Linux Audio

Check our new training course

Loading...
v6.2
   1/*
   2 * Copyright 2019 Advanced Micro Devices, Inc.
   3 *
   4 * Permission is hereby granted, free of charge, to any person obtaining a
   5 * copy of this software and associated documentation files (the "Software"),
   6 * to deal in the Software without restriction, including without limitation
   7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8 * and/or sell copies of the Software, and to permit persons to whom the
   9 * Software is furnished to do so, subject to the following conditions:
  10 *
  11 * The above copyright notice and this permission notice shall be included in
  12 * all copies or substantial portions of the Software.
  13 *
  14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
  18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  20 * OTHER DEALINGS IN THE SOFTWARE.
  21 *
  22 */
  23
  24#include <linux/firmware.h>
  25#include "amdgpu.h"
  26#include "amdgpu_vcn.h"
  27#include "amdgpu_pm.h"
  28#include "amdgpu_cs.h"
  29#include "soc15.h"
  30#include "soc15d.h"
  31#include "vcn_v2_0.h"
  32#include "mmsch_v3_0.h"
  33#include "vcn_sw_ring.h"
  34
  35#include "vcn/vcn_3_0_0_offset.h"
  36#include "vcn/vcn_3_0_0_sh_mask.h"
  37#include "ivsrcid/vcn/irqsrcs_vcn_2_0.h"
  38
  39#include <drm/drm_drv.h>
  40
  41#define VCN_VID_SOC_ADDRESS_2_0					0x1fa00
  42#define VCN1_VID_SOC_ADDRESS_3_0				0x48200
  43
  44#define mmUVD_CONTEXT_ID_INTERNAL_OFFSET			0x27
  45#define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET			0x0f
  46#define mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET			0x10
  47#define mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET			0x11
  48#define mmUVD_NO_OP_INTERNAL_OFFSET				0x29
  49#define mmUVD_GP_SCRATCH8_INTERNAL_OFFSET			0x66
  50#define mmUVD_SCRATCH9_INTERNAL_OFFSET				0xc01d
  51
  52#define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET			0x431
  53#define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET		0x3b4
  54#define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET		0x3b5
  55#define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET			0x25c
  56
  57#define VCN_INSTANCES_SIENNA_CICHLID				2
  58#define DEC_SW_RING_ENABLED					FALSE
  59
  60#define RDECODE_MSG_CREATE					0x00000000
  61#define RDECODE_MESSAGE_CREATE					0x00000001
  62
  63static int amdgpu_ih_clientid_vcns[] = {
  64	SOC15_IH_CLIENTID_VCN,
  65	SOC15_IH_CLIENTID_VCN1
  66};
  67
 
 
 
 
 
  68static int vcn_v3_0_start_sriov(struct amdgpu_device *adev);
  69static void vcn_v3_0_set_dec_ring_funcs(struct amdgpu_device *adev);
  70static void vcn_v3_0_set_enc_ring_funcs(struct amdgpu_device *adev);
  71static void vcn_v3_0_set_irq_funcs(struct amdgpu_device *adev);
  72static int vcn_v3_0_set_powergating_state(void *handle,
  73			enum amd_powergating_state state);
  74static int vcn_v3_0_pause_dpg_mode(struct amdgpu_device *adev,
  75			int inst_idx, struct dpg_pause_state *new_state);
  76
  77static void vcn_v3_0_dec_ring_set_wptr(struct amdgpu_ring *ring);
  78static void vcn_v3_0_enc_ring_set_wptr(struct amdgpu_ring *ring);
  79
  80/**
  81 * vcn_v3_0_early_init - set function pointers
  82 *
  83 * @handle: amdgpu_device pointer
  84 *
  85 * Set ring and irq function pointers
  86 */
  87static int vcn_v3_0_early_init(void *handle)
  88{
  89	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
  90
  91	if (amdgpu_sriov_vf(adev)) {
  92		adev->vcn.num_vcn_inst = VCN_INSTANCES_SIENNA_CICHLID;
 
 
  93		adev->vcn.harvest_config = 0;
  94		adev->vcn.num_enc_rings = 1;
  95
  96	} else {
  97		if (adev->vcn.harvest_config == (AMDGPU_VCN_HARVEST_VCN0 |
  98						 AMDGPU_VCN_HARVEST_VCN1))
  99			/* both instances are harvested, disable the block */
 100			return -ENOENT;
 
 
 
 
 
 101
 102		if (adev->ip_versions[UVD_HWIP][0] == IP_VERSION(3, 0, 33))
 
 
 
 
 
 
 
 103			adev->vcn.num_enc_rings = 0;
 104		else
 105			adev->vcn.num_enc_rings = 2;
 106	}
 107
 108	vcn_v3_0_set_dec_ring_funcs(adev);
 109	vcn_v3_0_set_enc_ring_funcs(adev);
 110	vcn_v3_0_set_irq_funcs(adev);
 111
 112	return 0;
 113}
 114
 115/**
 116 * vcn_v3_0_sw_init - sw init for VCN block
 117 *
 118 * @handle: amdgpu_device pointer
 119 *
 120 * Load firmware and sw initialization
 121 */
 122static int vcn_v3_0_sw_init(void *handle)
 123{
 124	struct amdgpu_ring *ring;
 125	int i, j, r;
 126	int vcn_doorbell_index = 0;
 127	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 128
 129	r = amdgpu_vcn_sw_init(adev);
 130	if (r)
 131		return r;
 132
 133	amdgpu_vcn_setup_ucode(adev);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 134
 135	r = amdgpu_vcn_resume(adev);
 136	if (r)
 137		return r;
 138
 139	/*
 140	 * Note: doorbell assignment is fixed for SRIOV multiple VCN engines
 141	 * Formula:
 142	 *   vcn_db_base  = adev->doorbell_index.vcn.vcn_ring0_1 << 1;
 143	 *   dec_ring_i   = vcn_db_base + i * (adev->vcn.num_enc_rings + 1)
 144	 *   enc_ring_i,j = vcn_db_base + i * (adev->vcn.num_enc_rings + 1) + 1 + j
 145	 */
 146	if (amdgpu_sriov_vf(adev)) {
 147		vcn_doorbell_index = adev->doorbell_index.vcn.vcn_ring0_1;
 148		/* get DWORD offset */
 149		vcn_doorbell_index = vcn_doorbell_index << 1;
 150	}
 151
 152	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
 153		volatile struct amdgpu_fw_shared *fw_shared;
 154
 155		if (adev->vcn.harvest_config & (1 << i))
 156			continue;
 157
 158		adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
 159		adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
 160		adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
 161		adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
 162		adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
 163		adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
 164
 165		adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
 166		adev->vcn.inst[i].external.scratch9 = SOC15_REG_OFFSET(VCN, i, mmUVD_SCRATCH9);
 167		adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
 168		adev->vcn.inst[i].external.data0 = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_DATA0);
 169		adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
 170		adev->vcn.inst[i].external.data1 = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_DATA1);
 171		adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
 172		adev->vcn.inst[i].external.cmd = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_CMD);
 173		adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
 174		adev->vcn.inst[i].external.nop = SOC15_REG_OFFSET(VCN, i, mmUVD_NO_OP);
 175
 176		/* VCN DEC TRAP */
 177		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
 178				VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst[i].irq);
 179		if (r)
 180			return r;
 181
 182		atomic_set(&adev->vcn.inst[i].sched_score, 0);
 183
 184		ring = &adev->vcn.inst[i].ring_dec;
 185		ring->use_doorbell = true;
 186		if (amdgpu_sriov_vf(adev)) {
 187			ring->doorbell_index = vcn_doorbell_index + i * (adev->vcn.num_enc_rings + 1);
 188		} else {
 189			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i;
 190		}
 191		sprintf(ring->name, "vcn_dec_%d", i);
 192		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
 193				     AMDGPU_RING_PRIO_DEFAULT,
 194				     &adev->vcn.inst[i].sched_score);
 195		if (r)
 196			return r;
 197
 198		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
 199			enum amdgpu_ring_priority_level hw_prio = amdgpu_vcn_get_enc_ring_prio(j);
 200
 201			/* VCN ENC TRAP */
 202			r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
 203				j + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);
 204			if (r)
 205				return r;
 206
 207			ring = &adev->vcn.inst[i].ring_enc[j];
 208			ring->use_doorbell = true;
 209			if (amdgpu_sriov_vf(adev)) {
 210				ring->doorbell_index = vcn_doorbell_index + i * (adev->vcn.num_enc_rings + 1) + 1 + j;
 211			} else {
 212				ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + j + 8 * i;
 213			}
 214			sprintf(ring->name, "vcn_enc_%d.%d", i, j);
 215			r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
 216					     hw_prio, &adev->vcn.inst[i].sched_score);
 
