Skip to contentSkip to conversation

drm/i915/skl: Don't try to update plane watermarks if they haven't changed

PATCH Superseded Lyude 3 participants 6 messages Aug 29, 2016 – Sep 2, 2016

Subsystems (2)
lore
Share & tools
lore

Lyude

author
#0 lore raw

PATCH 1/1 drm/i915/skl: Don't try to update plane watermarks if they haven't changed

i915 sometimes needs to disable planes in the middle of an atomic commit, and then reenable them later in the same commit. Because of this, we can't make the assumption that the state of the plane actually changed. Since the state of the plane hasn't actually changed, neither have it's watermarks. And if the watermarks hasn't changed then we haven't populated skl_results with anything, which means we'll end up zeroing out a plane's watermarks in the middle of the atomic commit without restoring them later.

Changes since v1:
- Fix incorrect use of "it's"

Cc Maarten Lankhorst <[email protected]>
Fixes 62e0fb880123 ("drm/i915/skl: Update plane watermarks atomically

during plane updates")

---
 drivers/gpu/drm/i915/intel_display.c | 7 ++++++-
 drivers/gpu/drm/i915/intel_sprite.c  | 9 +++++++--
 2 files changed, 13 insertions(+), 3 deletions(-)
drivers/gpu/drm/i915/intel_display.c +6 -1
index e4e6141..13e47a7 100644
@@ -3448,7 +3448,12 @@ static void skylake_disable_primary_plane(struct drm_plane *primary,
 	struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
 	int pipe = intel_crtc->pipe;
 
-	skl_write_plane_wm(intel_crtc, &dev_priv->wm.skl_results, 0);
+	/*
+	 * We only populate skl_results on watermark updates, and if the
+	 * plane's visiblity isn't actually changing neither is its watermarks.
+	 */
+	if (!crtc->primary->state->visible)
+		skl_write_plane_wm(intel_crtc, &dev_priv->wm.skl_results, 0);
 
 	I915_WRITE(PLANE_CTL(pipe, 0), 0);
 	I915_WRITE(PLANE_SURF(pipe, 0), 0);
drivers/gpu/drm/i915/intel_sprite.c +7 -2
index 0df783a..73a521f 100644
@@ -292,8 +292,13 @@ skl_disable_plane(struct drm_plane *dplane, struct drm_crtc *crtc)
 	const int pipe = intel_plane->pipe;
 	const int plane = intel_plane->plane + 1;
 
-	skl_write_plane_wm(to_intel_crtc(crtc), &dev_priv->wm.skl_results,
-			   plane);
+	/*
+	 * We only populate skl_results on watermark updates, and if the
+	 * plane's visiblity isn't actually changing neither is its watermarks.
+	 */
+	if (!dplane->state->visible)
+		skl_write_plane_wm(to_intel_crtc(crtc),
+				   &dev_priv->wm.skl_results, plane);
 
 	I915_WRITE(PLANE_CTL(pipe, plane), 0);
 
Signature · 6 lines
2.7.4

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
maintainer
#1 lore raw
Quoted 8 lines · Lyude

i915 sometimes needs to disable planes in the middle of an atomic commit, and then reenable them later in the same commit. Because of this, we can't make the assumption that the state of the plane actually changed. Since the state of the plane hasn't actually changed, neither have it's watermarks. And if the watermarks hasn't changed then we haven't populated skl_results with anything, which means we'll end up zeroing out a plane's watermarks in the middle of the atomic commit without restoring them later.

I would appreciate a (brief) description of what the failure mode is in this case.

BR,
Jani.

Quoted 52 lines

Changes since v1:
- Fix incorrect use of "it's"

Cc Maarten Lankhorst <[email protected]>
Fixes 62e0fb880123 ("drm/i915/skl: Update plane watermarks atomically

during plane updates")

---
 drivers/gpu/drm/i915/intel_display.c | 7 ++++++-
 drivers/gpu/drm/i915/intel_sprite.c  | 9 +++++++--
 2 files changed, 13 insertions(+), 3 deletions(-)
drivers/gpu/drm/i915/intel_display.c +6 -1
index e4e6141..13e47a7 100644
@@ -3448,7 +3448,12 @@ static void skylake_disable_primary_plane(struct drm_plane *primary,
 	struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
 	int pipe = intel_crtc->pipe;
 
-	skl_write_plane_wm(intel_crtc, &dev_priv->wm.skl_results, 0);
+	/*
+	 * We only populate skl_results on watermark updates, and if the
+	 * plane's visiblity isn't actually changing neither is its watermarks.
+	 */
+	if (!crtc->primary->state->visible)
+		skl_write_plane_wm(intel_crtc, &dev_priv->wm.skl_results, 0);
 
 	I915_WRITE(PLANE_CTL(pipe, 0), 0);
 	I915_WRITE(PLANE_SURF(pipe, 0), 0);
drivers/gpu/drm/i915/intel_sprite.c +7 -2
index 0df783a..73a521f 100644
@@ -292,8 +292,13 @@ skl_disable_plane(struct drm_plane *dplane, struct drm_crtc *crtc)
 	const int pipe = intel_plane->pipe;
 	const int plane = intel_plane->plane + 1;
 
-	skl_write_plane_wm(to_intel_crtc(crtc), &dev_priv->wm.skl_results,
-			   plane);
+	/*
+	 * We only populate skl_results on watermark updates, and if the
+	 * plane's visiblity isn't actually changing neither is its watermarks.
+	 */
+	if (!dplane->state->visible)
+		skl_write_plane_wm(to_intel_crtc(crtc),
+				   &dev_priv->wm.skl_results, plane);
 
 	I915_WRITE(PLANE_CTL(pipe, plane), 0);
Signature · 5 lines
Jani Nikula, Intel Open Source Technology Center
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
author
#2 lore raw

PATCH v3 1/1 drm/i915/skl: Don't try to update plane watermarks if they haven't changed

i915 sometimes needs to disable planes in the middle of an atomic commit, and then reenable them later in the same commit. Because of this, we can't make the assumption that the state of the plane actually changed. Since the state of the plane hasn't actually changed, neither have it's watermarks. And if the watermarks hasn't changed then we haven't populated skl_results with anything, which means we'll end up zeroing out a plane's watermarks in the middle of the atomic commit without restoring them later.

Simple reproduction recipe:
- Get a SKL laptop, launch any kind of X session
- Get two extra monitors
- Keep hotplugging both displays (so that the display configuration jumps from 1 active pipe to 3 active pipes and back)
- Eventually underrun

Changes since v1:
- Fix incorrect use of "it's"
Changes since v2:
- Add reproduction recipe

Cc Maarten Lankhorst <[email protected]>
Fixes 62e0fb880123 ("drm/i915/skl: Update plane watermarks atomically

during plane updates")

---
 drivers/gpu/drm/i915/intel_display.c | 7 ++++++-
 drivers/gpu/drm/i915/intel_sprite.c  | 9 +++++++--
 2 files changed, 13 insertions(+), 3 deletions(-)
drivers/gpu/drm/i915/intel_display.c +6 -1
index e4e6141..13e47a7 100644
@@ -3448,7 +3448,12 @@ static void skylake_disable_primary_plane(struct drm_plane *primary,
 	struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
 	int pipe = intel_crtc->pipe;
 
-	skl_write_plane_wm(intel_crtc, &dev_priv->wm.skl_results, 0);
+	/*
+	 * We only populate skl_results on watermark updates, and if the
+	 * plane's visiblity isn't actually changing neither is its watermarks.
+	 */
+	if (!crtc->primary->state->visible)
+		skl_write_plane_wm(intel_crtc, &dev_priv->wm.skl_results, 0);
 
 	I915_WRITE(PLANE_CTL(pipe, 0), 0);
 	I915_WRITE(PLANE_SURF(pipe, 0), 0);
drivers/gpu/drm/i915/intel_sprite.c +7 -2
index 0df783a..73a521f 100644
@@ -292,8 +292,13 @@ skl_disable_plane(struct drm_plane *dplane, struct drm_crtc *crtc)
 	const int pipe = intel_plane->pipe;
 	const int plane = intel_plane->plane + 1;
 
-	skl_write_plane_wm(to_intel_crtc(crtc), &dev_priv->wm.skl_results,
-			   plane);
+	/*
+	 * We only populate skl_results on watermark updates, and if the
+	 * plane's visiblity isn't actually changing neither is its watermarks.
+	 */
+	if (!dplane->state->visible)
+		skl_write_plane_wm(to_intel_crtc(crtc),
+				   &dev_priv->wm.skl_results, plane);
 
 	I915_WRITE(PLANE_CTL(pipe, plane), 0);
 
Signature · 6 lines
2.7.4

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
author
#3 lore raw

Since this patch has been on hold for a little bit, I did a bit of thinking of how we could this a little more cleanly. Unfortunately I couldn't think of a way, however I did think of an alternative solution:

