[Xcb] [Bug 98048] libpthread-stubs must NOT provide symbols which depend on actual implementation

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=98048

--- Comment #12 from Emil Velikov <emil.l.velikov at gmail.com> ---
(In reply to Rob Clark from comment #9)
> (In reply to Ian Romanick from comment #7)
> > My system has glibc-2.22-18.fc23.x86_64, and that has:
> > 
> > $ nm /lib64/libc.so.6 | grep 'T pthread'
> > 000000000010fae0 T pthread_attr_destroy
> > 000000000010fb40 T pthread_attr_getdetachstate
> > 000000000010fba0 T pthread_attr_getinheritsched
> > 000000000010fc00 T pthread_attr_getschedparam
> > 000000000010fc60 T pthread_attr_getschedpolicy
> > 000000000010fcc0 T pthread_attr_getscope
> > 000000000010fb10 T pthread_attr_init@@GLIBC_2.2.5
> > 000000000010fb70 T pthread_attr_setdetachstate
> > 000000000010fbd0 T pthread_attr_setinheritsched
> > 000000000010fc30 T pthread_attr_setschedparam
> > 000000000010fc90 T pthread_attr_setschedpolicy
> > 000000000010fcf0 T pthread_attr_setscope
> > 000000000010fd20 T pthread_condattr_destroy
> > 000000000010fd50 T pthread_condattr_init
> > 000000000013e8d0 T pthread_cond_broadcast at GLIBC_2.2.5
> > 000000000010fd80 T pthread_cond_broadcast@@GLIBC_2.3.2
> > 000000000013e900 T pthread_cond_destroy at GLIBC_2.2.5
> > 000000000010fdb0 T pthread_cond_destroy@@GLIBC_2.3.2
> > 000000000013e930 T pthread_cond_init at GLIBC_2.2.5
> > 000000000010fde0 T pthread_cond_init@@GLIBC_2.3.2
> > 000000000013e960 T pthread_cond_signal at GLIBC_2.2.5
> > 000000000010fe10 T pthread_cond_signal@@GLIBC_2.3.2
> > 000000000013e9c0 T pthread_cond_timedwait at GLIBC_2.2.5
> > 000000000010fe70 T pthread_cond_timedwait@@GLIBC_2.3.2
> > 000000000013e990 T pthread_cond_wait at GLIBC_2.2.5
> > 000000000010fe40 T pthread_cond_wait@@GLIBC_2.3.2
> > 000000000010fab0 T pthread_equal
> > 000000000010fea0 T pthread_exit
> > 000000000010fed0 T pthread_getschedparam
> > 000000000010ff30 T pthread_mutex_destroy
> > 000000000010ff60 T pthread_mutex_init
> > 000000000010ff90 T pthread_mutex_lock
> > 000000000010ffc0 T pthread_mutex_unlock
> > 000000000010fff0 T pthread_self
> > 0000000000110050 T pthread_setcanceltype
> > 000000000010ff00 T pthread_setschedparam
> > 
> > I can try reverting fa6db2f9 tomorrow.
> 
> Interesting.. I see the same thing.  So just fired up glxgears in gdb
> instead and set a breakpoint on pthread_mutexattr_settype().  Turns out some
> of the pthreads syms are coming from libc and some from libpthread-2.23.so. 
> Not entirely sure the reasoning for that.
> 
> Is it possibly you have the libc part but not libpthreads, so you end up
> mixing libc's partial pthreads w/ pthread-stubs instead of libpthreads.so?

Sure that list is pretty normal/constant. Yet you don't see the issues that Ian
has reported, do you ?

As a simple check - build libpthreads-stub before/after the offending commit
and rebuild libxcb/mesa against it.

In the former case there will be no libpthread-stub.so while in the latter
there will be one. Then at runtime you'll likely end up using the
libc/pthreads/pthreads-stub libraries leading to fun experience.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/xcb/attachments/20161007/325d63d6/attachment-0001.html>