[Intel-gfx] SNA crashes

[Zdenek Kabelac]

2012/5/24 Chris Wilson <chris at chris-wilson.co.uk>:
> On Thu, 24 May 2012 16:15:41 +0200, Zdenek Kabelac <zdenek.kabelac at gmail.com> wrote:
>> Hi
>>
>> Since I've started to use  SNA driver (when UXA appeared to be very
>> slow), I'm  facing higher number of Xorg crashes.
>> SNA is faster, but less stable then UXA is.
>
> They are all debugging assertions caused by the earlier GPU hang, to
> catch instances were we are losing commands and thus potentially
> corrupting the output due to the loss of the GPU. Tip, don't hang the
> GPU or don't ask for the bt.
> -Chris
>
> --


Ok - I think I've figured out relatively simple way to trigger this
issue on my machine.
I'll also probably provide a list of memleaks which popped out and
other sets of problem.

So I just have to run  Xorg server with valgrind and run couple programs.
What I've not yet figured out is - if the is really fault of intel
driver - or synaptic handle
(which seems to have higher probability)

Anyway - here is the log from valgring just in the moment of kernel
reported GPU hang.

Process terminating with default action of signal 11 (SIGSEGV): dumping core
 General Protection Fault
   at 0x6B23165: __sprintf_chk (sprintf_chk.c:27)
   by 0x468E0C: LogVWrite (stdio2.h:34)
   by 0x468B3B: ErrorF (log.c:641)
   by 0x550765: mieqEnqueue (mieq.c:280)
   by 0x447F01: queueEventList (getevents.c:974)
   by 0x48F9CF: xf86PostMotionEvent (xf86Xinput.c:1045)
   by 0xB294D14: HandleState (synaptics.c:3370)
   by 0xB296D91: ReadInput (synaptics.c:1721)
   by 0x480206: xf86SigioReadInput (xf86Events.c:298)
   by 0x4A439F: xf86SIGIO (sigio.c:108)
   by 0x582054F: ??? (in /usr/lib64/libpthread-2.15.90.so)
   by 0x6B06316: ioctl (syscall-template.S:81)
   by 0x8775807: drmIoctl (xf86drm.c:167)
   by 0x8777B55: drmCommandNone (xf86drm.c:2357)
   by 0x8BBC84F: kgem_throttle (kgem.c:1924)
   by 0x8BE4A16: sna_accel_block_handler (sna_accel.c:12069)
   by 0x438149: BlockHandler (dixutils.c:387)
   by 0x462313: WaitForSomething (WaitFor.c:210)
   by 0x434080: Dispatch (dispatch.c:357)
   by 0x423404: main (main.c:288)

(using git 8ea4ba081de0206351394481f54dcbe6922a085b for intel driver
(to match code lines))

I'm unsure how the handling of synaptic events and mouse cursor
movement are related together.

Is there something I should enable for kernel part of intel driver to
get better explanation for GPU hang ?

Zdenek