drm/amdgpu: add VCN4 ip block support
authorLeo Liu <leo.liu@amd.com>
Wed, 13 Apr 2022 15:45:03 +0000 (11:45 -0400)
committerAlex Deucher <alexander.deucher@amd.com>
Wed, 4 May 2022 14:43:56 +0000 (10:43 -0400)
Add VCN 4.0 initialization and decoder/encoder ring functions.

Signed-off-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: James Zhu <James.Zhu@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drivers/gpu/drm/amd/amdgpu/Makefile
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c [new file with mode: 0644]
drivers/gpu/drm/amd/amdgpu/vcn_v4_0.h [new file with mode: 0644]

index 81088b0..0659391 100644 (file)
@@ -171,6 +171,7 @@ amdgpu-y += \
        vcn_v2_0.o \
        vcn_v2_5.o \
        vcn_v3_0.o \
+       vcn_v4_0.o \
        amdgpu_jpeg.o \
        jpeg_v1_0.o \
        jpeg_v2_0.o \
index a0ee828..29f26db 100644 (file)
@@ -53,6 +53,7 @@
 #define FIRMWARE_BEIGE_GOBY    "amdgpu/beige_goby_vcn.bin"
 #define FIRMWARE_YELLOW_CARP   "amdgpu/yellow_carp_vcn.bin"
 #define FIRMWARE_VCN_3_1_2     "amdgpu/vcn_3_1_2.bin"
+#define FIRMWARE_VCN4_0_0      "amdgpu/vcn_4_0_0.bin"
 
 MODULE_FIRMWARE(FIRMWARE_RAVEN);
 MODULE_FIRMWARE(FIRMWARE_PICASSO);
@@ -71,6 +72,7 @@ MODULE_FIRMWARE(FIRMWARE_DIMGREY_CAVEFISH);
 MODULE_FIRMWARE(FIRMWARE_BEIGE_GOBY);
 MODULE_FIRMWARE(FIRMWARE_YELLOW_CARP);
 MODULE_FIRMWARE(FIRMWARE_VCN_3_1_2);
+MODULE_FIRMWARE(FIRMWARE_VCN4_0_0);
 
 static void amdgpu_vcn_idle_work_handler(struct work_struct *work);
 
@@ -175,6 +177,12 @@ int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
                    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
                        adev->vcn.indirect_sram = true;
                break;
+       case IP_VERSION(4, 0, 0):
+               fw_name = FIRMWARE_VCN4_0_0;
+               if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
+                       (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
+                       adev->vcn.indirect_sram = true;
+               break;
        default:
                return -EINVAL;
        }
@@ -228,8 +236,15 @@ int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
        bo_size = AMDGPU_VCN_STACK_SIZE + AMDGPU_VCN_CONTEXT_SIZE;
        if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
                bo_size += AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
-       fw_shared_size = AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
-       log_offset = offsetof(struct amdgpu_fw_shared, fw_log);
+
+       if (adev->ip_versions[UVD_HWIP][0] >= IP_VERSION(4, 0, 0)){
+               fw_shared_size = AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared));
+               log_offset = offsetof(struct amdgpu_vcn4_fw_shared, fw_log);
+       } else {
+               fw_shared_size = AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
+               log_offset = offsetof(struct amdgpu_fw_shared, fw_log);
+       }
+
        bo_size += fw_shared_size;
 
        if (amdgpu_vcnfw_log)
index 912ead2..8e7aec8 100644 (file)
                }                                                                               \
        } while (0)
 
+#define AMDGPU_FW_SHARED_FLAG_0_UNIFIED_QUEUE (1 << 2)
 #define AMDGPU_VCN_FW_SHARED_FLAG_0_RB (1 << 6)
 #define AMDGPU_VCN_MULTI_QUEUE_FLAG    (1 << 8)
 #define AMDGPU_VCN_SW_RING_FLAG                (1 << 9)
@@ -286,6 +287,13 @@ struct amdgpu_fw_shared_sw_ring {
        uint8_t padding[3];
 };
 
+struct amdgpu_fw_shared_unified_queue_struct {
+       uint8_t is_enabled;
+       uint8_t queue_mode;
+       uint8_t queue_status;
+       uint8_t padding[5];
+};
+
 struct amdgpu_fw_shared_fw_logging {
        uint8_t is_enabled;
        uint32_t addr_lo;
@@ -309,6 +317,14 @@ struct amdgpu_fw_shared {
        struct amdgpu_fw_shared_smu_interface_info smu_interface_info;
 };
 
+struct amdgpu_vcn4_fw_shared {
+       uint32_t present_flag_0;
+       uint8_t pad[12];
+       struct amdgpu_fw_shared_unified_queue_struct sq;
+       uint8_t pad1[8];
+       struct amdgpu_fw_shared_fw_logging fw_log;
+};
+
 struct amdgpu_vcn_fwlog {
        uint32_t rptr;
        uint32_t wptr;
diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c
new file mode 100644 (file)
index 0000000..dada3e3
--- /dev/null
@@ -0,0 +1,1877 @@
+/*
+ * Copyright 2021 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <linux/firmware.h>
+#include "amdgpu.h"
+#include "amdgpu_vcn.h"
+#include "amdgpu_pm.h"
+#include "soc15.h"
+#include "soc15d.h"
+#include "soc15_hw_ip.h"
+#include "vcn_v2_0.h"
+#include "vcn_v3_0.h"
+
+#include "vcn/vcn_4_0_0_offset.h"
+#include "vcn/vcn_4_0_0_sh_mask.h"
+#include "ivsrcid/vcn/irqsrcs_vcn_4_0.h"
+
+#include <drm/drm_drv.h>
+
+#define mmUVD_DPG_LMA_CTL                                                      regUVD_DPG_LMA_CTL
+#define mmUVD_DPG_LMA_CTL_BASE_IDX                                             regUVD_DPG_LMA_CTL_BASE_IDX
+#define mmUVD_DPG_LMA_DATA                                                     regUVD_DPG_LMA_DATA
+#define mmUVD_DPG_LMA_DATA_BASE_IDX                                            regUVD_DPG_LMA_DATA_BASE_IDX
+
+#define VCN_VID_SOC_ADDRESS_2_0                                                        0x1fb00
+#define VCN1_VID_SOC_ADDRESS_3_0                                               0x48300
+
+bool unifiedQ_enabled = false;
+
+static int amdgpu_ih_clientid_vcns[] = {
+       SOC15_IH_CLIENTID_VCN,
+       SOC15_IH_CLIENTID_VCN1
+};
+
+static void vcn_v4_0_set_dec_ring_funcs(struct amdgpu_device *adev);
+static void vcn_v4_0_set_enc_ring_funcs(struct amdgpu_device *adev);
+static void vcn_v4_0_set_irq_funcs(struct amdgpu_device *adev);
+static int vcn_v4_0_set_powergating_state(void *handle,
+        enum amd_powergating_state state);
+static int vcn_v4_0_pause_dpg_mode(struct amdgpu_device *adev,
+        int inst_idx, struct dpg_pause_state *new_state);
+
+/**
+ * vcn_v4_0_early_init - set function pointers
+ *
+ * @handle: amdgpu_device pointer
+ *
+ * Set ring and irq function pointers
+ */
+static int vcn_v4_0_early_init(void *handle)
+{
+       struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+
+       if (unifiedQ_enabled) {
+               adev->vcn.num_vcn_inst = 1;
+               adev->vcn.num_enc_rings = 1;
+       } else {
+               adev->vcn.num_enc_rings = 2;
+       }
+
+       if (!unifiedQ_enabled)
+               vcn_v4_0_set_dec_ring_funcs(adev);
+
+       vcn_v4_0_set_enc_ring_funcs(adev);
+       vcn_v4_0_set_irq_funcs(adev);
+
+       return 0;
+}
+
+static void amdgpu_vcn_setup_unified_queue_ucode(struct amdgpu_device *adev)
+{
+       if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
+               const struct common_firmware_header *hdr;
+
+               hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
+               adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].ucode_id = AMDGPU_UCODE_ID_VCN;
+               adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].fw = adev->vcn.fw;
+               adev->firmware.fw_size +=
+                       ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
+               DRM_INFO("PSP loading VCN firmware\n");
+       }
+}
+
+/**
+ * vcn_v4_0_sw_init - sw init for VCN block
+ *
+ * @handle: amdgpu_device pointer
+ *
+ * Load firmware and sw initialization
+ */
+static int vcn_v4_0_sw_init(void *handle)
+{
+       struct amdgpu_ring *ring;
+       int i, j, r;
+       struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+
+       r = amdgpu_vcn_sw_init(adev);
+       if (r)
+               return r;
+
+       if (unifiedQ_enabled)
+               amdgpu_vcn_setup_unified_queue_ucode(adev);
+       else
+               amdgpu_vcn_setup_ucode(adev);
+
+       r = amdgpu_vcn_resume(adev);
+       if (r)
+               return r;
+
+       for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
+               volatile struct amdgpu_vcn4_fw_shared *fw_shared;
+               if (adev->vcn.harvest_config & (1 << i))
+                       continue;
+               /* VCN DEC TRAP */
+               r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
+                               VCN_4_0__SRCID__UVD_TRAP, &adev->vcn.inst[i].irq);
+               if (r)
+                       return r;
+
+               atomic_set(&adev->vcn.inst[i].sched_score, 0);
+               if (!unifiedQ_enabled) {
+                       ring = &adev->vcn.inst[i].ring_dec;
+                       ring->use_doorbell = true;
+
+                       /* VCN4 doorbell layout
+                        * 1: VCN_JPEG_DB_CTRL UVD_JRBC_RB_WPTR; (jpeg)
+                        * 2: VCN_RB1_DB_CTRL  UVD_RB_WPTR; (decode/encode for unified queue)
+                        * 3: VCN_RB2_DB_CTRL  UVD_RB_WPTR2; (encode only for swqueue)
+                        * 4: VCN_RB3_DB_CTRL  UVD_RB_WPTR3; (Reserved)
+                        * 5: VCN_RB4_DB_CTRL  UVD_RB_WPTR4; (decode only for swqueue)
+                        */
+
+                       ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1)
+                                               + 5 + 8 * i;
+
+                       sprintf(ring->name, "vcn_dec_%d", i);
+                       r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
+                                            AMDGPU_RING_PRIO_DEFAULT,
+                                            &adev->vcn.inst[i].sched_score);
+                       if (r)
+                               return r;
+               }
+               for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
+                       /* VCN ENC TRAP */
+                       r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
+                               j + VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);
+                       if (r)
+                               return r;
+
+                       ring = &adev->vcn.inst[i].ring_enc[j];
+                       ring->use_doorbell = true;
+
+                       ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + j + 8 * i;
+
+                       if (unifiedQ_enabled) {
+                               sprintf(ring->name, "vcn_unified%d", i);
+                               r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0,
+                                    AMDGPU_RING_PRIO_DEFAULT, NULL);
+                       } else {
+                               enum amdgpu_ring_priority_level hw_prio;
+
+                               hw_prio = amdgpu_vcn_get_enc_ring_prio(j);
+                               sprintf(ring->name, "vcn_enc_%d.%d", i, j);
+                               r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
+                                            hw_prio, &adev->vcn.inst[i].sched_score);
+                       }
+                       if (r)
+                               return r;
+               }
+
+               fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
+               fw_shared->present_flag_0 = 0;
+
+               if (unifiedQ_enabled) {
+                       fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_FW_SHARED_FLAG_0_UNIFIED_QUEUE);
+                       fw_shared->sq.is_enabled = 1;
+               }
+
+               if (amdgpu_vcnfw_log)
+                       amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
+       }
+
+       if (!unifiedQ_enabled) {
+               if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
+                       adev->vcn.pause_dpg_mode = vcn_v4_0_pause_dpg_mode;
+       }
+       return 0;
+}
+
+/**
+ * vcn_v4_0_sw_fini - sw fini for VCN block
+ *
+ * @handle: amdgpu_device pointer
+ *
+ * VCN suspend and free up sw allocation
+ */
+static int vcn_v4_0_sw_fini(void *handle)
+{
+       struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+       int i, r, idx;
+
+       if (drm_dev_enter(&adev->ddev, &idx)) {
+                for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
+                        volatile struct amdgpu_vcn4_fw_shared *fw_shared;
+
+                        if (adev->vcn.harvest_config & (1 << i))
+                                continue;
+
+                        fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
+                        fw_shared->present_flag_0 = 0;
+                        fw_shared->sq.is_enabled = 0;
+                }
+
+                drm_dev_exit(idx);
+        }
+
+       r = amdgpu_vcn_suspend(adev);
+       if (r)
+               return r;
+
+       r = amdgpu_vcn_sw_fini(adev);
+
+       return r;
+}
+
+/**
+ * vcn_v4_0_hw_init - start and test VCN block
+ *
+ * @handle: amdgpu_device pointer
+ *
+ * Initialize the hardware, boot up the VCPU and do some testing
+ */
+static int vcn_v4_0_hw_init(void *handle)
+{
+       struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+       struct amdgpu_ring *ring;
+       int i, j, r;
+
+       for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
+               if (adev->vcn.harvest_config & (1 << i))
+                       continue;
+               if (unifiedQ_enabled)
+                       ring = &adev->vcn.inst[i].ring_enc[0];
+               else
+                       ring = &adev->vcn.inst[i].ring_dec;
+
+               adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
+                               ((adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i), i);
+
+               r = amdgpu_ring_test_helper(ring);
+               if (r)
+                       goto done;
+
+               for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
+                       ring = &adev->vcn.inst[i].ring_enc[j];
+                       r = amdgpu_ring_test_helper(ring);
+                       if (r)
+                               goto done;
+               }
+       }
+
+done:
+       if (!r)
+               DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
+                       (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
+
+       return r;
+}
+
+/**
+ * vcn_v4_0_hw_fini - stop the hardware block
+ *
+ * @handle: amdgpu_device pointer
+ *
+ * Stop the VCN block, mark ring as not ready any more
+ */
+static int vcn_v4_0_hw_fini(void *handle)
+{
+       struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+       int i;
+
+       cancel_delayed_work_sync(&adev->vcn.idle_work);
+
+       for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
+               if (adev->vcn.harvest_config & (1 << i))
+                       continue;
+
+               if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
+                        (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
+                                RREG32_SOC15(VCN, i, regUVD_STATUS))) {
+                        vcn_v4_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
+               }
+       }
+
+       return 0;
+}
+
+/**
+ * vcn_v4_0_suspend - suspend VCN block
+ *
+ * @handle: amdgpu_device pointer
+ *
+ * HW fini and suspend VCN block
+ */
+static int vcn_v4_0_suspend(void *handle)
+{
+       int r;
+       struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+
+       r = vcn_v4_0_hw_fini(adev);
+       if (r)
+               return r;
+
+       r = amdgpu_vcn_suspend(adev);
+
+       return r;
+}
+
+/**
+ * vcn_v4_0_resume - resume VCN block
+ *
+ * @handle: amdgpu_device pointer
+ *
+ * Resume firmware and hw init VCN block
+ */
+static int vcn_v4_0_resume(void *handle)
+{
+       int r;
+       struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+
+       r = amdgpu_vcn_resume(adev);
+       if (r)
+               return r;
+
+       r = vcn_v4_0_hw_init(adev);
+
+       return r;
+}
+
+/**
+ * vcn_v4_0_mc_resume - memory controller programming
+ *
+ * @adev: amdgpu_device pointer
+ * @inst: instance number
+ *
+ * Let the VCN memory controller know it's offsets
+ */
+static void vcn_v4_0_mc_resume(struct amdgpu_device *adev, int inst)
+{
+       uint32_t offset, size;
+       const struct common_firmware_header *hdr;
+
+       hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
+       size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
+
+       /* cache window 0: fw */
+       if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
+               WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
+                       (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo));
+               WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
+                       (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi));
+               WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, 0);
+               offset = 0;
+       } else {
+               WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
+                       lower_32_bits(adev->vcn.inst[inst].gpu_addr));
+               WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
+                       upper_32_bits(adev->vcn.inst[inst].gpu_addr));
+               offset = size;
+                WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
+       }
+       WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE0, size);
+
+       /* cache window 1: stack */
+       WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
+               lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
+       WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
+               upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
+       WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET1, 0);
+       WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
+
+       /* cache window 2: context */
+       WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
+               lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
+       WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
+               upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
+       WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET2, 0);
+       WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
+
+       /* non-cache window */
+       WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
+               lower_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
+       WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
+               upper_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
+       WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_OFFSET0, 0);
+       WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_SIZE0,
+               AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
+}
+
+/**
+ * vcn_v4_0_mc_resume_dpg_mode - memory controller programming for dpg mode
+ *
+ * @adev: amdgpu_device pointer
+ * @inst_idx: instance number index
+ * @indirect: indirectly write sram
+ *
+ * Let the VCN memory controller know it's offsets with dpg mode
+ */
+static void vcn_v4_0_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
+{
+       uint32_t offset, size;
+       const struct common_firmware_header *hdr;
+       hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
+       size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
+
+       /* cache window 0: fw */
+       if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
+               if (!indirect) {
+                       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                               VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
+                               (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
+                       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                               VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
+                               (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
+                       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                               VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
