Skip to contentSkip to conversation

Another GPU hang

DISCUSSION Dmitry Nezhevenko 2 participants 6 messages May 21, 2011 – May 25, 2011

lore
Share & tools
lore
#0 lore raw

Hi,

I'm not sure why I'm getting this. I've performed a lot of configuration changes, so can't be sure, which one causes. I've switched to amd64 distro and upgraded everything to latest debian unstable. Also I'm on 2.6.39 now.

I'm getting such hang sometimes just after unplugging HDMI cable from laptop.

[ 8401.416156] [drm:i915_hangcheck_elapsed] *ERROR* Hangcheck timer elapsed... GPU hung
[ 8401.416169] [drm] capturing error event; look for more information in /debug/dri/0/i915_error_state
[ 8401.418055] [drm:i915_wait_request] *ERROR* i915_wait_request returns -11 (awaiting 154589 at 154588, next 154591)
[ 8401.920228] [drm:i915_reset] *ERROR* Failed to reset chip.

Actually the monitor is DVI, connected via HDMI->DVI cable to ASUS F6A laptop with Intel GM45 graphics.

Ii915_error_state file is something around 1.5 mb, so I've uploaded it to
http://dion.org.ua/uploads/2011/05/2.6.39_intel_hung.txt

xserver-xorg 1:7.6+6
xserver-xorg-core 2:1.10.1-2
xserver-xorg-video-intel 2:2.15.0-3

Kernel version 2.6.39.

Any ideas?

Signature · 1 line
WBR, Dmitry
#1 lore raw
Quoted 9 lines · Dmitry Nezhevenko

Hi,

I'm not sure why I'm getting this. I've performed a lot of configuration changes, so can't be sure, which one causes. I've switched to amd64 distro and upgraded everything to latest debian unstable. Also I'm on 2.6.39 now.

I'm getting such hang sometimes just after unplugging HDMI cable from laptop.

Hmm. Are you absolutely sure? I've an open bug 35576, but nothing that indicated a correlation with modeswitching. The hang would appear to be due to waiting on a dead pipe then. We've had a long history with such bugs and purposely changing modes and so are now careful to flush any pending waits from userspace and in KMS before modeswitching. This raises the question that maybe there is a window for a hotplug event to turn-off the pipe before userspace has finished flushing its queue of pending ops.

The important question is do you see this 0x01820000 at other times? -Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#2 lore raw
Quoted 11 lines · Chris Wilson
Quoted 9 lines · Dmitry Nezhevenko

Hi,

I'm not sure why I'm getting this. I've performed a lot of configuration changes, so can't be sure, which one causes. I've switched to amd64 distro and upgraded everything to latest debian unstable. Also I'm on 2.6.39 now.

I'm getting such hang sometimes just after unplugging HDMI cable from laptop.

Quoted 7 lines

Hmm. Are you absolutely sure? I've an open bug 35576, but nothing that indicated a correlation with modeswitching. The hang would appear to be due to waiting on a dead pipe then. We've had a long history with such bugs and purposely changing modes and so are now careful to flush any pending waits from userspace and in KMS before modeswitching. This raises the question that maybe there is a window for a hotplug event to turn-off the pipe before userspace has finished flushing its queue of pending ops.

I've just tried to reproduce it again. It looks like I'm unable to do it when external display (DVI connected via DMI->HDMI cable) is turned off via xrandr.

However if display is active (LVDS turned off and HDMI1 active) it was easy enough to reproduce this by just playing with HDMI connector. So after 5-6 attempts to plug/unplug monitor I've got hung again:

