[Nouveau] Trouble on kernel

[Ilia Mirkin]

On Mon, Oct 28, 2013 at 12:31 PM, Martin Kaffanke
<martin.kaffanke at gmx.at> wrote:
> Am 2013-10-28 15:42, schrieb Ilia Mirkin:
>
>> On Mon, Oct 28, 2013 at 3:45 AM, Martin Kaffanke <martin.kaffanke at gmx.at>
>> wrote:
>>>
>>> Hi there,
>>>
>>> Sometimes my computer crashes since i use now ubuntu 13.10 with
>>> unity/compiz.
>>
>>
>> You're on NV4C. Do you have mesa 9.2.1 or later (or pre-9.2)? If you
>> have 9.2.0, there are known issues with shader compilation.
>
>
> it is 9.2.1 as far as i can see in the package names containing mesa (glx
> and dri)

OK, good.

>
>
>>> dmesg tells me:
>>>
>>> [  686.209019] nouveau E[    PBUS][0000:00:0d.0] MMIO write of 0x00000000
>>> FAULT at 0x00b020
>>
>>
>> Odd. That's VPE. You should have one... maybe not. Nothing to worry
>> about though. (Especially odd that it's showing up multiple times --
>> are you trying to use xvmc?)
>
>
> At the moment I'm trying nothing, its the default installation, i didn't
> change something.
>
> Where do I have to look to find it out?

Meh. Like I said, don't worry about it. On closer inspection, it's
most likely the nv31_mpeg_tile_prog function. Although why it's
getting addresses of 0... I don't know. (And also why it's getting
run, I don't know either. Something must be starting up either xvmc or
vdpau, the latter of which won't work for your card.)

>
>
>>> and later:
>>>
>>> [ 1425.565441] nouveau E[  PGRAPH][0000:00:0d.0]  ERROR nsource:
>>> DATA_ERROR
>>> nstatus: BAD_ARGUMENT
>>> [ 1425.565462] nouveau E[  PGRAPH][0000:00:0d.0] ch 3 [0x00056000
>>> compiz[1970]] subc 2 class 0x0039 mthd 0x0314 data 0x031dc000
>>
>>
>> A bit more concerning:
>>
>> $ lookup -a 4c -d NV01_SUBCHAN -- -v obj-class NV03_M2MF 0x0314 0x031dc000
>> PITCH_IN => 52281344
>
>
> lookup seems not to work, the package 'lookup' seems to be a different tool,
> right?

envytools. rnn/lookup. But you don't need to run it, I was just using
it to decode that error message to figure out what command triggered
the bad argument error.

>
>
>> I'm sure that's above some limit. A pitch of 50MB... I wonder how
>> that's happening... If you feel like debugging, the place to look is
>> nv30_transfer.c (in mesa). Although I would think this would merely
>> cause misrendering, not crashes.
>
>
> I'm not going to do that, because I really not that competent for that, even
> if i know about c programming.
>
> I also think that this can cause missrendering, and not crashes, but it also
> seems
> that there aren't stored any messages on the crash, maybe it crashes before
> it can
> write something to the harddisk?

Well, if compiz/something else *crashes* (i.e.
segfault/sigbus/whatever), it's likely there's a log of it. Perhaps
look in .xsession-errors, or run it inside gdb? Or are you saying that
you get a hard hang? Can you ssh into the machine when that happens?

  -ilia