diff mbox series

[v3,10/11] drm/i915/display: Check if pipe fastset is allowed by external dependencies

Message ID 20191216220742.34332-10-jose.souza@intel.com (mailing list archive)
State New, archived
Headers show
Series [v3,01/11] drm: Add __drm_atomic_helper_crtc_state_reset() & co. | expand

Commit Message

Souza, Jose Dec. 16, 2019, 10:07 p.m. UTC 
Check if fastset is allowed by external dependencies like other pipes
and transcoders.

Right now this patch only forces a fullmodeset in MST slaves of MST
masters that needs a fullmodeset but it will be needed for port sync
as well.

v3:
- moved handling to intel_atomic_check() this way is guarantee that
all pipes will have its state computed

Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Manasi Navare <manasi.d.navare@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
---
 drivers/gpu/drm/i915/display/intel_display.c | 63 ++++++++++++++++----
 1 file changed, 50 insertions(+), 13 deletions(-)

Comments

Ville Syrjälä Dec. 18, 2019, 5 p.m. UTC | #1 
On Mon, Dec 16, 2019 at 02:07:41PM -0800, José Roberto de Souza wrote:
> Check if fastset is allowed by external dependencies like other pipes
> and transcoders.
> 
> Right now this patch only forces a fullmodeset in MST slaves of MST
> masters that needs a fullmodeset but it will be needed for port sync
> as well.
> 
> v3:
> - moved handling to intel_atomic_check() this way is guarantee that
> all pipes will have its state computed
> 
> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
> Cc: Lucas De Marchi <lucas.demarchi@intel.com>
> Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> Cc: Manasi Navare <manasi.d.navare@intel.com>
> Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
> ---
>  drivers/gpu/drm/i915/display/intel_display.c | 63 ++++++++++++++++----
>  1 file changed, 50 insertions(+), 13 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/display/intel_display.c b/drivers/gpu/drm/i915/display/intel_display.c
> index f1c0687bf69b..2627a7bcb45c 100644
> --- a/drivers/gpu/drm/i915/display/intel_display.c
> +++ b/drivers/gpu/drm/i915/display/intel_display.c
> @@ -13941,19 +13941,6 @@ static void intel_crtc_check_fastset(const struct intel_crtc_state *old_crtc_sta
>  
>  	new_crtc_state->uapi.mode_changed = false;
>  	new_crtc_state->update_pipe = true;
> -
> -	/*
> -	 * If we're not doing the full modeset we want to
> -	 * keep the current M/N values as they may be
> -	 * sufficiently different to the computed values
> -	 * to cause problems.
> -	 *
> -	 * FIXME: should really copy more fuzzy state here
> -	 */
> -	new_crtc_state->fdi_m_n = old_crtc_state->fdi_m_n;
> -	new_crtc_state->dp_m_n = old_crtc_state->dp_m_n;
> -	new_crtc_state->dp_m2_n2 = old_crtc_state->dp_m2_n2;
> -	new_crtc_state->has_drrs = old_crtc_state->has_drrs;
>  }
>  
>  static int intel_crtc_add_planes_to_state(struct intel_atomic_state *state,
> @@ -14121,6 +14108,56 @@ static int intel_atomic_check(struct drm_device *dev,
>  			any_ms = true;
>  	}
>  
> +	/**
> +	 * Check if fastset is allowed by external dependencies like other
> +	 * pipes and transcoders.
> +	 *
> +	 * Right now it only forces a fullmodeset when the MST master
> +	 * transcoder did not changed but the pipe of the master transcoder
> +	 * needs a fullmodeset so all slaves also needs to do a fullmodeset.
> +	 */
> +	for_each_new_intel_crtc_in_state(state, crtc, new_crtc_state, i) {
> +		struct intel_crtc_state *new_crtc_state_iter;
> +		struct intel_crtc *crtc_iter;
> +		int j;
> +
> +		if (!intel_dp_mst_is_slave_trans(new_crtc_state) ||
> +		    needs_modeset(new_crtc_state))

shouldn't this be
if (!mst_master || !needs_modeset)
	continue;
?

> +			continue;
> +
> +		for_each_new_intel_crtc_in_state(state, crtc_iter,
> +						 new_crtc_state_iter, j) {
> +			if (new_crtc_state->mst_master_transcoder !=
> +			    new_crtc_state_iter->cpu_transcoder)
> +				continue;
> +
> +			if (needs_modeset(new_crtc_state_iter)) {
> +				new_crtc_state->uapi.mode_changed = true;
> +				new_crtc_state->update_pipe = false;
> +			}
> +			break;

Hmm. So I guess the break here is because we're going to keep iterating
the outer loop anyway? Don't think that works with the s/slave/master/
fix above.

