drm/amdgpu: allow split of queues with kfd at queue granularity v4

Previously the queue/pipe split with kfd operated with pipe
granularity. This patch allows amdgpu to take ownership of an arbitrary
set of queues.

It also consolidates the last few magic numbers in the compute
initialization process into mec_init.

v2: support for gfx9
v3: renamed AMDGPU_MAX_QUEUES to AMDGPU_MAX_COMPUTE_QUEUES
v4: fix off-by-one in num_mec checks in *_compute_queue_acquire

Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

1 files changed, 64 insertions, 17 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
index 8c2241631ade..2af325500ea0 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
@@ -52,7 +52,6 @@
 #include "smu/smu_7_1_3_d.h"
 
 #define GFX8_NUM_GFX_RINGS     1
-#define GFX8_NUM_COMPUTE_RINGS 8
 #define GFX8_MEC_HPD_SIZE 2048
 
 #define TOPAZ_GB_ADDR_CONFIG_GOLDEN 0x22010001
@@ -1422,12 +1421,45 @@ static void gfx_v8_0_kiq_free_ring(struct amdgpu_ring *ring,
 	amdgpu_ring_fini(ring);
 }
 
+static void gfx_v8_0_compute_queue_acquire(struct amdgpu_device *adev)
+{
+	int i, queue, pipe, mec;
+
+	/* policy for amdgpu compute queue ownership */
+	for (i = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; ++i) {
+		queue = i % adev->gfx.mec.num_queue_per_pipe;
+		pipe = (i / adev->gfx.mec.num_queue_per_pipe)
+			% adev->gfx.mec.num_pipe_per_mec;
+		mec = (i / adev->gfx.mec.num_queue_per_pipe)
+			/ adev->gfx.mec.num_pipe_per_mec;
+
+		/* we've run out of HW */
+		if (mec >= adev->gfx.mec.num_mec)
+			break;
+
+		/* policy: amdgpu owns all queues in the first pipe */
+		if (mec == 0 && pipe == 0)
+			set_bit(i, adev->gfx.mec.queue_bitmap);
+	}
+
+	/* update the number of active compute rings */
+	adev->gfx.num_compute_rings =
+		bitmap_weight(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
+
+	/* If you hit this case and edited the policy, you probably just
+	 * need to increase AMDGPU_MAX_COMPUTE_RINGS */
+	if (WARN_ON(adev->gfx.num_compute_rings > AMDGPU_MAX_COMPUTE_RINGS))
+		adev->gfx.num_compute_rings = AMDGPU_MAX_COMPUTE_RINGS;
+}
+
 static int gfx_v8_0_mec_init(struct amdgpu_device *adev)
 {
 	int r;
 	u32 *hpd;
 	size_t mec_hpd_size;
 
+	bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
+
 	switch (adev->asic_type) {
 	case CHIP_FIJI:
 	case CHIP_TONGA:
@@ -1447,8 +1479,10 @@ static int gfx_v8_0_mec_init(struct amdgpu_device *adev)
 	adev->gfx.mec.num_pipe_per_mec = 4;
 	adev->gfx.mec.num_queue_per_pipe = 8;
 
-	/* only 1 pipe of the first MEC is owned by amdgpu */
-	mec_hpd_size = 1 * 1 * adev->gfx.mec.num_queue_per_pipe * GFX8_MEC_HPD_SIZE;
+	/* take ownership of the relevant compute queues */
+	gfx_v8_0_compute_queue_acquire(adev);
+
+	mec_hpd_size = adev->gfx.num_compute_rings * GFX8_MEC_HPD_SIZE;
 
 	if (adev->gfx.mec.hpd_eop_obj == NULL) {
 		r = amdgpu_bo_create(adev,
@@ -2107,7 +2141,7 @@ static int gfx_v8_0_gpu_early_init(struct amdgpu_device *adev)
 
 static int gfx_v8_0_sw_init(void *handle)
 {
-	int i, r;
+	int i, r, ring_id;
 	struct amdgpu_ring *ring;
 	struct amdgpu_kiq *kiq;
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
@@ -2174,29 +2208,42 @@ static int gfx_v8_0_sw_init(void *handle)
 	}
 
 	/* set up the compute queues */
-	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
+	for (i = 0, ring_id = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; i++) {
 		unsigned irq_type;
 
-		/* max 32 queues per MEC */
-		if ((i >= 32) || (i >= AMDGPU_MAX_COMPUTE_RINGS)) {
-			DRM_ERROR("Too many (%d) compute rings!\n", i);
+		if (!test_bit(i, adev->gfx.mec.queue_bitmap))
+			continue;
+
+		if (WARN_ON(ring_id >= AMDGPU_MAX_COMPUTE_RINGS))
 			break;
-		}
-		ring = &adev->gfx.compute_ring[i];
+
+		ring = &adev->gfx.compute_ring[ring_id];
+
+		/* mec0 is me1 */
+		ring->me = ((i / adev->gfx.mec.num_queue_per_pipe)
+				/ adev->gfx.mec.num_pipe_per_mec)
+				+ 1;
+		ring->pipe = (i / adev->gfx.mec.num_queue_per_pipe)
+				% adev->gfx.mec.num_pipe_per_mec;
+		ring->queue = i % adev->gfx.mec.num_queue_per_pipe;
+
 		ring->ring_obj = NULL;
 		ring->use_doorbell = true;
-		ring->doorbell_index = AMDGPU_DOORBELL_MEC_RING0 + i;
-		ring->me = 1; /* first MEC */
-		ring->pipe = i / 8;
-		ring->queue = i % 8;
-		ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + (i * GFX8_MEC_HPD_SIZE);
+		ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + (ring_id * GFX8_MEC_HPD_SIZE);
+		ring->doorbell_index = AMDGPU_DOORBELL_MEC_RING0 + ring_id;
 		sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
-		irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP + ring->pipe;
+
+		irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
+			+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
+			+ ring->pipe;
+
 		/* type-2 packets are deprecated on MEC, use type-3 instead */
 		r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq,
 				     irq_type);
 		if (r)
 			return r;
+
+		ring_id++;
 	}
 
 	r = gfx_v8_0_kiq_init(adev);
@@ -5678,7 +5725,7 @@ static int gfx_v8_0_early_init(void *handle)
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
 	adev->gfx.num_gfx_rings = GFX8_NUM_GFX_RINGS;
-	adev->gfx.num_compute_rings = GFX8_NUM_COMPUTE_RINGS;
+	adev->gfx.num_compute_rings = AMDGPU_MAX_COMPUTE_RINGS;
 	adev->gfx.funcs = &gfx_v8_0_gfx_funcs;
 	gfx_v8_0_set_ring_funcs(adev);
 	gfx_v8_0_set_irq_funcs(adev);