Since #6003 a fiber with a context of 64 KiB or more runs on the
context itself, so the 64 KiB fiber of agc_driver_gpu_map_on_fiber
imports its mapping on a host thread on Windows too, and the test's
Windows-only expectation of no thread failed. The test now also maps
on a 16 KiB fiber, which keeps a host stack and imports on it on
Windows, and expects one import thread for the 64 KiB fiber on every
platform.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Titles that bundle OpenSSL load the console's CA file from the sandbox
directory and fail every HTTPS request without it. ProsperoRadio
(PPSA99001, libcurl 8.18.0 + OpenSSL 3.5.2) on main + #4796 + #4525:
[ProsperoRadio][http] https://all.api.radio-browser.info/json/servers: Problem with the SSL CA cert (path? access rights?) (error adding trust anchors from file: /gkbHtXc3Wq/common/cert/CA_LIST.cer)
SwanStation (PPSA98510) stops on the same file (#5001).
The guest path resolver now accepts exact-path providers, and libkernel
registers one for /<sceKernelGetFsSandboxRandomWord()>/common/cert/CA_LIST.cer.
On first use it writes a PEM bundle once per process to
cert_cache/CA_LIST.cer in the run directory, read-only: on Windows the
current user's ROOT and CA system stores, elsewhere SSL_CERT_FILE or the
first distro bundle that exists. Without a certificate the path keeps
resolving under the guest root and does not exist, as before. Every other
path under the sandbox word is unchanged. The console file's format and
the rest of its common/cert directory are unverified and recorded in
docs/dev/debt/libkernel.md.
With this change, #4796, #4525 and #5025 on Windows 11, 60 s each:
[ProsperoRadio][http] 11846160 bytes in 1458 ms: 1161 passes, 803 waits (679 woken) for 1428 ms
[ProsperoRadio][catalog] page offset=0 bytes=11846160 objects=10000 accepted=9549 total=9549
...
[ProsperoRadio][catalog] page offset=60000 bytes=1712035 objects=1494 accepted=1436 total=59371
[SwanStationPS5] update: latest 1.2.5, this 1.1.3, available
[SwanStationPS5] gamedb: index written
Co-Authored-By: Claude Fable 5.1 <[email protected]>
PS4 programs load system libraries with sceKernelLoadStartModule and
their console path. The OpenOrbis piglet sample loads
/<sandbox word>/common/lib/libScePigletv2VSH.sprx and
libScePrecompiledShaders.sprx, the sandboxed form of
/system/common/lib/ that sceKernelGetFsSandboxRandomWord names. The path
was opened as a guest file and the load failed with ENOENT.
A path in /system/common/lib/ or in the sandboxed /<word>/common/lib/
whose file name ends in .sprx now loads the host libs/<name>.prx, as
libSceSysmodule does for module IDs. Names with a directory part are
not mapped, so such a path cannot leave libs/.
PS4 titles ship sce_sys/param.sfo instead of param.json. When param.json
is absent, the application metadata now reads the title, title ID,
download data size and user defined parameters from param.sfo, and
/data and /download0 are mounted for it as for param.json. param.json
is still preferred when both files exist.
On Windows a test that returns from main ends through exit and ExitProcess,
which kills every other thread before the prx static destructors run at
DLL_PROCESS_DETACH. Detached host threads (the driver's async log thread,
libkernel's file watch thread, ...) die holding winpthreads and queue locks,
and a destructor that then locks a mutex, waits for the dead logger or
destroys a condition variable never returns: the test prints its last line
and hangs until the CTest timeout (#6791).
Every test executable now wraps main on MinGW. After main returns the
wrapper runs the executable's own exit handlers and static destructors
(_cexit), as exit does today, flushes the streams and ends the process with
LibcTerminate_nid_no_patch, the path the guest exit already takes. That
flushes the driver log while its thread is alive and calls TerminateProcess,
so no DLL_PROCESS_DETACH work runs after the threads are gone. The libc
shutdown callbacks are not run, as on a plain return from main on Linux. A
test that does not load libc.prx returns as before.
A coordinate just below 1 truncates to the wrong vertex. The generated evaluation shader is checked at each corner, and a draw records the vertex that lands on each corner of the rectangle.
Co-authored-by: Cursor <[email protected]>
Resolve the XCuZoBSVFG8 import required by Killzone Liberation by adding sceRazorCpuNamedSync as a no-op returning success when capture is inactive.
Extend the RazorCpu behavior test and document the NID source and unverified console signature and capture behavior.
Co-authored-by: Copilot <[email protected]>
Value::setTypeMismatchHandler, setElementAccessFailureHandler and
setSpecialFloatHandler were stubs that threw NotImplemented. PPSA07411
calls all three while it sets up its documents.
The title calls them as members of Value: on its root Value before
parsing into it, with the handler and its context, and treats an int
result of 0 as success. They now store the handler and context on that
Value and on the values it holds, and return 0. A Value keeps its
handlers when its content changes (set, clear, operator=,
Parser::parse), and Parser::parse gives the values it creates the
handlers of the Value it parses into, since the title reads the errors
its handlers record after accessing parsed members. A copy does not
carry them.
A typed getter on a non-null value of another type calls the type
mismatch handler with the requested type, the value, null and the
context, and returns its result when it has the requested type, else
the getter's default.
The element access failure and special float handlers are stored but
not called: the error code the console passes to the first and the
FunctionType it passes to the second are unknown. A failed element
access on a value carrying the first throws instead of reaching the
null-access callback. Serializing a non-finite real already threw; with
the second set it throws naming the unknown FunctionType. The technical
debt entry records what is inferred.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
NVIDIA's compiler predicates a branched LDS load at no cost, while the
selected form costs about 2.5 instructions per dword and splits a
ds_read_b128 into four loads (RTX 4060 measurements by @karanvora2599 in
loads therefore keep the branch of main, in workgroup, function and device
memory alike, and the sentinel dword that only they read goes away: the LDS
array is the program's dwords plus the lock again, and the device-memory
decision reserves the lock alone. The sentinel budget tests become lock
budget tests at that boundary, and the guarded-load test pins the branch.
Co-Authored-By: Claude Fable 5.1 <[email protected]>
A load selected instead of branched still accesses memory, so its fallback
element must be one no write can race with: scratch keeps element 0, LDS in
workgroup or function memory reads a sentinel dword appended after the
program's LDS and lock, and storage buffers read element 0xfffffffe, which
robustBufferAccess2 turns into a zero load without a memory access. The
device enables robustBufferAccess2 when it has it and the target carries the
flag; without it, and for LDS in device memory, buffer loads keep their
selections. The aligned high word of a buffer read re-reads the low word's
element instead of word 0. agc_guarded_memory_loads traces the emitted loads
in both device modes.
Co-Authored-By: Claude Fable 5.1 <[email protected]>
LDS, scratch and direct storage-buffer loads load element 0 when the lane is
inactive or the address is out of bounds and select 0 as the result, instead of
opening a selection per guard. Stores, atomics, formatted and physical-address
accesses keep their branches. A branch guard cost four blocks per lane half, and
the host compiler's time grows with the block count: PPSA11142's merged PSP
vertex program emitted 16,983 blocks and 1,710 conditional branches, 5,907 and
868 with the selects.
Co-Authored-By: Claude Fable 5.1 <[email protected]>
Adrià