[ycarogabriel-lgtm]

What

Keep synchronous host fault signals deliverable while scePthreadCreate creates a guest thread, and unblock them when the guest thread starts on macOS. The guest signal mask remains emulated by libkernel. This prevents a guest-created thread under Rosetta from inheriting a host mask that hides SIGBUS/SIGSEGV write-watch faults.

The focused test checks that the creator's host mask is restored and the guest worker can receive fault signals. This extracts the issue #1 finding by @Cloud-Multiplayer from the former integration PR #7332. No title run is claimed for this branch.

Tested

• Merged current main through f2d874c1fd in adefe80eea; resolved the Thread.cpp conflict while preserving both Windows pending-exception delivery and the macOS fault-signal mask behavior.
• python3 tools/check_conventions.py --base origin/main: passed.
• ctest --test-dir build-mac -R '^public_headers$' -j1 --output-on-failure: passed.
• The M4 Pro, M1 Pro and M2 Pro integrated macOS test reports below confirm the focused regression test and related pthread/write-watch paths pass with this change.
• A standalone macOS build against current main remains blocked by existing host portability gaps: pthread_getattr_np and pthread_getcpuclockid in Thread.cpp, st_atim/st_mtim/st_ctim in NativeStat.cpp, MFD_CLOEXEC in SharedMemory.cpp, and sys/vfs.h in Stdio.cpp. These errors prevent building the focused test on this base; the test has not been rerun locally after the merge.
• Linux and Windows were not tested locally. No GPU code changed.

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; silent stubs are listed in technical debt
• No notes, investigation .md files or images added to the repository
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: none for this isolated signal-mask fix
• Fixes: none
• Unblocks: write-watch progress in Stray on the integrated macOS build reported by @imellag (a later depth-bounds limitation remains)
• GPUs tested: n/a
• AI-assisted: yes

[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 3986	core/libs/prx/libkernel/Pthread/src/Thread.cpp
PR 4563	core/libs/prx/libkernel/CMakeLists.txt core/libs/prx/libkernel/Pthread/src/Thread.cpp
PR 5021	core/libs/prx/libkernel/CMakeLists.txt core/libs/prx/libkernel/Pthread/src/Thread.cpp
PR 6936	core/libs/prx/libkernel/CMakeLists.txt
PR 7149	core/libs/prx/libkernel/Pthread/src/Thread.cpp
PR 7192	core/libs/prx/libkernel/CMakeLists.txt
PR 7804	core/libs/prx/libkernel/Pthread/src/Thread.cpp
PR 8592	core/libs/prx/libkernel/Pthread/src/Thread.cpp
PR 9019	core/libs/prx/libkernel/CMakeLists.txt
PR 9156	core/libs/prx/libkernel/CMakeLists.txt
PR 9293	core/libs/prx/libkernel/CMakeLists.txt
PR 9297	core/libs/prx/libkernel/CMakeLists.txt
PR 9302	core/libs/prx/libkernel/CMakeLists.txt
PR 9383	core/libs/prx/libkernel/CMakeLists.txt
PR 9389	core/libs/prx/libkernel/CMakeLists.txt
PR 9557	core/libs/prx/libkernel/CMakeLists.txt
PR 9566	core/libs/prx/libkernel/CMakeLists.txt
PR 9938	core/libs/prx/libkernel/CMakeLists.txt core/libs/prx/libkernel/Pthread/src/Thread.cpp
PR 9947	core/libs/prx/libkernel/CMakeLists.txt
PR 11625	core/libs/prx/libkernel/CMakeLists.txt
PR 11831	core/libs/prx/libkernel/CMakeLists.txt
PR 11896	core/libs/prx/libkernel/CMakeLists.txt
PR 11939	core/libs/prx/libkernel/CMakeLists.txt
PR 12544	core/libs/prx/libkernel/CMakeLists.txt core/libs/prx/libkernel/Pthread/src/Thread.cpp
PR 12642	core/libs/prx/libkernel/Pthread/src/Thread.cpp
PR 14299	core/libs/prx/libkernel/CMakeLists.txt

[imellag]

Ran the new test standalone on Apple silicon, since the description says it could not run on this branch alone.

Machine: Apple M4 Pro (24 GB), macOS 27.0.1 (26A434), x86-64 under Rosetta 2, Apple clang 21.0.0, CMake 4.4.4, Ninja 1.13.2, Release, CMAKE_OSX_ARCHITECTURES=x86_64, no Vulkan loader.

