Nouveau - Jan 2018 - [PATCH] drm/nouveau/therm/gp100: Do not report temperature when subdev is shadowed

Not sure if i understand completely what you intend to say here, with 
this we prevent hwmon from reporting utterly wrong temperature values 
returning an error (we could return -EBUSY or somehting instead, 
granted), yet if the device is shadowed, getting a sane temp value out 
of is seems unlikely to me!

Greetings,

Tobias

On 1/26/18 12:40 PM, Karol Herbst wrote:
> no, we can't do that. We actually have to prevent this from hwom. The
> issue here is, that the reg read returns 0xffffffff and parsing that
> is the first step in the first place.
>
> On Thu, Jan 25, 2018 at 7:16 PM, Tobias Klausmann
> <tobias.johannes.klausmann at mni.thm.de> wrote:
>> This fixes wrong temperature outputs e.g. 511°C if the card is asleep.
>>
>> Signed-off-by: Tobias Klausmann <tobias.johannes.klausmann at
mni.thm.de>
>> ---
>>   drivers/gpu/drm/nouveau/nvkm/subdev/therm/gp100.c | 4 +++-
>>   1 file changed, 3 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/therm/gp100.c
b/drivers/gpu/drm/nouveau/nvkm/subdev/therm/gp100.c
>> index 9f0dea3f61dc..45d0ec632b5a 100644
>> --- a/drivers/gpu/drm/nouveau/nvkm/subdev/therm/gp100.c
>> +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/therm/gp100.c
>> @@ -32,8 +32,10 @@ gp100_temp_get(struct nvkm_therm *therm)
>>          u32 inttemp = (tsensor & 0x0001fff8);
>>
>>          /* device SHADOWed */
>> -       if (tsensor & 0x40000000)
>> +       if (tsensor & 0x40000000) {
>>                  nvkm_trace(subdev, "reading temperature from
SHADOWed sensor\n");
>> +               return -ENODEV;
>> +       }
>>
>>          /* device valid */
>>          if (tsensor & 0x20000000)
>> --
>> 2.16.1
>>
>> _______________________________________________
>> Nouveau mailing list
>> Nouveau at lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/nouveau

well I just tried to say, that you are not fixing the issue you think
were fixing. In your case the GPU is powered off and you get garbage
values from any mmio read, so parsing those values is just wrong and
we need to prevent doing anything on the hw whenever it is powered off
directly in hwmon.

On Fri, Jan 26, 2018 at 2:40 PM, Tobias Klausmann
<tobias.johannes.klausmann at mni.thm.de> wrote:
> Not sure if i understand completely what you intend to say here, with this
> we prevent hwmon from reporting utterly wrong temperature values returning
> an error (we could return -EBUSY or somehting instead, granted), yet if the
> device is shadowed, getting a sane temp value out of is seems unlikely to
> me!
>
> Greetings,
>
> Tobias
>
>
> On 1/26/18 12:40 PM, Karol Herbst wrote:
>>
>> no, we can't do that. We actually have to prevent this from hwom.
The
>> issue here is, that the reg read returns 0xffffffff and parsing that
>> is the first step in the first place.
>>
>> On Thu, Jan 25, 2018 at 7:16 PM, Tobias Klausmann
>> <tobias.johannes.klausmann at mni.thm.de> wrote:
>>>
>>> This fixes wrong temperature outputs e.g. 511°C if the card is
asleep.
>>>
>>> Signed-off-by: Tobias Klausmann <tobias.johannes.klausmann at
mni.thm.de>
>>> ---
>>>   drivers/gpu/drm/nouveau/nvkm/subdev/therm/gp100.c | 4 +++-
>>>   1 file changed, 3 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/therm/gp100.c
>>> b/drivers/gpu/drm/nouveau/nvkm/subdev/therm/gp100.c
>>> index 9f0dea3f61dc..45d0ec632b5a 100644
>>> --- a/drivers/gpu/drm/nouveau/nvkm/subdev/therm/gp100.c
>>> +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/therm/gp100.c
>>> @@ -32,8 +32,10 @@ gp100_temp_get(struct nvkm_therm *therm)
>>>          u32 inttemp = (tsensor & 0x0001fff8);
>>>
>>>          /* device SHADOWed */
>>> -       if (tsensor & 0x40000000)
>>> +       if (tsensor & 0x40000000) {
>>>                  nvkm_trace(subdev, "reading temperature from
SHADOWed
>>> sensor\n");
>>> +               return -ENODEV;
>>> +       }
>>>
>>>          /* device valid */
>>>          if (tsensor & 0x20000000)
>>> --
>>> 2.16.1
>>>
>>> _______________________________________________
>>> Nouveau mailing list
>>> Nouveau at lists.freedesktop.org
>>> https://lists.freedesktop.org/mailman/listinfo/nouveau