I suggest moving this inner loop into a separate function entirely.
Could simply pass in the master_transcoder to that function and then
we wouldn't even have this clash with variable names for the two
crtcs/crtc_states.

Somethig like (with better funciton names ofc :)

mst_modeset_slaves(state, master_transcoder)
{
	for_each() {
		if (master_transcoder == crtc_state->master_transcoder)
			modeset=1;
	}
}

atomic_check()
{
	for_each()
		compute();

	for_each() {
		if (master && needs_modeset()
			mst_modeset_slaves(state, crtc_state->cpu_transcoder);
	}

	for_each() {
		if (fastset)
			copy();
	}
}

Or do we also need to force a modeset for the master if a slave needs a
modeset? I'm trying to think if there's a way that we could end up with
a slave changing its master (thus needing modeset) without the master
being flagged for modeset...

> +		}
> +	}
> +
> +	for_each_oldnew_intel_crtc_in_state(state, crtc, old_crtc_state,
> +					    new_crtc_state, i) {
> +		if (needs_modeset(new_crtc_state))
> +			continue;
> +
> +		/*
> +		 * If we're not doing the full modeset we want to
> +		 * keep the current M/N values as they may be
> +		 * sufficiently different to the computed values
> +		 * to cause problems.
> +		 *
> +		 * FIXME: should really copy more fuzzy state here
> +		 */
> +		new_crtc_state->fdi_m_n = old_crtc_state->fdi_m_n;
> +		new_crtc_state->dp_m_n = old_crtc_state->dp_m_n;
> +		new_crtc_state->dp_m2_n2 = old_crtc_state->dp_m2_n2;
> +		new_crtc_state->has_drrs = old_crtc_state->has_drrs;

A separate function for this fastset state copy might also be nice so we
don't have to see these mundane details in this higher level code.

