xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c (revision c4a11bf4)
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 
58 static int amdgpu_ih_clientid_vcns[] = {
59 	SOC15_IH_CLIENTID_VCN,
60 	SOC15_IH_CLIENTID_VCN1
61 };
62 
63 static int vcn_v3_0_start_sriov(struct amdgpu_device *adev);
64 static void vcn_v3_0_set_dec_ring_funcs(struct amdgpu_device *adev);
65 static void vcn_v3_0_set_enc_ring_funcs(struct amdgpu_device *adev);
66 static void vcn_v3_0_set_irq_funcs(struct amdgpu_device *adev);
67 static int vcn_v3_0_set_powergating_state(void *handle,
68 			enum amd_powergating_state state);
69 static int vcn_v3_0_pause_dpg_mode(struct amdgpu_device *adev,
70 			int inst_idx, struct dpg_pause_state *new_state);
71 
72 static void vcn_v3_0_dec_ring_set_wptr(struct amdgpu_ring *ring);
73 static void vcn_v3_0_enc_ring_set_wptr(struct amdgpu_ring *ring);
74 
75 /**
76  * vcn_v3_0_early_init - set function pointers
77  *
78  * @handle: amdgpu_device pointer
79  *
80  * Set ring and irq function pointers
81  */
82 static int vcn_v3_0_early_init(void *handle)
83 {
84 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
85 
86 	if (amdgpu_sriov_vf(adev)) {
87 		adev->vcn.num_vcn_inst = VCN_INSTANCES_SIENNA_CICHLID;
88 		adev->vcn.harvest_config = 0;
89 		adev->vcn.num_enc_rings = 1;
90 
91 	} else {
92 		if (adev->vcn.harvest_config == (AMDGPU_VCN_HARVEST_VCN0 |
93 						 AMDGPU_VCN_HARVEST_VCN1))
94 			/* both instances are harvested, disable the block */
95 			return -ENOENT;
96 
97 		if (adev->ip_versions[UVD_HWIP][0] == IP_VERSION(3, 0, 33))
98 			adev->vcn.num_enc_rings = 0;
99 		else
100 			adev->vcn.num_enc_rings = 2;
101 	}
102 
103 	vcn_v3_0_set_dec_ring_funcs(adev);
104 	vcn_v3_0_set_enc_ring_funcs(adev);
105 	vcn_v3_0_set_irq_funcs(adev);
106 
107 	return 0;
108 }
109 
110 /**
111  * vcn_v3_0_sw_init - sw init for VCN block
112  *
113  * @handle: amdgpu_device pointer
114  *
115  * Load firmware and sw initialization
116  */
117 static int vcn_v3_0_sw_init(void *handle)
118 {
119 	struct amdgpu_ring *ring;
120 	int i, j, r;
121 	int vcn_doorbell_index = 0;
122 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
123 
124 	r = amdgpu_vcn_sw_init(adev);
125 	if (r)
126 		return r;
127 
128 	amdgpu_vcn_setup_ucode(adev);
129 
130 	r = amdgpu_vcn_resume(adev);
131 	if (r)
132 		return r;
133 
134 	/*
135 	 * Note: doorbell assignment is fixed for SRIOV multiple VCN engines
136 	 * Formula:
137 	 *   vcn_db_base  = adev->doorbell_index.vcn.vcn_ring0_1 << 1;
138 	 *   dec_ring_i   = vcn_db_base + i * (adev->vcn.num_enc_rings + 1)
139 	 *   enc_ring_i,j = vcn_db_base + i * (adev->vcn.num_enc_rings + 1) + 1 + j
140 	 */
141 	if (amdgpu_sriov_vf(adev)) {
142 		vcn_doorbell_index = adev->doorbell_index.vcn.vcn_ring0_1;
143 		/* get DWORD offset */
144 		vcn_doorbell_index = vcn_doorbell_index << 1;
145 	}
146 
147 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
148 		volatile struct amdgpu_fw_shared *fw_shared;
149 
150 		if (adev->vcn.harvest_config & (1 << i))
151 			continue;
152 
153 		adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
154 		adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
155 		adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
156 		adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
157 		adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
158 		adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
159 
160 		adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
161 		adev->vcn.inst[i].external.scratch9 = SOC15_REG_OFFSET(VCN, i, mmUVD_SCRATCH9);
162 		adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
163 		adev->vcn.inst[i].external.data0 = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_DATA0);
164 		adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
165 		adev->vcn.inst[i].external.data1 = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_DATA1);
166 		adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
167 		adev->vcn.inst[i].external.cmd = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_CMD);
168 		adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
169 		adev->vcn.inst[i].external.nop = SOC15_REG_OFFSET(VCN, i, mmUVD_NO_OP);
170 
171 		/* VCN DEC TRAP */
172 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
173 				VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst[i].irq);
174 		if (r)
175 			return r;
176 
177 		atomic_set(&adev->vcn.inst[i].sched_score, 0);
178 
179 		ring = &adev->vcn.inst[i].ring_dec;
180 		ring->use_doorbell = true;
181 		if (amdgpu_sriov_vf(adev)) {
182 			ring->doorbell_index = vcn_doorbell_index + i * (adev->vcn.num_enc_rings + 1);
183 		} else {
184 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i;
185 		}
186 		sprintf(ring->name, "vcn_dec_%d", i);
187 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
188 				     AMDGPU_RING_PRIO_DEFAULT,