[38440.632948] [drm:drm_edid_block_valid] *ERROR* EDID checksum is invalid, remainder is 81
[38440.632956] [drm:drm_edid_block_valid] *ERROR* Raw EDID:
[38440.632962] <3>00 ff ff ff ff ff ff 00 38 a3 8c 67 01 01 01 01  ........8..g....
[38440.632968] <3>07 14 01 03 80 34 20 78 ea fc 85 a4 55 4d 9d 25  .....4 x....UM.%
[38440.632973] <3>12 50 54 bf ef 80 81 c0 81 80 90 40 8b c0 95 00  .PT........@....
[38440.632978] <3>a9 40 b3 00 d1 00 28 3c 80 a0 70 b0 23 40 30 20  .@....(<..p.#@0
[38440.632982] <3>36 00 06 44 21 00 00 1a 00 00 00 fd 00 32 55 1f  6..D!........2U.
[38440.632987] <3>5c 11 00 0a 20 20 20 20 20 20 00 00 00 fc 00 4c  \...      .....L
[38440.632992] <3>43 44 32 34 39 30 57 55 58 69 32 0a 00 00 00 ff  CD2490WUXi2.....
[38440.632997] <3>00 30 32 33 30 32 34 37 31 55 4f ff ff ff ff ff  .02302471UO.....
[38440.633000]
[38461.596072] [drm:i915_hangcheck_elapsed] *ERROR* Hangcheck timer elapsed... GPU hung
[38461.596083] [drm] capturing error event; look for more information in /debug/dri/0/i915_error_state
[38461.597995] [drm:i915_wait_request] *ERROR* i915_wait_request returns -11 (awaiting 2888267 at 2888261, next 2888268)
[38462.100045] [drm:i915_reset] *ERROR* Failed to reset chip.
Quoted 2 lines

The important question is do you see this 0x01820000 at other times? -Chris

You are asking about "IPEHR:" line in i915_error_state file, right? Now it was
IPEHR: 0x01800002.

I've uploaded full file to:
http://dion.org.ua/uploads/2011/05/2.6.39_intel_hung_2.txt

As about other hungs, some times ago there were hungs while playing video using mplayer -vo xv. I've asked about it here and you replied that it's fixed in commit:

commit 23f9b14df7c102c1036134835dd5d1a508059858

Author: Chris Wilson <[email protected]>
Date:   Sat Feb 12 10:42:34 2011 +0000                                                                                                                                                                                                  

i965: Remove broken maximum base addresses from video

So after upgrading to proper debian package everything was OK. At least currently I don't remember any hung except display plug/unplug

Signature · 1 line
WBR, Dmitry
#3 lore raw
Quoted 6 lines · Dmitry Nezhevenko

I've just tried to reproduce it again. It looks like I'm unable to do it when external display (DVI connected via DMI->HDMI cable) is turned off via xrandr.

However if display is active (LVDS turned off and HDMI1 active) it was easy enough to reproduce this by just playing with HDMI connector. So after 5-6 attempts to plug/unplug monitor I've got hung again:

Ok, and I just got another bug (37450) which makes me think that, at least, part of this problem is due to the stale DPMS values. (We depend upon the DPMS state for not feeding WAITs onto a disabled pipe).
https://bugzilla.kernel.org/show_bug.cgi?id=24982
-Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#4 lore raw
Quoted 8 lines · Chris Wilson
Quoted 6 lines · Dmitry Nezhevenko

I've just tried to reproduce it again. It looks like I'm unable to do it when external display (DVI connected via DMI->HDMI cable) is turned off via xrandr.

However if display is active (LVDS turned off and HDMI1 active) it was easy enough to reproduce this by just playing with HDMI connector. So after 5-6 attempts to plug/unplug monitor I've got hung again:

Quoted 6 lines

Ok, and I just got another bug (37450) which makes me think that, at least, part of this problem is due to the stale DPMS values. (We depend upon the DPMS state for not feeding WAITs onto a disabled pipe).
https://bugzilla.kernel.org/show_bug.cgi?id=24982
-Chris

There is a reference to commit 811aaa55ba21ab37407018cfc01770d6b037d3fb in bugreport. I've double checked that this commit is present in 2.6.39 kernel I'm using. So this is probably not applicable to me.

Should I create bugreport at kernel.org for this issue?

PS. In case if you've some patch or other things to do, I can try it.

Signature · 1 line
WBR, Dmitry
#5 lore raw
Quoted 7 lines · Dmitry Nezhevenko
Quoted 6 lines

Ok, and I just got another bug (37450) which makes me think that, at least, part of this problem is due to the stale DPMS values. (We depend upon the DPMS state for not feeding WAITs onto a disabled pipe).
https://bugzilla.kernel.org/show_bug.cgi?id=24982
-Chris

Quoted 7 lines

There is a reference to commit 811aaa55ba21ab37407018cfc01770d6b037d3fb in bugreport. I've double checked that this commit is present in 2.6.39 kernel I'm using. So this is probably not applicable to me.

Should I create bugreport at kernel.org for this issue?

PS. In case if you've some patch or other things to do, I can try it.

Just a note. I've once reproduced my issue without even plugging connectors. It was enough to just switch from HDMI1 to LVDS output using xrandr.

Regards

Signature · 1 line
WBR, Dmitry