Created attachment 130752 [details]
GPU crash dump saved to /sys/class/drm/card0/error

Encountered GPU hang on Chromebook while doing Google hangouts + video(youtube) + few (5) tabs opened simultaneously. Reproducility is nearly 20% but requires overnight to reproduce this issue.

GPU crash dump saved to /sys/class/drm/card0/error and logs are attached.

Kernel log says:
2017-03-26T15:51:27.709098-07:00 INFO kernel: [ 8104.736419] [drm] stuck on render ring
2017-03-26T15:51:27.709145-07:00 INFO kernel: [ 8104.737420] [drm] GPU HANG: ecode 9:0:0xef9ffffd, in chrome [1869], reason: Ring hung, action: reset
2017-03-26T15:51:27.709150-07:00 INFO kernel: [ 8104.737439] [drm] GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.


Few more details:
From "change_card0_drm.20170406.193114.0.log" attachment, I saw it suffered GPU command (MI_STORE_DATA_IMM) blocked in render ring and guess some batch buffers from mesa cause this issue.
...
0x01028a90:      0x7a000004: PIPE_CONTROL: no write, no depth stall, no RC write flush, no inst flush
0x01028a94:      0x00101001:    destination address
0x01028a98:      0x00000080:    immediate dword low
0x01028a9c:      0x00000000:    immediate dword high
0x01028aa8:      0x10400002: MI_STORE_DATA_IMM <------------blocked
0x01028aac:      0x000020c0:    dword 1
0x01028ab0:      0x00000000:    dword 2
0x01028ab4:      0x0014d95b:    dword 3
0x01028ab8:      0x01000000: MI_USER_INTERRUPT
0x01028abc:      0x00000000: MI_NOOP
0x01028ac0:      0x00000000: MI_NOOP
0x01028ac4:      0x00000000: MI_NOOP
0x01028ac8:      0x7a000004: PIPE_CONTROL: no write, no depth stall, no RC write flush, no inst flush
0x01028acc:      0x00000000:    destination address
0x01028ad0:      0x00000000:    immediate dword low
0x01028ad4:      0x00000000:    immediate dword high

...