[Mesa-dev] [PATCH 1/6] glx: Convert to automake.

[Matt Turner]

On Sun, Mar 18, 2012 at 10:17 PM, Matt Turner <mattst88 at gmail.com> wrote:
> On Mon, Mar 5, 2012 at 9:37 PM, Eric Anholt <eric at anholt.net> wrote:
>> On Thu, 1 Mar 2012 19:18:50 +0100, Marek Olšák <maraeo at gmail.com> wrote:
>>> On Thu, Mar 1, 2012 at 5:28 PM, Eric Anholt <eric at anholt.net> wrote:
>>> > On Thu, 1 Mar 2012 16:50:40 +0100, Marek Olšák <maraeo at gmail.com> wrote:
>>> >> Hi Eric,
>>> >>
>>> >> this commit breaks StarCraft II running on Wine traced by apitrace
>>> >> on r600g. Wine is most probably stuck in a deadlock. It prints these
>>> >> error messages:
>>> >>
>>> >> err:seh:setup_exception_record stack overflow 2412 bytes in thread
>>> >> 0024 eip 00437fdb esp 00c409c4 stack 0xc40000-0xc41000-0xd40000
>>> >> err:ntdll:RtlpWaitForCriticalSection section 0x7dce41e0
>>> >> "x11drv_main.c: X11DRV_CritSection" wait timed out in thread 0025,
>>> >> blocked by 0024, retrying (60 sec)
>>> >> err:ntdll:RtlpWaitForCriticalSection section 0x72e200 "?" wait timed
>>> >> out in thread 0023, blocked by 0024, retrying (60 sec)
>>> >> err:ntdll:RtlpWaitForCriticalSection section 0x72e200 "?" wait timed
>>> >> out in thread 0009, blocked by 0024, retrying (60 sec)
>>> >> etc.
>>> >>
>>> >> Reverting this commit (and also the egl automake commits to avoid
>>> >> conflicts) fixes the issue.
>>> >>
>>> >> Do you have any idea about what might have gone wrong?
>>> >
>>> > Hmm, weird.  Defines were my first idea, the diff of all defines used in
>>> > the compile is:
>> [...]
>>> >
>>> > Could you stick a debugger on it and see where it's stopped, at least?
>>>
>>> Sorry, I haven't been able to obtain a useful backtrace from wine. I
>>> have no idea how to debug this.
>>
>> Just to be clear: Is the failure mode only when apitracing wine, and not
>> when running wine standalone or when running glretrace of any resulting
>> trace?
>>
>> Does it require SC2, or can other apps be used to reproduce the failure?
>> Any freely downloadable apps by chance?
>>
>> (That said, wine appears to be totally hosed for me currently, on 8.0 or
>> master, with a segfault almost all the time due to a missing buffer
>> object for one of my winsys renderbuffers)
>
> In the process of trying to figure out why SC2 was so slow today, I
> hit a GPU hang, so I tried apitracing it and hit this problem.
>
> I reverted automaking glx and was able to trace it. I tried to replay
> the trace (this time with a 64-bit apitrace and i965_dri.so, which
> crashed). I rebuilt 64-bit Mesa with automake glx reverted and sure
> enough it worked, so I've got a trace that works before and fails
> after. Hopefully it'll point to the problem of SC2 hanging under
> WINE/apitrace.
>
> I'll figure out how to trim it (it's 5.4GB now) and send it to you.
>
> Matt

Or maybe it was just a bug in apitrace. I cannot reproduce it now.