[PATCH 3/3] drm/amdgpu: Add JPEG v5.0.2 SRIOV code path

From: Bokun Zhang <[email protected]>

- Add JPEG v5.0.2 SRIOV code path
- Add JPEG MMSCH init table transfer
- Skip unsupported actions such as power gating
  and per queue reset for SRIOV VF.

Signed-off-by: Bokun Zhang <[email protected]>
Reviewed-by: Alex Deucher <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
---
 drivers/gpu/drm/amd/amdgpu/jpeg_v5_0_2.c | 156 ++++++++++++++++++++++-
 1 file changed, 152 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/jpeg_v5_0_2.c b/drivers/gpu/drm/amd/amdgpu/jpeg_v5_0_2.c
index e79ffec251d75..dd0abce73687c 100644
--- a/drivers/gpu/drm/amd/amdgpu/jpeg_v5_0_2.c
+++ b/drivers/gpu/drm/amd/amdgpu/jpeg_v5_0_2.c
@@ -34,6 +34,7 @@
 #include "vcn/vcn_5_0_0_sh_mask.h"
 #include "ivsrcid/vcn/irqsrcs_vcn_5_0.h"
 
+static int jpeg_v5_0_2_start_sriov(struct amdgpu_device *adev);
 static void jpeg_v5_0_2_set_dec_ring_funcs(struct amdgpu_device *adev);
 static void jpeg_v5_0_2_set_irq_funcs(struct amdgpu_device *adev);
 static int jpeg_v5_0_2_set_powergating_state(struct amdgpu_ip_block *ip_block,
@@ -164,9 +165,15 @@ static int jpeg_v5_0_2_sw_init(struct amdgpu_ip_block *ip_block)
 			ring = &adev->jpeg.inst[i].ring_dec[j];
 			ring->use_doorbell = true;
 			ring->vm_hub = AMDGPU_MMHUB0(adev->jpeg.inst[i].aid_id);
-			ring->doorbell_index =
-				(adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
-				1 + j + 11 * jpeg_inst;
+			if (amdgpu_sriov_vf(adev)) {
+				ring->doorbell_index =
+					(adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
+					2 + j + 32 * jpeg_inst;
+			} else {
+				ring->doorbell_index =
+					(adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
+					1 + j + 11 * jpeg_inst;
+			}
 			sprintf(ring->name, "jpeg_dec_%d.%d", i, j);
 			r = amdgpu_ring_init(adev, ring, 512, &adev->jpeg.inst->irq, 0,
 					     AMDGPU_RING_PRIO_DEFAULT, NULL);
@@ -187,7 +194,8 @@ static int jpeg_v5_0_2_sw_init(struct amdgpu_ip_block *ip_block)
 
 	adev->jpeg.supported_reset =
 		amdgpu_get_soft_full_reset_mask(&adev->jpeg.inst[0].ring_dec[0]);
-	adev->jpeg.supported_reset |= AMDGPU_RESET_TYPE_PER_QUEUE;
+	if (!amdgpu_sriov_vf(adev))
+		adev->jpeg.supported_reset |= AMDGPU_RESET_TYPE_PER_QUEUE;
 	r = amdgpu_jpeg_sysfs_reset_mask_init(adev);
 
 	return r;
@@ -228,6 +236,23 @@ static int jpeg_v5_0_2_hw_init(struct amdgpu_ip_block *ip_block)
 	struct amdgpu_ring *ring;
 	int i, j, r, jpeg_inst, tmp;
 
+	if (amdgpu_sriov_vf(adev)) {
+		r = jpeg_v5_0_2_start_sriov(adev);
+		if (r)
+			return r;
+
+		for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) {
+			for (j = 0; j < adev->jpeg.num_jpeg_rings; ++j) {
+				ring = &adev->jpeg.inst[i].ring_dec[j];
+				ring->wptr = 0;
+				ring->wptr_old = 0;
+				jpeg_v5_0_2_dec_ring_set_wptr(ring);
+				ring->sched.ready = true;
+			}
+		}
+		return 0;
+	}
+
 	if (RREG32_SOC15(VCN, GET_INST(VCN, 0), regVCN_RRMT_CNTL) & 0x100)
 		adev->jpeg.caps |= AMDGPU_JPEG_CAPS(RRMT_ENABLED);
 
@@ -276,8 +301,10 @@ static int jpeg_v5_0_2_hw_fini(struct amdgpu_ip_block *ip_block)
 
 	cancel_delayed_work_sync(&adev->jpeg.idle_work);
 
+	if (!amdgpu_sriov_vf(adev)) {
 		if (adev->jpeg.cur_state != AMD_PG_STATE_GATE)
 			ret = jpeg_v5_0_2_set_powergating_state(ip_block, AMD_PG_STATE_GATE);
+	}
 
 	return ret;
 }
@@ -407,6 +434,122 @@ static void jpeg_v5_0_2_init_jrbc(struct amdgpu_ring *ring)
 					 reg_offset);
 }
 
