol module if it grows pure logic.
+
+**Test:** new QEMU scenario `driver-restart` — the xHCI driver takes a test-only
+argv flag to fault after hello on its first run; assert: fault, exit reason
+recorded, manager respawns with backoff, second run claims the controller
+(M17.1) and hellos clean. Assert the crash-loop cap by a driver that always
+faults (a tiny test driver, not xhci).
+
+## M18.2 — bus tree reports
+
+- Protocol: `child_added { parent, identity, resources }` / `child_removed { id }`.
+- usb-xhci-bus: bring-up to **port scan only** — map the MMIO window (claimed in
+  M16-era work), controller reset/start per xHCI spec, walk the port registers,
+  report one `child_added` per connected port with speed + port number as
+  identity. **No transfer rings, no descriptors** — reading device/interface
+  descriptors (and therefore USB class triples for matching) is the follow-on USB
+  track, not this plan.
+- Device manager: mirror reports into its tree; prune the subtree (emitting
+  `child_removed`) when a bus driver dies; assert re-report on restart.
+
+**Test:** QEMU already attaches usb-kbd + usb-mouse on xhci.0 — assert two
+`child_added` events reach the manager and appear in its tree dump; kill the
+driver, assert two `child_removed` then two fresh `child_added` after respawn.
+
+## M18.3 — the application surface
+
+- Protocol: `enumerate` (tree snapshot) + `subscribe` (published add/remove
+  events, input-service pattern).
+- A small client (`device-list`, the `ps` analog) exercising both; the manager
+  becomes the one answer to "what devices exist" for user space.
+  `device_enumerate` stays for drivers/kernel seeding — its retreat is tied to the
+  discovery migration, out of this plan.
+
+**Test:** QEMU scenario — `device-list` shows the tree including USB children;
+during a driver restart the subscribing client logs remove + add events.
+
+---
+
+**Explicitly out of scope** (own tracks, after M18): discovery migration (pci-bus
+driver, acpi service, retiring the kernel scan), USB control transfers +
+descriptors + class-driver matching, the musl layer, `interrupt`/`quit` senders
+(needs a console), job control.

