Skip to contentSkip to conversation

Allow ASYNC flip with atomic helpers.

PATCH Superseded Andrey Grodzovsky 11 participants 27 messages Jan 16, 2017 – Jan 26, 2017 6 messages not in the archive

Subsystems (5)

v1 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

cover Allow ASYNC flip with atomic helpers.

This series is a folow-up on
https://patchwork.kernel.org/patch/9501787/

The first patch makes changes to atomic helpers to allow for drives with ASYNC flip support to use them. Patches 2 and 3 are to use this in AMDGPU/DC and patch 4 is possible cleanup in nouveau/kms who seems to have the duplicate the helper as we did to support ASYNC flips.

Andrey Grodzovsky (4):
drm/atomic: Save flip flags in drm_plane_state drm/amdgpu: Remove flip_flag from amdgpu_crtc drm/amd/display: Switch to using atomic_helper for flip. drm/nouveau/kms/nv50: Switch to using atomic helper for flip.

 drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h           |  1 -
 .../drm/amd/display/amdgpu_dm/amdgpu_dm_types.c    | 92 ++--------------------
 drivers/gpu/drm/drm_atomic_helper.c                | 10 +--
 drivers/gpu/drm/nouveau/nv50_display.c             | 77 ++----------------
 include/drm/drm_plane.h                            |  8 ++
 5 files changed, 22 insertions(+), 166 deletions(-)
Signature · 6 lines
1.9.1

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#1 lore raw

PATCH 1/4 drm/atomic: Save flip flags in drm_plane_state

Allows using atomic flip helpers for drivers
using ASYNC flip.
Remove ASYNC_FLIP restriction in helpers and
caches the page flip flags in drm_plane_state
to be used in the low level drivers.

---
 drivers/gpu/drm/drm_atomic_helper.c | 10 +++-------
 include/drm/drm_plane.h             |  8 ++++++++
 2 files changed, 11 insertions(+), 7 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +3 -7
index a4e5477..f83dc43 100644
@@ -2737,7 +2737,8 @@ static int page_flip_common(
 				struct drm_atomic_state *state,
 				struct drm_crtc *crtc,
 				struct drm_framebuffer *fb,
-				struct drm_pending_vblank_event *event)
+				struct drm_pending_vblank_event *event,
+				uint32_t flags)
 {
 	struct drm_plane *plane = crtc->primary;
 	struct drm_plane_state *plane_state;
@@ -2754,6 +2755,7 @@ static int page_flip_common(
 	if (IS_ERR(plane_state))
 		return PTR_ERR(plane_state);
 
+	plane_state->pflip_flags = flags;
 
 	ret = drm_atomic_set_crtc_for_plane(plane_state, crtc);
 	if (ret != 0)
@@ -2800,9 +2802,6 @@ int drm_atomic_helper_page_flip(struct drm_crtc *crtc,
 	struct drm_atomic_state *state;
 	int ret = 0;
 
-	if (flags & DRM_MODE_PAGE_FLIP_ASYNC)
-		return -EINVAL;
-
 	state = drm_atomic_state_alloc(plane->dev);
 	if (!state)
 		return -ENOMEM;
@@ -2865,9 +2864,6 @@ int drm_atomic_helper_page_flip_target(
 	struct drm_crtc_state *crtc_state;
 	int ret = 0;
 
-	if (flags & DRM_MODE_PAGE_FLIP_ASYNC)
-		return -EINVAL;
-
 	state = drm_atomic_state_alloc(plane->dev);
 	if (!state)
 		return -ENOMEM;
include/drm/drm_plane.h +8 -0
index db3bbde..86d8ffc 100644
@@ -122,6 +122,14 @@ struct drm_plane_state {
 	 */
 	bool visible;
 
+
+	/**
+	 * @pflip_flags:
+	 *
+	 * Flip related config options
+	 */
+	u32 pflip_flags;
+
 	struct drm_atomic_state *state;
 };
 
Signature · 6 lines
1.9.1

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

A message that is not in the archive (sent off-list or not yet received): 1484581498-32309-2-git-send-email-Andrey.Grodzovsky-5C7GfCeVMHo@public.gmane.org

#3 lore raw

Re: [PATCH 1/4] drm/atomic: Save flip flags in drm_plane_state

Hi Andrey,

2017-01-16 Andrey Grodzovsky <[email protected]>:

Quoted 17 lines

Allows using atomic flip helpers for drivers
using ASYNC flip.
Remove ASYNC_FLIP restriction in helpers and
caches the page flip flags in drm_plane_state
to be used in the low level drivers.

---
 drivers/gpu/drm/drm_atomic_helper.c | 10 +++-------
 include/drm/drm_plane.h             |  8 ++++++++
 2 files changed, 11 insertions(+), 7 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +0 -0
index a4e5477..f83dc43 100644

drivers/gpu/drm/drm_atomic_helper.c static int page_flip_common( · line 2741

Quoted 6 lines
drivers/gpu/drm/drm_atomic_helper.c +2 -1
@@ -2737,7 +2737,8 @@ static int page_flip_common(
 				struct drm_atomic_state *state,
 				struct drm_crtc *crtc,
 				struct drm_framebuffer *fb,
-				struct drm_pending_vblank_event *event)
+				struct drm_pending_vblank_event *event,
+				uint32_t flags)

Did you build this patch? It is changing the signature of page_flip_common() but no changes to the callers.

Gustavo

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

#4 lore raw

RE: [PATCH 1/4] drm/atomic: Save flip flags in drm_plane_state

Quoted 40 lines

-----Original Message-----
From: Gustavo Padovan [mailto:[email protected]]
Sent: Monday, January 16, 2017 3:22 PM
To: Grodzovsky, Andrey
Cc: [email protected]; [email protected]; amd- [email protected]; Deucher, Alexander; [email protected]
Subject: Re: [PATCH 1/4] drm/atomic: Save flip flags in drm_plane_state

Hi Andrey,

2017-01-16 Andrey Grodzovsky <[email protected]>:

Quoted 16 lines

Allows using atomic flip helpers for drivers using ASYNC flip. Remove ASYNC_FLIP restriction in helpers and caches the page flip flags in drm_plane_state to be used in the low level drivers.

---
 drivers/gpu/drm/drm_atomic_helper.c | 10 +++-------
 include/drm/drm_plane.h             |  8 ++++++++
 2 files changed, 11 insertions(+), 7 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +0 -0
index a4e5477..f83dc43 100644

drivers/gpu/drm/drm_atomic_helper.c static int page_flip_common( · line 2741

Quoted 6 lines
drivers/gpu/drm/drm_atomic_helper.c +2 -1
@@ -2737,7 +2737,8 @@ static int page_flip_common(
 				struct drm_atomic_state *state,
 				struct drm_crtc *crtc,
 				struct drm_framebuffer *fb,
-				struct drm_pending_vblank_event *event)
+				struct drm_pending_vblank_event *event,
+				uint32_t flags)

Did you build this patch? It is changing the signature of page_flip_common() but no changes to the callers.

Quoted 1 line

Gustavo

Thanks for spotting this, I am afraid I've sent not the final version of the patch. I will resend the latest version later today.

Thanks
Andrey

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#5 lore raw

Re: [PATCH 1/4] drm/atomic: Save flip flags in drm_plane_state

In reply to a message not in the archive
Quoted 7 lines · Andrey Grodzovsky

Allows using atomic flip helpers for drivers
using ASYNC flip.
Remove ASYNC_FLIP restriction in helpers and
caches the page flip flags in drm_plane_state
to be used in the low level drivers.

It's mostly guesswork, but I think we should have the flip flags in the crtc, not in each plane. Similar to how we move the event from planes to crtc.
-Daniel

Quoted 73 lines
---
 drivers/gpu/drm/drm_atomic_helper.c | 10 +++-------
 include/drm/drm_plane.h             |  8 ++++++++
 2 files changed, 11 insertions(+), 7 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +3 -7
index a4e5477..f83dc43 100644
@@ -2737,7 +2737,8 @@ static int page_flip_common(
 				struct drm_atomic_state *state,
 				struct drm_crtc *crtc,
 				struct drm_framebuffer *fb,
-				struct drm_pending_vblank_event *event)
+				struct drm_pending_vblank_event *event,
+				uint32_t flags)
 {
 	struct drm_plane *plane = crtc->primary;
 	struct drm_plane_state *plane_state;
@@ -2754,6 +2755,7 @@ static int page_flip_common(
 	if (IS_ERR(plane_state))
 		return PTR_ERR(plane_state);
 
+	plane_state->pflip_flags = flags;
 
 	ret = drm_atomic_set_crtc_for_plane(plane_state, crtc);
 	if (ret != 0)
@@ -2800,9 +2802,6 @@ int drm_atomic_helper_page_flip(struct drm_crtc *crtc,
 	struct drm_atomic_state *state;
 	int ret = 0;
 
-	if (flags & DRM_MODE_PAGE_FLIP_ASYNC)
-		return -EINVAL;
-
 	state = drm_atomic_state_alloc(plane->dev);
 	if (!state)
 		return -ENOMEM;
@@ -2865,9 +2864,6 @@ int drm_atomic_helper_page_flip_target(
 	struct drm_crtc_state *crtc_state;
 	int ret = 0;
 
-	if (flags & DRM_MODE_PAGE_FLIP_ASYNC)
-		return -EINVAL;
-
 	state = drm_atomic_state_alloc(plane->dev);
 	if (!state)
 		return -ENOMEM;
include/drm/drm_plane.h +8 -0
index db3bbde..86d8ffc 100644
@@ -122,6 +122,14 @@ struct drm_plane_state {
 	 */
 	bool visible;
 
+
+	/**
+	 * @pflip_flags:
+	 *
+	 * Flip related config options
+	 */
+	u32 pflip_flags;
+
 	struct drm_atomic_state *state;
 };
 
Signature · 6 lines
1.9.1

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#6 lore raw

RE: [PATCH 1/4] drm/atomic: Save flip flags in drm_plane_state

Quoted 17 lines

-----Original Message-----
From: dri-devel [mailto:[email protected]] On Behalf Of Daniel Vetter
Sent: Monday, January 23, 2017 3:55 AM
To: Grodzovsky, Andrey <[email protected]>
Cc: Deucher, Alexander <[email protected]>; [email protected]; [email protected]; dri- [email protected]; [email protected]
Subject: Re: [PATCH 1/4] drm/atomic: Save flip flags in drm_plane_state

Quoted 4 lines · Andrey Grodzovsky

Allows using atomic flip helpers for drivers using ASYNC flip. Remove ASYNC_FLIP restriction in helpers and caches the page flip flags in drm_plane_state to be used in the low level drivers.

Quoted 5 lines
Quoted 1 line

It's mostly guesswork, but I think we should have the flip flags in the crtc, not in each plane. Similar to how we move the event from planes to crtc. -Daniel

What does ASYNC flip mean? HW flip as soon as possible and result in tearing on screen? If so I could imaging some use case where you have some UI control/menu overlay on top, and some game running on a underlay plane, and the game want to be able to flip as soon as possible. Or Daniel do you think ASYNC property will apply to all planes in CRTC?

Quoted 72 lines

drivers/gpu/drm/drm_atomic_helper.c int drm_atomic_helper_page_flip(struct drm_crtc

Quoted 30 lines
---
 drivers/gpu/drm/drm_atomic_helper.c | 10 +++-------
 include/drm/drm_plane.h             |  8 ++++++++
 2 files changed, 11 insertions(+), 7 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +2 -1
index a4e5477..f83dc43 100644
@@ -2737,7 +2737,8 @@ static int page_flip_common(
 				struct drm_atomic_state *state,
 				struct drm_crtc *crtc,
 				struct drm_framebuffer *fb,
-				struct drm_pending_vblank_event *event)
+				struct drm_pending_vblank_event *event,
+				uint32_t flags)
 {
 	struct drm_plane *plane = crtc->primary;
 	struct drm_plane_state *plane_state; @@ -2754,6 +2755,7 @@ static

int page_flip_common(

drivers/gpu/drm/drm_atomic_helper.c +1 -0
 	if (IS_ERR(plane_state))
 		return PTR_ERR(plane_state);
 
+	plane_state->pflip_flags = flags;
 
 	ret = drm_atomic_set_crtc_for_plane(plane_state, crtc);
 	if (ret != 0)
@@ -2800,9 +2802,6 @@ int drm_atomic_helper_page_flip(struct drm_crtc

*crtc,

Quoted 41 lines
drivers/gpu/drm/drm_atomic_helper.c +0 -6
 	struct drm_atomic_state *state;
 	int ret = 0;
 
-	if (flags & DRM_MODE_PAGE_FLIP_ASYNC)
-		return -EINVAL;
-
 	state = drm_atomic_state_alloc(plane->dev);
 	if (!state)
 		return -ENOMEM;
@@ -2865,9 +2864,6 @@ int drm_atomic_helper_page_flip_target(
 	struct drm_crtc_state *crtc_state;
 	int ret = 0;
 
-	if (flags & DRM_MODE_PAGE_FLIP_ASYNC)
-		return -EINVAL;
-
 	state = drm_atomic_state_alloc(plane->dev);
 	if (!state)
 		return -ENOMEM;
include/drm/drm_plane.h index renamed from include/drm/drm_plane.h +0 -0

db3bbde..86d8ffc 100644

include/drm/drm_plane.h +8 -1
@@ -122,6 +122,14 @@ struct drm_plane_state {
 	 */
 	bool visible;
 
+
+	/**
+	 * @pflip_flags:
+	 *
+	 * Flip related config options
+	 */
+	u32 pflip_flags;
+
 	struct drm_atomic_state *state;
 };
 
--

1.9.1

Quoted 13 lines
Quoted 4 lines

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

A message that is not in the archive (sent off-list or not yet received): MWHPR12MB14066404FBCEF3C03EF24BF998720-Gy0DoCVfaSXS1+HfpYZxvgdYzm3356FpvxpqHgZTriW3zl9H0oFU5g@public.gmane.org

#8 lore raw

RE: [PATCH 1/4] drm/atomic: Save flip flags in drm_plane_state

Quoted 29 lines

-----Original Message-----
From: Cheng, Tony
Sent: Monday, January 23, 2017 2:49 PM
To: Daniel Vetter; Grodzovsky, Andrey
Cc: Deucher, Alexander; [email protected]; amd- [email protected]; [email protected]; [email protected]; dc_upstream
Subject: RE: [PATCH 1/4] drm/atomic: Save flip flags in drm_plane_state

Quoted 16 lines

-----Original Message-----
From: dri-devel [mailto:[email protected]] On Behalf Of Daniel Vetter
Sent: Monday, January 23, 2017 3:55 AM
To: Grodzovsky, Andrey <[email protected]>
Cc: Deucher, Alexander <[email protected]>; [email protected]; [email protected]; dri- [email protected]; [email protected]
Subject: Re: [PATCH 1/4] drm/atomic: Save flip flags in drm_plane_state

Quoted 4 lines · Andrey Grodzovsky

Allows using atomic flip helpers for drivers using ASYNC flip. Remove ASYNC_FLIP restriction in helpers and caches the page flip flags in drm_plane_state to be used in the low level drivers.

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

It's mostly guesswork, but I think we should have the flip flags in the crtc, not in each plane. Similar to how we move the event from planes

to crtc.

Quoted 1 line

-Daniel

Do you mean here crtc or crtc_state ?

Quoted 6 lines

What does ASYNC flip mean? HW flip as soon as possible and result in tearing on screen? If so I could imaging some use case where you have some UI control/menu overlay on top, and some game running on a underlay plane, and the game want to be able to flip as soon as possible. Or Daniel do you think ASYNC property will apply to all planes in CRTC?

Also the client code both in nouveau and AMD/DAL iterates over new planes and their new states so it seems to me easier to retrieve the parameter from the plane_state and not adding loop to find matching crtc or its state, especially if it's before drm_atomic_helper_swap_state is called,
when crtc->state still points to the old state ...
[Grodzovsky, Andrey]

Quoted 73 lines
Quoted 73 lines
Quoted 30 lines
---
 drivers/gpu/drm/drm_atomic_helper.c | 10 +++-------
 include/drm/drm_plane.h             |  8 ++++++++
 2 files changed, 11 insertions(+), 7 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +2 -1
index a4e5477..f83dc43 100644
@@ -2737,7 +2737,8 @@ static int page_flip_common(
 				struct drm_atomic_state *state,
 				struct drm_crtc *crtc,
 				struct drm_framebuffer *fb,
-				struct drm_pending_vblank_event *event)
+				struct drm_pending_vblank_event *event,
+				uint32_t flags)
 {
 	struct drm_plane *plane = crtc->primary;
 	struct drm_plane_state *plane_state; @@ -2754,6 +2755,7 @@ static

int page_flip_common(

drivers/gpu/drm/drm_atomic_helper.c +1 -0
 	if (IS_ERR(plane_state))
 		return PTR_ERR(plane_state);
 
+	plane_state->pflip_flags = flags;
 
 	ret = drm_atomic_set_crtc_for_plane(plane_state, crtc);
 	if (ret != 0)
@@ -2800,9 +2802,6 @@ int drm_atomic_helper_page_flip(struct
Quoted 1 line

drm_crtc

*crtc,

Quoted 41 lines
drivers/gpu/drm/drm_atomic_helper.c +0 -6
 	struct drm_atomic_state *state;
 	int ret = 0;
 
-	if (flags & DRM_MODE_PAGE_FLIP_ASYNC)
-		return -EINVAL;
-
 	state = drm_atomic_state_alloc(plane->dev);
 	if (!state)
 		return -ENOMEM;
@@ -2865,9 +2864,6 @@ int drm_atomic_helper_page_flip_target(
 	struct drm_crtc_state *crtc_state;
 	int ret = 0;
 
-	if (flags & DRM_MODE_PAGE_FLIP_ASYNC)
-		return -EINVAL;
-
 	state = drm_atomic_state_alloc(plane->dev);
 	if (!state)
 		return -ENOMEM;
include/drm/drm_plane.h index renamed from include/drm/drm_plane.h +0 -0

db3bbde..86d8ffc 100644

include/drm/drm_plane.h +8 -1
@@ -122,6 +122,14 @@ struct drm_plane_state {
 	 */
 	bool visible;
 
+
+	/**
+	 * @pflip_flags:
+	 *
+	 * Flip related config options
+	 */
+	u32 pflip_flags;
+
 	struct drm_atomic_state *state;
 };
 
--

1.9.1

Quoted 13 lines
Quoted 5 lines
Quoted 4 lines

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Quoted 8 lines

--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

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

A message that is not in the archive (sent off-list or not yet received): DM5PR12MB1147AEF3F8CFFAD4B69E114EEA770-2J9CzHegvk9IwjGPM8RDJwdYzm3356FpvxpqHgZTriW3zl9H0oFU5g@public.gmane.org

#10 lore raw

Re: [PATCH 1/4] drm/atomic: Save flip flags in drm_plane_state

Quoted 29 lines · Andrey
Quoted 29 lines

-----Original Message-----
From: Cheng, Tony
Sent: Monday, January 23, 2017 2:49 PM
To: Daniel Vetter; Grodzovsky, Andrey
Cc: Deucher, Alexander; [email protected]; amd- [email protected]; [email protected]; [email protected]; dc_upstream
Subject: RE: [PATCH 1/4] drm/atomic: Save flip flags in drm_plane_state

Quoted 16 lines

-----Original Message-----
From: dri-devel [mailto:[email protected]] On Behalf Of Daniel Vetter
Sent: Monday, January 23, 2017 3:55 AM
To: Grodzovsky, Andrey <[email protected]>
Cc: Deucher, Alexander <[email protected]>; [email protected]; [email protected]; dri- [email protected]; [email protected]
Subject: Re: [PATCH 1/4] drm/atomic: Save flip flags in drm_plane_state

Quoted 4 lines · Andrey Grodzovsky

Allows using atomic flip helpers for drivers using ASYNC flip. Remove ASYNC_FLIP restriction in helpers and caches the page flip flags in drm_plane_state to be used in the low level drivers.

Quoted 8 lines
Quoted 7 lines
Quoted 4 lines
Quoted 1 line

It's mostly guesswork, but I think we should have the flip flags in the crtc, not in each plane. Similar to how we move the event from planes

to crtc.

