drm/i915: Postpone plane readout until after encoder readout, v2.
diff mbox

Message ID 55C1E95C.6030101@linux.intel.com
State New
Headers show

Commit Message

Maarten Lankhorst Aug. 5, 2015, 10:45 a.m. UTC
From: Patrik Jakobsson <patrik.jakobsson@linux.intel.com>

When reading out hw state for planes we disable inactive planes which in
turn triggers an update of the watermarks. The update depends on the
crtc_clock being set which is done when reading out encoders. Thus
postpone the plane readout until after encoder readout.

This prevents a warning in skl_compute_linetime_wm() where pixel_rate
becomes 0 when crtc_clock is 0.

Changes since v1:
- Set all modes after all state is read out. (Maarten)

References: https://bugs.freedesktop.org/show_bug.cgi?id=91428
Signed-off-by: Patrik Jakobsson <patrik.jakobsson@linux.intel.com>
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---

Comments

Daniel Vetter Aug. 26, 2015, 2:43 p.m. UTC | #1
On Wed, Aug 05, 2015 at 12:45:48PM +0200, Maarten Lankhorst wrote:
> From: Patrik Jakobsson <patrik.jakobsson@linux.intel.com>
> 
> When reading out hw state for planes we disable inactive planes which in
> turn triggers an update of the watermarks. The update depends on the
> crtc_clock being set which is done when reading out encoders. Thus
> postpone the plane readout until after encoder readout.
> 
> This prevents a warning in skl_compute_linetime_wm() where pixel_rate
> becomes 0 when crtc_clock is 0.
> 
> Changes since v1:
> - Set all modes after all state is read out. (Maarten)
> 
> References: https://bugs.freedesktop.org/show_bug.cgi?id=91428
> Signed-off-by: Patrik Jakobsson <patrik.jakobsson@linux.intel.com>
> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>

Hm I still think we have a bit a mess here - the plane readout code is
still spread out between modeset_readout_hw_state and the per-plane loop
in intel_modeset_init after setup_hw_state.

I thought that we only ever need to do the plane state readout on initial
load (they're all off on resume anyway and we force a full modeset to make
sure plane state is correct again). Can't we just have a setup_plane_state
which has that loop from the end of modeset_init with all the other plane
state unified there?
-Daniel

> ---
> diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
> index df237ad4a780..85d52158d476 100644
> --- a/drivers/gpu/drm/i915/intel_display.c
> +++ b/drivers/gpu/drm/i915/intel_display.c
> @@ -15042,27 +15042,25 @@ static bool primary_get_hw_state(struct intel_crtc *crtc)
>  	return !!(I915_READ(DSPCNTR(crtc->plane)) & DISPLAY_PLANE_ENABLE);
>  }
>  
> -static void readout_plane_state(struct intel_crtc *crtc,
> -				struct intel_crtc_state *crtc_state)
> +static void intel_sanitize_plane(struct intel_plane *plane)
>  {
> -	struct intel_plane *p;
>  	struct intel_plane_state *plane_state;
> -	bool active = crtc_state->base.active;
> +	struct intel_crtc *crtc;
>  
> -	for_each_intel_plane(crtc->base.dev, p) {
> -		if (crtc->pipe != p->pipe)
> -			continue;
> +	plane_state = to_intel_plane_state(plane->base.state);
>  
> -		plane_state = to_intel_plane_state(p->base.state);
> +	if (!plane_state->base.crtc)
> +		return;
>  
> -		if (p->base.type == DRM_PLANE_TYPE_PRIMARY)
> -			plane_state->visible = primary_get_hw_state(crtc);
> -		else {
> -			if (active)
> -				p->disable_plane(&p->base, &crtc->base);
> +	crtc = to_intel_crtc(plane_state->base.crtc);
>  
> -			plane_state->visible = false;
> -		}
> +	if (plane->base.type == DRM_PLANE_TYPE_PRIMARY) {
> +		plane_state->visible = primary_get_hw_state(crtc);
> +	} else {
> +		if (crtc->base.state->active)
> +			plane->disable_plane(&plane->base, &crtc->base);
> +
> +		plane_state->visible = false;
>  	}
>  }
>  
> @@ -15086,35 +15084,6 @@ static void intel_modeset_readout_hw_state(struct drm_device *dev)
>  		crtc->base.state->active = crtc->active;
>  		crtc->base.enabled = crtc->active;
>  
> -		memset(&crtc->base.mode, 0, sizeof(crtc->base.mode));
> -		if (crtc->base.state->active) {
> -			intel_mode_from_pipe_config(&crtc->base.mode, crtc->config);
> -			intel_mode_from_pipe_config(&crtc->base.state->adjusted_mode, crtc->config);
> -			WARN_ON(drm_atomic_set_mode_for_crtc(crtc->base.state, &crtc->base.mode));
> -
> -			/*
> -			 * The initial mode needs to be set in order to keep
> -			 * the atomic core happy. It wants a valid mode if the
> -			 * crtc's enabled, so we do the above call.
> -			 *
> -			 * At this point some state updated by the connectors
> -			 * in their ->detect() callback has not run yet, so
> -			 * no recalculation can be done yet.
> -			 *
> -			 * Even if we could do a recalculation and modeset
> -			 * right now it would cause a double modeset if
> -			 * fbdev or userspace chooses a different initial mode.
> -			 *
> -			 * If that happens, someone indicated they wanted a
> -			 * mode change, which means it's safe to do a full
> -			 * recalculation.
> -			 */
> -			crtc->base.state->mode.private_flags = I915_MODE_FLAG_INHERITED;
> -		}
> -
> -		crtc->base.hwmode = crtc->config->base.adjusted_mode;
> -		readout_plane_state(crtc, to_intel_crtc_state(crtc->base.state));
> -
>  		DRM_DEBUG_KMS("[CRTC:%d] hw state readout: %s\n",
>  			      crtc->base.base.id,
>  			      crtc->active ? "enabled" : "disabled");
> @@ -15184,6 +15153,7 @@ intel_modeset_setup_hw_state(struct drm_device *dev)
>  	enum pipe pipe;
>  	struct intel_crtc *crtc;
>  	struct intel_encoder *encoder;
> +	struct intel_plane *plane;
>  	int i;
>  
>  	intel_modeset_readout_hw_state(dev);
> @@ -15195,6 +15165,40 @@ intel_modeset_setup_hw_state(struct drm_device *dev)
>  
>  	for_each_pipe(dev_priv, pipe) {
>  		crtc = to_intel_crtc(dev_priv->pipe_to_crtc_mapping[pipe]);
> +		crtc->base.hwmode = crtc->config->base.adjusted_mode;
> +		memset(&crtc->base.mode, 0, sizeof(crtc->base.mode));
> +		if (crtc->base.state->active) {
> +			intel_mode_from_pipe_config(&crtc->base.mode, crtc->config);
> +			intel_mode_from_pipe_config(&crtc->base.state->adjusted_mode, crtc->config);
> +			WARN_ON(drm_atomic_set_mode_for_crtc(crtc->base.state, &crtc->base.mode));
> +
> +			/*
> +			 * The initial mode needs to be set in order to keep
> +			 * the atomic core happy. It wants a valid mode if the
> +			 * crtc's enabled, so we do the above call.
> +			 *
> +			 * At this point some state updated by the connectors
> +			 * in their ->detect() callback has not run yet, so
> +			 * no recalculation can be done yet.
> +			 *
> +			 * Even if we could do a recalculation and modeset
> +			 * right now it would cause a double modeset if
> +			 * fbdev or userspace chooses a different initial mode.
> +			 *
> +			 * If that happens, someone indicated they wanted a
> +			 * mode change, which means it's safe to do a full
> +			 * recalculation.
> +			 */
> +			crtc->base.state->mode.private_flags = I915_MODE_FLAG_INHERITED;
> +		}
> +
> +		for_each_intel_plane(crtc->base.dev, plane) {
> +			if (crtc->pipe != plane->pipe)
> +				continue;
> +
> +			intel_sanitize_plane(plane);
> +		}
> +
>  		intel_sanitize_crtc(crtc);
>  		intel_dump_pipe_config(crtc, crtc->config,
>  				       "[setup_hw_state]");
> 
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Maarten Lankhorst Aug. 27, 2015, 11:05 a.m. UTC | #2
Hey,

Op 26-08-15 om 16:43 schreef Daniel Vetter:
> On Wed, Aug 05, 2015 at 12:45:48PM +0200, Maarten Lankhorst wrote:
>> From: Patrik Jakobsson <patrik.jakobsson@linux.intel.com>
>>
>> When reading out hw state for planes we disable inactive planes which in
>> turn triggers an update of the watermarks. The update depends on the
>> crtc_clock being set which is done when reading out encoders. Thus
>> postpone the plane readout until after encoder readout.
>>
>> This prevents a warning in skl_compute_linetime_wm() where pixel_rate
>> becomes 0 when crtc_clock is 0.
>>
>> Changes since v1:
>> - Set all modes after all state is read out. (Maarten)
>>
>> References: https://bugs.freedesktop.org/show_bug.cgi?id=91428
>> Signed-off-by: Patrik Jakobsson <patrik.jakobsson@linux.intel.com>
>> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> Hm I still think we have a bit a mess here - the plane readout code is
> still spread out between modeset_readout_hw_state and the per-plane loop
> in intel_modeset_init after setup_hw_state.
>
> I thought that we only ever need to do the plane state readout on initial
> load (they're all off on resume anyway and we force a full modeset to make
> sure plane state is correct again). Can't we just have a setup_plane_state
> which has that loop from the end of modeset_init with all the other plane
> state unified there?
>
Perhaps, but intel_crtc_disable_noatomic cares about visibility state.

~Maarten
Daniel Vetter Sept. 1, 2015, 9:52 a.m. UTC | #3
On Thu, Aug 27, 2015 at 01:05:12PM +0200, Maarten Lankhorst wrote:
> Hey,
> 
> Op 26-08-15 om 16:43 schreef Daniel Vetter:
> > On Wed, Aug 05, 2015 at 12:45:48PM +0200, Maarten Lankhorst wrote:
> >> From: Patrik Jakobsson <patrik.jakobsson@linux.intel.com>
> >>
> >> When reading out hw state for planes we disable inactive planes which in
> >> turn triggers an update of the watermarks. The update depends on the
> >> crtc_clock being set which is done when reading out encoders. Thus
> >> postpone the plane readout until after encoder readout.
> >>
> >> This prevents a warning in skl_compute_linetime_wm() where pixel_rate
> >> becomes 0 when crtc_clock is 0.
> >>
> >> Changes since v1:
> >> - Set all modes after all state is read out. (Maarten)
> >>
> >> References: https://bugs.freedesktop.org/show_bug.cgi?id=91428
> >> Signed-off-by: Patrik Jakobsson <patrik.jakobsson@linux.intel.com>
> >> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> > Hm I still think we have a bit a mess here - the plane readout code is
> > still spread out between modeset_readout_hw_state and the per-plane loop
> > in intel_modeset_init after setup_hw_state.
> >
> > I thought that we only ever need to do the plane state readout on initial
> > load (they're all off on resume anyway and we force a full modeset to make
> > sure plane state is correct again). Can't we just have a setup_plane_state
> > which has that loop from the end of modeset_init with all the other plane
> > state unified there?
> >
> Perhaps, but intel_crtc_disable_noatomic cares about visibility state.

I don't see a need for that any more at all, since disable_noatomic is
only used in sanitize_crtc. And that's only called when the bios fucks
something up and i915.ko needs to patch it up. No way a pageflip or
anything else is pending there. Also we don't seem to set up plane_mask
early enough for that function to correctly kill the primary plane.

That might explain some of the display faults Chris is seeing ...
-Daniel

Patch
diff mbox

diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
index df237ad4a780..85d52158d476 100644
--- a/drivers/gpu/drm/i915/intel_display.c
+++ b/drivers/gpu/drm/i915/intel_display.c
@@ -15042,27 +15042,25 @@  static bool primary_get_hw_state(struct intel_crtc *crtc)
 	return !!(I915_READ(DSPCNTR(crtc->plane)) & DISPLAY_PLANE_ENABLE);
 }
 
-static void readout_plane_state(struct intel_crtc *crtc,
-				struct intel_crtc_state *crtc_state)
+static void intel_sanitize_plane(struct intel_plane *plane)
 {
-	struct intel_plane *p;
 	struct intel_plane_state *plane_state;
-	bool active = crtc_state->base.active;
+	struct intel_crtc *crtc;
 
-	for_each_intel_plane(crtc->base.dev, p) {
-		if (crtc->pipe != p->pipe)
-			continue;
+	plane_state = to_intel_plane_state(plane->base.state);
 
-		plane_state = to_intel_plane_state(p->base.state);
+	if (!plane_state->base.crtc)
+		return;
 
-		if (p->base.type == DRM_PLANE_TYPE_PRIMARY)
-			plane_state->visible = primary_get_hw_state(crtc);
-		else {
-			if (active)
-				p->disable_plane(&p->base, &crtc->base);
+	crtc = to_intel_crtc(plane_state->base.crtc);
 
-			plane_state->visible = false;
-		}
+	if (plane->base.type == DRM_PLANE_TYPE_PRIMARY) {
+		plane_state->visible = primary_get_hw_state(crtc);
+	} else {
+		if (crtc->base.state->active)
+			plane->disable_plane(&plane->base, &crtc->base);
+
+		plane_state->visible = false;
 	}
 }
 
@@ -15086,35 +15084,6 @@  static void intel_modeset_readout_hw_state(struct drm_device *dev)
 		crtc->base.state->active = crtc->active;
 		crtc->base.enabled = crtc->active;
 
-		memset(&crtc->base.mode, 0, sizeof(crtc->base.mode));
-		if (crtc->base.state->active) {
-			intel_mode_from_pipe_config(&crtc->base.mode, crtc->config);
-			intel_mode_from_pipe_config(&crtc->base.state->adjusted_mode, crtc->config);
-			WARN_ON(drm_atomic_set_mode_for_crtc(crtc->base.state, &crtc->base.mode));
-
-			/*
-			 * The initial mode needs to be set in order to keep
-			 * the atomic core happy. It wants a valid mode if the
-			 * crtc's enabled, so we do the above call.
-			 *
-			 * At this point some state updated by the connectors
-			 * in their ->detect() callback has not run yet, so
-			 * no recalculation can be done yet.
-			 *
-			 * Even if we could do a recalculation and modeset
-			 * right now it would cause a double modeset if
-			 * fbdev or userspace chooses a different initial mode.
-			 *
-			 * If that happens, someone indicated they wanted a
-			 * mode change, which means it's safe to do a full
-			 * recalculation.
-			 */
-			crtc->base.state->mode.private_flags = I915_MODE_FLAG_INHERITED;
-		}
-
-		crtc->base.hwmode = crtc->config->base.adjusted_mode;
-		readout_plane_state(crtc, to_intel_crtc_state(crtc->base.state));
-
 		DRM_DEBUG_KMS("[CRTC:%d] hw state readout: %s\n",
 			      crtc->base.base.id,
 			      crtc->active ? "enabled" : "disabled");
@@ -15184,6 +15153,7 @@  intel_modeset_setup_hw_state(struct drm_device *dev)
 	enum pipe pipe;
 	struct intel_crtc *crtc;
 	struct intel_encoder *encoder;
+	struct intel_plane *plane;
 	int i;
 
 	intel_modeset_readout_hw_state(dev);
@@ -15195,6 +15165,40 @@  intel_modeset_setup_hw_state(struct drm_device *dev)
 
 	for_each_pipe(dev_priv, pipe) {
 		crtc = to_intel_crtc(dev_priv->pipe_to_crtc_mapping[pipe]);
+		crtc->base.hwmode = crtc->config->base.adjusted_mode;
+		memset(&crtc->base.mode, 0, sizeof(crtc->base.mode));
+		if (crtc->base.state->active) {
+			intel_mode_from_pipe_config(&crtc->base.mode, crtc->config);
+			intel_mode_from_pipe_config(&crtc->base.state->adjusted_mode, crtc->config);
+			WARN_ON(drm_atomic_set_mode_for_crtc(crtc->base.state, &crtc->base.mode));
+
+			/*
+			 * The initial mode needs to be set in order to keep
+			 * the atomic core happy. It wants a valid mode if the
+			 * crtc's enabled, so we do the above call.
+			 *
+			 * At this point some state updated by the connectors
+			 * in their ->detect() callback has not run yet, so
+			 * no recalculation can be done yet.
+			 *
+			 * Even if we could do a recalculation and modeset
+			 * right now it would cause a double modeset if
+			 * fbdev or userspace chooses a different initial mode.
+			 *
+			 * If that happens, someone indicated they wanted a
+			 * mode change, which means it's safe to do a full
+			 * recalculation.
+			 */
+			crtc->base.state->mode.private_flags = I915_MODE_FLAG_INHERITED;
+		}
+
+		for_each_intel_plane(crtc->base.dev, plane) {
+			if (crtc->pipe != plane->pipe)
+				continue;
+
+			intel_sanitize_plane(plane);
+		}
+
 		intel_sanitize_crtc(crtc);
 		intel_dump_pipe_config(crtc, crtc->config,
 				       "[setup_hw_state]");