[xXPBSXx]

What

Pm4::Validate rejects a REWIND packet whose payload has any bit other than bit 31 set (unsupported REWIND payload bits, Pm4.cpp). PPSA15246 writes 0x81000000: bit 31 (valid) and bit 24. shadPS4's PM4CmdRewind (src/video_core/amdgpu/pm4_cmds.h) describes the payload as two fields, bit 31 valid ("subsequent packets are valid") and bit 24 offload_enable (lets the CP offload polling of the valid bit to the IQ), and its Valid() tests bit 31 only. KytyPS5's CpOpRewind (src/graphics/guest_gpu/command_processor/pm4Handlers.cpp) accepts exactly bits 31 and 24 of the payload, stops with a not-implemented error on any other bit, and waits on bit 31 only. The driver here also waits on bit 31 only (executeRewindTail), so bit 24 needs no handling beyond being accepted.

This lets the validator accept bit 24 together with bit 31 and keeps rejecting every other bit. New testRewindPayloadBits cases in agc_driver_pm4_tests: 0x00000000, 0x80000000, 0x01000000 and 0x81000000 validate; 0x82000000, 0x80000001 and 0x00800000 are still rejected.

Title evidence: PPSA15246 (Teardown), Windows 10, main 3ddbc4acc plus an unmerged local change that lets a zero header word through (run_062434.err, line 149): driver failure: AGC driver: REWIND at DWORD 190: unsupported REWIND payload bits, dump dwords c0005900 81000000. On unmodified main this title stops earlier, at unsupported PM4 packet type: header 0x00000000 at DWORD 15, in front of a 169-dword prefix of zeros; that stop is not changed here, so this PR alone does not get Teardown further. It removes the next one.

Tested

Windows 10 (10.0.19045.6466), MinGW-w64 GCC 15.2.0 (WinLibs), Release, on main 7a36de66b9 + this commit (rebased afterwards onto 77b3edf32a as c0b68e443; no commit in between touches the changed files):

• ctest -N -R '^agc_driver_pm4$' lists exactly one test; ctest -R '^agc_driver_pm4$' --no-tests=error: passes 3 of 3 (testRewindPayloadBits included).
• With Execution/src/Pm4.cpp restored to main (git checkout upstream/main -- <file>, git status shows only that file staged, 1 insertion / 1 deletion; the new test kept): the same ctest entry fails 3 of 3 (ctest exit code 8).
• python tools/check_conventions.py --base upstream/main with this title and body passes.
• No GPU or Vulkan device is involved (command-stream validation only), and no Khronos validation layer is installed on this machine.
• Not tested: Linux, a title run past the zero-header stop in front of the REWIND (see What).

Checklist

• Based on current main; no other open PR implements the same functions
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw (NotImplemented_nid_no_patch); silent stubs are listed under Silent stubs in their library's technical debt file
• No notes, investigation .md files or images added to the repository (attach them to this PR)
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: none
• Fixes: none
• Unblocks: none yet (PPSA15246 stops earlier at the zero-filled command prefix; no Game bring-up discussion exists for it)
• GPUs tested: n/a (command validation only)
• AI-assisted: yes

AI-assisted (Claude Code).

🤖 Generated with Claude Code

[github-actions]

Some comparisons could not be completed because a pull request conflicts with the base branch. Review these shared files manually before concluding that there is no overlap.

Pull request	Shared files not compared
PR 6186	core/libs/prx/libSceAgcDriver/tests/Pm4.cpp
PR 6188	core/libs/prx/libSceAgcDriver/Execution/include/Pm4.hpp core/libs/prx/libSceAgcDriver/Execution/src/Pm4.cpp
PR 6549	core/libs/prx/libSceAgcDriver/Execution/include/Pm4.hpp core/libs/prx/libSceAgcDriver/Execution/src/Pm4.cpp core/libs/prx/libSceAgcDriver/tests/Pm4.cpp
PR 9893	core/libs/prx/libSceAgcDriver/Execution/src/Pm4.cpp core/libs/prx/libSceAgcDriver/tests/Pm4.cpp

[fogydes]

Verified on Windows 11 with MinGW-w64 GCC 15.2.0 (Release):

• Architecture & Diff: In core/libs/prx/libSceAgcDriver/Execution/src/Pm4.cpp, allowing RewindOffloadEnable (0x01000000u, bit 24) alongside RewindValid (0x80000000u, bit 31) in Pm4::Validate aligns packet validation with CP offload polling semantics seen in PPSA15246 without altering driver wait behavior in executeRewindTail.
• Unit Testing: Verified testRewindPayloadBits in core/libs/prx/libSceAgcDriver/tests/Pm4.cpp. Built and ran ctest --test-dir build-libs -R "^agc_driver_pm4$" -V passing in 0.24s:

  1/1 Test #422: agc_driver_pm4 ...................   Passed    0.24 sec
  

  All catalog, register, state, memory, and conditional execution fixtures passed.
• Conventions: No code comments added; tools/check_conventions.py passes cleanly.

[savvaniss]

AMD and Intel Windows data: with this PR's test changes, agc_driver_pm4 fails on main and passes with the PR, on an AMD RX 9070 XT and an Intel UHD 770.

Windows 11 Pro 26300, WinLibs GCC 15.2.0 r7, Release, every execution test built, ANYPS5_NO_SHADER_CACHE=1, serial CTest. Intel UHD Graphics 770 (8086:a780), driver 32.0.101.7088. AMD Radeon RX 9070 XT (1002:7550), driver 32.0.31041.1004, with VkCache emptied first. This head c0b68e443 merged into main 85029a00f. "Without" means only this PR's diff to tests/Pm4.cpp, applied to main 85029a00f.

Tests, ×3	UHD 770	RX 9070 XT
This head: the 13 tests built from tests/Pm4.cpp	13 of 13 pass	13 of 13 pass
Without	agc_driver_pm4 fails: unsupported REWIND payload bits; the other 12 pass	same

AI-assisted: yes (Claude Code)

[adriadam10]

Checked the validator change against shadPS4 PM4CmdRewind and KytyPS5 CpOpRewind (bits 31 and 24 only), the driver wait on bit 31 only, and the new testRewindPayloadBits rows (valid and still-rejected payloads). Re-checked on current main after #6794: the patch applies cleanly and the REWIND line is unchanged. Third-party runs on Windows (fogydes, savvaniss) confirm the test fails on main and passes with the change. Checked on the merge queue full Linux and Windows build; merged.