docs/process-lifecycle.md

@@ -0,0 +1,327 @@
+# Process lifecycle: signals over IPC
+
+**Status: increments 1–4 built** (2026-07-12): claim release on death, exit
+reasons, published exit events, and signals + one-shot timers + the service
+harness are all in — the interface below is as-built. The primitives underneath
+predate this design ([process-management.md](process-management.md):
+spawn, the supervision link, kill, child-exit notifications); this document designs
+the layer above them — the standard vocabulary a danos process speaks about its own
+life, and the stable `runtime.process` interface that carries it. Nothing here is
+device- or driver-specific: a driver, the VFS, and a user application all stop,
+reload, and die the same way. The device manager is simply this design's first
+serious customer ([device-manager.md](device-manager.md)).
+
+**"POSIX" in this document means the concepts, never the letter of the standard.**
+danos borrows the ideas and the hard-won lessons (what SIGTERM *means*, why SIGPIPE
+was a mistake) without inheriting the mechanism, the API, or the names. The naming
+rule is danos's own and it is strict: plain words that communicate intent
+(`terminate`, `reload`, `exited`) and the IPC vocabulary the system already speaks
+(`bind`, `subscribe`, `publish`, `endpoint`) — never `SIG*`, never a second word for
+a concept that already has one. Literal POSIX arrives later and lives elsewhere: a
+**musl-based C layer** (growing out of library/posix) that wires C programs to the
+danos runtime — musl's syscall surface retargeted at danos system calls and IPC
+protocols (files onto the VFS protocol, `sigaction`/`wait` onto this lifecycle,
+sockets onto whatever networking becomes). Ported programs see POSIX; the system
+underneath never does.
+
+## Why a standard vocabulary
+
+A supervisor can only manage processes it has never heard of if "please exit" means
+the same thing to all of them. That is the one thing POSIX signals got deeply right:
+`SIGTERM` means the same thing to nginx and to a five-line script, which is why
+process supervision on Unix (init systems, container runtimes) is possible at all.
+danos wants that property from day one, because supervision-and-restart is the
+system's core motivation ([resilience.md](resilience.md)).
+
+What POSIX got wrong — for a system like this — is the **delivery mechanism**:
+asynchronous control-flow hijack. A Unix handler runs on a stolen stack at an
+arbitrary instruction boundary, which is why the async-signal-safe function list
+exists, why `errno` must be saved, and why the canonical signal bug is a SIGTERM
+handler innocently calling `printf` mid-`malloc`. That entire bug class comes from
+the mechanism, not the vocabulary, and none of it is worth importing.
+
+A microkernel already has the right channel: **a signal is a message.** QNX delivers
+POSIX signals over its message passing; seL4 has notification objects; Erlang turned
+"death is a message to whoever linked" into a reliability philosophy. danos has
+already done it once without naming it: a child's death arrives as a notification
+badge on the supervisor's endpoint — the microkernel's SIGCHLD, the IRQ-as-IPC
+pattern reused. Signals are the same pattern reused a third time.
+
+## The mechanism
+
+- **`signal_bind(endpoint)`** — a process nominates the endpoint its signals arrive
+  on, exactly as `irq_bind` nominates where a device's interrupts land. The runtime
+  does this at startup for any program that opts in.
+- **`process_signal(id, signal)`** — posts the signal as an asynchronous
+  notification to the target's bound endpoint: badge = `notify_badge_bit |
+  notify_signal_bit | pending signals`. Non-blocking for the sender, always.
+- **Pending signals coalesce** in a per-process bitmask until the target next waits
+  — exactly like interrupt notifications, and exactly POSIX's own semantics for
+  non-realtime signals (two pending SIGTERMs are one SIGTERM). The bitmask *is* the
+  design: signals carry no payload. Anything with a payload is a protocol message.
+- **Authority**: the supervisor may signal its children — the same link that is
+  already the kill authority. A process may signal itself. Anything broader waits
+  for transferable process handles.
+- **No binding, no problem**: a process that never calls `signal_bind` is not
+  broken — its signals pend unread and only `process_kill` works on it. Simple
+  programs stay simple; the vocabulary is opt-in, the kill authority is not.
+
+Because delivery is a message into the process's own event loop, there is no
+async-signal-safe list in danos: a handler is ordinary code running at a point the
+process chose. The bug class is gone by construction, not by discipline.
+
+## The vocabulary: POSIX.1-1990, sorted honestly
+
+The full 1990 set, and what each becomes. Two intrinsically problematic cases get a
+defense below the table.
+
+| POSIX.1-1990 | danos disposition | Notes |
+|---|---|---|
+| SIGTERM | signal `terminate` | finish up and exit; the supervisor's polite half |
+| SIGHUP | signal `reload` | re-read configuration / re-scan |
+| SIGINT | signal `interrupt` | interactive interrupt; meaningful once a console can send it, in the vocabulary now so numbering is stable |
+| SIGQUIT | signal `quit` | as SIGINT, without the core-dump baggage |
+| SIGALRM | signal `alarm` | timer expiry as a message; the Unix SIGALRM+`longjmp` timeout hacks are impossible here. In the vocabulary, unbuilt: no consumer yet, and when one appears it is runtime sugar over the existing timer — zero kernel work |
+| SIGUSR1, SIGUSR2 | signals `user_1`, `user_2` | service-defined |
+| SIGCHLD | **already exists** — the exit notification | the badge carries the child id, dodging the classic coalescing bug (Unix code must loop `waitpid`) |
+| SIGKILL | `process_kill` — kernel mechanism | its definition is "cannot be handled"; it was never really a signal |
+| SIGABRT | exit reason `abort` | `abort()` is synchronous self-termination, not an event |
+| SIGSEGV, SIGILL, SIGFPE | exit reasons, **never delivered** | see below |
+| SIGPIPE | **an error return**, not a signal | see below |
+| SIGSTOP, SIGTSTP, SIGTTIN, SIGTTOU, SIGCONT | deferred | job control needs terminals, sessions, and process groups; stop/continue is scheduler territory |
+
+**The fault signals (SIGSEGV, SIGILL, SIGFPE) are intrinsically wrong for messages.**
+They are *synchronous* — raised at a specific faulting instruction, not "sometime
+soon". A message cannot be delivered to a process whose next instruction re-faults;
+it never reaches its event loop to read it. POSIX only makes fault handlers "work"
+via the async hijack (run the handler *instead of* the instruction), and even there,
+returning from a SIGSEGV handler without curing the cause is undefined behavior.
+danos's architecture already has the better answer: fault → the kernel kills the
+process ([resilience.md](resilience.md) step 2, built) → the supervisor reads the
+reason → restart. Recovery is restart, not a handler. This is also truer to the 1990
+standard than handling is: the standard's default action for all three was
+"terminate the process".
+
+**SIGPIPE deserves special contempt.** Its default kills a process that writes to a
+closed pipe — which is why "the whole server died because one client disconnected"
+is roughly every network daemon's first production bug, and why every mature codebase
+contains the same fix: ignore SIGPIPE, handle the `EPIPE` error return. danos made
+the right choice natively already — a reply owed to a dead peer fails with `-EPEER`.
+Errors from operations are error returns from those operations. The posix layer can
+synthesize SIGPIPE for ported code that expects it.
+
+### Statements, not questions
+
+A signal and a protocol message both travel over IPC — the difference is the
+**contract**, not the transport. danos IPC has two primitives, both already in
+daily use: the **asynchronous notification** (a badge — bits that coalesce into a
+pending mask; the sender never blocks; no payload, *no reply path*; how IRQs and
+exit events arrive) and the **synchronous call** (a rendezvous — payload both
+ways, the caller waits for the reply; how VFS requests work). A signal is the
+first kind: a *statement*. `terminate` wants no reply — the exit notification is
+its acknowledgement.
+
+A health probe is the second kind: a *question*, worthless without its answer —
+and the answer's absence within a deadline is the very thing being measured.
+Asked as a signal it has no reply channel (a coalescing bit can't carry an answer,
+and the authority rule forbids a child signalling its supervisor back); asked as a
+call, the timeout-is-the-diagnosis semantics come free. So there is no `health`
+signal. Liveness is the common **`ping`**: a reserved request every harness-run
+service answers automatically on its main endpoint — still free for the service
+author, still one obvious way — and a supervisor's probe is a `ping` call with a
+deadline.
+
+## The two iron rules
+
+1. **Cleanup is the kernel's job.** A process can die with no warning — fault,
+   kill, power. Correctness must never depend on a `terminate` handler running. On
+   any death the kernel releases the address space, IPC handles, IRQ bindings, and
+   owed replies (built), and must also release **device, I/O-port, and interrupt
+   claims and MSI vectors** (the known gap in
+   [process-management.md](process-management.md); increment 1). A signal handler is
+   for *graceful* work — flushing, deregistering, saving — never for *necessary*
+   work.
+2. **Kill is not a signal, and exit reasons are load-bearing.** The standard stop
+   sequence is *terminate → deadline → `process_kill`*; the unhandleable kill stays
+   a kernel mechanism. And a supervisor deciding whether to restart must know *how*
+   the child died: clean exit (meant to — don't restart), fault (restart with
+   backoff), killed (the supervisor did it). The exit notification today carries
+   only the id; it grows a reason. Restart policy cannot be written without it.
+
+## Who learns of a death
+
+A death has three audiences, and conflating them is how systems end up with either
+zombie state or privileged snooping:
+
+1. **The supervisor** — gets the exit notification on the endpoint it gave at spawn
+   (built), which grows the `ExitReason` (increment 2). The supervisor is the only
+   audience that needs the *reason*, because it is the only one deciding whether to
+   restart.
+2. **The peer owed a reply** — already built: a client that dies mid-request fails
+   the server's reply with `-EPEER`; a server that dies fails its waiting clients
+   the same way. This covers the *synchronous* case only.
+3. **The subscribers** — the new piece, and it is the input service's
+   publish/subscribe shape ([input.md](input.md)) applied to exits. A stateful
+   service accumulates per-client state across many requests: the VFS holds a dead
+   client's open file handles, the input service holds its subscriptions, a future
+   network stack holds its sockets. None of these are the client's supervisor, and
+   none learn anything from a failed reply if the client simply never calls again.
+   So the kernel **publishes every exit** to whoever subscribed:
+   `process_subscribe(endpoint)` adds a subscriber, and each death posts a
+   notification to every subscriber (badge = `notify_exit_bit | process id` — the
+   same encoding supervisors already decode, the IRQ-as-IPC pattern once more). The
+   subscriber filters for ids it holds state for and releases what the dead client
+   held. Correlating is free of bookkeeping: an IPC sender's badge already *is* its
+   task id (`runtime.ipc.Received`), so the id a service has been keying client
+   state by all along is the id the exit event carries.
+
+   Subscription, not broadcast-to-everyone: only processes that asked receive
+   events, the kernel keeps a bounded subscriber table, and delivery is the same
+   non-blocking coalescing notification as everything else — a dying process never
+   waits on its mourners. Subscribing is ungated, like `process_enumerate`: what is
+   running (and dying) is not a secret between cooperating processes. Subscribers
+   do not receive the exit reason — the VFS does not care *why* the client died.
+
+This is the service-side mirror of iron rule 1: **a service must never depend on
+its clients cleaning up after themselves.** Handle release on client death is the
+service's job, triggered by the published exit event — never by a courtesy
+"closing now" message that a crashed client will never send.
+
+## The stable interface: `runtime.process`
+
+`runtime.process` already owns what a process receives at birth (`Init`, the
+argv contract). It grows to own the other end of life.
+
+**The runtime is the stable interface; the numbers are not.** danos applications do
+not make system calls — they call the runtime library, and the system-call numbers,
+notification bits, and signal bit positions beneath it are a **private kernel ↔
+runtime contract** that may change at any time (settled 2026-07-12). This is why
+the runtime exists. Today kernel and runtime ship from one tree in one image, so
+"stability" is simply building them together. When driver binaries start shipping
+as separately-versioned applications — the whole point of the restart design — the
+binary's embedded runtime version becomes compatibility metadata (the same idea as
+the protocol version in the device manager's `hello`), and the kernel refuses what
+it cannot serve. Signals therefore need no reserved numbering scheme: the enum
+below is vocabulary, not ABI.
+
+```zig
+/// The signal vocabulary. The value is the bit position in the pending mask — a
+/// private kernel/runtime detail, free to change while they ship together.
+pub const Signal = enum(u5) {
+    terminate = 0, // SIGTERM: finish up and exit
+    reload = 1, // SIGHUP: re-read configuration
+    interrupt = 2, // SIGINT
+    quit = 3, // SIGQUIT
+    alarm = 4, // SIGALRM
+    user_1 = 5, // SIGUSR1
+    user_2 = 6, // SIGUSR2
+};
+
+/// A decoded pending mask: the coalesced set of signals a notification delivered.
+pub const SignalSet = struct {
+    pending: u32,
+    pub fn has(set: SignalSet, signal: Signal) bool { ... }
+    pub fn iterate(set: SignalSet) Iterator { ... }
+};
+
+/// Nominate `endpoint` as this process's signal endpoint (signal_bind). The
+/// runtime's service harness calls this; a bare program may call it directly and
+/// fold signals into its own replyWait loop.
+pub fn bindSignals(endpoint: usize) bool { ... }
+
+/// Decode a received badge into signals, or null if the badge is not a signal
+/// notification (mirrors ipc.Received.isChildExit).
+pub fn signalsFrom(badge: usize) ?SignalSet { ... }
+
+/// Send `signal` to process `id`. Supervisor-gated, like kill; non-blocking.
+pub fn sendSignal(id: u32, signal: Signal) bool { ... }
+
+/// The standard stop sequence: terminate, wait up to `deadline_ms` for the exit
+/// notification, then process_kill. The one call a supervisor needs.
+pub fn stop(id: u32, deadline_ms: u64) void { ... }
+
+/// Subscribe `endpoint` to published exit events (process_subscribe). Every
+/// process death posts an asynchronous notification: badge = notify_exit_bit |
+/// process id — the same encoding a supervisor's exit notification uses, decoded
+/// by the same ipc.Received helpers. For stateful services: release what the dead
+/// client held (file handles, subscriptions, sockets). Ungated, like
+/// process_enumerate.
+pub fn subscribeExits(endpoint: usize) bool { ... }
+
+/// How a process ended — queried after the exit notification (the kernel records
+/// it first, so the two never race). What restart policy reads. (Built in M17.2.)
+pub const ExitReason = enum(u8) {
+    exited, // returned from main / clean exit
+    aborted, // abort() — deliberate self-termination (SIGABRT's ghost; reserved)
+    segmentation_fault, // SIGSEGV's ghost
+    illegal_instruction, // SIGILL's ghost
+    arithmetic_fault, // SIGFPE's ghost
+    protection_fault, // general protection fault
+    fault, // any other CPU exception
+    killed, // process_kill
+};
+```
+
+Two deliberate absences. There is no `mask`/`block` API — a process that is not
+ready for a signal simply has not waited on its endpoint yet; the pending mask *is*
+the blocked set. And there is no per-signal handler registration at this layer —
+dispatch is the process's own `switch` over `SignalSet`, or the service harness's
+callbacks (`on_terminate`, `on_reload`) for programs that want defaults.
+
+### The service harness
+
+`runtime.service` owns the `replyWait` loop and folds every event source — signals,
+child exits, protocol messages — into callbacks, with the vocabulary's defaults:
+`terminate` returns from the loop (clean exit), the common `ping` is answered automatically,
+`reload` is ignored unless overridden. One loop, no locking, nothing reentrant. A
+service author writes domain logic; the lifecycle contract is satisfied by the
+harness. A process that bypasses the harness and ignores its signals meets the
+deadline-then-kill escalation — you cannot force a process to implement an
+interface, but you can make compliance free and non-compliance fatal.
+
+### The musl layer later
+
+The POSIX C layer is a **musl port**: musl's arch/syscall layer retargeted so that
+what musl believes are kernel syscalls become danos runtime calls and IPC — `open`
+and `read` onto the VFS protocol, `kill`/`sigaction`/`waitpid` onto this document's
+vocabulary, `exit` onto the runtime's exit path. `sigaction` handlers registered
+through it are invoked by the runtime's loop when the signal message arrives —
+synchronous underneath, async-looking to ported code, delivered at wait boundaries
+the way most Unix programs already experience signals (at syscalls). No stack hijack
+ever happens, `SA_RESTART` semantics come free because nothing was interrupted, and
+SIGPIPE can be synthesized from `-EPEER` for the programs that expect it. C programs
+get POSIX; danos-native programs never pay for it.
+
+## Increments
+
+1. **Kernel: release device/port/IRQ claims and MSI vectors on death** — the
+   cleanup half of iron rule 1, and the prerequisite for any restart story. Test:
+   kill a claiming driver, spawn it again, the claim succeeds.
+2. **Exit reason in the death notification** (`ExitReason` above).
+3. **Exit events**: `process_subscribe` in the kernel (bounded subscriber table,
+   publishes on every death), `runtime.process.subscribeExits`; the VFS becomes the
+   first subscriber — releasing a dead client's handles is its proof test.
+4. **Signals**: `signal_bind` + `process_signal` + the pending mask in the kernel;
+   `runtime.process` grows the interface above; the service harness handles
+   `terminate` and answers the common `ping`; `stop()` for supervisors.
+
+[device-manager.md](device-manager.md) builds directly on all four.
+
+## Settled questions (2026-07-12)
+
+- **Signal numbering is not ABI**: the runtime is the stable interface; the numbers
+  beneath it are a private kernel ↔ runtime contract (see "The stable interface").
+- **Liveness is a `ping` call, not a signal**: signals are statements, questions
+  are synchronous calls (see "Statements, not questions"). A service wanting *deep*
+  health ("can I reach my hardware?") defines its own protocol message on top.
+- **Process handles: deferred.** Pids + the supervisor gate cover everything
+  planned; transferable handles (Fuchsia-style, delegating signalling without
+  delegating kill) wait for the capability table to grow types beyond endpoints.
+- **`alarm`: in the vocabulary, unbuilt.** No consumer yet; when one appears it is
+  runtime sugar over the existing timer (arm a timer that posts your own signal) —
+  zero kernel work, so deferring costs nothing.
+- **Subscription granularity: all exits**, subscriber-side filtering — one
+  subscription per service, a bounded kernel table. Per-id subscriptions only if
+  event volume ever matters (hundreds of processes, not before).
+- **Client identity across the exit boundary: no convention needed** — an IPC
+  sender's badge already is its task id (see "Who learns of a death").

docs/process-management.md

@@ -95,11 +95,18 @@ the architecture layer calls up into `tick`.
 
