Created attachment 63870 [details] Xorg log from session Occasionally while playing video (either in Totem or a Flash player in a web-page), playback and other elements on the screen pause. A message like [drm:i915_hangcheck_ring_idle] *ERROR* Hangcheck timer elapsed... blt ring idle [waiting on 651163, at 651163], missed IRQ? appears in dmesg. Full dmesg and Xorg log attached. Software: xorg-x11-drv-intel-2.19.0-5.20120612.fc17.x86_64 libdrm-2.4.33-3.fc17.x86_64 libdrm-2.4.33-3.fc17.i686 mesa-dri-drivers-8.0.3-1.fc17.x86_64 Kernel from kernel-3.4.4-3.fc17 source The only intel-specific xorg.conf.d entry is Section "Device" Identifier "Card0" Driver "intel" Option "AccelMethod" "sna" EndSection although I've seen this happen with UXA acceleration as well. Hardware is an Intel i7 2760QM processor, HD 3000 graphics.
Created attachment 63871 [details] full dmesg, up to "missed IRQ?"
Well, well, well. 3.4 should be up to date with all the known workarounds for the missed IRQs. Did this issue only start occurring recently? Have you used any older kernels without this issue?
(In reply to comment #2) > Did this issue only start occurring recently? Have you used any older kernels > without this issue? Maybe, but I'm not entirely sure. Trawling back through the logs, the earliest I *noted* this was on 3.4.0. However that's as far back as my logs go. I could reinstall a 3.3-series kernel if that might help.
Please do, it would be very very helpful to know if your system has always suffered, or if the more recent workarounds are not as effective.
I've not seen it so far playing videos using kernel 3.3.8...
From 3.3.8 to 3.4.8, things go south. Daniel, Ben does this match your expectations?
Well, semaphores=1 should have made these less likely. But since this is libva, which (iirc) still syncs in userspace this might not apply. Can you please test what happens when you boot 3.4 with i915.semaphores=0 ?
Is it always the blt ring?
Looking back in the logs: All the i915_hangcheck messages I see in my logs are either of the form [drm:i915_hangcheck_ring_idle] *ERROR* Hangcheck timer elapsed... blt ring idle [waiting on 2195097, at 2195097], missed IRQ? with 3.4-series kernels, or more recently under the 3.5 series: [drm:i915_hangcheck_ring_idle] *ERROR* Hangcheck timer elapsed... blitter ring idle I'm now using kernel 3.5.1, I'll try i915.semaphores=0 at the next opportunity.
3.5 should dump the i915_error_state into debugfs if this happens, please attach that?
(In reply to comment #10) > 3.5 should dump the i915_error_state into debugfs if this happens, please > attach that? Sorry for the delay. After seeing the associated pause and message in dmesg (kernel 3.5.3), i915_error_state reports no error state collected.
Hmm, perhaps try again with a more recent kernel? It should be dumping the error-state on the missed IRQ!
(In reply to comment #12) > Hmm, perhaps try again with a more recent kernel? It should be dumping the > error-state on the missed IRQ! Currently on 3.6.2. All I've seen so far is [drm:__gen6_gt_force_wake_get] *ERROR* Force wake wait timed out Don't know if this is the same issue, though. Still nothing in i915_error_state -- do I have to enable a flag somewhere?
Can you please attach your current dmesg? If that is a warning from only the first forcewake get, then it is harmless and a patch has already been applied.
Created attachment 68882 [details] dmesg output from 3.6.2, with [drm:__gen6_gt_force_wake_get] *ERROR* Force wake wait timed out There are 9 such messages so far. By the way, I am looking in the right place for i915_error_state (/sys/kernel/debug/dri/*/i915_error_state), yes?
Ok, that's the more worrying forcewake warning, though I suspect the patch to increase the timeout only landed in 3.7, so maybe not as scary as it appears. Something to keep an eye on though. However, the good news is that you have yet to see a missed IRQ. /sys/kernel/debug/dri/0/i915_error_state is the right file to read should you see another hang.
So I think we have the original missed IRQ fixed. Only we now have the rc6 warnings, which are being tracked in bug 50619. Closing for the original bug, but please do join the bug hunt on #50619.
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.