 217			if (r)
 218				return r;
 219		}
 220
 221		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
 222		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_SW_RING_FLAG) |
 223					     cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG) |
 224					     cpu_to_le32(AMDGPU_VCN_FW_SHARED_FLAG_0_RB);
 225		fw_shared->sw_ring.is_enabled = cpu_to_le32(DEC_SW_RING_ENABLED);
 226		fw_shared->present_flag_0 |= AMDGPU_VCN_SMU_VERSION_INFO_FLAG;
 227		if (adev->ip_versions[UVD_HWIP][0] == IP_VERSION(3, 1, 2))
 228			fw_shared->smu_interface_info.smu_interface_type = 2;
 229		else if (adev->ip_versions[UVD_HWIP][0] == IP_VERSION(3, 1, 1))
 230			fw_shared->smu_interface_info.smu_interface_type = 1;
 231
 232		if (amdgpu_vcnfw_log)
 233			amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
 234	}
 235
 236	if (amdgpu_sriov_vf(adev)) {
 237		r = amdgpu_virt_alloc_mm_table(adev);
 238		if (r)
 239			return r;
 240	}
 241	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
 242		adev->vcn.pause_dpg_mode = vcn_v3_0_pause_dpg_mode;
 243
 244	return 0;
 245}
 246
 247/**
 248 * vcn_v3_0_sw_fini - sw fini for VCN block
 249 *
 250 * @handle: amdgpu_device pointer
 251 *
 252 * VCN suspend and free up sw allocation
 253 */
 254static int vcn_v3_0_sw_fini(void *handle)
 255{
 256	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 257	int i, r, idx;
 258
 259	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
 260		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
 261			volatile struct amdgpu_fw_shared *fw_shared;
 262
 263			if (adev->vcn.harvest_config & (1 << i))
 264				continue;
 265			fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
 266			fw_shared->present_flag_0 = 0;
 267			fw_shared->sw_ring.is_enabled = false;
 268		}
 269
 270		drm_dev_exit(idx);
 271	}
 272
 273	if (amdgpu_sriov_vf(adev))
 274		amdgpu_virt_free_mm_table(adev);
 275
 276	r = amdgpu_vcn_suspend(adev);
 277	if (r)
 278		return r;
 279
 280	r = amdgpu_vcn_sw_fini(adev);
 281
 282	return r;
 283}
 284
 285/**
 286 * vcn_v3_0_hw_init - start and test VCN block
 287 *
 288 * @handle: amdgpu_device pointer
 289 *
 290 * Initialize the hardware, boot up the VCPU and do some testing
 291 */
 292static int vcn_v3_0_hw_init(void *handle)
 293{
 294	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 295	struct amdgpu_ring *ring;
 296	int i, j, r;
 297
 298	if (amdgpu_sriov_vf(adev)) {
 299		r = vcn_v3_0_start_sriov(adev);
 300		if (r)
 301			goto done;
 302
 303		/* initialize VCN dec and enc ring buffers */
 304		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
 305			if (adev->vcn.harvest_config & (1 << i))
 306				continue;
 307
 308			ring = &adev->vcn.inst[i].ring_dec;
 309			if (amdgpu_vcn_is_disabled_vcn(adev, VCN_DECODE_RING, i)) {
 310				ring->sched.ready = false;
 311				ring->no_scheduler = true;
 312				dev_info(adev->dev, "ring %s is disabled by hypervisor\n", ring->name);
 313			} else {
 314				ring->wptr = 0;
 315				ring->wptr_old = 0;
 316				vcn_v3_0_dec_ring_set_wptr(ring);
 317				ring->sched.ready = true;
 318			}
 319
 320			for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
 321				ring = &adev->vcn.inst[i].ring_enc[j];
 322				if (amdgpu_vcn_is_disabled_vcn(adev, VCN_ENCODE_RING, i)) {
 323					ring->sched.ready = false;
 324					ring->no_scheduler = true;
 325					dev_info(adev->dev, "ring %s is disabled by hypervisor\n", ring->name);
 326				} else {
 327					ring->wptr = 0;
 328					ring->wptr_old = 0;
 329					vcn_v3_0_enc_ring_set_wptr(ring);
 330					ring->sched.ready = true;
 331				}
 332			}
 333		}
 334	} else {
 335		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
 336			if (adev->vcn.harvest_config & (1 << i))
 337				continue;
 338
 339			ring = &adev->vcn.inst[i].ring_dec;
 340
 341			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
 342						     ring->doorbell_index, i);
 343
 344			r = amdgpu_ring_test_helper(ring);
 345			if (r)
 346				goto done;
 347
 348			for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
 349				ring = &adev->vcn.inst[i].ring_enc[j];
 350				r = amdgpu_ring_test_helper(ring);
 351				if (r)
 352					goto done;
 353			}
 354		}
 355	}
 356
 357done:
 358	if (!r)
 359		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
 360			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
 361
 362	return r;
 363}
 364
 365/**
 366 * vcn_v3_0_hw_fini - stop the hardware block
 367 *
 368 * @handle: amdgpu_device pointer
 369 *
 370 * Stop the VCN block, mark ring as not ready any more
 371 */
 372static int vcn_v3_0_hw_fini(void *handle)
 373{
 374	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 375	int i;
 376
 377	cancel_delayed_work_sync(&adev->vcn.idle_work);
 378
 379	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
 380		if (adev->vcn.harvest_config & (1 << i))
 381			continue;
 382
 383		if (!amdgpu_sriov_vf(adev)) {
 384			if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
 385					(adev->vcn.cur_state != AMD_PG_STATE_GATE &&
 386					 RREG32_SOC15(VCN, i, mmUVD_STATUS))) {
 387				vcn_v3_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
 388			}
 389		}
 390	}
 391
 392	return 0;
 393}
 394
 395/**
 396 * vcn_v3_0_suspend - suspend VCN block
 397 *
 398 * @handle: amdgpu_device pointer
 399 *
 400 * HW fini and suspend VCN block
 401 */
 402static int vcn_v3_0_suspend(void *handle)
 403{
 404	int r;
 405	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 406
 407	r = vcn_v3_0_hw_fini(adev);
 408	if (r)
 409		return r;
 410
 411	r = amdgpu_vcn_suspend(adev);
 412
 413	return r;
 414}
 415
 416/**
 417 * vcn_v3_0_resume - resume VCN block
 418 *
 419 * @handle: amdgpu_device pointer
 420 *
 421 * Resume firmware and hw init VCN block
 422 */
 423static int vcn_v3_0_resume(void *handle)
 424{
 425	int r;
 426	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 427
 428	r = amdgpu_vcn_resume(adev);
 429	if (r)
 430		return r;
 431
 432	r = vcn_v3_0_hw_init(adev);
 433
 434	return r;
 435}
 436
 437/**
 438 * vcn_v3_0_mc_resume - memory controller programming
 439 *
 440 * @adev: amdgpu_device pointer
 441 * @inst: instance number
 442 *
 443 * Let the VCN memory controller know it's offsets
 444 */
 445static void vcn_v3_0_mc_resume(struct amdgpu_device *adev, int inst)
 446{
 447	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
 448	uint32_t offset;
 449
 450	/* cache window 0: fw */
 451	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
 452		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
 453			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo));
 454		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
 455			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi));
 456		WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET0, 0);
 457		offset = 0;
 458	} else {
 459		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
 460			lower_32_bits(adev->vcn.inst[inst].gpu_addr));
 461		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
 462			upper_32_bits(adev->vcn.inst[inst].gpu_addr));
 463		offset = size;
 464		WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET0,
 465			AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
 466	}
 467	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE0, size);
 468
 469	/* cache window 1: stack */
 470	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
 471		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
 472	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
 473		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
 474	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET1, 0);
 475	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
 476
 477	/* cache window 2: context */
 478	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
 479		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
 480	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
 481		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
 482	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET2, 0);
 483	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
 484
 485	/* non-cache window */
 486	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
 487		lower_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
 488	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
 489		upper_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
 490	WREG32_SOC15(VCN, inst, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
 491	WREG32_SOC15(VCN, inst, mmUVD_VCPU_NONCACHE_SIZE0,
 492		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
 493}
 494
 495static void vcn_v3_0_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
 496{
 497	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
 498	uint32_t offset;
 499
 500	/* cache window 0: fw */
 501	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
 502		if (!indirect) {
 503			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 504				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
 505				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
 506			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 507				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
 508				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
 509			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 510				VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
 511		} else {
 512			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 513				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
 514			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 515				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
 516			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 517				VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
 518		}
 519		offset = 0;
 520	} else {
 521		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 522			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
 523			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
 524		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 525			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
 526			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
 527		offset = size;
 528		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 529			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0),
 530			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
 531	}
 532
 533	if (!indirect)
 534		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 535			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
 536	else
 537		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 538			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
 539
 540	/* cache window 1: stack */
 541	if (!indirect) {
 542		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 543			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
 544			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
 545		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 546			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
 547			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
 548		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 549			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
 550	} else {
 551		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 552			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
 553		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 554			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
 555		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 556			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
 557	}
 558	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 559			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
 560
 561	/* cache window 2: context */
 562	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 563			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
 564			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
 565	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 566			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
 567			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
 568	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 569			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
 570	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 571			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
 572
 573	/* non-cache window */
 574	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 575			VCN, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
 576			lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
 577	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 578			VCN, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
 579			upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
 580	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 581			VCN, inst_idx, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
 582	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 583			VCN, inst_idx, mmUVD_VCPU_NONCACHE_SIZE0),
 584			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
 585
 586	/* VCN global tiling registers */
 587	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 588		UVD, inst_idx, mmUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
 589}
 590
 591static void vcn_v3_0_disable_static_power_gating(struct amdgpu_device *adev, int inst)
 592{
 593	uint32_t data = 0;
 594
 595	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
 596		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
 597			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
 598			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
 599			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
 600			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
 601			| 2 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
 602			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
 603			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
 604			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
 605			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
 606			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
 607			| 2 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
 608			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
 609			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
 610
 611		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
 612		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS,
 613			UVD_PGFSM_STATUS__UVDM_UVDU_UVDLM_PWR_ON_3_0, 0x3F3FFFFF);
 614	} else {
 615		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
 616			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
 617			| 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
 618			| 1 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
 619			| 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
 620			| 1 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
 621			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
 622			| 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
 623			| 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
 624			| 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
 625			| 1 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
 626			| 1 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
 627			| 1 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
 628			| 1 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
 629		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
 630		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS, 0,  0x3F3FFFFF);
 631	}
 632
 633	data = RREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS);
 634	data &= ~0x103;
 635	if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
 636		data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
 637			UVD_POWER_STATUS__UVD_PG_EN_MASK;
 638
 639	WREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS, data);
 640}
 641
 642static void vcn_v3_0_enable_static_power_gating(struct amdgpu_device *adev, int inst)
 643{
 644	uint32_t data;
 645
 646	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
 647		/* Before power off, this indicator has to be turned on */
 648		data = RREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS);
 649		data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
 650		data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
 651		WREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS, data);
 652
 653		data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
 654			| 2 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
 655			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
 656			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
 657			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
 658			| 2 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
 659			| 2 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
 660			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
 661			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
 662			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
 663			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
 664			| 2 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
 665			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
 666			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
 667		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
 668
 669		data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
 670			| 2 << UVD_PGFSM_STATUS__UVDU_PWR_STATUS__SHIFT
 671			| 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
 672			| 2 << UVD_PGFSM_STATUS__UVDC_PWR_STATUS__SHIFT
 673			| 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
 674			| 2 << UVD_PGFSM_STATUS__UVDIRL_PWR_STATUS__SHIFT
 675			| 2 << UVD_PGFSM_STATUS__UVDLM_PWR_STATUS__SHIFT
 676			| 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
 677			| 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
 678			| 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT
 679			| 2 << UVD_PGFSM_STATUS__UVDAB_PWR_STATUS__SHIFT
 680			| 2 << UVD_PGFSM_STATUS__UVDATD_PWR_STATUS__SHIFT
 681			| 2 << UVD_PGFSM_STATUS__UVDNA_PWR_STATUS__SHIFT
 682			| 2 << UVD_PGFSM_STATUS__UVDNB_PWR_STATUS__SHIFT);
 683		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS, data, 0x3F3FFFFF);
 684	}
 685}
 686
 687/**
 688 * vcn_v3_0_disable_clock_gating - disable VCN clock gating
 689 *
 690 * @adev: amdgpu_device pointer
 691 * @inst: instance number
 692 *
 693 * Disable clock gating for VCN block
 694 */
 695static void vcn_v3_0_disable_clock_gating(struct amdgpu_device *adev, int inst)
 696{
 697	uint32_t data;
 698
 699	/* VCN disable CGC */
 700	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
 701	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
 702		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
 703	else
 704		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
 705	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
 706	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
 707	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
 708
 709	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_GATE);
 710	data &= ~(UVD_CGC_GATE__SYS_MASK
 711		| UVD_CGC_GATE__UDEC_MASK
 712		| UVD_CGC_GATE__MPEG2_MASK
 713		| UVD_CGC_GATE__REGS_MASK
 714		| UVD_CGC_GATE__RBC_MASK
 715		| UVD_CGC_GATE__LMI_MC_MASK
 716		| UVD_CGC_GATE__LMI_UMC_MASK
 717		| UVD_CGC_GATE__IDCT_MASK
 718		| UVD_CGC_GATE__MPRD_MASK
 719		| UVD_CGC_GATE__MPC_MASK
 720		| UVD_CGC_GATE__LBSI_MASK
 721		| UVD_CGC_GATE__LRBBM_MASK
 722		| UVD_CGC_GATE__UDEC_RE_MASK
 723		| UVD_CGC_GATE__UDEC_CM_MASK
 724		| UVD_CGC_GATE__UDEC_IT_MASK
 725		| UVD_CGC_GATE__UDEC_DB_MASK
 726		| UVD_CGC_GATE__UDEC_MP_MASK
 727		| UVD_CGC_GATE__WCB_MASK
 728		| UVD_CGC_GATE__VCPU_MASK
 729		| UVD_CGC_GATE__MMSCH_MASK);
 730
 731	WREG32_SOC15(VCN, inst, mmUVD_CGC_GATE, data);
 732
 733	SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_CGC_GATE, 0,  0xFFFFFFFF);
 734
 735	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
 736	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
 737		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
 738		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
 739		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
 740		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
 741		| UVD_CGC_CTRL__SYS_MODE_MASK
 742		| UVD_CGC_CTRL__UDEC_MODE_MASK
 743		| UVD_CGC_CTRL__MPEG2_MODE_MASK
 744		| UVD_CGC_CTRL__REGS_MODE_MASK
 745		| UVD_CGC_CTRL__RBC_MODE_MASK
 746		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
 747		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
 748		| UVD_CGC_CTRL__IDCT_MODE_MASK
 749		| UVD_CGC_CTRL__MPRD_MODE_MASK
 750		| UVD_CGC_CTRL__MPC_MODE_MASK
 751		| UVD_CGC_CTRL__LBSI_MODE_MASK
 752		| UVD_CGC_CTRL__LRBBM_MODE_MASK
 753		| UVD_CGC_CTRL__WCB_MODE_MASK
 754		| UVD_CGC_CTRL__VCPU_MODE_MASK
 755		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
 756	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
 757
 758	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE);
 759	data |= (UVD_SUVD_CGC_GATE__SRE_MASK
 760		| UVD_SUVD_CGC_GATE__SIT_MASK
 761		| UVD_SUVD_CGC_GATE__SMP_MASK
 762		| UVD_SUVD_CGC_GATE__SCM_MASK
 763		| UVD_SUVD_CGC_GATE__SDB_MASK
 764		| UVD_SUVD_CGC_GATE__SRE_H264_MASK
 765		| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
 766		| UVD_SUVD_CGC_GATE__SIT_H264_MASK
 767		| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
 768		| UVD_SUVD_CGC_GATE__SCM_H264_MASK
 769		| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
 770		| UVD_SUVD_CGC_GATE__SDB_H264_MASK
 771		| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
 772		| UVD_SUVD_CGC_GATE__SCLR_MASK
 773		| UVD_SUVD_CGC_GATE__ENT_MASK
 774		| UVD_SUVD_CGC_GATE__IME_MASK
 775		| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
 776		| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
 777		| UVD_SUVD_CGC_GATE__SITE_MASK
 778		| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
 779		| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
 780		| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
 781		| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
 782		| UVD_SUVD_CGC_GATE__IME_HEVC_MASK
 783		| UVD_SUVD_CGC_GATE__EFC_MASK
 784		| UVD_SUVD_CGC_GATE__SAOE_MASK
 785		| UVD_SUVD_CGC_GATE__SRE_AV1_MASK
 786		| UVD_SUVD_CGC_GATE__FBC_PCLK_MASK
 787		| UVD_SUVD_CGC_GATE__FBC_CCLK_MASK
 788		| UVD_SUVD_CGC_GATE__SCM_AV1_MASK
 789		| UVD_SUVD_CGC_GATE__SMPA_MASK);
 790	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE, data);
 791
 792	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE2);
 793	data |= (UVD_SUVD_CGC_GATE2__MPBE0_MASK
 794		| UVD_SUVD_CGC_GATE2__MPBE1_MASK
 795		| UVD_SUVD_CGC_GATE2__SIT_AV1_MASK
 796		| UVD_SUVD_CGC_GATE2__SDB_AV1_MASK
 797		| UVD_SUVD_CGC_GATE2__MPC1_MASK);
 798	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE2, data);
 799
 800	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL);
 801	data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
 802		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
 803		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
 804		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
 805		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
 806		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
 807		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
 808		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
 809		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK
 810		| UVD_SUVD_CGC_CTRL__EFC_MODE_MASK
 811		| UVD_SUVD_CGC_CTRL__SAOE_MODE_MASK
 812		| UVD_SUVD_CGC_CTRL__SMPA_MODE_MASK
 813		| UVD_SUVD_CGC_CTRL__MPBE0_MODE_MASK
 814		| UVD_SUVD_CGC_CTRL__MPBE1_MODE_MASK
 815		| UVD_SUVD_CGC_CTRL__SIT_AV1_MODE_MASK
 816		| UVD_SUVD_CGC_CTRL__SDB_AV1_MODE_MASK
 817		| UVD_SUVD_CGC_CTRL__MPC1_MODE_MASK
 818		| UVD_SUVD_CGC_CTRL__FBC_PCLK_MASK
 819		| UVD_SUVD_CGC_CTRL__FBC_CCLK_MASK);
 820	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL, data);
 821}
 822
 823static void vcn_v3_0_clock_gating_dpg_mode(struct amdgpu_device *adev,
 824		uint8_t sram_sel, int inst_idx, uint8_t indirect)
 825{
 826	uint32_t reg_data = 0;
 827
 828	/* enable sw clock gating control */
 829	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
 830		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
 831	else
 832		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
 833	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
 834	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
 835	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
 836		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
 837		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
 838		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
 839		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
 840		 UVD_CGC_CTRL__SYS_MODE_MASK |
 841		 UVD_CGC_CTRL__UDEC_MODE_MASK |
 842		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
 843		 UVD_CGC_CTRL__REGS_MODE_MASK |
 844		 UVD_CGC_CTRL__RBC_MODE_MASK |
 845		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
 846		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
 847		 UVD_CGC_CTRL__IDCT_MODE_MASK |
 848		 UVD_CGC_CTRL__MPRD_MODE_MASK |
 849		 UVD_CGC_CTRL__MPC_MODE_MASK |
 850		 UVD_CGC_CTRL__LBSI_MODE_MASK |
 851		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
 852		 UVD_CGC_CTRL__WCB_MODE_MASK |
 853		 UVD_CGC_CTRL__VCPU_MODE_MASK |
 854		 UVD_CGC_CTRL__MMSCH_MODE_MASK);
 855	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 856		VCN, inst_idx, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
 857
 858	/* turn off clock gating */
 859	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 860		VCN, inst_idx, mmUVD_CGC_GATE), 0, sram_sel, indirect);
 861
 862	/* turn on SUVD clock gating */
 863	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 864		VCN, inst_idx, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
 865
 866	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
 867	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 868		VCN, inst_idx, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
 869}
 870
 871/**
 872 * vcn_v3_0_enable_clock_gating - enable VCN clock gating
 873 *
 874 * @adev: amdgpu_device pointer
 875 * @inst: instance number
 876 *
 877 * Enable clock gating for VCN block
 878 */
 879static void vcn_v3_0_enable_clock_gating(struct amdgpu_device *adev, int inst)
 880{
 881	uint32_t data;
 882
 883	/* enable VCN CGC */
 884	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
 885	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
 886		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
 887	else
 888		data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
 889	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
 890	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
 891	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
 892
 893	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
 894	data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
 895		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
 896		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
 897		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
 898		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
 899		| UVD_CGC_CTRL__SYS_MODE_MASK
 900		| UVD_CGC_CTRL__UDEC_MODE_MASK
 901		| UVD_CGC_CTRL__MPEG2_MODE_MASK
 902		| UVD_CGC_CTRL__REGS_MODE_MASK
 903		| UVD_CGC_CTRL__RBC_MODE_MASK
 904		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
 905		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
 906		| UVD_CGC_CTRL__IDCT_MODE_MASK
 907		| UVD_CGC_CTRL__MPRD_MODE_MASK
 908		| UVD_CGC_CTRL__MPC_MODE_MASK
 909		| UVD_CGC_CTRL__LBSI_MODE_MASK
 910		| UVD_CGC_CTRL__LRBBM_MODE_MASK
 911		| UVD_CGC_CTRL__WCB_MODE_MASK
 912		| UVD_CGC_CTRL__VCPU_MODE_MASK
 913		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
 914	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
 915
 916	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL);
 917	data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
 918		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
 919		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
 920		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
 921		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
 922		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
 923		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
 924		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
 925		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK
 926		| UVD_SUVD_CGC_CTRL__EFC_MODE_MASK
 927		| UVD_SUVD_CGC_CTRL__SAOE_MODE_MASK
 928		| UVD_SUVD_CGC_CTRL__SMPA_MODE_MASK
 929		| UVD_SUVD_CGC_CTRL__MPBE0_MODE_MASK
 930		| UVD_SUVD_CGC_CTRL__MPBE1_MODE_MASK
 931		| UVD_SUVD_CGC_CTRL__SIT_AV1_MODE_MASK
 932		| UVD_SUVD_CGC_CTRL__SDB_AV1_MODE_MASK
 933		| UVD_SUVD_CGC_CTRL__MPC1_MODE_MASK
 934		| UVD_SUVD_CGC_CTRL__FBC_PCLK_MASK
 935		| UVD_SUVD_CGC_CTRL__FBC_CCLK_MASK);
 936	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL, data);
 937}
 938
 939static int vcn_v3_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
 940{
 941	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
 942	struct amdgpu_ring *ring;
 943	uint32_t rb_bufsz, tmp;
 944
 945	/* disable register anti-hang mechanism */
 946	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1,
 947		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
 948	/* enable dynamic power gating mode */
 949	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS);
 950	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
 951	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
 952	WREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS, tmp);
 953
 954	if (indirect)
 955		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
 956
 957	/* enable clock gating */
 958	vcn_v3_0_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
 959
 960	/* enable VCPU clock */
 961	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
 962	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
 963	tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
 964	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 965		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
 966
 967	/* disable master interupt */
 968	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 969		VCN, inst_idx, mmUVD_MASTINT_EN), 0, 0, indirect);
 970
 971	/* setup mmUVD_LMI_CTRL */
 972	tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
 973		UVD_LMI_CTRL__REQ_MODE_MASK |
 974		UVD_LMI_CTRL__CRC_RESET_MASK |
 975		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
 976		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
 977		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
 978		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
 979		0x00100000L);
 980	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 981		VCN, inst_idx, mmUVD_LMI_CTRL), tmp, 0, indirect);
 982
 983	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 984		VCN, inst_idx, mmUVD_MPC_CNTL),
 985		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
 986
 987	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 988		VCN, inst_idx, mmUVD_MPC_SET_MUXA0),
 989		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
 990		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
 991		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
 992		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
 993
 994	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 995		VCN, inst_idx, mmUVD_MPC_SET_MUXB0),
 996		 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
 997		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
 998		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
 999		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
1000
1001	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1002		VCN, inst_idx, mmUVD_MPC_SET_MUX),
1003		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1004		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1005		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
1006
1007	vcn_v3_0_mc_resume_dpg_mode(adev, inst_idx, indirect);
1008
1009	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1010		VCN, inst_idx, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
1011	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1012		VCN, inst_idx, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
1013
1014	/* enable LMI MC and UMC channels */
1015	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1016		VCN, inst_idx, mmUVD_LMI_CTRL2), 0, 0, indirect);
1017
1018	/* unblock VCPU register access */
1019	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1020		VCN, inst_idx, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
1021
1022	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
1023	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
1024	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1025		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
1026
1027	/* enable master interrupt */
1028	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1029		VCN, inst_idx, mmUVD_MASTINT_EN),
1030		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
1031
1032	/* add nop to workaround PSP size check */
1033	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1034		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
1035
1036	if (indirect)
1037		psp_update_vcn_sram(adev, inst_idx, adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
1038			(uint32_t)((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
1039				(uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr));
1040
1041	ring = &adev->vcn.inst[inst_idx].ring_dec;
1042	/* force RBC into idle state */
1043	rb_bufsz = order_base_2(ring->ring_size);
1044	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1045	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1046	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1047	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1048	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1049	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
1050
1051	/* Stall DPG before WPTR/RPTR reset */
1052	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1053		UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1054		~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1055	fw_shared->multi_queue.decode_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1056
1057	/* set the write pointer delay */
1058	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
1059
1060	/* set the wb address */
1061	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR_ADDR,
1062		(upper_32_bits(ring->gpu_addr) >> 2));
1063
1064	/* programm the RB_BASE for ring buffer */
1065	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1066		lower_32_bits(ring->gpu_addr));
1067	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1068		upper_32_bits(ring->gpu_addr));
1069
1070	/* Initialize the ring buffer's read and write pointers */
1071	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, 0);
1072
1073	WREG32_SOC15(VCN, inst_idx, mmUVD_SCRATCH2, 0);
1074
1075	ring->wptr = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR);
1076	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR,
1077		lower_32_bits(ring->wptr));
1078
1079	/* Reset FW shared memory RBC WPTR/RPTR */
1080	fw_shared->rb.rptr = 0;
1081	fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1082
1083	/*resetting done, fw can check RB ring */
1084	fw_shared->multi_queue.decode_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1085
1086	/* Unstall DPG */
1087	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1088		0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1089
1090	return 0;
1091}
1092
1093static int vcn_v3_0_start(struct amdgpu_device *adev)
1094{
1095	volatile struct amdgpu_fw_shared *fw_shared;
1096	struct amdgpu_ring *ring;
1097	uint32_t rb_bufsz, tmp;
1098	int i, j, k, r;
1099
1100	if (adev->pm.dpm_enabled)
1101		amdgpu_dpm_enable_uvd(adev, true);
1102
1103	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1104		if (adev->vcn.harvest_config & (1 << i))
1105			continue;
1106
1107		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG){
1108			r = vcn_v3_0_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
1109			continue;
1110		}
1111
1112		/* disable VCN power gating */
1113		vcn_v3_0_disable_static_power_gating(adev, i);
1114
1115		/* set VCN status busy */
1116		tmp = RREG32_SOC15(VCN, i, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY;
1117		WREG32_SOC15(VCN, i, mmUVD_STATUS, tmp);
1118
1119		/*SW clock gating */
1120		vcn_v3_0_disable_clock_gating(adev, i);
1121
1122		/* enable VCPU clock */
1123		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1124			UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1125
1126		/* disable master interrupt */
1127		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN), 0,
1128			~UVD_MASTINT_EN__VCPU_EN_MASK);
1129
1130		/* enable LMI MC and UMC channels */
1131		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_LMI_CTRL2), 0,
1132			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1133
1134		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1135		tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1136		tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1137		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1138
1139		/* setup mmUVD_LMI_CTRL */
1140		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL);
1141		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL, tmp |
1142			UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK	|
1143			UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1144			UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1145			UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1146
1147		/* setup mmUVD_MPC_CNTL */
1148		tmp = RREG32_SOC15(VCN, i, mmUVD_MPC_CNTL);
1149		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1150		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1151		WREG32_SOC15(VCN, i, mmUVD_MPC_CNTL, tmp);
1152
1153		/* setup UVD_MPC_SET_MUXA0 */
1154		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXA0,
1155			((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1156			(0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1157			(0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1158			(0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1159
1160		/* setup UVD_MPC_SET_MUXB0 */
1161		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXB0,
1162			((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1163			(0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1164			(0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1165			(0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1166
1167		/* setup mmUVD_MPC_SET_MUX */
1168		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUX,
1169			((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1170			(0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1171			(0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1172
1173		vcn_v3_0_mc_resume(adev, i);
1174
1175		/* VCN global tiling registers */
1176		WREG32_SOC15(VCN, i, mmUVD_GFX10_ADDR_CONFIG,
1177			adev->gfx.config.gb_addr_config);
1178
1179		/* unblock VCPU register access */
1180		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL), 0,
1181			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1182
1183		/* release VCPU reset to boot */
1184		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1185			~UVD_VCPU_CNTL__BLK_RST_MASK);
1186
1187		for (j = 0; j < 10; ++j) {
1188			uint32_t status;
1189
1190			for (k = 0; k < 100; ++k) {
1191				status = RREG32_SOC15(VCN, i, mmUVD_STATUS);
1192				if (status & 2)
1193					break;
1194				mdelay(10);
1195			}
1196			r = 0;
1197			if (status & 2)
1198				break;
1199
1200			DRM_ERROR("VCN[%d] decode not responding, trying to reset the VCPU!!!\n", i);
1201			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1202				UVD_VCPU_CNTL__BLK_RST_MASK,
1203				~UVD_VCPU_CNTL__BLK_RST_MASK);
1204			mdelay(10);
1205			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1206				~UVD_VCPU_CNTL__BLK_RST_MASK);
1207
1208			mdelay(10);
1209			r = -1;
1210		}
1211
1212		if (r) {
1213			DRM_ERROR("VCN[%d] decode not responding, giving up!!!\n", i);
1214			return r;
1215		}
1216
1217		/* enable master interrupt */
1218		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN),
1219			UVD_MASTINT_EN__VCPU_EN_MASK,
1220			~UVD_MASTINT_EN__VCPU_EN_MASK);
1221
1222		/* clear the busy bit of VCN_STATUS */
1223		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS), 0,
1224			~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1225
1226		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_VMID, 0);
1227
1228		ring = &adev->vcn.inst[i].ring_dec;
1229		/* force RBC into idle state */
1230		rb_bufsz = order_base_2(ring->ring_size);
1231		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1232		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1233		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1234		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1235		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1236		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_CNTL, tmp);
1237
1238		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1239		fw_shared->multi_queue.decode_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1240
1241		/* programm the RB_BASE for ring buffer */
1242		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1243			lower_32_bits(ring->gpu_addr));
1244		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1245			upper_32_bits(ring->gpu_addr));
1246
1247		/* Initialize the ring buffer's read and write pointers */
1248		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR, 0);
1249
1250		WREG32_SOC15(VCN, i, mmUVD_SCRATCH2, 0);
1251		ring->wptr = RREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR);
1252		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_WPTR,
1253			lower_32_bits(ring->wptr));
1254		fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1255		fw_shared->multi_queue.decode_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1256
1257		if (adev->ip_versions[UVD_HWIP][0] != IP_VERSION(3, 0, 33)) {
1258			fw_shared->multi_queue.encode_generalpurpose_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1259			ring = &adev->vcn.inst[i].ring_enc[0];
1260			WREG32_SOC15(VCN, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1261			WREG32_SOC15(VCN, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1262			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO, ring->gpu_addr);
1263			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1264			WREG32_SOC15(VCN, i, mmUVD_RB_SIZE, ring->ring_size / 4);
1265			fw_shared->multi_queue.encode_generalpurpose_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1266
1267			fw_shared->multi_queue.encode_lowlatency_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1268			ring = &adev->vcn.inst[i].ring_enc[1];
1269			WREG32_SOC15(VCN, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1270			WREG32_SOC15(VCN, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1271			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1272			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1273			WREG32_SOC15(VCN, i, mmUVD_RB_SIZE2, ring->ring_size / 4);
1274			fw_shared->multi_queue.encode_lowlatency_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1275		}
1276	}
1277
1278	return 0;
1279}
1280
1281static int vcn_v3_0_start_sriov(struct amdgpu_device *adev)
1282{
1283	int i, j;
1284	struct amdgpu_ring *ring;
1285	uint64_t cache_addr;
1286	uint64_t rb_addr;
1287	uint64_t ctx_addr;
1288	uint32_t param, resp, expected;
1289	uint32_t offset, cache_size;
1290	uint32_t tmp, timeout;
 
1291
1292	struct amdgpu_mm_table *table = &adev->virt.mm_table;
1293	uint32_t *table_loc;
1294	uint32_t table_size;
1295	uint32_t size, size_dw;
1296
1297	struct mmsch_v3_0_cmd_direct_write
1298		direct_wt = { {0} };
1299	struct mmsch_v3_0_cmd_direct_read_modify_write
1300		direct_rd_mod_wt = { {0} };
1301	struct mmsch_v3_0_cmd_end end = { {0} };
1302	struct mmsch_v3_0_init_header header;
1303
1304	direct_wt.cmd_header.command_type =
1305		MMSCH_COMMAND__DIRECT_REG_WRITE;
1306	direct_rd_mod_wt.cmd_header.command_type =
1307		MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1308	end.cmd_header.command_type =
1309		MMSCH_COMMAND__END;
1310
1311	header.version = MMSCH_VERSION;
1312	header.total_size = sizeof(struct mmsch_v3_0_init_header) >> 2;
1313	for (i = 0; i < AMDGPU_MAX_VCN_INSTANCES; i++) {
1314		header.inst[i].init_status = 0;
1315		header.inst[i].table_offset = 0;
1316		header.inst[i].table_size = 0;
1317	}
1318
1319	table_loc = (uint32_t *)table->cpu_addr;
1320	table_loc += header.total_size;
1321	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1322		if (adev->vcn.harvest_config & (1 << i))
1323			continue;
1324
1325		table_size = 0;
1326
1327		MMSCH_V3_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(VCN, i,
1328			mmUVD_STATUS),
1329			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1330
1331		cache_size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1332
1333		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
 
1334			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1335				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1336				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1337			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1338				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1339				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1340			offset = 0;
1341			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1342				mmUVD_VCPU_CACHE_OFFSET0),
1343				0);
1344		} else {
1345			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1346				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1347				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1348			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1349				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1350				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1351			offset = cache_size;
1352			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1353				mmUVD_VCPU_CACHE_OFFSET0),
1354				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1355		}
1356
1357		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1358			mmUVD_VCPU_CACHE_SIZE0),
1359			cache_size);
1360
1361		cache_addr = adev->vcn.inst[i].gpu_addr + offset;
1362		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1363			mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1364			lower_32_bits(cache_addr));
1365		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1366			mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1367			upper_32_bits(cache_addr));
1368		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1369			mmUVD_VCPU_CACHE_OFFSET1),
1370			0);
1371		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1372			mmUVD_VCPU_CACHE_SIZE1),
1373			AMDGPU_VCN_STACK_SIZE);
1374
1375		cache_addr = adev->vcn.inst[i].gpu_addr + offset +
1376			AMDGPU_VCN_STACK_SIZE;
1377		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1378			mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1379			lower_32_bits(cache_addr));
1380		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1381			mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1382			upper_32_bits(cache_addr));
1383		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1384			mmUVD_VCPU_CACHE_OFFSET2),
1385			0);
1386		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1387			mmUVD_VCPU_CACHE_SIZE2),
1388			AMDGPU_VCN_CONTEXT_SIZE);
1389
1390		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
1391			ring = &adev->vcn.inst[i].ring_enc[j];
1392			ring->wptr = 0;
1393			rb_addr = ring->gpu_addr;
1394			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1395				mmUVD_RB_BASE_LO),
1396				lower_32_bits(rb_addr));
1397			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1398				mmUVD_RB_BASE_HI),
1399				upper_32_bits(rb_addr));
1400			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1401				mmUVD_RB_SIZE),
1402				ring->ring_size / 4);
1403		}
1404
1405		ring = &adev->vcn.inst[i].ring_dec;
1406		ring->wptr = 0;
1407		rb_addr = ring->gpu_addr;
1408		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1409			mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1410			lower_32_bits(rb_addr));
1411		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1412			mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1413			upper_32_bits(rb_addr));
1414		/* force RBC into idle state */
1415		tmp = order_base_2(ring->ring_size);
1416		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, tmp);
1417		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1418		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1419		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1420		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1421		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1422			mmUVD_RBC_RB_CNTL),
1423			tmp);
1424
1425		/* add end packet */
1426		MMSCH_V3_0_INSERT_END();
1427
1428		/* refine header */
1429		header.inst[i].init_status = 0;
1430		header.inst[i].table_offset = header.total_size;
1431		header.inst[i].table_size = table_size;
1432		header.total_size += table_size;
1433	}
1434
1435	/* Update init table header in memory */
1436	size = sizeof(struct mmsch_v3_0_init_header);
1437	table_loc = (uint32_t *)table->cpu_addr;
1438	memcpy((void *)table_loc, &header, size);
1439
1440	/* message MMSCH (in VCN[0]) to initialize this client
1441	 * 1, write to mmsch_vf_ctx_addr_lo/hi register with GPU mc addr
1442	 * of memory descriptor location
1443	 */
1444	ctx_addr = table->gpu_addr;
1445	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(ctx_addr));
1446	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(ctx_addr));
1447
1448	/* 2, update vmid of descriptor */
1449	tmp = RREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID);
1450	tmp &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1451	/* use domain0 for MM scheduler */
1452	tmp |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1453	WREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID, tmp);
1454
1455	/* 3, notify mmsch about the size of this descriptor */
1456	size = header.total_size;
1457	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_SIZE, size);
1458
1459	/* 4, set resp to zero */
1460	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1461
1462	/* 5, kick off the initialization and wait until
1463	 * MMSCH_VF_MAILBOX_RESP becomes non-zero
1464	 */
1465	param = 0x10000001;
1466	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_HOST, param);
1467	tmp = 0;
1468	timeout = 1000;
1469	resp = 0;
1470	expected = param + 1;
1471	while (resp != expected) {
1472		resp = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1473		if (resp == expected)
1474			break;
1475
1476		udelay(10);
1477		tmp = tmp + 10;
1478		if (tmp >= timeout) {
1479			DRM_ERROR("failed to init MMSCH. TIME-OUT after %d usec"\
1480				" waiting for mmMMSCH_VF_MAILBOX_RESP "\
1481				"(expected=0x%08x, readback=0x%08x)\n",
1482				tmp, expected, resp);
1483			return -EBUSY;
1484		}
1485	}
1486
1487	return 0;
1488}
1489
1490static int vcn_v3_0_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1491{
1492	struct dpg_pause_state state = {.fw_based = VCN_DPG_STATE__UNPAUSE};
1493	uint32_t tmp;
1494
1495	vcn_v3_0_pause_dpg_mode(adev, inst_idx, &state);
1496
1497	/* Wait for power status to be 1 */
1498	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1499		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1500
1501	/* wait for read ptr to be equal to write ptr */
1502	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR);
1503	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1504
1505	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2);
1506	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
1507
1508	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1509	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF);
1510
1511	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1512		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1513
1514	/* disable dynamic power gating mode */
1515	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 0,
1516		~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1517
1518	return 0;
1519}
1520
1521static int vcn_v3_0_stop(struct amdgpu_device *adev)
1522{
1523	uint32_t tmp;
1524	int i, r = 0;
1525
1526	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1527		if (adev->vcn.harvest_config & (1 << i))
1528			continue;
1529
1530		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1531			r = vcn_v3_0_stop_dpg_mode(adev, i);
1532			continue;
1533		}
1534
1535		/* wait for vcn idle */
1536		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1537		if (r)
1538			return r;
1539
1540		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1541			UVD_LMI_STATUS__READ_CLEAN_MASK |
1542			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1543			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1544		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1545		if (r)
1546			return r;
1547
1548		/* disable LMI UMC channel */
1549		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2);
1550		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1551		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2, tmp);
1552		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1553			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1554		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1555		if (r)
1556			return r;
1557
1558		/* block VCPU register access */
1559		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL),
1560			UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1561			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1562
1563		/* reset VCPU */
1564		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1565			UVD_VCPU_CNTL__BLK_RST_MASK,
1566			~UVD_VCPU_CNTL__BLK_RST_MASK);
1567
1568		/* disable VCPU clock */
1569		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1570			~(UVD_VCPU_CNTL__CLK_EN_MASK));
1571
1572		/* apply soft reset */
1573		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1574		tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1575		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1576		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1577		tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1578		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1579
1580		/* clear status */
1581		WREG32_SOC15(VCN, i, mmUVD_STATUS, 0);
1582
1583		/* apply HW clock gating */
1584		vcn_v3_0_enable_clock_gating(adev, i);
1585
1586		/* enable VCN power gating */
1587		vcn_v3_0_enable_static_power_gating(adev, i);
1588	}
1589
1590	if (adev->pm.dpm_enabled)
1591		amdgpu_dpm_enable_uvd(adev, false);
1592
1593	return 0;
1594}
1595
1596static int vcn_v3_0_pause_dpg_mode(struct amdgpu_device *adev,
1597		   int inst_idx, struct dpg_pause_state *new_state)
1598{
1599	volatile struct amdgpu_fw_shared *fw_shared;
1600	struct amdgpu_ring *ring;
1601	uint32_t reg_data = 0;
1602	int ret_code;
1603
1604	/* pause/unpause if state is changed */
1605	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1606		DRM_DEBUG("dpg pause state changed %d -> %d",
1607			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1608		reg_data = RREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE) &
1609			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1610
1611		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1612			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1613				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1614
1615			if (!ret_code) {
1616				/* pause DPG */
1617				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1618				WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1619
1620				/* wait for ACK */
1621				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_DPG_PAUSE,
1622					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1623					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1624
1625				/* Stall DPG before WPTR/RPTR reset */
1626				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1627					UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1628					~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1629
1630				if (adev->ip_versions[UVD_HWIP][0] != IP_VERSION(3, 0, 33)) {
1631					/* Restore */
1632					fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
1633					fw_shared->multi_queue.encode_generalpurpose_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1634					ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1635					ring->wptr = 0;
1636					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
1637					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1638					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
1639					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1640					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1641					fw_shared->multi_queue.encode_generalpurpose_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1642
1643					fw_shared->multi_queue.encode_lowlatency_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1644					ring = &adev->vcn.inst[inst_idx].ring_enc[1];
1645					ring->wptr = 0;
1646					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1647					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1648					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
1649					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1650					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1651					fw_shared->multi_queue.encode_lowlatency_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1652
1653					/* restore wptr/rptr with pointers saved in FW shared memory*/
1654					WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, fw_shared->rb.rptr);
1655					WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR, fw_shared->rb.wptr);
1656				}
1657
1658				/* Unstall DPG */
1659				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1660					0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1661
1662				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS,
1663					UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1664			}
1665		} else {
1666			/* unpause dpg, no need to wait */
1667			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1668			WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1669		}
1670		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1671	}
1672
1673	return 0;
1674}
1675
1676/**
1677 * vcn_v3_0_dec_ring_get_rptr - get read pointer
1678 *
1679 * @ring: amdgpu_ring pointer
1680 *
1681 * Returns the current hardware read pointer
1682 */
1683static uint64_t vcn_v3_0_dec_ring_get_rptr(struct amdgpu_ring *ring)
1684{
1685	struct amdgpu_device *adev = ring->adev;
1686
1687	return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_RPTR);
1688}
1689
1690/**
1691 * vcn_v3_0_dec_ring_get_wptr - get write pointer
1692 *
1693 * @ring: amdgpu_ring pointer
1694 *
1695 * Returns the current hardware write pointer
1696 */
1697static uint64_t vcn_v3_0_dec_ring_get_wptr(struct amdgpu_ring *ring)
1698{
1699	struct amdgpu_device *adev = ring->adev;
1700
1701	if (ring->use_doorbell)
1702		return *ring->wptr_cpu_addr;
1703	else
1704		return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR);
1705}
1706
1707/**
1708 * vcn_v3_0_dec_ring_set_wptr - set write pointer
1709 *
1710 * @ring: amdgpu_ring pointer
1711 *
1712 * Commits the write pointer to the hardware
1713 */
1714static void vcn_v3_0_dec_ring_set_wptr(struct amdgpu_ring *ring)
1715{
1716	struct amdgpu_device *adev = ring->adev;
1717	volatile struct amdgpu_fw_shared *fw_shared;
1718
1719	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1720		/*whenever update RBC_RB_WPTR, we save the wptr in shared rb.wptr and scratch2 */
1721		fw_shared = adev->vcn.inst[ring->me].fw_shared.cpu_addr;
1722		fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1723		WREG32_SOC15(VCN, ring->me, mmUVD_SCRATCH2,
1724			lower_32_bits(ring->wptr));
1725	}
1726
1727	if (ring->use_doorbell) {
1728		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1729		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1730	} else {
1731		WREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1732	}
1733}
1734
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1735static const struct amdgpu_ring_funcs vcn_v3_0_dec_sw_ring_vm_funcs = {
1736	.type = AMDGPU_RING_TYPE_VCN_DEC,
1737	.align_mask = 0x3f,
1738	.nop = VCN_DEC_SW_CMD_NO_OP,
1739	.secure_submission_supported = true,
1740	.vmhub = AMDGPU_MMHUB_0,
1741	.get_rptr = vcn_v3_0_dec_ring_get_rptr,
1742	.get_wptr = vcn_v3_0_dec_ring_get_wptr,
1743	.set_wptr = vcn_v3_0_dec_ring_set_wptr,
1744	.emit_frame_size =
1745		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1746		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1747		VCN_SW_RING_EMIT_FRAME_SIZE,
1748	.emit_ib_size = 5, /* vcn_dec_sw_ring_emit_ib */
1749	.emit_ib = vcn_dec_sw_ring_emit_ib,
1750	.emit_fence = vcn_dec_sw_ring_emit_fence,
1751	.emit_vm_flush = vcn_dec_sw_ring_emit_vm_flush,
 
