Update: same stall on macOS 27.0.1 (26A434)
Retested after updating the host from 26.7.1 (25G309) to 27.0.1 (26A434). Same M4, same VM configuration and storage, UTM front-end, guest macOS 26 (build 25G83) updating in place. This time it was the only VM running, and the host logged no AppleAVD errors.
Observed:
The in-OS phase ran ~20 minutes at ~2 cores, then the guest rebooted twice, 65 s apart. After the first reboot a display mode was set within 3 s.
After the second reboot the host logged PGDisplayNub[0]: Created and no display-mode change afterwards (through the end of my log capture, ~7 minutes later). The window stayed black.
Two spindumps of the VM host process, 11–12 minutes after the second reboot, show the six PGFifoThreads last ran within a few ms of that nub creation and not since. All four vCPU threads were active, alternating roughly 80/20 between wait_for_interrupt and guest execution (~0.5–0.75 core in total), so this is not the parked-in-WFI state I saw on 26.7.1. Neither 10 s spindump sampled any DiskImages2/preadv/pwritev frames; that is sampling evidence only, not proof of zero I/O.
Lifetime counters for the VM process at capture: ~585 GiB read, ~41 GiB written (no baseline to compare against).
So: a post-reboot guest stall with inactive graphics workers and continuing vCPU activity. Whether the guest failed to initialize its graphics device or stalled elsewhere and simply never used it, I can't tell from the host side.
Outcome: killing the stalled VM and cold-starting it boots the previous system (26.6.2, 25G83), and Software Update offers 26.7.1 again. The patched system is sealed (patchd SUCCESS, seal result=0, ota-result=success in NVRAM) but never completes its first boot ("Boot 1"), and the VM falls back to the prior version.