 ## Known gaps (bring-up honesty)
 
-- Device **claims** are not released on death (pre-existing: the fault path has
-  the same gap) — a killed driver's device stays claimed until reboot.
+- ~~Device claims are not released on death~~ Closed (M17.1): every path out of a
+  process releases its device claims alongside its IRQ and MSI bindings
+  (`releaseTaskResourcesLocked`), so a restarted driver can claim its hardware
+  again — the cleanup half of [process-lifecycle.md](process-lifecycle.md)'s iron
+  rule 1. The `claim-release` test proves the kill → release → re-claim cycle.
 - Kernel stacks of dead tasks are leaked, as on every exit path (no reaper yet).
-- There is no exit *status* in the notification, only the id; a supervisor that
-  needs the code can grow a wait-style call later.
+- ~~There is no exit status in the notification~~ Closed (M17.2): the kernel
+  records how every process ends — exited, a fault class, or killed — before it
+  posts the exit notification, and the supervisor reads it with
+  `process_exit_reason` (`runtime.process.exitReason`). This is the input to
+  restart policy ([process-lifecycle.md](process-lifecycle.md)); an exit *code*
+  for the clean case can still ride alongside later.
 - Enumerate writes through the caller's raw pointer under the bring-up trust
   model, like `device_enumerate` (an unmapped page is a self-DoS, not an
   isolation break).
@@ -109,4 +116,5 @@ the architecture layer calls up into `tick`.
 `process-list` (enumerate), `process-kill` (kernel-level kill paths, refusals,
 notifications), `supervision` (the whole user-side surface via the process-test
 service: spawn supervised → enumerate → kill blocked and spinning children →
-notifications → gone). See test/qemu_test.py.
+notifications → gone), `claim-release` (a killed claim-holder's device is
+claimable again). See test/qemu_test.py.

library/runtime/ipc.zig

@@ -94,6 +94,15 @@ pub fn send(h: Handle, message: []const u8) bool {
 /// GSI. See `isNotification`.
 pub const notify_badge_bit: u64 = abi.notify_badge_bit;
 
+/// Set alongside `notify_badge_bit` when the notification is a **signal** — the
+/// lifecycle vocabulary of docs/process-lifecycle.md, delivered to the endpoint
+/// nominated with `process.bindSignals`. Decode with `process.signalsFrom`.
+pub const notify_signal_bit: u64 = abi.notify_signal_bit;
+
+/// Set alongside `notify_badge_bit` when the notification is a **one-shot timer**
+/// landing (`system.timerOnce`).
+pub const notify_timer_bit: u64 = abi.notify_timer_bit;
+
 /// Set alongside `notify_badge_bit` when the notification is a **child-exit
 /// notice** — a process this one spawned (with an exit endpoint) has ended —
 /// rather than a device interrupt. The low bits carry the child's process id.
@@ -133,6 +142,17 @@ pub const Received = struct {
     }
 
     /// The task id of whoever posted a buffered message, meaningful only when
+    /// Whether this arrival is a signal notification — decode the set with
+    /// `process.signalsFrom(badge)`.
+    pub fn isSignal(self: Received) bool {
+        return self.isNotification() and self.badge & notify_signal_bit != 0;
+    }
+
+    /// Whether this arrival is a one-shot timer landing (`system.timerOnce`).
+    pub fn isTimer(self: Received) bool {
+        return self.isNotification() and self.badge & notify_timer_bit != 0;
+    }
+
     /// `isMessage`. (The badge's low bits, with the three high marker bits masked off.)
     pub fn senderTaskId(self: Received) u32 {
         return @intCast(self.badge & ~(notify_badge_bit | notify_exit_bit | notify_message_bit));

library/runtime/process.zig

@@ -1,8 +1,15 @@
-//! Process-level runtime types: what a user program receives at entry. Mirrors
+//! Process-level runtime types: what a user program receives at entry (`Init`,
+//! the argv contract) and the process end of the lifecycle
+//! (docs/process-lifecycle.md) — today the exit reason a supervisor reads to
+//! decide restart; signals and the stop sequence land here with M17.4. Mirrors
 //! the spirit of `std.process.Init.Minimal` in danos terms — std's `Args` holds
 //! no data on freestanding targets, so the type is danos's own.
 
 const std = @import("std");
+const abi = @import("abi");
+const sc = @import("system-call.zig");
+const ipc = @import("ipc.zig");
+const system = @import("system.zig");
 
 /// Everything a program receives at entry. Passed to
 /// `pub fn main(init: runtime.process.Init)`; programs that need nothing keep
@@ -45,3 +52,86 @@ pub const Arguments = struct {
         }
     };
 };
+
+/// How a process ended — what a supervisor's restart policy reads: a clean exit
+/// meant to stop, a fault wants a restart with backoff, killed means the
+/// supervisor did it itself (docs/process-lifecycle.md).
+pub const ExitReason = abi.ExitReason;
+
+/// How dead child `id` ended. Ask after the exit notification arrives — the
+/// kernel records the reason before it posts the notification, so this never
+/// races it. Returns null for an id that never lived, is still alive, was
+/// evicted from the kernel's bounded record, or is not this process's child
+/// (the same authority gate as `kill`).
+pub fn exitReason(id: u32) ?ExitReason {
+    const r = sc.systemCall1(.process_exit_reason, id);
+    if (r > ~@as(usize, 0) - 4095) return null; // a wrapped -errno
+    return @enumFromInt(r);
+}
+
+/// The signal vocabulary (docs/process-lifecycle.md): POSIX's concepts, danos's
+/// names, message delivery. A signal is a one-way coalescing statement — never a
+/// question (liveness is the zero-length ping call) and never kill (that is
+/// `system.kill`, unhandleable by definition).
+pub const Signal = abi.Signal;
+
+/// The coalesced set of signals one notification delivered: two pending
+/// terminates arrive as one. Decode a received badge with `signalsFrom`.
+pub const SignalSet = struct {
+    pending: u32,
+
+    pub fn has(set: SignalSet, signal: Signal) bool {
+        return set.pending & (@as(u32, 1) << @intFromEnum(signal)) != 0;
+    }
+};
+
+/// Nominate `endpoint` as this process's signal endpoint. Signals posted while
+/// unbound have pended; they are delivered immediately on bind, coalesced.
+pub fn bindSignals(endpoint: usize) bool {
+    return sc.systemCall1(.signal_bind, endpoint) == 0;
+}
+
+/// Decode a received badge into the signals it delivered, or null if it is not
+/// a signal notification.
+pub fn signalsFrom(badge: u64) ?SignalSet {
+    if (badge & abi.notify_badge_bit == 0 or badge & abi.notify_signal_bit == 0) return null;
+    return .{ .pending = @truncate(badge & ~(abi.notify_badge_bit | abi.notify_signal_bit)) };
+}
+
+/// Post `signal` to child `id` (or to yourself). Supervisor-gated, like kill;
+/// non-blocking, always — a statement, not a conversation.
+pub fn sendSignal(id: u32, signal: Signal) bool {
+    return sc.systemCall2(.process_signal, id, @intFromEnum(signal)) == 0;
+}
+
+/// The standard stop sequence (docs/process-lifecycle.md): terminate, wait up to
+/// `deadline_ms` for the exit notification on `exit_endpoint` (the endpoint the
+/// child was spawned with), then kill. Any *other* notifications arriving on
+/// that endpoint while stopping are consumed and dropped — a supervisor with
+/// concurrent traffic implements the same sequence inside its own event loop
+/// (arm `system.timerOnce`, keep serving) instead of calling this.
+pub fn stop(id: u32, deadline_ms: u64, exit_endpoint: usize) void {
+    _ = sendSignal(id, .terminate);
+    _ = system.timerOnce(exit_endpoint, deadline_ms);
+    var receive: [8]u8 = undefined;
+    while (true) {
+        const got = ipc.replyWait(exit_endpoint, &.{}, &receive, null);
+        if (got.isChildExit() and got.childProcessId() == id) return;
+        if (got.isTimer()) break; // the deadline passed first — escalate
+    }
+    _ = system.kill(id);
+    while (true) {
+        const got = ipc.replyWait(exit_endpoint, &.{}, &receive, null);
+        if (got.isChildExit() and got.childProcessId() == id) return;
+    }
+}
+
+/// Subscribe `endpoint` to published exit events: every process death posts an
+/// asynchronous notification with the same badge encoding as a supervisor's exit
+/// notice (decode with `ipc.Received.isChildExit`/`childProcessId`). For stateful
+/// services: release what the dead client held — file handles, subscriptions —
+/// because a service must never depend on clients cleaning up after themselves
+/// (docs/process-lifecycle.md). Ungated, like `system.processes`.
+pub fn subscribeExits(endpoint: usize) bool {
+    return sc.systemCall1(.process_subscribe, endpoint) == 0;
+}

library/runtime/runtime.zig

@@ -35,5 +35,9 @@ pub const panic = start.panic;
 /// Process entry types: the `Init` handed to `main`, and its `Arguments`.
 pub const process = @import("process.zig");
 
+/// The service harness: one replyWait loop folding requests, signals, and
+/// notifications into callbacks (docs/process-lifecycle.md).
+pub const service = @import("service.zig");
+
 /// The heap as a `std.mem.Allocator`, for Zig `std` containers in user code.
 pub const allocator = heap.allocator;