Base: main does not build the libraries on macOS, so I used mugurc's ci/macos-series-2 at 53db0badb, plus the three code commits of #7937 (c6f44a028, ca0068033, 4e5dba2db: without them DirectMemory/Jit.cpp stops on MFD_CLOEXEC), plus this PR's f22b57083 cherry-picked (applies cleanly).

	guest_pthread_host_fault_mask	full suite (ctest -j6, 1649 tests)
with f22b57083	passed 11/11	17 failed
Thread.cpp reverted to the base, test kept	aborted 5/5	19 failed: the same 17, plus guest_pthread_host_fault_mask and precise_wait

The 17 fail without the fix too (and again when rerun one by one with it), so they come from the base, not from this change. precise_wait is a timing test that is flaky on this host under load.

The 17 failures common to both runs

guest_math (does not compile: libc++ has no std::cyl_bessel_j/std::cyl_neumann), guest_math_exact, guest_standard_streams, agc_write_tracking ("a block the import window did not touch reads as written"), agc_performance_overlay_statistics, guest_sce_net, guest_ssl_transport_kernel, guest_ssl_transport_net, guest_memory_lifetime, guest_filesystem, guest_host_standard_error, guest_itimer (SIGALRM delivery to a guest handler not implemented), guest_shm_open, guest_poll, guest_socket_pair, guest_kernel_app_archive_mmap, guest_flexible_memory_declared.

Not checked: a title with and without the fix (Planet of Lana stops on this base at a pixel shader that reads barycentrics, before any write-watch path matters), Linux and Windows (the change is __APPLE__ only), and the GPU suite without the fix (with it, 761 of 807 AGC and shader tests pass on MoltenVK 1.4.2, but I have no run without it to compare).

AI-assisted: yes

[imellag]

A title that needs this fix: Stray (PPSA02100, v01.002.000, Unreal Engine 4) hangs twice on macOS without it, and gets past both with only f22b570 added.

Machine: Apple M4 Pro (24 GB), macOS 27.0.1, x86-64 under Rosetta 2, MoltenVK 1.4.2 (Vulkan SDK 1.4.363.0).
Build: #8658 (0a6c0fd) + #9193 + #9263 + the macOS RTLD_GLOBAL loader change now in #8658, Release.

	without f22b570	with f22b570
1st stop, about 10 s in	hangs at 0% CPU: the main guest thread is in RegisterShader → VulkanDevice → PrepareImportWatch → decideImportWatch, during the import write-protection probe	probe finishes: host imports resolve write protection: no (0 of 4 scratch pages ...)
2nd stop, with the probe skipped (APS5_WRITE_WATCH_IMPORTS=unwatch), after about 3,400 shaders are prepared	hangs at 0% CPU: the packet worker is in Driver::preparePacketMemory → GuestMemory::Write → storeOwn, inside the store to guest memory	no hang
result	no further progress in 120 s	runs on to the next, unrelated stop: device does not support the depth bounds test (MoltenVK has no depthBounds)

Both hangs are a plain store to a write-watched page that never returns, which is what the description explains: the threads inherit a mask with SIGSEGV/SIGBUS blocked, so the fault is never delivered. Each hang reproduced in two runs; the stacks are from sample. God of War (PPSA28997) and Planet of Lana (PPSA19450) reach gameplay on the same stack without this fix, so Stray is the first title I have that depends on it.

Not checked: the depth bounds stop and anything after it, Linux and Windows, the full CTest suite.

AI-assisted: yes

[maximilianfv]

Tested on Apple M1 Pro, macOS 27.0.1 (x86_64 build under Rosetta 2)

Applied at 9ea069639 to mugurc's macOS series (27e968974) with other open macOS fixes. "Without" keeps this PR's new test and leaves out only the Thread.cpp change.

	guest_pthread_host_fault_mask	other guest_pthread*, *write_watch*, guest_raise_exception* (39 tests)
Without the Thread.cpp change	Subprocess aborted: Require at GuestPthreadHostFaultMask.cpp:17 in CheckMask (a fault signal is blocked in the guest thread's host mask)	Passed
With this PR	Passed	Passed

AI-assisted: yes (Claude Code applied the PR, ran the tests and wrote this report).

