amd/amdgpu: add sched array to IPs with multiple run-queues

This sched array can be passed on to entity creation routine
instead of manually creating such sched array on every context creation.

v2: squash in missing break fix
Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c

@@ -74,7 +74,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 			   struct amdgpu_ctx *ctx)
 {
 	unsigned num_entities = amdgpu_ctx_total_num_entities();
-	unsigned i, j, k;
+	unsigned i, j;
 	int r;
 
 	if (priority < 0 || priority >= DRM_SCHED_PRIORITY_MAX)
@@ -121,73 +121,56 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 	ctx->override_priority = DRM_SCHED_PRIORITY_UNSET;
 
 	for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
-		struct amdgpu_ring *rings[AMDGPU_MAX_RINGS];
-		struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
-		unsigned num_rings = 0;
-		unsigned num_rqs = 0;
+		struct drm_gpu_scheduler **scheds;
+		struct drm_gpu_scheduler *sched;
+		unsigned num_scheds = 0;
 
 		switch (i) {
 		case AMDGPU_HW_IP_GFX:
-			rings[0] = &adev->gfx.gfx_ring[0];
-			num_rings = 1;
+			scheds = adev->gfx.gfx_sched;
+			num_scheds = 1;
 			break;
 		case AMDGPU_HW_IP_COMPUTE:
-			for (j = 0; j < adev->gfx.num_compute_rings; ++j)
-				rings[j] = &adev->gfx.compute_ring[j];
-			num_rings = adev->gfx.num_compute_rings;
+			scheds = adev->gfx.compute_sched;
+			num_scheds = adev->gfx.num_compute_sched;
 			break;
 		case AMDGPU_HW_IP_DMA:
-			for (j = 0; j < adev->sdma.num_instances; ++j)
-				rings[j] = &adev->sdma.instance[j].ring;
-			num_rings = adev->sdma.num_instances;
+			scheds = adev->sdma.sdma_sched;
+			num_scheds = adev->sdma.num_sdma_sched;
 			break;
 		case AMDGPU_HW_IP_UVD:
-			rings[0] = &adev->uvd.inst[0].ring;
-			num_rings = 1;
+			sched = &adev->uvd.inst[0].ring.sched;
+			scheds = &sched;
+			num_scheds = 1;
 			break;
 		case AMDGPU_HW_IP_VCE:
-			rings[0] = &adev->vce.ring[0];
-			num_rings = 1;
+			sched = &adev->vce.ring[0].sched;
+			scheds = &sched;
+			num_scheds = 1;
 			break;
 		case AMDGPU_HW_IP_UVD_ENC:
-			rings[0] = &adev->uvd.inst[0].ring_enc[0];
-			num_rings = 1;
+			sched = &adev->uvd.inst[0].ring_enc[0].sched;
+			scheds = &sched;
+			num_scheds = 1;
 			break;
 		case AMDGPU_HW_IP_VCN_DEC:
-			for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
-				if (adev->vcn.harvest_config & (1 << j))
-					continue;
-				rings[num_rings++] = &adev->vcn.inst[j].ring_dec;
-			}
+			scheds = adev->vcn.vcn_dec_sched;
+			num_scheds =  adev->vcn.num_vcn_dec_sched;
 			break;
 		case AMDGPU_HW_IP_VCN_ENC:
-			for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
-				if (adev->vcn.harvest_config & (1 << j))
-					continue;
-				for (k = 0; k < adev->vcn.num_enc_rings; ++k)
-					rings[num_rings++] = &adev->vcn.inst[j].ring_enc[k];
-			}
+			scheds = adev->vcn.vcn_enc_sched;
+			num_scheds =  adev->vcn.num_vcn_enc_sched;
 			break;
 		case AMDGPU_HW_IP_VCN_JPEG:
-			for (j = 0; j < adev->jpeg.num_jpeg_inst; ++j) {
-				if (adev->jpeg.harvest_config & (1 << j))
-					continue;
-				rings[num_rings++] = &adev->jpeg.inst[j].ring_dec;
-			}
+			scheds = adev->jpeg.jpeg_sched;
+			num_scheds =  adev->jpeg.num_jpeg_sched;
 			break;
 		}
 
-		for (j = 0; j < num_rings; ++j) {
-			if (!rings[j]->adev)
-				continue;
-
-			sched_list[num_rqs++] = &rings[j]->sched;
-		}
-
 		for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j)
 			r = drm_sched_entity_init(&ctx->entities[i][j].entity,
-						  priority, sched_list,
-						  num_rqs, &ctx->guilty);
+						  priority, scheds,
+						  num_scheds, &ctx->guilty);
 		if (r)
 			goto error_cleanup_entities;
 	}
@@ -628,3 +611,45 @@ void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr)
 	idr_destroy(&mgr->ctx_handles);
 	mutex_destroy(&mgr->lock);
 }
+
+void amdgpu_ctx_init_sched(struct amdgpu_device *adev)
+{
+	int i, j;
+
+	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
+		adev->gfx.gfx_sched[i] = &adev->gfx.gfx_ring[i].sched;
+		adev->gfx.num_gfx_sched++;
+	}
+
+	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
+		adev->gfx.compute_sched[i] = &adev->gfx.compute_ring[i].sched;
+		adev->gfx.num_compute_sched++;
+	}
+
+	for (i = 0; i < adev->sdma.num_instances; i++) {
+		adev->sdma.sdma_sched[i] = &adev->sdma.instance[i].ring.sched;
+		adev->sdma.num_sdma_sched++;
+	}
+
+	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
+		if (adev->vcn.harvest_config & (1 << i))
+			continue;
+		adev->vcn.vcn_dec_sched[adev->vcn.num_vcn_dec_sched++] =
+			&adev->vcn.inst[i].ring_dec.sched;
+	}
+
+	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
+		if (adev->vcn.harvest_config & (1 << i))
+			continue;
+		for (j = 0; j < adev->vcn.num_enc_rings; ++j)
+			adev->vcn.vcn_enc_sched[adev->vcn.num_vcn_enc_sched++] =
+				&adev->vcn.inst[i].ring_enc[j].sched;
+	}
+
+	for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) {
+		if (adev->jpeg.harvest_config & (1 << i))
+			continue;
+		adev->jpeg.jpeg_sched[adev->jpeg.num_jpeg_sched++] =
+			&adev->jpeg.inst[i].ring_dec.sched;
+	}
+}

drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h

@@ -87,4 +87,7 @@ void amdgpu_ctx_mgr_entity_fini(struct amdgpu_ctx_mgr *mgr);
 long amdgpu_ctx_mgr_entity_flush(struct amdgpu_ctx_mgr *mgr, long timeout);
 void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr);
 
+void amdgpu_ctx_init_sched(struct amdgpu_device *adev);
+
+
 #endif

drivers/gpu/drm/amd/amdgpu/amdgpu_device.c

@@ -3036,6 +3036,8 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 			adev->gfx.config.max_cu_per_sh,
 			adev->gfx.cu_info.number);
 
+	amdgpu_ctx_init_sched(adev);
+
 	adev->accel_working = true;
 
 	amdgpu_vm_check_compute_bug(adev);

drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h

@@ -269,8 +269,12 @@ struct amdgpu_gfx {
 	bool				me_fw_write_wait;
 	bool				cp_fw_write_wait;
 	struct amdgpu_ring		gfx_ring[AMDGPU_MAX_GFX_RINGS];
+	struct drm_gpu_scheduler	*gfx_sched[AMDGPU_MAX_GFX_RINGS];
+	uint32_t			num_gfx_sched;
 	unsigned			num_gfx_rings;
 	struct amdgpu_ring		compute_ring[AMDGPU_MAX_COMPUTE_RINGS];
+	struct drm_gpu_scheduler	*compute_sched[AMDGPU_MAX_COMPUTE_RINGS];
+	uint32_t			num_compute_sched;
 	unsigned			num_compute_rings;
 	struct amdgpu_irq_src		eop_irq;
 	struct amdgpu_irq_src		priv_reg_irq;

drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h

@@ -43,6 +43,8 @@ struct amdgpu_jpeg {
 	uint8_t	num_jpeg_inst;
 	struct amdgpu_jpeg_inst inst[AMDGPU_MAX_JPEG_INSTANCES];
 	struct amdgpu_jpeg_reg internal;
+	struct drm_gpu_scheduler *jpeg_sched[AMDGPU_MAX_JPEG_INSTANCES];
+	uint32_t num_jpeg_sched;
 	unsigned harvest_config;
 	struct delayed_work idle_work;
 	enum amd_powergating_state cur_state;

drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h

@@ -52,6 +52,8 @@ struct amdgpu_sdma_instance {
 
 struct amdgpu_sdma {
 	struct amdgpu_sdma_instance instance[AMDGPU_MAX_SDMA_INSTANCES];
+	struct drm_gpu_scheduler    *sdma_sched[AMDGPU_MAX_SDMA_INSTANCES];
+	uint32_t		    num_sdma_sched;
 	struct amdgpu_irq_src	trap_irq;
 	struct amdgpu_irq_src	illegal_inst_irq;
 	struct amdgpu_irq_src	ecc_irq;

drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h

@@ -31,6 +31,7 @@
 #define AMDGPU_VCN_MAX_ENC_RINGS	3
 
 #define AMDGPU_MAX_VCN_INSTANCES	2
+#define AMDGPU_MAX_VCN_ENC_RINGS  AMDGPU_VCN_MAX_ENC_RINGS * AMDGPU_MAX_VCN_INSTANCES
 
 #define AMDGPU_VCN_HARVEST_VCN0 (1 << 0)
 #define AMDGPU_VCN_HARVEST_VCN1 (1 << 1)
@@ -189,8 +190,12 @@ struct amdgpu_vcn {
 	uint32_t		*dpg_sram_curr_addr;
 
 	uint8_t	num_vcn_inst;
-	struct amdgpu_vcn_inst	inst[AMDGPU_MAX_VCN_INSTANCES];
-	struct amdgpu_vcn_reg	internal;
+	struct amdgpu_vcn_inst	 inst[AMDGPU_MAX_VCN_INSTANCES];
+	struct amdgpu_vcn_reg	 internal;
+	struct drm_gpu_scheduler *vcn_enc_sched[AMDGPU_MAX_VCN_ENC_RINGS];
+	struct drm_gpu_scheduler *vcn_dec_sched[AMDGPU_MAX_VCN_INSTANCES];
+	uint32_t		 num_vcn_enc_sched;
+	uint32_t		 num_vcn_dec_sched;
 
 	unsigned	harvest_config;
 	int (*pause_dpg_mode)(struct amdgpu_device *adev,