library/runtime/service.zig

@@ -0,0 +1,81 @@
+//! The service harness (docs/process-lifecycle.md): one replyWait loop that
+//! folds protocol requests, signals, and subscribed notifications into
+//! callbacks — so the lifecycle contract ("answers ping, exits on terminate")
+//! is satisfied by construction and a service author writes domain logic only.
+//! Nothing is asynchronous inside the process: a callback runs at a point the
+//! loop chose, never on a hijacked stack — the whole reason signals are
+//! messages.
+//!
+//! The liveness probe: a **zero-length request is the universal ping**, answered
+//! with a zero-length reply by the harness itself. No protocol's requests start
+//! at length zero, so the encoding cannot collide, and there is nothing for a
+//! service author to implement — a wedged service simply fails to answer, which
+//! is the diagnosis (see docs/ipc.md).
+
+const abi = @import("abi");
+const ipc = @import("ipc.zig");
+const process = @import("process.zig");
+
+pub const Callbacks = struct {
+    /// Called once with the service's endpoint before the loop starts — the
+    /// place to subscribe to exit events, bind IRQs, or announce readiness.
+    /// Return false to abort startup (the process exits).
+    init: ?*const fn (endpoint: ipc.Handle) bool = null,
+    /// One protocol request from `sender` (a task id): write the reply into
+    /// `reply`, return its length. The zero-length ping never reaches this.
+    on_message: *const fn (message: []const u8, reply: []u8, sender: u32) usize,
+    /// A notification that is not a signal — a subscribed exit event, a bound
+    /// IRQ, a timer landing. The raw badge; decode with the ipc helpers.
+    on_notification: ?*const fn (badge: u64) void = null,
+    /// The reload signal. Default: ignored.
+    on_reload: ?*const fn () void = null,
+    /// The terminate signal, called before the loop returns. The clean exit is
+    /// the return itself — never put *necessary* work here (iron rule 1: a kill
+    /// arrives with no warning; this is for graceful extras only).
+    on_terminate: ?*const fn () void = null,
+    /// Publish the endpoint under a well-known service id at startup.
+    service: ?abi.ServiceId = null,
+};
+
+/// Run the service: create and (optionally) register the endpoint, bind signals
+/// to it, call `init`, then serve until `terminate` arrives — at which point the
+/// loop returns and main's return is the clean exit the supervisor reads as
+/// `ExitReason.exited`. `maximum_message` sizes the receive and reply buffers
+/// (a service passes its protocol's message maximum).
+pub fn run(comptime maximum_message: usize, callbacks: Callbacks) void {
+    const endpoint = ipc.createIpcEndpoint() orelse return;
+    if (callbacks.service) |id| {
+        if (!ipc.register(id, endpoint)) return;
+    }
+    _ = process.bindSignals(endpoint);
+    if (callbacks.init) |initialise| {
+        if (!initialise(endpoint)) return;
+    }
+
+    var reply_buffer: [maximum_message]u8 = undefined;
+    var reply_len: usize = 0;
+    var receive: [maximum_message]u8 = undefined;
+    while (true) {
+        const got = ipc.replyWait(endpoint, reply_buffer[0..reply_len], &receive, null);
+        if (got.isNotification()) {
+            reply_len = 0; // nothing owed for a notification
+            if (process.signalsFrom(got.badge)) |signals| {
+                if (signals.has(.reload)) {
+                    if (callbacks.on_reload) |onReload| onReload();
+                }
+                if (signals.has(.terminate)) {
+                    if (callbacks.on_terminate) |onTerminate| onTerminate();
+                    return; // the loop's return IS the clean exit
+                }
+                continue;
+            }
+            if (callbacks.on_notification) |onNotification| onNotification(got.badge);
+            continue;
+        }
+        if (got.len == 0) {
+            reply_len = 0; // the universal ping: a zero-length reply, from the harness
+            continue;
+        }
+        reply_len = callbacks.on_message(receive[0..got.len], &reply_buffer, got.senderTaskId());
+    }
+}

library/runtime/system.zig

@@ -32,6 +32,15 @@ pub fn sleep(ms: usize) void {
     _ = sc.systemCall1(.sleep, ms);
 }
 
+/// Arm a one-shot timer: after `ms` milliseconds the kernel posts a timer
+/// notification (`ipc.Received.isTimer`) to `endpoint`. The timed wait of
+/// docs/process-lifecycle.md — a service arms a deadline and keeps serving,
+/// instead of blocking in sleep; what stop-sequence escalation, hello deadlines,
+/// and restart backoff are built from.
+pub fn timerOnce(endpoint: usize, ms: u64) bool {
+    return sc.systemCall2(.timer_bind, endpoint, ms) == 0;
+}
+
 /// Monotonic nanoseconds since boot — a time source for timeouts and short delays. It
 /// only ever moves forward. This is *not* wall-clock time (no date, no timezone — that
 /// is a user-space service layered on top). Deadline pattern for a bounded poll loop:

system/abi.zig

@@ -53,9 +53,31 @@ pub const SystemCall = enum(u64) {
     process_enumerate = 24, // process_enumerate(buffer, maximum) -> total: snapshot the task table
     process_kill = 25, // process_kill(id) -> 0/-errno: end a process this process spawned
     ipc_send = 26, // ipc_send(handle, message_ptr, message_len) -> 0/-errno: post a payload to an endpoint's async queue without blocking
+    process_exit_reason = 27, // process_exit_reason(id) -> ExitReason/-errno: how a dead child ended (its supervisor only)
+    process_subscribe = 28, // process_subscribe(endpoint) -> 0/-errno: subscribe to published exit events — every death posts a notification
+    signal_bind = 29, // signal_bind(endpoint) -> 0/-errno: nominate the endpoint this process's signals arrive on
+    process_signal = 30, // process_signal(id, signal) -> 0/-errno: post a signal to a child (or to yourself)
+    timer_bind = 31, // timer_bind(endpoint, ms) -> 0/-errno: one-shot timer — posts a notification when ms elapse
     _,
 };
 
+/// How a process ended — recorded by the kernel at death, queried by the
+/// supervisor with `process_exit_reason`, and the input to its restart decision
+/// (docs/process-lifecycle.md): a clean exit meant to stop, a fault wants a
+/// restart with backoff, killed means the supervisor did it itself. The faults
+/// mirror the CPU exceptions a ring-3 process can die of; they are exit reasons,
+/// never delivered to the faulting process (recovery is restart, not a handler).
+pub const ExitReason = enum(u8) {
+    exited = 0, // returned from main / called exit
+    aborted = 1, // deliberate self-termination (reserved: no abort path yet)
+    segmentation_fault = 2, // page fault
+    illegal_instruction = 3, // invalid opcode
+    arithmetic_fault = 4, // divide error, x87 or SIMD fault
+    protection_fault = 5, // general protection fault
+    fault = 6, // any other CPU exception
+    killed = 7, // process_kill
+};
+
 /// The x86 MSI message address base (`0xFEE0_0000`): a device raises an MSI by writing
 /// `data` to this address, which the Local APIC turns into an interrupt at the vector
 /// in `data`. The kernel returns the concrete (address, data) from `msi_bind`; this is
@@ -93,6 +115,35 @@ pub const notify_exit_bit: u64 = 1 << 62;
 /// broadcasts where a rendezvous is the wrong shape (the input service is the first user).
 pub const notify_message_bit: u64 = 1 << 61;
 
+/// Set (alongside `notify_badge_bit`) in the badge of a **signal notification** —
+/// the process-lifecycle vocabulary of docs/process-lifecycle.md, delivered to the
+/// endpoint the process nominated with `signal_bind`. The low bits carry the
+/// coalesced pending mask (bit positions = `Signal` values): signals are
+/// statements, not questions, and two pending terminates are one terminate.
+pub const notify_signal_bit: u64 = 1 << 60;
+
+/// Set (alongside `notify_badge_bit`) in the badge of a **timer notification** —
+/// a one-shot `timer_bind` deadline landing. No payload bits: what to do when the
+/// deadline fires is whatever the receiver armed it for (a stop-sequence
+/// escalation, a restart backoff, an alarm).
+pub const notify_timer_bit: u64 = 1 << 59;
+
+/// The signal vocabulary (docs/process-lifecycle.md): POSIX's concepts, danos's
+/// names, message delivery. The value is the bit position in the pending mask — a
+/// private kernel/runtime detail, free to change while they ship together. Kill
+/// is not here (it is `process_kill`, unhandleable by definition); faults are not
+/// here (they are `ExitReason`s — recovery is restart, not a handler); liveness is
+/// not here (a question, asked as the zero-length ping call, not a statement).
+pub const Signal = enum(u5) {
+    terminate = 0, // finish up and exit (the polite half of the stop sequence)
+    reload = 1, // re-read configuration / re-scan
+    interrupt = 2, // interactive interrupt (no sender until a console exists)
+    quit = 3, // as interrupt, by convention more final
+    alarm = 4, // a timer the process armed for itself (unbuilt: no consumer yet)
+    user_1 = 5, // service-defined
+    user_2 = 6, // service-defined
+};
+
 /// Capacity of `ProcessDescriptor.name` — matches the longest name `system_spawn`
 /// accepts, so a process's recorded name (its argv[0]) is never truncated.
 pub const maximum_process_name = 64;

system/kernel/devices-broker.zig

@@ -104,6 +104,19 @@ pub fn ownerOf(id: u64) ?u32 {
     return claimed[@intCast(id)];
 }
 
+/// Release every claim held by `owner` — called by the process layer on every
+/// path out of a process (exit, fault, kill), so a restarted driver can claim its
+/// hardware again (docs/process-lifecycle.md iron rule 1: cleanup is the kernel's
+/// job). The devices stay in the table — they describe hardware, which did not go
+/// away — only their ownership clears.
+pub fn releaseAllOwnedBy(owner: u32) void {
+    for (claimed[0..count]) |*slot| {
+        if (slot.*) |o| {
+            if (o == owner) slot.* = null;
+        }
+    }
+}
+
 /// Resource `index` of device `id`, or null if out of range.
 pub fn resourceOf(id: u64, index: u64) ?device_abi.ResourceDescriptor {
     if (id >= count) return null;

system/kernel/kernel.zig

@@ -412,13 +412,26 @@ fn recoverableFault(vector: u64) bool {
 /// plus a POST code and a persistent breadcrumb. (A ring-3 fault on a *borrowed*
 /// kernel thread — process.run, the user-pf isolation probe — also lands here: there
 /// is no scheduled process to kill.)
+/// Classify a CPU exception vector as the ExitReason a supervisor reads — the
+/// fault classes of docs/process-lifecycle.md. Faults are exit reasons, never
+/// signals delivered to the faulting process: recovery is restart, not a handler.
+fn exitReasonForVector(vector: u64) abi.ExitReason {
+    return switch (vector) {
+        14 => .segmentation_fault, // page fault
+        6 => .illegal_instruction, // invalid opcode
+        0, 16, 19 => .arithmetic_fault, // divide error, x87, SIMD
+        13 => .protection_fault, // general protection
+        else => .fault,
+    };
+}
+
 fn onException(state: *const architecture.CpuState) noreturn {
     if (architecture.fromUser(state) and scheduler.currentIsUserProcess() and recoverableFault(state.vector)) {
         statusPrint("\ndanos: process {d} ({s}) killed by {s} (vector {d}) on core {d}\n", .{ scheduler.currentId(), scheduler.current().name(), architecture.exceptionName(state.vector), state.vector, scheduler.currentCpuIndex() });
         statusPrint("  error code : 0x{x}\n", .{state.error_code});
         statusPrint("  IP         : 0x{x:0>16}\n", .{architecture.instructionPointer(state)});
         if (architecture.faultAddress(state)) |address| statusPrint("  fault addr : 0x{x:0>16}\n", .{address});
-        process.killCurrentProcess(); // reclaims everything, reschedules; never returns
+        process.killCurrentProcess(exitReasonForVector(state.vector)); // reclaims everything, reschedules; never returns
     }
 
     log.checkpoint(cp_exception);

system/kernel/process.zig

@@ -137,6 +137,7 @@ pub fn init() void {
     architecture.setSystemCallHandler(system_call);
     scheduler.terminate_current_hook = terminateCurrentLocked;
     scheduler.reap_task_hook = reapTaskLocked;
+    scheduler.timer_tick_hook = timerSweepLocked;
 }
 
