during a mpeg2 video playback, GPU hang happen at exactly same frames.
there 8 hang frames of 1499 total frames in the clip.

divide the clip into 1 second long small pieces by ffmpeg with -c copy, just copy the stream, GPU hang happen on the pieces include hang frame.

software decode OK

1) reproduced on Fedora 18(kernel 3.8) + OTC driver on haswell, tried gstreamer vaapi-mplayer, and GPU hang even gstreamer fakesink used
2) cannot reproduce on SUSE(kernel 3.0) + intel media SDK on haswell, output yuv file OK
both kernels use i915 drm module, it seems more likely intel VA driver do something wrong

dump the headers of each frame, no special parameter or extra big frame.
dump batch buffer(not include reloc trees) and ring buffer, seems no overfolw

dmesg messages
[  421.443060] [drm:i915_hangcheck_elapsed] *ERROR* Hangcheck timer elapsed... GPU hung
[  421.443071] [drm] capturing error event; look for more information in /debug/dri/0/i915_error_state
[  421.469452] [drm:i915_wait_request] *ERROR* i915_wait_request returns -11 (awaiting 128404 at 128398, next 128407)
[  541.189805] [drm:i915_hangcheck_elapsed] *ERROR* Hangcheck timer elapsed... GPU hung
[  541.189870] [drm:i915_wait_request] *ERROR* i915_wait_request returns -11 (awaiting 175021 at 175015, next 175024)
[ 1896.540267] [drm:i915_hangcheck_elapsed] *ERROR* Hangcheck timer elapsed... GPU hung
[ 1896.540309] [drm:i915_wait_request] *ERROR* i915_wait_request returns -11 (awaiting 427313 at 427308, next 427324)
[ 5782.218952] [drm:i915_hangcheck_elapsed] *ERROR* Hangcheck timer elapsed... GPU hung
[ 5782.219024] [drm:i915_wait_request] *ERROR* i915_wait_request returns -11 (awaiting 885116 at 885111, next 885120)


may attach the video clip later