chore(main): release 3.1.1#2
Open
github-actions[bot] wants to merge 1 commit into
Open
Conversation
github-actions
Bot
force-pushed
the
release-please--branches--main
branch
from
0602bdf to
7150fd8
Compare
github-actions
Bot
force-pushed
the
release-please--branches--main
branch
from
7150fd8 to
06d0551
Compare
bernie-laberge
added a commit
that referenced
this pull request
May 30, 2026
…ux 9 (part 2) (AcademySoftwareFoundation#1279) ### fix: SG-42890: Fix SIGSEGV error when running RVIO commands on Rocky Linux 9 (part 2) ### Linked issues NA ### Summarize your change. Fix an `rvio` SIGSEGV that fires at shutdown on Linux with the NVIDIA proprietary GL driver. Conversions complete successfully — the output `.mov` is written intact — but the process then exits with a segfault, which produces a crash dump and a non-zero exit code that breaks RV/CI/render-farm pipelines. Symptoms: - All frames are rendered and the output file is fully written. - The process then segfaults after `main()` returns. - Crashpad-symbolicated minidump shows the crashing thread's `rip` in an unmapped region with no usable call stack (only `stack scanning` frames in `libc.so.6`). The crash address looks like garbage because the page is already gone by the time the dump is written. Constant register signatures across ASLR-randomized runs (`rax=0x6e`, `r14=0x1e8480`, `r15=0xfffffffffffffb28`) confirmed the crash always lands at the same instruction. ### Root cause Captured under `gdb` (where the offending library is still mapped at the instant of the SIGSEGV), the crashing thread is an NVIDIA driver background worker: ``` Thread N "RVIO Main" received signal SIGSEGV 0x00007fffcd9f59ec in ?? () from /lib64/libnvidia-glcore.so.595.58.03 #0 libnvidia-glcore.so.595.58.03 #1 libnvidia-glcore.so.595.58.03 #2 libnvidia-glcore.so.595.58.03 #3 start_thread () from libc.so.6 #4 clone3 () from libc.so.6 ``` …while the main thread is mid-teardown: ``` #0 _dl_close_worker (...) at dl-close.c:769 #1 _dl_close #2 __GI__dl_catch_exception AcademySoftwareFoundation#5 dlclose@GLIBC_2.2.5 () AcademySoftwareFoundation#6 __glXMappingTeardown () from libGLX.so.0 AcademySoftwareFoundation#7 __glXFini () from libGLX.so.0 AcademySoftwareFoundation#8 _dl_fini () at dl-fini.c:148 AcademySoftwareFoundation#9 __run_exit_handlers () from libc.so.6 AcademySoftwareFoundation#10 exit () from libc.so.6 ``` This is the well-known **libglvnd ↔ NVIDIA driver shutdown race**: 1. `main()` returns. 2. libc's `exit()` runs `__run_exit_handlers` → `_dl_fini`. 3. `_dl_fini` invokes `libGLX.so.0`'s ELF destructor `__glXFini`, which `dlclose`s the NVIDIA GLX vendor (`libGLX_nvidia.so.0`) and transitively unmaps `libnvidia-glcore.so`. 4. NVIDIA's proprietary driver maintains its own internal helper threads. They are **not** joined by libglvnd's teardown — NVIDIA's blob has no such hook. The threads are still running when `dlclose` finishes. 5. The kernel unmaps `libnvidia-glcore.so`'s code pages out from under one of those running threads → next instruction fetch hits unmapped memory → SIGSEGV. The thread is named `"RVIO Main"` only because Linux child threads inherit their parent's `prctl(PR_SET_NAME)` value (set by `TwkUtil::setThreadName`). It is NOT the application's main thread. By the time crashpad's signal handler finishes serializing the minidump, `dlclose` has fully completed and the page is no longer in `/proc/self/maps`, which is why the dump appeared to crash "nowhere." ### Solution `src/bin/imgtools/rvio/main.cpp`: on Linux, replace the final `return 0;` in `utf8Main` with `std::_Exit(0);` — but only after all of rvio's legitimate shutdown has already run (`MovieRV::uninit`, `TwkMovie::GenericIO::shutdown`, `TwkFB::GenericIO::shutdown`, `TwkFB::ThreadPool::shutdown`). `std::_Exit` is the POSIX "skip atexit / `_dl_fini` and terminate now" primitive. It hands control directly to the kernel, which reaps memory, file descriptors, and threads — including NVIDIA's straggler threads — atomically, with no opportunity for the unload race. Nothing in our shutdown is left undone: - The output movie has already been finalized and `close`d by the writer before `utf8Main` reaches this point. - All Open RV-owned plugins have been shut down explicitly above the `_Exit` call. Windows and macOS still go through `return 0;` because they don't use libglvnd and don't exhibit this race. The change is `#ifdef PLATFORM_LINUX` guarded. Signed-off-by: Bernard Laberge <bernard.laberge@autodesk.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
🤖 I have created a release beep boop
3.1.1 (2026-02-26)
Bug Fixes
Build System
GitHub Actions
This PR was generated with Release Please. See documentation.