[Scramblrrrr]

What

A small committed access in a large registered allocation can lose host import because its 64 MiB import window also includes inaccessible pages. Main rejects the entire window even when the actual GPU access is wholly committed. The synthetic Vulkan case reserves 128 MiB, commits 64 KiB and fills 4 KiB: main refuses the fill; this branch imports the committed prefix and preserves the GPU-written pattern.

windowRun now checks commitment after its usual cell selection and overlap growth. For a partially accessible window it selects the aligned committed span containing the whole requested access. It keeps the original refusal path if the span cannot cover the access or narrowing would cut an overlapping import. Existing held-import, registry-owner, budget and device-import checks still apply. Whole-run callers and registrations no larger than the window retain their existing policy.

Related: #6442 caches rejected windows, #6097 changes probe/hold accounting, #6182 retires sibling windows on unmap, #6457 combines adjacent registered blocks, #5372 joins crossed registrations. This is the single-registration/partially committed window case; no dependency on those branches. Coordination: 007 investigation.

Tested

Windows 11 Pro, NVIDIA RTX 5070 Ti, driver 617.42, MinGW GCC 15.2. Release, SPIRV-Tools OFF. Clean main 93e54ce plus this branch; no other title fixes.

• 7/7 pass, no skips, 6.95 s after restoring and rebuilding the fixed production source: agc_driver_graphics, agc_driver_host_import_windows, agc_host_import_unmap, agc_host_import_retirement, agc_sparse_run_import, agc_sparse_run_import_copied, agc_host_import_second_mapping.
• Meaningful negative: replace only GuestBufferMemory.cpp with main93e54cedd, keep both final fixtures, rebuild. Both new cases fail (1.10 s): a committed access was refused because its import window includes uncommitted pages; host import windows: cannot fill the committed prefix. Restored source/rebuilt: 7/7 above.
• Mock fixture covers repeat reuse, real-hole rejection, separate committed islands, held-prefix refusal without a recorder, then growth in the next build. Real Vulkan fixture requires imported coverage and verifies the CPU-visible GPU fill pattern.
• Conventions and diff checks pass. Full upstream CI not run.
• Independent Linux cross-test by SleglJan: head 955e7c3c7 on main 93e54cedd, Arch Linux container, GCC 16.2.1, Mesa 26.2.4 lavapipe/LLVM 23.1.1, Release. Ten host-import/graphics/BDA tests: nine pass, sparse-import control skips. Restoring only main's production file makes both new assertions fail; the other eight keep their pass/skip result. userfaultfd is unavailable, so this container uses the memory-comparison tracking fallback. No hardware GPU or title result.
• Independent AMD Windows cross-test by keravyn-ops: head 955e7c3c7, Radeon 780M, AMD 25.10.20.01, Windows 11 26100, GCC 15.2 Release. Host-import windows passes (0.44 s), graphics passes (5.14 s).

Investigated during PPSA11386 bring-up. No title FPS improvement or title unblock claimed: the title runtime test of this candidate is pending. Keeping this draft until that cross-test and review of commitment-query cost. Raw title assets, requests, descriptors and logs stay local.

Checklist

• Based on current main; no other open PR implements the same functions (related work listed above; collaborator independently checked overlap)
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw; no new stubs
• No notes, investigation .md files or images added to the repository
• New third-party code is a submodule built from source (none added)
• Generic behaviour, not specific to one title
• Depends on: none
• Unblocks: none confirmed; investigation for PPSA11386, discussion 007 First Light (PPSA11386) #5714
• GPUs tested: NVIDIA GeForce RTX 5070 Ti, driver 617.42, Windows 11 Pro
• AI-assisted: yes (Codex)

[Scramblrrrr]

Title diagnostic qualification before a user-requested pause: the original dominant12.5MiB copied span was misclassified by an experimental sampler checking only its first4bytes. It crosses2MiB registrations, so it is not title evidence for this single-registration fix. Corrected findings: #5714 (comment) . This PR's clean-main synthetic negative/positive and real Vulkan fill remain valid independently (7/7 restored, two new fixtures fail with only production restored to main).

This fix was included in title integrationc70803a19 with #5372 plus an opt-in16MiB window. That combined stack renders the offline menu at1.80194FPS, but no isolated benefit is attributed to this PR. Still draft; Linux and commitment-query-cost review remain pending. No new runs will start until the user resumes. AI-assisted: Codex.

[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 1191	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 2434	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 4448	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 4449	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 5021	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 5372	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 6146	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 6182	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 6185	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 6186	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 6197	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 6292	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 6412	core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 6481	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 6548	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 6615	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 6628	core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 6636	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 6682	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 6913	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 6947	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 6951	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 7201	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 7233	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 7261	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 7292	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 7371	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 7446	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 7457	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 7465	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp core/libs/prx/libSceAgcDriver/tests/execution/HostImportWindows.cpp
PR 7466	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp core/libs/prx/libSceAgcDriver/tests/execution/HostImportWindows.cpp
PR 7514	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 7618	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 7622	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 7629	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 7642	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 7647	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 7717	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 7844	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 7849	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 7919	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 7961	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 8016	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 8091	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 8104	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 8126	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 8279	core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 8336	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 8380	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 8527	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 8622	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 8658	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 8667	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 8672	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 8762	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 8874	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 8895	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 8896	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 8926	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 8963	core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 8971	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp core/libs/prx/libSceAgcDriver/tests/BdaResources.cpp
PR 9011	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp
PR 9012	core/libs/prx/libSceAgcDriver/Graphics/src/GuestBufferMemory.cpp

[SleglJan]

Linux cross-test of 955e7c3 on main 93e54ce. Setup: Arch Linux container, GCC 16.2.1, lavapipe (Mesa 26.2.4, LLVM 23.1.1), Release, BUILD_TESTING=ON. Test command: ctest -R 'host_import|driver_graphics|bda', 10 tests.

• This head: 9 pass, 1 skipped (agc_host_import_sparse).
• Same tests with only GuestBufferMemory.cpp taken from main: 2 fail, both at the new cases.
  • agc_driver_graphics fails with a committed access was refused because its import window includes uncommitted pages.
  • agc_driver_host_import_windows fails with host import windows: cannot fill the committed prefix.
  • The other 8 pass or skip as with this head.
• userfaultfd is not permitted in this container, so the write watch falls back to comparing guest memory (logged by both failing tests in the second run; passing tests print no output).

Not run: a hardware GPU, the full suite, any title.

AI-assisted: yes (Claude Code)

[keravyn-ops]

AMD / Windows run of this PR at 955e7c3: Radeon 780M (RDNA 3, 0x15bf), AMD driver 25.10.20.01 (32.0.21020.1007), Windows 11 Home 10.0.26100, WinLibs GCC 15.2.0 r7, Release.

• agc_driver_host_import_windows: pass (host import windows tests passed, 0.44 s)
• agc_driver_graphics: pass (5.14 s)

AI-assisted: yes (runs made on the device above).

[Scramblrrrr]

Thanks @SleglJan and @keravyn-ops for the independent Linux red/green and AMD Windows runs. I have recorded both in the PR description. The Linux comparison isolates the intended production failure at both new assertions, with the container write-watch limitation retained; the AMD run covers actual host imports on Windows. This remains draft pending the import-query cost review. Our current 007 integration passes its selected host-import/graphics controls, but no isolated title FPS benefit is claimed for this fix. AI-assisted: Codex.