Well fixing the return of wrong values in this function is reasonable by 
any means, of course not reading the mem in the first place would be 
nice, but deciding this is imho not in the scope of a temp_get function 
but somewhere in the code calling temp_get.

On 1/26/18 3:03 PM, Karol Herbst wrote:
> well I just tried to say, that you are not fixing the issue you think
> were fixing. In your case the GPU is powered off and you get garbage
> values from any mmio read, so parsing those values is just wrong and
> we need to prevent doing anything on the hw whenever it is powered off
> directly in hwmon.
>
> On Fri, Jan 26, 2018 at 2:40 PM, Tobias Klausmann
> <tobias.johannes.klausmann at mni.thm.de> wrote:
>> Not sure if i understand completely what you intend to say here, with
this
>> we prevent hwmon from reporting utterly wrong temperature values
returning
>> an error (we could return -EBUSY or somehting instead, granted), yet if
the
>> device is shadowed, getting a sane temp value out of is seems unlikely
to
>> me!
>>
>> Greetings,
>>
>> Tobias
>>
>>
>> On 1/26/18 12:40 PM, Karol Herbst wrote:
>>> no, we can't do that. We actually have to prevent this from
hwom. The
>>> issue here is, that the reg read returns 0xffffffff and parsing
that
>>> is the first step in the first place.
>>>
>>> On Thu, Jan 25, 2018 at 7:16 PM, Tobias Klausmann
>>> <tobias.johannes.klausmann at mni.thm.de> wrote:
>>>> This fixes wrong temperature outputs e.g. 511°C if the card is
asleep.
>>>>
>>>> Signed-off-by: Tobias Klausmann <tobias.johannes.klausmann
at mni.thm.de>
>>>> ---
>>>>    drivers/gpu/drm/nouveau/nvkm/subdev/therm/gp100.c | 4 +++-
>>>>    1 file changed, 3 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/therm/gp100.c
>>>> b/drivers/gpu/drm/nouveau/nvkm/subdev/therm/gp100.c
>>>> index 9f0dea3f61dc..45d0ec632b5a 100644
>>>> --- a/drivers/gpu/drm/nouveau/nvkm/subdev/therm/gp100.c
>>>> +++ b/drivers/gpu/drm/nouveau/nvkm/subdev/therm/gp100.c
>>>> @@ -32,8 +32,10 @@ gp100_temp_get(struct nvkm_therm *therm)
>>>>           u32 inttemp = (tsensor & 0x0001fff8);
>>>>
>>>>           /* device SHADOWed */
>>>> -       if (tsensor & 0x40000000)
>>>> +       if (tsensor & 0x40000000) {
>>>>                   nvkm_trace(subdev, "reading temperature
from SHADOWed
>>>> sensor\n");
>>>> +               return -ENODEV;
>>>> +       }
>>>>
>>>>           /* device valid */
>>>>           if (tsensor & 0x20000000)
>>>> --
>>>> 2.16.1
>>>>
>>>> _______________________________________________
>>>> Nouveau mailing list
>>>> Nouveau at lists.freedesktop.org
>>>> https://lists.freedesktop.org/mailman/listinfo/nouveau