I'm planning on backporting all of the skl wm fixes already, so I'm going to use this patch for that since it's very small. As for mainline, I'm going to do a whole reorganization of the skl wm/ddb structs in i915 like Matt had suggested before. Things might look a little more like this (taken from my half-complete reorganization):

struct skl_plane_ddb_allocation {
	struct skl_ddb_entry plane;
	struct skl_ddb_entry y_plane;
};
struct skl_plane_wm_values {
	struct skl_plane_ddb_allocation ddb;
	uint32_t wm[8];
	uint32_t trans_wm;
};
struct skl_pipe_wm_values {
	struct skl_ddb_entry ddb;
	uint32_t linetime;
};

struct skl_hw_wm_values { /* (only used for skl_get_hw_ddb and friends) */

	struct skl_pipe_wm_values pipe[I915_MAX_PIPES];
	struct skl_plane_wm_values plane[I915_MAX_PIPES][I915_MAX_PLANES];
};

As well, I'm also just going to completely remove the skl_results and skl_hw structs from struct drm_i915_private. This makes sense for a lot of reasons:
* This completely gets rid of the need for a global watermark lock (on Skylake at least) and will make things a lot easier for atomic support in the future
* Skylake doesn't have any actual global watermark hooks anyway, aside from skl_update_wm() which is now only used for writing watermarks for inactive pipes during haswell_crtc_enable()
* This makes passing watermarks around way less of a mess
* Saves a tiny bit of data, and so far being able to grab watermarks/ddbs right from the plane states seems to be a lot easier then messing with a large array

As for this fix, I'll probably still need someone to review it so I can get it into 4.7.y.

Let me know what you think.

Quoted 85 lines · Lyude

i915 sometimes needs to disable planes in the middle of an atomic commit, and then reenable them later in the same commit. Because of this, we can't make the assumption that the state of the plane actually
changed. Since the state of the plane hasn't actually changed, neither
have it's watermarks. And if the watermarks hasn't changed then we haven't populated skl_results with anything, which means we'll end up zeroing out a plane's watermarks in the middle of the atomic commit without restoring them later.

Simple reproduction recipe:
 - Get a SKL laptop, launch any kind of X session
 - Get two extra monitors
 - Keep hotplugging both displays (so that the display configuration jumps from 1 active pipe to 3 active pipes and back)
 - Eventually underrun

Changes since v1:
 - Fix incorrect use of "it's"
Changes since v2:
 - Add reproduction recipe

Cc Maarten Lankhorst <[email protected]>
Fixes 62e0fb880123 ("drm/i915/skl: Update plane watermarks

atomically
during plane updates")

---
 drivers/gpu/drm/i915/intel_display.c | 7 ++++++- drivers/gpu/drm/i915/intel_sprite.c  | 9 +++++++--

 2 files changed, 13 insertions(+), 3 deletions(-)
drivers/gpu/drm/i915/intel_display.c +0 -0
index e4e6141..13e47a7 100644
@@ -3448,7 +3448,12 @@ static void
 skylake_disable_primary_plane(struct drm_plane *primary,
  	struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
  	int pipe = intel_crtc->pipe;

drivers/gpu/drm/i915/intel_display.c +0 -1
-	skl_write_plane_wm(intel_crtc, &dev_priv->wm.skl_results,

0);

drivers/gpu/drm/i915/intel_display.c +2 -0
+	/*
+	 * We only populate skl_results on watermark updates, and if

the

drivers/gpu/drm/i915/intel_display.c +1 -0
+	 * plane's visiblity isn't actually changing neither is its

watermarks.

drivers/gpu/drm/i915/intel_display.c +3 -0
+	 */
+	if (!crtc->primary->state->visible)
+		skl_write_plane_wm(intel_crtc, &dev_priv-
Quoted 1 line

wm.skl_results, 0);


  I915_WRITE(PLANE_CTL(pipe, 0), 0);
  I915_WRITE(PLANE_SURF(pipe, 0), 0);

drivers/gpu/drm/i915/intel_sprite.c +0 -0
index 0df783a..73a521f 100644
@@ -292,8 +292,13 @@ skl_disable_plane(struct drm_plane *dplane,
 struct drm_crtc *crtc)
  	const int pipe = intel_plane->pipe;
  	const int plane = intel_plane->plane + 1;

drivers/gpu/drm/i915/intel_sprite.c +0 -1
-	skl_write_plane_wm(to_intel_crtc(crtc), &dev_priv-
Quoted 1 line

wm.skl_results,

drivers/gpu/drm/i915/intel_sprite.c +2 -1
-			   plane);
+	/*
+	 * We only populate skl_results on watermark updates, and if

the

drivers/gpu/drm/i915/intel_sprite.c +1 -0
+	 * plane's visiblity isn't actually changing neither is its

watermarks.

drivers/gpu/drm/i915/intel_sprite.c +4 -0
+	 */
+	if (!dplane->state->visible)
+		skl_write_plane_wm(to_intel_crtc(crtc),
+				   &dev_priv->wm.skl_results,

plane);

  I915_WRITE(PLANE_CTL(pipe, plane), 0);

Signature · 2 lines
Cheers,
	Lyude
#4 lore raw

✗ Fi.CI.BAT: failure for drm/i915/skl: Don't try to update plane watermarks if they haven't changed (rev2)

In reply to Lyude
== Series Details ==

== Series Details ==

Series: drm/i915/skl: Don't try to update plane watermarks if they haven't changed (rev2)
URL : https://patchwork.freedesktop.org/series/11654/
State : failure

== Summary ==

Series 11654v2 drm/i915/skl: Don't try to update plane watermarks if they haven't changed
http://patchwork.freedesktop.org/api/1.0/series/11654/revisions/2/mbox/

Test kms_cursor_legacy:

        Subgroup basic-cursor-vs-flip-legacy:
                pass       -> FAIL       (fi-bsw-n3050)
Test kms_pipe_crc_basic:
        Subgroup suspend-read-crc-pipe-a:
                incomplete -> PASS       (fi-hsw-4770k)
fi-bdw-5557u     total:252  pass:235  dwarn:0   dfail:0   fail:2   skip:15 
fi-bsw-n3050     total:252  pass:204  dwarn:0   dfail:0   fail:2   skip:46 
fi-byt-n2820     total:252  pass:208  dwarn:0   dfail:0   fail:3   skip:41 
fi-hsw-4770k     total:252  pass:228  dwarn:0   dfail:0   fail:2   skip:22 
fi-hsw-4770r     total:252  pass:224  dwarn:0   dfail:0   fail:2   skip:26 
fi-ivb-3520m     total:252  pass:220  dwarn:0   dfail:0   fail:1   skip:31 
fi-skl-6260u     total:252  pass:236  dwarn:0   dfail:0   fail:2   skip:14 
fi-skl-6700k     total:252  pass:222  dwarn:0   dfail:0   fail:2   skip:28 
fi-snb-2520m     total:252  pass:207  dwarn:0   dfail:0   fail:2   skip:43 
fi-snb-2600      total:252  pass:207  dwarn:0   dfail:0   fail:2   skip:43 

Results at /archive/results/CI_IGT_test/Patchwork_2446/

f4f46e5544894b2198cdfd5a226ee587d9834cc4 drm-intel-nightly: 2016y-08m-29d-14h-09m-23s UTC integration manifest 9d8dab1 drm/i915/skl: Don't try to update plane watermarks if they haven't changed

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#5 lore raw

✓ Fi.CI.BAT: success for drm/i915/skl: Don't try to update plane watermarks if they haven't changed (rev3)

In reply to Lyude
== Series Details ==

== Series Details ==

Series: drm/i915/skl: Don't try to update plane watermarks if they haven't changed (rev3)
URL : https://patchwork.freedesktop.org/series/11654/
State : success

== Summary ==

Series 11654v3 drm/i915/skl: Don't try to update plane watermarks if they haven't changed
http://patchwork.freedesktop.org/api/1.0/series/11654/revisions/3/mbox/

fi-bdw-5557u     total:252  pass:235  dwarn:0   dfail:0   fail:2   skip:15 
fi-bsw-n3050     total:252  pass:205  dwarn:0   dfail:0   fail:1   skip:46 
fi-hsw-4770k     total:107  pass:91   dwarn:0   dfail:0   fail:0   skip:15 
fi-hsw-4770r     total:252  pass:223  dwarn:0   dfail:0   fail:2   skip:27 
fi-ivb-3520m     total:252  pass:220  dwarn:0   dfail:0   fail:1   skip:31 
fi-skl-6260u     total:252  pass:236  dwarn:0   dfail:0   fail:2   skip:14 
fi-skl-6700k     total:252  pass:222  dwarn:0   dfail:0   fail:2   skip:28 
fi-snb-2520m     total:252  pass:207  dwarn:0   dfail:0   fail:2   skip:43 
fi-snb-2600      total:252  pass:207  dwarn:0   dfail:0   fail:2   skip:43 

Results at /archive/results/CI_IGT_test/Patchwork_2447/

57de27e40b9741c17c6749a366e891faf8b22fcb drm-intel-nightly: 2016y-08m-29d-15h-38m-26s UTC integration manifest 6c4a762 drm/i915/skl: Don't try to update plane watermarks if they haven't changed

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx