Strange X crash...

[marmarek]

Reported by joanna on 1 Jul 2011 22:15 UTC
Just got an X server crash in Dom0:

Jul  2 00:12:01 dom0 kernel: [[drm:i915_gem_mmap_gtt_ioctl](19495.868185]) *ERROR* Attempting to mmap a purgeable buffer
Jul  2 00:12:02 dom0 kdm[X server for display :0 terminated unexpectedly
Jul  2 00:12:43 dom0 kdm[2730](2730]:): Cannot execute 'grub-set-default': not in $PATH.
Jul  2 00:13:02 dom0 pulseaudio[ratelimit.c: 199 events suppressed
Jul  2 00:13:07 dom0 init: tty (/dev/tty2) main process (2735) killed by TERM signal
Jul  2 00:13:07 dom0 init: tty (/dev/tty3) main process (2737) killed by TERM signal
Jul  2 00:13:07 dom0 init: tty (/dev/tty4) main process (2739) killed by TERM signal
Jul  2 00:13:07 dom0 init: tty (/dev/tty5) main process (2741) killed by TERM signal
Jul  2 00:13:07 dom0 init: tty (/dev/tty6) main process (2743) killed by TERM signal
Jul  2 00:13:07 dom0 init: system-setup-keyboard main process (1639) killed by TERM signal
Jul  2 00:13:19 dom0 logger: /etc/xen/scripts/block-snapshot: Device /dev/mapper/snapshot-fd01:7213618-fd01:7213893 still in use - not removing
Jul  2 00:13:20 dom0 logger: /etc/xen/scripts/block-snapshot: Device /dev/mapper/snapshot-fd01:7083000-fd01:7082994 still in use - not removing
Jul  2 00:13:22 dom0 logger: /etc/xen/scripts/block-snapshot: Device /dev/mapper/snapshot-fd01:7083000-fd01:7082994 still in use - not removing
Jul  2 00:13:22 dom0 logger: /etc/xen/scripts/block-snapshot: Device /dev/mapper/snapshot-fd01:7083000-fd01:7082994 still in use - not removing
Jul  2 00:13:22 dom0 logger: /etc/xen/scripts/block-snapshot: Device /dev/mapper/snapshot-fd01:7083000-fd01:7082994 still in use - not removing
Jul  2 00:13:25 dom0 kernel: [19579.131128](3009]:) xl[segfault at fffffffffffffffb ip 00007f2c5abec83c sp 00007fff4d512aa8 error 4 in libc-2.12.so[7f2c5ab72000+175000](14012]:)
Jul  2 00:13:25 dom0 kernel: [xl[14014](19579.140306]): segfault at fffffffffffffffb ip 00007f2f756eb83c sp 00007fff41569818 error 4 in libc-2.12.so[7f2f75671000+175000]
Jul  2 00:13:25 dom0 kernel: Kernel logging (proc) stopped.

I admit, I have run a lot of AppVMs. However, qmemman should have ensure that Dom0 always have some minimum of mem (it used to work this way in Beta 2).

Migrated-From: https://wiki.qubes-os.org/ticket/256

[marmarek]

Comment by rafal on 4 Jul 2011 09:25 UTC
I do not see any OOM messages in the attached log, so I am not convinced it is a result of OOM.
The "Attempting to mmap a purgeable buffer" seems to be a common fatal error in google results.

In fact the xl segfault is more interesting - but again, I see no evidence OOM condition occurred. Did you do "free" in dom0 after crash ?

Qmemman will not take away memory from dom0 when it is under memory pressure. However, if no other VMs are above their prefmem, no memory will be available for dom0, even if it is below its prefmem as well. So, it is quite possible that if these is something memory hungry in dom0, it can get OOM even with qmmeman functioning as designed - just the above logs does not necessarily show it happened.

[marmarek]

Comment by rafal on 4 Jul 2011 09:25 UTC
I do not see any OOM messages in the attached log, so I am not convinced it is a result of OOM.
The "Attempting to mmap a purgeable buffer" seems to be a common fatal error in google results.

In fact the xl segfault is more interesting - but again, I see no evidence OOM condition occurred. Did you do "free" in dom0 after crash ?

Qmemman will not take away memory from dom0 when it is under memory pressure. However, if no other VMs are above their prefmem, no memory will be available for dom0, even if it is below its prefmem as well. So, it is quite possible that if these is something memory hungry in dom0, it can get OOM even with qmmeman functioning as designed - just the above logs does not necessarily show it happened.

[marmarek]

Comment by joanna on 12 Jul 2011 14:20 UTC
This might be relevant:

https://bugs.launchpad.net/ubuntu/+source/linux/+bug/682712

[marmarek]

Comment by joanna on 12 Jul 2011 14:20 UTC
This might be relevant:

https://bugs.launchpad.net/ubuntu/+source/linux/+bug/682712

[marmarek]

Modified by joanna on 12 Jul 2011 19:01 UTC

[marmarek]

Modified by joanna on 12 Jul 2011 19:01 UTC

[marmarek]

Comment by joanna on 18 Jul 2011 10:18 UTC
This happens quite regularly on my system :( Just to repeat: KDE, Intel Core i5 IGD, Composition Enabled (Open GL).

This is what was reported in dmesg immediately before (or just after) this crash:

[[drm:i915_gem_mmap_gtt_ioctl](12227.329391]) *ERROR* Attempting to mmap a purgeable buffer

And in /var/log/messages:

Jul 18 11:36:03 dom0 kernel: [[drm:i915_gem_mmap_gtt_ioctl](12227.329391]) *ERROR* Attempting to mmap a purgeable buffer
Jul 18 11:36:04 dom0 kdm[3133]: X server for display :0 terminated unexpectedly

This is very serious issue, so I'm bumping this to Critical...

[marmarek]

Comment by joanna on 18 Jul 2011 10:18 UTC
This happens quite regularly on my system :( Just to repeat: KDE, Intel Core i5 IGD, Composition Enabled (Open GL).

This is what was reported in dmesg immediately before (or just after) this crash:

[[drm:i915_gem_mmap_gtt_ioctl](12227.329391]) *ERROR* Attempting to mmap a purgeable buffer

And in /var/log/messages:

Jul 18 11:36:03 dom0 kernel: [[drm:i915_gem_mmap_gtt_ioctl](12227.329391]) *ERROR* Attempting to mmap a purgeable buffer
Jul 18 11:36:04 dom0 kdm[3133]: X server for display :0 terminated unexpectedly

This is very serious issue, so I'm bumping this to Critical...

[marmarek]

Comment by joanna on 1 Aug 2011 19:07 UTC
Sigh... have been unable to reproduce it for the last 2 weeks or so... Again, this seems nothing Qubes- or Xen-related, so I think we should just close it for now. In case somebody finds a way to reproduce it we might reopen it in Beta 3. Hopefully an xorg and kde update in beta3 will definitely solve this.

[marmarek]

Comment by joanna on 1 Aug 2011 19:07 UTC
Sigh... have been unable to reproduce it for the last 2 weeks or so... Again, this seems nothing Qubes- or Xen-related, so I think we should just close it for now. In case somebody finds a way to reproduce it we might reopen it in Beta 3. Hopefully an xorg and kde update in beta3 will definitely solve this.