Quoted 1 line

-Daniel

Do you mean here crtc or crtc_state ?

Yes.

Quoted 10 lines
Quoted 6 lines

What does ASYNC flip mean? HW flip as soon as possible and result in tearing on screen? If so I could imaging some use case where you have some UI control/menu overlay on top, and some game running on a underlay plane, and the game want to be able to flip as soon as possible. Or Daniel do you think ASYNC property will apply to all planes in CRTC?

Also the client code both in nouveau and AMD/DAL iterates over new planes and their new states so it seems to me easier to retrieve the parameter from the plane_state and not adding loop to find matching crtc or its state, especially if it's before drm_atomic_helper_swap_state is called,
when crtc->state still points to the old state ...

It's pretty easy to go from plane_state to crtc_state. No need to loop at all. And the implementation shouldn't justify where we put it, the important part is the semantics.
-Daniel

Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#11 lore raw

Re: [PATCH 1/4] drm/atomic: Save flip flags in drm_plane_state

In reply to Cheng, Tony
Quoted 23 lines · Tony
Quoted 17 lines

-----Original Message-----
From: dri-devel [mailto:[email protected]] On Behalf Of Daniel Vetter
Sent: Monday, January 23, 2017 3:55 AM
To: Grodzovsky, Andrey <[email protected]>
Cc: Deucher, Alexander <[email protected]>; [email protected]; [email protected]; dri- [email protected]; [email protected]
Subject: Re: [PATCH 1/4] drm/atomic: Save flip flags in drm_plane_state

Quoted 4 lines · Andrey Grodzovsky

Allows using atomic flip helpers for drivers using ASYNC flip. Remove ASYNC_FLIP restriction in helpers and caches the page flip flags in drm_plane_state to be used in the low level drivers.

Quoted 5 lines
Quoted 1 line

It's mostly guesswork, but I think we should have the flip flags in the crtc, not in each plane. Similar to how we move the event from planes to crtc. -Daniel

Quoted 5 lines

What does ASYNC flip mean? HW flip as soon as possible and result in tearing on screen? If so I could imaging some use case where you have some UI control/menu overlay on top, and some game running on a underlay plane, and the game want to be able to flip as soon as possible. Or Daniel do you think ASYNC property will apply to all planes in CRTC?

Those kind of questions are exactly why I think we should wait with exposing async through the atomic ioctl until someone needs it. And yes async means "as fast as possible, with tearing".
-Daniel

Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#12 lore raw

Re: [PATCH 1/4] drm/atomic: Save flip flags in drm_plane_state

In reply to Andrey Grodzovsky

Hi Andrey,

Thank you for the patch.

Quoted 35 lines · On Monday 16 Jan 2017 10:44:55 Andrey Grodzovsky wrote:

Allows using atomic flip helpers for drivers
using ASYNC flip.
Remove ASYNC_FLIP restriction in helpers and
caches the page flip flags in drm_plane_state
to be used in the low level drivers.

---
 drivers/gpu/drm/drm_atomic_helper.c | 10 +++-------
 include/drm/drm_plane.h             |  8 ++++++++
 2 files changed, 11 insertions(+), 7 deletions(-)
(quoted diff) +0 -0

b/drivers/gpu/drm/drm_atomic_helper.c index a4e5477..f83dc43 100644

drivers/gpu/drm/drm_atomic_helper.c +3 -1
@@ -2737,7 +2737,8 @@ static int page_flip_common(
 				struct drm_atomic_state *state,
 				struct drm_crtc *crtc,
 				struct drm_framebuffer *fb,
-				struct drm_pending_vblank_event *event)
+				struct drm_pending_vblank_event *event,
+				uint32_t flags)
 {
 	struct drm_plane *plane = crtc->primary;
 	struct drm_plane_state *plane_state;
@@ -2754,6 +2755,7 @@ static int page_flip_common(
 	if (IS_ERR(plane_state))
 		return PTR_ERR(plane_state);
 
+	plane_state->pflip_flags = flags;
 
 	ret = drm_atomic_set_crtc_for_plane(plane_state, crtc);
 	if (ret != 0)

drivers/gpu/drm/drm_atomic_helper.c int drm_atomic_helper_page_flip(struct drm_crtc *crtc, · line 2805

Quoted 6 lines
drivers/gpu/drm/drm_atomic_helper.c +0 -3
@@ -2800,9 +2802,6 @@ int drm_atomic_helper_page_flip(struct drm_crtc *crtc,
 struct drm_atomic_state *state;
 	int ret = 0;
 
-	if (flags & DRM_MODE_PAGE_FLIP_ASYNC)
-		return -EINVAL;
-

With this change all drivers using the helper will not reject that async flag, even if they don't implement support for async page flip. You need to either patch them all to reject the flag, or implement async page flip support for all of them (preferable in the helpers, and then remove the

* Note that for now so called async page flips (i.e. updates which are not
* synchronized to vblank) are not supported, since the atomic interfaces have
* no provisions for this yet.

comment).

Quoted 31 lines
drivers/gpu/drm/drm_atomic_helper.c +0 -3
 	state = drm_atomic_state_alloc(plane->dev);
 	if (!state)
 		return -ENOMEM;
@@ -2865,9 +2864,6 @@ int drm_atomic_helper_page_flip_target(
 	struct drm_crtc_state *crtc_state;
 	int ret = 0;
 
-	if (flags & DRM_MODE_PAGE_FLIP_ASYNC)
-		return -EINVAL;
-
 	state = drm_atomic_state_alloc(plane->dev);
 	if (!state)
 		return -ENOMEM;
include/drm/drm_plane.h +8 -0
index db3bbde..86d8ffc 100644
@@ -122,6 +122,14 @@ struct drm_plane_state {
 	 */
 	bool visible;
 
+
+	/**
+	 * @pflip_flags:
+	 *
+	 * Flip related config options
+	 */
+	u32 pflip_flags;
+
 	struct drm_atomic_state *state;
 };
Signature · 8 lines
Regards,

Laurent Pinchart

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#13 lore raw

RE: [PATCH 1/4] drm/atomic: Save flip flags in drm_plane_state

Quoted 49 lines

-----Original Message-----
From: Laurent Pinchart [mailto:[email protected]]
Sent: Monday, January 16, 2017 5:18 PM
To: [email protected]
Cc: Grodzovsky, Andrey; [email protected]; amd- [email protected]; Deucher, Alexander; [email protected]
Subject: Re: [PATCH 1/4] drm/atomic: Save flip flags in drm_plane_state

Hi Andrey,

Thank you for the patch.

Quoted 36 lines · On Monday 16 Jan 2017 10:44:55 Andrey Grodzovsky wrote:

Allows using atomic flip helpers for drivers using ASYNC flip. Remove ASYNC_FLIP restriction in helpers and caches the page flip flags in drm_plane_state to be used in the low level drivers.

---
 drivers/gpu/drm/drm_atomic_helper.c | 10 +++-------
 include/drm/drm_plane.h             |  8 ++++++++
 2 files changed, 11 insertions(+), 7 deletions(-)
(quoted diff) +0 -0

b/drivers/gpu/drm/drm_atomic_helper.c index a4e5477..f83dc43 100644

drivers/gpu/drm/drm_atomic_helper.c +2 -1
@@ -2737,7 +2737,8 @@ static int page_flip_common(
 				struct drm_atomic_state *state,
 				struct drm_crtc *crtc,
 				struct drm_framebuffer *fb,
-				struct drm_pending_vblank_event *event)
+				struct drm_pending_vblank_event *event,
+				uint32_t flags)
 {
 	struct drm_plane *plane = crtc->primary;
 	struct drm_plane_state *plane_state; @@ -2754,6 +2755,7 @@ static

int page_flip_common(

drivers/gpu/drm/drm_atomic_helper.c +1 -0
 	if (IS_ERR(plane_state))
 		return PTR_ERR(plane_state);
 
+	plane_state->pflip_flags = flags;
 
 	ret = drm_atomic_set_crtc_for_plane(plane_state, crtc);
 	if (ret != 0)
@@ -2800,9 +2802,6 @@ int drm_atomic_helper_page_flip(struct drm_crtc
 *crtc, struct drm_atomic_state *state;
 	int ret = 0;
 
Quoted 8 lines

drivers/gpu/drm/drm_atomic_helper.c int drm_atomic_helper_page_flip(struct drm_crtc *crtc, · line 2805

Quoted 3 lines
drivers/gpu/drm/drm_atomic_helper.c +0 -3
-	if (flags & DRM_MODE_PAGE_FLIP_ASYNC)
-		return -EINVAL;
-

With this change all drivers using the helper will not reject that async flag, even if they don't implement support for async page flip. You need to either patch them all to reject the flag, or implement async page flip support for all of them (preferable in the helpers, and then remove the

Please check drm_mode_page_flip_ioctl, one of the checks in the beginning is

if ((page_flip->flags & DRM_MODE_PAGE_FLIP_ASYNC) && !dev->mode_config.async_page_flip)

		return -EINVAL;

We in DC explicitly set dev->mode_config.async_page_flip = true, any driver which is Not supporting ASYNC flip will fail the IOCTL at this point.
Same in drm_mode_atomic_ioctl

Quoted 7 lines

* Note that for now so called async page flips (i.e. updates which are not
* synchronized to vblank) are not supported, since the atomic interfaces
have
* no provisions for this yet.

comment).

Thanks, that a good catch, will remove.

