* kernel: do not release a waiter on a zero-count syncOnAddress wake
sceKernelSyncOnAddressWake with a count of 0 must release no waiters. On
Linux it released exactly one.
WakeLinux passes the count straight to FUTEX_WAKE, which does not treat 0
as "wake nobody": the kernel's wake loop tests `if (++ret >= nr_wake) break`
after waking a waiter, so any waiter present is released before the loop can
break. A standalone futex program confirms it -- FUTEX_WAKE with val = 0
returns 1, and the waiter observes its wait return 0 with the word unchanged.
WakePortable already short-circuits count == 0, so the two backends of the
same ABI disagreed, with only the Linux one wrong.
TestWakeZeroIsNoOp in SyncOnAddressTests covers exactly this and fails on
Linux without the fix (15 consecutive failures on the assertion "zero-count
wake releases no waiters"); with the fix the suite passes 15 consecutive runs.
* kernel: shorten zero-count futex comment
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
Why: Thread-atexit registrations were discarded, leaving objects alive after their guest TLS storage was released.
What: Store registrations per host thread and run them in LIFO order before pthread keys and guest TLS are destroyed.
Why safe: Only callbacks registered on the exiting thread run, once, before existing teardown continues.
* fix: keep hint-less guest mappings at the canonical PS5 base
FindGuestFreeRange searched the low system-managed range first for
mappings with no address hint, so the first hint-less direct-memory map
could land as low as 0x200000. The PS5 kernel never places hint-less
user mappings below 0x200000000 and guest code relies on that: Sony's
libc maps 4 MiB of direct memory for its internal heap, fails its
mspace setup when the returned address is that low, and the first
malloc then dereferences a null mspace (a read at 0x38, the mspace
magic check). On macOS this made Raiden III crash on the main guest
thread a couple of seconds after boot, 100 percent reproducible with
--printf-direction Silent.
Search from the canonical base first, fall back to the user range, and
keep the low system-managed range only as a last resort. The mmap path
already anchored hint-less searches at 0x200000000; this aligns the
shared search helper with it.
Adds two regression tests: the libc-shaped allocation must come back at
or above the canonical base and hold writes, and direct-memory content
must survive an unmap and remap of the same physical range.
* macos: make the fatal-report memory dumps fault-safe
IsReadableRange returned true for any nonzero address on macOS, so the
fatal report's guest memory dumps dereferenced whatever the crashed
thread had in its registers. A fault inside the reporter re-enters the
signal handler and wedges the reporting thread, which hid real guest
crashes whenever logging was enabled: the game kept running with a dead
thread and the report was never completed.
Walk the Mach regions covering the range and require read permission
before dumping, the same contract the Linux implementation provides.
* do not fallthrough HOST_SYSTEM_MANAGED_MIN
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
* macos: POSIX platform layer (host fault handler, virtual memory)
- hostException: Mach/POSIX signal-based host fault handler mirroring the
Windows vectored handler (SIGSEGV/SIGBUS/SIGILL), behind
'#elif defined(__APPLE__)'. Windows and Linux branches are untouched.
- sysLinuxVirtual: mach_vm_region-based is_mapped (no /proc/self/maps on
macOS), PTHREAD_MUTEX_NORMAL, and a MAP_FIXED carve-in-place allocator that
never leaves an unmapped hole for dyld/Rosetta/Metal to claim. All
__APPLE__-guarded; the original Linux allocator path is preserved verbatim.
- sysLinuxDbg/sysLinuxFileIO: libgen.h for basename(), and a POSIX
opendir/readdir implementation of SysFileGetDents (previously a stub).
* macos: guest kernel backing, threads, and POSIX libs
- memory/memoryAddressSpace: anonymous shm_open backing (macOS has no
memfd_create) and re-reserve-on-unmap so a guest MAP_FIXED remap never
destroys a host mapping (__APPLE__-guarded). Drops host <sys/mman.h> MAP_*
macros so the guest's constexpr MAP_* constants compile (no-op on Windows).
- pthread: undef the host PTHREAD_STACK_MIN macro; on non-Windows, pin the
stack-switch asm's guest rsp/rbp to callee-saved r14/r15 (the template
clobbers r12/r13); __APPLE__ no-op for the absent pthread_condattr_setclock.
The Windows stack-switch asm is byte-identical.
- network: define SOCKET/INVALID_SOCKET for the non-Windows paths (were
undefined at 20+ use sites), and rename the non-Windows kernel_clock_*
helpers to the CamelCase names the callers already use. Both repair the
shared non-Windows build. Integer reinterpret_cast -> static_cast.
* loader: pin stack-switch asm operands to callee-saved registers
RunEntry switches to the guest stack with an asm template that clobbers
r12/r13 before consuming its inputs. With plain "r" constraints the compiler
may place func/guest_rsp/guest_rbp into r12/r13, so 'callq *func' jumps
through the saved host rsp -- a latent miscompile on every platform that any
unrelated codegen change can trigger. Pin the inputs to rbx/r14/r15, which the
template never touches and the SysV callee preserves.
General correctness fix (not macOS-specific); the same fix is applied to
pthread RunOnGuestStack.
* macos: Vulkan/MoltenVK graphics enablement
- pageManager: GPU write-tracking via Mach VM queries + mprotect (mprotect
cannot report the previous protection, so the expected-old check is dropped),
thread id via pthread_mach_thread_np, 4 KB page-size check -- __APPLE__-guarded.
- shaders/renderDraw/vulkanWindow: MoltenVK lacks VK_EXT_color_write_enable,
VK_EXT_depth_clip_enable and depthBounds; fall back to static color-write
masks and default depth clipping, request VK_KHR_portability_subset, and
reuse the graphics queue for present when only one queue is exposed. Non-Apple
pipeline/extension setup is unchanged.
- window: optional borderless window (KYTY_BORDERLESS) to sidestep a Rosetta
NSException in macOS window chrome.
* graphics: macOS thread identity for region tracking
Region-ownership locks resolve the current thread via GetCurrentThreadId()
on Windows and EXIT elsewhere. Use the Mach thread port on macOS (nonzero
per-thread id; 0 stays the no-owner sentinel).
* macos: marshal AppKit window operations to the main thread
The present worker thread shows the window, updates its icon and title, and
recreates a lost Vulkan surface. AppKit traps with 'Must only be used from the
main thread' when these run off the main thread. Route them through a small
task queue drained by the SDL main loop (an SDL_USEREVENT wakes the loop when
it is blocked in SDL_WaitEvent). Title updates are fire-and-forget; showing
the window and surface recreation wait for completion.
Windows and Linux call the SDL functions directly, as before.
* build: ignore the _Build output directory
_Build is the conventional out-of-tree build location (only _Build/vscode-clang
was ignored); a stray git add could sweep build artifacts into a commit.
* macos: isolate stack-switch workaround
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
Map paired PS5 depth-clip state through Vulkan pipeline creation and preserve asymmetric hard failures. Prepare GPU-mapped guest ranges before kernel file reads so buffer, texture, and alias ownership remains coherent. Add focused regression coverage and update the current launch target.
nmzik