[Nouveau] Possible 3.13-rc nouveau regression with GT 560 Ti

[Ilia Mirkin]

On Wed, Jan 1, 2014 at 9:36 PM, Sid Boyce <sboyce at blueyonder.co.uk> wrote:
> On 01/01/14 18:46, Ilia Mirkin wrote:
>>
>> On Wed, Jan 1, 2014 at 9:04 AM, Sid Boyce <sboyce at blueyonder.co.uk> wrote:
>>>
>>> On 01/01/14 00:55, Ilia Mirkin wrote:
>>>>
>>>> On Tue, Dec 31, 2013 at 7:41 PM, Sid Boyce <sboyce at blueyonder.co.uk>
>>>> wrote:
>>>>>
>>>>> On 31/12/13 10:36, Ilia Mirkin wrote:
>>>>>>
>>>>>> Having a dmesg would be nice. One thing I can think of off-hand is
>>>>>> that 3.13-rc has MSI turned on by default. You can turn it off by
>>>>>> adding "nouveau.config=NvMSI=0" to your kernel cmdline. If that
>>>>>> doesn't help, a bisect restricted to drivers/gpu/drm/nouveau should
>>>>>> show the offending commit fairly quickly.
>>>>>>
>>>>>>      -ilia
>>>>>>
>>>>> Adding "nouveau.config=NvMSI=0" to the command line fixed the problem.
>>>>> So it looks like commit 049ffa8ab33a63b3bff672d1a0ee6a35ad253fe8
>>>>> introduced
>>>>> it.
>>>>
>>>> Any chance you might mmiotrace the blob (version 325 or later) to see
>>>> which registers it fiddles with? Or alternatively, if you have a NVCE
>>>> card (you never did end up providing the logs which would have made
>>>> that apparent), could you try replacing nvc3_mc_oclass with
>>>> nvc0_mc_oclass for the 0xce case in
>>>> drivers/gpu/drm/nouveau/core/engine/device/nvc0.c? (and boot without
>>>> the MSI disabling.) The switch has already been made for NVC8 in
>>>> 0bae1d61c75 -- perhaps there are more "odd" ones.
>>>>
>>>>     -ilia
>>>>
>>> Fails exactly the same.
>>>          case 0xc3:
>>>                  device->cname = "GF106";
>>>
>>>                  device->oclass[NVDEV_SUBDEV_VBIOS  ] =
>>> &nouveau_bios_oclass;
>>>                  device->oclass[NVDEV_SUBDEV_GPIO   ] =
>>> &nv50_gpio_oclass;
>>>                  device->oclass[NVDEV_SUBDEV_I2C    ] = &nv94_i2c_oclass;
>>>                  device->oclass[NVDEV_SUBDEV_CLOCK  ] =
>>> &nvc0_clock_oclass;
>>>                  device->oclass[NVDEV_SUBDEV_THERM  ] =
>>> &nva3_therm_oclass;
>>>                  device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
>>>                  device->oclass[NVDEV_SUBDEV_DEVINIT] =
>>> &nvc0_devinit_oclass;
>>>                  device->oclass[NVDEV_SUBDEV_MC     ] = nvc0_mc_oclass;
>>> <<<<<====
>>>                  device->oclass[NVDEV_SUBDEV_BUS    ] = nvc0_bus_oclass;
>>>                  device->oclass[NVDEV_SUBDEV_TIMER  ] =
>>> &nv04_timer_oclass;
>>
>> That's the 0xc3 case... you have a nvce card, not nvc3 -- you would
>> need to change the NVDEV_SUBDEV_MC line to nvc0_mc_oclass for the 0xce
>> case.
>>
>>> The dmesg and Xorg.0.log with the problem captured across a ssh link.
>>>
>>> # ps fax|grep X
>>>   5633 pts/0    S+     0:00          \_ grep --color=auto X
>>>   5160 tty7     Ss+    0:08  \_ /usr/bin/Xorg -br :0 vt7 -nolisten tcp
>>> -auth
>>> /var/lib/kdm/AuthFiles/A:0-yqspza
>>>
>>> Also
>>> # echo mmiotrace > /sys/kernel/debug/tracing/current_tracer
>>> -bash: echo: write error: Invalid argument.
>>
>> Take a look at https://wiki.ubuntu.com/X/MMIOTracing
>>
>>    -ilia
>>
> Of course it's a GF114.
> Made the change and it boots without the command line change.

Great! Care to send a patch?

  -ilia