+               } else {
+                       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                               VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
+                       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                               VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
+                       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                               VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
+               }
+               offset = 0;
+       } else {
+               WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                       VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
+                       lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
+               WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                       VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
+                       upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
+               offset = size;
+               WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                       VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0),
+                       AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
+
+       }
+
+       if (!indirect)
+               WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                       VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
+       else
+               WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                       VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
+
+       /* cache window 1: stack */
+       if (!indirect) {
+               WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                       VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
+                       lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
+               WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                       VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
+                       upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
+               WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                       VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
+       } else {
+               WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                       VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
+               WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                       VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
+               WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                       VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
+       }
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                       VCN, inst_idx, regUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
+
+       /* cache window 2: context */
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                       VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
+                       lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                       VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
+                       upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                       VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                       VCN, inst_idx, regUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
+
+       /* non-cache window */
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                       VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
+                       lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                       VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
+                       upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                       VCN, inst_idx, regUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+                       VCN, inst_idx, regUVD_VCPU_NONCACHE_SIZE0),
+                       AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)), 0, indirect);
+
+       /* VCN global tiling registers */
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+               VCN, 0, regUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
+}
+
+/**
+ * vcn_v4_0_disable_static_power_gating - disable VCN static power gating
+ *
+ * @adev: amdgpu_device pointer
+ * @inst: instance number
+ *
+ * Disable static power gating for VCN block
+ */
+static void vcn_v4_0_disable_static_power_gating(struct amdgpu_device *adev, int inst)
+{
+       uint32_t data = 0;
+
+       if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
+               data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
+                       | 1 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
+                       | 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
+
+               WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
+               SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS,
+                       UVD_PGFSM_STATUS__UVDM_UVDU_UVDLM_PWR_ON_3_0, 0x3F3FFFFF);
+       } else {
+               uint32_t value;
+
+               value = (inst) ? 0x2200800 : 0;
+               data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
+                       | 1 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
+                       | 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
+                       | 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
+                       | 1 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
+                       | 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
+                       | 1 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
+                       | 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
+                       | 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
+                       | 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
+                       | 1 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
+                       | 1 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
+                       | 1 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
+                       | 1 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
+
+                WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
+                SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS, value,  0x3F3FFFFF);
+        }
+
+        data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
+        data &= ~0x103;
+        if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
+                data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
+                        UVD_POWER_STATUS__UVD_PG_EN_MASK;
+
+        WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
+
+        return;
+}
+
+/**
+ * vcn_v4_0_enable_static_power_gating - enable VCN static power gating
+ *
+ * @adev: amdgpu_device pointer
+ * @inst: instance number
+ *
+ * Enable static power gating for VCN block
+ */
+static void vcn_v4_0_enable_static_power_gating(struct amdgpu_device *adev, int inst)
+{
+       uint32_t data;
+
+       if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
+               /* Before power off, this indicator has to be turned on */
+               data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
+               data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
+               data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
+               WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
+
+               data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
+                       | 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
+               WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
+
+               data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
+                       | 2 << UVD_PGFSM_STATUS__UVDS_PWR_STATUS__SHIFT
+                       | 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
+                       | 2 << UVD_PGFSM_STATUS__UVDTC_PWR_STATUS__SHIFT
+                       | 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
+                       | 2 << UVD_PGFSM_STATUS__UVDTA_PWR_STATUS__SHIFT
+                       | 2 << UVD_PGFSM_STATUS__UVDLM_PWR_STATUS__SHIFT
+                       | 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
+                       | 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
+                       | 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT
+                       | 2 << UVD_PGFSM_STATUS__UVDAB_PWR_STATUS__SHIFT
+                       | 2 << UVD_PGFSM_STATUS__UVDTB_PWR_STATUS__SHIFT
+                       | 2 << UVD_PGFSM_STATUS__UVDNA_PWR_STATUS__SHIFT
+                       | 2 << UVD_PGFSM_STATUS__UVDNB_PWR_STATUS__SHIFT);
+               SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS, data, 0x3F3FFFFF);
+       }
+
+        return;
+}
+
+/**
+ * vcn_v4_0_disable_clock_gating - disable VCN clock gating
+ *
+ * @adev: amdgpu_device pointer
+ * @inst: instance number
+ *
+ * Disable clock gating for VCN block
+ */
+static void vcn_v4_0_disable_clock_gating(struct amdgpu_device *adev, int inst)
+{
+       uint32_t data;
+
+       if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
+               return;
+
+       /* VCN disable CGC */
+       data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
+       data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
+       data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
+       data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
+       WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
+
+       data = RREG32_SOC15(VCN, inst, regUVD_CGC_GATE);
