On 2020-11-30 at 17:28:32 +0200, Imre Deak wrote:
> On Mon, Nov 30, 2020 at 02:46:46PM +0530, Anshuman Gupta wrote:
> > At usual case DC3CO exit happen automatically by DMC f/w whenever
> > PSR2 clears idle. This happens smoothly by DMC f/w to work with flips.
> > But there are certain scenario where DC3CO Disallowed by driver
> > asynchronous with flips. In such scenario display engine could
> > be already in DC3CO state and driver has disallowed it,
> > It initiates DC3CO exit sequence in DMC f/w which requires a
> > dc3co exit delay of 200us in driver.
> > It requires to protect intel_pipe_update_{update_end} with
> > dc3co exit delay.
> >
> > Cc: Imre Deak <imre.deak@xxxxxxxxx>
> > Cc: <stable@xxxxxxxxxxxxxxx>
> > Signed-off-by: Anshuman Gupta <anshuman.gupta@xxxxxxxxx>
>
> To make sure that it doesn't hide the root cause (or affects unrelated
> platforms), I'd only add locking around DC3co changes with a new lock,
> using lock/unlock helpers in intel_display_power.c called from
> intel_pipe_update_start/end.
>
> Also please submit this patch separately, w/o the optimization in patch
> 1/2, so we know that this change fixes the problem.
This patch doesn't seems to fix the issue.
Looks like there is some other set of display register updates before
completing the dc3co exit delay beyond intel_pipe_update_start/end causing this issue.
Thanks,
Anshuman Gupta.
>
> --Imre
>
> > ---
> > drivers/gpu/drm/i915/display/intel_display.c | 3 +++
> > 1 file changed, 3 insertions(+)
> >
> > diff --git a/drivers/gpu/drm/i915/display/intel_display.c b/drivers/gpu/drm/i915/display/intel_display.c
> > index ba26545392bc..3b81b98c0daf 100644
> > --- a/drivers/gpu/drm/i915/display/intel_display.c
> > +++ b/drivers/gpu/drm/i915/display/intel_display.c
> > @@ -15924,6 +15924,8 @@ static void intel_update_crtc(struct intel_atomic_state *state,
> > else
> > intel_fbc_enable(state, crtc);
> >
> > + /* Protect intel_pipe_update_{start,end} with power_domians lock */
> > + mutex_lock(&dev_priv->power_domains.lock);
> > /* Perform vblank evasion around commit operation */
> > intel_pipe_update_start(new_crtc_state);
> >
> > @@ -15935,6 +15937,7 @@ static void intel_update_crtc(struct intel_atomic_state *state,
> > i9xx_update_planes_on_crtc(state, crtc);
> >
> > intel_pipe_update_end(new_crtc_state);
> > + mutex_unlock(&dev_prive->power_domains.lock);
> >
> > /*
> > * We usually enable FIFO underrun interrupts as part of the
> > --
> > 2.26.2
> >
