Bug 22283

Created attachment 26774 [details]
GPU dump of crashed chip

Here is the gpu dump of a crashed chip.  When the chip goes down the batch
buffer head is located at,

> 0x0c1db014: HEAD 0x54f08806: XY_SRC_COPY_BLT (rgb enabled, alpha enabled, src
> tile 1, dst tile 1)
> 0x0c1db018: HEAD 0x03cc0780:    format 8888, dst pitch 1920, clipping disabled
> 0x0c1db01c: HEAD 0x04980766:    dst (1894,1176)
> 0x0c1db020: HEAD 0x04b00780:    dst (1920,1200)
> 0x0c1db024: HEAD 0x04331000:    dst offset 0x04331000
> 0x0c1db028: HEAD 0x04980766:    src (1894,1176)
> 0x0c1db02c: HEAD 0x00000780:    src pitch 1920
> 0x0c1db030: HEAD 0x06696000:    src offset 0x06696000

I'm not certain, but it seems like the dst and src offsets seem to be invalid
given that tiling is enabled:

> Destination Base Address: (base address of the destination surface: X=0, Y=0)
> When Dest Tiling is enabled (Bit 11 enabled), this address is limited to
> 4Kbytes.
> 
> Source Base Address: (base address of the source surface: X=0, Y=0)
> When Src Tiling is enabled (Bit 15 enabled), this address is limited to
> 4Kbytes.

Then again, ickle suggested the quite reasonable proposal that the
specification is actually trying to say that the address must be aligned to
4kB, consistent with the addresses in the command. Need feedback from someone
at intel to confirm the actual nature of this requirement.

This is intended to place kernel.org Bug #13490 (http://bugzilla.kernel.org/show_bug.cgi?id=13490) in the right bugzilla. I am seeing this on GM965, although others on that bug have reported seeing it on G45 as well.

Same here on Intel GMA950. This problem occurs on linux-2.6.30 and xserver-xorg-video-intel-2.7.99.901-2 (from Debian experimental).

The problem doesn't exist on linux-2.6.30-rc8 with xserver-xorg-video-intel-2.7.99.901-2

or linux-2.6.30 and xserver-xorg-video-intel-2.7.1 (from Debian unstable).

This is fixed for me as of xf86-video-intel 246cec965958e94babf5377e6f221522b05fb458.

Ok, tentatively closing then..  Mateusz please re-open if you see the issue with git master.

(In reply to comment #5)
> Ok, tentatively closing then..  Mateusz please re-open if you see the issue
> with git master.
> 

I think I must have spoke too soon. My GM965 (running master) crashed yet again while I was sleep last night. Post-mortem data coming.

Created attachment 26891 [details]
GPU dump of last night's crash

Very similar crash point,

0x047ab014: HEAD 0x54f08806: XY_SRC_COPY_BLT (rgb enabled, alpha enabled, src tile 1, dst tile 1)
0x047ab018: HEAD 0x03cc0780:    format 8888, dst pitch 1920, clipping disabled
0x047ab01c: HEAD 0x00000000:    dst (0,0)
0x047ab020: HEAD 0x04b00780:    dst (1920,1200)
0x047ab024: HEAD 0x020c9000:    dst offset 0x020c9000
0x047ab028: HEAD 0x00000000:    src (0,0)
0x047ab02c: HEAD 0x00000780:    src pitch 1920
0x047ab030: HEAD 0x0ae2c000:    src offset 0x0ae2c000

Created attachment 26892 [details]
More post-mortem data

Includes the following, 

date
dmesg
versions
Xorg.log
debugfs/i915_ringbuffer_data
debugfs/i915_gem_flushing
debugfs/i915_batchbuffers
debugfs/name
debugfs/i915_gem_inactive
debugfs/vma
debugfs/i915_gem_fence_regs
debugfs/i915_gem_request
debugfs/vm
debugfs/i915_gem_active
debugfs/i915_gem_interrupt
debugfs/clients
debugfs/gem_objects
debugfs/i915_gem_hws
debugfs/queues
debugfs/bufs
debugfs/i915_gem_seqno
debugfs/i915_ringbuffer_info
debugfs/gem_names

Definitely not fixed, reopening.

Who is still seeing this? My harddrive decided to quit late last week and after rebuilding from scratch, the issue seems to have disappeared (fingers crossed) on my primary machine. My G45, however, still seems to be dead every morning after running blanked all night.

*** This bug has been marked as a duplicate of bug 22383 ***