[CinCV TNG] lock problems with suse/leap, which can cause segvs in 5.1

Petter Reinholdtsen pere at hungry.com
Sun Nov 6 23:58:48 CET 2016


[Good Guy]
> As it turns out, intel's lock elision code was causing SEGVs when
> attempting to unlock a locked mutex.  Since intel seems to be now
> disavowing TSX locks in the microcode updates,

I believe this is a misunderstanding.  Intel have disabled TSX locks on
some CPU models where it was broken, but the feature is enabled and used
on other and more recent CPU models.  Where the feature is disabled,
using TSX style locks causes SIGILL instead of SIGSEGV.

I've been told that on s390, s390x, Hurd and kFreeBSD, unlocking an
non-locked mutex will also cause a crash.

> I am not sure that this is not a really issue, unless you have real
> evidence that there is a problem.

I have not tested if the issue exist in cinelerra.  I know incorrect
handling of mutexes exist elsewhere, which will cause a segfault on CPUs
with working TSX.

> Have you run 'valgrind --tool=helgrind' and actually seen problems??

Only with ogg123 so far, which is one of the packages I maintain in
Debian and which caused the start of the thread on debian-devel that I
pointed to in my previous message.

-- 
Happy hacking
Petter Reinholdtsen


More information about the Cinelerra mailing list