[hypermoist]

What

Implements sceVrTracker2Initialize in place of its NotImplemented stub. The behaviour is not measured on a console.

• Validation, in the module's order, all 0x8A740006: null record or size other than 72; mode other than 0, 1000 or 1001; null work memory; work memory below 32 MiB (mode 0), 7 MiB (1000) or 3 MiB (1001); alignment below 0x10000 (only a minimum, not a power of two); any reserved byte at 0x3a..0x47 set; either thread priority outside 256..767; affinity masks outside bits 0..12 (mode 0) or outside 0xEA00 (1000/1001). Then the module state moves from 0 to 1, else 0x8A740002.
• After that every failure undoes what the call opened and puts the state back to 0. Mode 0 clears the first 32 MiB of work memory first. The console's application attribute checks on the sceKernelGetAppInfo2 record (0x8A74001E, 0x8A740020) are skipped: the host has no such record, and a zeroed one passes them. The signed option at 0x39 accepts -1 and 0, and 1 only for titles compiled with SDK 7.00 or later.
• Modes 0 and 1000 then call libSceHmd2 TrackingOpen, TrackingStart, LedDetectorOpen and LedDetectorStart and return the first non-zero result as is. With no headset main's libSceHmd2 answers 0 and -1 (its measured pre-initialization answers, docs/dev/debt/libSceHmd2.md), so Initialize returns -1, the module stays uninitialized and every other export keeps answering 0x8A740001.
• Still throws: modes 1000/1001, which need the application category the host doesn't have, and a fully successful Hmd2 sequence, which would need the tracker runtime. The daemon IPC client and the work-memory heap aren't created on the host; neither is visible from the caller's side on the failing path.
• The eight tracking and LED-detector exports in libSceHmd2 become named functions (same NIDs, same answers) so libSceVrTracker2 can import them by name across the NID patcher.
• The debt entry is cut to what stays open.

Tested

Linux (CachyOS, GCC, Release): ctest --test-dir build -R 'guest_vr_tracker2|guest_hmd2$' --output-on-failure 2/2 passed.
Windows (MinGW-w64 GCC 15.2.0 winlibs cross build, run under Wine): ctest --test-dir build-win15 -R 'guest_vr_tracker2|guest_hmd2$' --timeout 120 --output-on-failure 2/2 passed. The NID-patched libSceVrTracker2.prx imports B0b8DeIUGes, 1oPZizn0SNA and the other six from libSceHmd2.prx, which exports them, on both platforms.
guest_vr_tracker2 covers each validation error in order, the -1 pass-through, the 32 MiB clearing in mode 0 (and that a validation failure writes nothing), a retry after the failure, sceVrTracker2Finalize and an uninitialized export still answering 0x8A740001, the late coordinate-system check coming after the Hmd2 error, and the 0x39 option against SDK 6.xx and 7.00.

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: PPSA05666 Horizon Call of the Mountain (PPSA05666) #6214 (the stop on sceVrTracker2Initialize is reached with feat(libSceVrTracker2): answer as the console's module before initialization #7463, feat(libSceHmd2): return the console's display and mirroring buffer sizes #7690 and feat(libs): the Hmd2 reprojection session #7658 applied)
• GPUs tested: n/a
• AI-assisted: yes

Assisted With AI: Yes. All tests were run locally on computer hardware (AMD Ryzen 5 5600G, 32 GB RAM, NVIDIA GeForce RTX 3060, CachyOS Linux 7.3, GCC 16.2), not on an actual console.

