On Thu, Aug 05, 2021 at 02:58:03PM +0100, Daniel P. Berrangé wrote:
On Thu, Aug 05, 2021 at 03:33:57PM +0200, Tim Wiederhake wrote:
On Thu, 2021-08-05 at 14:24 +0100, Daniel P. Berrangé wrote:
On Thu, Aug 05, 2021 at 03:08:50PM +0200, Tim Wiederhake wrote:
`pthread_mutex_destroy`, `pthread_mutex_lock` and `pthread_mutex_unlock` return an error code that is currently ignored.
Add debug information if one of these operations failed, e.g. when there is an attempt to destroy a still locked mutex or unlock and already unlocked mutex.
We ignore the errors because in practice we found impls don't return any error code when using PTHREAD_MUTEX_NORMAL - it would require use of PTHREAD_MUTEX_ERRORCHECK.
Did you actually see errors returned for real ?
Yes. Please note that I apprently copy-pasted the wrong link in the cover letter. The pipeline for this series is correct (https://gitlab.com/twiederh/libvirt/-/pipelines/348505233), the pipeline for only patches 5, 6, 7 that demonstrate the non-zero return values of pthread_mutex_{lock,unlock,destroy} is here:
All that's showing is
(killed by signal 6 SIGABRT)
where's the VIR_WARN output that ought to have ended up on stderr from the failed test before aborting ?
Ok, so the VIR_WARN output is getting squelched by the tests somewhere, as its missing when i run locally too. If I add a test that does virMutex m; ignore_value(virMutexInit(&m)); virMutexUnlock(&m); then I get no warnings with this patch, nor abort with the next patch. It only triggers with PTHREAD_MUTEX_ERRORCHECK The pthread_mutex_destroy call is the only one that I see returning errors with PTHREAD_MUTEX_NORMAL. So I don't think there's benefit to adding the code to lock/unlock paths. Regards, Daniel -- |: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|