+/**
+ * jpeg_v5_0_2_start_sriov - start JPEG block under SRIOV env
+ */
+static int jpeg_v5_0_2_start_sriov(struct amdgpu_device *adev)
+{
+	struct amdgpu_ring *ring;
+	uint64_t ctx_addr;
+	uint32_t param, resp, expected;
+	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, item_offset;
+	uint32_t init_status;
+	int i, j, jpeg_inst;
+
+	struct mmsch_v5_0_cmd_direct_write
+		direct_wt = { {0} };
+	struct mmsch_v5_0_cmd_end end = { {0} };
+	struct mmsch_v5_0_init_header header;
+
+	direct_wt.cmd_header.command_type =
+		MMSCH_COMMAND__DIRECT_REG_WRITE;
+	end.cmd_header.command_type =
+		MMSCH_COMMAND__END;
+
+	for (i = 0; i < adev->jpeg.num_jpeg_inst; i++) {
+		jpeg_inst = GET_INST(JPEG, i);
+
+		memset(&header, 0, sizeof(struct mmsch_v5_0_init_header));
+		header.version = MMSCH_VERSION;
+		header.total_size = sizeof(struct mmsch_v5_0_init_header) >> 2;
+
+		table_loc = (uint32_t *)table->cpu_addr;
+		table_loc += header.total_size;
+
+		item_offset = header.total_size;
+
+		for (j = 0; j < adev->jpeg.num_jpeg_rings; j++) {
+			ring = &adev->jpeg.inst[i].ring_dec[j];
+			table_size = 0;
+
+			tmp = SOC15_REG_OFFSET(JPEG, 0, regUVD_LMI_JRBC_RB_64BIT_BAR_LOW);
+			MMSCH_V5_0_INSERT_DIRECT_WT(tmp, lower_32_bits(ring->gpu_addr));
+			tmp = SOC15_REG_OFFSET(JPEG, 0, regUVD_LMI_JRBC_RB_64BIT_BAR_HIGH);
+			MMSCH_V5_0_INSERT_DIRECT_WT(tmp, upper_32_bits(ring->gpu_addr));
+			tmp = SOC15_REG_OFFSET(JPEG, 0, regUVD_JRBC_RB_SIZE);
+			MMSCH_V5_0_INSERT_DIRECT_WT(tmp, ring->ring_size / 4);
+
+			if (j < 5) {
+				header.mjpegdec0[j].table_offset = item_offset;
+				header.mjpegdec0[j].init_status = 0;
+				header.mjpegdec0[j].table_size = table_size;
+			} else {
+				header.mjpegdec1[j - 5].table_offset = item_offset;
+				header.mjpegdec1[j - 5].init_status = 0;
+				header.mjpegdec1[j - 5].table_size = table_size;
+			}
+			header.total_size += table_size;
+			item_offset += table_size;
+		}
+
+		MMSCH_V5_0_INSERT_END();
+
+		/* send init table to MMSCH */
+		size = sizeof(struct mmsch_v5_0_init_header);
+		table_loc = (uint32_t *)table->cpu_addr;
+		memcpy((void *)table_loc, &header, size);
+
+		ctx_addr = table->gpu_addr;
+		WREG32_SOC15(VCN, jpeg_inst, regMMSCH_VF_CTX_ADDR_LO, lower_32_bits(ctx_addr));
+		WREG32_SOC15(VCN, jpeg_inst, regMMSCH_VF_CTX_ADDR_HI, upper_32_bits(ctx_addr));
+
+		tmp = RREG32_SOC15(VCN, jpeg_inst, regMMSCH_VF_VMID);
+		tmp &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
+		tmp |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
+		WREG32_SOC15(VCN, jpeg_inst, regMMSCH_VF_VMID, tmp);
+
+		size = header.total_size;
+		WREG32_SOC15(VCN, jpeg_inst, regMMSCH_VF_CTX_SIZE, size);
+
+		WREG32_SOC15(VCN, jpeg_inst, regMMSCH_VF_MAILBOX_RESP, 0);
+
+		param = 0x00000001;
+		WREG32_SOC15(VCN, jpeg_inst, regMMSCH_VF_MAILBOX_HOST, param);
+		tmp = 0;
+		timeout = amdgpu_emu_mode ? 1000000 : 1000;
+		resp = 0;
+		expected = MMSCH_VF_MAILBOX_RESP__OK;
+		init_status =
+			((struct mmsch_v5_0_init_header *)(table_loc))->mjpegdec0[i].init_status;
+		while (resp != expected) {
+			resp = RREG32_SOC15(VCN, jpeg_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;
+			}
+		}
+		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 jpeg: %x\n",
+					resp, init_status);
+
+	}
+	return 0;
+}
+
 /**
  * jpeg_v5_0_2_start - start JPEG block
  *
@@ -566,6 +709,11 @@ static int jpeg_v5_0_2_set_powergating_state(struct amdgpu_ip_block *ip_block,
 	struct amdgpu_device *adev = ip_block->adev;
 	int ret;
 
+	if (amdgpu_sriov_vf(adev)) {
+		adev->jpeg.cur_state = AMD_PG_STATE_UNGATE;
+		return 0;
+	}
+
 	if (state == adev->jpeg.cur_state)
 		return 0;
 
-- 
2.55.0