+       data &= ~(UVD_CGC_GATE__SYS_MASK
+               | UVD_CGC_GATE__UDEC_MASK
+               | UVD_CGC_GATE__MPEG2_MASK
+               | UVD_CGC_GATE__REGS_MASK
+               | UVD_CGC_GATE__RBC_MASK
+               | UVD_CGC_GATE__LMI_MC_MASK
+               | UVD_CGC_GATE__LMI_UMC_MASK
+               | UVD_CGC_GATE__IDCT_MASK
+               | UVD_CGC_GATE__MPRD_MASK
+               | UVD_CGC_GATE__MPC_MASK
+               | UVD_CGC_GATE__LBSI_MASK
+               | UVD_CGC_GATE__LRBBM_MASK
+               | UVD_CGC_GATE__UDEC_RE_MASK
+               | UVD_CGC_GATE__UDEC_CM_MASK
+               | UVD_CGC_GATE__UDEC_IT_MASK
+               | UVD_CGC_GATE__UDEC_DB_MASK
+               | UVD_CGC_GATE__UDEC_MP_MASK
+               | UVD_CGC_GATE__WCB_MASK
+               | UVD_CGC_GATE__VCPU_MASK
+               | UVD_CGC_GATE__MMSCH_MASK);
+
+       WREG32_SOC15(VCN, inst, regUVD_CGC_GATE, data);
+       SOC15_WAIT_ON_RREG(VCN, inst, regUVD_CGC_GATE, 0,  0xFFFFFFFF);
+
+       data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
+       data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
+               | UVD_CGC_CTRL__UDEC_CM_MODE_MASK
+               | UVD_CGC_CTRL__UDEC_IT_MODE_MASK
+               | UVD_CGC_CTRL__UDEC_DB_MODE_MASK
+               | UVD_CGC_CTRL__UDEC_MP_MODE_MASK
+               | UVD_CGC_CTRL__SYS_MODE_MASK
+               | UVD_CGC_CTRL__UDEC_MODE_MASK
+               | UVD_CGC_CTRL__MPEG2_MODE_MASK
+               | UVD_CGC_CTRL__REGS_MODE_MASK
+               | UVD_CGC_CTRL__RBC_MODE_MASK
+               | UVD_CGC_CTRL__LMI_MC_MODE_MASK
+               | UVD_CGC_CTRL__LMI_UMC_MODE_MASK
+               | UVD_CGC_CTRL__IDCT_MODE_MASK
+               | UVD_CGC_CTRL__MPRD_MODE_MASK
+               | UVD_CGC_CTRL__MPC_MODE_MASK
+               | UVD_CGC_CTRL__LBSI_MODE_MASK
+               | UVD_CGC_CTRL__LRBBM_MODE_MASK
+               | UVD_CGC_CTRL__WCB_MODE_MASK
+               | UVD_CGC_CTRL__VCPU_MODE_MASK
+               | UVD_CGC_CTRL__MMSCH_MODE_MASK);
+       WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
+
+       data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE);
+       data |= (UVD_SUVD_CGC_GATE__SRE_MASK
+               | UVD_SUVD_CGC_GATE__SIT_MASK
+               | UVD_SUVD_CGC_GATE__SMP_MASK
+               | UVD_SUVD_CGC_GATE__SCM_MASK
+               | UVD_SUVD_CGC_GATE__SDB_MASK
+               | UVD_SUVD_CGC_GATE__SRE_H264_MASK
+               | UVD_SUVD_CGC_GATE__SRE_H264_MASK
+               | UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
+               | UVD_SUVD_CGC_GATE__SIT_H264_MASK
+               | UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
+               | UVD_SUVD_CGC_GATE__SCM_H264_MASK
+               | UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
+               | UVD_SUVD_CGC_GATE__SDB_H264_MASK
+               | UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
+               | UVD_SUVD_CGC_GATE__SCLR_MASK
+               | UVD_SUVD_CGC_GATE__UVD_SC_MASK
+               | UVD_SUVD_CGC_GATE__ENT_MASK
+               | UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
+               | UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
+               | UVD_SUVD_CGC_GATE__SITE_MASK
+               | UVD_SUVD_CGC_GATE__SRE_VP9_MASK
+               | UVD_SUVD_CGC_GATE__SCM_VP9_MASK
+               | UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
+               | UVD_SUVD_CGC_GATE__SDB_VP9_MASK
+               | UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
+       WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE, data);
+
+       data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
+       data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
+               | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
+               | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
+               | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
+               | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
+               | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
+               | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
+               | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
+               | UVD_SUVD_CGC_CTRL__IME_MODE_MASK
+               | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
+       WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
+}
+
+/**
+ * vcn_v4_0_disable_clock_gating_dpg_mode - disable VCN clock gating dpg mode
+ *
+ * @adev: amdgpu_device pointer
+ * @sram_sel: sram select
+ * @inst_idx: instance number index
+ * @indirect: indirectly write sram
+ *
+ * Disable clock gating for VCN block with dpg mode
+ */
+static void vcn_v4_0_disable_clock_gating_dpg_mode(struct amdgpu_device *adev, uint8_t sram_sel,
+      int inst_idx, uint8_t indirect)
+{
+       uint32_t reg_data = 0;
+
+       if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
+               return;
+
+       /* enable sw clock gating control */
+       reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
+       reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
+       reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
+       reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
+                UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
+                UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
+                UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
+                UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
+                UVD_CGC_CTRL__SYS_MODE_MASK |
+                UVD_CGC_CTRL__UDEC_MODE_MASK |
+                UVD_CGC_CTRL__MPEG2_MODE_MASK |
+                UVD_CGC_CTRL__REGS_MODE_MASK |
+                UVD_CGC_CTRL__RBC_MODE_MASK |
+                UVD_CGC_CTRL__LMI_MC_MODE_MASK |
+                UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
+                UVD_CGC_CTRL__IDCT_MODE_MASK |
+                UVD_CGC_CTRL__MPRD_MODE_MASK |
+                UVD_CGC_CTRL__MPC_MODE_MASK |
+                UVD_CGC_CTRL__LBSI_MODE_MASK |
+                UVD_CGC_CTRL__LRBBM_MODE_MASK |
+                UVD_CGC_CTRL__WCB_MODE_MASK |
+                UVD_CGC_CTRL__VCPU_MODE_MASK);
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+               VCN, inst_idx, regUVD_CGC_CTRL), reg_data, sram_sel, indirect);
+
+       /* turn off clock gating */
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+               VCN, inst_idx, regUVD_CGC_GATE), 0, sram_sel, indirect);
+
+       /* turn on SUVD clock gating */
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+               VCN, inst_idx, regUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
+
+       /* turn on sw mode in UVD_SUVD_CGC_CTRL */
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+               VCN, inst_idx, regUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
+}
+
+/**
+ * vcn_v4_0_enable_clock_gating - enable VCN clock gating
+ *
+ * @adev: amdgpu_device pointer
+ * @inst: instance number
+ *
+ * Enable clock gating for VCN block
+ */
+static void vcn_v4_0_enable_clock_gating(struct amdgpu_device *adev, int inst)
+{
+       uint32_t data;
+
+       if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
+               return;
+
+       /* enable VCN CGC */
+       data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
+       data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
+       data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
+       data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
+       WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
+
+       data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
+       data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
+               | UVD_CGC_CTRL__UDEC_CM_MODE_MASK
+               | UVD_CGC_CTRL__UDEC_IT_MODE_MASK
+               | UVD_CGC_CTRL__UDEC_DB_MODE_MASK
+               | UVD_CGC_CTRL__UDEC_MP_MODE_MASK
+               | UVD_CGC_CTRL__SYS_MODE_MASK
+               | UVD_CGC_CTRL__UDEC_MODE_MASK
+               | UVD_CGC_CTRL__MPEG2_MODE_MASK
+               | UVD_CGC_CTRL__REGS_MODE_MASK
+               | UVD_CGC_CTRL__RBC_MODE_MASK
+               | UVD_CGC_CTRL__LMI_MC_MODE_MASK
+               | UVD_CGC_CTRL__LMI_UMC_MODE_MASK
+               | UVD_CGC_CTRL__IDCT_MODE_MASK
+               | UVD_CGC_CTRL__MPRD_MODE_MASK
+               | UVD_CGC_CTRL__MPC_MODE_MASK
+               | UVD_CGC_CTRL__LBSI_MODE_MASK
+               | UVD_CGC_CTRL__LRBBM_MODE_MASK
+               | UVD_CGC_CTRL__WCB_MODE_MASK
+               | UVD_CGC_CTRL__VCPU_MODE_MASK
+               | UVD_CGC_CTRL__MMSCH_MODE_MASK);
+       WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
+
+       data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
+       data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
+               | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
+               | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
+               | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
+               | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
+               | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
+               | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
+               | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
+               | UVD_SUVD_CGC_CTRL__IME_MODE_MASK
+               | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
+       WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
+
+       return;
+}
+
+/**
+ * vcn_v4_0_start_dpg_mode - VCN start with dpg mode
+ *
+ * @adev: amdgpu_device pointer
+ * @inst_idx: instance number index
+ * @indirect: indirectly write sram
+ *
+ * Start VCN block with dpg mode
+ */
+static int vcn_v4_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
+{
+       volatile struct amdgpu_vcn4_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
+       struct amdgpu_ring *ring;
+       uint32_t tmp;
+       int i;
+
+       /* disable register anti-hang mechanism */
+       WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 1,
+               ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
+       /* enable dynamic power gating mode */
+       tmp = RREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS);
+       tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
+       tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
+       WREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS, tmp);
+
+       if (indirect)
+               adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
+
+       /* enable clock gating */
+       vcn_v4_0_disable_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
+
+       /* enable VCPU clock */
+       tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
+       tmp |= UVD_VCPU_CNTL__CLK_EN_MASK | UVD_VCPU_CNTL__BLK_RST_MASK;
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+               VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
+
+       /* disable master interupt */
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+               VCN, inst_idx, regUVD_MASTINT_EN), 0, 0, indirect);
+
+       /* setup regUVD_LMI_CTRL */
+       tmp = (UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
+               UVD_LMI_CTRL__REQ_MODE_MASK |
+               UVD_LMI_CTRL__CRC_RESET_MASK |
+               UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
+               UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
+               UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
+               (8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
+               0x00100000L);
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+               VCN, inst_idx, regUVD_LMI_CTRL), tmp, 0, indirect);
+
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+               VCN, inst_idx, regUVD_MPC_CNTL),
+               0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
+
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+               VCN, inst_idx, regUVD_MPC_SET_MUXA0),
+               ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
+                (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
+                (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
+                (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
+
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+               VCN, inst_idx, regUVD_MPC_SET_MUXB0),
+                ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
+                (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
+                (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
+                (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
+
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+               VCN, inst_idx, regUVD_MPC_SET_MUX),
+               ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
+                (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
+                (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
+
+       vcn_v4_0_mc_resume_dpg_mode(adev, inst_idx, indirect);
+
+       tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
+       tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+               VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
+
+       /* enable LMI MC and UMC channels */
+       tmp = 0x1f << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT;
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+               VCN, inst_idx, regUVD_LMI_CTRL2), tmp, 0, indirect);
+
+       /* enable master interrupt */
+       WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
+               VCN, inst_idx, regUVD_MASTINT_EN),
+               UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
+
+
+       if (indirect)
+               psp_update_vcn_sram(adev, inst_idx, adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
+                       (uint32_t)((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
+                               (uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr));
+
+       if (unifiedQ_enabled) {
+               ring = &adev->vcn.inst[inst_idx].ring_enc[0];
+               fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
+       } else
+               ring = &adev->vcn.inst[inst_idx].ring_dec;
+
+       WREG32_SOC15(VCN, inst_idx, regVCN_RB4_DB_CTRL,
+               ring->doorbell_index << VCN_RB4_DB_CTRL__OFFSET__SHIFT |
+               VCN_RB4_DB_CTRL__EN_MASK);
+
+       /* program the RB_BASE for ring buffer */
+       WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_LO4,
+               lower_32_bits(ring->gpu_addr));
+       WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_HI4,
+               upper_32_bits(ring->gpu_addr));
+
+       WREG32_SOC15(VCN, inst_idx, regUVD_RB_SIZE4, ring->ring_size / sizeof(uint32_t));
+
+       /* reseting ring, fw should not check RB ring */
+       tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
+       tmp &= ~(VCN_RB_ENABLE__RB4_EN_MASK);
+       WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
+
+       /* Initialize the ring buffer's read and write pointers */
+       tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR4);
+       WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR4, tmp);
+       ring->wptr = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR4);
+
+       tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
+       tmp |= VCN_RB_ENABLE__RB4_EN_MASK;
+       WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
+
+       WREG32_SOC15(VCN, inst_idx, regUVD_SCRATCH2, 0);
+
+       if (unifiedQ_enabled)
+               fw_shared->sq.queue_mode &= ~FW_QUEUE_RING_RESET;
+
+       for (i = 0; i < adev->vcn.num_enc_rings; i++) {
+               ring = &adev->vcn.inst[inst_idx].ring_enc[i];
+
+               if (i) {
+                       ring = &adev->vcn.inst[inst_idx].ring_enc[1];
+
+                       WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_LO2, ring->gpu_addr);
+                       WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
+                       WREG32_SOC15(VCN, inst_idx, regUVD_RB_SIZE2, ring->ring_size / 4);
+                       tmp= RREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR2);
+                       WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR2, tmp);
+                       ring->wptr = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR2);
+
+                       WREG32_SOC15(VCN, inst_idx, regVCN_RB2_DB_CTRL,
+                               ring->doorbell_index << VCN_RB2_DB_CTRL__OFFSET__SHIFT |
+                               VCN_RB2_DB_CTRL__EN_MASK);
+               } else {
+                       ring = &adev->vcn.inst[inst_idx].ring_enc[0];
+
+                       WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_LO, ring->gpu_addr);
+                       WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
+                       WREG32_SOC15(VCN, inst_idx, regUVD_RB_SIZE, ring->ring_size / 4);
+                       tmp= RREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR);
+                       WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, tmp);
+                       ring->wptr = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
+
+                       WREG32_SOC15(VCN, inst_idx, regVCN_RB1_DB_CTRL,
+                               ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
+                               VCN_RB1_DB_CTRL__EN_MASK);
+               }
+       }
+       return 0;
+}
+
+
+/**
+ * vcn_v4_0_start - VCN start
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Start VCN block
+ */
+static int vcn_v4_0_start(struct amdgpu_device *adev)
+{
+       volatile struct amdgpu_vcn4_fw_shared *fw_shared;
+       struct amdgpu_ring *ring;
+       uint32_t tmp;
+       int i, j, k, r;
+
+       if (adev->pm.dpm_enabled)
+               amdgpu_dpm_enable_uvd(adev, true);
+
+       for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
+               if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
+                       r = vcn_v4_0_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
+                       continue;
+               }
+
+               /* disable VCN power gating */
+               vcn_v4_0_disable_static_power_gating(adev, i);
+
+               /* set VCN status busy */
+               tmp = RREG32_SOC15(VCN, i, regUVD_STATUS) | UVD_STATUS__UVD_BUSY;
+               WREG32_SOC15(VCN, i, regUVD_STATUS, tmp);
+
+               /*SW clock gating */
+               vcn_v4_0_disable_clock_gating(adev, i);
+
+               /* enable VCPU clock */
+               WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
+                       UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
+
+               /* disable master interrupt */
+               WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN), 0,
+                       ~UVD_MASTINT_EN__VCPU_EN_MASK);
+
+               /* enable LMI MC and UMC channels */
+               WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_LMI_CTRL2), 0,
+                       ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
+
+               tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
+               tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
+               tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
+               WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
+
+               /* setup regUVD_LMI_CTRL */
+               tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL);
+               WREG32_SOC15(VCN, i, regUVD_LMI_CTRL, tmp |
+               UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