189 				     &adev->vcn.inst[i].sched_score);
190 		if (r)
191 			return r;
192 
193 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
194 			enum amdgpu_ring_priority_level hw_prio = amdgpu_vcn_get_enc_ring_prio(j);
195 
196 			/* VCN ENC TRAP */
197 			r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
198 				j + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);
199 			if (r)
200 				return r;
201 
202 			ring = &adev->vcn.inst[i].ring_enc[j];
203 			ring->use_doorbell = true;
204 			if (amdgpu_sriov_vf(adev)) {
205 				ring->doorbell_index = vcn_doorbell_index + i * (adev->vcn.num_enc_rings + 1) + 1 + j;
206 			} else {
207 				ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + j + 8 * i;
208 			}
209 			sprintf(ring->name, "vcn_enc_%d.%d", i, j);
210 			r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
211 					     hw_prio, &adev->vcn.inst[i].sched_score);
212 			if (r)
213 				return r;
214 		}
215 
216 		fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
217 		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_SW_RING_FLAG) |
218 					     cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG) |
219 					     cpu_to_le32(AMDGPU_VCN_FW_SHARED_FLAG_0_RB);
220 		fw_shared->sw_ring.is_enabled = cpu_to_le32(DEC_SW_RING_ENABLED);
221 	}
222 
223 	if (amdgpu_sriov_vf(adev)) {
224 		r = amdgpu_virt_alloc_mm_table(adev);
225 		if (r)
226 			return r;
227 	}
228 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
229 		adev->vcn.pause_dpg_mode = vcn_v3_0_pause_dpg_mode;
230 
231 	return 0;
232 }
233 
234 /**
235  * vcn_v3_0_sw_fini - sw fini for VCN block
236  *
237  * @handle: amdgpu_device pointer
238  *
239  * VCN suspend and free up sw allocation
240  */
241 static int vcn_v3_0_sw_fini(void *handle)
242 {
243 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
244 	int i, r, idx;
245 
246 	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
247 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
248 			volatile struct amdgpu_fw_shared *fw_shared;
249 
250 			if (adev->vcn.harvest_config & (1 << i))
251 				continue;
252 			fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
253 			fw_shared->present_flag_0 = 0;
254 			fw_shared->sw_ring.is_enabled = false;
255 		}
256 
257 		drm_dev_exit(idx);
258 	}
259 
260 	if (amdgpu_sriov_vf(adev))
261 		amdgpu_virt_free_mm_table(adev);
262 
263 	r = amdgpu_vcn_suspend(adev);
264 	if (r)
265 		return r;
266 
267 	r = amdgpu_vcn_sw_fini(adev);
268 
269 	return r;
270 }
271 
272 /**
273  * vcn_v3_0_hw_init - start and test VCN block
274  *
275  * @handle: amdgpu_device pointer
276  *
277  * Initialize the hardware, boot up the VCPU and do some testing
278  */
279 static int vcn_v3_0_hw_init(void *handle)
280 {
281 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
282 	struct amdgpu_ring *ring;
283 	int i, j, r;
284 
285 	if (amdgpu_sriov_vf(adev)) {
286 		r = vcn_v3_0_start_sriov(adev);
287 		if (r)
288 			goto done;
289 
290 		/* initialize VCN dec and enc ring buffers */
291 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
292 			if (adev->vcn.harvest_config & (1 << i))
293 				continue;
294 
295 			ring = &adev->vcn.inst[i].ring_dec;
296 			if (amdgpu_vcn_is_disabled_vcn(adev, VCN_DECODE_RING, i)) {
297 				ring->sched.ready = false;
298 				dev_info(adev->dev, "ring %s is disabled by hypervisor\n", ring->name);
299 			} else {
300 				ring->wptr = 0;
301 				ring->wptr_old = 0;
302 				vcn_v3_0_dec_ring_set_wptr(ring);
303 				ring->sched.ready = true;
304 			}
305 
306 			for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
307 				ring = &adev->vcn.inst[i].ring_enc[j];
308 				if (amdgpu_vcn_is_disabled_vcn(adev, VCN_ENCODE_RING, i)) {
309 					ring->sched.ready = false;
310 					dev_info(adev->dev, "ring %s is disabled by hypervisor\n", ring->name);
311 				} else {
312 					ring->wptr = 0;
313 					ring->wptr_old = 0;
314 					vcn_v3_0_enc_ring_set_wptr(ring);
315 					ring->sched.ready = true;
316 				}
317 			}
318 		}
319 	} else {
320 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
321 			if (adev->vcn.harvest_config & (1 << i))
322 				continue;
323 
324 			ring = &adev->vcn.inst[i].ring_dec;
325 
326 			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
327 						     ring->doorbell_index, i);
328 
329 			r = amdgpu_ring_test_helper(ring);
330 			if (r)
331 				goto done;
332 
333 			for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
334 				ring = &adev->vcn.inst[i].ring_enc[j];
335 				r = amdgpu_ring_test_helper(ring);
336 				if (r)
337 					goto done;
338 			}
339 		}
340 	}
341 
342 done:
343 	if (!r)
344 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
345 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
346 
347 	return r;
348 }
349 
350 /**
351  * vcn_v3_0_hw_fini - stop the hardware block
352  *
353  * @handle: amdgpu_device pointer
354  *
355  * Stop the VCN block, mark ring as not ready any more
356  */
357 static int vcn_v3_0_hw_fini(void *handle)
358 {
359 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
360 	int i;
361 
362 	cancel_delayed_work_sync(&adev->vcn.idle_work);
363 
364 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
365 		if (adev->vcn.harvest_config & (1 << i))
366 			continue;
367 
368 		if (!amdgpu_sriov_vf(adev)) {
369 			if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
370 					(adev->vcn.cur_state != AMD_PG_STATE_GATE &&
371 					 RREG32_SOC15(VCN, i, mmUVD_STATUS))) {
372 				vcn_v3_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
373 			}
374 		}
375 	}
376 
377 	return 0;
378 }
379 
380 /**
381  * vcn_v3_0_suspend - suspend VCN block
382  *
383  * @handle: amdgpu_device pointer
384  *
385  * HW fini and suspend VCN block
386  */
387 static int vcn_v3_0_suspend(void *handle)
388 {
389 	int r;
390 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
391 
392 	r = vcn_v3_0_hw_fini(adev);
393 	if (r)
394 		return r;
395 
396 	r = amdgpu_vcn_suspend(adev);
397 
398 	return r;
399 }
400 
401 /**
402  * vcn_v3_0_resume - resume VCN block
403  *
404  * @handle: amdgpu_device pointer
405  *
406  * Resume firmware and hw init VCN block
407  */
408 static int vcn_v3_0_resume(void *handle)
409 {
410 	int r;
411 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
412 
413 	r = amdgpu_vcn_resume(adev);
414 	if (r)
415 		return r;
416 
417 	r = vcn_v3_0_hw_init(adev);
418 
419 	return r;
420 }
421 
422 /**
423  * vcn_v3_0_mc_resume - memory controller programming
424  *
425  * @adev: amdgpu_device pointer
426  * @inst: instance number
427  *
428  * Let the VCN memory controller know it's offsets
429  */
430 static void vcn_v3_0_mc_resume(struct amdgpu_device *adev, int inst)
431 {
432 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
433 	uint32_t offset;
434 
435 	/* cache window 0: fw */
436 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
437 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
438 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo));
439 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
440 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi));
441 		WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET0, 0);
442 		offset = 0;
443 	} else {
444 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
445 			lower_32_bits(adev->vcn.inst[inst].gpu_addr));
446 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
447 			upper_32_bits(adev->vcn.inst[inst].gpu_addr));
448 		offset = size;
449 		WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET0,
450 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
451 	}
452 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE0, size);
453 
454 	/* cache window 1: stack */
455 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
456 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
457 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
458 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
459 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET1, 0);
460 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
461 
462 	/* cache window 2: context */
463 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
464 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
465 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
466 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
467 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET2, 0);
468 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
469 
470 	/* non-cache window */
471 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
472 		lower_32_bits(adev->vcn.inst[inst].fw_shared_gpu_addr));
473 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
474 		upper_32_bits(adev->vcn.inst[inst].fw_shared_gpu_addr));
475 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
476 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_NONCACHE_SIZE0,
477 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
478 }
479 
480 static void vcn_v3_0_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
481 {
482 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
483 	uint32_t offset;
484 
485 	/* cache window 0: fw */
486 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
487 		if (!indirect) {
488 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
489 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
490 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
491 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
492 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
493 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
494 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
495 				VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
496 		} else {
497 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
498 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
499 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
