[PATCH AUTOSEL 5.4 11/15] drm/amdkfd: fix a memory leak issue

[PATCH AUTOSEL 5.4 11/15] drm/amdkfd: fix a memory leak issue

[Sasha Levin]

From: Dennis Li <Dennis.Li@amd.com>

[ Upstream commit 087d764159996ae378b08c0fdd557537adfd6899 ]

In the resume stage of GPU recovery, start_cpsch will call pm_init
which set pm->allocated as false, cause the next pm_release_ib has
no chance to release ib memory.

Add pm_release_ib in stop_cpsch which will be called in the suspend
stage of GPU recovery.

Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Dennis Li <Dennis.Li@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
index a2ed9c257cb0d..e9a2784400792 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -1075,6 +1075,8 @@ static int stop_cpsch(struct device_queue_manager *dqm)
 	unmap_queues_cpsch(dqm, KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES, 0);
 	dqm_unlock(dqm);
 
+	pm_release_ib(&dqm->packets);
+
 	kfd_gtt_sa_free(dqm->dev, dqm->fence_mem);
 	pm_uninit(&dqm->packets);
 
-- 
2.25.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH AUTOSEL 5.4 12/15] drm/amd/display: update nv1x stutter latencies

[Sasha Levin]

From: Jun Lei <jun.lei@amd.com>

[ Upstream commit c4790a8894232f39c25c7c546c06efe074e63384 ]

[why]
Recent characterization shows increased stutter latencies on some SKUs,
leading to underflow.

[how]
Update SOC params to account for this worst case latency.

Signed-off-by: Jun Lei <jun.lei@amd.com>
Acked-by: Aurabindo Pillai <aurabindo.pillai@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c b/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c
index 05b98eadc2899..46afd048c1ea3 100644
--- a/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c
+++ b/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c
@@ -340,8 +340,8 @@ struct _vcs_dpi_soc_bounding_box_st dcn2_0_nv14_soc = {
 			},
 		},
 	.num_states = 5,
-	.sr_exit_time_us = 8.6,
-	.sr_enter_plus_exit_time_us = 10.9,
+	.sr_exit_time_us = 11.6,
+	.sr_enter_plus_exit_time_us = 13.9,
 	.urgent_latency_us = 4.0,
 	.urgent_latency_pixel_data_only_us = 4.0,
 	.urgent_latency_pixel_mixed_with_vm_data_us = 4.0,
-- 
2.25.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH AUTOSEL 5.4 13/15] drm/amdgpu/dc: Require primary plane to be enabled whenever the CRTC is

[Sasha Levin]

From: Michel Dänzer <mdaenzer@redhat.com>

[ Upstream commit 2f228aab21bbc74e90e267a721215ec8be51daf7 ]

Don't check drm_crtc_state::active for this either, per its
documentation in include/drm/drm_crtc.h:

 * Hence drivers must not consult @active in their various
 * &drm_mode_config_funcs.atomic_check callback to reject an atomic
 * commit.

atomic_remove_fb disables the CRTC as needed for disabling the primary
plane.

This prevents at least the following problems if the primary plane gets
disabled (e.g. due to destroying the FB assigned to the primary plane,
as happens e.g. with mutter in Wayland mode):

* The legacy cursor ioctl returned EINVAL for a non-0 cursor FB ID
  (which enables the cursor plane).
* If the cursor plane was enabled, changing the legacy DPMS property
  value from off to on returned EINVAL.

v2:
* Minor changes to code comment and commit log, per review feedback.

GitLab: https://gitlab.gnome.org/GNOME/mutter/-/issues/1108
GitLab: https://gitlab.gnome.org/GNOME/mutter/-/issues/1165
GitLab: https://gitlab.gnome.org/GNOME/mutter/-/issues/1344
Suggested-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Signed-off-by: Michel Dänzer <mdaenzer@redhat.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 32 ++++++-------------
 1 file changed, 10 insertions(+), 22 deletions(-)

diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
index 60e50181f6d39..2384aa018993d 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
@@ -4299,19 +4299,6 @@ static void dm_crtc_helper_disable(struct drm_crtc *crtc)
 {
 }
 
-static bool does_crtc_have_active_cursor(struct drm_crtc_state *new_crtc_state)
-{
-	struct drm_device *dev = new_crtc_state->crtc->dev;
-	struct drm_plane *plane;
-
-	drm_for_each_plane_mask(plane, dev, new_crtc_state->plane_mask) {
-		if (plane->type == DRM_PLANE_TYPE_CURSOR)
-			return true;
-	}
-
-	return false;
-}
-
 static int count_crtc_active_planes(struct drm_crtc_state *new_crtc_state)
 {
 	struct drm_atomic_state *state = new_crtc_state->state;
@@ -4391,19 +4378,20 @@ static int dm_crtc_helper_atomic_check(struct drm_crtc *crtc,
 		return ret;
 	}
 
-	/* In some use cases, like reset, no stream is attached */
-	if (!dm_crtc_state->stream)
-		return 0;
-
 	/*
-	 * We want at least one hardware plane enabled to use
-	 * the stream with a cursor enabled.
+	 * We require the primary plane to be enabled whenever the CRTC is, otherwise
+	 * drm_mode_cursor_universal may end up trying to enable the cursor plane while all other
+	 * planes are disabled, which is not supported by the hardware. And there is legacy
+	 * userspace which stops using the HW cursor altogether in response to the resulting EINVAL.
 	 */
-	if (state->enable && state->active &&
-	    does_crtc_have_active_cursor(state) &&
-	    dm_crtc_state->active_planes == 0)
+	if (state->enable &&
+	    !(state->plane_mask & drm_plane_mask(crtc->primary)))
 		return -EINVAL;
 
+	/* In some use cases, like reset, no stream is attached */
+	if (!dm_crtc_state->stream)
+		return 0;
+
 	if (dc_validate_stream(dc, dm_crtc_state->stream) == DC_OK)
 		return 0;
 
-- 
2.25.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH AUTOSEL 5.4 13/15] drm/amdgpu/dc: Require primary plane to be enabled whenever the CRTC is

[Michel Dänzer]

On 2020-09-21 4:40 p.m., Sasha Levin wrote:
> From: Michel Dänzer <mdaenzer@redhat.com>
> 
> [ Upstream commit 2f228aab21bbc74e90e267a721215ec8be51daf7 ]
> 
> Don't check drm_crtc_state::active for this either, per its
> documentation in include/drm/drm_crtc.h:
> 
>   * Hence drivers must not consult @active in their various
>   * &drm_mode_config_funcs.atomic_check callback to reject an atomic
>   * commit.
> 
> atomic_remove_fb disables the CRTC as needed for disabling the primary
> plane.
> 
> This prevents at least the following problems if the primary plane gets
> disabled (e.g. due to destroying the FB assigned to the primary plane,
> as happens e.g. with mutter in Wayland mode):
> 
> * The legacy cursor ioctl returned EINVAL for a non-0 cursor FB ID
>    (which enables the cursor plane).
> * If the cursor plane was enabled, changing the legacy DPMS property
>    value from off to on returned EINVAL.
> 
> v2:
> * Minor changes to code comment and commit log, per review feedback.
> 
> GitLab: https://gitlab.gnome.org/GNOME/mutter/-/issues/1108
> GitLab: https://gitlab.gnome.org/GNOME/mutter/-/issues/1165
> GitLab: https://gitlab.gnome.org/GNOME/mutter/-/issues/1344
> Suggested-by: Daniel Vetter <daniel.vetter@ffwll.ch>
> Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
> Reviewed-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
> Signed-off-by: Michel Dänzer <mdaenzer@redhat.com>
> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
> Signed-off-by: Sasha Levin <sashal@kernel.org>

I'm a bit nervous about this getting backported so far back so quickly. 
I'd prefer waiting for 5.9 final first at least.


-- 
Earthling Michel Dänzer               |               https://redhat.com
Libre software enthusiast             |             Mesa and X developer
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH AUTOSEL 5.4 13/15] drm/amdgpu/dc: Require primary plane to be enabled whenever the CRTC is

[Sasha Levin]

On Mon, Sep 21, 2020 at 04:48:05PM +0200, Michel Dänzer wrote:
>On 2020-09-21 4:40 p.m., Sasha Levin wrote:
>>From: Michel Dänzer <mdaenzer@redhat.com>
>>
>>[ Upstream commit 2f228aab21bbc74e90e267a721215ec8be51daf7 ]
>>
>>Don't check drm_crtc_state::active for this either, per its
>>documentation in include/drm/drm_crtc.h:
>>
>>  * Hence drivers must not consult @active in their various
>>  * &drm_mode_config_funcs.atomic_check callback to reject an atomic
>>  * commit.
>>
>>atomic_remove_fb disables the CRTC as needed for disabling the primary
>>plane.
>>
>>This prevents at least the following problems if the primary plane gets
>>disabled (e.g. due to destroying the FB assigned to the primary plane,
>>as happens e.g. with mutter in Wayland mode):
>>
>>* The legacy cursor ioctl returned EINVAL for a non-0 cursor FB ID
>>   (which enables the cursor plane).
>>* If the cursor plane was enabled, changing the legacy DPMS property
>>   value from off to on returned EINVAL.
>>
>>v2:
>>* Minor changes to code comment and commit log, per review feedback.
>>
>>GitLab: https://gitlab.gnome.org/GNOME/mutter/-/issues/1108
>>GitLab: https://gitlab.gnome.org/GNOME/mutter/-/issues/1165
>>GitLab: https://gitlab.gnome.org/GNOME/mutter/-/issues/1344
>>Suggested-by: Daniel Vetter <daniel.vetter@ffwll.ch>
>>Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
>>Reviewed-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
>>Signed-off-by: Michel Dänzer <mdaenzer@redhat.com>
>>Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
>>Signed-off-by: Sasha Levin <sashal@kernel.org>
>
>I'm a bit nervous about this getting backported so far back so 
>quickly. I'd prefer waiting for 5.9 final first at least.

Will drop it for now, thanks.

-- 
Thanks,
Sasha
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx