drm/amdgpu: implement harvesting support for UVD 7.2 (v3)

Properly handle cases where one or more instance of the IP
block may be harvested.

v2: make sure ip_num_rings is initialized amdgpu_queue_mgr.c
v3: rebase on Christian's UVD changes, drop unused var
Reviewed-by: James Zhu <James.Zhu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

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

@@ -286,7 +286,7 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 	struct drm_crtc *crtc;
 	uint32_t ui32 = 0;
 	uint64_t ui64 = 0;
-	int i, found;
+	int i, j, found;
 	int ui32_size = sizeof(ui32);
 
 	if (!info->return_size || !info->return_pointer)
@@ -348,7 +348,11 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 			break;
 		case AMDGPU_HW_IP_UVD:
 			type = AMD_IP_BLOCK_TYPE_UVD;
-			ring_mask |= adev->uvd.inst[0].ring.ready;
+			for (i = 0; i < adev->uvd.num_uvd_inst; i++) {
+				if (adev->uvd.harvest_config & (1 << i))
+					continue;
+				ring_mask |= adev->uvd.inst[i].ring.ready;
+			}
 			ib_start_alignment = 64;
 			ib_size_alignment = 64;
 			break;
@@ -361,9 +365,12 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
 			break;
 		case AMDGPU_HW_IP_UVD_ENC:
 			type = AMD_IP_BLOCK_TYPE_UVD;
-			for (i = 0; i < adev->uvd.num_enc_rings; i++)
-				ring_mask |=
-					adev->uvd.inst[0].ring_enc[i].ready << i;
+			for (i = 0; i < adev->uvd.num_uvd_inst; i++) {
+				if (adev->uvd.harvest_config & (1 << i))
+					continue;
+				for (j = 0; j < adev->uvd.num_enc_rings; j++)
+					ring_mask |= adev->uvd.inst[i].ring_enc[j].ready << j;
+			}
 			ib_start_alignment = 64;
 			ib_size_alignment = 64;
 			break;

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

@@ -214,7 +214,7 @@ int amdgpu_queue_mgr_map(struct amdgpu_device *adev,
 			 u32 hw_ip, u32 instance, u32 ring,
 			 struct amdgpu_ring **out_ring)
 {
-	int r, ip_num_rings;
+	int i, r, ip_num_rings = 0;
 	struct amdgpu_queue_mapper *mapper = &mgr->mapper[hw_ip];
 
 	if (!adev || !mgr || !out_ring)
@@ -243,14 +243,21 @@ int amdgpu_queue_mgr_map(struct amdgpu_device *adev,
 		ip_num_rings = adev->sdma.num_instances;
 		break;
 	case AMDGPU_HW_IP_UVD:
-		ip_num_rings = adev->uvd.num_uvd_inst;
+		for (i = 0; i < adev->uvd.num_uvd_inst; i++) {
+			if (!(adev->uvd.harvest_config & (1 << i)))
+				ip_num_rings++;
+		}
 		break;
 	case AMDGPU_HW_IP_VCE:
 		ip_num_rings = adev->vce.num_rings;
 		break;
 	case AMDGPU_HW_IP_UVD_ENC:
+		for (i = 0; i < adev->uvd.num_uvd_inst; i++) {
+			if (!(adev->uvd.harvest_config & (1 << i)))
+				ip_num_rings++;
+		}
 		ip_num_rings =
-			adev->uvd.num_enc_rings * adev->uvd.num_uvd_inst;
+			adev->uvd.num_enc_rings * ip_num_rings;
 		break;
 	case AMDGPU_HW_IP_VCN_DEC:
 		ip_num_rings = 1;

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

@@ -255,7 +255,8 @@ int amdgpu_uvd_sw_init(struct amdgpu_device *adev)
 		bo_size += AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
 
 	for (j = 0; j < adev->uvd.num_uvd_inst; j++) {
-
+		if (adev->uvd.harvest_config & (1 << j))
+			continue;
 		r = amdgpu_bo_create_kernel(adev, bo_size, PAGE_SIZE,
 					    AMDGPU_GEM_DOMAIN_VRAM, &adev->uvd.inst[j].vcpu_bo,
 					    &adev->uvd.inst[j].gpu_addr, &adev->uvd.inst[j].cpu_addr);
@@ -308,6 +309,8 @@ int amdgpu_uvd_sw_fini(struct amdgpu_device *adev)
 	drm_sched_entity_destroy(&adev->uvd.entity);
 
 	for (j = 0; j < adev->uvd.num_uvd_inst; ++j) {
+		if (adev->uvd.harvest_config & (1 << j))
+			continue;
 		kfree(adev->uvd.inst[j].saved_bo);
 
 		amdgpu_bo_free_kernel(&adev->uvd.inst[j].vcpu_bo,
@@ -343,6 +346,8 @@ int amdgpu_uvd_suspend(struct amdgpu_device *adev)
 	}
 
 	for (j = 0; j < adev->uvd.num_uvd_inst; ++j) {
+		if (adev->uvd.harvest_config & (1 << j))
+			continue;
 		if (adev->uvd.inst[j].vcpu_bo == NULL)
 			continue;
 
@@ -365,6 +370,8 @@ int amdgpu_uvd_resume(struct amdgpu_device *adev)
 	int i;
 
 	for (i = 0; i < adev->uvd.num_uvd_inst; i++) {
+		if (adev->uvd.harvest_config & (1 << i))
+			continue;
 		if (adev->uvd.inst[i].vcpu_bo == NULL)
 			return -EINVAL;
 
@@ -1159,6 +1166,8 @@ static void amdgpu_uvd_idle_work_handler(struct work_struct *work)
 	unsigned fences = 0, i, j;
 
 	for (i = 0; i < adev->uvd.num_uvd_inst; ++i) {
+		if (adev->uvd.harvest_config & (1 << i))
+			continue;
 		fences += amdgpu_fence_count_emitted(&adev->uvd.inst[i].ring);
 		for (j = 0; j < adev->uvd.num_enc_rings; ++j) {
 			fences += amdgpu_fence_count_emitted(&adev->uvd.inst[i].ring_enc[j]);

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

@@ -48,6 +48,9 @@ struct amdgpu_uvd_inst {
 	uint32_t                srbm_soft_reset;
 };
 
+#define AMDGPU_UVD_HARVEST_UVD0 (1 << 0)
+#define AMDGPU_UVD_HARVEST_UVD1 (1 << 1)
+
 struct amdgpu_uvd {
 	const struct firmware	*fw;	/* UVD firmware */
 	unsigned		fw_version;
@@ -61,6 +64,7 @@ struct amdgpu_uvd {
 	atomic_t		handles[AMDGPU_MAX_UVD_HANDLES];
 	struct drm_sched_entity entity;
 	struct delayed_work	idle_work;
+	unsigned		harvest_config;
 };
 
 int amdgpu_uvd_sw_init(struct amdgpu_device *adev);

drivers/gpu/drm/amd/amdgpu/uvd_v7_0.c

@@ -41,6 +41,12 @@
 #include "mmhub/mmhub_1_0_sh_mask.h"
 #include "ivsrcid/uvd/irqsrcs_uvd_7_0.h"
 
+#define mmUVD_PG0_CC_UVD_HARVESTING                                                                    0x00c7
+#define mmUVD_PG0_CC_UVD_HARVESTING_BASE_IDX                                                           1
+//UVD_PG0_CC_UVD_HARVESTING
+#define UVD_PG0_CC_UVD_HARVESTING__UVD_DISABLE__SHIFT                                                         0x1
+#define UVD_PG0_CC_UVD_HARVESTING__UVD_DISABLE_MASK                                                           0x00000002L
+
 #define UVD7_MAX_HW_INSTANCES_VEGA20			2
 
 static void uvd_v7_0_set_ring_funcs(struct amdgpu_device *adev);
@@ -370,10 +376,25 @@ static int uvd_v7_0_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout)
 static int uvd_v7_0_early_init(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-	if (adev->asic_type == CHIP_VEGA20)
+
+	if (adev->asic_type == CHIP_VEGA20) {
+		u32 harvest;
+		int i;
+
 		adev->uvd.num_uvd_inst = UVD7_MAX_HW_INSTANCES_VEGA20;
-	else
+		for (i = 0; i < adev->uvd.num_uvd_inst; i++) {
+			harvest = RREG32_SOC15(UVD, i, mmUVD_PG0_CC_UVD_HARVESTING);
+			if (harvest & UVD_PG0_CC_UVD_HARVESTING__UVD_DISABLE_MASK) {
+				adev->uvd.harvest_config |= 1 << i;
+			}
+		}
+		if (adev->uvd.harvest_config == (AMDGPU_UVD_HARVEST_UVD0 |
+						 AMDGPU_UVD_HARVEST_UVD1))
+			/* both instances are harvested, disable the block */
+			return -ENOENT;
+	} else {
 		adev->uvd.num_uvd_inst = 1;
+	}
 
 	if (amdgpu_sriov_vf(adev))
 		adev->uvd.num_enc_rings = 1;
@@ -393,6 +414,8 @@ static int uvd_v7_0_sw_init(void *handle)
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
 	for (j = 0; j < adev->uvd.num_uvd_inst; j++) {
+		if (adev->uvd.harvest_config & (1 << j))
+			continue;
 		/* UVD TRAP */
 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_uvds[j], UVD_7_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->uvd.inst[j].irq);
 		if (r)
@@ -425,6 +448,8 @@ static int uvd_v7_0_sw_init(void *handle)
 		return r;
 
 	for (j = 0; j < adev->uvd.num_uvd_inst; j++) {
+		if (adev->uvd.harvest_config & (1 << j))
+			continue;
 		if (!amdgpu_sriov_vf(adev)) {
 			ring = &adev->uvd.inst[j].ring;
 			sprintf(ring->name, "uvd<%d>", j);
@@ -472,6 +497,8 @@ static int uvd_v7_0_sw_fini(void *handle)
 		return r;
 
 	for (j = 0; j < adev->uvd.num_uvd_inst; ++j) {
+		if (adev->uvd.harvest_config & (1 << j))
+			continue;
 		for (i = 0; i < adev->uvd.num_enc_rings; ++i)
 			amdgpu_ring_fini(&adev->uvd.inst[j].ring_enc[i]);
 	}
@@ -500,6 +527,8 @@ static int uvd_v7_0_hw_init(void *handle)
 		goto done;
 
 	for (j = 0; j < adev->uvd.num_uvd_inst; ++j) {
+		if (adev->uvd.harvest_config & (1 << j))
+			continue;
 		ring = &adev->uvd.inst[j].ring;
 
 		if (!amdgpu_sriov_vf(adev)) {
@@ -579,8 +608,11 @@ static int uvd_v7_0_hw_fini(void *handle)
 		DRM_DEBUG("For SRIOV client, shouldn't do anything.\n");
 	}
 
-	for (i = 0; i < adev->uvd.num_uvd_inst; ++i)
+	for (i = 0; i < adev->uvd.num_uvd_inst; ++i) {
+		if (adev->uvd.harvest_config & (1 << i))
+			continue;
 		adev->uvd.inst[i].ring.ready = false;
+	}
 
 	return 0;
 }
@@ -623,6 +655,8 @@ static void uvd_v7_0_mc_resume(struct amdgpu_device *adev)
 	int i;
 
 	for (i = 0; i < adev->uvd.num_uvd_inst; ++i) {
+		if (adev->uvd.harvest_config & (1 << i))
+			continue;
 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
 			WREG32_SOC15(UVD, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
 				lower_32_bits(adev->firmware.ucode[AMDGPU_UCODE_ID_UVD].mc_addr));
@@ -695,6 +729,8 @@ static int uvd_v7_0_mmsch_start(struct amdgpu_device *adev,
 	WREG32_SOC15(VCE, 0, mmVCE_MMSCH_VF_MAILBOX_RESP, 0);
 
