[RFC] drm/i915/dp: PPS registers doesn't require AUX power

Commit Message

Gupta, Anshuman Nov. 24, 2020, 9:58 a.m. UTC

Platforms with South Display Engine on PCH, doesn't
require to get/put the AUX power domain in order to
access PPS register because PPS registers are always on
with South display on PCH.

Cc: Imre Deak <[e-mail]>
Cc: <[e-mail]>
Signed-off-by: Anshuman Gupta <[e-mail]>
---
 drivers/gpu/drm/i915/display/intel_dp.c | 13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)

Comments

Imre Deak Nov. 24, 2020, 4:44 p.m. UTC | #1

On Tue, Nov 24, 2020 at 03:28:47PM +0530, Anshuman Gupta wrote:
> Platforms with South Display Engine on PCH, doesn't
> require to get/put the AUX power domain in order to
> access PPS register because PPS registers are always on
> with South display on PCH.
> 
> Cc: Imre Deak <[e-mail]>
> Cc: <[e-mail]>
> Signed-off-by: Anshuman Gupta <[e-mail]>

Could you describe the issue the patch is fixing?

For accessing PPS registers the AUX power well may not be needed, but
I'm not sure if this also applies to PPS functionality in general. For
instance forcing VDD is required for AUX functionality.

In any case we do need a power reference for any register access, so I
don't think not getting any power reference for PPS is ok.

--Imre

> ---
>  drivers/gpu/drm/i915/display/intel_dp.c | 13 ++++++++-----
>  1 file changed, 8 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/display/intel_dp.c b/drivers/gpu/drm/i915/display/intel_dp.c
> index 3896d08c4177..84a2c49e154c 100644
> --- a/drivers/gpu/drm/i915/display/intel_dp.c
> +++ b/drivers/gpu/drm/i915/display/intel_dp.c
> @@ -872,8 +872,9 @@ pps_lock(struct intel_dp *intel_dp)
>  	 * See intel_power_sequencer_reset() why we need
>  	 * a power domain reference here.
>  	 */
> -	wakeref = intel_display_power_get(dev_priv,
> -					  intel_aux_power_domain(dp_to_dig_port(intel_dp)));
> +	if (!HAS_PCH_SPLIT(dev_priv))
> +		wakeref = intel_display_power_get(dev_priv,
> +						  intel_aux_power_domain(dp_to_dig_port(intel_dp)));
>  
>  	mutex_lock(&dev_priv->pps_mutex);
>  
> @@ -886,9 +887,11 @@ pps_unlock(struct intel_dp *intel_dp, intel_wakeref_t wakeref)
>  	struct drm_i915_private *dev_priv = dp_to_i915(intel_dp);
>  
>  	mutex_unlock(&dev_priv->pps_mutex);
> -	intel_display_power_put(dev_priv,
> -				intel_aux_power_domain(dp_to_dig_port(intel_dp)),
> -				wakeref);
> +
> +	if (!HAS_PCH_SPLIT(dev_priv))
> +		intel_display_power_put(dev_priv,
> +					intel_aux_power_domain(dp_to_dig_port(intel_dp)),
> +					wakeref);
>  	return 0;
>  }
>  
> -- 
> 2.26.2
>

Gupta, Anshuman Nov. 25, 2020, 7:46 a.m. UTC | #2

On 2020-11-24 at 18:44:06 +0200, Imre Deak wrote:
> On Tue, Nov 24, 2020 at 03:28:47PM +0530, Anshuman Gupta wrote:
> > Platforms with South Display Engine on PCH, doesn't
> > require to get/put the AUX power domain in order to
> > access PPS register because PPS registers are always on
> > with South display on PCH.
> > 
> > Cc: Imre Deak <[e-mail]>
> > Cc: <[e-mail]>
> > Signed-off-by: Anshuman Gupta <[e-mail]>
> 
> Could you describe the issue the patch is fixing?
This fixes the display glitches causes by race between brightness update thread 
and flip thread.
while brightness is being updated it reads pp_ctrl reg to check whether backlight
is enabled and get/put the AUX power domain, this enables and disable DC Off 
power well(DC3CO) back and forth.
IMO there are two work item for above race needed to be addressed.
1. Don't get AUX power for PPS register access (this patch addressed this).
2. skl_program_plane() should wait for DC3CO exit delay to avoid any race with
   DC3CO disable sequence. (WIP)      
> 
> For accessing PPS registers the AUX power well may not be needed, but
> I'm not sure if this also applies to PPS functionality in general. For
> instance forcing VDD is required for AUX functionality.
AFAIU edp_panel_vdd_on explicitly get AUX power in order to force the VDD.
> 
> In any case we do need a power reference for any register access, so I
> don't think not getting any power reference for PPS is ok.
IMO if PPS register lies in PCH(South Display), it is not correct to take
any power domain which are associated with north display power wells.

