drm/amdgpu/vcn: sriov support for vcn_v4_0_3

initialization table handshake with mmsch

Signed-off-by: Samir Dhume <samir.dhume@amd.com>
Acked-by: Leo Liu <leo.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Author: Samir Dhume

drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c

@@ -31,6 +31,7 @@
 #include "soc15d.h"
 #include "soc15_hw_ip.h"
 #include "vcn_v2_0.h"
+#include "mmsch_v4_0_3.h"
 
 #include "vcn/vcn_4_0_3_offset.h"
 #include "vcn/vcn_4_0_3_sh_mask.h"
@@ -44,6 +45,7 @@
 #define VCN_VID_SOC_ADDRESS_2_0		0x1fb00
 #define VCN1_VID_SOC_ADDRESS_3_0	0x48300
 
+static int vcn_v4_0_3_start_sriov(struct amdgpu_device *adev);
 static void vcn_v4_0_3_set_unified_ring_funcs(struct amdgpu_device *adev);
 static void vcn_v4_0_3_set_irq_funcs(struct amdgpu_device *adev);
 static int vcn_v4_0_3_set_powergating_state(void *handle,
@@ -137,6 +139,12 @@ static int vcn_v4_0_3_sw_init(void *handle)
 			amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
 	}
 
+	if (amdgpu_sriov_vf(adev)) {
+		r = amdgpu_virt_alloc_mm_table(adev);
+		if (r)
+			return r;
+	}
+
 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
 		adev->vcn.pause_dpg_mode = vcn_v4_0_3_pause_dpg_mode;
 
@@ -174,6 +182,9 @@ static int vcn_v4_0_3_sw_fini(void *handle)
 		drm_dev_exit(idx);
 	}
 
+	if (amdgpu_sriov_vf(adev))
+		amdgpu_virt_free_mm_table(adev);
+
 	r = amdgpu_vcn_suspend(adev);
 	if (r)
 		return r;
@@ -196,33 +207,47 @@ static int vcn_v4_0_3_hw_init(void *handle)
 	struct amdgpu_ring *ring;
 	int i, r, vcn_inst;
 
-	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
-		vcn_inst = GET_INST(VCN, i);
-		ring = &adev->vcn.inst[i].ring_enc[0];
+	if (amdgpu_sriov_vf(adev)) {
+		r = vcn_v4_0_3_start_sriov(adev);
+		if (r)
+			goto done;
 
-		if (ring->use_doorbell) {
-			adev->nbio.funcs->vcn_doorbell_range(
-				adev, ring->use_doorbell,
-				(adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
-					9 * vcn_inst,
-				adev->vcn.inst[i].aid_id);
-
-			WREG32_SOC15(
-				VCN, GET_INST(VCN, ring->me),
-				regVCN_RB1_DB_CTRL,
-				ring->doorbell_index
-						<< VCN_RB1_DB_CTRL__OFFSET__SHIFT |
-					VCN_RB1_DB_CTRL__EN_MASK);
-
-			/* Read DB_CTRL to flush the write DB_CTRL command. */
-			RREG32_SOC15(
-				VCN, GET_INST(VCN, ring->me),
-				regVCN_RB1_DB_CTRL);
+		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
+			ring = &adev->vcn.inst[i].ring_enc[0];
+			ring->wptr = 0;
+			ring->wptr_old = 0;
+			vcn_v4_0_3_unified_ring_set_wptr(ring);
+			ring->sched.ready = true;
 		}
+	} else {
+		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
+			vcn_inst = GET_INST(VCN, i);
+			ring = &adev->vcn.inst[i].ring_enc[0];
+
+			if (ring->use_doorbell) {
+				adev->nbio.funcs->vcn_doorbell_range(
+					adev, ring->use_doorbell,
+					(adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
+						9 * vcn_inst,
+					adev->vcn.inst[i].aid_id);
+
+				WREG32_SOC15(
+					VCN, GET_INST(VCN, ring->me),
+					regVCN_RB1_DB_CTRL,
+					ring->doorbell_index
+							<< VCN_RB1_DB_CTRL__OFFSET__SHIFT |
+						VCN_RB1_DB_CTRL__EN_MASK);
+
+				/* Read DB_CTRL to flush the write DB_CTRL command. */
+				RREG32_SOC15(
+					VCN, GET_INST(VCN, ring->me),
+					regVCN_RB1_DB_CTRL);
+			}
 
-		r = amdgpu_ring_test_helper(ring);
-		if (r)
-			goto done;
+			r = amdgpu_ring_test_helper(ring);
+			if (r)
+				goto done;
+		}
 	}
 
 done:
@@ -820,6 +845,193 @@ static int vcn_v4_0_3_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, b
 	return 0;
 }
 
+static int vcn_v4_0_3_start_sriov(struct amdgpu_device *adev)
+{
+	int i, vcn_inst;
+	struct amdgpu_ring *ring_enc;
+	uint64_t cache_addr;
+	uint64_t rb_enc_addr;
+	uint64_t ctx_addr;
+	uint32_t param, resp, expected;
+	uint32_t offset, cache_size;
+	uint32_t tmp, timeout;
+
+	struct amdgpu_mm_table *table = &adev->virt.mm_table;
+	uint32_t *table_loc;
+	uint32_t table_size;
+	uint32_t size, size_dw;
+	uint32_t init_status;
+	uint32_t enabled_vcn;
+
+	struct mmsch_v4_0_cmd_direct_write
+		direct_wt = { {0} };
+	struct mmsch_v4_0_cmd_direct_read_modify_write
+		direct_rd_mod_wt = { {0} };
+	struct mmsch_v4_0_cmd_end end = { {0} };
+	struct mmsch_v4_0_3_init_header header;
+
+	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
+	volatile struct amdgpu_fw_shared_rb_setup *rb_setup;
+
+	direct_wt.cmd_header.command_type =
+		MMSCH_COMMAND__DIRECT_REG_WRITE;
+	direct_rd_mod_wt.cmd_header.command_type =
+		MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
+	end.cmd_header.command_type = MMSCH_COMMAND__END;
+
+	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
+		vcn_inst = GET_INST(VCN, i);
+
+		memset(&header, 0, sizeof(struct mmsch_v4_0_3_init_header));
+		header.version = MMSCH_VERSION;
+		header.total_size = sizeof(struct mmsch_v4_0_3_init_header) >> 2;
+
+		table_loc = (uint32_t *)table->cpu_addr;
+		table_loc += header.total_size;
+
+		table_size = 0;
+
+		MMSCH_V4_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(VCN, 0, regUVD_STATUS),
+			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
+
+		cache_size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
+
+		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
+			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
+				regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
+				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
+
+			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
+				regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
+				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
+
+			offset = 0;
+			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
+				regUVD_VCPU_CACHE_OFFSET0), 0);
+		} else {
+			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
+				regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
+				lower_32_bits(adev->vcn.inst[i].gpu_addr));
+			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
+				regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
+				upper_32_bits(adev->vcn.inst[i].gpu_addr));
+			offset = cache_size;
+			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
+				regUVD_VCPU_CACHE_OFFSET0),
+				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
+		}
+
+		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
+			regUVD_VCPU_CACHE_SIZE0),
+			cache_size);
+
+		cache_addr = adev->vcn.inst[vcn_inst].gpu_addr + offset;
+		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
+			regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), lower_32_bits(cache_addr));
+		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
+			regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), upper_32_bits(cache_addr));
+		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
+			regUVD_VCPU_CACHE_OFFSET1), 0);
+		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
+			regUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE);
+