 /// Return -1 (as an unsigned bit pattern) in the system_call result register.
@@ -164,6 +165,7 @@ fn system_call(state: *architecture.CpuState) void {
             // A scheduled process tears down fully (terminateCurrent); a borrowed
             // test thread unwinds back to the kernel that entered it.
             if (scheduler.currentIsUserProcess()) {
+                scheduler.current().exit_reason = .exited;
                 terminateCurrent();
             } else architecture.userExit();
         },
@@ -199,6 +201,11 @@ fn system_call(state: *architecture.CpuState) void {
         .clock => systemClock(state),
         .process_enumerate => systemProcessEnumerate(state),
         .process_kill => systemProcessKill(state),
+        .process_exit_reason => systemProcessExitReason(state),
+        .process_subscribe => systemProcessSubscribe(state),
+        .signal_bind => systemSignalBind(state),
+        .process_signal => systemProcessSignal(state),
+        .timer_bind => systemTimerBind(state),
         _ => fail(state),
     }
 }
@@ -552,7 +559,11 @@ pub var fault_kill_count: u64 = 0;
 ///   endpoint reference destroys the Endpoint, and a still-bound GSI would have an
 ///   ISR call notifyFromIsr on freed memory the next time the device fired.
 ///   `releaseOwner` also leaves the line masked, so a dead driver's device goes
-///   quiet rather than storming.
+///   quiet rather than storming. (It drops MSI vectors by the same owner sweep.)
+/// - Device claims are released with the IRQ bindings, so a restarted driver can
+///   claim the same hardware again — the cleanup half of process-lifecycle.md's
+///   iron rule 1. Claims hold no pointers, so ordering is free; they go here so
+///   the exit notification (below, last) observes a fully-released child.
 /// - A client this task still owes a reply to (it died between receive and reply)
 ///   is failed with -EPEER rather than left blocked forever — a dead server must
 ///   not hang its callers.
@@ -565,7 +576,33 @@ pub var fault_kill_count: u64 = 0;
 ///   reference taken at spawn is dropped with it.
 /// Precondition: the big kernel lock is held.
 fn releaseTaskResourcesLocked(t: *scheduler.Task) void {
+    recordExitLocked(t);
     irq.releaseOwner(t.id);
+    devices_broker.releaseAllOwnedBy(t.id);
+    // The dying task's signal endpoint and one-shot timers go with it.
+    if (t.signal_endpoint) |raw| {
+        ipc.dropRef(@ptrCast(@alignCast(raw)));
+        t.signal_endpoint = null;
+    }
+    t.pending_signals = 0;
+    for (&one_shot_timers) |*slot| {
+        if (slot.*) |timer| {
+            if (timer.owner == t.id) {
+                ipc.dropRef(timer.endpoint);
+                slot.* = null;
+            }
+        }
+    }
+    // A dead subscriber's own subscriptions go first: it must not hear about
+    // itself, and the slots' endpoint references drop with it.
+    for (&exit_subscribers) |*slot| {
+        if (slot.*) |subscriber| {
+            if (subscriber.owner == t.id) {
+                ipc.dropRef(subscriber.endpoint);
+                slot.* = null;
+            }
+        }
+    }
     if (t.ipc_client) |client| {
         t.ipc_client = null;
         client.ipc_status = -ipc.EPEER;
@@ -575,6 +612,12 @@ fn releaseTaskResourcesLocked(t: *scheduler.Task) void {
     scheduler.removeFromWaitQueueLocked(t);
     scheduler.forgetIpcClientLocked(t);
     ipc.closeHandles(t);
+    // Publish the exit to every subscriber (docs/process-lifecycle.md): the same
+    // badge encoding as the supervisor's notification, and equally late, so a
+    // subscriber also observes a fully-released child.
+    for (&exit_subscribers) |*slot| {
+        if (slot.*) |subscriber| ipc.notifyLocked(subscriber.endpoint, abi.notify_exit_bit | t.id);
+    }
     if (t.exit_endpoint) |raw| {
         const endpoint: *ipc.Endpoint = @ptrCast(@alignCast(raw));
         t.exit_endpoint = null;
@@ -628,6 +671,7 @@ pub fn killProcess(caller_id: u32, target_id: u32) i64 {
     const target = scheduler.taskByIdLocked(target_id) orelse return -ipc.ESRCH;
     if (target.aspace == 0) return -ipc.ESRCH; // kernel tasks are not processes
     if (target.supervisor != caller_id) return -ipc.EPERM;
+    target.exit_reason = .killed;
     if (target.state == .running) {
         target.kill_pending = true;
     } else {
@@ -640,12 +684,170 @@ pub fn killProcess(caller_id: u32, target_id: u32) i64 {
 /// The fault is confined to the process — the kernel trapped it on the task's own
 /// kernel stack and is intact — so everything the process held is reclaimed and the
 /// core reschedules. The system keeps running; only the faulting process dies
-/// (docs/resilience.md: fault -> kill -> continue).
-pub fn killCurrentProcess() noreturn {
+/// (docs/resilience.md: fault -> kill -> continue). `reason` is the fault class
+/// (from the vector), recorded for the supervisor's `process_exit_reason`.
+pub fn killCurrentProcess(reason: abi.ExitReason) noreturn {
+    scheduler.current().exit_reason = reason;
     fault_kill_count += 1;
     terminateCurrent();
 }
 
+/// The bounded record of recent deaths, for `process_exit_reason`: ids are never
+/// reused, so a ring keyed by id is enough — a record evicted by wraparound reads
+/// as -ESRCH, the same as an id that never lived, which a supervisor treats as
+/// "too late to ask". Written under the big kernel lock by the reap.
+const exit_record_capacity = 64;
+const ExitRecord = struct { id: u32 = 0, supervisor: u32 = 0, reason: abi.ExitReason = .exited, valid: bool = false };
+var exit_records: [exit_record_capacity]ExitRecord = .{ExitRecord{}} ** exit_record_capacity;
+var exit_record_next: usize = 0;
+
+/// Record a dying task's (id, supervisor, reason) — called by the reap before the
+/// exit notification is posted, so a supervisor that hears the notification can
+/// always still query the reason. Precondition: the big kernel lock is held.
+fn recordExitLocked(t: *scheduler.Task) void {
+    exit_records[exit_record_next] = .{ .id = t.id, .supervisor = t.supervisor, .reason = t.exit_reason, .valid = true };
+    exit_record_next = (exit_record_next + 1) % exit_record_capacity;
+}
+
+/// How dead process `id` ended, for `caller` — the kernel half of the
+/// process_exit_reason system call. Returns the ExitReason value, -ESRCH (never
+/// lived, still alive, or evicted from the ring), or -EPERM (the caller was not
+/// its supervisor — the same authority gate as process_kill).
+pub fn exitReasonOf(caller_id: u32, target_id: u32) i64 {
+    const flags = sync.enter();
+    defer sync.leave(flags);
+    for (&exit_records) |*record| {
+        if (record.valid and record.id == target_id) {
+            if (record.supervisor != caller_id) return -ipc.EPERM;
+            return @intFromEnum(record.reason);
+        }
+    }
+    return -ipc.ESRCH;
+}
+
+/// The published exit events' subscribers (docs/process-lifecycle.md "Who learns
+/// of a death"): stateful services — the VFS's file handles, input's
+/// subscriptions — that must release what a dead client held and cannot learn it
+/// any other way (a client that simply never calls again looks like silence).
+/// Bounded like every kernel table; each entry holds its own endpoint reference.
+const exit_subscriber_capacity = 8;
+const ExitSubscriber = struct { endpoint: *ipc.Endpoint, owner: u32 };
+var exit_subscribers: [exit_subscriber_capacity]?ExitSubscriber = .{null} ** exit_subscriber_capacity;
+
+/// process_subscribe(endpoint): subscribe the caller's endpoint to published exit
+/// events. Ungated, like process_enumerate — what is running (and dying) is not a
+/// secret between cooperating processes. -ENOSPC when the table is full.
+fn systemProcessSubscribe(state: *architecture.CpuState) void {
+    const t = scheduler.current();
+    if (t.aspace == 0) return fail(state);
+    const endpoint = ipc.resolveHandle(t, architecture.systemCallArg(state, 0)) orelse return failErr(state, ipc.EBADF);
+    const flags = sync.enter();
+    defer sync.leave(flags);
+    for (&exit_subscribers) |*slot| {
+        if (slot.* == null) {
+            endpoint.refcount += 1; // the slot's own reference, dropped on unsubscribe-by-death
+            slot.* = .{ .endpoint = endpoint, .owner = t.id };
+            return architecture.setSystemCallResult(state, 0);
+        }
+    }
+    failErr(state, ipc.ENOSPC);
+}
+
+/// signal_bind(endpoint): nominate where this process's signals arrive — the
+/// IRQ-as-IPC pattern a fourth time (docs/process-lifecycle.md). Replacing a
+/// binding drops the old reference; signals that pended while unbound are
+/// delivered immediately on bind, coalesced into one notification.
+fn systemSignalBind(state: *architecture.CpuState) void {
+    const t = scheduler.current();
+    if (t.aspace == 0) return fail(state);
+    const endpoint = ipc.resolveHandle(t, architecture.systemCallArg(state, 0)) orelse return failErr(state, ipc.EBADF);
+    const flags = sync.enter();
+    defer sync.leave(flags);
+    if (t.signal_endpoint) |raw| ipc.dropRef(@ptrCast(@alignCast(raw)));
+    endpoint.refcount += 1;
+    t.signal_endpoint = @ptrCast(endpoint);
+    if (t.pending_signals != 0) {
+        ipc.notifyLocked(endpoint, abi.notify_signal_bit | t.pending_signals);
+        t.pending_signals = 0;
+    }
+    architecture.setSystemCallResult(state, 0);
+}
+
+/// process_signal(id, signal): post a signal — a one-way, coalescing statement,
+/// never a question (docs/process-lifecycle.md). The authority gate is the
+/// supervision link, like kill; a process may also signal itself. Unbound
+/// targets accumulate the signal in their pending mask.
+fn systemProcessSignal(state: *architecture.CpuState) void {
+    const t = scheduler.current();
+    if (t.aspace == 0) return fail(state);
+    const id = architecture.systemCallArg(state, 0);
+    const signal = architecture.systemCallArg(state, 1);
+    if (id > std.math.maxInt(u32)) return failErr(state, ipc.ESRCH);
+    if (signal > 31) return failErr(state, ipc.EBADF); // not a Signal bit position
+    const flags = sync.enter();
+    defer sync.leave(flags);
+    const target = scheduler.taskByIdLocked(@intCast(id)) orelse return failErr(state, ipc.ESRCH);
+    if (target.aspace == 0) return failErr(state, ipc.ESRCH);
+    if (target.supervisor != t.id and target.id != t.id) return failErr(state, ipc.EPERM);
+    target.pending_signals |= @as(u32, 1) << @intCast(signal);
+    if (target.signal_endpoint) |raw| {
+        const endpoint: *ipc.Endpoint = @ptrCast(@alignCast(raw));
+        ipc.notifyLocked(endpoint, abi.notify_signal_bit | target.pending_signals);
+        target.pending_signals = 0;
+    }
+    architecture.setSystemCallResult(state, 0);
+}
+
+/// The one-shot timers of timer_bind: the missing timed wait. A service arms a
+/// deadline and keeps serving; the expiry arrives in the same replyWait as
+/// everything else (notify_timer_bit). What stop-sequence escalation, hello
+/// deadlines, and restart backoff are built from — and later, `alarm`.
+const timer_capacity = 16;
+const OneShotTimer = struct { deadline: u64, endpoint: *ipc.Endpoint, owner: u32 };
+var one_shot_timers: [timer_capacity]?OneShotTimer = .{null} ** timer_capacity;
+
+/// Sweep expired timers — hung on scheduler.timer_tick_hook, so it runs on every
+/// tick with the big kernel lock held, like the sleeper wake it rides beside.
+fn timerSweepLocked() void {
+    const now = architecture.millis();
+    for (&one_shot_timers) |*slot| {
+        if (slot.*) |timer| {
+            if (now >= timer.deadline) {
+                ipc.notifyLocked(timer.endpoint, abi.notify_timer_bit);
+                ipc.dropRef(timer.endpoint);
+                slot.* = null;
+            }
+        }
+    }
+}
+
+/// timer_bind(endpoint, ms): arm a one-shot timer. -ENOSPC when the table is full.
+fn systemTimerBind(state: *architecture.CpuState) void {
+    const t = scheduler.current();
+    if (t.aspace == 0) return fail(state);
+    const endpoint = ipc.resolveHandle(t, architecture.systemCallArg(state, 0)) orelse return failErr(state, ipc.EBADF);
+    const ms = architecture.systemCallArg(state, 1);
+    const flags = sync.enter();
+    defer sync.leave(flags);
+    for (&one_shot_timers) |*slot| {
+        if (slot.* == null) {
+            endpoint.refcount += 1;
+            slot.* = .{ .deadline = architecture.millis() + ms, .endpoint = endpoint, .owner = t.id };
+            return architecture.setSystemCallResult(state, 0);
+        }
+    }
+    failErr(state, ipc.ENOSPC);
+}
+
+fn systemProcessExitReason(state: *architecture.CpuState) void {
+    const t = scheduler.current();
+    if (t.aspace == 0) return fail(state);
+    const id = architecture.systemCallArg(state, 0);
+    if (id > std.math.maxInt(u32)) return failErr(state, ipc.ESRCH);
+    const r = exitReasonOf(t.id, @intCast(id));
+    architecture.setSystemCallResult(state, @bitCast(r));
+}
+
 /// Resolve `(device_id, resource_index)` to a GSI this process is entitled to bind, or null.
 /// The two checks are the whole security story: the device must be *claimed* by the
 /// caller, and the resource must be one of that device's `irq` resources as recorded

system/kernel/scheduler.zig

@@ -50,6 +50,17 @@ pub const Task = struct {
     // null. Holds its own reference, dropped when the notification is posted.
     // Opaque here for the same reason as `handles` below.
     exit_endpoint: ?*anyopaque = null,
+    // How this process ended — set by the death paths (exit, fault, kill) just
+    // before the reap records it for `process_exit_reason`. Meaningless while
+    // the task lives.
+    exit_reason: abi.ExitReason = .exited,
+    // Endpoint this process's signals arrive on (signal_bind), or null — same
+    // ownership rules as exit_endpoint (holds a reference; opaque here).
+    signal_endpoint: ?*anyopaque = null,
+    // Signals posted but not yet delivered: the coalescing pending mask
+    // (docs/process-lifecycle.md). Bits are abi.Signal values. Signals pend here
+    // until an endpoint is bound; two pending terminates are one terminate.
+    pending_signals: u32 = 0,
     // Set by process_kill on a task that is running on another core; the kernel
     // finishes the kill at that task's next system call or timer tick.
     kill_pending: bool = false,
@@ -643,9 +654,15 @@ fn reapKillPendingLocked() void {
 /// other critical section, but releases it *without* touching the interrupt flag
 /// — the handler's `iretq` restores the interrupted context's flags, so
 /// re-enabling here would open a nested-interrupt window before the return.
+/// Called from the tick with the big kernel lock held — process.zig hangs the
+/// one-shot timer sweep here (timer_bind), the same call-up pattern as the
+/// teardown hooks below.
+pub var timer_tick_hook: ?*const fn () void = null;
+
 pub fn tick() void {
     _ = sync.enter();
     wakeExpired();
+    if (timer_tick_hook) |hook| hook();
     reapKillPendingLocked();
     if (preemption_enabled) schedule();
     sync.leaveIsr();

system/kernel/tests.zig

@@ -132,6 +132,12 @@ pub fn run(case: []const u8, boot_information: *const BootInformation) void {
         processKillTest(boot_information);
     } else if (eql(case, "supervision")) {
         supervisionTest(boot_information);
+    } else if (eql(case, "claim-release")) {
+        claimReleaseTest(boot_information);
+    } else if (eql(case, "vfs-client-death")) {
+        vfsClientDeathTest(boot_information);
+    } else if (eql(case, "signals")) {
+        signalsTest(boot_information);
     } else if (eql(case, "initial-ramdisk")) {
         initialRamdiskTest(boot_information);
     } else if (eql(case, "vfs")) {
@@ -1207,15 +1213,15 @@ fn userPfTest() void {
 /// hand — address space, code page RO+X, stack page RW+NX — because the blob is a
 /// raw code fragment, not an ELF `spawnProcess` could load. Returns false if any
 /// allocation fails.
-fn spawnFaultingProcess() bool {
+fn spawnFaultingProcess() ?u32 {
     const blob = process.pfBlob();
     const flags = sync.enter();
     defer sync.leave(flags);
 
-    const aspace = architecture.createAddressSpace() orelse return false;
+    const aspace = architecture.createAddressSpace() orelse return null;
     const code_frame = pmm.alloc() orelse {
         architecture.destroyAddressSpace(aspace);
-        return false;
+        return null;
     };
     // Fill through the physmap (the user mapping is read-only); pad with int3 so a
     // stray jump traps instead of sliding.
@@ -1226,15 +1232,16 @@ fn spawnFaultingProcess() bool {
 
     const stack_frame = pmm.alloc() orelse {
         architecture.destroyAddressSpace(aspace); // frees code_frame too — it's mapped
-        return false;
+        return null;
     };
     architecture.mapUserPageInto(aspace, process.stack_base_virtual, stack_frame, true, false); // RW + NX
 
-    if (scheduler.spawnUserLocked(aspace, process.code_virtual, process.stack_base_virtual + abi.page_size, 4, "fault-probe", 0, null) == null) {
+    // Supervised by the calling test task, so exitReasonOf can read the verdict.
+    const id = scheduler.spawnUserLocked(aspace, process.code_virtual, process.stack_base_virtual + abi.page_size, 4, "fault-probe", scheduler.currentId(), null) orelse {
         architecture.destroyAddressSpace(aspace);
-        return false;
-    }
-    return true;
+        return null;
+    };
+    return id;
 }
 
 /// Fault recovery (docs/resilience.md step 2): a scheduled ring-3 process that
@@ -1263,7 +1270,8 @@ fn faultRecoveryTest(boot_information: *const BootInformation) void {
     scheduler.setPriority(4);
     check("init heartbeat before the fault", process.write_count >= 1);
 
-    check("faulting process spawned", spawnFaultingProcess());
+    const probe = spawnFaultingProcess() orelse 0;
+    check("faulting process spawned", probe != 0);
 
     // The kill: the faulting process #PFs on its first instruction and the kernel
     // reaps it instead of halting.
@@ -1272,6 +1280,7 @@ fn faultRecoveryTest(boot_information: *const BootInformation) void {
     while (process.fault_kill_count < 1 and architecture.millis() < deadline) scheduler.yield();
     scheduler.setPriority(4);
     check("faulting process was killed (not the machine)", process.fault_kill_count == 1);
+    check("the probe's reason reads segmentation_fault", process.exitReasonOf(scheduler.currentId(), probe) == @intFromEnum(abi.ExitReason.segmentation_fault));
 
     // Life after the kill: init must keep beating on the same core.
     const beats_at_kill = process.write_count;
@@ -1423,6 +1432,12 @@ fn processKillTest(boot_information: *const BootInformation) void {
     check("the sleeper's exit notification arrived (length 0)", r == 0);
     check("its badge carries the exit bit and the child id", badge == abi.notify_badge_bit | abi.notify_exit_bit | sleeper);
 
+    // M17.2: the recorded reason — the notification is the fence, so it is
+    // already readable, and gated by the same supervisor check as the kill.
+    check("the sleeper's reason reads killed", process.exitReasonOf(me, sleeper) == @intFromEnum(abi.ExitReason.killed));
+    check("a non-supervisor may not read the reason (-EPERM)", process.exitReasonOf(me + 12345, sleeper) == -ipcsync.EPERM);
+    check("an unknown id has no reason (-ESRCH)", process.exitReasonOf(me, 0xFFFF_FF00) == -ipcsync.ESRCH);
+
     const beats_at_kill = process.write_count;
     scheduler.sleep(1500); // more than one heartbeat period
     check("the heartbeat stopped with the kill", process.write_count == beats_at_kill);
@@ -1443,6 +1458,21 @@ fn processKillTest(boot_information: *const BootInformation) void {
     check("the spinner's exit notification arrived (length 0)", r == 0);
     check("its badge carries the exit bit and the child id", badge == abi.notify_badge_bit | abi.notify_exit_bit | spinner);
 
+    // M17.2: a child that ends on its own must read exited, not killed —
+    // args-echo with arguments echoes once and returns from main.
+    var clean: u32 = 0;
+    i = 0;
+    while (i < rd.count) : (i += 1) {
+        const item = rd.entry(i) orelse continue;
+        if (!eql(item.name, "args-echo")) continue;
+        clean = process.spawnProcessSupervised(item.blob, 4, &.{ "args-echo", "clean-exit" }, me, endpoint) catch 0;
+        break;
+    }
+    check("args-echo spawned as the clean-exit child", clean != 0);
+    r = ipcsync.replyWait(endpoint, 0, 0, 0, 0, abi.no_cap, &badge, &received_cap);
+    check("the clean child's exit notification arrived", badge == abi.notify_badge_bit | abi.notify_exit_bit | clean);
+    check("the clean child's reason reads exited", process.exitReasonOf(me, clean) == @intFromEnum(abi.ExitReason.exited));
+
     var table: [32]abi.ProcessDescriptor = undefined;
     const total = scheduler.enumerate(&table);
     var still_listed = false;
@@ -1454,6 +1484,176 @@ fn processKillTest(boot_information: *const BootInformation) void {
     result();
 }
 
