hybrid: the driver is killed, and the driver comes back

H2 of docs/plan/hybrid.md -- the milestone the plan itself calls the one
that proves the whole idea, and the one that cannot be skipped.

    # cat /proc/schinit
    1 4 11 os.schos.vgpud
    # kill 4
    schinitd: restarting os.schos.vgpud
    vgpud: claimed
    vgpud: pid 8 drew a test pattern from userspace

The test pattern ends in a twelve-block tally of the drawing process's
pid, because a second run that redraws an identical picture is
indistinguishable from a first run that never stopped. Read off the
virtio-gpu scanout across four generations it says 4, 8, 10, 12, and the
DMA arena hands back the same physical address every time.

What it took:

* A halt the kernel can apply to a device it does not understand.
  Clearing PCI_CMD_BUS_MASTER is the only one there is, and it has to
  happen before the DMA goes back to the arena -- a dead driver's device
  keeps reading a framebuffer the kernel is about to give to somebody
  else. Claim sets the bit, release clears it; pcitest shows it in a bm
  column, with a third state so a failed read is never printed as off.

* A refusal. vgpu_release() asks vga_fb_restore() whether there is a
  framebuffer to put the console back on, and declines to hand over the
  device if there is not. The rule that the kernel keeps a minimal
  driver for everything on the path to a shell is worth writing down
  only if something enforces it.

* The DMA arena not being owned by two allocators at once. This is the
  bug that stopped H2: freevm() frees every mapped page below PHYSTOP
  and the arena is the top 16 MB of exactly that range, so a DMA block
  mapped into a driver went onto the page allocator's free list when the
  driver died, while pciuser_release_all() handed the same pages back to
  the arena. The restarted vgpud asked for its framebuffer, got pages
  kalloc() had already given out -- including its own page tables -- and
  dma_alloc()'s zeroing wiped them, so it died inside the ioctl on the
  return from a syscall whose stack had stopped existing. do_dma() now
  records a VMA with no_free set, the way shared memory does.

* One exit status. KeepAlive must restart a driver that crashed and must
  not restart one that has looked at the machine and found its device
  absent: on hardware with no virtio-gpu that was fifteen spawns in
  forty-five seconds, each printing a line to the console, for ever.
  SCHINIT_EXIT_NORESTART is 78, sysexits.h's EX_CONFIG. It is not a new
  manifest key -- the wait status has always travelled from PID 1 to the
  supervisor, and note_exit() simply stopped discarding it.

The milestone's third clause, "the compositor reattaches", is not done
and is not claimed: there is no compositor running. That is recorded in
the plan as a Schcom milestone rather than counted here.

Closes the register row "A released PCI device" in docs/plan/1.0.md.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VQ7Qx8E1q1M1bbCq8QLHS2

docs/architecture/devices.md

@@ -113,6 +113,44 @@ console falls back to the VBE framebuffer.
 **No syscall on that path knows it is talking about a GPU.** The kernel hands
 over a PCI device; the process decides what it is.
 
+### Taking it back
+
+A driver that dies is the case this whole arrangement exists for, so release is a
+symmetric operation rather than a courtesy. Three things happen that are not
+obvious from "the process exited":
+
+**Bus mastering is cleared before the DMA is freed.** A device whose driver has
+just died is still writing into memory the kernel is about to hand to somebody
+else. The kernel cannot quiesce a device it does not understand — that is a
+different register on every chip — but it can clear `PCI_CMD_BUS_MASTER`, which
+stops the writes whatever the device is, and it is the only halt available to a
+kernel that knows nothing but the header. So `PCIIOC_CLAIM` sets the bit and
+release clears it, in that order relative to freeing the memory. `pcitest` prints
+the bit in a `bm` column, with a separate mark for "could not read it", because a
+failed read rendered as "off" is worse than no column at all.
+
+**DMA memory a driver mapped is not `freevm()`'s to release.** `freevm()` frees
+every mapped page below `PHYSTOP` and the DMA arena is the top 16 MB of exactly
+that range, so a DMA block left for it went onto the page allocator's free list
+while `pciuser_release_all()` handed the same pages back to the arena — two
+allocators, one set of pages. `PCIIOC_DMA` records a VMA with `no_free` set, the
+same way shared memory does, and `proc_release_shared()` unmaps the range before
+the page tables are torn down. The same rule covers every future case: **anything
+mapped into a process that the process does not own the pages of needs a VMA, or
+exit will free it twice.**
+
+**A device that is the only display will not be released.** `vgpu_release()` asks
+`vga_fb_restore()` whether there is a framebuffer to put the console back on, and
+declines if the answer is no. The rule that the kernel keeps a minimal driver for
+everything on the path to a shell is only worth writing down if something
+enforces it.
+
+Coming back is a job, not a kernel feature. `etc/schinit/vgpud.job` carries
+`KeepAlive = yes`, so killing the driver is the entire test: `schinitd` starts
+another one, it claims the device, and the pid of whichever process drew the
+frame is legible in the frame itself. See
+[../plan/hybrid.md](../plan/hybrid.md), "The driver came back".
+
 ### What this is not
 
 **It is not a security boundary.** There is no IOMMU, so a process holding a DMA
@@ -170,9 +208,11 @@ existing mechanisms for keeping that true mechanically rather than by intention.
   kernel is an ordinary program and can be started whenever — but it reaches its
   device through `/dev/pci` and publishes nothing to the class vtables, so the
   console cannot be handed to one yet.
-- **A released device does not come back by itself.** `PCIIOC_RELEASE`, or the
-  process exiting, leaves the interrupt line masked and the in-kernel driver
-  unbound. Something has to start a replacement; nothing watches for one.
+- **The kernel does not rebind its own driver.** `PCIIOC_RELEASE`, or the process
+  exiting, leaves the interrupt line masked and the in-kernel driver unbound —
+  the console stays wherever the release put it. A *supervisor* can start a
+  replacement driver, and `schinitd` does; the kernel itself watches for
+  nothing.
 - **Only two class vtables are defined**: `display_ops` and `netdev_ops`.
   Storage and input devices appear in the list with `ops = 0` and are reached
   through their own subsystems instead.