+		cache_addr = adev->vcn.inst[vcn_inst].gpu_addr + offset +
+			AMDGPU_VCN_STACK_SIZE;
+
+		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
+			regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW), lower_32_bits(cache_addr));
+
+		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
+			regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH), upper_32_bits(cache_addr));
+
+		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
+			regUVD_VCPU_CACHE_OFFSET2), 0);
+
+		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
+			regUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE);
+
+		fw_shared = adev->vcn.inst[vcn_inst].fw_shared.cpu_addr;
+		rb_setup = &fw_shared->rb_setup;
+
+		ring_enc = &adev->vcn.inst[vcn_inst].ring_enc[0];
+		ring_enc->wptr = 0;
+		rb_enc_addr = ring_enc->gpu_addr;
+
+		rb_setup->is_rb_enabled_flags |= RB_ENABLED;
+		rb_setup->rb_addr_lo = lower_32_bits(rb_enc_addr);
+		rb_setup->rb_addr_hi = upper_32_bits(rb_enc_addr);
+		rb_setup->rb_size = ring_enc->ring_size / 4;
+		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
+
+		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
+			regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
+			lower_32_bits(adev->vcn.inst[vcn_inst].fw_shared.gpu_addr));
+		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
+			regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
+			upper_32_bits(adev->vcn.inst[vcn_inst].fw_shared.gpu_addr));
+		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
+			regUVD_VCPU_NONCACHE_SIZE0),
+			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
+		MMSCH_V4_0_INSERT_END();
+
+		header.vcn0.init_status = 0;
+		header.vcn0.table_offset = header.total_size;
+		header.vcn0.table_size = table_size;
+		header.total_size += table_size;
+
+		/* Send init table to mmsch */
+		size = sizeof(struct mmsch_v4_0_3_init_header);
+		table_loc = (uint32_t *)table->cpu_addr;
+		memcpy((void *)table_loc, &header, size);
+
+		ctx_addr = table->gpu_addr;
+		WREG32_SOC15(VCN, vcn_inst, regMMSCH_VF_CTX_ADDR_LO, lower_32_bits(ctx_addr));
+		WREG32_SOC15(VCN, vcn_inst, regMMSCH_VF_CTX_ADDR_HI, upper_32_bits(ctx_addr));
+
+		tmp = RREG32_SOC15(VCN, vcn_inst, regMMSCH_VF_VMID);
+		tmp &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
+		tmp |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
+		WREG32_SOC15(VCN, vcn_inst, regMMSCH_VF_VMID, tmp);
+
+		size = header.total_size;
+		WREG32_SOC15(VCN, vcn_inst, regMMSCH_VF_CTX_SIZE, size);
+
+		WREG32_SOC15(VCN, vcn_inst, regMMSCH_VF_MAILBOX_RESP, 0);
+
+		param = 0x00000001;
+		WREG32_SOC15(VCN, vcn_inst, regMMSCH_VF_MAILBOX_HOST, param);
+		tmp = 0;
+		timeout = 1000;
+		resp = 0;
+		expected = MMSCH_VF_MAILBOX_RESP__OK;
+		while (resp != expected) {
+			resp = RREG32_SOC15(VCN, vcn_inst, regMMSCH_VF_MAILBOX_RESP);
+			if (resp != 0)
+				break;
+
+			udelay(10);
+			tmp = tmp + 10;
+			if (tmp >= timeout) {
+				DRM_ERROR("failed to init MMSCH. TIME-OUT after %d usec"\
+					" waiting for regMMSCH_VF_MAILBOX_RESP "\
+					"(expected=0x%08x, readback=0x%08x)\n",
+					tmp, expected, resp);
+				return -EBUSY;
+			}
+		}
+
+		enabled_vcn = amdgpu_vcn_is_disabled_vcn(adev, VCN_DECODE_RING, 0) ? 1 : 0;
+		init_status = ((struct mmsch_v4_0_3_init_header *)(table_loc))->vcn0.init_status;
+		if (resp != expected && resp != MMSCH_VF_MAILBOX_RESP__INCOMPLETE
+					&& init_status != MMSCH_VF_ENGINE_STATUS__PASS) {
+			DRM_ERROR("MMSCH init status is incorrect! readback=0x%08x, header init "\
+				"status for VCN%x: 0x%x\n", resp, enabled_vcn, init_status);
+		}
+	}
+
+	return 0;
+}
+
 /**
  * vcn_v4_0_3_start - VCN start
  *