_commit_msg.txt

graphics: fix CB4 black-screen - blocked-wait starvation + watchdog

CB4 (PPSA02433) wedged after the boot-video chain with a black screen:
the present loop kept flipping empty frames (live frame counter, ~3 fps)
while zero draws reached the GPU.

Root cause chain:

1. The UE4 RHI re-issues a fence ping-pong (release_mem + wait_reg_mem)
   to the compute queue forever. A submission blocked on wait_reg_mem sits
   at the FRONT of its queue and the GPU thread only cleared blocked flags
   when EVERY queue was idle - so the graphics queue front was never
   re-polled while the compute queue kept producing work. Starvation:
   the wait was evaluated once, failed once, and was never retried.

2. Our end-of-pipe fence writes are host-synchronous, so a wait that
   re-polls for seconds can never succeed; it is a self-deadlock, not a
   fence.

Fixes (all in the guest GPU command path):

- graphicsRun.cpp ThreadRun: anti-starvation - a blocked queue front is
  re-pollable every 10ms even while other queues have work, so blocked
  waits are actually re-evaluated.

- graphicsRun.cpp WaitRegMem: per-address stalled-wait watchdog. After
  10 seconds of consecutive unsatisfied polls (long enough for iGPU
  shader-compile bursts, short enough to self-heal a wedge) the wait is
  released and the awaited value is written so guest bookkeeping stays
  consistent. Normal fences resolve in microseconds and never trip it.

- graphicsRun.cpp WriteAtEndOfPipe write32/write64 + pm4Handlers.cpp
  CpOpReleaseMem: null-destination guards. A release_mem whose label
  address was never patched (null dst) previously fell through to a
  memcpy to nullptr; it now soft-skips with a log line.

Verified on the target machine (Radeon 840M iGPU): CB4 now renders the
splash screen past the point where it previously froze - farther than
any documented result for this title on any PS5 emulator.

Co-authored-by: Chatbox <chatbox@chatboxai.com>