 
1752	.test_ring = amdgpu_vcn_dec_sw_ring_test_ring,
1753	.test_ib = NULL,//amdgpu_vcn_dec_sw_ring_test_ib,
1754	.insert_nop = amdgpu_ring_insert_nop,
1755	.insert_end = vcn_dec_sw_ring_insert_end,
1756	.pad_ib = amdgpu_ring_generic_pad_ib,
1757	.begin_use = amdgpu_vcn_ring_begin_use,
1758	.end_use = amdgpu_vcn_ring_end_use,
1759	.emit_wreg = vcn_dec_sw_ring_emit_wreg,
1760	.emit_reg_wait = vcn_dec_sw_ring_emit_reg_wait,
1761	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1762};
1763
1764static int vcn_v3_0_limit_sched(struct amdgpu_cs_parser *p,
1765				struct amdgpu_job *job)
1766{
1767	struct drm_gpu_scheduler **scheds;
1768
1769	/* The create msg must be in the first IB submitted */
1770	if (atomic_read(&job->base.entity->fence_seq))
1771		return -EINVAL;
1772
1773	scheds = p->adev->gpu_sched[AMDGPU_HW_IP_VCN_DEC]
1774		[AMDGPU_RING_PRIO_DEFAULT].sched;
1775	drm_sched_entity_modify_sched(job->base.entity, scheds, 1);
1776	return 0;
1777}
1778
1779static int vcn_v3_0_dec_msg(struct amdgpu_cs_parser *p, struct amdgpu_job *job,
1780			    uint64_t addr)
1781{
1782	struct ttm_operation_ctx ctx = { false, false };
1783	struct amdgpu_bo_va_mapping *map;
1784	uint32_t *msg, num_buffers;
1785	struct amdgpu_bo *bo;
1786	uint64_t start, end;
1787	unsigned int i;
1788	void * ptr;
1789	int r;
1790
1791	addr &= AMDGPU_GMC_HOLE_MASK;
1792	r = amdgpu_cs_find_mapping(p, addr, &bo, &map);
1793	if (r) {
1794		DRM_ERROR("Can't find BO for addr 0x%08Lx\n", addr);
1795		return r;
1796	}
1797
1798	start = map->start * AMDGPU_GPU_PAGE_SIZE;
1799	end = (map->last + 1) * AMDGPU_GPU_PAGE_SIZE;
1800	if (addr & 0x7) {
1801		DRM_ERROR("VCN messages must be 8 byte aligned!\n");
1802		return -EINVAL;
1803	}
1804
1805	bo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
1806	amdgpu_bo_placement_from_domain(bo, bo->allowed_domains);
1807	r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
1808	if (r) {
1809		DRM_ERROR("Failed validating the VCN message BO (%d)!\n", r);
1810		return r;
1811	}
1812
1813	r = amdgpu_bo_kmap(bo, &ptr);
1814	if (r) {
1815		DRM_ERROR("Failed mapping the VCN message (%d)!\n", r);
1816		return r;
1817	}
1818
1819	msg = ptr + addr - start;
1820
1821	/* Check length */
1822	if (msg[1] > end - addr) {
1823		r = -EINVAL;
1824		goto out;
1825	}
1826
1827	if (msg[3] != RDECODE_MSG_CREATE)
1828		goto out;
1829
1830	num_buffers = msg[2];
1831	for (i = 0, msg = &msg[6]; i < num_buffers; ++i, msg += 4) {
1832		uint32_t offset, size, *create;
1833
1834		if (msg[0] != RDECODE_MESSAGE_CREATE)
1835			continue;
1836
1837		offset = msg[1];
1838		size = msg[2];
1839
1840		if (offset + size > end) {
1841			r = -EINVAL;
1842			goto out;
1843		}
1844
1845		create = ptr + addr + offset - start;
1846
1847		/* H246, HEVC and VP9 can run on any instance */
1848		if (create[0] == 0x7 || create[0] == 0x10 || create[0] == 0x11)
1849			continue;
1850
1851		r = vcn_v3_0_limit_sched(p, job);
1852		if (r)
1853			goto out;
1854	}
1855
1856out:
1857	amdgpu_bo_kunmap(bo);
1858	return r;
1859}
1860
1861static int vcn_v3_0_ring_patch_cs_in_place(struct amdgpu_cs_parser *p,
1862					   struct amdgpu_job *job,
1863					   struct amdgpu_ib *ib)
1864{
1865	struct amdgpu_ring *ring = amdgpu_job_ring(job);
 
1866	uint32_t msg_lo = 0, msg_hi = 0;
1867	unsigned i;
1868	int r;
1869
1870	/* The first instance can decode anything */
1871	if (!ring->me)
1872		return 0;
1873
1874	for (i = 0; i < ib->length_dw; i += 2) {
1875		uint32_t reg = amdgpu_ib_get_value(ib, i);
1876		uint32_t val = amdgpu_ib_get_value(ib, i + 1);
1877
1878		if (reg == PACKET0(p->adev->vcn.internal.data0, 0)) {
1879			msg_lo = val;
1880		} else if (reg == PACKET0(p->adev->vcn.internal.data1, 0)) {
1881			msg_hi = val;
1882		} else if (reg == PACKET0(p->adev->vcn.internal.cmd, 0) &&
1883			   val == 0) {
1884			r = vcn_v3_0_dec_msg(p, job,
1885					     ((u64)msg_hi) << 32 | msg_lo);
1886			if (r)
1887				return r;
1888		}
1889	}
1890	return 0;
1891}
1892
1893static const struct amdgpu_ring_funcs vcn_v3_0_dec_ring_vm_funcs = {
1894	.type = AMDGPU_RING_TYPE_VCN_DEC,
1895	.align_mask = 0xf,
1896	.secure_submission_supported = true,
1897	.vmhub = AMDGPU_MMHUB_0,
1898	.get_rptr = vcn_v3_0_dec_ring_get_rptr,
1899	.get_wptr = vcn_v3_0_dec_ring_get_wptr,
1900	.set_wptr = vcn_v3_0_dec_ring_set_wptr,
1901	.patch_cs_in_place = vcn_v3_0_ring_patch_cs_in_place,
1902	.emit_frame_size =
1903		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1904		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1905		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1906		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1907		6,
1908	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1909	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
1910	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
1911	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1912	.test_ring = vcn_v2_0_dec_ring_test_ring,
1913	.test_ib = amdgpu_vcn_dec_ring_test_ib,
1914	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
1915	.insert_start = vcn_v2_0_dec_ring_insert_start,
1916	.insert_end = vcn_v2_0_dec_ring_insert_end,
1917	.pad_ib = amdgpu_ring_generic_pad_ib,
1918	.begin_use = amdgpu_vcn_ring_begin_use,
1919	.end_use = amdgpu_vcn_ring_end_use,
1920	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1921	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1922	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1923};
1924
1925/**
1926 * vcn_v3_0_enc_ring_get_rptr - get enc read pointer
1927 *
1928 * @ring: amdgpu_ring pointer
1929 *
1930 * Returns the current hardware enc read pointer
1931 */
1932static uint64_t vcn_v3_0_enc_ring_get_rptr(struct amdgpu_ring *ring)
1933{
1934	struct amdgpu_device *adev = ring->adev;
1935
1936	if (ring == &adev->vcn.inst[ring->me].ring_enc[0])
1937		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR);
1938	else
1939		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR2);
1940}
1941
1942/**
1943 * vcn_v3_0_enc_ring_get_wptr - get enc write pointer
1944 *
1945 * @ring: amdgpu_ring pointer
1946 *
1947 * Returns the current hardware enc write pointer
1948 */
1949static uint64_t vcn_v3_0_enc_ring_get_wptr(struct amdgpu_ring *ring)
1950{
1951	struct amdgpu_device *adev = ring->adev;
1952
1953	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1954		if (ring->use_doorbell)
1955			return *ring->wptr_cpu_addr;
1956		else
1957			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR);
1958	} else {
1959		if (ring->use_doorbell)
1960			return *ring->wptr_cpu_addr;
1961		else
1962			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2);
1963	}
1964}
1965
1966/**
1967 * vcn_v3_0_enc_ring_set_wptr - set enc write pointer
1968 *
1969 * @ring: amdgpu_ring pointer
1970 *
1971 * Commits the enc write pointer to the hardware
1972 */
1973static void vcn_v3_0_enc_ring_set_wptr(struct amdgpu_ring *ring)
1974{
1975	struct amdgpu_device *adev = ring->adev;
1976
1977	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1978		if (ring->use_doorbell) {
1979			*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1980			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1981		} else {
1982			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1983		}
1984	} else {
1985		if (ring->use_doorbell) {
1986			*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1987			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1988		} else {
1989			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1990		}
1991	}
1992}
1993
1994static const struct amdgpu_ring_funcs vcn_v3_0_enc_ring_vm_funcs = {
1995	.type = AMDGPU_RING_TYPE_VCN_ENC,
1996	.align_mask = 0x3f,
1997	.nop = VCN_ENC_CMD_NO_OP,
1998	.vmhub = AMDGPU_MMHUB_0,
1999	.get_rptr = vcn_v3_0_enc_ring_get_rptr,
2000	.get_wptr = vcn_v3_0_enc_ring_get_wptr,
2001	.set_wptr = vcn_v3_0_enc_ring_set_wptr,
2002	.emit_frame_size =
2003		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2004		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
2005		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
2006		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
2007		1, /* vcn_v2_0_enc_ring_insert_end */
2008	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
2009	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
2010	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
2011	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
2012	.test_ring = amdgpu_vcn_enc_ring_test_ring,
2013	.test_ib = amdgpu_vcn_enc_ring_test_ib,
2014	.insert_nop = amdgpu_ring_insert_nop,
2015	.insert_end = vcn_v2_0_enc_ring_insert_end,
2016	.pad_ib = amdgpu_ring_generic_pad_ib,
2017	.begin_use = amdgpu_vcn_ring_begin_use,
2018	.end_use = amdgpu_vcn_ring_end_use,
2019	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
2020	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
2021	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2022};
2023
2024static void vcn_v3_0_set_dec_ring_funcs(struct amdgpu_device *adev)
2025{
2026	int i;
2027
2028	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2029		if (adev->vcn.harvest_config & (1 << i))
2030			continue;
2031
2032		if (!DEC_SW_RING_ENABLED)
2033			adev->vcn.inst[i].ring_dec.funcs = &vcn_v3_0_dec_ring_vm_funcs;
2034		else
2035			adev->vcn.inst[i].ring_dec.funcs = &vcn_v3_0_dec_sw_ring_vm_funcs;
2036		adev->vcn.inst[i].ring_dec.me = i;
2037		DRM_INFO("VCN(%d) decode%s is enabled in VM mode\n", i,
2038			  DEC_SW_RING_ENABLED?"(Software Ring)":"");
2039	}
2040}
2041
2042static void vcn_v3_0_set_enc_ring_funcs(struct amdgpu_device *adev)
2043{
2044	int i, j;
2045
2046	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2047		if (adev->vcn.harvest_config & (1 << i))
2048			continue;
2049
2050		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
2051			adev->vcn.inst[i].ring_enc[j].funcs = &vcn_v3_0_enc_ring_vm_funcs;
2052			adev->vcn.inst[i].ring_enc[j].me = i;
2053		}
2054		if (adev->vcn.num_enc_rings > 0)
2055			DRM_INFO("VCN(%d) encode is enabled in VM mode\n", i);
2056	}
2057}
2058
2059static bool vcn_v3_0_is_idle(void *handle)
2060{
2061	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2062	int i, ret = 1;
2063
2064	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2065		if (adev->vcn.harvest_config & (1 << i))
2066			continue;
2067
2068		ret &= (RREG32_SOC15(VCN, i, mmUVD_STATUS) == UVD_STATUS__IDLE);
2069	}
2070
2071	return ret;
2072}
2073
2074static int vcn_v3_0_wait_for_idle(void *handle)
2075{
2076	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2077	int i, ret = 0;
2078
2079	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2080		if (adev->vcn.harvest_config & (1 << i))
2081			continue;
2082
2083		ret = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE,
2084			UVD_STATUS__IDLE);
2085		if (ret)
2086			return ret;
2087	}
2088
2089	return ret;
2090}
2091
2092static int vcn_v3_0_set_clockgating_state(void *handle,
2093					  enum amd_clockgating_state state)
2094{
2095	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2096	bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
2097	int i;
2098
2099	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2100		if (adev->vcn.harvest_config & (1 << i))
2101			continue;
2102
2103		if (enable) {
2104			if (RREG32_SOC15(VCN, i, mmUVD_STATUS) != UVD_STATUS__IDLE)
2105				return -EBUSY;
2106			vcn_v3_0_enable_clock_gating(adev, i);
2107		} else {
2108			vcn_v3_0_disable_clock_gating(adev, i);
2109		}
2110	}
2111
2112	return 0;
2113}
2114
2115static int vcn_v3_0_set_powergating_state(void *handle,
2116					  enum amd_powergating_state state)
2117{
2118	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2119	int ret;
2120
2121	/* for SRIOV, guest should not control VCN Power-gating
2122	 * MMSCH FW should control Power-gating and clock-gating
2123	 * guest should avoid touching CGC and PG
2124	 */
2125	if (amdgpu_sriov_vf(adev)) {
2126		adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
2127		return 0;
2128	}
2129
2130	if(state == adev->vcn.cur_state)
2131		return 0;
2132
2133	if (state == AMD_PG_STATE_GATE)
2134		ret = vcn_v3_0_stop(adev);
2135	else
2136		ret = vcn_v3_0_start(adev);
2137
2138	if(!ret)
2139		adev->vcn.cur_state = state;
2140
2141	return ret;
2142}
2143
2144static int vcn_v3_0_set_interrupt_state(struct amdgpu_device *adev,
2145					struct amdgpu_irq_src *source,
2146					unsigned type,
2147					enum amdgpu_interrupt_state state)
2148{
2149	return 0;
2150}
2151
2152static int vcn_v3_0_process_interrupt(struct amdgpu_device *adev,
2153				      struct amdgpu_irq_src *source,
2154				      struct amdgpu_iv_entry *entry)
2155{
2156	uint32_t ip_instance;
2157
2158	switch (entry->client_id) {
2159	case SOC15_IH_CLIENTID_VCN:
2160		ip_instance = 0;
2161		break;
2162	case SOC15_IH_CLIENTID_VCN1:
2163		ip_instance = 1;
2164		break;
2165	default:
2166		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
2167		return 0;
2168	}
2169
2170	DRM_DEBUG("IH: VCN TRAP\n");
2171
2172	switch (entry->src_id) {
2173	case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
2174		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec);
2175		break;
2176	case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
2177		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
2178		break;
2179	case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
2180		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]);
2181		break;
2182	default:
2183		DRM_ERROR("Unhandled interrupt: %d %d\n",
2184			  entry->src_id, entry->src_data[0]);
2185		break;
2186	}
2187
2188	return 0;
2189}
2190
2191static const struct amdgpu_irq_src_funcs vcn_v3_0_irq_funcs = {
2192	.set = vcn_v3_0_set_interrupt_state,
2193	.process = vcn_v3_0_process_interrupt,
2194};
2195
2196static void vcn_v3_0_set_irq_funcs(struct amdgpu_device *adev)
2197{
2198	int i;
2199
2200	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2201		if (adev->vcn.harvest_config & (1 << i))
2202			continue;
2203
2204		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
2205		adev->vcn.inst[i].irq.funcs = &vcn_v3_0_irq_funcs;
2206	}
2207}
2208
2209static const struct amd_ip_funcs vcn_v3_0_ip_funcs = {
2210	.name = "vcn_v3_0",
2211	.early_init = vcn_v3_0_early_init,
2212	.late_init = NULL,
2213	.sw_init = vcn_v3_0_sw_init,
2214	.sw_fini = vcn_v3_0_sw_fini,
2215	.hw_init = vcn_v3_0_hw_init,
2216	.hw_fini = vcn_v3_0_hw_fini,
2217	.suspend = vcn_v3_0_suspend,
2218	.resume = vcn_v3_0_resume,
2219	.is_idle = vcn_v3_0_is_idle,
2220	.wait_for_idle = vcn_v3_0_wait_for_idle,
2221	.check_soft_reset = NULL,
2222	.pre_soft_reset = NULL,
2223	.soft_reset = NULL,
2224	.post_soft_reset = NULL,
2225	.set_clockgating_state = vcn_v3_0_set_clockgating_state,
2226	.set_powergating_state = vcn_v3_0_set_powergating_state,
2227};
2228
2229const struct amdgpu_ip_block_version vcn_v3_0_ip_block =
2230{
2231	.type = AMD_IP_BLOCK_TYPE_VCN,
2232	.major = 3,
2233	.minor = 0,
2234	.rev = 0,
2235	.funcs = &vcn_v3_0_ip_funcs,
2236};
v5.14.15
   1/*
   2 * Copyright 2019 Advanced Micro Devices, Inc.
   3 *
   4 * Permission is hereby granted, free of charge, to any person obtaining a
   5 * copy of this software and associated documentation files (the "Software"),
   6 * to deal in the Software without restriction, including without limitation
   7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8 * and/or sell copies of the Software, and to permit persons to whom the
   9 * Software is furnished to do so, subject to the following conditions:
  10 *
  11 * The above copyright notice and this permission notice shall be included in
  12 * all copies or substantial portions of the Software.
  13 *
  14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
  18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  20 * OTHER DEALINGS IN THE SOFTWARE.
  21 *
  22 */
  23
  24#include <linux/firmware.h>
  25#include "amdgpu.h"
  26#include "amdgpu_vcn.h"
  27#include "amdgpu_pm.h"
 
  28#include "soc15.h"
  29#include "soc15d.h"
  30#include "vcn_v2_0.h"
  31#include "mmsch_v3_0.h"
 
  32
  33#include "vcn/vcn_3_0_0_offset.h"
  34#include "vcn/vcn_3_0_0_sh_mask.h"
  35#include "ivsrcid/vcn/irqsrcs_vcn_2_0.h"
  36
  37#include <drm/drm_drv.h>
  38
 
 
 
  39#define mmUVD_CONTEXT_ID_INTERNAL_OFFSET			0x27
  40#define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET			0x0f
  41#define mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET			0x10
  42#define mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET			0x11
  43#define mmUVD_NO_OP_INTERNAL_OFFSET				0x29
  44#define mmUVD_GP_SCRATCH8_INTERNAL_OFFSET			0x66
  45#define mmUVD_SCRATCH9_INTERNAL_OFFSET				0xc01d
  46
  47#define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET			0x431
  48#define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET		0x3b4
  49#define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET		0x3b5
  50#define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET			0x25c
  51
  52#define VCN_INSTANCES_SIENNA_CICHLID				2
  53#define DEC_SW_RING_ENABLED					FALSE
  54
  55#define RDECODE_MSG_CREATE					0x00000000
  56#define RDECODE_MESSAGE_CREATE					0x00000001
  57
  58static int amdgpu_ih_clientid_vcns[] = {
  59	SOC15_IH_CLIENTID_VCN,
  60	SOC15_IH_CLIENTID_VCN1
  61};
  62
  63static int amdgpu_ucode_id_vcns[] = {
  64	AMDGPU_UCODE_ID_VCN,
  65	AMDGPU_UCODE_ID_VCN1
  66};
  67
  68static int vcn_v3_0_start_sriov(struct amdgpu_device *adev);
  69static void vcn_v3_0_set_dec_ring_funcs(struct amdgpu_device *adev);
  70static void vcn_v3_0_set_enc_ring_funcs(struct amdgpu_device *adev);
  71static void vcn_v3_0_set_irq_funcs(struct amdgpu_device *adev);
  72static int vcn_v3_0_set_powergating_state(void *handle,
  73			enum amd_powergating_state state);
  74static int vcn_v3_0_pause_dpg_mode(struct amdgpu_device *adev,
  75			int inst_idx, struct dpg_pause_state *new_state);
  76
  77static void vcn_v3_0_dec_ring_set_wptr(struct amdgpu_ring *ring);
  78static void vcn_v3_0_enc_ring_set_wptr(struct amdgpu_ring *ring);
  79
  80/**
  81 * vcn_v3_0_early_init - set function pointers
  82 *
  83 * @handle: amdgpu_device pointer
  84 *
  85 * Set ring and irq function pointers
  86 */
  87static int vcn_v3_0_early_init(void *handle)
  88{
  89	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
  90	int i;
  91
  92	if (amdgpu_sriov_vf(adev)) {
  93		for (i = 0; i < VCN_INSTANCES_SIENNA_CICHLID; i++)
  94			if (amdgpu_vcn_is_disabled_vcn(adev, VCN_DECODE_RING, i))
  95				adev->vcn.num_vcn_inst++;
  96		adev->vcn.harvest_config = 0;
  97		adev->vcn.num_enc_rings = 1;
  98
  99	} else {
 100		if (adev->asic_type == CHIP_SIENNA_CICHLID) {
 101			u32 harvest;
 102
 103			adev->vcn.num_vcn_inst = VCN_INSTANCES_SIENNA_CICHLID;
 104			for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
 105				harvest = RREG32_SOC15(VCN, i, mmCC_UVD_HARVESTING);
 106				if (harvest & CC_UVD_HARVESTING__UVD_DISABLE_MASK)
 107					adev->vcn.harvest_config |= 1 << i;
 108			}
 109
 110			if (adev->vcn.harvest_config == (AMDGPU_VCN_HARVEST_VCN0 |
 111						AMDGPU_VCN_HARVEST_VCN1))
 112				/* both instances are harvested, disable the block */
 113				return -ENOENT;
 114		} else
 115			adev->vcn.num_vcn_inst = 1;
 116
 117		if (adev->asic_type == CHIP_BEIGE_GOBY)
 118			adev->vcn.num_enc_rings = 0;
 119		else
 120			adev->vcn.num_enc_rings = 2;
 121	}
 122
 123	vcn_v3_0_set_dec_ring_funcs(adev);
 124	vcn_v3_0_set_enc_ring_funcs(adev);
 125	vcn_v3_0_set_irq_funcs(adev);
 126
 127	return 0;
 128}
 129
 130/**
 131 * vcn_v3_0_sw_init - sw init for VCN block
 132 *
 133 * @handle: amdgpu_device pointer
 134 *
 135 * Load firmware and sw initialization
 136 */
 137static int vcn_v3_0_sw_init(void *handle)
 138{
 139	struct amdgpu_ring *ring;
 140	int i, j, r;
 141	int vcn_doorbell_index = 0;
 142	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 143
 144	r = amdgpu_vcn_sw_init(adev);
 145	if (r)
 146		return r;
 147
 148	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
 149		const struct common_firmware_header *hdr;
 150		hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
 151		adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].ucode_id = AMDGPU_UCODE_ID_VCN;
 152		adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].fw = adev->vcn.fw;
 153		adev->firmware.fw_size +=
 154			ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
 155
 156		if ((adev->vcn.num_vcn_inst == VCN_INSTANCES_SIENNA_CICHLID) ||
 157		    (amdgpu_sriov_vf(adev) && adev->asic_type == CHIP_SIENNA_CICHLID)) {
 158			adev->firmware.ucode[AMDGPU_UCODE_ID_VCN1].ucode_id = AMDGPU_UCODE_ID_VCN1;
 159			adev->firmware.ucode[AMDGPU_UCODE_ID_VCN1].fw = adev->vcn.fw;
 160			adev->firmware.fw_size +=
 161				ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
 162		}
 163		dev_info(adev->dev, "Will use PSP to load VCN firmware\n");
 164	}
 165
 166	r = amdgpu_vcn_resume(adev);
 167	if (r)
 168		return r;
 169
 170	/*
 171	 * Note: doorbell assignment is fixed for SRIOV multiple VCN engines
 172	 * Formula:
 173	 *   vcn_db_base  = adev->doorbell_index.vcn.vcn_ring0_1 << 1;
 174	 *   dec_ring_i   = vcn_db_base + i * (adev->vcn.num_enc_rings + 1)
 175	 *   enc_ring_i,j = vcn_db_base + i * (adev->vcn.num_enc_rings + 1) + 1 + j
 176	 */
 177	if (amdgpu_sriov_vf(adev)) {
 178		vcn_doorbell_index = adev->doorbell_index.vcn.vcn_ring0_1;
 179		/* get DWORD offset */
 180		vcn_doorbell_index = vcn_doorbell_index << 1;
 181	}
 182
 183	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
 184		volatile struct amdgpu_fw_shared *fw_shared;
 185
 186		if (adev->vcn.harvest_config & (1 << i))
 187			continue;
 188
 189		adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
 190		adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
 191		adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
 192		adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
 193		adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
 194		adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
 195
 196		adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
 197		adev->vcn.inst[i].external.scratch9 = SOC15_REG_OFFSET(VCN, i, mmUVD_SCRATCH9);
 198		adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
 199		adev->vcn.inst[i].external.data0 = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_DATA0);
 200		adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
 201		adev->vcn.inst[i].external.data1 = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_DATA1);
 202		adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
 203		adev->vcn.inst[i].external.cmd = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_CMD);
 204		adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
 205		adev->vcn.inst[i].external.nop = SOC15_REG_OFFSET(VCN, i, mmUVD_NO_OP);
 206
 207		/* VCN DEC TRAP */
 208		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
 209				VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst[i].irq);
 210		if (r)
 211			return r;
 212
 213		atomic_set(&adev->vcn.inst[i].sched_score, 0);
 214
 215		ring = &adev->vcn.inst[i].ring_dec;
 216		ring->use_doorbell = true;
 217		if (amdgpu_sriov_vf(adev)) {
 218			ring->doorbell_index = vcn_doorbell_index + i * (adev->vcn.num_enc_rings + 1);
 219		} else {
 220			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i;
 221		}
 222		sprintf(ring->name, "vcn_dec_%d", i);
 223		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
 224				     AMDGPU_RING_PRIO_DEFAULT,
 225				     &adev->vcn.inst[i].sched_score);
 226		if (r)
 227			return r;
 228
 229		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
 
 
 230			/* VCN ENC TRAP */
 231			r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
 232				j + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);
 233			if (r)
 234				return r;
 235
 236			ring = &adev->vcn.inst[i].ring_enc[j];
 237			ring->use_doorbell = true;
 238			if (amdgpu_sriov_vf(adev)) {
 239				ring->doorbell_index = vcn_doorbell_index + i * (adev->vcn.num_enc_rings + 1) + 1 + j;
 240			} else {
 241				ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + j + 8 * i;
 242			}
 243			sprintf(ring->name, "vcn_enc_%d.%d", i, j);
 244			r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
 245					     AMDGPU_RING_PRIO_DEFAULT,
 246					     &adev->vcn.inst[i].sched_score);
 247			if (r)
 248				return r;
 249		}
 250
 251		fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
 252		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_SW_RING_FLAG) |
 253					     cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG) |
 254					     cpu_to_le32(AMDGPU_VCN_FW_SHARED_FLAG_0_RB);
 255		fw_shared->sw_ring.is_enabled = cpu_to_le32(DEC_SW_RING_ENABLED);
 
 
 
 
 
 
 
 
 256	}
 257
 258	if (amdgpu_sriov_vf(adev)) {
 259		r = amdgpu_virt_alloc_mm_table(adev);
 260		if (r)
 261			return r;
 262	}
 263	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
 264		adev->vcn.pause_dpg_mode = vcn_v3_0_pause_dpg_mode;
 265
 266	return 0;
 267}
 268
 269/**
 270 * vcn_v3_0_sw_fini - sw fini for VCN block
 271 *
 272 * @handle: amdgpu_device pointer
 273 *
 274 * VCN suspend and free up sw allocation
 275 */
 276static int vcn_v3_0_sw_fini(void *handle)
 277{
 278	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 279	int i, r, idx;
 280
 281	if (drm_dev_enter(&adev->ddev, &idx)) {
 282		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
 283			volatile struct amdgpu_fw_shared *fw_shared;
 284
 285			if (adev->vcn.harvest_config & (1 << i))
 286				continue;
 287			fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
 288			fw_shared->present_flag_0 = 0;
 289			fw_shared->sw_ring.is_enabled = false;
 290		}
 291
 292		drm_dev_exit(idx);
 293	}
 294
 295	if (amdgpu_sriov_vf(adev))
 296		amdgpu_virt_free_mm_table(adev);
 297
 298	r = amdgpu_vcn_suspend(adev);
 299	if (r)
 300		return r;
 301
 302	r = amdgpu_vcn_sw_fini(adev);
 303
 304	return r;
 305}
 306
 307/**
 308 * vcn_v3_0_hw_init - start and test VCN block
 309 *
 310 * @handle: amdgpu_device pointer
 311 *
 312 * Initialize the hardware, boot up the VCPU and do some testing
 313 */
 314static int vcn_v3_0_hw_init(void *handle)
 315{
 316	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 317	struct amdgpu_ring *ring;
 318	int i, j, r;
 319
 320	if (amdgpu_sriov_vf(adev)) {
 321		r = vcn_v3_0_start_sriov(adev);
 322		if (r)
 323			goto done;
 324
 325		/* initialize VCN dec and enc ring buffers */
 326		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
 327			if (adev->vcn.harvest_config & (1 << i))
 328				continue;
 329
 330			ring = &adev->vcn.inst[i].ring_dec;
 331			ring->wptr = 0;
 332			ring->wptr_old = 0;
 333			vcn_v3_0_dec_ring_set_wptr(ring);
 334			ring->sched.ready = true;
 
 
 
 
 
 
 335
 336			for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
 337				ring = &adev->vcn.inst[i].ring_enc[j];
 338				ring->wptr = 0;
 339				ring->wptr_old = 0;
 340				vcn_v3_0_enc_ring_set_wptr(ring);
 341				ring->sched.ready = true;
 
 
 
 
 
 
 342			}
 343		}
 344	} else {
 345		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
 346			if (adev->vcn.harvest_config & (1 << i))
 347				continue;
 348
 349			ring = &adev->vcn.inst[i].ring_dec;
 350
 351			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
 352						     ring->doorbell_index, i);
 353
 354			r = amdgpu_ring_test_helper(ring);
 355			if (r)
 356				goto done;
 357
 358			for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
 359				ring = &adev->vcn.inst[i].ring_enc[j];
 360				r = amdgpu_ring_test_helper(ring);
 361				if (r)
 362					goto done;
 363			}
 364		}
 365	}
 366
 367done:
 368	if (!r)
 369		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
 370			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
 371
 372	return r;
 373}
 374
 375/**
 376 * vcn_v3_0_hw_fini - stop the hardware block
 377 *
 378 * @handle: amdgpu_device pointer
 379 *
 380 * Stop the VCN block, mark ring as not ready any more
 381 */
 382static int vcn_v3_0_hw_fini(void *handle)
 383{
 384	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 385	int i;
 386
 387	cancel_delayed_work_sync(&adev->vcn.idle_work);
 388
 389	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
 390		if (adev->vcn.harvest_config & (1 << i))
 391			continue;
 392
 393		if (!amdgpu_sriov_vf(adev)) {
 394			if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
 395					(adev->vcn.cur_state != AMD_PG_STATE_GATE &&
 396					 RREG32_SOC15(VCN, i, mmUVD_STATUS))) {
 397				vcn_v3_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
 398			}
 399		}
 400	}
 401
 402	return 0;
 403}
 404
 405/**
 406 * vcn_v3_0_suspend - suspend VCN block
 407 *
 408 * @handle: amdgpu_device pointer
 409 *
 410 * HW fini and suspend VCN block
 411 */
 412static int vcn_v3_0_suspend(void *handle)
 413{
 414	int r;
 415	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 416
 417	r = vcn_v3_0_hw_fini(adev);
 418	if (r)
 419		return r;
 420
 421	r = amdgpu_vcn_suspend(adev);
 422
 423	return r;
 424}
 425
 426/**
 427 * vcn_v3_0_resume - resume VCN block
 428 *
 429 * @handle: amdgpu_device pointer
 430 *
 431 * Resume firmware and hw init VCN block
 432 */
 433static int vcn_v3_0_resume(void *handle)
 434{
 435	int r;
 436	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 437
 438	r = amdgpu_vcn_resume(adev);
 439	if (r)
 440		return r;
 441
 442	r = vcn_v3_0_hw_init(adev);
 443
 444	return r;
 445}
 446
 447/**
 448 * vcn_v3_0_mc_resume - memory controller programming
 449 *
 450 * @adev: amdgpu_device pointer
 451 * @inst: instance number
 452 *
 453 * Let the VCN memory controller know it's offsets
 454 */
 455static void vcn_v3_0_mc_resume(struct amdgpu_device *adev, int inst)
 456{
 457	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
 458	uint32_t offset;
 459
 460	/* cache window 0: fw */
 461	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
 462		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
 463			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo));
 464		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
 465			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi));
 466		WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET0, 0);
 467		offset = 0;
 468	} else {
 469		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
 470			lower_32_bits(adev->vcn.inst[inst].gpu_addr));
 471		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
 472			upper_32_bits(adev->vcn.inst[inst].gpu_addr));
 473		offset = size;
 474		WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET0,
 475			AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
 476	}
 477	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE0, size);
 478
 479	/* cache window 1: stack */
 480	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
 481		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
 482	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
 483		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
 484	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET1, 0);
 485	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
 486
 487	/* cache window 2: context */
 488	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
 489		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
 490	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
 491		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
 492	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET2, 0);
 493	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
 494
 495	/* non-cache window */
 496	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
 497		lower_32_bits(adev->vcn.inst[inst].fw_shared_gpu_addr));
 498	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
 499		upper_32_bits(adev->vcn.inst[inst].fw_shared_gpu_addr));
 500	WREG32_SOC15(VCN, inst, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
 501	WREG32_SOC15(VCN, inst, mmUVD_VCPU_NONCACHE_SIZE0,
 502		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
 503}
 504
 505static void vcn_v3_0_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
 506{
 507	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
 508	uint32_t offset;
 509
 510	/* cache window 0: fw */
 511	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
 512		if (!indirect) {
 513			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 514				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
 515				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
 516			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 517				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
 518				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
 519			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 520				VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
 521		} else {
 522			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 523				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
 524			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 525				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
 526			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 527				VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
 528		}
 529		offset = 0;
 530	} else {
 531		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 532			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
 533			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
 534		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 535			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
 536			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
 537		offset = size;
 538		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 539			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0),
 540			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
 541	}
 542
 543	if (!indirect)
 544		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 545			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
 546	else
 547		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 548			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
 549
 550	/* cache window 1: stack */
 551	if (!indirect) {
 552		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 553			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
 554			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
 555		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 556			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
 557			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
 558		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 559			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
 560	} else {
 561		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 562			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
 563		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 564			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
 565		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 566			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
 567	}
 568	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 569			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
 570
 571	/* cache window 2: context */
 572	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 573			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
 574			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
 575	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 576			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
 577			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
 578	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 579			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
 580	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 581			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
 582
 583	/* non-cache window */
 584	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 585			VCN, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
 586			lower_32_bits(adev->vcn.inst[inst_idx].fw_shared_gpu_addr), 0, indirect);
 587	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 588			VCN, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
 589			upper_32_bits(adev->vcn.inst[inst_idx].fw_shared_gpu_addr), 0, indirect);
 590	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 591			VCN, inst_idx, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
 592	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 593			VCN, inst_idx, mmUVD_VCPU_NONCACHE_SIZE0),
 594			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
 595
 596	/* VCN global tiling registers */
 597	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
 598		UVD, 0, mmUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
 599}
 600
 601static void vcn_v3_0_disable_static_power_gating(struct amdgpu_device *adev, int inst)
 602{
 603	uint32_t data = 0;
 604
 605	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
 606		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
 607			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
 608			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
 609			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
 610			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
 611			| 2 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
 612			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
 613			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
 614			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
 615			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
 616			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
 617			| 2 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
 618			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
 619			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
 620
 621		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
 622		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS,
 623			UVD_PGFSM_STATUS__UVDM_UVDU_UVDLM_PWR_ON_3_0, 0x3F3FFFFF);
 624	} else {
 625		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
 626			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
 627			| 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
 628			| 1 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
 629			| 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
 630			| 1 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
 631			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
 632			| 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
 633			| 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
 634			| 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
 635			| 1 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
 636			| 1 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
 637			| 1 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
 638			| 1 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
 639		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
 640		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS, 0,  0x3F3FFFFF);
 641	}
 642
 643	data = RREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS);
 644	data &= ~0x103;
 645	if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
 646		data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
 647			UVD_POWER_STATUS__UVD_PG_EN_MASK;
 648
 649	WREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS, data);
 650}
 651
 652static void vcn_v3_0_enable_static_power_gating(struct amdgpu_device *adev, int inst)
 653{
 654	uint32_t data;
 655
 656	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
 657		/* Before power off, this indicator has to be turned on */
 658		data = RREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS);
 659		data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
 660		data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
 661		WREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS, data);
 662
 663		data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
 664			| 2 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
 665			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
 666			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
 667			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
 668			| 2 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
 669			| 2 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
 670			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
 671			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
 672			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
 673			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
 674			| 2 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
 675			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
 676			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
 677		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
 678
 679		data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
 680			| 2 << UVD_PGFSM_STATUS__UVDU_PWR_STATUS__SHIFT
 681			| 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
 682			| 2 << UVD_PGFSM_STATUS__UVDC_PWR_STATUS__SHIFT
 683			| 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
 684			| 2 << UVD_PGFSM_STATUS__UVDIRL_PWR_STATUS__SHIFT
 685			| 2 << UVD_PGFSM_STATUS__UVDLM_PWR_STATUS__SHIFT
 686			| 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
 687			| 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
 688			| 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT
 689			| 2 << UVD_PGFSM_STATUS__UVDAB_PWR_STATUS__SHIFT
 690			| 2 << UVD_PGFSM_STATUS__UVDATD_PWR_STATUS__SHIFT
 691			| 2 << UVD_PGFSM_STATUS__UVDNA_PWR_STATUS__SHIFT
 692			| 2 << UVD_PGFSM_STATUS__UVDNB_PWR_STATUS__SHIFT);
 693		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS, data, 0x3F3FFFFF);
 694	}
 695}
 696
 697/**
 698 * vcn_v3_0_disable_clock_gating - disable VCN clock gating
 699 *
 700 * @adev: amdgpu_device pointer
 701 * @inst: instance number
 702 *
 703 * Disable clock gating for VCN block
 704 */
 705static void vcn_v3_0_disable_clock_gating(struct amdgpu_device *adev, int inst)
 706{
 707	uint32_t data;
 708
 709	/* VCN disable CGC */
 710	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
 711	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
 712		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
 713	else
 714		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
 715	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
 716	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
 717	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
 718
 719	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_GATE);
 720	data &= ~(UVD_CGC_GATE__SYS_MASK
 721		| UVD_CGC_GATE__UDEC_MASK
 722		| UVD_CGC_GATE__MPEG2_MASK
 723		| UVD_CGC_GATE__REGS_MASK
 724		| UVD_CGC_GATE__RBC_MASK
 725		| UVD_CGC_GATE__LMI_MC_MASK
 726		| UVD_CGC_GATE__LMI_UMC_MASK
 727		| UVD_CGC_GATE__IDCT_MASK
 728		| UVD_CGC_GATE__MPRD_MASK
 729		| UVD_CGC_GATE__MPC_MASK
 730		| UVD_CGC_GATE__LBSI_MASK
 731		| UVD_CGC_GATE__LRBBM_MASK
 732		| UVD_CGC_GATE__UDEC_RE_MASK
 733		| UVD_CGC_GATE__UDEC_CM_MASK
 734		| UVD_CGC_GATE__UDEC_IT_MASK
 735		| UVD_CGC_GATE__UDEC_DB_MASK
 736		| UVD_CGC_GATE__UDEC_MP_MASK
 737		| UVD_CGC_GATE__WCB_MASK
 738		| UVD_CGC_GATE__VCPU_MASK
 739		| UVD_CGC_GATE__MMSCH_MASK);
 740
 741	WREG32_SOC15(VCN, inst, mmUVD_CGC_GATE, data);
 742
 743	SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_CGC_GATE, 0,  0xFFFFFFFF);
 744
 745	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
 746	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
 747		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
 748		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
 749		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
 750		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
 751		| UVD_CGC_CTRL__SYS_MODE_MASK
 752		| UVD_CGC_CTRL__UDEC_MODE_MASK
 753		| UVD_CGC_CTRL__MPEG2_MODE_MASK
 754		| UVD_CGC_CTRL__REGS_MODE_MASK
 755		| UVD_CGC_CTRL__RBC_MODE_MASK
 756		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
 757		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
 758		| UVD_CGC_CTRL__IDCT_MODE_MASK
 759		| UVD_CGC_CTRL__MPRD_MODE_MASK
 760		| UVD_CGC_CTRL__MPC_MODE_MASK
 761		| UVD_CGC_CTRL__LBSI_MODE_MASK
 762		| UVD_CGC_CTRL__LRBBM_MODE_MASK
 763		| UVD_CGC_CTRL__WCB_MODE_MASK
 764		| UVD_CGC_CTRL__VCPU_MODE_MASK
 765		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
 766	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
 767
 768	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE);
 769	data |= (UVD_SUVD_CGC_GATE__SRE_MASK
 770		| UVD_SUVD_CGC_GATE__SIT_MASK
 771		| UVD_SUVD_CGC_GATE__SMP_MASK
 772		| UVD_SUVD_CGC_GATE__SCM_MASK
 773		| UVD_SUVD_CGC_GATE__SDB_MASK
 774		| UVD_SUVD_CGC_GATE__SRE_H264_MASK
 775		| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
 776		| UVD_SUVD_CGC_GATE__SIT_H264_MASK
 777		| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
 778		| UVD_SUVD_CGC_GATE__SCM_H264_MASK
 779		| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
 780		| UVD_SUVD_CGC_GATE__SDB_H264_MASK
 781		| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
 782		| UVD_SUVD_CGC_GATE__SCLR_MASK
 783		| UVD_SUVD_CGC_GATE__ENT_MASK
 784		| UVD_SUVD_CGC_GATE__IME_MASK
 785		| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
 786		| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
 787		| UVD_SUVD_CGC_GATE__SITE_MASK
 788		| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
 789		| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
 790		| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
 791		| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
 792		| UVD_SUVD_CGC_GATE__IME_HEVC_MASK
 793		| UVD_SUVD_CGC_GATE__EFC_MASK
 794		| UVD_SUVD_CGC_GATE__SAOE_MASK
 795		| UVD_SUVD_CGC_GATE__SRE_AV1_MASK
 796		| UVD_SUVD_CGC_GATE__FBC_PCLK_MASK
 797		| UVD_SUVD_CGC_GATE__FBC_CCLK_MASK
 798		| UVD_SUVD_CGC_GATE__SCM_AV1_MASK
 799		| UVD_SUVD_CGC_GATE__SMPA_MASK);
 800	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE, data);
 801
 802	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE2);
 803	data |= (UVD_SUVD_CGC_GATE2__MPBE0_MASK
 804		| UVD_SUVD_CGC_GATE2__MPBE1_MASK
 805		| UVD_SUVD_CGC_GATE2__SIT_AV1_MASK
 806		| UVD_SUVD_CGC_GATE2__SDB_AV1_MASK
 807		| UVD_SUVD_CGC_GATE2__MPC1_MASK);
 808	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE2, data);
 809
 810	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL);
 811	data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
 812		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
 813		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
 814		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
 815		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
 816		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
 817		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
 818		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
 819		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK
 820		| UVD_SUVD_CGC_CTRL__EFC_MODE_MASK
 821		| UVD_SUVD_CGC_CTRL__SAOE_MODE_MASK
 822		| UVD_SUVD_CGC_CTRL__SMPA_MODE_MASK
 823		| UVD_SUVD_CGC_CTRL__MPBE0_MODE_MASK
 824		| UVD_SUVD_CGC_CTRL__MPBE1_MODE_MASK
 825		| UVD_SUVD_CGC_CTRL__SIT_AV1_MODE_MASK
 826		| UVD_SUVD_CGC_CTRL__SDB_AV1_MODE_MASK
 827		| UVD_SUVD_CGC_CTRL__MPC1_MODE_MASK
 828		| UVD_SUVD_CGC_CTRL__FBC_PCLK_MASK
 829		| UVD_SUVD_CGC_CTRL__FBC_CCLK_MASK);
 830	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL, data);
 831}
 832
 833static void vcn_v3_0_clock_gating_dpg_mode(struct amdgpu_device *adev,
 834		uint8_t sram_sel, int inst_idx, uint8_t indirect)
 835{
 836	uint32_t reg_data = 0;
 837
 838	/* enable sw clock gating control */
 839	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
 840		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
 841	else
 842		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
 843	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
 844	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
 845	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
 846		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
 847		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
 848		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
 849		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
 850		 UVD_CGC_CTRL__SYS_MODE_MASK |
 851		 UVD_CGC_CTRL__UDEC_MODE_MASK |
 852		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
 853		 UVD_CGC_CTRL__REGS_MODE_MASK |
 854		 UVD_CGC_CTRL__RBC_MODE_MASK |
 855		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
 856		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
 857		 UVD_CGC_CTRL__IDCT_MODE_MASK |
 858		 UVD_CGC_CTRL__MPRD_MODE_MASK |
 859		 UVD_CGC_CTRL__MPC_MODE_MASK |
 860		 UVD_CGC_CTRL__LBSI_MODE_MASK |
 861		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
 862		 UVD_CGC_CTRL__WCB_MODE_MASK |
 863		 UVD_CGC_CTRL__VCPU_MODE_MASK |
 864		 UVD_CGC_CTRL__MMSCH_MODE_MASK);
 865	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 866		VCN, inst_idx, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
 867
 868	/* turn off clock gating */
 869	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 870		VCN, inst_idx, mmUVD_CGC_GATE), 0, sram_sel, indirect);
 871
 872	/* turn on SUVD clock gating */
 873	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 874		VCN, inst_idx, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
 875
 876	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
 877	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 878		VCN, inst_idx, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
 879}
 880
 881/**
 882 * vcn_v3_0_enable_clock_gating - enable VCN clock gating
 883 *
 884 * @adev: amdgpu_device pointer
 885 * @inst: instance number
 886 *
 887 * Enable clock gating for VCN block
 888 */
 889static void vcn_v3_0_enable_clock_gating(struct amdgpu_device *adev, int inst)
 890{
 891	uint32_t data;
 892
 893	/* enable VCN CGC */
 894	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
 895	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
 896		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
 897	else
 898		data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
 899	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
 900	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
 901	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
 902
 903	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
 904	data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
 905		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
 906		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
 907		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
 908		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
 909		| UVD_CGC_CTRL__SYS_MODE_MASK
 910		| UVD_CGC_CTRL__UDEC_MODE_MASK
 911		| UVD_CGC_CTRL__MPEG2_MODE_MASK
 912		| UVD_CGC_CTRL__REGS_MODE_MASK
 913		| UVD_CGC_CTRL__RBC_MODE_MASK
 914		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
 915		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
 916		| UVD_CGC_CTRL__IDCT_MODE_MASK
 917		| UVD_CGC_CTRL__MPRD_MODE_MASK
 918		| UVD_CGC_CTRL__MPC_MODE_MASK
 919		| UVD_CGC_CTRL__LBSI_MODE_MASK
 920		| UVD_CGC_CTRL__LRBBM_MODE_MASK
 921		| UVD_CGC_CTRL__WCB_MODE_MASK
 922		| UVD_CGC_CTRL__VCPU_MODE_MASK
 923		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
 924	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
 925
 926	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL);
 927	data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
 928		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
 929		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
 930		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
 931		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
 932		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
 933		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
 934		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
 935		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK
 936		| UVD_SUVD_CGC_CTRL__EFC_MODE_MASK
 937		| UVD_SUVD_CGC_CTRL__SAOE_MODE_MASK
 938		| UVD_SUVD_CGC_CTRL__SMPA_MODE_MASK
 939		| UVD_SUVD_CGC_CTRL__MPBE0_MODE_MASK
 940		| UVD_SUVD_CGC_CTRL__MPBE1_MODE_MASK
 941		| UVD_SUVD_CGC_CTRL__SIT_AV1_MODE_MASK
 942		| UVD_SUVD_CGC_CTRL__SDB_AV1_MODE_MASK
 943		| UVD_SUVD_CGC_CTRL__MPC1_MODE_MASK
 944		| UVD_SUVD_CGC_CTRL__FBC_PCLK_MASK
 945		| UVD_SUVD_CGC_CTRL__FBC_CCLK_MASK);
 946	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL, data);
 947}
 948
 949static int vcn_v3_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
 950{
 951	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared_cpu_addr;
 952	struct amdgpu_ring *ring;
 953	uint32_t rb_bufsz, tmp;
 954
 955	/* disable register anti-hang mechanism */
 956	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1,
 957		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
 958	/* enable dynamic power gating mode */
 959	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS);
 960	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
 961	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
 962	WREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS, tmp);
 963
 964	if (indirect)
 965		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
 966
 967	/* enable clock gating */
 968	vcn_v3_0_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
 969
 970	/* enable VCPU clock */
 971	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
 972	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
 973	tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
 974	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 975		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
 976
 977	/* disable master interupt */
 978	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 979		VCN, inst_idx, mmUVD_MASTINT_EN), 0, 0, indirect);
 980
 981	/* setup mmUVD_LMI_CTRL */
 982	tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
 983		UVD_LMI_CTRL__REQ_MODE_MASK |
 984		UVD_LMI_CTRL__CRC_RESET_MASK |
 985		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
 986		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
 987		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
 988		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
 989		0x00100000L);
 990	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 991		VCN, inst_idx, mmUVD_LMI_CTRL), tmp, 0, indirect);
 992
 993	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 994		VCN, inst_idx, mmUVD_MPC_CNTL),
 995		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
 996
 997	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
 998		VCN, inst_idx, mmUVD_MPC_SET_MUXA0),
 999		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1000		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1001		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1002		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
1003
1004	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1005		VCN, inst_idx, mmUVD_MPC_SET_MUXB0),
1006		 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1007		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1008		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1009		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
1010
1011	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1012		VCN, inst_idx, mmUVD_MPC_SET_MUX),
1013		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1014		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1015		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
1016
1017	vcn_v3_0_mc_resume_dpg_mode(adev, inst_idx, indirect);
1018
1019	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1020		VCN, inst_idx, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
1021	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1022		VCN, inst_idx, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
1023
1024	/* enable LMI MC and UMC channels */
1025	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1026		VCN, inst_idx, mmUVD_LMI_CTRL2), 0, 0, indirect);
1027
1028	/* unblock VCPU register access */
1029	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1030		VCN, inst_idx, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
1031
1032	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
1033	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
1034	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1035		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
1036
1037	/* enable master interrupt */
1038	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1039		VCN, inst_idx, mmUVD_MASTINT_EN),
1040		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
1041
1042	/* add nop to workaround PSP size check */
1043	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1044		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
1045
1046	if (indirect)
1047		psp_update_vcn_sram(adev, inst_idx, adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
1048			(uint32_t)((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
1049				(uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr));
1050
1051	ring = &adev->vcn.inst[inst_idx].ring_dec;
1052	/* force RBC into idle state */
1053	rb_bufsz = order_base_2(ring->ring_size);
1054	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1055	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1056	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1057	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1058	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1059	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
1060
1061	/* Stall DPG before WPTR/RPTR reset */
1062	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1063		UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1064		~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1065	fw_shared->multi_queue.decode_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1066
1067	/* set the write pointer delay */
1068	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
1069
1070	/* set the wb address */
1071	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR_ADDR,
1072		(upper_32_bits(ring->gpu_addr) >> 2));
1073
1074	/* programm the RB_BASE for ring buffer */
1075	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1076		lower_32_bits(ring->gpu_addr));
1077	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1078		upper_32_bits(ring->gpu_addr));
1079
1080	/* Initialize the ring buffer's read and write pointers */
1081	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, 0);
1082
1083	WREG32_SOC15(VCN, inst_idx, mmUVD_SCRATCH2, 0);
1084
1085	ring->wptr = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR);
1086	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR,
1087		lower_32_bits(ring->wptr));
1088
1089	/* Reset FW shared memory RBC WPTR/RPTR */
1090	fw_shared->rb.rptr = 0;
1091	fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1092
1093	/*resetting done, fw can check RB ring */
1094	fw_shared->multi_queue.decode_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1095
1096	/* Unstall DPG */
1097	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1098		0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1099
1100	return 0;
1101}
1102
1103static int vcn_v3_0_start(struct amdgpu_device *adev)
1104{
1105	volatile struct amdgpu_fw_shared *fw_shared;
1106	struct amdgpu_ring *ring;
1107	uint32_t rb_bufsz, tmp;
1108	int i, j, k, r;
1109
1110	if (adev->pm.dpm_enabled)
1111		amdgpu_dpm_enable_uvd(adev, true);
1112
1113	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1114		if (adev->vcn.harvest_config & (1 << i))
1115			continue;
1116
1117		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG){
1118			r = vcn_v3_0_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
1119			continue;
1120		}
1121
1122		/* disable VCN power gating */
1123		vcn_v3_0_disable_static_power_gating(adev, i);
1124
1125		/* set VCN status busy */
1126		tmp = RREG32_SOC15(VCN, i, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY;
1127		WREG32_SOC15(VCN, i, mmUVD_STATUS, tmp);
1128
1129		/*SW clock gating */
1130		vcn_v3_0_disable_clock_gating(adev, i);
1131
1132		/* enable VCPU clock */
1133		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1134			UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1135
1136		/* disable master interrupt */
1137		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN), 0,
1138			~UVD_MASTINT_EN__VCPU_EN_MASK);
1139
1140		/* enable LMI MC and UMC channels */
1141		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_LMI_CTRL2), 0,
1142			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1143
1144		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1145		tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1146		tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1147		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1148
1149		/* setup mmUVD_LMI_CTRL */
1150		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL);
1151		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL, tmp |
1152			UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK	|
1153			UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1154			UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1155			UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1156
1157		/* setup mmUVD_MPC_CNTL */
1158		tmp = RREG32_SOC15(VCN, i, mmUVD_MPC_CNTL);
1159		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1160		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1161		WREG32_SOC15(VCN, i, mmUVD_MPC_CNTL, tmp);
1162
1163		/* setup UVD_MPC_SET_MUXA0 */
1164		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXA0,
1165			((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1166			(0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1167			(0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1168			(0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1169
1170		/* setup UVD_MPC_SET_MUXB0 */
1171		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXB0,
1172			((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1173			(0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1174			(0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1175			(0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1176
1177		/* setup mmUVD_MPC_SET_MUX */
1178		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUX,
1179			((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1180			(0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1181			(0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1182
1183		vcn_v3_0_mc_resume(adev, i);
1184
1185		/* VCN global tiling registers */
1186		WREG32_SOC15(VCN, i, mmUVD_GFX10_ADDR_CONFIG,
1187			adev->gfx.config.gb_addr_config);
1188
1189		/* unblock VCPU register access */
1190		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL), 0,
1191			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1192
1193		/* release VCPU reset to boot */
1194		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1195			~UVD_VCPU_CNTL__BLK_RST_MASK);
1196
1197		for (j = 0; j < 10; ++j) {
1198			uint32_t status;
1199
1200			for (k = 0; k < 100; ++k) {
1201				status = RREG32_SOC15(VCN, i, mmUVD_STATUS);
1202				if (status & 2)
1203					break;
1204				mdelay(10);
1205			}
1206			r = 0;
1207			if (status & 2)
1208				break;
1209
1210			DRM_ERROR("VCN[%d] decode not responding, trying to reset the VCPU!!!\n", i);
1211			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1212				UVD_VCPU_CNTL__BLK_RST_MASK,
1213				~UVD_VCPU_CNTL__BLK_RST_MASK);
1214			mdelay(10);
1215			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1216				~UVD_VCPU_CNTL__BLK_RST_MASK);
1217
1218			mdelay(10);
1219			r = -1;
1220		}
1221
1222		if (r) {
1223			DRM_ERROR("VCN[%d] decode not responding, giving up!!!\n", i);
1224			return r;
1225		}
1226
1227		/* enable master interrupt */
1228		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN),
1229			UVD_MASTINT_EN__VCPU_EN_MASK,
1230			~UVD_MASTINT_EN__VCPU_EN_MASK);
1231
1232		/* clear the busy bit of VCN_STATUS */
1233		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS), 0,
1234			~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1235
1236		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_VMID, 0);
1237
1238		ring = &adev->vcn.inst[i].ring_dec;
1239		/* force RBC into idle state */
1240		rb_bufsz = order_base_2(ring->ring_size);
1241		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1242		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1243		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1244		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1245		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1246		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_CNTL, tmp);
1247
1248		fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
1249		fw_shared->multi_queue.decode_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1250
1251		/* programm the RB_BASE for ring buffer */
1252		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1253			lower_32_bits(ring->gpu_addr));
1254		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1255			upper_32_bits(ring->gpu_addr));
1256
1257		/* Initialize the ring buffer's read and write pointers */
1258		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR, 0);
1259
1260		WREG32_SOC15(VCN, i, mmUVD_SCRATCH2, 0);
1261		ring->wptr = RREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR);
1262		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_WPTR,
1263			lower_32_bits(ring->wptr));
1264		fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1265		fw_shared->multi_queue.decode_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1266
1267		if (adev->asic_type != CHIP_BEIGE_GOBY) {
1268			fw_shared->multi_queue.encode_generalpurpose_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1269			ring = &adev->vcn.inst[i].ring_enc[0];
1270			WREG32_SOC15(VCN, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1271			WREG32_SOC15(VCN, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1272			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO, ring->gpu_addr);
1273			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1274			WREG32_SOC15(VCN, i, mmUVD_RB_SIZE, ring->ring_size / 4);
1275			fw_shared->multi_queue.encode_generalpurpose_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1276
1277			fw_shared->multi_queue.encode_lowlatency_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1278			ring = &adev->vcn.inst[i].ring_enc[1];
1279			WREG32_SOC15(VCN, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1280			WREG32_SOC15(VCN, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1281			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1282			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1283			WREG32_SOC15(VCN, i, mmUVD_RB_SIZE2, ring->ring_size / 4);
1284			fw_shared->multi_queue.encode_lowlatency_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1285		}
1286	}
1287
1288	return 0;
1289}
1290
1291static int vcn_v3_0_start_sriov(struct amdgpu_device *adev)
1292{
1293	int i, j;
1294	struct amdgpu_ring *ring;
1295	uint64_t cache_addr;
1296	uint64_t rb_addr;
1297	uint64_t ctx_addr;
1298	uint32_t param, resp, expected;
1299	uint32_t offset, cache_size;
1300	uint32_t tmp, timeout;
1301	uint32_t id;
1302
1303	struct amdgpu_mm_table *table = &adev->virt.mm_table;
1304	uint32_t *table_loc;
1305	uint32_t table_size;
1306	uint32_t size, size_dw;
1307
1308	struct mmsch_v3_0_cmd_direct_write
1309		direct_wt = { {0} };
1310	struct mmsch_v3_0_cmd_direct_read_modify_write
1311		direct_rd_mod_wt = { {0} };
1312	struct mmsch_v3_0_cmd_end end = { {0} };
1313	struct mmsch_v3_0_init_header header;
1314
1315	direct_wt.cmd_header.command_type =
1316		MMSCH_COMMAND__DIRECT_REG_WRITE;
1317	direct_rd_mod_wt.cmd_header.command_type =
1318		MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1319	end.cmd_header.command_type =
1320		MMSCH_COMMAND__END;
1321
1322	header.version = MMSCH_VERSION;
1323	header.total_size = sizeof(struct mmsch_v3_0_init_header) >> 2;
1324	for (i = 0; i < AMDGPU_MAX_VCN_INSTANCES; i++) {
1325		header.inst[i].init_status = 0;
1326		header.inst[i].table_offset = 0;
1327		header.inst[i].table_size = 0;
1328	}
1329
1330	table_loc = (uint32_t *)table->cpu_addr;
1331	table_loc += header.total_size;
1332	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1333		if (adev->vcn.harvest_config & (1 << i))
1334			continue;
1335
1336		table_size = 0;
1337
1338		MMSCH_V3_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(VCN, i,
1339			mmUVD_STATUS),
1340			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1341
1342		cache_size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1343
1344		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1345			id = amdgpu_ucode_id_vcns[i];
1346			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1347				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1348				adev->firmware.ucode[id].tmr_mc_addr_lo);
1349			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1350				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1351				adev->firmware.ucode[id].tmr_mc_addr_hi);
1352			offset = 0;
1353			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1354				mmUVD_VCPU_CACHE_OFFSET0),
1355				0);
1356		} else {
1357			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1358				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1359				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1360			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1361				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1362				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1363			offset = cache_size;
1364			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1365				mmUVD_VCPU_CACHE_OFFSET0),
1366				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1367		}
1368
1369		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1370			mmUVD_VCPU_CACHE_SIZE0),
1371			cache_size);
1372
1373		cache_addr = adev->vcn.inst[i].gpu_addr + offset;
1374		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1375			mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1376			lower_32_bits(cache_addr));
1377		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1378			mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1379			upper_32_bits(cache_addr));
1380		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1381			mmUVD_VCPU_CACHE_OFFSET1),
1382			0);
1383		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1384			mmUVD_VCPU_CACHE_SIZE1),
1385			AMDGPU_VCN_STACK_SIZE);
1386
1387		cache_addr = adev->vcn.inst[i].gpu_addr + offset +
1388			AMDGPU_VCN_STACK_SIZE;
1389		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1390			mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1391			lower_32_bits(cache_addr));
1392		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1393			mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1394			upper_32_bits(cache_addr));
1395		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1396			mmUVD_VCPU_CACHE_OFFSET2),
1397			0);
1398		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1399			mmUVD_VCPU_CACHE_SIZE2),
1400			AMDGPU_VCN_CONTEXT_SIZE);
1401
1402		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
1403			ring = &adev->vcn.inst[i].ring_enc[j];
1404			ring->wptr = 0;
1405			rb_addr = ring->gpu_addr;
1406			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1407				mmUVD_RB_BASE_LO),
1408				lower_32_bits(rb_addr));
1409			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1410				mmUVD_RB_BASE_HI),
1411				upper_32_bits(rb_addr));
1412			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1413				mmUVD_RB_SIZE),
1414				ring->ring_size / 4);
1415		}
1416
1417		ring = &adev->vcn.inst[i].ring_dec;
1418		ring->wptr = 0;
1419		rb_addr = ring->gpu_addr;
1420		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1421			mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1422			lower_32_bits(rb_addr));
1423		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1424			mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1425			upper_32_bits(rb_addr));
1426		/* force RBC into idle state */
1427		tmp = order_base_2(ring->ring_size);
1428		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, tmp);
1429		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1430		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1431		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1432		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1433		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1434			mmUVD_RBC_RB_CNTL),
1435			tmp);
1436
1437		/* add end packet */
1438		MMSCH_V3_0_INSERT_END();
1439
1440		/* refine header */
1441		header.inst[i].init_status = 0;
1442		header.inst[i].table_offset = header.total_size;
1443		header.inst[i].table_size = table_size;
1444		header.total_size += table_size;
1445	}
1446
1447	/* Update init table header in memory */
1448	size = sizeof(struct mmsch_v3_0_init_header);
1449	table_loc = (uint32_t *)table->cpu_addr;
1450	memcpy((void *)table_loc, &header, size);
1451
1452	/* message MMSCH (in VCN[0]) to initialize this client
1453	 * 1, write to mmsch_vf_ctx_addr_lo/hi register with GPU mc addr
1454	 * of memory descriptor location
1455	 */
1456	ctx_addr = table->gpu_addr;
1457	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(ctx_addr));
1458	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(ctx_addr));
1459
1460	/* 2, update vmid of descriptor */
1461	tmp = RREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID);
1462	tmp &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1463	/* use domain0 for MM scheduler */
1464	tmp |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1465	WREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID, tmp);
1466
1467	/* 3, notify mmsch about the size of this descriptor */
1468	size = header.total_size;
1469	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_SIZE, size);
1470
1471	/* 4, set resp to zero */
1472	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1473
1474	/* 5, kick off the initialization and wait until
1475	 * MMSCH_VF_MAILBOX_RESP becomes non-zero
1476	 */
1477	param = 0x10000001;
1478	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_HOST, param);
1479	tmp = 0;
1480	timeout = 1000;
1481	resp = 0;
1482	expected = param + 1;
1483	while (resp != expected) {
1484		resp = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1485		if (resp == expected)
1486			break;
1487
1488		udelay(10);
1489		tmp = tmp + 10;
1490		if (tmp >= timeout) {
1491			DRM_ERROR("failed to init MMSCH. TIME-OUT after %d usec"\
1492				" waiting for mmMMSCH_VF_MAILBOX_RESP "\
1493				"(expected=0x%08x, readback=0x%08x)\n",
1494				tmp, expected, resp);
1495			return -EBUSY;
1496		}
1497	}
1498
1499	return 0;
1500}
1501
1502static int vcn_v3_0_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1503{
 
1504	uint32_t tmp;
1505
 
 
1506	/* Wait for power status to be 1 */
1507	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1508		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1509
1510	/* wait for read ptr to be equal to write ptr */
1511	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR);
1512	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1513
1514	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2);
1515	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
1516
1517	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1518	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF);
1519
1520	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1521		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1522
1523	/* disable dynamic power gating mode */
1524	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 0,
1525		~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1526
1527	return 0;
1528}
1529
1530static int vcn_v3_0_stop(struct amdgpu_device *adev)
1531{
1532	uint32_t tmp;
1533	int i, r = 0;
1534
1535	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1536		if (adev->vcn.harvest_config & (1 << i))
1537			continue;
1538
1539		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1540			r = vcn_v3_0_stop_dpg_mode(adev, i);
1541			continue;
1542		}
1543
1544		/* wait for vcn idle */
1545		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1546		if (r)
1547			return r;
1548
1549		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1550			UVD_LMI_STATUS__READ_CLEAN_MASK |
1551			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1552			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1553		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1554		if (r)
1555			return r;
1556
1557		/* disable LMI UMC channel */
1558		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2);
1559		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1560		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2, tmp);
1561		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1562			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1563		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1564		if (r)
1565			return r;
1566
1567		/* block VCPU register access */
1568		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL),
1569			UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1570			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1571
1572		/* reset VCPU */
1573		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1574			UVD_VCPU_CNTL__BLK_RST_MASK,
1575			~UVD_VCPU_CNTL__BLK_RST_MASK);
1576
1577		/* disable VCPU clock */
1578		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1579			~(UVD_VCPU_CNTL__CLK_EN_MASK));
1580
1581		/* apply soft reset */
1582		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1583		tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1584		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1585		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1586		tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1587		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1588
1589		/* clear status */
1590		WREG32_SOC15(VCN, i, mmUVD_STATUS, 0);
1591
1592		/* apply HW clock gating */
1593		vcn_v3_0_enable_clock_gating(adev, i);
1594
1595		/* enable VCN power gating */
1596		vcn_v3_0_enable_static_power_gating(adev, i);
1597	}
1598
1599	if (adev->pm.dpm_enabled)
1600		amdgpu_dpm_enable_uvd(adev, false);
1601
1602	return 0;
1603}
1604
1605static int vcn_v3_0_pause_dpg_mode(struct amdgpu_device *adev,
1606		   int inst_idx, struct dpg_pause_state *new_state)
1607{
1608	volatile struct amdgpu_fw_shared *fw_shared;
1609	struct amdgpu_ring *ring;
1610	uint32_t reg_data = 0;
1611	int ret_code;
1612
1613	/* pause/unpause if state is changed */
1614	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1615		DRM_DEBUG("dpg pause state changed %d -> %d",
1616			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1617		reg_data = RREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE) &
1618			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1619
1620		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1621			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1622				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1623
1624			if (!ret_code) {
1625				/* pause DPG */
1626				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1627				WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1628
1629				/* wait for ACK */
1630				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_DPG_PAUSE,
1631					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1632					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1633
1634				/* Stall DPG before WPTR/RPTR reset */
1635				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1636					UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1637					~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1638
1639				if (adev->asic_type != CHIP_BEIGE_GOBY) {
1640					/* Restore */
1641					fw_shared = adev->vcn.inst[inst_idx].fw_shared_cpu_addr;
1642					fw_shared->multi_queue.encode_generalpurpose_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1643					ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1644					ring->wptr = 0;
1645					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
1646					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1647					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
1648					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1649					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1650					fw_shared->multi_queue.encode_generalpurpose_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1651
1652					fw_shared->multi_queue.encode_lowlatency_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1653					ring = &adev->vcn.inst[inst_idx].ring_enc[1];
1654					ring->wptr = 0;
1655					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1656					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1657					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
1658					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1659					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1660					fw_shared->multi_queue.encode_lowlatency_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1661
1662					/* restore wptr/rptr with pointers saved in FW shared memory*/
1663					WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, fw_shared->rb.rptr);
1664					WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR, fw_shared->rb.wptr);
1665				}
1666
1667				/* Unstall DPG */
1668				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1669					0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1670
1671				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS,
1672					UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1673			}
1674		} else {
1675			/* unpause dpg, no need to wait */
1676			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1677			WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1678		}
1679		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1680	}
1681
1682	return 0;
1683}
1684
1685/**
1686 * vcn_v3_0_dec_ring_get_rptr - get read pointer
1687 *
1688 * @ring: amdgpu_ring pointer
1689 *
1690 * Returns the current hardware read pointer
1691 */
1692static uint64_t vcn_v3_0_dec_ring_get_rptr(struct amdgpu_ring *ring)
1693{
1694	struct amdgpu_device *adev = ring->adev;
1695
1696	return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_RPTR);
1697}
1698
1699/**
1700 * vcn_v3_0_dec_ring_get_wptr - get write pointer
1701 *
1702 * @ring: amdgpu_ring pointer
1703 *
1704 * Returns the current hardware write pointer
1705 */
1706static uint64_t vcn_v3_0_dec_ring_get_wptr(struct amdgpu_ring *ring)
1707{
1708	struct amdgpu_device *adev = ring->adev;
1709
1710	if (ring->use_doorbell)
1711		return adev->wb.wb[ring->wptr_offs];
1712	else
1713		return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR);
1714}
1715
1716/**
1717 * vcn_v3_0_dec_ring_set_wptr - set write pointer
1718 *
1719 * @ring: amdgpu_ring pointer
1720 *
1721 * Commits the write pointer to the hardware
1722 */
1723static void vcn_v3_0_dec_ring_set_wptr(struct amdgpu_ring *ring)
1724{
1725	struct amdgpu_device *adev = ring->adev;
1726	volatile struct amdgpu_fw_shared *fw_shared;
1727
1728	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1729		/*whenever update RBC_RB_WPTR, we save the wptr in shared rb.wptr and scratch2 */
1730		fw_shared = adev->vcn.inst[ring->me].fw_shared_cpu_addr;
1731		fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1732		WREG32_SOC15(VCN, ring->me, mmUVD_SCRATCH2,
1733			lower_32_bits(ring->wptr));
1734	}
1735
1736	if (ring->use_doorbell) {
1737		adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1738		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1739	} else {
1740		WREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1741	}
1742}
1743
1744static void vcn_v3_0_dec_sw_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
1745				u64 seq, uint32_t flags)
1746{
1747	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
1748
1749	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_FENCE);
1750	amdgpu_ring_write(ring, addr);
1751	amdgpu_ring_write(ring, upper_32_bits(addr));
1752	amdgpu_ring_write(ring, seq);
1753	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_TRAP);
1754}
1755
1756static void vcn_v3_0_dec_sw_ring_insert_end(struct amdgpu_ring *ring)
1757{
1758	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_END);
1759}
1760
1761static void vcn_v3_0_dec_sw_ring_emit_ib(struct amdgpu_ring *ring,
1762			       struct amdgpu_job *job,
1763			       struct amdgpu_ib *ib,
1764			       uint32_t flags)
1765{
1766	uint32_t vmid = AMDGPU_JOB_GET_VMID(job);
1767
1768	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_IB);
1769	amdgpu_ring_write(ring, vmid);
1770	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
1771	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
1772	amdgpu_ring_write(ring, ib->length_dw);
1773}
1774
1775static void vcn_v3_0_dec_sw_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
1776				uint32_t val, uint32_t mask)
1777{
1778	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_REG_WAIT);
1779	amdgpu_ring_write(ring, reg << 2);
1780	amdgpu_ring_write(ring, mask);
1781	amdgpu_ring_write(ring, val);
1782}
1783
1784static void vcn_v3_0_dec_sw_ring_emit_vm_flush(struct amdgpu_ring *ring,
1785				uint32_t vmid, uint64_t pd_addr)
1786{
1787	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
1788	uint32_t data0, data1, mask;
1789
1790	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1791
1792	/* wait for register write */
1793	data0 = hub->ctx0_ptb_addr_lo32 + vmid * hub->ctx_addr_distance;
1794	data1 = lower_32_bits(pd_addr);
1795	mask = 0xffffffff;
1796	vcn_v3_0_dec_sw_ring_emit_reg_wait(ring, data0, data1, mask);
1797}
1798
1799static void vcn_v3_0_dec_sw_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg, uint32_t val)
1800{
1801	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_REG_WRITE);
1802	amdgpu_ring_write(ring,	reg << 2);
1803	amdgpu_ring_write(ring, val);
1804}
1805
1806static const struct amdgpu_ring_funcs vcn_v3_0_dec_sw_ring_vm_funcs = {
1807	.type = AMDGPU_RING_TYPE_VCN_DEC,
1808	.align_mask = 0x3f,
1809	.nop = VCN_DEC_SW_CMD_NO_OP,
 
