Re: [PATCH 5.11 079/104] drm/amdgpu: enable only one high prio compute queue

Christian König <christian.koenig@xxxxxxx> · Fri, 5 Mar 2021 16:18:00 +0100

Am 05.03.21 um 15:48 schrieb Deucher, Alexander:
[AMD Public Use]

-----Original Message-----
From: Koenig, Christian <Christian.Koenig@xxxxxxx>
Sent: Friday, March 5, 2021 8:03 AM
To: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>; linux-
kernel@xxxxxxxxxxxxxxx
Cc: stable@xxxxxxxxxxxxxxx; Das, Nirmoy <Nirmoy.Das@xxxxxxx>; Deucher,
Alexander <Alexander.Deucher@xxxxxxx>; Sasha Levin
<sashal@xxxxxxxxxx>
Subject: Re: [PATCH 5.11 079/104] drm/amdgpu: enable only one high prio
compute queue

Mhm, I'm not sure this one needs to be backported.

Why did you pick it up Greg?
It was picked up by Sasha's fixes checker.

Well the change who needs this isn't in any earlier kernel, isn't it?

Christian.


Alex


Thanks,
Christian.

Am 05.03.21 um 13:21 schrieb Greg Kroah-Hartman:
From: Nirmoy Das <nirmoy.das@xxxxxxx>

[ Upstream commit 8c0225d79273968a65e73a4204fba023ae02714d ]

For high priority compute to work properly we need to enable wave
limiting on gfx pipe. Wave limiting is done through writing into
mmSPI_WCL_PIPE_PERCENT_GFX register. Enable only one high priority
compute queue to avoid race condition between multiple high priority
compute queues writing that register simultaneously.

Signed-off-by: Nirmoy Das <nirmoy.das@xxxxxxx>
Acked-by: Christian König <christian.koenig@xxxxxxx>
Reviewed-by: Alex Deucher <alexander.deucher@xxxxxxx>
Signed-off-by: Alex Deucher <alexander.deucher@xxxxxxx>
Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
---
   drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 15 ++++++++-------
   drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h |  2 +-
   drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c  |  6 ++----
   drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c   |  6 ++----
   drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c   |  7 ++-----
   5 files changed, 15 insertions(+), 21 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
index cd2c676a2797..8e0a6c62322e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
@@ -193,15 +193,16 @@ static bool
amdgpu_gfx_is_multipipe_capable(struct amdgpu_device *adev)
   }

   bool amdgpu_gfx_is_high_priority_compute_queue(struct
amdgpu_device *adev,
-					       int pipe, int queue)
+					       struct amdgpu_ring *ring)
   {
-	bool multipipe_policy = amdgpu_gfx_is_multipipe_capable(adev);
-	int cond;
-	/* Policy: alternate between normal and high priority */
-	cond = multipipe_policy ? pipe : queue;
-
-	return ((cond % 2) != 0);
+	/* Policy: use 1st queue as high priority compute queue if we
+	 * have more than one compute queue.
+	 */
+	if (adev->gfx.num_compute_rings > 1 &&
+	    ring == &adev->gfx.compute_ring[0])
+		return true;

+	return false;
   }

   void amdgpu_gfx_compute_queue_acquire(struct amdgpu_device
*adev)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
index 6b5a8f4642cc..72dbcd2bc6a6 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
@@ -380,7 +380,7 @@ void
amdgpu_queue_mask_bit_to_mec_queue(struct amdgpu_device *adev, int
bit,
   bool amdgpu_gfx_is_mec_queue_enabled(struct amdgpu_device *adev,
int mec,
   				     int pipe, int queue);
   bool amdgpu_gfx_is_high_priority_compute_queue(struct
amdgpu_device *adev,
-					       int pipe, int queue);
+					       struct amdgpu_ring *ring);
   int amdgpu_gfx_me_queue_to_bit(struct amdgpu_device *adev, int me,
   			       int pipe, int queue);
   void amdgpu_gfx_bit_to_me_queue(struct amdgpu_device *adev, int
bit,
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
index e7d6da05011f..3a291befcddc 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
@@ -4495,8 +4495,7 @@ static int gfx_v10_0_compute_ring_init(struct
amdgpu_device *adev, int ring_id,
   	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
   		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
   		+ ring->pipe;
-	hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev,
ring->pipe,
-							    ring->queue) ?
+	hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev,
ring) ?
   			AMDGPU_GFX_PIPE_PRIO_HIGH :
AMDGPU_GFX_PIPE_PRIO_NORMAL;
   	/* type-2 packets are deprecated on MEC, use type-3 instead */
   	r = amdgpu_ring_init(adev, ring, 1024, @@ -6545,8 +6544,7 @@ static
void gfx_v10_0_compute_mqd_set_priority(struct amdgpu_ring *ring,
struct
   	struct amdgpu_device *adev = ring->adev;

   	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
-		if (amdgpu_gfx_is_high_priority_compute_queue(adev,
ring->pipe,
-							      ring->queue)) {
+		if (amdgpu_gfx_is_high_priority_compute_queue(adev,
ring)) {
   			mqd->cp_hqd_pipe_priority =
AMDGPU_GFX_PIPE_PRIO_HIGH;
   			mqd->cp_hqd_queue_priority =

	AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM;
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
index 37639214cbbb..b0284c4659ba 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
@@ -1923,8 +1923,7 @@ static int gfx_v8_0_compute_ring_init(struct
amdgpu_device *adev, int ring_id,
   		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
   		+ ring->pipe;

-	hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev,
ring->pipe,
-							    ring->queue) ?
+	hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev,
ring) ?
   			AMDGPU_GFX_PIPE_PRIO_HIGH :
AMDGPU_RING_PRIO_DEFAULT;
   	/* type-2 packets are deprecated on MEC, use type-3 instead */
   	r = amdgpu_ring_init(adev, ring, 1024, @@ -4442,8 +4441,7 @@ static
void gfx_v8_0_mqd_set_priority(struct amdgpu_ring *ring, struct vi_mqd
*m
   	struct amdgpu_device *adev = ring->adev;

   	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
-		if (amdgpu_gfx_is_high_priority_compute_queue(adev,
ring->pipe,
-							      ring->queue)) {
+		if (amdgpu_gfx_is_high_priority_compute_queue(adev,
ring)) {
   			mqd->cp_hqd_pipe_priority =
AMDGPU_GFX_PIPE_PRIO_HIGH;
   			mqd->cp_hqd_queue_priority =

	AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM;
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
index 5f4805e4d04a..3e800193a604 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
@@ -2228,8 +2228,7 @@ static int gfx_v9_0_compute_ring_init(struct
amdgpu_device *adev, int ring_id,
   	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
   		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
   		+ ring->pipe;
-	hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev,
ring->pipe,
-							    ring->queue) ?
+	hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev,
ring) ?
   			AMDGPU_GFX_PIPE_PRIO_HIGH :
AMDGPU_GFX_PIPE_PRIO_NORMAL;
   	/* type-2 packets are deprecated on MEC, use type-3 instead */
   	return amdgpu_ring_init(adev, ring, 1024, @@ -3391,9 +3390,7 @@
static void gfx_v9_0_mqd_set_priority(struct amdgpu_ring *ring, struct
v9_mqd *m
   	struct amdgpu_device *adev = ring->adev;

   	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
-		if (amdgpu_gfx_is_high_priority_compute_queue(adev,
-							      ring->pipe,
-							      ring->queue)) {
+		if (amdgpu_gfx_is_high_priority_compute_queue(adev,
ring)) {
   			mqd->cp_hqd_pipe_priority =
AMDGPU_GFX_PIPE_PRIO_HIGH;
   			mqd->cp_hqd_queue_priority =

	AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM;