This patch is inspired from the comment in pps_lock, quoting that
"See intel_power_sequencer_reset() why we need a power domain reference here."

intel_power_sequencer_reset is not being called for platforms with split PCH,
stating that PPS registers are always on.
https://patchwork.freedesktop.org/patch/259077/ ((v4: (James Ausmus)))

Could you please provide your opinion to use intel_runtime_pm_get()
before accessing PPS register in order to get a wakeref.

Thanks,
Anshuman Gupta.
> 
> --Imre
> 
> > ---
> >  drivers/gpu/drm/i915/display/intel_dp.c | 13 ++++++++-----
> >  1 file changed, 8 insertions(+), 5 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/i915/display/intel_dp.c b/drivers/gpu/drm/i915/display/intel_dp.c
> > index 3896d08c4177..84a2c49e154c 100644
> > --- a/drivers/gpu/drm/i915/display/intel_dp.c
> > +++ b/drivers/gpu/drm/i915/display/intel_dp.c
> > @@ -872,8 +872,9 @@ pps_lock(struct intel_dp *intel_dp)
> >  	 * See intel_power_sequencer_reset() why we need
> >  	 * a power domain reference here.
> >  	 */
> > -	wakeref = intel_display_power_get(dev_priv,
> > -					  intel_aux_power_domain(dp_to_dig_port(intel_dp)));
> > +	if (!HAS_PCH_SPLIT(dev_priv))
> > +		wakeref = intel_display_power_get(dev_priv,
> > +						  intel_aux_power_domain(dp_to_dig_port(intel_dp)));
> >  
> >  	mutex_lock(&dev_priv->pps_mutex);
> >  
> > @@ -886,9 +887,11 @@ pps_unlock(struct intel_dp *intel_dp, intel_wakeref_t wakeref)
> >  	struct drm_i915_private *dev_priv = dp_to_i915(intel_dp);
> >  
> >  	mutex_unlock(&dev_priv->pps_mutex);
> > -	intel_display_power_put(dev_priv,
> > -				intel_aux_power_domain(dp_to_dig_port(intel_dp)),
> > -				wakeref);
> > +
> > +	if (!HAS_PCH_SPLIT(dev_priv))
> > +		intel_display_power_put(dev_priv,
> > +					intel_aux_power_domain(dp_to_dig_port(intel_dp)),
> > +					wakeref);
> >  	return 0;
> >  }
> >  
> > -- 
> > 2.26.2
> >

Imre Deak Nov. 25, 2020, 4:24 p.m. UTC | #3

+Ville.

On Wed, Nov 25, 2020 at 01:16:27PM +0530, Anshuman Gupta wrote:
> On 2020-11-24 at 18:44:06 +0200, Imre Deak wrote:
> > On Tue, Nov 24, 2020 at 03:28:47PM +0530, Anshuman Gupta wrote:
> > > Platforms with South Display Engine on PCH, doesn't
> > > require to get/put the AUX power domain in order to
> > > access PPS register because PPS registers are always on
> > > with South display on PCH.
> > > 
> > > Cc: Imre Deak <[e-mail]>
> > > Cc: <[e-mail]>
> > > Signed-off-by: Anshuman Gupta <[e-mail]>
> > 
> > Could you describe the issue the patch is fixing?
>
> This fixes the display glitches causes by race between brightness
> update thread and flip thread.

Flips should work even with asynchronous DC3co (or any DC state)
disabling, at least according to the spec the HW handles this. Only
modesetting and AUX transfers have restriction wrt. DC state handling
(where DC states need to get disabled).

I think the exact restriction needs to be clarified with HW people: Is
only the DC3co disable -> flip or also the opposite sequence
problematic? Is it only DC3co or also DC5/6 affected?

> While brightness is being updated it reads pp_ctrl reg to check
> whether backlight is enabled and get/put the AUX power domain, this
> enables and disable DC Off power well(DC3CO) back and forth.
>
> IMO there are two work item for above race needed to be addressed.
> 1. Don't get AUX power for PPS register access (this patch addressed this).
> 2. skl_program_plane() should wait for DC3CO exit delay to avoid any race with
>    DC3CO disable sequence. (WIP)      

DC states can be disabled asynchronously with a flip modeset, not only
for panel brightness setting, but also AUX transfers for instance. So I
think we'd need to add locking against DC state changes to
intel_pipe_update_start()/end(). Probably the easiest would be to use
the power_domains->lock for this.

--Imre

Gupta, Anshuman Nov. 26, 2020, 9:39 a.m. UTC | #4

On 2020-11-25 at 18:24:44 +0200, Imre Deak wrote:
> +Ville.
Hi Ville ,
Let me provide you some context over the issue which requires your input.
TGL on chorome OS has observed some display glitches when brightness is being updated
at very fast rate. This has surfaced out two issue.
1. Getting the AUX power when accessing the PPS registers on platform with split PCH.
2. The race between DC3CO disabling delay and flips. (B.Spec says 200us dc3co exit delay)
   I will send a separate RFC patch to fix this issue.

Current patch is addressing issue1, 
IMHO it is unnecessary to take AUX power for pps register read for checking
whether backlight was enabled. This is causing flip to race with
DC3CO exit delay.
Could you please provide your input to this . 

Thanks,
Anshuman Gupta.   
> 
> On Wed, Nov 25, 2020 at 01:16:27PM +0530, Anshuman Gupta wrote:
> > On 2020-11-24 at 18:44:06 +0200, Imre Deak wrote:
> > > On Tue, Nov 24, 2020 at 03:28:47PM +0530, Anshuman Gupta wrote:
> > > > Platforms with South Display Engine on PCH, doesn't
> > > > require to get/put the AUX power domain in order to
> > > > access PPS register because PPS registers are always on
> > > > with South display on PCH.
> > > > 
> > > > Cc: Imre Deak <[e-mail]>
> > > > Cc: <[e-mail]>
> > > > Signed-off-by: Anshuman Gupta <[e-mail]>
> > > 
> > > Could you describe the issue the patch is fixing?
> >
> > This fixes the display glitches causes by race between brightness
> > update thread and flip thread.
> 
> Flips should work even with asynchronous DC3co (or any DC state)
> disabling, at least according to the spec the HW handles this. Only
> modesetting and AUX transfers have restriction wrt. DC state handling
> (where DC states need to get disabled).
> 
> I think the exact restriction needs to be clarified with HW people: Is
> only the DC3co disable -> flip or also the opposite sequence
> problematic? Is it only DC3co or also DC5/6 affected?
> 
> > While brightness is being updated it reads pp_ctrl reg to check
> > whether backlight is enabled and get/put the AUX power domain, this
> > enables and disable DC Off power well(DC3CO) back and forth.
> >
> > IMO there are two work item for above race needed to be addressed.
> > 1. Don't get AUX power for PPS register access (this patch addressed this).
> > 2. skl_program_plane() should wait for DC3CO exit delay to avoid any race with
> >    DC3CO disable sequence. (WIP)      
> 
> DC states can be disabled asynchronously with a flip modeset, not only
> for panel brightness setting, but also AUX transfers for instance. So I
> think we'd need to add locking against DC state changes to
> intel_pipe_update_start()/end(). Probably the easiest would be to use
> the power_domains->lock for this.
> 
> --Imre

Ville Syrjälä Nov. 26, 2020, 1:52 p.m. UTC | #5

On Thu, Nov 26, 2020 at 03:09:50PM +0530, Anshuman Gupta wrote:
> On 2020-11-25 at 18:24:44 +0200, Imre Deak wrote:
> > +Ville.
> Hi Ville ,
> Let me provide you some context over the issue which requires your input.
> TGL on chorome OS has observed some display glitches when brightness is being updated
> at very fast rate. This has surfaced out two issue.
> 1. Getting the AUX power when accessing the PPS registers on platform with split PCH.

There can be all kinds of reasons for taking the AUX power
domain. If that somehow causes display glitches then someone
needs to figure out why and fix it. This looks like just duct
tape over one specific case.