+               UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
+               UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
+               UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
+
+               /* setup regUVD_MPC_CNTL */
+               tmp = RREG32_SOC15(VCN, i, regUVD_MPC_CNTL);
+               tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
+               tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
+               WREG32_SOC15(VCN, i, regUVD_MPC_CNTL, tmp);
+
+               /* setup UVD_MPC_SET_MUXA0 */
+               WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXA0,
+                       ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
+                       (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
+                       (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
+                       (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
+
+               /* setup UVD_MPC_SET_MUXB0 */
+               WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXB0,
+                       ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
+                       (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
+                       (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
+                       (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
+
+               /* setup UVD_MPC_SET_MUX */
+               WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUX,
+                       ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
+                       (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
+                       (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
+
+               vcn_v4_0_mc_resume(adev, i);
+
+               /* VCN global tiling registers */
+               WREG32_SOC15(VCN, i, regUVD_GFX10_ADDR_CONFIG,
+                       adev->gfx.config.gb_addr_config);
+
+               /* unblock VCPU register access */
+               WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL), 0,
+                       ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
+
+               /* release VCPU reset to boot */
+               WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
+                       ~UVD_VCPU_CNTL__BLK_RST_MASK);
+
+               for (j = 0; j < 10; ++j) {
+                       uint32_t status;
+
+                       for (k = 0; k < 100; ++k) {
+                               status = RREG32_SOC15(VCN, i, regUVD_STATUS);
+                               if (status & 2)
+                                       break;
+                               mdelay(10);
+                               if (amdgpu_emu_mode==1)
+                                       msleep(1);
+                       }
+
+                       if (amdgpu_emu_mode==1) {
+                               if (status & 2) {
+                                       r = 0;
+                                       break;
+                               }
+                       } else {
+                               r = 0;
+                               if (status & 2)
+                                       break;
+
+                               dev_err(adev->dev, "VCN[%d] decode not responding, trying to reset the VCPU!!!\n", i);
+                                       WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
+                                       UVD_VCPU_CNTL__BLK_RST_MASK,
+                                       ~UVD_VCPU_CNTL__BLK_RST_MASK);
+                               mdelay(10);
+                               WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
+                                       ~UVD_VCPU_CNTL__BLK_RST_MASK);
+
+                               mdelay(10);
+                               r = -1;
+                       }
+               }
+
+               if (r) {
+                       dev_err(adev->dev, "VCN[%d] decode not responding, giving up!!!\n", i);
+                       return r;
+               }
+
+               /* enable master interrupt */
+               WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN),
+                       UVD_MASTINT_EN__VCPU_EN_MASK,
+                       ~UVD_MASTINT_EN__VCPU_EN_MASK);
+
+               /* clear the busy bit of VCN_STATUS */
+               WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_STATUS), 0,
+                       ~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
+
+               fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
+               if (unifiedQ_enabled) {
+                       ring = &adev->vcn.inst[i].ring_enc[0];
+                       fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
+               } else {
+                       ring = &adev->vcn.inst[i].ring_dec;
+
+                       WREG32_SOC15(VCN, i, regVCN_RB4_DB_CTRL,
+                               ring->doorbell_index << VCN_RB4_DB_CTRL__OFFSET__SHIFT |
+                               VCN_RB4_DB_CTRL__EN_MASK);
+
+                       /* program the RB_BASE for ring buffer */
+                       WREG32_SOC15(VCN, i, regUVD_RB_BASE_LO4,
+                               lower_32_bits(ring->gpu_addr));
+                       WREG32_SOC15(VCN, i, regUVD_RB_BASE_HI4,
+                               upper_32_bits(ring->gpu_addr));
+
+                       WREG32_SOC15(VCN, i, regUVD_RB_SIZE4, ring->ring_size / sizeof(uint32_t));
+
+                       /* resetting ring, fw should not check RB ring */
+                       tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
+                       tmp &= ~(VCN_RB_ENABLE__RB4_EN_MASK);
+                       WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
+
+                       /* Initialize the ring buffer's read and write pointers */
+                       tmp = RREG32_SOC15(VCN, i, regUVD_RB_RPTR4);
+                       WREG32_SOC15(VCN, i, regUVD_RB_WPTR4, tmp);
+                       ring->wptr = RREG32_SOC15(VCN, i, regUVD_RB_WPTR4);
+
+                       tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
+                       tmp |= VCN_RB_ENABLE__RB4_EN_MASK;
+                       WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
+
+                       ring->wptr = RREG32_SOC15(VCN, i, regUVD_RB_RPTR4);
+               }
+               ring = &adev->vcn.inst[i].ring_enc[0];
+               WREG32_SOC15(VCN, i, regVCN_RB1_DB_CTRL,
+                       ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
+                       VCN_RB1_DB_CTRL__EN_MASK);
+               tmp = RREG32_SOC15(VCN, i, regUVD_RB_RPTR);
+               WREG32_SOC15(VCN, i, regUVD_RB_WPTR, tmp);
+               ring->wptr = RREG32_SOC15(VCN, i, regUVD_RB_WPTR);
+               WREG32_SOC15(VCN, i, regUVD_RB_BASE_LO, ring->gpu_addr);
+               WREG32_SOC15(VCN, i, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
+               WREG32_SOC15(VCN, i, regUVD_RB_SIZE, ring->ring_size / 4);
+               if (unifiedQ_enabled)
+                       fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
+               else {
+                       ring = &adev->vcn.inst[i].ring_enc[1];
+                       WREG32_SOC15(VCN, i, regVCN_RB2_DB_CTRL,
+                               ring->doorbell_index << VCN_RB2_DB_CTRL__OFFSET__SHIFT |
+                               VCN_RB2_DB_CTRL__EN_MASK);
+                       tmp = RREG32_SOC15(VCN, i, regUVD_RB_RPTR2);
+                       WREG32_SOC15(VCN, i, regUVD_RB_WPTR2, tmp);
+                       ring->wptr = RREG32_SOC15(VCN, i, regUVD_RB_WPTR2);
+                       WREG32_SOC15(VCN, i, regUVD_RB_BASE_LO2, ring->gpu_addr);
+                       WREG32_SOC15(VCN, i, regUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
+                       WREG32_SOC15(VCN, i, regUVD_RB_SIZE2, ring->ring_size / 4);
+               }
+       }
+
+       return 0;
+}
+
+/**
+ * vcn_v4_0_stop_dpg_mode - VCN stop with dpg mode
+ *
+ * @adev: amdgpu_device pointer
+ * @inst_idx: instance number index
+ *
+ * Stop VCN block with dpg mode
+ */
+static int vcn_v4_0_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
+{
+       uint32_t tmp;
+
+       /* Wait for power status to be 1 */
+       SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
+               UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
+
+       /* wait for read ptr to be equal to write ptr */
+       tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
+       SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_RB_RPTR, tmp, 0xFFFFFFFF);
+
+       tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR2);
+       SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
+
+       tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR4);
+       SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_RB_RPTR4, tmp, 0xFFFFFFFF);
+
+       SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
+               UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
+
+       /* disable dynamic power gating mode */
+       WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 0,
+               ~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
+       return 0;
+}
+
+/**
+ * vcn_v4_0_stop - VCN stop
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Stop VCN block
+ */
+static int vcn_v4_0_stop(struct amdgpu_device *adev)
+{
+       uint32_t tmp;
+       int i, r = 0;
+
+       for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
+               if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
+                       r = vcn_v4_0_stop_dpg_mode(adev, i);
+                       continue;
+               }
+
+               /* wait for vcn idle */
+               r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE, 0x7);
+               if (r)
+                       return r;
+
+               tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
+                       UVD_LMI_STATUS__READ_CLEAN_MASK |
+                       UVD_LMI_STATUS__WRITE_CLEAN_MASK |
+                       UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
+               r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
+               if (r)
+                       return r;
+
+               /* disable LMI UMC channel */
+               tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL2);
+               tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
+               WREG32_SOC15(VCN, i, regUVD_LMI_CTRL2, tmp);
+               tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK |
+                       UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
+               r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
+               if (r)
+                       return r;
+
+               /* block VCPU register access */
+               WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL),
+                               UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
+                               ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
+
+               /* reset VCPU */
+               WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
+                               UVD_VCPU_CNTL__BLK_RST_MASK,
+                               ~UVD_VCPU_CNTL__BLK_RST_MASK);
+
+               /* disable VCPU clock */
+               WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
+                               ~(UVD_VCPU_CNTL__CLK_EN_MASK));
+
+               /* apply soft reset */
+               tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
+               tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
+               WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
+               tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
+               tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
+               WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
+
+               /* clear status */
+               WREG32_SOC15(VCN, i, regUVD_STATUS, 0);
+
+               /* apply HW clock gating */
+               vcn_v4_0_enable_clock_gating(adev, i);
+
+               /* enable VCN power gating */
+               vcn_v4_0_enable_static_power_gating(adev, i);
+       }
+
+       if (adev->pm.dpm_enabled)
+               amdgpu_dpm_enable_uvd(adev, false);
+
+       return 0;
+}
+
+/**
+ * vcn_v4_0_pause_dpg_mode - VCN pause with dpg mode
+ *
+ * @adev: amdgpu_device pointer
+ * @inst_idx: instance number index
+ * @new_state: pause state
+ *
+ * Pause dpg mode for VCN block
+ */
+static int vcn_v4_0_pause_dpg_mode(struct amdgpu_device *adev, int inst_idx,
+      struct dpg_pause_state *new_state)
+{
+       uint32_t reg_data = 0;
+       int ret_code;
+
+       /* pause/unpause if state is changed */
+       if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
+               DRM_DEV_DEBUG(adev->dev, "dpg pause state changed %d -> %d",
+                       adev->vcn.inst[inst_idx].pause_state.fw_based,  new_state->fw_based);
+               reg_data = RREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE) &
+                       (~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
+
+               if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
+                       ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 0x1,
+                               UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
+
+                       if (!ret_code) {
+                               /* pause DPG */
+                               reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
+                               WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
+
+                               /* wait for ACK */
+                               SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_DPG_PAUSE,
+                                       UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
+                                       UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
+
+                               SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS,
+                                       UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
+                       }
+               } else {
+                       /* unpause dpg, no need to wait */
+                       reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
+                       WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
+                       SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 0x1,
+                               UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
+               }
+               adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
+       }
+
+       return 0;
+}
+
+/**
+ * vcn_v4_0_dec_ring_get_rptr - get read pointer
+ *
+ * @ring: amdgpu_ring pointer
+ *
+ * Returns the current hardware read pointer
+ */
+static uint64_t vcn_v4_0_dec_ring_get_rptr(struct amdgpu_ring *ring)
+{
+       struct amdgpu_device *adev = ring->adev;
+
+       return RREG32_SOC15(VCN, ring->me, regUVD_RB_RPTR4);
+}
+
+/**
+ * vcn_v4_0_dec_ring_get_wptr - get write pointer
+ *
+ * @ring: amdgpu_ring pointer
+ *
+ * Returns the current hardware write pointer
+ */
+static uint64_t vcn_v4_0_dec_ring_get_wptr(struct amdgpu_ring *ring)
+{
+       struct amdgpu_device *adev = ring->adev;
+
+       if (ring->use_doorbell)
+               return *ring->wptr_cpu_addr;
+       else
+               return RREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR4);
+}
+
+/**
+ * vcn_v4_0_dec_ring_set_wptr - set write pointer
+ *
+ * @ring: amdgpu_ring pointer
+ *
+ * Commits the write pointer to the hardware
+ */
+static void vcn_v4_0_dec_ring_set_wptr(struct amdgpu_ring *ring)
+{
+       struct amdgpu_device *adev = ring->adev;
+
+       if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
+               WREG32_SOC15(VCN, ring->me, regUVD_SCRATCH2,
+                       lower_32_bits(ring->wptr));
+       }
+
+       if (ring->use_doorbell) {
+               *ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
+               WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
+       } else {
+               WREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR4, lower_32_bits(ring->wptr));
+       }
+}
+
+static const struct amdgpu_ring_funcs vcn_v4_0_dec_sw_ring_vm_funcs = {
+       .type = AMDGPU_RING_TYPE_VCN_DEC,
+       .align_mask = 0x3f,
+       .nop = VCN_DEC_SW_CMD_NO_OP,
+       .vmhub = AMDGPU_MMHUB_0,
+       .get_rptr = vcn_v4_0_dec_ring_get_rptr,
+       .get_wptr = vcn_v4_0_dec_ring_get_wptr,
+       .set_wptr = vcn_v4_0_dec_ring_set_wptr,
+       .emit_frame_size =
+               SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
+               SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
+               4 + /* vcn_v3_0_dec_sw_ring_emit_vm_flush */
+               5 + 5 + /* vcn_v3_0_dec_sw_ring_emit_fdec_swe x2 vm fdec_swe */
+               1, /* vcn_v3_0_dec_sw_ring_insert_end */
+       .emit_ib_size = 5, /* vcn_v3_0_dec_sw_ring_emit_ib */
+       .emit_ib = vcn_v3_0_dec_sw_ring_emit_ib,
+       .emit_fence = vcn_v3_0_dec_sw_ring_emit_fence,
+       .emit_vm_flush = vcn_v3_0_dec_sw_ring_emit_vm_flush,
+       .test_ring = amdgpu_vcn_dec_sw_ring_test_ring,
+       .test_ib = amdgpu_vcn_dec_sw_ring_test_ib,
+       .insert_nop = amdgpu_ring_insert_nop,
+       .insert_end = vcn_v3_0_dec_sw_ring_insert_end,
+       .pad_ib = amdgpu_ring_generic_pad_ib,
+       .begin_use = amdgpu_vcn_ring_begin_use,
+       .end_use = amdgpu_vcn_ring_end_use,
+       .emit_wreg = vcn_v3_0_dec_sw_ring_emit_wreg,
+       .emit_reg_wait = vcn_v3_0_dec_sw_ring_emit_reg_wait,
+       .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
+};
+
+/**
+ * vcn_v4_0_enc_ring_get_rptr - get enc read pointer
+ *
+ * @ring: amdgpu_ring pointer
+ *
+ * Returns the current hardware enc read pointer
+ */
+static uint64_t vcn_v4_0_enc_ring_get_rptr(struct amdgpu_ring *ring)
+{
+       struct amdgpu_device *adev = ring->adev;
+
+       if (ring == &adev->vcn.inst[ring->me].ring_enc[0])
+               return RREG32_SOC15(VCN, ring->me, regUVD_RB_RPTR);
+       else
+               return RREG32_SOC15(VCN, ring->me, regUVD_RB_RPTR2);
+}
+
+/**
+ * vcn_v4_0_enc_ring_get_wptr - get enc write pointer
+ *
+ * @ring: amdgpu_ring pointer
+ *
+ * Returns the current hardware enc write pointer
+ */
+static uint64_t vcn_v4_0_enc_ring_get_wptr(struct amdgpu_ring *ring)
+{
+       struct amdgpu_device *adev = ring->adev;
+
+       if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
+               if (ring->use_doorbell)
+                       return *ring->wptr_cpu_addr;
+               else
+                       return RREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR);
+       } else {
+               if (ring->use_doorbell)
+                       return *ring->wptr_cpu_addr;
+               else
+                       return RREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR2);
+       }
+}
+
+/**
+ * vcn_v4_0_enc_ring_set_wptr - set enc write pointer
+ *
+ * @ring: amdgpu_ring pointer
+ *
+ * Commits the enc write pointer to the hardware
+ */
+static void vcn_v4_0_enc_ring_set_wptr(struct amdgpu_ring *ring)
+{
+       struct amdgpu_device *adev = ring->adev;
+
+       if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
+               if (ring->use_doorbell) {
+                       *ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
+                       WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
+               } else {