500 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
501 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
502 				VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
503 		}
504 		offset = 0;
505 	} else {
506 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
507 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
508 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
509 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
510 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
511 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
512 		offset = size;
513 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
514 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0),
515 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
516 	}
517 
518 	if (!indirect)
519 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
520 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
521 	else
522 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
523 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
524 
525 	/* cache window 1: stack */
526 	if (!indirect) {
527 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
528 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
529 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
530 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
531 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
532 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
533 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
534 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
535 	} else {
536 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
537 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
538 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
539 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
540 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
541 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
542 	}
543 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
544 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
545 
546 	/* cache window 2: context */
547 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
548 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
549 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
550 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
551 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
552 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
553 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
554 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
555 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
556 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
557 
558 	/* non-cache window */
559 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
560 			VCN, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
561 			lower_32_bits(adev->vcn.inst[inst_idx].fw_shared_gpu_addr), 0, indirect);
562 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
563 			VCN, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
564 			upper_32_bits(adev->vcn.inst[inst_idx].fw_shared_gpu_addr), 0, indirect);
565 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
566 			VCN, inst_idx, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
567 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
568 			VCN, inst_idx, mmUVD_VCPU_NONCACHE_SIZE0),
569 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
570 
571 	/* VCN global tiling registers */
572 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
573 		UVD, 0, mmUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
574 }
575 
576 static void vcn_v3_0_disable_static_power_gating(struct amdgpu_device *adev, int inst)
577 {
578 	uint32_t data = 0;
579 
580 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
581 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
582 			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
583 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
584 			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
585 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
586 			| 2 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
587 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
588 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
589 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
590 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
591 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
592 			| 2 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
593 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
594 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
595 
596 		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
597 		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS,
598 			UVD_PGFSM_STATUS__UVDM_UVDU_UVDLM_PWR_ON_3_0, 0x3F3FFFFF);
599 	} else {
600 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