Andrey

Quoted 36 lines
Quoted 22 lines
drivers/gpu/drm/drm_atomic_helper.c +0 -3
 	state = drm_atomic_state_alloc(plane->dev);
 	if (!state)
 		return -ENOMEM;
@@ -2865,9 +2864,6 @@ int drm_atomic_helper_page_flip_target(
 	struct drm_crtc_state *crtc_state;
 	int ret = 0;
 
-	if (flags & DRM_MODE_PAGE_FLIP_ASYNC)
-		return -EINVAL;
-
 	state = drm_atomic_state_alloc(plane->dev);
 	if (!state)
 		return -ENOMEM;
include/drm/drm_plane.h index renamed from include/drm/drm_plane.h +0 -0

db3bbde..86d8ffc 100644

include/drm/drm_plane.h +0 -0
@@ -122,6 +122,14 @@ struct drm_plane_state {
 	 */
 	bool visible;
 

include/drm/drm_plane.h struct drm_plane_state { · line 134

Quoted 10 lines
include/drm/drm_plane.h +8 -0
+
+	/**
+	 * @pflip_flags:
+	 *
+	 * Flip related config options
+	 */
+	u32 pflip_flags;
+
 	struct drm_atomic_state *state;
 };

--
Regards,

Quoted 1 line

Laurent Pinchart

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

A message that is not in the archive (sent off-list or not yet received): DM5PR12MB1147FF9E484370F1572CB9C3EA7C0-2J9CzHegvk9IwjGPM8RDJwdYzm3356FpvxpqHgZTriW3zl9H0oFU5g@public.gmane.org

#15 lore raw

Re: [PATCH 1/4] drm/atomic: Save flip flags in drm_plane_state

Hi Andrey,

Quoted 49 lines · Andrey
Quoted 39 lines · Laurent Pinchart
Quoted 38 lines · On Monday 16 Jan 2017 10:44:55 Andrey Grodzovsky wrote:

Allows using atomic flip helpers for drivers using ASYNC flip. Remove ASYNC_FLIP restriction in helpers and caches the page flip flags in drm_plane_state to be used in the low level drivers.

---

 drivers/gpu/drm/drm_atomic_helper.c | 10 +++-------
 include/drm/drm_plane.h             |  8 ++++++++
 2 files changed, 11 insertions(+), 7 deletions(-)
(quoted diff) +0 -0

b/drivers/gpu/drm/drm_atomic_helper.c index a4e5477..f83dc43 100644

drivers/gpu/drm/drm_atomic_helper.c +3 -1
@@ -2737,7 +2737,8 @@ static int page_flip_common(
 				struct drm_atomic_state *state,
 				struct drm_crtc *crtc,
 				struct drm_framebuffer *fb,
-				struct drm_pending_vblank_event *event)
+				struct drm_pending_vblank_event *event,
+				uint32_t flags)
 {
 	struct drm_plane *plane = crtc->primary;
 	struct drm_plane_state *plane_state;
@@ -2754,6 +2755,7 @@ static int page_flip_common(
 	if (IS_ERR(plane_state))
 		return PTR_ERR(plane_state);
 
+	plane_state->pflip_flags = flags;
 
 	ret = drm_atomic_set_crtc_for_plane(plane_state, crtc);
 	if (ret != 0)
@@ -2800,9 +2802,6 @@ int drm_atomic_helper_page_flip(struct drm_crtc
 *crtc,
 	struct drm_atomic_state *state;
 	int ret = 0;
 
Quoted 8 lines

drivers/gpu/drm/drm_atomic_helper.c int drm_atomic_helper_page_flip(struct drm_crtc *crtc, · line 2805

Quoted 3 lines
drivers/gpu/drm/drm_atomic_helper.c +0 -3
-	if (flags & DRM_MODE_PAGE_FLIP_ASYNC)
-		return -EINVAL;
-

With this change all drivers using the helper will not reject that async flag, even if they don't implement support for async page flip. You need to either patch them all to reject the flag, or implement async page flip support for all of them (preferable in the helpers, and then remove the

Quoted 4 lines

Please check drm_mode_page_flip_ioctl, one of the checks in the beginning is

if ((page_flip->flags & DRM_MODE_PAGE_FLIP_ASYNC) && !dev->mode_config.async_page_flip) return -EINVAL;

I think you're right. Sorry for the noise.

Quoted 13 lines

We in DC explicitly set dev->mode_config.async_page_flip = true, any driver which is Not supporting ASYNC flip will fail the IOCTL at this point. Same in drm_mode_atomic_ioctl

Quoted 7 lines

* Note that for now so called async page flips (i.e. updates which are not
* synchronized to vblank) are not supported, since the atomic interfaces
have
* no provisions for this yet.

comment).

Thanks, that a good catch, will remove.

Signature · 8 lines
Regards,

Laurent Pinchart

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#16 lore raw

Re: [PATCH 1/4] drm/atomic: Save flip flags in drm_plane_state

In reply to Andrey Grodzovsky
Quoted 27 lines · Andrey Grodzovsky

Allows using atomic flip helpers for drivers
using ASYNC flip.
Remove ASYNC_FLIP restriction in helpers and
caches the page flip flags in drm_plane_state
to be used in the low level drivers.

---
 drivers/gpu/drm/drm_atomic_helper.c | 10 +++-------
 include/drm/drm_plane.h             |  8 ++++++++
 2 files changed, 11 insertions(+), 7 deletions(-)
drivers/gpu/drm/drm_atomic_helper.c +2 -1
index a4e5477..f83dc43 100644
@@ -2737,7 +2737,8 @@ static int page_flip_common(
 				struct drm_atomic_state *state,
 				struct drm_crtc *crtc,
 				struct drm_framebuffer *fb,
-				struct drm_pending_vblank_event *event)
+				struct drm_pending_vblank_event *event,
+				uint32_t flags)
 {
 	struct drm_plane *plane = crtc->primary;
 	struct drm_plane_state *plane_state;

drivers/gpu/drm/drm_atomic_helper.c static int page_flip_common( · line 2758

Quoted 4 lines
drivers/gpu/drm/drm_atomic_helper.c +1 -0
@@ -2754,6 +2755,7 @@ static int page_flip_common(
 	if (IS_ERR(plane_state))
 		return PTR_ERR(plane_state);
 
+	plane_state->pflip_flags = flags;

"pflip" looks off. Better just spell it out.

These flags need to be reset in duplicate_state otherwise they leak into subsequent operations. This is why I don't really like the concept of "state" containing flags and stuff that are not really part of the state but rather part of the operation of moving from the old state to the new state. But maybe we don't have a better place for this sort of stuff? I have suggested at some point that we might rename drm_atomic_state to drm_atomic_transaction or something. Stuffing the flags (or just a bool perhaps?) in there might be less confusing.

Quoted 49 lines
drivers/gpu/drm/drm_atomic_helper.c +0 -6
 	ret = drm_atomic_set_crtc_for_plane(plane_state, crtc);
 	if (ret != 0)
@@ -2800,9 +2802,6 @@ int drm_atomic_helper_page_flip(struct drm_crtc *crtc,
 	struct drm_atomic_state *state;
 	int ret = 0;
 
-	if (flags & DRM_MODE_PAGE_FLIP_ASYNC)
-		return -EINVAL;
-
 	state = drm_atomic_state_alloc(plane->dev);
 	if (!state)
 		return -ENOMEM;
@@ -2865,9 +2864,6 @@ int drm_atomic_helper_page_flip_target(
 	struct drm_crtc_state *crtc_state;
 	int ret = 0;
 
-	if (flags & DRM_MODE_PAGE_FLIP_ASYNC)
-		return -EINVAL;
-
 	state = drm_atomic_state_alloc(plane->dev);
 	if (!state)
 		return -ENOMEM;
include/drm/drm_plane.h +8 -0
index db3bbde..86d8ffc 100644
@@ -122,6 +122,14 @@ struct drm_plane_state {
 	 */
 	bool visible;
 
+
+	/**
+	 * @pflip_flags:
+	 *
+	 * Flip related config options
+	 */
+	u32 pflip_flags;
+
 	struct drm_atomic_state *state;
 };
 
Signature · 6 lines
1.9.1

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Signature · 6 lines
Ville Syrjälä
Intel OTC
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

A message that is not in the archive (sent off-list or not yet received): 1484581498-32309-1-git-send-email-Andrey.Grodzovsky-5C7GfCeVMHo@public.gmane.org

#18 lore raw

PATCH 2/4 drm/amdgpu: Remove flip_flag from amdgpu_crtc

Follwing introduction of pflip_flags in drm_plane_state this is not needed anymore.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h | 1 -
 1 file changed, 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h +0 -1
index 4c0a86e..3ff3c14 100644
@@ -443,7 +443,6 @@ struct amdgpu_crtc {
 	enum amdgpu_interrupt_state vsync_timer_enabled;
 
 	int otg_inst;
-	uint32_t flip_flags;
 	/* After Set Mode target will be non-NULL */
 	struct dc_target *target;
 };
Signature · 6 lines
1.9.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#19 lore raw

Re: [PATCH 2/4] drm/amdgpu: Remove flip_flag from amdgpu_crtc

Hi Andrey,

Thank you for the patch.

Quoted 13 lines · On Monday 16 Jan 2017 10:44:56 Andrey Grodzovsky wrote:

Follwing introduction of pflip_flags in drm_plane_state this is not needed anymore.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h | 1 -
 1 file changed, 1 deletion(-)
(quoted diff) +0 -0

b/drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h index 4c0a86e..3ff3c14 100644

drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h struct amdgpu_crtc { · line 446

Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h +0 -1
@@ -443,7 +443,6 @@ struct amdgpu_crtc {
 	enum amdgpu_interrupt_state vsync_timer_enabled;
 
 	int otg_inst;
-	uint32_t flip_flags;

This breaks compilation of the amdgpu driver. You should squash this patch with 3/4.

Quoted 3 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h +0 -0
 	/* After Set Mode target will be non-NULL */
 	struct dc_target *target;
 };
Signature · 8 lines
Regards,

Laurent Pinchart

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#20 lore raw

PATCH 3/4 drm/amd/display: Switch to using atomic_helper for flip.