> 2. The race between DC3CO disabling delay and flips. (B.Spec says 200us dc3co exit delay)
>    I will send a separate RFC patch to fix this issue.
> 
> Current patch is addressing issue1, 
> IMHO it is unnecessary to take AUX power for pps register read for checking
> whether backlight was enabled. This is causing flip to race with
> DC3CO exit delay.
> Could you please provide your input to this . 
> 
> Thanks,
> Anshuman Gupta.   
> > 
> > On Wed, Nov 25, 2020 at 01:16:27PM +0530, Anshuman Gupta wrote:
> > > On 2020-11-24 at 18:44:06 +0200, Imre Deak wrote:
> > > > On Tue, Nov 24, 2020 at 03:28:47PM +0530, Anshuman Gupta wrote:
> > > > > Platforms with South Display Engine on PCH, doesn't
> > > > > require to get/put the AUX power domain in order to
> > > > > access PPS register because PPS registers are always on
> > > > > with South display on PCH.
> > > > > 
> > > > > Cc: Imre Deak <[e-mail]>
> > > > > Cc: <[e-mail]>
> > > > > Signed-off-by: Anshuman Gupta <[e-mail]>
> > > > 
> > > > Could you describe the issue the patch is fixing?
> > >
> > > This fixes the display glitches causes by race between brightness
> > > update thread and flip thread.
> > 
> > Flips should work even with asynchronous DC3co (or any DC state)
> > disabling, at least according to the spec the HW handles this. Only
> > modesetting and AUX transfers have restriction wrt. DC state handling
> > (where DC states need to get disabled).
> > 
> > I think the exact restriction needs to be clarified with HW people: Is
> > only the DC3co disable -> flip or also the opposite sequence
> > problematic? Is it only DC3co or also DC5/6 affected?
> > 
> > > While brightness is being updated it reads pp_ctrl reg to check
> > > whether backlight is enabled and get/put the AUX power domain, this
> > > enables and disable DC Off power well(DC3CO) back and forth.
> > >
> > > IMO there are two work item for above race needed to be addressed.
> > > 1. Don't get AUX power for PPS register access (this patch addressed this).
> > > 2. skl_program_plane() should wait for DC3CO exit delay to avoid any race with
> > >    DC3CO disable sequence. (WIP)      
> > 
> > DC states can be disabled asynchronously with a flip modeset, not only
> > for panel brightness setting, but also AUX transfers for instance. So I
> > think we'd need to add locking against DC state changes to
> > intel_pipe_update_start()/end(). Probably the easiest would be to use
> > the power_domains->lock for this.
> > 
> > --Imre

Patch

diff --git a/drivers/gpu/drm/i915/display/intel_dp.c b/drivers/gpu/drm/i915/display/intel_dp.c
index 3896d08c4177..84a2c49e154c 100644
--- a/drivers/gpu/drm/i915/display/intel_dp.c
+++ b/drivers/gpu/drm/i915/display/intel_dp.c
@@ -872,8 +872,9 @@  pps_lock(struct intel_dp *intel_dp)
 	 * See intel_power_sequencer_reset() why we need
 	 * a power domain reference here.
 	 */
-	wakeref = intel_display_power_get(dev_priv,
-					  intel_aux_power_domain(dp_to_dig_port(intel_dp)));
+	if (!HAS_PCH_SPLIT(dev_priv))
+		wakeref = intel_display_power_get(dev_priv,
+						  intel_aux_power_domain(dp_to_dig_port(intel_dp)));
 
 	mutex_lock(&dev_priv->pps_mutex);
 
@@ -886,9 +887,11 @@  pps_unlock(struct intel_dp *intel_dp, intel_wakeref_t wakeref)
 	struct drm_i915_private *dev_priv = dp_to_i915(intel_dp);
 
 	mutex_unlock(&dev_priv->pps_mutex);
-	intel_display_power_put(dev_priv,
-				intel_aux_power_domain(dp_to_dig_port(intel_dp)),
-				wakeref);
+
+	if (!HAS_PCH_SPLIT(dev_priv))
+		intel_display_power_put(dev_priv,
+					intel_aux_power_domain(dp_to_dig_port(intel_dp)),
+					wakeref);
 	return 0;
 }