+                       WREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR, lower_32_bits(ring->wptr));
+               }
+       } else {
+               if (ring->use_doorbell) {
+                       *ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
+                       WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
+               } else {
+                       WREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR2, lower_32_bits(ring->wptr));
+               }
+       }
+}
+
+static const struct amdgpu_ring_funcs vcn_v4_0_enc_ring_vm_funcs = {
+       .type = AMDGPU_RING_TYPE_VCN_ENC,
+       .align_mask = 0x3f,
+       .nop = VCN_ENC_CMD_NO_OP,
+       .vmhub = AMDGPU_MMHUB_0,
+       .get_rptr = vcn_v4_0_enc_ring_get_rptr,
+       .get_wptr = vcn_v4_0_enc_ring_get_wptr,
+       .set_wptr = vcn_v4_0_enc_ring_set_wptr,
+       .emit_frame_size =
+               SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
+               SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
+               4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
+               5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
+               1, /* vcn_v2_0_enc_ring_insert_end */
+       .emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
+       .emit_ib = vcn_v2_0_enc_ring_emit_ib,
+       .emit_fence = vcn_v2_0_enc_ring_emit_fence,
+       .emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
+       .test_ring = amdgpu_vcn_enc_ring_test_ring,
+       .test_ib = amdgpu_vcn_enc_ring_test_ib,
+       .insert_nop = amdgpu_ring_insert_nop,
+       .insert_end = vcn_v2_0_enc_ring_insert_end,
+       .pad_ib = amdgpu_ring_generic_pad_ib,
+       .begin_use = amdgpu_vcn_ring_begin_use,
+       .end_use = amdgpu_vcn_ring_end_use,
+       .emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
+       .emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
+       .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
+};
+
+/**
+ * vcn_v4_0_set_dec_ring_funcs - set dec ring functions
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Set decode ring functions
+ */
+static void vcn_v4_0_set_dec_ring_funcs(struct amdgpu_device *adev)
+{
+       int i;
+
+       for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
+               if (adev->vcn.harvest_config & (1 << i))
+                       continue;
+
+               adev->vcn.inst[i].ring_dec.funcs = &vcn_v4_0_dec_sw_ring_vm_funcs;
+               adev->vcn.inst[i].ring_dec.me = i;
+               DRM_INFO("VCN(%d) decode software ring is enabled in VM mode\n", i);
+       }
+}
+
+/**
+ * vcn_v4_0_set_enc_ring_funcs - set enc ring functions
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Set encode ring functions
+ */
+static void vcn_v4_0_set_enc_ring_funcs(struct amdgpu_device *adev)
+{
+       int i, j;
+
+       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.inst[i].ring_enc[j].funcs = &vcn_v4_0_enc_ring_vm_funcs;
+                       adev->vcn.inst[i].ring_enc[j].me = i;
+               }
+               DRM_INFO("VCN(%d) encode is enabled in VM mode\n", i);
+       }
+}
+
+/**
+ * vcn_v4_0_is_idle - check VCN block is idle
+ *
+ * @handle: amdgpu_device pointer
+ *
+ * Check whether VCN block is idle
+ */
+static bool vcn_v4_0_is_idle(void *handle)
+{
+       struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+       int i, ret = 1;
+
+       for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
+               if (adev->vcn.harvest_config & (1 << i))
+                       continue;
+
+               ret &= (RREG32_SOC15(VCN, i, regUVD_STATUS) == UVD_STATUS__IDLE);
+       }
+
+       return ret;
+}
+
+/**
+ * vcn_v4_0_wait_for_idle - wait for VCN block idle
+ *
+ * @handle: amdgpu_device pointer
+ *
+ * Wait for VCN block idle
+ */
+static int vcn_v4_0_wait_for_idle(void *handle)
+{
+       struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+       int i, ret = 0;
+
+       for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
+               if (adev->vcn.harvest_config & (1 << i))
+                       continue;
+
+               ret = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE,
+                       UVD_STATUS__IDLE);
+               if (ret)
+                       return ret;
+       }
+
+       return ret;
+}
+
+/**
+ * vcn_v4_0_set_clockgating_state - set VCN block clockgating state
+ *
+ * @handle: amdgpu_device pointer
+ * @state: clock gating state
+ *
+ * Set VCN block clockgating state
+ */
+static int vcn_v4_0_set_clockgating_state(void *handle, enum amd_clockgating_state state)
+{
+       struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+       bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
+       int i;
+
+       for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
+               if (adev->vcn.harvest_config & (1 << i))
+                       continue;
+
+               if (enable) {
+                       if (RREG32_SOC15(VCN, i, regUVD_STATUS) != UVD_STATUS__IDLE)
+                               return -EBUSY;
+                       vcn_v4_0_enable_clock_gating(adev, i);
+               } else {
+                       vcn_v4_0_disable_clock_gating(adev, i);
+               }
+       }
+
+       return 0;
+}
+
+/**
+ * vcn_v4_0_set_powergating_state - set VCN block powergating state
+ *
+ * @handle: amdgpu_device pointer
+ * @state: power gating state
+ *
+ * Set VCN block powergating state
+ */
+static int vcn_v4_0_set_powergating_state(void *handle, enum amd_powergating_state state)
+{
+       struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+       int ret;
+
+       if(state == adev->vcn.cur_state)
+               return 0;
+
+       if (state == AMD_PG_STATE_GATE)
+               ret = vcn_v4_0_stop(adev);
+       else
+               ret = vcn_v4_0_start(adev);
+
+       if(!ret)
+               adev->vcn.cur_state = state;
+
+       return ret;
+}
+
+/**
+ * vcn_v4_0_set_interrupt_state - set VCN block interrupt state
+ *
+ * @adev: amdgpu_device pointer
+ * @source: interrupt sources
+ * @type: interrupt types
+ * @state: interrupt states
+ *
+ * Set VCN block interrupt state
+ */
+static int vcn_v4_0_set_interrupt_state(struct amdgpu_device *adev, struct amdgpu_irq_src *source,
+      unsigned type, enum amdgpu_interrupt_state state)
+{
+       return 0;
+}
+
+/**
+ * vcn_v4_0_process_interrupt - process VCN block interrupt
+ *
+ * @adev: amdgpu_device pointer
+ * @source: interrupt sources
+ * @entry: interrupt entry from clients and sources
+ *
+ * Process VCN block interrupt
+ */
+static int vcn_v4_0_process_interrupt(struct amdgpu_device *adev, struct amdgpu_irq_src *source,
+      struct amdgpu_iv_entry *entry)
+{
+       uint32_t ip_instance;
+
+       switch (entry->client_id) {
+       case SOC15_IH_CLIENTID_VCN:
+               ip_instance = 0;
+               break;
+       case SOC15_IH_CLIENTID_VCN1:
+               ip_instance = 1;
+               break;
+       default:
+               DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
+               return 0;
+       }
+
+       DRM_DEBUG("IH: VCN TRAP\n");
+
+       switch (entry->src_id) {
+       case VCN_4_0__SRCID__UVD_TRAP:
+               if (!unifiedQ_enabled) {
+                       amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec);
+                       break;
+               }
+               break;
+       case VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
+               amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
+               break;
+       case VCN_4_0__SRCID__UVD_ENC_LOW_LATENCY:
+               amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]);
+               break;
+       default:
+               DRM_ERROR("Unhandled interrupt: %d %d\n",
+                         entry->src_id, entry->src_data[0]);
+               break;
+       }
+
+       return 0;
+}
+
+static const struct amdgpu_irq_src_funcs vcn_v4_0_irq_funcs = {
+       .set = vcn_v4_0_set_interrupt_state,
+       .process = vcn_v4_0_process_interrupt,
+};
+
+/**
+ * vcn_v4_0_set_irq_funcs - set VCN block interrupt irq functions
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Set VCN block interrupt irq functions
+ */
+static void vcn_v4_0_set_irq_funcs(struct amdgpu_device *adev)
+{
+       int i;
+
+       for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
+               if (adev->vcn.harvest_config & (1 << i))
+                       continue;
+
+               adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
+               adev->vcn.inst[i].irq.funcs = &vcn_v4_0_irq_funcs;
+       }
+}
+
+static const struct amd_ip_funcs vcn_v4_0_ip_funcs = {
+       .name = "vcn_v4_0",
+       .early_init = vcn_v4_0_early_init,
+       .late_init = NULL,
+       .sw_init = vcn_v4_0_sw_init,
+       .sw_fini = vcn_v4_0_sw_fini,
+       .hw_init = vcn_v4_0_hw_init,
+       .hw_fini = vcn_v4_0_hw_fini,
+       .suspend = vcn_v4_0_suspend,
+       .resume = vcn_v4_0_resume,
+       .is_idle = vcn_v4_0_is_idle,
+       .wait_for_idle = vcn_v4_0_wait_for_idle,
+       .check_soft_reset = NULL,
+       .pre_soft_reset = NULL,
+       .soft_reset = NULL,
+       .post_soft_reset = NULL,
+       .set_clockgating_state = vcn_v4_0_set_clockgating_state,
+       .set_powergating_state = vcn_v4_0_set_powergating_state,
+};
+
+const struct amdgpu_ip_block_version vcn_v4_0_ip_block =
+{
+       .type = AMD_IP_BLOCK_TYPE_VCN,
+       .major = 4,
+       .minor = 0,
+       .rev = 0,
+       .funcs = &vcn_v4_0_ip_funcs,
+};
diff --git a/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.h b/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.h
new file mode 100644 (file)
index 0000000..7c5c9d9
--- /dev/null
@@ -0,0 +1,29 @@
+/*
+ * Copyright 2021 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __VCN_V4_0_H__
+#define __VCN_V4_0_H__
+
+extern const struct amdgpu_ip_block_version vcn_v4_0_ip_block;
+
+#endif /* __VCN_V4_0_H__ */