> +	}
> +
>  	if (any_ms && !check_digital_port_conflicts(state)) {
>  		DRM_DEBUG_KMS("rejecting conflicting digital port configuration\n");
>  		ret = EINVAL;
> -- 
> 2.24.1
Ville Syrjälä Dec. 18, 2019, 5:10 p.m. UTC | #2 
On Wed, Dec 18, 2019 at 07:00:19PM +0200, Ville Syrjälä wrote:
> On Mon, Dec 16, 2019 at 02:07:41PM -0800, José Roberto de Souza wrote:
> > Check if fastset is allowed by external dependencies like other pipes
> > and transcoders.
> > 
> > Right now this patch only forces a fullmodeset in MST slaves of MST
> > masters that needs a fullmodeset but it will be needed for port sync
> > as well.
> > 
> > v3:
> > - moved handling to intel_atomic_check() this way is guarantee that
> > all pipes will have its state computed
> > 
> > Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
> > Cc: Lucas De Marchi <lucas.demarchi@intel.com>
> > Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> > Cc: Manasi Navare <manasi.d.navare@intel.com>
> > Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
> > ---
> >  drivers/gpu/drm/i915/display/intel_display.c | 63 ++++++++++++++++----
> >  1 file changed, 50 insertions(+), 13 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/i915/display/intel_display.c b/drivers/gpu/drm/i915/display/intel_display.c
> > index f1c0687bf69b..2627a7bcb45c 100644
> > --- a/drivers/gpu/drm/i915/display/intel_display.c
> > +++ b/drivers/gpu/drm/i915/display/intel_display.c
> > @@ -13941,19 +13941,6 @@ static void intel_crtc_check_fastset(const struct intel_crtc_state *old_crtc_sta
> >  
> >  	new_crtc_state->uapi.mode_changed = false;
> >  	new_crtc_state->update_pipe = true;
> > -
> > -	/*
> > -	 * If we're not doing the full modeset we want to
> > -	 * keep the current M/N values as they may be
> > -	 * sufficiently different to the computed values
> > -	 * to cause problems.
> > -	 *
> > -	 * FIXME: should really copy more fuzzy state here
> > -	 */
> > -	new_crtc_state->fdi_m_n = old_crtc_state->fdi_m_n;
> > -	new_crtc_state->dp_m_n = old_crtc_state->dp_m_n;
> > -	new_crtc_state->dp_m2_n2 = old_crtc_state->dp_m2_n2;
> > -	new_crtc_state->has_drrs = old_crtc_state->has_drrs;
> >  }
> >  
> >  static int intel_crtc_add_planes_to_state(struct intel_atomic_state *state,
> > @@ -14121,6 +14108,56 @@ static int intel_atomic_check(struct drm_device *dev,
> >  			any_ms = true;
> >  	}
> >  
> > +	/**
> > +	 * Check if fastset is allowed by external dependencies like other
> > +	 * pipes and transcoders.
> > +	 *
> > +	 * Right now it only forces a fullmodeset when the MST master
> > +	 * transcoder did not changed but the pipe of the master transcoder
> > +	 * needs a fullmodeset so all slaves also needs to do a fullmodeset.
> > +	 */
> > +	for_each_new_intel_crtc_in_state(state, crtc, new_crtc_state, i) {
> > +		struct intel_crtc_state *new_crtc_state_iter;
> > +		struct intel_crtc *crtc_iter;
> > +		int j;
> > +
> > +		if (!intel_dp_mst_is_slave_trans(new_crtc_state) ||
> > +		    needs_modeset(new_crtc_state))
> 
> shouldn't this be
> if (!mst_master || !needs_modeset)
> 	continue;
> ?
> 
> > +			continue;
> > +
> > +		for_each_new_intel_crtc_in_state(state, crtc_iter,
> > +						 new_crtc_state_iter, j) {
> > +			if (new_crtc_state->mst_master_transcoder !=
> > +			    new_crtc_state_iter->cpu_transcoder)
> > +				continue;
> > +
> > +			if (needs_modeset(new_crtc_state_iter)) {
> > +				new_crtc_state->uapi.mode_changed = true;
> > +				new_crtc_state->update_pipe = false;
> > +			}
> > +			break;
> 
> Hmm. So I guess the break here is because we're going to keep iterating
> the outer loop anyway? Don't think that works with the s/slave/master/
> fix above.
> 
> I suggest moving this inner loop into a separate function entirely.
> Could simply pass in the master_transcoder to that function and then
> we wouldn't even have this clash with variable names for the two
> crtcs/crtc_states.
> 
> Somethig like (with better funciton names ofc :)
> 
> mst_modeset_slaves(state, master_transcoder)
> {
> 	for_each() {
> 		if (master_transcoder == crtc_state->master_transcoder)
> 			modeset=1;
> 	}
> }
> 
> atomic_check()
> {
> 	for_each()
> 		compute();
> 
> 	for_each() {
> 		if (master && needs_modeset()
> 			mst_modeset_slaves(state, crtc_state->cpu_transcoder);
> 	}
> 
> 	for_each() {
> 		if (fastset)
> 			copy();
> 	}
> }
> 
> Or do we also need to force a modeset for the master if a slave needs a
> modeset? I'm trying to think if there's a way that we could end up with
> a slave changing its master (thus needing modeset) without the master
> being flagged for modeset...

How about this:

1. Enable pipe B (master) + pipe C (slave)
2. Enable pipe A (should become master)

Hmm. Scratch that. Pipe B will have it's master_transcoder change from
INVALID to A so it'll get flagged for a modeset. Seems I need a more
heavy duty thinking cap for this one...
Souza, Jose Dec. 18, 2019, 5:38 p.m. UTC | #3 
On Wed, 2019-12-18 at 19:10 +0200, Ville Syrjälä wrote:
> On Wed, Dec 18, 2019 at 07:00:19PM +0200, Ville Syrjälä wrote:
> > On Mon, Dec 16, 2019 at 02:07:41PM -0800, José Roberto de Souza
> > wrote:
> > > Check if fastset is allowed by external dependencies like other
> > > pipes
> > > and transcoders.
> > > 
> > > Right now this patch only forces a fullmodeset in MST slaves of
> > > MST
> > > masters that needs a fullmodeset but it will be needed for port
> > > sync
> > > as well.
> > > 
> > > v3:
> > > - moved handling to intel_atomic_check() this way is guarantee
> > > that
> > > all pipes will have its state computed
> > > 
> > > Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
> > > Cc: Lucas De Marchi <lucas.demarchi@intel.com>
> > > Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> > > Cc: Manasi Navare <manasi.d.navare@intel.com>
> > > Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
> > > ---
> > >  drivers/gpu/drm/i915/display/intel_display.c | 63
> > > ++++++++++++++++----
> > >  1 file changed, 50 insertions(+), 13 deletions(-)
> > > 
> > > diff --git a/drivers/gpu/drm/i915/display/intel_display.c
> > > b/drivers/gpu/drm/i915/display/intel_display.c
> > > index f1c0687bf69b..2627a7bcb45c 100644
> > > --- a/drivers/gpu/drm/i915/display/intel_display.c
> > > +++ b/drivers/gpu/drm/i915/display/intel_display.c
> > > @@ -13941,19 +13941,6 @@ static void
> > > intel_crtc_check_fastset(const struct intel_crtc_state
> > > *old_crtc_sta
> > >  
> > >  	new_crtc_state->uapi.mode_changed = false;
> > >  	new_crtc_state->update_pipe = true;
> > > -
> > > -	/*
> > > -	 * If we're not doing the full modeset we want to
> > > -	 * keep the current M/N values as they may be
> > > -	 * sufficiently different to the computed values
> > > -	 * to cause problems.
> > > -	 *
> > > -	 * FIXME: should really copy more fuzzy state here
> > > -	 */
> > > -	new_crtc_state->fdi_m_n = old_crtc_state->fdi_m_n;
> > > -	new_crtc_state->dp_m_n = old_crtc_state->dp_m_n;
> > > -	new_crtc_state->dp_m2_n2 = old_crtc_state->dp_m2_n2;
> > > -	new_crtc_state->has_drrs = old_crtc_state->has_drrs;
> > >  }
> > >  
> > >  static int intel_crtc_add_planes_to_state(struct
> > > intel_atomic_state *state,
> > > @@ -14121,6 +14108,56 @@ static int intel_atomic_check(struct
> > > drm_device *dev,
> > >  			any_ms = true;
> > >  	}
> > >  
> > > +	/**
> > > +	 * Check if fastset is allowed by external dependencies like
> > > other
> > > +	 * pipes and transcoders.
> > > +	 *
> > > +	 * Right now it only forces a fullmodeset when the MST master
> > > +	 * transcoder did not changed but the pipe of the master
> > > transcoder
> > > +	 * needs a fullmodeset so all slaves also needs to do a
> > > fullmodeset.
> > > +	 */
> > > +	for_each_new_intel_crtc_in_state(state, crtc, new_crtc_state,
> > > i) {
> > > +		struct intel_crtc_state *new_crtc_state_iter;
> > > +		struct intel_crtc *crtc_iter;
> > > +		int j;
> > > +
> > > +		if (!intel_dp_mst_is_slave_trans(new_crtc_state) ||
> > > +		    needs_modeset(new_crtc_state))
> > 
> > shouldn't this be
> > if (!mst_master || !needs_modeset)
> > 	continue;

When iterating over MST pipes in older gens
intel_dp_mst_is_master_trans() and intel_dp_mst_is_slave_trans() will
always return false and if this one is intel_dp_mst_is_master_trans()
we can skip it.
In case it is a intel_dp_mst_is_slave_trans() and if it needs a modeset
already we don't need to force anything.


> > ?
> > 
> > > +			continue;
> > > +
> > > +		for_each_new_intel_crtc_in_state(state, crtc_iter,
> > > +						 new_crtc_state_iter,
> > > j) {
> > > +			if (new_crtc_state->mst_master_transcoder !=
> > > +			    new_crtc_state_iter->cpu_transcoder)
> > > +				continue;
> > > +
> > > +			if (needs_modeset(new_crtc_state_iter)) {
> > > +				new_crtc_state->uapi.mode_changed =
> > > true;
> > > +				new_crtc_state->update_pipe = false;
> > > +			}
> > > +			break;
> > 
> > Hmm. So I guess the break here is because we're going to keep
> > iterating
> > the outer loop anyway? Don't think that works with the
> > s/slave/master/
> > fix above.

This inner loop is to find the master of the pipe being check, when we
find it we can break and process the next pipe.

> > 
> > I suggest moving this inner loop into a separate function entirely.
> > Could simply pass in the master_transcoder to that function and
> > then
> > we wouldn't even have this clash with variable names for the two
> > crtcs/crtc_states.
> > 
> > Somethig like (with better funciton names ofc :)
> > 
> > mst_modeset_slaves(state, master_transcoder)
> > {
> > 	for_each() {
> > 		if (master_transcoder == crtc_state->master_transcoder)
> > 			modeset=1;
> > 	}
> > }

Sounds good the function above.

> > 
> > atomic_check()
> > {
> > 	for_each()
> > 		compute();
> > 
> > 	for_each() {
> > 		if (master && needs_modeset()
> > 			mst_modeset_slaves(state, crtc_state-
> > >cpu_transcoder);
> > 	}
> > 
> > 	for_each() {
> > 		if (fastset)
> > 			copy();
> > 	}
> > }
> > 
> > Or do we also need to force a modeset for the master if a slave
> > needs a
> > modeset? I'm trying to think if there's a way that we could end up
> > with
> > a slave changing its master (thus needing modeset) without the
> > master
> > being flagged for modeset...
> 
> How about this:
> 
> 1. Enable pipe B (master) + pipe C (slave)
> 2. Enable pipe A (should become master)
> 
> Hmm. Scratch that. Pipe B will have it's master_transcoder change
> from
> INVALID to A so it'll get flagged for a modeset. Seems I need a more
> heavy duty thinking cap for this one...
>
diff mbox series

Patch

diff --git a/drivers/gpu/drm/i915/display/intel_display.c b/drivers/gpu/drm/i915/display/intel_display.c
index f1c0687bf69b..2627a7bcb45c 100644
--- a/drivers/gpu/drm/i915/display/intel_display.c
+++ b/drivers/gpu/drm/i915/display/intel_display.c
@@ -13941,19 +13941,6 @@  static void intel_crtc_check_fastset(const struct intel_crtc_state *old_crtc_sta
 
 	new_crtc_state->uapi.mode_changed = false;
 	new_crtc_state->update_pipe = true;
-
-	/*
-	 * If we're not doing the full modeset we want to
-	 * keep the current M/N values as they may be
-	 * sufficiently different to the computed values
-	 * to cause problems.
-	 *
-	 * FIXME: should really copy more fuzzy state here
-	 */
-	new_crtc_state->fdi_m_n = old_crtc_state->fdi_m_n;
-	new_crtc_state->dp_m_n = old_crtc_state->dp_m_n;
-	new_crtc_state->dp_m2_n2 = old_crtc_state->dp_m2_n2;
-	new_crtc_state->has_drrs = old_crtc_state->has_drrs;
 }
 
 static int intel_crtc_add_planes_to_state(struct intel_atomic_state *state,
@@ -14121,6 +14108,56 @@  static int intel_atomic_check(struct drm_device *dev,
 			any_ms = true;
 	}
 
+	/**
+	 * Check if fastset is allowed by external dependencies like other
+	 * pipes and transcoders.
+	 *
+	 * Right now it only forces a fullmodeset when the MST master
+	 * transcoder did not changed but the pipe of the master transcoder
+	 * needs a fullmodeset so all slaves also needs to do a fullmodeset.
+	 */
+	for_each_new_intel_crtc_in_state(state, crtc, new_crtc_state, i) {
+		struct intel_crtc_state *new_crtc_state_iter;
+		struct intel_crtc *crtc_iter;
+		int j;
+
+		if (!intel_dp_mst_is_slave_trans(new_crtc_state) ||
+		    needs_modeset(new_crtc_state))
+			continue;
+
+		for_each_new_intel_crtc_in_state(state, crtc_iter,
+						 new_crtc_state_iter, j) {
+			if (new_crtc_state->mst_master_transcoder !=
+			    new_crtc_state_iter->cpu_transcoder)
+				continue;
+
+			if (needs_modeset(new_crtc_state_iter)) {
+				new_crtc_state->uapi.mode_changed = true;
+				new_crtc_state->update_pipe = false;
+			}
+			break;
+		}
+	}
+
+	for_each_oldnew_intel_crtc_in_state(state, crtc, old_crtc_state,
+					    new_crtc_state, i) {
+		if (needs_modeset(new_crtc_state))
+			continue;
+
+		/*
+		 * If we're not doing the full modeset we want to
+		 * keep the current M/N values as they may be
+		 * sufficiently different to the computed values
+		 * to cause problems.
+		 *
+		 * FIXME: should really copy more fuzzy state here
+		 */
+		new_crtc_state->fdi_m_n = old_crtc_state->fdi_m_n;
+		new_crtc_state->dp_m_n = old_crtc_state->dp_m_n;
+		new_crtc_state->dp_m2_n2 = old_crtc_state->dp_m2_n2;
+		new_crtc_state->has_drrs = old_crtc_state->has_drrs;
+	}
+
 	if (any_ms && !check_digital_port_conflicts(state)) {
 		DRM_DEBUG_KMS("rejecting conflicting digital port configuration\n");
 		ret = EINVAL;