DRM driver and runtime suspend-resume handling?

DRM driver and runtime suspend-resume handling?

[Jyri Sarha]

Hi,
While working with CRTC color related properties (gamma and CTM for
instance) and making them persistent over suspend-resume cycle it
occurred to me if I am just wasting resources by storing the property
values in the driver and restoring them in dev_pm_ops runtime_resume()..

Wouldn't it work if I would just:

1. Add a flag in the driver to indicate that the context may have been
lost since the previous atomic commit and set in runtime_resume().

2. And write the color properties to HW if the context lost flag is set
even if the drm_crtc_state color_mgmt_changed is false.

The color property values are there despite the color_mgmt_changed ==
false, aren't they?

Best regards,
Jyri

-- 
Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: DRM driver and runtime suspend-resume handling?

[Daniel Vetter]

On Mon, Jan 13, 2020 at 01:03:11PM +0200, Jyri Sarha wrote:
> Hi,
> While working with CRTC color related properties (gamma and CTM for
> instance) and making them persistent over suspend-resume cycle it
> occurred to me if I am just wasting resources by storing the property
> values in the driver and restoring them in dev_pm_ops runtime_resume()..
> 
> Wouldn't it work if I would just:
> 
> 1. Add a flag in the driver to indicate that the context may have been
> lost since the previous atomic commit and set in runtime_resume().
> 
> 2. And write the color properties to HW if the context lost flag is set
> even if the drm_crtc_state color_mgmt_changed is false.

Still feels a bit too complicated, but might be needed for your hw. The
usual approach is:

- runtime pm within modeset enable/disable. Since atomic helpers always
  enable the entire pipeline for a crtc enable/disable you can put the
  runtime pm into each component (drm_crtc/encoder/bridge/...).

- just unconditionally restore everything in modeset enable, assuming
  everything got lost.

- drm_atomic_helper_suspend/resume for system suspend/resume.

And you should be covevered. State save/restore in your driver code is
indeed an anti-pattern for modeset drivers, don't do that - ime at least
with fragile hw you'll get divergence between the two paths in minor
details, with some really hard to track down bugs as a result.

Ime tracking state changes and trying to be clever with when to restore
stuff (maybe outside of some atomic flip fastpath) in driver code only
attracts bugs :-)

> The color property values are there despite the color_mgmt_changed ==
> false, aren't they?

Yes.

Cheers, Daniel
> 
> Best regards,
> Jyri
> 
> -- 
> Texas Instruments Finland Oy, Porkkalankatu 22, 00180 Helsinki.
> Y-tunnus/Business ID: 0615521-4. Kotipaikka/Domicile: Helsinki
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel