The stem of a request and of a relinked module name was cut at the first dot,
so a request for Engine.Render.prx that missed could load Engine.Core.prx.guest.prx.
Only the last extension is removed now, as the review of the stem lookup asked,
and the variant-names test requests a dotted name that must not load the sibling
fixture module.
Co-Authored-By: lendinglibrary <[email protected]>
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
Culprit: none in 892a4f1328..b85d972f42; race in the Windows vnode watch (50b06110b guarded only the loop). Run 39473930499, guest_kqueue: a watched file deleted before registration is snapshotted under $Extend\$Deleted, WatchParent got ERROR_ACCESS_DENIED and threw.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
A title asks for a module by its console name and the relinker names the
converted module after the ELF it was built from: Unity IL2CPP titles load
Media/Modules/<name>.prx and Media/Plugins/<name>.prx while the dump holds
<name>.elf.guest.prx, so sceKernelLoadStartModule failed with 0x80020002
and the title stopped 3 s after start. After the exact and the
case-insensitive <name>.guest.prx lookups miss, take the sibling whose stem
matches the request and ends with .guest.prx, shortest name first.
A path whose directory part is a regular file, such as file/child, failed
with ENOENT in readlink and readlinkat on Windows, where FreeBSD 11.4 fails
it with ENOTDIR (vfs_cache_lookup, sys/kern/vfs_cache.c). The not-found
error now checks whether the nearest existing ancestor is not a directory,
as was previously done for chmod, truncate, utimes, rename, remove, chdir,
opendir, and fopen.
Co-Authored-By: Antigravity <[email protected]>
FreeBSD 11.4 kern_readlinkat returns EINVAL for a count above IOSIZE_MAX
before namei looks at the path. readlink and readlinkat checked a NULL or
empty path first, so readlink(NULL, buf, SIZE_MAX) returned EFAULT and
readlink("", buf, SIZE_MAX) returned ENOENT where FreeBSD returns EINVAL.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
The test builds a set over 1500 textures and expects its first proof on
the fast path. A ShaderResources object stamps its last full walk when it
is constructed, before that build, so when the build and the first proof
take more than the default 1000 ms template audit interval (as under
parallel test load) the audit turns the proof into a full walk and the
test fails. The test now sets APS5_TEMPLATE_AUDIT_MS to 0 unless it is
already set, as the sampled cache budget test pins it, so its checks see
only memory changes.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Trimming the end of any PT_INTERP inside an executable segment dropped
the code in front of one that follows it, so an image without an SCE
block whose interpreter sits after its code lost its entry points. The
trim now applies to an interpreter that starts the segment, and the SCE
path keeps taking any interpreter inside the scanned range, as before.
The code analysis trims an interpreter path from the start of an
executable segment only for images that have a PT_SCE_DYNLIBDATA block.
An executable without one keeps that path in the scanned range, so
/libexec/ld-elf.so.1 is walked as code: a RELATIVE relocation with
addend 0 roots the control flow graph at the first byte of the segment,
and the 0x64 of "ld-elf" decodes as an FS-prefixed instruction that the
Windows and macOS TLS rewriters refuse. The trim now runs for both
layouts, ending at the segment for an image without the block.
The Windows sceKernelGetdirentries stored 0 in *basep on every read, so a
guest looping on the cookie saw it stall (FreeBSD stores the directory
offset the read started at, kern_getdirentries: *basep = loff). Report the
index of the first entry of the read, the same unit lseek takes on these
descriptors, and run guest_directory_entries on Windows, where it was never
built.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
kevent threw NotImplemented for EVFILT_WRITE with NOTE_LOWAT above 1 on a socket, although the socket readiness check already reports free send-buffer space (SO_SNDBUF less SIOCOUTQ on Linux). The write event now fires once that space reaches the mark, as FreeBSD 11's filt_sowrite compares sbspace with kn_sdata, and stays silent while the space is below it. The debt entry no longer says a write mark is limited to 1. Reference: https://github.com/freebsd/freebsd-src/blob/release/11.0.0/sys/kern/uipc_socket.c
sceNet translated only 21 host socket errors and reported every other one
as EIO. libkernel already owns the complete FreeBSD table in
GuestSockets::GuestErrorFromHost, so use it for net_error. The new check
makes a stream socket with the UDP protocol fail with EPROTONOSUPPORT (43)
instead of EIO (5).
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
pad_timeline.py now takes its sequence from pad_input.next_sequence, and
pad_input.previous_sequence reads the sequence from an APS5_PAD_TIMELINE
header, so a pulse or timeline queued after a host reboot still exceeds
the sequence the runtime loaded from the existing command file.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
source_files compared every component of the absolute path against SKIP,
so a --root inside a directory named build or 3rdparty yielded no files.
Only components below the scan root are compared now.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Tripchipp