1810	.vmhub = AMDGPU_MMHUB_0,
1811	.get_rptr = vcn_v3_0_dec_ring_get_rptr,
1812	.get_wptr = vcn_v3_0_dec_ring_get_wptr,
1813	.set_wptr = vcn_v3_0_dec_ring_set_wptr,
1814	.emit_frame_size =
1815		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1816		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1817		4 + /* vcn_v3_0_dec_sw_ring_emit_vm_flush */
1818		5 + 5 + /* vcn_v3_0_dec_sw_ring_emit_fdec_swe x2 vm fdec_swe */
1819		1, /* vcn_v3_0_dec_sw_ring_insert_end */
1820	.emit_ib_size = 5, /* vcn_v3_0_dec_sw_ring_emit_ib */
1821	.emit_ib = vcn_v3_0_dec_sw_ring_emit_ib,
1822	.emit_fence = vcn_v3_0_dec_sw_ring_emit_fence,
1823	.emit_vm_flush = vcn_v3_0_dec_sw_ring_emit_vm_flush,
1824	.test_ring = amdgpu_vcn_dec_sw_ring_test_ring,
1825	.test_ib = NULL,//amdgpu_vcn_dec_sw_ring_test_ib,
1826	.insert_nop = amdgpu_ring_insert_nop,
1827	.insert_end = vcn_v3_0_dec_sw_ring_insert_end,
1828	.pad_ib = amdgpu_ring_generic_pad_ib,
1829	.begin_use = amdgpu_vcn_ring_begin_use,
1830	.end_use = amdgpu_vcn_ring_end_use,
1831	.emit_wreg = vcn_v3_0_dec_sw_ring_emit_wreg,
1832	.emit_reg_wait = vcn_v3_0_dec_sw_ring_emit_reg_wait,
1833	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1834};
1835
1836static int vcn_v3_0_limit_sched(struct amdgpu_cs_parser *p)
 
1837{
1838	struct drm_gpu_scheduler **scheds;
1839
1840	/* The create msg must be in the first IB submitted */
1841	if (atomic_read(&p->entity->fence_seq))
1842		return -EINVAL;
1843
1844	scheds = p->adev->gpu_sched[AMDGPU_HW_IP_VCN_DEC]
1845		[AMDGPU_RING_PRIO_DEFAULT].sched;
1846	drm_sched_entity_modify_sched(p->entity, scheds, 1);
1847	return 0;
1848}
1849
1850static int vcn_v3_0_dec_msg(struct amdgpu_cs_parser *p, uint64_t addr)
 
1851{
1852	struct ttm_operation_ctx ctx = { false, false };
1853	struct amdgpu_bo_va_mapping *map;
1854	uint32_t *msg, num_buffers;
1855	struct amdgpu_bo *bo;
1856	uint64_t start, end;
1857	unsigned int i;
1858	void * ptr;
1859	int r;
1860
1861	addr &= AMDGPU_GMC_HOLE_MASK;
1862	r = amdgpu_cs_find_mapping(p, addr, &bo, &map);
1863	if (r) {
1864		DRM_ERROR("Can't find BO for addr 0x%08Lx\n", addr);
1865		return r;
1866	}
1867
1868	start = map->start * AMDGPU_GPU_PAGE_SIZE;
1869	end = (map->last + 1) * AMDGPU_GPU_PAGE_SIZE;
1870	if (addr & 0x7) {
1871		DRM_ERROR("VCN messages must be 8 byte aligned!\n");
1872		return -EINVAL;
1873	}
1874
1875	bo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
1876	amdgpu_bo_placement_from_domain(bo, bo->allowed_domains);
1877	r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
1878	if (r) {
1879		DRM_ERROR("Failed validating the VCN message BO (%d)!\n", r);
1880		return r;
1881	}
1882
1883	r = amdgpu_bo_kmap(bo, &ptr);
1884	if (r) {
1885		DRM_ERROR("Failed mapping the VCN message (%d)!\n", r);
1886		return r;
1887	}
1888
1889	msg = ptr + addr - start;
1890
1891	/* Check length */
1892	if (msg[1] > end - addr) {
1893		r = -EINVAL;
1894		goto out;
1895	}
1896
1897	if (msg[3] != RDECODE_MSG_CREATE)
1898		goto out;
1899
1900	num_buffers = msg[2];
1901	for (i = 0, msg = &msg[6]; i < num_buffers; ++i, msg += 4) {
1902		uint32_t offset, size, *create;
1903
1904		if (msg[0] != RDECODE_MESSAGE_CREATE)
1905			continue;
1906
1907		offset = msg[1];
1908		size = msg[2];
1909
1910		if (offset + size > end) {
1911			r = -EINVAL;
1912			goto out;
1913		}
1914
1915		create = ptr + addr + offset - start;
1916
1917		/* H246, HEVC and VP9 can run on any instance */
1918		if (create[0] == 0x7 || create[0] == 0x10 || create[0] == 0x11)
1919			continue;
1920
1921		r = vcn_v3_0_limit_sched(p);
1922		if (r)
1923			goto out;
1924	}
1925
1926out:
1927	amdgpu_bo_kunmap(bo);
1928	return r;
1929}
1930
1931static int vcn_v3_0_ring_patch_cs_in_place(struct amdgpu_cs_parser *p,
1932					   uint32_t ib_idx)
 
1933{
1934	struct amdgpu_ring *ring = to_amdgpu_ring(p->entity->rq->sched);
1935	struct amdgpu_ib *ib = &p->job->ibs[ib_idx];
1936	uint32_t msg_lo = 0, msg_hi = 0;
1937	unsigned i;
1938	int r;
1939
1940	/* The first instance can decode anything */
1941	if (!ring->me)
1942		return 0;
1943
1944	for (i = 0; i < ib->length_dw; i += 2) {
1945		uint32_t reg = amdgpu_get_ib_value(p, ib_idx, i);
1946		uint32_t val = amdgpu_get_ib_value(p, ib_idx, i + 1);
1947
1948		if (reg == PACKET0(p->adev->vcn.internal.data0, 0)) {
1949			msg_lo = val;
1950		} else if (reg == PACKET0(p->adev->vcn.internal.data1, 0)) {
1951			msg_hi = val;
1952		} else if (reg == PACKET0(p->adev->vcn.internal.cmd, 0) &&
1953			   val == 0) {
1954			r = vcn_v3_0_dec_msg(p, ((u64)msg_hi) << 32 | msg_lo);
 
1955			if (r)
1956				return r;
1957		}
1958	}
1959	return 0;
1960}
1961
1962static const struct amdgpu_ring_funcs vcn_v3_0_dec_ring_vm_funcs = {
1963	.type = AMDGPU_RING_TYPE_VCN_DEC,
1964	.align_mask = 0xf,
 
1965	.vmhub = AMDGPU_MMHUB_0,
1966	.get_rptr = vcn_v3_0_dec_ring_get_rptr,
1967	.get_wptr = vcn_v3_0_dec_ring_get_wptr,
1968	.set_wptr = vcn_v3_0_dec_ring_set_wptr,
1969	.patch_cs_in_place = vcn_v3_0_ring_patch_cs_in_place,
1970	.emit_frame_size =
1971		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1972		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1973		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1974		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1975		6,
1976	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1977	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
1978	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
1979	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1980	.test_ring = vcn_v2_0_dec_ring_test_ring,
1981	.test_ib = amdgpu_vcn_dec_ring_test_ib,
1982	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
1983	.insert_start = vcn_v2_0_dec_ring_insert_start,
1984	.insert_end = vcn_v2_0_dec_ring_insert_end,
1985	.pad_ib = amdgpu_ring_generic_pad_ib,
1986	.begin_use = amdgpu_vcn_ring_begin_use,
1987	.end_use = amdgpu_vcn_ring_end_use,
1988	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1989	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1990	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1991};
1992
1993/**
1994 * vcn_v3_0_enc_ring_get_rptr - get enc read pointer
1995 *
1996 * @ring: amdgpu_ring pointer
1997 *
1998 * Returns the current hardware enc read pointer
1999 */
2000static uint64_t vcn_v3_0_enc_ring_get_rptr(struct amdgpu_ring *ring)
2001{
2002	struct amdgpu_device *adev = ring->adev;
2003
2004	if (ring == &adev->vcn.inst[ring->me].ring_enc[0])
2005		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR);
2006	else
2007		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR2);
2008}
2009
2010/**
2011 * vcn_v3_0_enc_ring_get_wptr - get enc write pointer
2012 *
2013 * @ring: amdgpu_ring pointer
2014 *
2015 * Returns the current hardware enc write pointer
2016 */
2017static uint64_t vcn_v3_0_enc_ring_get_wptr(struct amdgpu_ring *ring)
2018{
2019	struct amdgpu_device *adev = ring->adev;
2020
2021	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
2022		if (ring->use_doorbell)
2023			return adev->wb.wb[ring->wptr_offs];
2024		else
2025			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR);
2026	} else {
2027		if (ring->use_doorbell)
2028			return adev->wb.wb[ring->wptr_offs];
2029		else
2030			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2);
2031	}
2032}
2033
2034/**
2035 * vcn_v3_0_enc_ring_set_wptr - set enc write pointer
2036 *
2037 * @ring: amdgpu_ring pointer
2038 *
2039 * Commits the enc write pointer to the hardware
2040 */
2041static void vcn_v3_0_enc_ring_set_wptr(struct amdgpu_ring *ring)
2042{
2043	struct amdgpu_device *adev = ring->adev;
2044
2045	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
2046		if (ring->use_doorbell) {
2047			adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
2048			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
2049		} else {
2050			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
2051		}
2052	} else {
2053		if (ring->use_doorbell) {
2054			adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
2055			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
2056		} else {
2057			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
2058		}
2059	}
2060}
2061
2062static const struct amdgpu_ring_funcs vcn_v3_0_enc_ring_vm_funcs = {
2063	.type = AMDGPU_RING_TYPE_VCN_ENC,
2064	.align_mask = 0x3f,
2065	.nop = VCN_ENC_CMD_NO_OP,
2066	.vmhub = AMDGPU_MMHUB_0,
2067	.get_rptr = vcn_v3_0_enc_ring_get_rptr,
2068	.get_wptr = vcn_v3_0_enc_ring_get_wptr,
2069	.set_wptr = vcn_v3_0_enc_ring_set_wptr,
2070	.emit_frame_size =
2071		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2072		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
2073		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
2074		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
2075		1, /* vcn_v2_0_enc_ring_insert_end */
2076	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
2077	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
2078	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
2079	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
2080	.test_ring = amdgpu_vcn_enc_ring_test_ring,
2081	.test_ib = amdgpu_vcn_enc_ring_test_ib,
2082	.insert_nop = amdgpu_ring_insert_nop,
2083	.insert_end = vcn_v2_0_enc_ring_insert_end,
2084	.pad_ib = amdgpu_ring_generic_pad_ib,
2085	.begin_use = amdgpu_vcn_ring_begin_use,
2086	.end_use = amdgpu_vcn_ring_end_use,
2087	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
2088	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
2089	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2090};
2091
2092static void vcn_v3_0_set_dec_ring_funcs(struct amdgpu_device *adev)
2093{
2094	int i;
2095
2096	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2097		if (adev->vcn.harvest_config & (1 << i))
2098			continue;
2099
2100		if (!DEC_SW_RING_ENABLED)
2101			adev->vcn.inst[i].ring_dec.funcs = &vcn_v3_0_dec_ring_vm_funcs;
2102		else
2103			adev->vcn.inst[i].ring_dec.funcs = &vcn_v3_0_dec_sw_ring_vm_funcs;
2104		adev->vcn.inst[i].ring_dec.me = i;
2105		DRM_INFO("VCN(%d) decode%s is enabled in VM mode\n", i,
2106			  DEC_SW_RING_ENABLED?"(Software Ring)":"");
2107	}
2108}
2109
2110static void vcn_v3_0_set_enc_ring_funcs(struct amdgpu_device *adev)
2111{
2112	int i, j;
2113
2114	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2115		if (adev->vcn.harvest_config & (1 << i))
2116			continue;
2117
2118		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
2119			adev->vcn.inst[i].ring_enc[j].funcs = &vcn_v3_0_enc_ring_vm_funcs;
2120			adev->vcn.inst[i].ring_enc[j].me = i;
2121		}
2122		if (adev->vcn.num_enc_rings > 0)
2123			DRM_INFO("VCN(%d) encode is enabled in VM mode\n", i);
2124	}
2125}
2126
2127static bool vcn_v3_0_is_idle(void *handle)
2128{
2129	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2130	int i, ret = 1;
2131
2132	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2133		if (adev->vcn.harvest_config & (1 << i))
2134			continue;
2135
2136		ret &= (RREG32_SOC15(VCN, i, mmUVD_STATUS) == UVD_STATUS__IDLE);
2137	}
2138
2139	return ret;
2140}
2141
2142static int vcn_v3_0_wait_for_idle(void *handle)
2143{
2144	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2145	int i, ret = 0;
2146
2147	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2148		if (adev->vcn.harvest_config & (1 << i))
2149			continue;
2150
2151		ret = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE,
2152			UVD_STATUS__IDLE);
2153		if (ret)
2154			return ret;
2155	}
2156
2157	return ret;
2158}
2159
2160static int vcn_v3_0_set_clockgating_state(void *handle,
2161					  enum amd_clockgating_state state)
2162{
2163	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2164	bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
2165	int i;
2166
2167	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2168		if (adev->vcn.harvest_config & (1 << i))
2169			continue;
2170
2171		if (enable) {
2172			if (RREG32_SOC15(VCN, i, mmUVD_STATUS) != UVD_STATUS__IDLE)
2173				return -EBUSY;
2174			vcn_v3_0_enable_clock_gating(adev, i);
2175		} else {
2176			vcn_v3_0_disable_clock_gating(adev, i);
2177		}
2178	}
2179
2180	return 0;
2181}
2182
2183static int vcn_v3_0_set_powergating_state(void *handle,
2184					  enum amd_powergating_state state)
2185{
2186	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2187	int ret;
2188
2189	/* for SRIOV, guest should not control VCN Power-gating
2190	 * MMSCH FW should control Power-gating and clock-gating
2191	 * guest should avoid touching CGC and PG
2192	 */
2193	if (amdgpu_sriov_vf(adev)) {
2194		adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
2195		return 0;
2196	}
2197
2198	if(state == adev->vcn.cur_state)
2199		return 0;
2200
2201	if (state == AMD_PG_STATE_GATE)
2202		ret = vcn_v3_0_stop(adev);
2203	else
2204		ret = vcn_v3_0_start(adev);
2205
2206	if(!ret)
2207		adev->vcn.cur_state = state;
2208
2209	return ret;
2210}
2211
2212static int vcn_v3_0_set_interrupt_state(struct amdgpu_device *adev,
2213					struct amdgpu_irq_src *source,
2214					unsigned type,
2215					enum amdgpu_interrupt_state state)
2216{
2217	return 0;
2218}
2219
2220static int vcn_v3_0_process_interrupt(struct amdgpu_device *adev,
2221				      struct amdgpu_irq_src *source,
2222				      struct amdgpu_iv_entry *entry)
2223{
2224	uint32_t ip_instance;
2225
2226	switch (entry->client_id) {
2227	case SOC15_IH_CLIENTID_VCN:
2228		ip_instance = 0;
2229		break;
2230	case SOC15_IH_CLIENTID_VCN1:
2231		ip_instance = 1;
2232		break;
2233	default:
2234		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
2235		return 0;
2236	}
2237
2238	DRM_DEBUG("IH: VCN TRAP\n");
2239
2240	switch (entry->src_id) {
2241	case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
2242		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec);
2243		break;
2244	case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
2245		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
2246		break;
2247	case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
2248		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]);
2249		break;
2250	default:
2251		DRM_ERROR("Unhandled interrupt: %d %d\n",
2252			  entry->src_id, entry->src_data[0]);
2253		break;
2254	}
2255
2256	return 0;
2257}
2258
2259static const struct amdgpu_irq_src_funcs vcn_v3_0_irq_funcs = {
2260	.set = vcn_v3_0_set_interrupt_state,
2261	.process = vcn_v3_0_process_interrupt,
2262};
2263
2264static void vcn_v3_0_set_irq_funcs(struct amdgpu_device *adev)
2265{
2266	int i;
2267
2268	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2269		if (adev->vcn.harvest_config & (1 << i))
2270			continue;
2271
2272		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
2273		adev->vcn.inst[i].irq.funcs = &vcn_v3_0_irq_funcs;
2274	}
2275}
2276
2277static const struct amd_ip_funcs vcn_v3_0_ip_funcs = {
2278	.name = "vcn_v3_0",
2279	.early_init = vcn_v3_0_early_init,
2280	.late_init = NULL,
2281	.sw_init = vcn_v3_0_sw_init,
2282	.sw_fini = vcn_v3_0_sw_fini,
2283	.hw_init = vcn_v3_0_hw_init,
2284	.hw_fini = vcn_v3_0_hw_fini,
2285	.suspend = vcn_v3_0_suspend,
2286	.resume = vcn_v3_0_resume,
2287	.is_idle = vcn_v3_0_is_idle,
2288	.wait_for_idle = vcn_v3_0_wait_for_idle,
2289	.check_soft_reset = NULL,
2290	.pre_soft_reset = NULL,
2291	.soft_reset = NULL,
2292	.post_soft_reset = NULL,
2293	.set_clockgating_state = vcn_v3_0_set_clockgating_state,
2294	.set_powergating_state = vcn_v3_0_set_powergating_state,
2295};
2296
2297const struct amdgpu_ip_block_version vcn_v3_0_ip_block =
2298{
2299	.type = AMD_IP_BLOCK_TYPE_VCN,
2300	.major = 3,
2301	.minor = 0,
2302	.rev = 0,
2303	.funcs = &vcn_v3_0_ip_funcs,
2304};