https://bugs.kde.org/show_bug.cgi?id=371396
Bug ID: 371396
Summary: helgrind and drd pth_cond_destroy_busy testcase hang
with new glibc cond var implementation
Product: valgrind
Version: 3.12 SVN
Platform: Other
OS: Linux
Status: UNCONFIRMED
Severity: normal
Priority: NOR
Component: helgrind
Assignee: jsew...@acm.org
Reporter: m...@redhat.com
A new conditional variable implementation being tested in fedora rawhide atm
will hang the helgrind and drd pth_cond_destroy_busy tests. This is because
they test a condition that triggers undefined behaviour. "Attempting to destroy
a condition variable upon which other threads are currently blocked results in
undefined behavior." In the new implementation this causes the thread calling
pthread_cond_destroy will just hang in this situation.
There are a couple of ways we could work around this.
- Add some timer thread that kills the whole test after some time.
This is what we did for the bar_bad tests. It causes some non-determinism,
extra thread events and you
have to handle different code paths/warnings/error conditions.
- Add a generic timer/watchdog to vgregtest that kills hanging tests
This would be nice in general to make sure make regtest does at least finish.
- Skip the test if a newer glibc is detected. This is what I am doing for now.
- Try to call pthread_cond_broadcast in HG_PTHREAD_COND_DESTROY_PRE when we
detect that pthread_cond_destroy is called on a variable that other threads are
waiting on. That should in theory wake them up making the pthread_cond_destroy
defined again.
- Just not call pthread_cond_destroy if we detect that it might hang.
The new implementation is:
https://sourceware.org/ml/libc-alpha/2016-06/msg00554.html
It is currently only applied in fedora rawhide glibc 2.24.90, but is expected
to eventually hit glibc master.
Reproducible: Always
--
You are receiving this mail because:
You are watching all bug changes.
