[PATCH] drm/amdgpu: Add sysfs file for VBIOS

[Deucher, Alexander]

> -----Original Message-----
> From: amd-gfx [mailto:amd-gfx-bounces at lists.freedesktop.org] On Behalf
> Of Deucher, Alexander
> Sent: Friday, August 25, 2017 4:07 PM
> To: Kuehling, Felix; amd-gfx at lists.freedesktop.org
> Subject: RE: [PATCH] drm/amdgpu: Add sysfs file for VBIOS
> 
> > -----Original Message-----
> > From: Kuehling, Felix
> > Sent: Friday, August 25, 2017 3:59 PM
> > To: Deucher, Alexander; amd-gfx at lists.freedesktop.org
> > Subject: Re: [PATCH] drm/amdgpu: Add sysfs file for VBIOS
> >
> > On 2017-08-25 03:40 PM, Deucher, Alexander wrote:
> > >> -----Original Message-----
> > >> From: amd-gfx [mailto:amd-gfx-bounces at lists.freedesktop.org] On
> > Behalf
> > >> Of Felix Kuehling
> > >> Sent: Friday, August 25, 2017 3:34 PM
> > >> To: amd-gfx at lists.freedesktop.org
> > >> Subject: Re: [PATCH] drm/amdgpu: Add sysfs file for VBIOS
> > >>
> > >> I think the power measurement is a bit of a hack right now. It requires
> > >> pinging the SMU twice to start and stop the measurement, and waiting
> > for
> > >> some time (but not too long) in between. I think if we want to expose
> > >> this via hwmon, we'll need to have some kind of timer that polls it in
> > >> regular intervals in the background and updates a value that can be
> > >> queried through HWMon without delay.
> > > I don't know that hwmon has any requirements with respect to delay.  I
> > think a slight delay in reading it back through hwmon is preferable to a
> > background thread polling it. Regular polling may also have negative
> impacts
> > on performance or stability, depending on how it was validated.
> >
> > If an application is reading hwmon data from 4 GPUs and each query for
> > power usage from each GPU takes one second, that would seriously limit
> > the update frequency of the application.
> >
> > Maybe the first read could block long enough to get useful data. But
> > subsequent reads could probably avoid blocking if we keep track of the
> > time stamp of the last query. If it was very recently, we can return the
> > last value. If it was long enough ago, we can get an updated value from
> > the SMU. If the last query was too long ago, we start over and block for
> > a second.
> 
> Could we just set the sampling period down?  If the application is polling for
> instantaneous power we'd probably want a short smu sampling period
> anyway.  Otherwise, the application shouldn’t be polling so frequently in the
> first place.

Another alternative would be to expose the SMU interface directly.  So getting the power would be something like:

echo 1 > power
#wait
echo 0 > power
cat power

or:
echo 2 > power # seconds to sample
cat power #returns -EBUSY until the sampling period is done

Alex

> 
> Alex
> 
> >
> > Regards,
> >   Felix
> >
> > > Alex
> > >
> > >> Regards,
> > >>   Felix
> > >>
> > >>
> > >> On 2017-08-25 11:56 AM, Deucher, Alexander wrote:
> > >>>> -----Original Message-----
> > >>>> From: amd-gfx [mailto:amd-gfx-bounces at lists.freedesktop.org] On
> > >> Behalf
> > >>>> Of Russell, Kent
> > >>>> Sent: Friday, August 25, 2017 9:00 AM
> > >>>> To: StDenis, Tom; Christian König; amd-gfx at lists.freedesktop.org
> > >>>> Subject: RE: [PATCH] drm/amdgpu: Add sysfs file for VBIOS
> > >>>>
> > >>>> There is GPU Power usage reported through amdgpu_pm_info,
> which
> > >> also
> > >>>> has some other information as well. I'd like that in sysfs, but I am
> > unsure if
> > >>>> we are  allowed to due to the other information reported there.
> > >>> For power and voltage, I believe there are standard hwmon interfaces.
> > It
> > >> would probably be best to expose it via those.  For clocks/pcie, I think
> you
> > >> can already determine them via the existing pp sysfs interfaces.
> > >>> Alex
> > >>>
> > >>>>  Kent
> > >>>>
> > >>>> -----Original Message-----
> > >>>> From: StDenis, Tom
> > >>>> Sent: Friday, August 25, 2017 8:58 AM
> > >>>> To: Christian König; Russell, Kent; amd-gfx at lists.freedesktop.org
> > >>>> Subject: Re: [PATCH] drm/amdgpu: Add sysfs file for VBIOS
> > >>>>
> > >>>> On 25/08/17 08:56 AM, Christian König wrote:
> > >>>>> Hi Kent,
> > >>>>>
> > >>>>> agree on the VBIOS dump file, that clearly belongs to debugsf.
> > >>>>>
> > >>>>> The power usage stuff I can't say much about cause I'm not deeply
> > into
> > >>>>> this, but keep in mind the restriction for sysfs:
> > >>>>> 1. It's a stable interface. So it must be very well designed.
> > >>>>> 2. Only one value per file. I think the power stuff doesn't fulfill
> > >>>>> that requirement at the moment.
> > >>>> What "power" stuff are we talking about?  The sensors interface or
> the
> > >>>> pm_info or something else?
> > >>>>
> > >>>> Keep in mind umr uses the sensors debugfs file in --top mode.
> > >>>>
> > >>>> Tom
> > >>>> _______________________________________________
> > >>>> amd-gfx mailing list
> > >>>> amd-gfx at lists.freedesktop.org
> > >>>> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
> > >>> _______________________________________________
> > >>> amd-gfx mailing list
> > >>> amd-gfx at lists.freedesktop.org
> > >>> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
> > >> _______________________________________________
> > >> amd-gfx mailing list
> > >> amd-gfx at lists.freedesktop.org
> > >> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
> 
> _______________________________________________
> amd-gfx mailing list
> amd-gfx at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx