2.6.37: BUG at drivers/gpu/drm/i915/i915_gem.c:4190!

[Chris Wilson]

On Thu, 20 Jan 2011 11:49:18 +0100, Toralf Förster <toralf.foerster at gmx.de> wrote:
> Hello,
> 
> after the following BUG the system crashed :
> ...
> 2011-01-12T10:31:59.597+01:00 n22 kernel: kernel BUG at drivers/gpu/drm/i915/i915_gem.c:4190!
> 2011-01-12T10:31:59.597+01:00 n22 kernel: invalid opcode: 0000 [#1] SMP
> 2011-01-12T10:31:59.597+01:00 n22 kernel: last sysfs file: /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq
> 2011-01-12T10:31:59.597+01:00 n22 kernel: Modules linked in: bridge stp llc ipt_MASQUERADE ipt_REJECT xt_recent xt_tcpudp nf_conntrack_ftp xt_state 
> xt_limit ipt_LOG iptable_nat nf_nat nf_conntrack_ipv4 nf_conntrack nf_defrag_ipv4 iptable_filter xt_owner xt_multiport ip_tables x_tables snd_seq_oss 
> snd_seq_midi_event snd_seq snd_seq_device snd_pcm_oss snd_mixer_oss af_packet fuse hdaps input_polldev fbcon font bitblit softcursor acpi_cpufreq mperf 
> dm_mod arc4 ecb crypto_blkcipher snd_hda_codec_conexant cryptomgr i915 snd_hda_intel drm_kms_helper aead crypto_algapi usbhid hid thinkpad_acpi 
> snd_hda_codec snd_pcm drm iwlagn fb hwmon uhci_hcd ehci_hcd iwlcore fbdev snd_timer snd usbcore processor 8250_pci sr_mod e1000e intel_agp soundcore 
> i2c_algo_bit wmi cfbcopyarea cfbimgblt i2c_i801 battery ac psmouse mac80211 cdrom evdev nvram intel_gtt agpgart video button i2c_core snd_page_alloc cfbfillrect 
> thermal output nls_base cfg80211 rfkill 8250 serial_core [last unloaded: microcode]
> 2011-01-12T10:31:59.597+01:00 n22 kernel:
> 2011-01-12T10:31:59.597+01:00 n22 kernel: Pid: 5057, comm: X Not tainted 2.6.37 #4 6474B84/6474B84
> 2011-01-12T10:31:59.597+01:00 n22 kernel: EIP: 0060:[<f89f9f6e>] EFLAGS: 00013246 CPU: 0
> 2011-01-12T10:31:59.597+01:00 n22 kernel: EIP is at i915_gem_object_pin+0x15e/0x190 [i915]
> 2011-01-12T10:31:59.597+01:00 n22 kernel: EAX: 0003c47c EBX: f4cd95c0 ECX: 00000000 EDX: 0003c000
> 2011-01-12T10:33:34.621+01:00 n22 syslog-ng[2754]: The current log file has a mismatching size/inode information, restarting from the beginning; 
> filename='/proc/kmsg'
> 2011-01-12T10:33:34.648+01:00 n22 kernel: Linux version 2.6.37 (root at n22) (gcc version 4.4.4 (Gentoo 4.4.4-r2 p1.2, pie-0.4.5) ) #4 SMP Sun Jan 9 16:29:30 CET 
It's an old pin leak [https://bugzilla.kernel.org/show_bug.cgi?id=17181].
I keep finding little buglets that might contribute, but so far
inconclusive. The latest one was:

commit 0ba41e449fd0f45f5b29c1009020ab1b298bedda
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date:   Sat Jan 8 15:10:41 2011 +0000

    drm: Restore the old_fb upon modeset failure
    
    ... or else we may end up disabling the wrong framebuffer, leading to an
    OOPS, e.g:
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre