[RFC 4/4] drm/i915: Expose RPS thresholds in sysfs
Rodrigo Vivi
rodrigo.vivi at kernel.org
Fri May 19 21:00:40 UTC 2023
On Fri, May 19, 2023 at 07:36:56PM +0000, Prahlad Kilambi wrote:
> > One question is are we able to find a "one size fits all" values.
>
> > However regardless of that, given we already expose frequency controls in sysfs
> > with the same reasoning of allowing system owners explicit control if so wanted,
> > I think exposing the thresholds can be equally justified.
>
> Exposing these RPS thresholds via sysfs allows for dynamic tuning of these values at runtime. Common scenarios where we can benefit from variable frequency ramping include plugged in vs battery where differing thresholds allow to weight more for performance vs power. Data from testing on ChromeOS Gen12 platforms where GuC isn't enabled indicates gains > 10% across several games. See https://gitlab.freedesktop.org/drm/intel/-/issues/8389#note_1890428 for details.
>
In general we should always try to reduce the knobs and specially with a register
that doesn't work with the new platforms with FW on control of all these variations.
But this is a compelling argument.
Acked-by: Rodrigo Vivi <rodrigo.vivi at intel.com>
(if patch 3 doesn't break compilation and the other chunk has the placement explained
this is a rv-b for the series)
More information about the dri-devel
mailing list