Crash fix in case of forks: remove usage of pthread_once #483
Conversation
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
When compiling on Alpine Linux (musl libc) and running on glibc systems, pthread_once has incompatible internal state machines that cause crashes after fork(). This was observed as SIGSEGV in PHP binaries that use forking when returning from the pthread_once API. Root cause: - musl uses state value 1 during pthread_once initialization - glibc uses state value (__fork_generation | 1), typically 5 - After fork(), glibc sees musl's state=1, treats it as invalid for the current fork generation, and re-initializes - This causes double initialization of pthread keys Solution: Replaced pthread_once with atomic-based initialization in AllocationTracker::get_tl_state(): - Uses std::atomic<int> with three states: kKeyNotInitialized, kKeyInitializing, kKeyInitialized - Hot path optimized with relaxed atomic load - Slow path uses compare-and-swap with proper memory ordering - Added safety timeout to prevent infinite loops - Returns nullptr on timeout
- Cover TLS init with forking
Benchmark results for collatzParameters
SummaryFound 0 performance improvements and 0 performance regressions! Performance is the same for 1 metrics, 0 unstable metrics. See unchanged results
|
Benchmark results for BadBoggleSolver_runParameters
SummaryFound 0 performance improvements and 0 performance regressions! Performance is the same for 1 metrics, 0 unstable metrics. See unchanged results
|
r1viollet
commented
Jan 26, 2026
Collaborator
Author
|
Sadly, we will have to rethink this PR. |
Collaborator
Author
|
What could be happening:
A fix could be about securing the init after the fork ? |
r1viollet
commented
Jan 29, 2026
| pthread_key_t key = _tl_state_key.load(std::memory_order_relaxed); | ||
|
|
||
| // In debug builds, verify our assumption | ||
| assert(key != kInvalidKey && "pthread key should be initialized before use"); |
Collaborator
Author
There was a problem hiding this comment.
this might trigger with a relaxed order, but at least I will know about it
I might change mem order
r1viollet
force-pushed
the
r1viollet/pthread_once_crash_fix
branch
from
5dd2c62 to
dc0d1b6
Compare
- Use the key as synchronization - Avoid init during a get of the thread local state - Add a fork handler to ensure we have the key initialized
r1viollet
force-pushed
the
r1viollet/pthread_once_crash_fix
branch
from
dc0d1b6 to
57ba49e
Compare
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.
What does this PR do?
Disclaimer: I do not fully understand why we have a crash where the key was not initialized.
However we know that in the case of forks we can be in a state where we try to init in the hot path (during an allocation).
This could be the reason for the crash.
This PR addresses the symptoms.
We could go one step further here, to make sure we wrap the full init sequence (not just thread local state)
Motivation
Avoid a crash.
Additional Notes
NA
How to test the change?
I added a test for forking. Though it does not reproduce the crash.