Re: [PATCH v1 3/3] msm: skip the atomic commit of self refresh while PSR running

[Dmitry Baryshkov <dmitry.baryshkov@linaro.org>]

On 03/04/2023 19:11, Dmitry Baryshkov wrote:
> On Mon, 3 Apr 2023 at 15:01, Vinod Polimera <vpolimer@qti.qualcomm.com> wrote:
>>
>>> On Fri, 31 Mar 2023 at 16:59, Vinod Polimera <quic_vpolimer@quicinc.com>
>>> wrote:
>>>>
>>>> In certain CPU stress conditions, there can be a delay in scheduling commit
>>>> work and it was observed that PSR commit from a different work queue
>>> was
>>>> scheduled. Avoid these commits as display is already in PSR mode.
>>>>
>>>> Signed-off-by: Vinod Polimera <quic_vpolimer@quicinc.com>
>>>> ---
>>>>   drivers/gpu/drm/msm/msm_atomic.c | 3 +++
>>>>   1 file changed, 3 insertions(+)
>>>>
>>>> diff --git a/drivers/gpu/drm/msm/msm_atomic.c
>>> b/drivers/gpu/drm/msm/msm_atomic.c
>>>> index 645fe53..f8141bb 100644
>>>> --- a/drivers/gpu/drm/msm/msm_atomic.c
>>>> +++ b/drivers/gpu/drm/msm/msm_atomic.c
>>>> @@ -192,6 +192,9 @@ int msm_atomic_check(struct drm_device *dev,
>>> struct drm_atomic_state *state)
>>>>                          new_crtc_state->mode_changed = true;
>>>>                          state->allow_modeset = true;
>>>>                  }
>>>> +
>>>> +               if (old_crtc_state->self_refresh_active && new_crtc_state-
>>>> self_refresh_active)
>>>> +                       return -EINVAL;
>>>
>>> EINVAL here means that atomic_check will fail if both old and new
>>> states are in SR mode. For example, there might be a mode set for
>>> another CRTC (while keeping this one in SR mode). I don't think this
>>> is correct. We should skip/shortcut the commit, that's true. But I
>>> doubt that returning an error here is a proper way to do this. Please
>>> correct me if I'm wrong.
>>
>> If there is a modeset on same crtc with a different connector. The new_crtc_state will not have self_refresh_active set.
>> Self_refresh_active is set from the helper library, which will duplicate the old_state and just adds self_refresh_active to true and active to false.
>> so we can be confident that if we are checking for self_refresh_active status then it should be coming from the library call.
>>
>> Also the EINVAL is returned to the self_refresh library API and the function will be retired.
> 
> Maybe I misunderstand you here. However, in this way EINVAL is
> returned to drm_atomic_check_only() and not to the SR code.

Unless anybody objects, I'm going to drop this patch now. The issue 
should be solved in the framework itself.

> 
>> And self_refresh_active is cleared on every commit : https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/tree/drivers/gpu/drm/drm_atomic_state_helper.c#n158
> 
> And this means that this check will not trigger at all, if I'm not
> mistaken. You've added code to msm_atomic_check(), so
> drm_self_refresh_helper_alter_state() was not called (yet) and thus
> new_crtc_state->self_refresh_active is set to false, fresh after
> crtc's duplicate_state.
> 
> --
> With best wishes
> Dmitry

-- 
With best wishes
Dmitry