 	for (i = 0; i < adev->uvd.num_uvd_inst; ++i) {
+		if (adev->uvd.harvest_config & (1 << i))
+			continue;
 		WDOORBELL32(adev->uvd.inst[i].ring_enc[0].doorbell_index, 0);
 		adev->wb.wb[adev->uvd.inst[i].ring_enc[0].wptr_offs] = 0;
 		adev->uvd.inst[i].ring_enc[0].wptr = 0;
@@ -751,6 +787,8 @@ static int uvd_v7_0_sriov_start(struct amdgpu_device *adev)
 		init_table += header->uvd_table_offset;
 
 		for (i = 0; i < adev->uvd.num_uvd_inst; ++i) {
+			if (adev->uvd.harvest_config & (1 << i))
+				continue;
 			ring = &adev->uvd.inst[i].ring;
 			ring->wptr = 0;
 			size = AMDGPU_GPU_PAGE_ALIGN(adev->uvd.fw->size + 4);
@@ -890,6 +928,8 @@ static int uvd_v7_0_start(struct amdgpu_device *adev)
 	int i, j, k, r;
 
 	for (k = 0; k < adev->uvd.num_uvd_inst; ++k) {
+		if (adev->uvd.harvest_config & (1 << k))
+			continue;
 		/* disable DPG */
 		WREG32_P(SOC15_REG_OFFSET(UVD, k, mmUVD_POWER_STATUS), 0,
 				~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
@@ -902,6 +942,8 @@ static int uvd_v7_0_start(struct amdgpu_device *adev)
 	uvd_v7_0_mc_resume(adev);
 
 	for (k = 0; k < adev->uvd.num_uvd_inst; ++k) {
+		if (adev->uvd.harvest_config & (1 << k))
+			continue;
 		ring = &adev->uvd.inst[k].ring;
 		/* disable clock gating */
 		WREG32_P(SOC15_REG_OFFSET(UVD, k, mmUVD_CGC_CTRL), 0,
@@ -1069,6 +1111,8 @@ static void uvd_v7_0_stop(struct amdgpu_device *adev)
 	uint8_t i = 0;
 
 	for (i = 0; i < adev->uvd.num_uvd_inst; ++i) {
+		if (adev->uvd.harvest_config & (1 << i))
+			continue;
 		/* force RBC into idle state */
 		WREG32_SOC15(UVD, i, mmUVD_RBC_RB_CNTL, 0x11010101);
 
@@ -1785,6 +1829,8 @@ static void uvd_v7_0_set_ring_funcs(struct amdgpu_device *adev)
 	int i;
 
 	for (i = 0; i < adev->uvd.num_uvd_inst; i++) {
+		if (adev->uvd.harvest_config & (1 << i))
+			continue;
 		adev->uvd.inst[i].ring.funcs = &uvd_v7_0_ring_vm_funcs;
 		adev->uvd.inst[i].ring.me = i;
 		DRM_INFO("UVD(%d) is enabled in VM mode\n", i);
@@ -1796,6 +1842,8 @@ static void uvd_v7_0_set_enc_ring_funcs(struct amdgpu_device *adev)
 	int i, j;
 
 	for (j = 0; j < adev->uvd.num_uvd_inst; j++) {
+		if (adev->uvd.harvest_config & (1 << j))
+			continue;
 		for (i = 0; i < adev->uvd.num_enc_rings; ++i) {
 			adev->uvd.inst[j].ring_enc[i].funcs = &uvd_v7_0_enc_ring_vm_funcs;
 			adev->uvd.inst[j].ring_enc[i].me = j;
@@ -1815,6 +1863,8 @@ static void uvd_v7_0_set_irq_funcs(struct amdgpu_device *adev)
 	int i;
 
 	for (i = 0; i < adev->uvd.num_uvd_inst; i++) {
+		if (adev->uvd.harvest_config & (1 << i))
+			continue;
 		adev->uvd.inst[i].irq.num_types = adev->uvd.num_enc_rings + 1;
 		adev->uvd.inst[i].irq.funcs = &uvd_v7_0_irq_funcs;
 	}