In reply to a message not in the archive
Change-Id Iad3e0b9b3546e4e4dc79be9233daf4fe4dba83e0
---
 .../drm/amd/display/amdgpu_dm/amdgpu_dm_types.c    | 92 ++--------------------
 1 file changed, 6 insertions(+), 86 deletions(-)
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +6 -86
index a443b70..d4664bf 100644
@@ -1060,83 +1060,6 @@ static int dm_crtc_funcs_atomic_set_property(
 	return 0;
 }
 
-
-static int amdgpu_atomic_helper_page_flip(struct drm_crtc *crtc,
-				struct drm_framebuffer *fb,
-				struct drm_pending_vblank_event *event,
-				uint32_t flags)
-{
-	struct drm_plane *plane = crtc->primary;
-	struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
-	struct drm_atomic_state *state;
-	struct drm_plane_state *plane_state;
-	struct drm_crtc_state *crtc_state;
-	int ret = 0;
-
-	state = drm_atomic_state_alloc(plane->dev);
-	if (!state)
-		return -ENOMEM;
-
-	ret = drm_crtc_vblank_get(crtc);
-	if (ret)
-		return ret;
-
-	state->acquire_ctx = drm_modeset_legacy_acquire_ctx(crtc);
-retry:
-	crtc_state = drm_atomic_get_crtc_state(state, crtc);
-	if (IS_ERR(crtc_state)) {
-		ret = PTR_ERR(crtc_state);
-		goto fail;
-	}
-	crtc_state->event = event;
-
-	plane_state = drm_atomic_get_plane_state(state, plane);
-	if (IS_ERR(plane_state)) {
-		ret = PTR_ERR(plane_state);
-		goto fail;
-	}
-
-	ret = drm_atomic_set_crtc_for_plane(plane_state, crtc);
-	if (ret != 0)
-		goto fail;
-	drm_atomic_set_fb_for_plane(plane_state, fb);
-
-	/* Make sure we don't accidentally do a full modeset. */
-	state->allow_modeset = false;
-	if (!crtc_state->active) {
-		DRM_DEBUG_ATOMIC("[CRTC:%d] disabled, rejecting legacy flip\n",
-				 crtc->base.id);
-		ret = -EINVAL;
-		goto fail;
-	}
-	acrtc->flip_flags = flags;
-
-	ret = drm_atomic_nonblocking_commit(state);
-
-fail:
-	if (ret == -EDEADLK)
-		goto backoff;
-
-	if (ret)
-		drm_crtc_vblank_put(crtc);
-
-	drm_atomic_state_put(state);
-
-	return ret;
-backoff:
-	drm_atomic_state_clear(state);
-	drm_atomic_legacy_backoff(state);
-
-	/*
-	 * Someone might have exchanged the framebuffer while we dropped locks
-	 * in the backoff code. We need to fix up the fb refcount tracking the
-	 * core does for us.
-	 */
-	plane->old_fb = plane->fb;
-
-	goto retry;
-}
-
 /* Implemented only the options currently availible for the driver */
 static const struct drm_crtc_funcs amdgpu_dm_crtc_funcs = {
 	.reset = drm_atomic_helper_crtc_reset,
@@ -1145,7 +1068,7 @@ static int amdgpu_atomic_helper_page_flip(struct drm_crtc *crtc,
 	.destroy = amdgpu_dm_crtc_destroy,
 	.gamma_set = amdgpu_dm_atomic_crtc_gamma_set,
 	.set_config = drm_atomic_helper_set_config,
-	.page_flip = amdgpu_atomic_helper_page_flip,
+	.page_flip_target = drm_atomic_helper_page_flip_target,
 	.atomic_duplicate_state = drm_atomic_helper_crtc_duplicate_state,
 	.atomic_destroy_state = drm_atomic_helper_crtc_destroy_state,
 	.atomic_set_property = dm_crtc_funcs_atomic_set_property
@@ -1679,7 +1602,7 @@ static bool page_flip_needed(
 				    sizeof(old_state_tmp)) == 0 ? true:false;
 	if (new_state->crtc && page_flip_required == false) {
 		acrtc_new = to_amdgpu_crtc(new_state->crtc);
-		if (acrtc_new->flip_flags & DRM_MODE_PAGE_FLIP_ASYNC)
+		if (new_state->pflip_flags & DRM_MODE_PAGE_FLIP_ASYNC)
 			page_flip_required = true;
 	}
 	return page_flip_required;
@@ -2760,7 +2683,6 @@ int amdgpu_dm_atomic_commit(
 	for_each_plane_in_state(state, plane, old_plane_state, i) {
 		struct drm_plane_state *plane_state = plane->state;
 		struct drm_crtc *crtc = plane_state->crtc;
-		struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
 		struct drm_framebuffer *fb = plane_state->fb;
 
 		if (!fb || !crtc || !crtc->state->planes_changed ||
@@ -2771,10 +2693,9 @@ int amdgpu_dm_atomic_commit(
 			ret = amdgpu_crtc_page_flip_target(crtc,
 							   fb,
 							   crtc->state->event,
-							   acrtc->flip_flags,
-							   drm_crtc_vblank_count(crtc));
-			/*clean up the flags for next usage*/
-			acrtc->flip_flags = 0;
+							   plane_state->pflip_flags,
+							   crtc->state->target_vblank);
+
 			if (ret)
 				break;
 		}
@@ -3143,8 +3064,7 @@ int amdgpu_dm_atomic_check(struct drm_device *dev,
 			 * 1. This commit is not a page flip.
 			 * 2. This commit is a page flip, and targets are created.
 			 */
-			if (!page_flip_needed(plane_state, old_plane_state,
-					      true) ||
+			if (!page_flip_needed(plane_state, old_plane_state, true) ||
 					action == DM_COMMIT_ACTION_DPMS_ON ||
 					action == DM_COMMIT_ACTION_SET) {
 				struct dc_surface *surface;
Signature · 6 lines
1.9.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#21 lore raw

Re: [PATCH 3/4] drm/amd/display: Switch to using atomic_helper for flip.

Hi Andrey,

Thank you for the patch.

Quoted 23 lines · On Monday 16 Jan 2017 10:44:57 Andrey Grodzovsky wrote:
Change-Id Iad3e0b9b3546e4e4dc79be9233daf4fe4dba83e0
---
 .../drm/amd/display/amdgpu_dm/amdgpu_dm_types.c    | 92 ++----------------
 1 file changed, 6 insertions(+), 86 deletions(-)
(quoted diff) +0 -0

b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c index a443b70..d4664bf 100644

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +0 -8
@@ -1060,83 +1060,6 @@ static int dm_crtc_funcs_atomic_set_property(
 	return 0;
 }
 
-
-static int amdgpu_atomic_helper_page_flip(struct drm_crtc *crtc,
-				struct drm_framebuffer *fb,
-				struct drm_pending_vblank_event *event,
-				uint32_t flags)
-{
-	struct drm_plane *plane = crtc->primary;
-	struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c static int dm_crtc_funcs_atomic_set_property( · line 1080

Quoted 10 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +0 -10
-	struct drm_atomic_state *state;
-	struct drm_plane_state *plane_state;
-	struct drm_crtc_state *crtc_state;
-	int ret = 0;
-
-	state = drm_atomic_state_alloc(plane->dev);
-	if (!state)
-		return -ENOMEM;
-
-	ret = drm_crtc_vblank_get(crtc);

The DRM core's atomic page flip helper doesn't get/put vblank. Have you double-checked that removing them isn't a problem ?

Quoted 17 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +0 -17
-	if (ret)
-		return ret;
-
-	state->acquire_ctx = drm_modeset_legacy_acquire_ctx(crtc);
-retry:
-	crtc_state = drm_atomic_get_crtc_state(state, crtc);
-	if (IS_ERR(crtc_state)) {
-		ret = PTR_ERR(crtc_state);
-		goto fail;
-	}
-	crtc_state->event = event;
-
-	plane_state = drm_atomic_get_plane_state(state, plane);
-	if (IS_ERR(plane_state)) {
-		ret = PTR_ERR(plane_state);
-		goto fail;
-	}

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c static int dm_crtc_funcs_atomic_set_property(

Quoted 10 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +0 -10
-
-	ret = drm_atomic_set_crtc_for_plane(plane_state, crtc);
-	if (ret != 0)
-		goto fail;
-	drm_atomic_set_fb_for_plane(plane_state, fb);
-
-	/* Make sure we don't accidentally do a full modeset. */
-	state->allow_modeset = false;
-	if (!crtc_state->active) {
-		DRM_DEBUG_ATOMIC("[CRTC:%d] disabled, rejecting legacy 

flip\n",

Quoted 62 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +2 -35
-				 crtc->base.id);
-		ret = -EINVAL;
-		goto fail;
-	}
-	acrtc->flip_flags = flags;
-
-	ret = drm_atomic_nonblocking_commit(state);
-
-fail:
-	if (ret == -EDEADLK)
-		goto backoff;
-
-	if (ret)
-		drm_crtc_vblank_put(crtc);
-
-	drm_atomic_state_put(state);
-
-	return ret;
-backoff:
-	drm_atomic_state_clear(state);
-	drm_atomic_legacy_backoff(state);
-
-	/*
-	 * Someone might have exchanged the framebuffer while we dropped locks
-	 * in the backoff code. We need to fix up the fb refcount tracking the
-	 * core does for us.
-	 */
-	plane->old_fb = plane->fb;
-
-	goto retry;
-}
-
 /* Implemented only the options currently availible for the driver */
 static const struct drm_crtc_funcs amdgpu_dm_crtc_funcs = {
 	.reset = drm_atomic_helper_crtc_reset,
@@ -1145,7 +1068,7 @@ static int amdgpu_atomic_helper_page_flip(struct
 drm_crtc *crtc, .destroy = amdgpu_dm_crtc_destroy,
 	.gamma_set = amdgpu_dm_atomic_crtc_gamma_set,
 	.set_config = drm_atomic_helper_set_config,
-	.page_flip = amdgpu_atomic_helper_page_flip,
+	.page_flip_target = drm_atomic_helper_page_flip_target,
 	.atomic_duplicate_state = drm_atomic_helper_crtc_duplicate_state,
 	.atomic_destroy_state = drm_atomic_helper_crtc_destroy_state,
 	.atomic_set_property = dm_crtc_funcs_atomic_set_property
@@ -1679,7 +1602,7 @@ static bool page_flip_needed(
 				    sizeof(old_state_tmp)) == 0 ? true:false;
 	if (new_state->crtc && page_flip_required == false) {
 		acrtc_new = to_amdgpu_crtc(new_state->crtc);
-		if (acrtc_new->flip_flags & DRM_MODE_PAGE_FLIP_ASYNC)
+		if (new_state->pflip_flags & DRM_MODE_PAGE_FLIP_ASYNC)
 			page_flip_required = true;
 	}
 	return page_flip_required;
@@ -2760,7 +2683,6 @@ int amdgpu_dm_atomic_commit(
 	for_each_plane_in_state(state, plane, old_plane_state, i) {
 		struct drm_plane_state *plane_state = plane->state;
 		struct drm_crtc *crtc = plane_state->crtc;
-		struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
 		struct drm_framebuffer *fb = plane_state->fb;
 
 		if (!fb || !crtc || !crtc->state->planes_changed ||

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c int amdgpu_dm_atomic_commit( · line 2775

Quoted 5 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +0 -2
@@ -2771,10 +2693,9 @@ int amdgpu_dm_atomic_commit(
 			ret = amdgpu_crtc_page_flip_target(crtc,
 							   fb,
 							   crtc->state->event,
-							   acrtc->flip_flags,
-							   

drm_crtc_vblank_count(crtc));

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c int amdgpu_dm_atomic_check(struct drm_device *dev, · line 3065

Quoted 13 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +3 -2
-			/*clean up the flags for next usage*/
-			acrtc->flip_flags = 0;
+							   plane_state-
 pflip_flags,
+							   crtc->state-
 target_vblank);
+
 			if (ret)
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +0 -0
 				break;
 		}
@@ -3143,8 +3064,7 @@ int amdgpu_dm_atomic_check(struct drm_device *dev,
 			 * 1. This commit is not a page flip.
 			 * 2. This commit is a page flip, and targets are 

created.

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c int amdgpu_dm_atomic_check(struct drm_device *dev,

Quoted 4 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +1 -2
 			 */
-			if (!page_flip_needed(plane_state, old_plane_state,
-					      true) ||
+			if (!page_flip_needed(plane_state, old_plane_state, 

true) ||

This seems to be an unrelated change.

Quoted 3 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +0 -0
 					action == DM_COMMIT_ACTION_DPMS_ON ||
 					action == DM_COMMIT_ACTION_SET) {
 				struct dc_surface *surface;
Signature · 8 lines
Regards,

Laurent Pinchart

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#22 lore raw

Re: [PATCH 3/4] drm/amd/display: Switch to using atomic_helper for flip.

Quoted 29 lines · Laurent Pinchart
Quoted 28 lines · On Monday 16 Jan 2017 10:44:57 Andrey Grodzovsky wrote:
Change-Id Iad3e0b9b3546e4e4dc79be9233daf4fe4dba83e0
---
 .../drm/amd/display/amdgpu_dm/amdgpu_dm_types.c    | 92 ++----------------
 1 file changed, 6 insertions(+), 86 deletions(-)
(quoted diff) +0 -0

b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c index a443b70..d4664bf 100644

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +0 -13
@@ -1060,83 +1060,6 @@ static int dm_crtc_funcs_atomic_set_property(
 	return 0;
 }
 
-
-static int amdgpu_atomic_helper_page_flip(struct drm_crtc *crtc,
-				struct drm_framebuffer *fb,
-				struct drm_pending_vblank_event *event,
-				uint32_t flags)
-{
-	struct drm_plane *plane = crtc->primary;
-	struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
-	struct drm_atomic_state *state;
-	struct drm_plane_state *plane_state;
-	struct drm_crtc_state *crtc_state;
-	int ret = 0;
-
Quoted 8 lines

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c static int dm_crtc_funcs_atomic_set_property( · line 1080

Quoted 5 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +0 -5
-	state = drm_atomic_state_alloc(plane->dev);
-	if (!state)
-		return -ENOMEM;
-
-	ret = drm_crtc_vblank_get(crtc);

The DRM core's atomic page flip helper doesn't get/put vblank. Have you double-checked that removing them isn't a problem ?

This patch makes the amdgpu DM code use the page_flip_target hook. drm_mode_page_flip_ioctl calls drm_crtc_vblank_get before the page_flip_target hook.

You're right though that the fact that drm_atomic_helper_page_flip doesn't call drm_crtc_vblank_get is a bit alarming. From the DRM_IOCTL_MODE_PAGE_FLIP POV, drm_crtc_vblank_get must be called when userspace calls the ioctl (either by drm_mode_page_flip_ioctl or the page_flip hook implementation), and drm_crtc_vblank_put must be called when the flip completes and the event is sent to userspace. How is this supposed to be handled with atomic?

Andrey, did you check via code audit and/or testing that the vblank reference count is still balanced after this change?

Quoted 11 lines

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c int amdgpu_dm_atomic_check(struct drm_device *dev, · line 3065

Quoted 3 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +0 -0
@@ -3143,8 +3064,7 @@ int amdgpu_dm_atomic_check(struct drm_device *dev,
 			 * 1. This commit is not a page flip.
 			 * 2. This commit is a page flip, and targets are 

created.

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c int amdgpu_dm_atomic_check(struct drm_device *dev,

Quoted 4 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +1 -2
 			 */
-			if (!page_flip_needed(plane_state, old_plane_state,
-					      true) ||
+			if (!page_flip_needed(plane_state, old_plane_state, 

true) ||

This seems to be an unrelated change.

Yeah, such whitespace-only changes should be dropped.

Signature · 6 lines
Earthling Michel Dänzer               |               http://www.amd.com
Libre software enthusiast             |             Mesa and X developer
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#23 lore raw

Re: [PATCH 3/4] drm/amd/display: Switch to using atomic_helper for flip.

Hi Michel,

Quoted 47 lines · On Wednesday 18 Jan 2017 10:50:01 Michel Dänzer wrote:
Quoted 33 lines · Laurent Pinchart
Quoted 32 lines · On Monday 16 Jan 2017 10:44:57 Andrey Grodzovsky wrote:
Change-Id Iad3e0b9b3546e4e4dc79be9233daf4fe4dba83e0

---

 .../drm/amd/display/amdgpu_dm/amdgpu_dm_types.c    | 92
 ++----------------
 1 file changed, 6 insertions(+), 86 deletions(-)
(quoted diff) +0 -0

b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c index a443b70..d4664bf 100644

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +0 -13
@@ -1060,83 +1060,6 @@ static int dm_crtc_funcs_atomic_set_property(
 
 	return 0;
 
 }
 
-
-static int amdgpu_atomic_helper_page_flip(struct drm_crtc *crtc,
-				struct drm_framebuffer *fb,
-				struct drm_pending_vblank_event *event,
-				uint32_t flags)
-{
-	struct drm_plane *plane = crtc->primary;
-	struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
-	struct drm_atomic_state *state;
-	struct drm_plane_state *plane_state;
-	struct drm_crtc_state *crtc_state;
-	int ret = 0;
-
Quoted 8 lines

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c static int dm_crtc_funcs_atomic_set_property( · line 1080

Quoted 5 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +0 -5
-	state = drm_atomic_state_alloc(plane->dev);
-	if (!state)
-		return -ENOMEM;
-
-	ret = drm_crtc_vblank_get(crtc);

The DRM core's atomic page flip helper doesn't get/put vblank. Have you double-checked that removing them isn't a problem ?

This patch makes the amdgpu DM code use the page_flip_target hook. drm_mode_page_flip_ioctl calls drm_crtc_vblank_get before the page_flip_target hook.

Quoted 7 lines

You're right though that the fact that drm_atomic_helper_page_flip doesn't call drm_crtc_vblank_get is a bit alarming. From the DRM_IOCTL_MODE_PAGE_FLIP POV, drm_crtc_vblank_get must be called when userspace calls the ioctl (either by drm_mode_page_flip_ioctl or the page_flip hook implementation), and drm_crtc_vblank_put must be called when the flip completes and the event is sent to userspace. How is this supposed to be handled with atomic?

I'll let Daniel comment on that.

Quoted 21 lines

Andrey, did you check via code audit and/or testing that the vblank reference count is still balanced after this change?

Quoted 9 lines

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c int amdgpu_dm_atomic_check(struct drm_device *dev, · line 3066

Quoted 4 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +0 -0
@@ -3143,8 +3064,7 @@ int amdgpu_dm_atomic_check(struct drm_device *dev,
 
 			 * 1. This commit is not a page flip.
 			 * 2. This commit is a page flip, and targets are

created.

Quoted 2 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +0 -0
 			 */
 
Quoted 7 lines

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c int amdgpu_dm_atomic_check(struct drm_device *dev,

Quoted 3 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +1 -2
-			if (!page_flip_needed(plane_state, old_plane_state,
-					      true) ||
+			if (!page_flip_needed(plane_state, old_plane_state,

true) ||

This seems to be an unrelated change.

Yeah, such whitespace-only changes should be dropped.

Signature · 8 lines
Regards,

Laurent Pinchart

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#24 lore raw

RE: [PATCH 3/4] drm/amd/display: Switch to using atomic_helper for flip.

In reply to Michel Dänzer
Quoted 56 lines

-----Original Message-----
From: Michel Dänzer [mailto:[email protected]]
Sent: Tuesday, January 17, 2017 8:50 PM
To: Laurent Pinchart
Cc: [email protected]; Grodzovsky, Andrey; [email protected]; [email protected]; [email protected]
Subject: Re: [PATCH 3/4] drm/amd/display: Switch to using atomic_helper for flip.

Quoted 5 lines · Laurent Pinchart
Quoted 4 lines · On Monday 16 Jan 2017 10:44:57 Andrey Grodzovsky wrote:
Change-Id Iad3e0b9b3546e4e4dc79be9233daf4fe4dba83e0
---
 .../drm/amd/display/amdgpu_dm/amdgpu_dm_types.c    | 92 ++---------

-------

Quoted 3 lines
Quoted 3 lines
 1 file changed, 6 insertions(+), 86 deletions(-)

diff --git

a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c

Quoted 1 line
Quoted 1 line

b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c

index

Quoted 20 lines
Quoted 20 lines

a443b70..d4664bf 100644

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +0 -13
@@ -1060,83 +1060,6 @@ static int dm_crtc_funcs_atomic_set_property(
 	return 0;
 }
 
-
-static int amdgpu_atomic_helper_page_flip(struct drm_crtc *crtc,
-				struct drm_framebuffer *fb,
-				struct drm_pending_vblank_event *event,
-				uint32_t flags)
-{
-	struct drm_plane *plane = crtc->primary;
-	struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
-	struct drm_atomic_state *state;
-	struct drm_plane_state *plane_state;
-	struct drm_crtc_state *crtc_state;
-	int ret = 0;
-
Quoted 8 lines

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c static int dm_crtc_funcs_atomic_set_property( · line 1080

Quoted 5 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +0 -5
-	state = drm_atomic_state_alloc(plane->dev);
-	if (!state)
-		return -ENOMEM;
-
-	ret = drm_crtc_vblank_get(crtc);

The DRM core's atomic page flip helper doesn't get/put vblank. Have you double-checked that removing them isn't a problem ?

This patch makes the amdgpu DM code use the page_flip_target hook. drm_mode_page_flip_ioctl calls drm_crtc_vblank_get before the page_flip_target hook.

Quoted 7 lines

You're right though that the fact that drm_atomic_helper_page_flip doesn't call drm_crtc_vblank_get is a bit alarming. From the DRM_IOCTL_MODE_PAGE_FLIP POV, drm_crtc_vblank_get must be called when userspace calls the ioctl (either by drm_mode_page_flip_ioctl or the page_flip hook implementation), and drm_crtc_vblank_put must be called when the flip completes and the event is sent to userspace. How is this supposed to be handled with atomic?

First let me ask you - why we have to enable vlbank as part of pflip, I understand why it has to be enbbled in WAIT_FOR_VBLANK IOCTL but not sure about PFLIP IOCTL.

IMHO in atomic as before in legacy page_flip, it's up to the driver
implementation in atomic_commit hook to manage vblank ISR on,off.

Quoted 4 lines

Andrey, did you check via code audit and/or testing that the vblank reference count is still balanced after this change?

With the page_flip_target yes, if I switch back to page_flip hook then vblank ISR never gets enabled on FLIP IPCTL and then I see warning in DAL's pflip done IRQ handler from vblank_put which is obvious, which BTW didn't impact the flips, I still was able to run glxgears in vsync mode.

Quoted 12 lines
Quoted 2 lines

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c int amdgpu_dm_atomic_check(struct

Quoted 2 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +0 -0
@@ -3143,8 +3064,7 @@ int amdgpu_dm_atomic_check(struct

drm_device *dev,

Quoted 9 lines

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c int amdgpu_dm_atomic_check(struct · line 3065

Quoted 2 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +0 -0
 			 * 1. This commit is not a page flip.
 			 * 2. This commit is a page flip, and targets are

created.

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c int amdgpu_dm_atomic_check(struct drm_device *dev,

Quoted 4 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +1 -2
 			 */
-			if (!page_flip_needed(plane_state, old_plane_state,
-					      true) ||
+			if (!page_flip_needed(plane_state, old_plane_state,

true) ||

Quoted 3 lines
Quoted 1 line

This seems to be an unrelated change.

Yeah, such whitespace-only changes should be dropped.

Will remove this.

Quoted 5 lines

--
Earthling Michel Dänzer | http://www.amd.com
Libre software enthusiast | Mesa and X developer

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

A message that is not in the archive (sent off-list or not yet received): DM5PR12MB1147CFCD90203061E0F2EB95EA7F0-2J9CzHegvk9IwjGPM8RDJwdYzm3356FpvxpqHgZTriW3zl9H0oFU5g@public.gmane.org

#26 lore raw

Re: [PATCH 3/4] drm/amd/display: Switch to using atomic_helper for flip.

Quoted 55 lines · Andrey
Quoted 47 lines

From: Michel Dänzer [mailto:[email protected]]

Quoted 5 lines · Laurent Pinchart
Quoted 4 lines · On Monday 16 Jan 2017 10:44:57 Andrey Grodzovsky wrote:
Change-Id Iad3e0b9b3546e4e4dc79be9233daf4fe4dba83e0
---
 .../drm/amd/display/amdgpu_dm/amdgpu_dm_types.c    | 92 ++---------

-------

Quoted 3 lines
Quoted 3 lines
 1 file changed, 6 insertions(+), 86 deletions(-)

diff --git

a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c

Quoted 1 line
Quoted 1 line

b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c

index

Quoted 20 lines
Quoted 20 lines

a443b70..d4664bf 100644

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +0 -13
@@ -1060,83 +1060,6 @@ static int dm_crtc_funcs_atomic_set_property(
 	return 0;
 }
 
-
-static int amdgpu_atomic_helper_page_flip(struct drm_crtc *crtc,
-				struct drm_framebuffer *fb,
-				struct drm_pending_vblank_event *event,
-				uint32_t flags)
-{
-	struct drm_plane *plane = crtc->primary;
-	struct amdgpu_crtc *acrtc = to_amdgpu_crtc(crtc);
-	struct drm_atomic_state *state;
-	struct drm_plane_state *plane_state;
-	struct drm_crtc_state *crtc_state;
-	int ret = 0;
-
Quoted 8 lines

drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c static int dm_crtc_funcs_atomic_set_property( · line 1080

Quoted 5 lines
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +0 -5
-	state = drm_atomic_state_alloc(plane->dev);
-	if (!state)
-		return -ENOMEM;
-
-	ret = drm_crtc_vblank_get(crtc);

The DRM core's atomic page flip helper doesn't get/put vblank. Have you double-checked that removing them isn't a problem ?

This patch makes the amdgpu DM code use the page_flip_target hook. drm_mode_page_flip_ioctl calls drm_crtc_vblank_get before the page_flip_target hook.

Quoted 7 lines

You're right though that the fact that drm_atomic_helper_page_flip doesn't call drm_crtc_vblank_get is a bit alarming. From the DRM_IOCTL_MODE_PAGE_FLIP POV, drm_crtc_vblank_get must be called when userspace calls the ioctl (either by drm_mode_page_flip_ioctl or the page_flip hook implementation), and drm_crtc_vblank_put must be called when the flip completes and the event is sent to userspace. How is this supposed to be handled with atomic?

Quoted 3 lines

First let me ask you - why we have to enable vlbank as part of pflip, I understand why it has to be enbbled in WAIT_FOR_VBLANK IOCTL but not sure about PFLIP IOCTL.

It's required for the vblank sequence counter to be accurate in the page flip completion event sent to userspace.

Quoted 2 lines

IMHO in atomic as before in legacy page_flip, it's up to the driver
implementation in atomic_commit hook to manage vblank ISR on,off.

When using the page_flip hook, it's all up to the implementation of that hook. When using the page_flip_target hook, drm_mode_page_flip_ioctl calls vblank_get, so the hook implementation must make sure that a matching vblank_put call is made when the flip completes.

Quoted 6 lines
Quoted 3 lines

Andrey, did you check via code audit and/or testing that the vblank reference count is still balanced after this change?

With the page_flip_target yes, if I switch back to page_flip hook then vblank ISR never gets enabled on FLIP IPCTL and then I see warning in DAL's pflip done IRQ handler from vblank_put which is obvious,

For drivers using the atomic_helpers for flip, maybe there should be (or might be already) a corresponding helper which deals with things like this when the flip completes, so drivers don't have to call vblank_put and such.

Anyway, it sounds like the vblank_get/puts are balanced with this patch for now.

Quoted 2 lines

which BTW didn't impact the flips, I still was able to run glxgears in vsync mode.

I don't think glxgears relies on the vblank sequence numbers being accurate.

Signature · 6 lines
Earthling Michel Dänzer               |               http://www.amd.com
Libre software enthusiast             |             Mesa and X developer
_______________________________________________
Nouveau mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/nouveau
#27 lore raw

PATCH v3 2/3 drm/amd/display: Switch to using atomic_helper for flip.

In reply to Michel Dänzer

v3: Remove white space change.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h           |  1 -
 .../drm/amd/display/amdgpu_dm/amdgpu_dm_types.c    | 89 ++--------------------
 2 files changed, 5 insertions(+), 85 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h +0 -1
index 4c0a86e..3ff3c14 100644
@@ -443,7 +443,6 @@ struct amdgpu_crtc {
 	enum amdgpu_interrupt_state vsync_timer_enabled;
 
 	int otg_inst;
-	uint32_t flip_flags;
 	/* After Set Mode target will be non-NULL */
 	struct dc_target *target;
 };
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_types.c +5 -84

119 lines not on this page · raw message

Signature · 6 lines
1.9.1

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#28 lore raw

PATCH 4/4 drm/nouveau/kms/nv50: Switch to using atomic helper for flip.

In reply to Andrey Grodzovsky
---
 drivers/gpu/drm/nouveau/nv50_display.c | 77 +++-------------------------------
 1 file changed, 5 insertions(+), 72 deletions(-)
drivers/gpu/drm/nouveau/nv50_display.c +5 -72
index 2c2c645..419e00c 100644
@@ -846,6 +846,10 @@ struct nv50_wndw_func {
 	asyw->image.w = fb->base.width;
 	asyw->image.h = fb->base.height;
 	asyw->image.kind = (fb->nvbo->tile_flags & 0x0000ff00) >> 8;
+
+	asyw->interval =
+			asyw->state.pflip_flags & DRM_MODE_PAGE_FLIP_ASYNC ? 0 : 1;
+
 	if (asyw->image.kind) {
 		asyw->image.layout = 0;
 		if (drm->device.info.chipset >= 0xc0)
@@ -2221,77 +2225,6 @@ struct nv50_base {
 	.atomic_check = nv50_head_atomic_check,
 };
 
-/* This is identical to the version in the atomic helpers, except that
- * it supports non-vblanked ("async") page flips.
- */
-static int
-nv50_head_page_flip(struct drm_crtc *crtc, struct drm_framebuffer *fb,
-		    struct drm_pending_vblank_event *event, u32 flags)
-{
-	struct drm_plane *plane = crtc->primary;
-	struct drm_atomic_state *state;
-	struct drm_plane_state *plane_state;
-	struct drm_crtc_state *crtc_state;
-	int ret = 0;
-
-	state = drm_atomic_state_alloc(plane->dev);
-	if (!state)
-		return -ENOMEM;
-
-	state->acquire_ctx = drm_modeset_legacy_acquire_ctx(crtc);
-retry:
-	crtc_state = drm_atomic_get_crtc_state(state, crtc);
-	if (IS_ERR(crtc_state)) {
-		ret = PTR_ERR(crtc_state);
-		goto fail;
-	}
-	crtc_state->event = event;
-
-	plane_state = drm_atomic_get_plane_state(state, plane);
-	if (IS_ERR(plane_state)) {
-		ret = PTR_ERR(plane_state);
-		goto fail;
-	}
-
-	ret = drm_atomic_set_crtc_for_plane(plane_state, crtc);
-	if (ret != 0)
-		goto fail;
-	drm_atomic_set_fb_for_plane(plane_state, fb);
-
-	/* Make sure we don't accidentally do a full modeset. */
-	state->allow_modeset = false;
-	if (!crtc_state->active) {
-		DRM_DEBUG_ATOMIC("[CRTC:%d] disabled, rejecting legacy flip\n",
-				 crtc->base.id);
-		ret = -EINVAL;
-		goto fail;
-	}
-
-	if (flags & DRM_MODE_PAGE_FLIP_ASYNC)
-		nv50_wndw_atom(plane_state)->interval = 0;
-
-	ret = drm_atomic_nonblocking_commit(state);
-fail:
-	if (ret == -EDEADLK)
-		goto backoff;
-
-	drm_atomic_state_put(state);
-	return ret;
-
-backoff:
-	drm_atomic_state_clear(state);
-	drm_atomic_legacy_backoff(state);
-
-	/*
-	 * Someone might have exchanged the framebuffer while we dropped locks
-	 * in the backoff code. We need to fix up the fb refcount tracking the
-	 * core does for us.
-	 */
-	plane->old_fb = plane->fb;
-
-	goto retry;
-}
-
 static int
 nv50_head_gamma_set(struct drm_crtc *crtc, u16 *r, u16 *g, u16 *b,
 		    uint32_t size)
@@ -2386,7 +2319,7 @@ struct nv50_base {
 	.gamma_set = nv50_head_gamma_set,
 	.destroy = nv50_head_destroy,
 	.set_config = drm_atomic_helper_set_config,
-	.page_flip = nv50_head_page_flip,
+	.page_flip = drm_atomic_helper_page_flip,
 	.set_property = drm_atomic_helper_crtc_set_property,
 	.atomic_duplicate_state = nv50_head_atomic_duplicate_state,
 	.atomic_destroy_state = nv50_head_atomic_destroy_state,
Signature · 6 lines
1.9.1

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#29 lore raw
In reply to Andrey Grodzovsky

Hi Andrey,

Thank you for the patches.

Quoted 9 lines · On Monday 16 Jan 2017 10:44:54 Andrey Grodzovsky wrote:

This series is a folow-up on
https://patchwork.kernel.org/patch/9501787/

The first patch makes changes to atomic helpers to allow for drives with ASYNC flip support to use them. Patches 2 and 3 are to use this in AMDGPU/DC and patch 4 is possible cleanup in nouveau/kms who seems to have the duplicate the helper as we did to support ASYNC flips.

I have my doubts regarding this. I'd much rather see userspace moving to the atomic API instead of extending support for legacy APIs.

Quoted 12 lines

Andrey Grodzovsky (4):
drm/atomic: Save flip flags in drm_plane_state drm/amdgpu: Remove flip_flag from amdgpu_crtc drm/amd/display: Switch to using atomic_helper for flip. drm/nouveau/kms/nv50: Switch to using atomic helper for flip.

 drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h           |  1 -
 .../drm/amd/display/amdgpu_dm/amdgpu_dm_types.c    | 92 -------------------
 drivers/gpu/drm/drm_atomic_helper.c                | 10 +--
 drivers/gpu/drm/nouveau/nv50_display.c             | 77 ++----------------
 include/drm/drm_plane.h                            |  8 ++
 5 files changed, 22 insertions(+), 166 deletions(-)
Signature · 8 lines
Regards,

Laurent Pinchart

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#30 lore raw
Quoted 15 lines · Laurent Pinchart

Hi Andrey,

Thank you for the patches.

Quoted 9 lines · On Monday 16 Jan 2017 10:44:54 Andrey Grodzovsky wrote:

This series is a folow-up on
https://patchwork.kernel.org/patch/9501787/

The first patch makes changes to atomic helpers to allow for drives with ASYNC flip support to use them. Patches 2 and 3 are to use this in AMDGPU/DC and patch 4 is possible cleanup in nouveau/kms who seems to have the duplicate the helper as we did to support ASYNC flips.

Quoted 3 lines

I have my doubts regarding this. I'd much rather see userspace moving to the atomic API instead of extending support for legacy APIs.

This change is not about introducing the async flag but cleaning up the legacy helpers to make sure drivers that currently use it through the legacy IOCTLs can benefit from the helpers and not have to roll their own.

If the problem is with the pflip_flags, wouldn't drivers still need that after moving userspace to the atomic IOCTL?

I don't disagree with you on having userspace move to atomic but I don't expect to see all userspace drivers move to atomic in the next couple months. Why not clean this up in the meantime?

Harry

Quoted 6 lines
Quoted 6 lines

Andrey Grodzovsky (4):
drm/atomic: Save flip flags in drm_plane_state drm/amdgpu: Remove flip_flag from amdgpu_crtc drm/amd/display: Switch to using atomic_helper for flip. drm/nouveau/kms/nv50: Switch to using atomic helper for flip.

Quoted 7 lines
Quoted 6 lines
 drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h           |  1 -
 .../drm/amd/display/amdgpu_dm/amdgpu_dm_types.c    | 92 -------------------
 drivers/gpu/drm/drm_atomic_helper.c                | 10 +--
 drivers/gpu/drm/nouveau/nv50_display.c             | 77 ++----------------
 include/drm/drm_plane.h                            |  8 ++
 5 files changed, 22 insertions(+), 166 deletions(-)

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

#31 lore raw

Hi Harry,

Quoted 19 lines · On Monday 16 Jan 2017 16:13:39 Harry Wentland wrote:
Quoted 11 lines · Laurent Pinchart
Quoted 9 lines · On Monday 16 Jan 2017 10:44:54 Andrey Grodzovsky wrote:

This series is a folow-up on
https://patchwork.kernel.org/patch/9501787/

The first patch makes changes to atomic helpers to allow for drives with ASYNC flip support to use them. Patches 2 and 3 are to use this in AMDGPU/DC and patch 4 is possible cleanup in nouveau/kms who seems to have the duplicate the helper as we did to support ASYNC flips.

Quoted 2 lines

I have my doubts regarding this. I'd much rather see userspace moving to the atomic API instead of extending support for legacy APIs.

This change is not about introducing the async flag but cleaning up the legacy helpers to make sure drivers that currently use it through the legacy IOCTLs can benefit from the helpers and not have to roll their own.

Quoted 6 lines

If the problem is with the pflip_flags, wouldn't drivers still need that after moving userspace to the atomic IOCTL?

I don't disagree with you on having userspace move to atomic but I don't expect to see all userspace drivers move to atomic in the next couple months. Why not clean this up in the meantime?

If this patch series was just about moving common driver code into the core, sure, but it goes beyond that. Or, actually, it needs to go beyond that, but doesn't yet. Removing the DRM_MODE_PAGE_FLIP_ASYNC test in patch 1/4 means that the DRM core will not reject async page flips anymore, for any driver that uses the helper. You thus need to either patch all drivers that use the helper to reject the flag, or implement the feature in the drivers (and preferably in the helpers then). The current version of this patch series will make all existing users of the helpers accept async page flips without actually implementing them.

Quoted 12 lines
Quoted 12 lines
Quoted 12 lines

Andrey Grodzovsky (4):
drm/atomic: Save flip flags in drm_plane_state drm/amdgpu: Remove flip_flag from amdgpu_crtc drm/amd/display: Switch to using atomic_helper for flip. drm/nouveau/kms/nv50: Switch to using atomic helper for flip.

 drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h           |  1 -
 .../drm/amd/display/amdgpu_dm/amdgpu_dm_types.c    | 92 ---------------
 drivers/gpu/drm/drm_atomic_helper.c                | 10 +--
 drivers/gpu/drm/nouveau/nv50_display.c             | 77 ++-------------
 include/drm/drm_plane.h                            |  8 ++
 5 files changed, 22 insertions(+), 166 deletions(-)
Signature · 8 lines
Regards,

Laurent Pinchart

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#32 lore raw
Quoted 29 lines · Laurent Pinchart

Hi Harry,

Quoted 19 lines · On Monday 16 Jan 2017 16:13:39 Harry Wentland wrote:
Quoted 11 lines · Laurent Pinchart
Quoted 9 lines · On Monday 16 Jan 2017 10:44:54 Andrey Grodzovsky wrote:

This series is a folow-up on
https://patchwork.kernel.org/patch/9501787/

The first patch makes changes to atomic helpers to allow for drives with ASYNC flip support to use them. Patches 2 and 3 are to use this in AMDGPU/DC and patch 4 is possible cleanup in nouveau/kms who seems to have the duplicate the helper as we did to support ASYNC flips.

Quoted 2 lines

I have my doubts regarding this. I'd much rather see userspace moving to the atomic API instead of extending support for legacy APIs.

This change is not about introducing the async flag but cleaning up the legacy helpers to make sure drivers that currently use it through the legacy IOCTLs can benefit from the helpers and not have to roll their own.

Quoted 6 lines

If the problem is with the pflip_flags, wouldn't drivers still need that after moving userspace to the atomic IOCTL?

I don't disagree with you on having userspace move to atomic but I don't expect to see all userspace drivers move to atomic in the next couple months. Why not clean this up in the meantime?

Quoted 9 lines

If this patch series was just about moving common driver code into the core, sure, but it goes beyond that. Or, actually, it needs to go beyond that, but doesn't yet. Removing the DRM_MODE_PAGE_FLIP_ASYNC test in patch 1/4 means that the DRM core will not reject async page flips anymore, for any driver that uses the helper. You thus need to either patch all drivers that use the helper to reject the flag, or implement the feature in the drivers (and preferably in the helpers then). The current version of this patch series will make all existing users of the helpers accept async page flips without actually implementing them.

"How do we check for this driver feature" was one of the opens I raised here on irc, even before Andrey started typing :-) Definitely needs to be addressed. What's imo optional is exposing async flips through the atomic ioctl, since atm no atomic userspace seems to be in need of async flips yet. And the semantics of multi-plane async flips are tricky to define properly, so definitely want that userspace validation. That's why I think just doing the compat stuff is ok.
-Daniel

Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx