Xorg signal 11's spontaneously

[Dennis Jacobfeuerborn]

Adam Jackson wrote:
> Dennis Jacobfeuerborn wrote:
>> Dennis Jacobfeuerborn wrote:
>>> In the last 10 days Xorg crashed three times with a signal 11 
>>> spontaneously. This is the backtrace I found in the logfile:
>>>
>>> Backtrace:
>>> 0: /usr/bin/Xorg(xf86SigHandler+0x87) [0x80b6d07]
>>> 1: [0x49a420]
>>> 2: /usr/lib/xorg/modules/libfb.so [0x134b81]
>>> 3: /usr/lib/xorg/modules/libfb.so [0x13c80c]
>>> 4: /usr/lib/xorg/modules/libfb.so(fbCompositeGeneral+0x855) [0x13a4e5]
>>> 5: /usr/lib/xorg/modules/libfb.so(fbComposite+0x1cd) [0x1433ad]
>>> 6: /usr/lib/xorg/modules/libxaa.so(XAAComposite+0x261) [0x81f001]
>>> 7: /usr/bin/Xorg [0x8156d06]
>>> 8: /usr/bin/Xorg [0x8153915]
>>> 9: /usr/bin/Xorg(CompositePicture+0x153) [0x8140a13]
>>> 10: /usr/bin/Xorg [0x814687c]
>>> 11: /usr/bin/Xorg [0x8143bd5]
>>> 12: /usr/bin/Xorg(Dispatch+0x19a) [0x80869ba]
>>> 13: /usr/bin/Xorg(main+0x485) [0x806e3e5]
>>> 14: /lib/libc.so.6(__libc_start_main+0xdc) [0x4cd7e4]
>>> 15: /usr/bin/Xorg(FontFileCompleteXLFD+0xbd) [0x806d721]
>>>
>>> Fatal server error:
>>> Caught signal 11.  Server aborting
>>>
>>> All three crashes happened when I was loading a page in Firefox and 
>>> after the nightly Firefox builds got switched to Cairo as their 
>>> rendering backend so I wonder if Cairo could somehow trigger this.
>>>
>>> I'm running Fedora Core Rawhide with the following Xorg packages:
>>> xorg-x11-server-Xorg-1.0.99.901-5
>>> xorg-x11-drv-nv-1.1.1-2
>>>
>>> I'm using the open source nv driver with the AccelMethod set to EXA 
>>> and the 'load "dri"' and 'load "glx"' lines commented out.
> 
> Neither DRI nor EXA matter at all to the nv driver, which supports neither.
> 
>> Just happened again and the backtrace looks similar to the one above. 
>> If I can do something to provide better information to debug this 
>> please tell me.
> 
> Install the debuginfo packages, ssh in from another machine, su, attach 
> gdb to X, 'handle SIGPIPE nostop', and 'cont'.  When the server 
> segfaults, 'bt f' should now give you useful backtrace info.

I filed bug 6704 about this in the fdo bugzilla and there someone pointed 
me the appropriate wiki page for debugging the X server. Since I don't have 
a second machine nearby right now I'll try the single machine approach 
described in the wiki.

> May also be related to a crasher recently fixed for x4a4 glyphs; I'll 
> backpatch that into rawhide and stable shortly.

Is there a way to deliberately trigger that bug? I haven't found a way to 
trigger the crash I'm seeing but if the glyph bug would result in the same 
backtrace as above then we'd know if this is really the same problem.
Right now I've set the NoTrapSignal option and enabled core dumping and now 
have to wait until X dies again.

Regards,
   Dennis