[Nouveau] [PATCH 0/3] drm/gk20a: support for reclocking

[Alexandre Courbot]

On Fri, Jul 11, 2014 at 10:38 AM, Alexandre Courbot <acourbot at nvidia.com> wrote:
> Hi Ben,
>
>
> On 07/11/2014 10:07 AM, Ben Skeggs wrote:
>>
>> On Thu, Jul 10, 2014 at 5:34 PM, Alexandre Courbot <acourbot at nvidia.com>
>> wrote:
>>>
>>> This series adds support for reclocking on GK20A. The first two patches
>>> touch
>>> the clock subsystem to allow GK20A to operate, by making the presence of
>>> the
>>> thermal and voltage devices optional, and allowing pstates to be provided
>>> directly instead of being probed using the BIOS (which Tegra does not
>>> have).
>>
>> Hey Alex,
>>
>> I mentioned a while back I had some stuff in-progress to make
>> implementing this a bit cleaner for you, but as you can probably tell,
>> I didn't get to it yet.  It's likely I won't manage to before the next
>> merge window either.  So, I'll likely take these patches as-is
>> (assuming no objections on reviews here) and rebase my stuff on top.
>
>
> Thanks. You will probably notice that these patches won't apply to your tree
> and require some tweaking. I will probably end up sending a v2 anyway, so
> maybe you should wait for it. If you want to try this version anyway, I have
> fixed-up patches against your tree.
>
> Note that your tree currently won't build against -next because it uses
> drm_sysfs_connector_add/remove which are not available anymore (replaced by
> drm_connector_register/unregister I believe).
>
> Oh and while I'm at it, there seems to be a typo in line 131 of clock.h,
> which should read _nouveau_clock_fini and not _nouveau_clock_init.
>
>
>>>
>>> The last patch adds the GK20A clock device. Arguably the clock can be
>>> seen as a
>>> stripped-down version of what is seen on NVE0, however instead of using
>>> NVE0
>>> support has been written from scratch using the ChromeOS kernel as a
>>> basis.
>>> There are several reasons for this:
>>>
>>> - The ChromeOS driver uses a lookup table for the P coefficient which I
>>> could
>>>    not find in the NVE0 driver,
>>
>> Interesting.  Can you give more details on how "PL" works exactly,
>> we'd been operating on the assumption (mainly inherited from code that
>> appeared in xf86-video-nv) that it was always a straight division.
>
>
> The pl_to_div table in clock/gk20a.c should give the right coefficients, but
> I have seen contradictory information in our docs. Let me ask the right
> people so we can get to the bottom of this.

So as it turns out this seems to be gk20a-specific. Other Kepler GPUs
use it as a straight division, but gk20a uses a narrower range and
thus requires this table.