[iamthehimansh]

Tested on an Apple M2 Pro, macOS, x86_64 Release under Rosetta 2, MoltenVK 1.4.2, applied on top of #8658's feat/macos-port (27abf7e4b) together with #8477, #9263, #9746 and #9406. I built the whole tree and ran the full suite (1,962 tests), then ran any new failure 3x on its own with each PR alone.

• guest_pthread_host_fault_mask (added here): passes.
• With this PR alone, no test that passes without it fails. The suite's only differences from feat: macOS host support (Apple Silicon via Rosetta 2, Vulkan on MoltenVK) #8658 alone come from feat(recompiler): read I/J from barycentric built-ins on MoltenVK #9193, reported there.

AI-assisted: yes (Cana)

[maximilianfv]

Tested on Apple M1 Pro, macOS 27.0.1 (x86_64 build under Rosetta 2)

Retest at fc9716dad. Since my report at 9ea069639, the only new commit merges main through 137a8fab89. The PR's own changes, git diff 137a8fab89 fc9716dad against git diff 30fda0a03 9ea069639, add and remove the same lines, so the code is unchanged.

Applied at fc9716dad to mugurc's macOS series (27e968974) merged with main at 1e20650e84, with other open macOS fixes, then toggled only this PR. "Without" keeps this PR's new test and leaves out only the Thread.cpp change.

	guest_pthread_host_fault_mask	other guest_pthread*, *write_watch*, guest_raise_exception* (39 tests)
Without the Thread.cpp change	Subprocess aborted (3 of 3 runs): Require at GuestPthreadHostFaultMask.cpp:17 in CheckMask, called from StartNativeThread (a fault signal is blocked in the guest thread's host mask)	Passed
With this PR	Passed (5 of 5 runs)	Passed

AI-assisted: yes (Claude Code applied the PR, ran the tests and wrote this report).

[jotitadev]

Tested head fc9716dad merged with current main on macOS. The fix works, but after the merge the PR's test makes public_headers fail.

Machine: Apple M2 Max (32 GB), macOS 27.2 (26B5112a), Apple clang 21.0.0, x86_64 Release under Rosetta 2 (-DCMAKE_OSX_ARCHITECTURES=x86_64, Ninja, BUILD_TESTING=ON, Python 3.9 from Xcode).
Base: main b986c11a34 does not build the libraries on macOS by itself, so the tree is that main plus the open macOS build PRs (#10366, #7631, #10659, #10660, #11249, #11358, #10669, #10552, #7804, #10676, #11447, #11437) and #11554, plus #11705 (applied locally). Only this PR is toggled, with git merge --no-commit on top.

GuestPthreadHostFaultMask.cpp declares scePthreadCreate and scePthreadJoin itself. Since 35cf8d5 (canonical headers, 2026-10-24, after this branch's last main merge 137a8fab89), core/libs/prx/libkernel/include/Api.hpp declares both, and the public_headers audit allows only one declaration:

scePthreadCreate: expected exactly one header declaration; found 2: core/libs/prx/libkernel/include/Api.hpp:559, core/libs/tests/GuestPthreadHostFaultMask.cpp:7
scePthreadJoin: expected exactly one header declaration; found 2: core/libs/prx/libkernel/include/Api.hpp:560, core/libs/tests/GuestPthreadHostFaultMask.cpp:8

The audit reads the source tree, so it should fail on every host, not only macOS. #9263 hit the same thing after it merged, and main fixed it in 039408c. The same change works here:

+#include "prx/libkernel/include/Api.hpp"
 #include "SceTypes.hpp"
 #include <csignal>
 #include <cstdlib>
 #include <pthread.h>
 
-extern "C" {
-int APS5_VABI scePthreadCreate(Pthread* thread, const PthreadAttr* attr, PthreadEntry entry, void* arg, const char* name);
-int APS5_VABI scePthreadJoin(Pthread thread, void** retval);
-}

With that change, public_headers and the 64 public_headers_compile_* batches pass (65/65), and guest_pthread_host_fault_mask passes 3/3. I ran this in the tree above with this PR, #10876 and #9263 (plus 039408c) merged together, and in that tree the non-GPU suite (1374 tests) has no new failures against the base apart from the 65 public_headers* tests, which pass with the change. The merge itself is clean.

Written with AI assistance (Claude Code). I ran the builds and tests above on my machine.