[github-actions]

Progress report (811ffe1 → 83043ee)

📉 System libraries: 49.26% (-0.04%, +8 declared functions)

🆕 8 declared as stubs

Library	Name
libSceHmd2	sceHmd2InternalLedDetectorClose
libSceHmd2	sceHmd2InternalLedDetectorOpen
libSceHmd2	sceHmd2InternalLedDetectorStart
libSceHmd2	sceHmd2InternalLedDetectorStop
libSceHmd2	sceHmd2InternalTrackingClose
libSceHmd2	sceHmd2InternalTrackingOpen
libSceHmd2	sceHmd2InternalTrackingStart
libSceHmd2	sceHmd2InternalTrackingStop

Changed libraries (implemented exports)

Library	Before	After	Change
libSceHmd2	13/126 (10.32%)	13/134 (9.70%)	-0.62 pp

[MarcZX10]

One test coverage gap: WorkMemoryCleared checks a 4 KiB window centered at 16 MiB plus the tail after 32 MiB. That shows the clear reaches the middle, but a 16 MiB clear could still pass, and the first bytes of the cleared range and the byte immediately before the boundary are not checked. Could you assert the start and last byte of the 32 MiB range plus the first tail byte, or scan the whole expected range? The production memset length looks correct; this would make the reported 32 MiB boundary coverage precise.

[hypermoist]

@MarcZX10 good catch. Done in 16c9c31: WorkMemoryCleared now checks every byte of the 32 MiB range for 0 and every byte of the tail after it for the 0xAA fill, so the start, the byte before the boundary and the first tail byte are all covered. As a check, clearing half the size or one byte short of 32 MiB now makes guest_vr_tracker2 fail; the previous 4 KiB window passed both. With the production memset unchanged it passes on Linux and under Wine (WinLibs GCC 15.2.0), and the patched libSceVrTracker2.prx builds.

Assisted With AI: Yes But all test done locally on actual hardware

[MarcZX10]

The follow-up closes the coverage gap I raised: WorkMemoryCleared now checks every byte of the 32 MiB clear region and the entire sentinel tail, so the start, end, and boundary are covered. I reviewed the current implementation and test diff, including ordered validation, the state transition and rollback on Hmd2 failures, retry behavior, and the documented host-only limitation. The author reports the focused vr|hmd suite passing 8/8 on Linux and under Wine/MinGW; the latest hosted eligibility, conventions, labels, and report checks pass, while platform build jobs were skipped by the eligibility gate. I did not rerun those test suites locally.

[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 7658	core/libs/prx/libSceHmd2/Export.cpp
PR 8645	core/libs/prx/libSceHmd2/Export.cpp

[MarcZX10]

Reviewed exact head refs/review/pr-10566 ($(git rev-parse refs/review/pr-10566)). The recovered parameter checks precede the state transition and work-memory clearing; initialization claims the state atomically, tracks successful Hmd2 open/start operations, rolls them back in reverse order, and restores the uninitialized state before returning the downstream result or throwing for the still-unsupported system modes. The tests cover validation boundaries, memory clearing, retry after failed start, and SDK-gated option mode. git diff --check is clean; hosted eligibility, conventions, labels, and report pass. All hosted platform build jobs are skipped, so this approval relies on source review and the contributor’s reported local tests; no local build was run.

[oneandonlydean]

@hypermoist Please drop the new system-mode throw cases in GuestVrTracker2.cpp:96-101; those modes intentionally remain unimplemented and the test should not pin that limitation. Keep the invalid-argument, full memory-clear/tail, SDK-option and failure/retry checks. EnterApplicationRoot also removes a predictable PID-based temporary path before creating it, so PID reuse can delete an earlier run's files. Please allocate a fresh directory atomically and clean up only the directory this run created, preserving pre-existing paths. Then rerun guest_vr_tracker2 and guest_hmd2 on Linux and Windows.

[hypermoist]

@oneandonlydean thanks, addressed in 83043ee:

• Dropped the system-mode throw cases in GuestVrTracker2.cpp that tested modes 1000 and 1001 throwing. All invalid-argument, memory clearing, SDK option, failure and retry checks remain.
• EnterApplicationRoot now allocates a unique temporary directory atomically using a random suffix without removing any pre-existing directory, and cleans up only the directory created for the test run.
• Rebuilt and passed guest_vr_tracker2 and guest_hmd2 on Linux and under Wine (WinLibs GCC 15.2.0 cross-build). check_conventions.py is clean.

Assisted With AI: Yes But all test done locally on actual hardware