Created attachment 145881 [details]
is 04 12:59:27 linux kernel: i915 0000:00:02.0: GPU HANG: ecode 4:1:0x00930f09, in systemd-logind , hang on rcs0
lis 04 12:59:27 linux kernel: GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.
lis 04 12:59:27 linux kernel: Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel
lis 04 12:59:27 linux kernel: drm/i915 developers can then reassign to the right component if it's not a kernel issue.
lis 04 12:59:27 linux kernel: The GPU crash dump is required to analyze GPU hangs, so please always attach it.
lis 04 12:59:27 linux kernel: GPU crash dump saved to /sys/class/drm/card0/error
lis 04 12:59:27 linux kernel: i915 0000:00:02.0: Resetting chip for hang on rcs0
Using git bff5eca49b27cb47673123222a714d2a6f56287f
Author: Chris Wilson <email@example.com>
Date: Mon Oct 7 14:16:28 2019 +0100
Just started desktop with xfce4 (no fancy 3d GUI) on T61
Some pixel data was blitted into the middle of the batch: 0x00843000 - 0x008431a0. That the initial byte is page aligned makes me suspicious of a TLB error.
Earlier, we have
0x0083d18c: 0x54300804: XY_COLOR_BLT (rgb enabled, alpha enabled, src tile
0, dst tile 1)
0x0083d190: 0x03f00480: format 8888, pitch 1152, rop 0xf0, clipping disa
0x0083d194: 0x00000000: (0,0)
0x0083d198: 0x03250446: (1094,805)
0x0083d19c: 0x07529000: offset 0x07529000
0x0083d1a0: 0xfff6f5f4: color
as the only thing blitting with 0xfff6f5f4; which would fill
0x07529000 - 0x79f8e00 so not overlapping with the dead zone in the batch. None of the 3DSTATE overlaps the batch. It could have been a result of an earlier batch that we don't capture; but atm TLB is my prime suspect.