+/// M17.1: a dead process's device claims are released by the reap, so a restarted
+/// driver can claim its hardware again (docs/process-lifecycle.md iron rule 1).
+/// First the broker release in isolation — two owners, one released, the other's
+/// claim must survive. Then the death-path wiring with a real child: the claim is
+/// made on the child's behalf (the broker is kernel-callable), the child is
+/// killed, and once the exit notification arrives — posted last, after release —
+/// the device must be unclaimed and claimable again.
+fn claimReleaseTest(boot_information: *const BootInformation) void {
+    log("DANOS-TEST-BEGIN: claim-release\n", .{});
+
+    var buffer: [2]device_abi.DeviceDescriptor = undefined;
+    const total = devices_broker.enumerate(&buffer);
+    check("the device tree is seeded (>= 2 devices)", total >= 2);
+    if (total < 2) {
+        result();
+        return;
+    }
+
+    // The broker release in isolation.
+    check("device 0 claimed by owner 111", devices_broker.claim(0, 111));
+    check("device 1 claimed by owner 222", devices_broker.claim(1, 222));
+    devices_broker.releaseAllOwnedBy(111);
+    check("owner 111's claim is released", devices_broker.ownerOf(0) == null);
+    check("owner 222's claim survives", (devices_broker.ownerOf(1) orelse 0) == 222);
+    devices_broker.releaseAllOwnedBy(222);
+    check("cleanup released owner 222", devices_broker.ownerOf(1) == null);
+
+    // The death-path wiring: a real process dies holding a claim.
+    check("bootloader handed over /system/services/init", boot_information.init_len != 0);
+    if (boot_information.init_len == 0) {
+        result();
+        return;
+    }
+    const image = @as([*]const u8, @ptrFromInt(boot_handoff.physicalToVirtual(boot_information.init_base)))[0..boot_information.init_len];
+    const me = scheduler.currentId();
+    const endpoint = ipcsync.createIpcEndpoint() orelse {
+        check("exit endpoint allocated", false);
+        result();
+        return;
+    };
+    const child = process.spawnProcessSupervised(image, 4, &.{"/system/services/init"}, me, endpoint) catch 0;
+    check("supervised child spawned", child != 0);
+    check("device 0 claimed on the child's behalf", devices_broker.claim(0, child));
+
+    check("the kill is accepted", process.killProcess(me, child) == 0);
+    var badge: u64 = 0;
+    var received_cap: u64 = 0;
+    _ = ipcsync.replyWait(endpoint, 0, 0, 0, 0, abi.no_cap, &badge, &received_cap);
+    check("the exit notification arrived", badge == abi.notify_badge_bit | abi.notify_exit_bit | child);
+    check("death released the child's claim", devices_broker.ownerOf(0) == null);
+    check("the device is claimable again", devices_broker.claim(0, me));
+    devices_broker.releaseAllOwnedBy(me);
+    result();
+}
+
+/// M17.3: the published exit events, proven by their first subscriber. The VFS
+/// subscribes at startup; a client opens a file and parks holding the handle;
+/// the kill posts the exit event to the VFS's endpoint; the VFS releases the
+/// dead client's handle and says so — the service-side mirror of iron rule 1
+/// (a service must never depend on clients cleaning up after themselves).
+fn vfsClientDeathTest(boot_information: *const BootInformation) void {
+    log("DANOS-TEST-BEGIN: vfs-client-death\n", .{});
+    if (boot_information.initial_ramdisk_len == 0) {
+        check("bootloader handed over an initial_ramdisk", false);
+        result();
+        return;
+    }
+    const image = @as([*]const u8, @ptrFromInt(boot_handoff.physicalToVirtual(boot_information.initial_ramdisk_base)))[0..boot_information.initial_ramdisk_len];
+    const rd = initial_ramdisk.Reader.init(image) orelse {
+        check("initial_ramdisk image is valid", false);
+        result();
+        return;
+    };
+
+    process.write_count = 0;
+    check("vfs spawned", spawnNamed(rd, "vfs"));
+
+    const me = scheduler.currentId();
+    const endpoint = ipcsync.createIpcEndpoint() orelse {
+        check("exit endpoint allocated", false);
+        result();
+        return;
+    };
+    var client: u32 = 0;
+    var i: u32 = 0;
+    while (i < rd.count) : (i += 1) {
+        const item = rd.entry(i) orelse continue;
+        if (!eql(item.name, "vfs-test")) continue;
+        client = process.spawnProcessSupervised(item.blob, 4, &.{ "vfs-test", "park" }, me, endpoint) catch 0;
+        break;
+    }
+    check("parked client spawned (supervised)", client != 0);
+
+    // Its heartbeat is the fence: once it beats, the handle is open.
+    const parked = "vfstest: parked";
+    scheduler.setPriority(1);
+    var deadline = architecture.millis() + 10000;
+    while (architecture.millis() < deadline) {
+        if (process.write_len >= parked.len and eql(process.write_buffer[0..parked.len], parked)) break;
+        scheduler.yield();
+    }
+    scheduler.setPriority(4);
+    check("client parked holding an open handle", process.write_len >= parked.len and eql(process.write_buffer[0..parked.len], parked));
+
+    check("the kill is accepted", process.killProcess(me, client) == 0);
+    var badge: u64 = 0;
+    var received_cap: u64 = 0;
+    _ = ipcsync.replyWait(endpoint, 0, 0, 0, 0, abi.no_cap, &badge, &received_cap);
+    check("the exit notification arrived", badge == abi.notify_badge_bit | abi.notify_exit_bit | client);
+
+    // The VFS heard the same published event; its release line is the proof.
+    const released = "vfs: released 1 handle(s) for dead client";
+    scheduler.setPriority(1);
+    deadline = architecture.millis() + 10000;
+    while (architecture.millis() < deadline) {
+        if (process.write_len >= released.len and eql(process.write_buffer[0..released.len], released)) break;
+        scheduler.yield();
+    }
+    scheduler.setPriority(4);
+    check("the VFS released the dead client's handle", process.write_len >= released.len and eql(process.write_buffer[0..released.len], released));
+    result();
+}
+
+/// M17.4 from ring 3: process-test's signal-run role drives the whole lifecycle
+/// surface — the zero-length ping (answered by the harness), signals as
+/// statements (reload logged, terminate = clean exit), the one-shot timer, and
+/// both endings of the stop sequence (polite -> exited, deaf -> killed at the
+/// deadline). Its "process-test: signals ok" is the pass marker.
+fn signalsTest(boot_information: *const BootInformation) void {
+    log("DANOS-TEST-BEGIN: signals\n", .{});
+    if (boot_information.initial_ramdisk_len == 0) {
+        check("bootloader handed over an initial_ramdisk", false);
+        result();
+        return;
+    }
+    const image = @as([*]const u8, @ptrFromInt(boot_handoff.physicalToVirtual(boot_information.initial_ramdisk_base)))[0..boot_information.initial_ramdisk_len];
+    const rd = initial_ramdisk.Reader.init(image) orelse {
+        check("initial_ramdisk image is valid", false);
+        result();
+        return;
+    };
+
+    process.setInitialRamdisk(image); // the parent system_spawns its children by name
+    process.write_count = 0;
+    var runner: u32 = 0;
+    var i: u32 = 0;
+    while (i < rd.count) : (i += 1) {
+        const item = rd.entry(i) orelse continue;
+        if (!eql(item.name, "process-test")) continue;
+        runner = process.spawnProcessSupervised(item.blob, 4, &.{ "process-test", "signal-run" }, scheduler.currentId(), null) catch 0;
+        break;
+    }
+    check("signal-run parent spawned", runner != 0);
+
+    const pass_marker = "process-test: signals ok";
+    const fail_marker = "process-test: FAIL";
+    scheduler.setPriority(1);
+    const deadline = architecture.millis() + 15000;
+    var saw_pass = false;
+    var saw_fail = false;
+    while (architecture.millis() < deadline and !saw_pass and !saw_fail) {
+        if (process.write_len >= pass_marker.len and eql(process.write_buffer[0..pass_marker.len], pass_marker)) saw_pass = true;
+        if (process.write_len >= fail_marker.len and eql(process.write_buffer[0..fail_marker.len], fail_marker)) saw_fail = true;
+        scheduler.yield();
+    }
+    scheduler.setPriority(4);
+    check("the signal-run parent reported ok", saw_pass and !saw_fail);
+    result();
+}
+
 /// The whole user-side surface at once: spawn process-test's supervisor role,
 /// which — entirely from ring 3 — creates an exit endpoint, spawns its two
 /// children supervised, sees them in process_enumerate, kills them (one blocked,

system/services/process-test/process-test.zig

@@ -48,11 +48,91 @@ fn awaitChildExit(endpoint: runtime.ipc.Handle) u32 {
     return received.childProcessId();
 }
 
