research: VZ plan — vtgpu spin fixed, fbdev gap closed, PSCI tracer is all that's left

Two of the three blockers this page tracked are now closed in tree, and
neither of them was the boot hang:

- the vtgpu attach spin (§4 #4) is fixed by kernel patch 0011, which bounds
  the control-queue wait and calls virtio_stop() on timeout rather than
  spinning a vCPU forever.  No nodevice needed.
- the fbdev gap (§6) is closed outright rather than merely "smaller than we
  thought".  drm_fbdev_shmem is vendored, built and on by default, and patch
  0012 supplies the DMA tags whose absence page-faulted the first
  dma_map_sgtable().  A stock arm64 boot now reaches card0 with a drmfb
  console unattended.

§7 is reversed rather than qualified: V7 and Option D are withdrawn for
virtio-gpu.  Base virtio_gpu(4) stays compiled in as the FIRST console --
on a VZ guest it is the only one -- and the kext takes the device from it
via an atomic detach/re-probe.  Records the measurement that makes the
atomicity non-negotiable: detaching whichever driver owns the console
panics the box, `clear driver -f` included.

What is left for VZ is exactly one thing, and it needs no console: the PSCI
tracer.  Phase 2 is restated to say so, and Phase 3 is marked done on qemu
with the VZ half explicitly flagged as reasoned rather than measured.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

Author: pkgdemon