Bug#883615: Acknowledgement ([CRITICAL] Stretch p-u 9.3 breaks NVidia driver and X.org)

Aurelien Jarno aurelien at aurel32.net
Thu Dec 7 09:06:01 UTC 2017


control: reopen -1
control: tag -1 - unreproducible
control: retitle -1 nvidia-driver: crashes with /etc/ld.so.nohwcap

On 2017-12-07 05:43, Julien Aubin wrote:
> 2017-12-06 21:50 GMT+01:00 Aurelien Jarno <aurelien at aurel32.net>:
> 
> > On 2017-12-06 19:39, Julien Aubin wrote:
> > > Weird... this time I re-upgraded libc6 and things work fine... looks like
> > > something wrong went during the install. And I cannot reproduce the issue
> > > anymore... :'( WTF ???
> >
> > Hmm, a bug has been introduced in libc6 version 2.24-11+deb9u2, which in
> > some conditions leave the /etc/ld.so.nohwcap file instead of removing it
> > just after the upgrade (see bug#883394). One of the condition is to have
> > libc6-i686 installed (while it can be safely removed), which seems to be
> > your case.
> >
> > I consider this bug harmless as it should not deactivate anything now
> > that the default libc is already i686 optimized. Also I don't see how it
> > could trigger the issue you described. Anyway better be safe than sorry,
> > could you please try to create this file with "touch /etc/ld.so.nohwcap"
> > as root and see if it makes the issue to reappear? Once the test is done
> > you can then remove it.
> >
> > Thanks,
> > Aurelien
> >
> 
> 
> Bingo ! It was exactly this !
> 
> If I re-create the file for example it crashes glxgears. When I remove it
> glxgears works fine.
> 
> With GDB, the stack trace for when I run glxgears :
> 
> 0  0x00007ffff6b311a4 in pthread_mutex_lock (mutex=0x7ffff604e8c0) at
> forward.c:192
> #1  0x00007ffff5de1308 in __glDispatchNewVendorID () from
> /usr/lib/x86_64-linux-gnu/libGLdispatch.so.0
> #2  0x00007ffff60793c2 in ?? () from /usr/lib/x86_64-linux-gnu/libGLX.so.0
> #3  0x00007ffff607a1ac in ?? () from /usr/lib/x86_64-linux-gnu/libGLX.so.0
> #4  0x00007ffff6073170 in glXChooseVisual () from
> /usr/lib/x86_64-linux-gnu/libGLX.so.0
> #5  0x000055555555779f in ?? ()
> #6  0x0000555555555ae7 in ?? ()
> #7  0x00007ffff6a5c2e1 in __libc_start_main (main=0x555555555970, argc=1,
> argv=0x7fffffffe638, init=<optimized out>,
>    fini=<optimized out>, rtld_fini=<optimized out>,
> stack_end=0x7fffffffe628) at ../csu/libc-start.c:291
> #8  0x000055555555646a in ?? ()
> 

The libc6 package version 2.24-11+deb9u2 won't be in the next point
release so this issue won't trigger. That said there is clearly an
issue on the nvidia package, it should behave the same with and
without /etc/ld.so.nohwcap. I am therefore reopening this bug.

One of the first step to debug this issue would be to run ldd
/usr/bin/glxgears with and without /etc/ld.so.nohwcap and compare the
difference.

Aurelien

-- 
Aurelien Jarno                          GPG: 4096R/1DDD8C9B
aurelien at aurel32.net                 http://www.aurel32.net



More information about the pkg-nvidia-devel mailing list