+/// The harness-run child of the signals test: echoes requests, logs the two
+/// signals it handles. Terminate makes run() return, and returning from main is
+/// the clean exit the parent reads as ExitReason.exited.
+fn echo(message: []const u8, reply: []u8, sender: u32) usize {
+    _ = sender;
+    const n = @min(message.len, reply.len);
+    @memcpy(reply[0..n], message[0..n]);
+    return n;
+}
+
+fn onReload() void {
+    _ = runtime.system.write("process-test: reloaded\n");
+}
+
+fn onTerminate() void {
+    _ = runtime.system.write("process-test: terminating\n");
+}
+
+/// The parent of the signals test: drives ping, echo, reload, the one-shot
+/// timer, and both endings of the stop sequence (polite -> exited; deaf ->
+/// killed at the deadline). Prints "process-test: signals ok" as the marker.
+fn signalRun() void {
+    const endpoint = runtime.ipc.createIpcEndpoint() orelse fail("create exit endpoint");
+    const child = runtime.system.spawnSupervised("process-test", &.{"service"}, endpoint) orelse fail("spawn service child");
+
+    // Reach the child's endpoint through the registry (retry: it may not be up).
+    var service_handle: ?runtime.ipc.Handle = null;
+    var tries: u32 = 0;
+    while (service_handle == null and tries < 200) : (tries += 1) {
+        service_handle = runtime.ipc.lookup(.input);
+        if (service_handle == null) runtime.system.sleep(20);
+    }
+    const h = service_handle orelse fail("service child never registered");
+
+    // The universal ping: a zero-length call answered zero-length by the harness.
+    var reply: [16]u8 = undefined;
+    const pong = runtime.ipc.call(h, &.{}, &reply) catch fail("ping call failed");
+    if (pong != 0) fail("ping reply not empty");
+
+    // An ordinary request still reaches on_message.
+    const n = runtime.ipc.call(h, "echo!", &reply) catch fail("echo call failed");
+    if (n != 5 or !std.mem.eql(u8, reply[0..5], "echo!")) fail("echo mismatch");
+
+    // reload: a statement — the child logs it; the kernel test reads the serial.
+    if (!runtime.process.sendSignal(child, .reload)) fail("send reload");
+    runtime.system.sleep(200);
+
+    // The one-shot timer: armed on our endpoint, lands as isTimer.
+    if (!runtime.system.timerOnce(endpoint, 100)) fail("arm timer");
+    var scratch: [8]u8 = undefined;
+    const landing = runtime.ipc.replyWait(endpoint, scratch[0..0], &scratch, null);
+    if (!landing.isTimer()) fail("expected the timer landing");
+
+    // The stop sequence, polite path: terminate, clean exit inside the deadline.
+    runtime.process.stop(child, 2000, endpoint);
+    if ((runtime.process.exitReason(child) orelse .killed) != .exited) fail("service child reason not exited");
+
+    // The deaf child: binds nothing, hears nothing — the deadline kills it.
+    const deaf = runtime.system.spawnSupervised("process-test", &.{"sleeper"}, endpoint) orelse fail("spawn deaf child");
+    runtime.system.sleep(50); // let it reach its sleep
+    runtime.process.stop(deaf, 300, endpoint);
+    if ((runtime.process.exitReason(deaf) orelse .exited) != .killed) fail("deaf child reason not killed");
+
+    _ = runtime.system.write("process-test: signals ok\n");
+}
+
 pub fn main(init: runtime.process.Init) void {
     const role = init.arguments.get(1) orelse return; // spawned bare (ramdisk sweep): stay silent
     if (std.mem.eql(u8, role, "sleeper")) {
         while (true) runtime.system.sleep(500);
     }
+    if (std.mem.eql(u8, role, "service")) {
+        // Borrowed well-known id: the input service is not part of this scenario.
+        runtime.service.run(64, .{
+            .service = .input,
+            .on_message = echo,
+            .on_reload = onReload,
+            .on_terminate = onTerminate,
+        });
+        return; // terminate arrived; returning is the clean exit
+    }
+    if (std.mem.eql(u8, role, "signal-run")) {
+        signalRun();
+        return;
+    }
     if (std.mem.eql(u8, role, "spinner")) {
         var beat: u64 = 0;
         const touch: *volatile u64 = &beat;
@@ -89,6 +169,12 @@ pub fn main(init: runtime.process.Init) void {
     if (listed(sleeper, "process-test")) fail("sleeper still listed after kill");
     if (listed(spinner, "process-test")) fail("spinner still listed after kill");
 
+    // M17.2: both children were killed by us, and the reason says so — the whole
+    // restart-policy input, read through the runtime like a real supervisor would.
+    if ((runtime.process.exitReason(sleeper) orelse .exited) != .killed) fail("sleeper reason not killed");
+    if ((runtime.process.exitReason(spinner) orelse .exited) != .killed) fail("spinner reason not killed");
+    if (runtime.process.exitReason(0xFFFF_FFF0) != null) fail("unknown id had a reason");
+
     _ = runtime.system.write("process-test: ok\n");
 }
 

system/services/vfs/vfs-test.zig

@@ -6,10 +6,30 @@
 const std = @import("std");
 const runtime = @import("runtime");
 
-pub fn main() void {
+pub fn main(init: runtime.process.Init) void {
     const u = @import("posix").unistd;
     const payload = "hello-vfs";
 
+    // The "park" role (the vfs-client-death test): open a file, then hold the
+    // handle forever without closing — the kill and the VFS's release-on-death
+    // are the point.
+    if (init.arguments.count > 1) {
+        var fd: i32 = -1;
+        var tries: u32 = 0;
+        while (fd < 0 and tries < 200) : (tries += 1) {
+            fd = u.open("parked", u.O_CREAT);
+            if (fd < 0) runtime.system.sleep(20);
+        }
+        if (fd < 0) {
+            _ = runtime.system.write("vfstest: park open failed\n");
+            return;
+        }
+        while (true) {
+            _ = runtime.system.write("vfstest: parked\n");
+            runtime.system.sleep(500);
+        }
+    }
+
     // The VFS server may not have registered yet — retry open until it's up.
     var fd: i32 = -1;
     var tries: u32 = 0;

system/services/vfs/vfs.zig

@@ -23,6 +23,10 @@ const Node = struct {
 const OpenFile = struct {
     used: bool = false,
     node: usize = 0,
+    // The client (task id — an IPC badge is one) that opened this handle. What
+    // release-on-death sweeps by: a service must never depend on its clients
+    // cleaning up after themselves (docs/process-lifecycle.md).
+    owner: u32 = 0,
 };
 
 var nodes = [_]Node{.{}} ** 8;
@@ -65,8 +69,29 @@ fn fail(out: []u8) usize {
     return writeReply(out, .{ .status = -1 }, &.{});
 }
 
-/// Handle one request; write the reply into `out`, return its length.
-fn handle(message: []const u8, out: []u8) usize {
+/// Format one whole log line and emit it in a single `debug_write`, so lines from
+/// concurrent processes can never land in the middle of it.
+fn writeLine(comptime fmt: []const u8, arguments: anytype) void {
+    var line: [96]u8 = undefined;
+    _ = runtime.system.write(std.fmt.bufPrint(&line, fmt, arguments) catch return);
+}
+
+/// Release every open handle `client` held — called on that client's published
+/// exit event. The nodes (the files) stay: ramfs contents outlive their writers,
+/// only the dead client's handles go.
+fn releaseClientHandles(client: u32) void {
+    var released: u32 = 0;
+    for (&opens) |*o| {
+        if (o.used and o.owner == client) {
+            o.used = false;
+            released += 1;
+        }
+    }
+    if (released != 0) writeLine("vfs: released {d} handle(s) for dead client {d}\n", .{ released, client });
+}
+
+/// Handle one request from `sender`; write the reply into `out`, return its length.
+fn handle(message: []const u8, out: []u8, sender: u32) usize {
     if (message.len < protocol.request_size) return fail(out);
     const request = std.mem.bytesToValue(protocol.Request, message[0..protocol.request_size]);
     const payload = message[protocol.request_size..];
@@ -77,7 +102,7 @@ fn handle(message: []const u8, out: []u8) usize {
             const ni = findNode(name) orelse createNode(name) orelse return fail(out);
             for (&opens, 0..) |*o, i| {
                 if (!o.used) {
-                    o.* = .{ .used = true, .node = ni };
+                    o.* = .{ .used = true, .node = ni, .owner = sender };
                     return writeReply(out, .{ .status = 0, .node = i }, &.{});
                 }
             }
@@ -113,27 +138,36 @@ fn handle(message: []const u8, out: []u8) usize {
     }
 }
 
-pub fn main() void {
-    const endpoint = runtime.ipc.createIpcEndpoint() orelse {
-        _ = runtime.system.write("vfs: no endpoint\n");
-        return;
-    };
-    if (!runtime.ipc.register(.vfs, endpoint)) {
-        _ = runtime.system.write("vfs: register failed\n");
-        return;
+/// Startup, under the harness: subscribe to the published exit events — when a
+/// client dies holding open handles, the exit notification is how the VFS learns
+/// to release them (docs/process-lifecycle.md).
+fn initialise(endpoint: runtime.ipc.Handle) bool {
+    if (!runtime.process.subscribeExits(endpoint)) {
+        _ = runtime.system.write("vfs: exit subscription failed\n");
     }
     _ = runtime.system.write("vfs: ready\n");
+    return true;
+}
 
-    var reply_buffer: [protocol.message_maximum]u8 = undefined;
-    var reply_len: usize = 0;
-    var receive: [protocol.message_maximum]u8 = undefined;
-    while (true) {
-        const got = runtime.ipc.replyWait(endpoint, reply_buffer[0..reply_len], &receive, null);
-        // Ignore notifications (none expected here); handle a request.
-        reply_len = handle(receive[0..got.len], &reply_buffer);
+/// A non-signal notification: the only kind the VFS subscribes to is exit events.
+fn onNotification(badge: u64) void {
+    if (badge & runtime.ipc.notify_exit_bit != 0) {
+        releaseClientHandles(@intCast(badge & ~(runtime.ipc.notify_badge_bit | runtime.ipc.notify_exit_bit)));
     }
 }
 
+pub fn main() void {
+    // The harness owns the loop: requests dispatch to handle(), exit events to
+    // onNotification(), ping and terminate are answered for free — this service
+    // gained the whole lifecycle contract by deleting its hand-rolled loop.
+    runtime.service.run(protocol.message_maximum, .{
+        .service = .vfs,
+        .init = initialise,
+        .on_message = handle,
+        .on_notification = onNotification,
+    });
+}
+
 pub const panic = runtime.panic;
 comptime {
     _ = &runtime.start._start;

test/qemu_test.py

@@ -240,6 +240,24 @@ CASES = [
      "smp":    4,
      "expect": r"DANOS-TEST-RESULT: PASS",
      "fail":   r"DANOS-TEST-RESULT: FAIL"},
+    # M17.1: a dead process's device claims are released by the reap — kill a child
+    # holding a claim, the device must be claimable again (process-lifecycle.md).
+    {"name": "claim-release",
+     "smp":    4,
+     "expect": r"DANOS-TEST-RESULT: PASS",
+     "fail":   r"DANOS-TEST-RESULT: FAIL"},
+    # M17.3: published exit events — the VFS subscribes, a client dies holding an
+    # open handle, and the VFS releases it (process-lifecycle.md "Who learns of a death").
+    {"name": "vfs-client-death",
+     "smp":    4,
+     "expect": r"DANOS-TEST-RESULT: PASS",
+     "fail":   r"DANOS-TEST-RESULT: FAIL"},
+    # M17.4: signals over IPC — ping, reload, terminate (clean exit), the one-shot
+    # timer, and the stop sequence's two endings, all driven from ring 3.
+    {"name": "signals",
+     "smp":    4,
+     "expect": r"DANOS-TEST-RESULT: PASS",
+     "fail":   r"DANOS-TEST-RESULT: FAIL"},
     # The initial_ramdisk: the loader ferries a bundle of user binaries; the kernel parses
     # it and spawns each as a ring-3 process (here the VFS-server stub heartbeats).
     {"name": "initial-ramdisk",