MASKMOVQ, MASKMOVDQU, VMASKMOVDQU, MONITOR and MONITORX use a register-form ModRM byte but still access memory at DS:rDI or DS:rAX. Mark them as memory operands and cover the encodings in the Windows and macOS GS trap tests.
Implement sceNetShowRoute as a silent stub returning 0, matching the (void) signature in shadPS4 and the behavior of sceNetShowIfconfig and sceNetShowNetstat. Adds guest unit tests in GuestSceNet.cpp and updates technical debt documentation.
sceNetRecv, sceNetRecvfrom and sceNetRecvmsg on a datagram socket without a
local address waited in wait_for_local_address until the socket was bound,
connected or written to, because only a non-blocking socket made that wait
fail. FreeBSD 11.4 soreceive_dgram fails a MSG_DONTWAIT receive without data
with EWOULDBLOCK. The flags now reach wait_for_local_address and the test
covers an unbound socket.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
sceNetSend, sceNetSendto, sceNetSendmsg, sceNetRecv, sceNetRecvfrom and
sceNetRecvmsg failed with EOPNOTSUPP for MSG_DONTWAIT on Windows, where
FreeBSD 11.4 sosend_generic and soreceive_generic make the one call
non-blocking and fail with EWOULDBLOCK when it would wait. Winsock has no
per-call flag, so the call now asks WSAPoll with a zero timeout whether the
socket is ready and fails with EAGAIN when it is not. The Windows branch of
the test that expected EOPNOTSUPP now runs the checks of the other hosts.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
A path whose directory part is a regular file, such as file/child, failed
with ENOENT in chmod, truncate, utimes and rename on Windows, where FreeBSD
11.4 fails it with ENOTDIR (vfs_cache_lookup, sys/kern/vfs_cache.c). The
not-found error of these calls now asks whether the nearest existing
ancestor is not a directory, as open, stat, lstat, access and unlink do.
For rename the source ancestor is checked first and the destination ancestor
only while the source exists. A test line that expected ENOENT for truncate
of a path below a regular file now expects ENOTDIR.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
fmin, fmax and their f and l forms tested their operands with std::isnan,
which GCC compiles to a compare that raises FE_INVALID for a signalling
NaN. FreeBSD's s_fmax.c and s_fmin.c read the exponent and significand
bits instead, "to avoid raising spurious exceptions", and return the
other operand. The NaN test now reads the bits of float, double and the
80-bit long double.
GuestFminf checks that a signalling NaN on either side returns the other
operand and leaves FE_INVALID clear, for all three types.
The failure was first caught by the numeric helper test of
boykopovar/AnyPS5#2280 (Привалов Александр Владимирович).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A path whose directory part is a regular file, such as file/child, failed
with ENOENT in open, stat, lstat, access and unlink on Windows, where
FreeBSD 11.4 lookup fails it with ENOTDIR (vfs_cache_lookup, sys/kern/vfs_cache.c).
mkdir and rmdir already did. A not-found error of these calls now asks whether the
nearest existing ancestor is not a directory and reports ENOTDIR then.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
ftruncate, truncate and sceKernelFtruncate extended a file through _chsize_s,
which writes the whole new length as zero bytes. FreeBSD extends the file
without writing data (kern_ftruncate -> ffs_truncate) and fails a length the
file system cannot hold with EFBIG. NativeFtruncate now sets the end of file
with SetFileInformationByHandle and maps ERROR_DISK_FULL to ENOSPC and
ERROR_FILE_TOO_LARGE/ERROR_INVALID_PARAMETER to EFBIG.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
MathExact.cpp was built without -frounding-math, so GCC compiled
std::rint as an add and subtract of 2^52 on the magnitude, which is exact
only when rounding to nearest: under FE_DOWNWARD rint(-1.5) returned -1
and rintf(-1.5f) -1, where C and FreeBSD's libm give -2. lrint and llrint
were right: they call the host libm's llrint, which honours the mode.
MathExact.cpp now builds with -frounding-math. GuestMathExact checks
rint, rintf, rintl, lrint and llrint of -1.5 and 1.5 in all four rounding
modes and that the mode is unchanged.
The failure was first caught by the rounding test of
boykopovar/AnyPS5#2235 (Привалов Александр Владимирович).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Invocations of one dispatch that fault through different guards wrote
the guard word with plain stores, which the Vulkan memory model treats
as a data race. The write is now a relaxed device-scope OpAtomicStore,
so one of the records is reported whole. The guarded-pointer test now
also has two workgroups fault through a null and an unmapped pointer.
MarcZX100 pointed out the race.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A ReadConst of a guarded flattened-SRT slot called record_bda_fault
with the poison record's pc and address when the slot's flag was set.
NVIDIA's compiler pays for every call site in a large function:
Crimson Desert's 1,916,980-word compute module has 2,769 guarded
reads, and its pipeline took 99 s and 5.9 GiB to create.
The read now stores the record's flattened-SRT word, tagged with its
stage, to a new fault-buffer word after the written-page list, and
stops as before. The host reads the pc and address from that stage's
flattened SRT and raises the same Unmapped fault. Poisoned and clean
captures still share one variant, specialization and pipeline, and
the existing words keep their offsets, so cached modules from before
still report through record_bda_fault.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Dean