601 			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
602 			| 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
603 			| 1 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
604 			| 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
605 			| 1 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
606 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
607 			| 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
608 			| 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
609 			| 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
610 			| 1 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
611 			| 1 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
612 			| 1 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
613 			| 1 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
614 		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
615 		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS, 0,  0x3F3FFFFF);
616 	}
617 
618 	data = RREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS);
619 	data &= ~0x103;
620 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
621 		data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
622 			UVD_POWER_STATUS__UVD_PG_EN_MASK;
623 
624 	WREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS, data);
625 }
626 
627 static void vcn_v3_0_enable_static_power_gating(struct amdgpu_device *adev, int inst)
628 {
629 	uint32_t data;
630 
631 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
632 		/* Before power off, this indicator has to be turned on */
633 		data = RREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS);
634 		data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
635 		data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
636 		WREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS, data);
637 
638 		data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
639 			| 2 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
640 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
641 			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
642 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
643 			| 2 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
644 			| 2 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
645 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
646 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
647 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
648 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
649 			| 2 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
650 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
651 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
652 		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
653 
654 		data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
655 			| 2 << UVD_PGFSM_STATUS__UVDU_PWR_STATUS__SHIFT
656 			| 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
657 			| 2 << UVD_PGFSM_STATUS__UVDC_PWR_STATUS__SHIFT
658 			| 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
659 			| 2 << UVD_PGFSM_STATUS__UVDIRL_PWR_STATUS__SHIFT
660 			| 2 << UVD_PGFSM_STATUS__UVDLM_PWR_STATUS__SHIFT
661 			| 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
662 			| 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
663 			| 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT
664 			| 2 << UVD_PGFSM_STATUS__UVDAB_PWR_STATUS__SHIFT
665 			| 2 << UVD_PGFSM_STATUS__UVDATD_PWR_STATUS__SHIFT
666 			| 2 << UVD_PGFSM_STATUS__UVDNA_PWR_STATUS__SHIFT
667 			| 2 << UVD_PGFSM_STATUS__UVDNB_PWR_STATUS__SHIFT);
668 		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS, data, 0x3F3FFFFF);
669 	}
670 }
671 
672 /**
673  * vcn_v3_0_disable_clock_gating - disable VCN clock gating
674  *
675  * @adev: amdgpu_device pointer
676  * @inst: instance number
677  *
678  * Disable clock gating for VCN block
679  */
680 static void vcn_v3_0_disable_clock_gating(struct amdgpu_device *adev, int inst)
681 {
682 	uint32_t data;
683 
684 	/* VCN disable CGC */
685 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
686 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
687 		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
688 	else
689 		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
690 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
691 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
692 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
693 
694 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_GATE);
695 	data &= ~(UVD_CGC_GATE__SYS_MASK
696 		| UVD_CGC_GATE__UDEC_MASK
697 		| UVD_CGC_GATE__MPEG2_MASK
698 		| UVD_CGC_GATE__REGS_MASK
699 		| UVD_CGC_GATE__RBC_MASK
700 		| UVD_CGC_GATE__LMI_MC_MASK
701 		| UVD_CGC_GATE__LMI_UMC_MASK
702 		| UVD_CGC_GATE__IDCT_MASK
703 		| UVD_CGC_GATE__MPRD_MASK
704 		| UVD_CGC_GATE__MPC_MASK
705 		| UVD_CGC_GATE__LBSI_MASK
706 		| UVD_CGC_GATE__LRBBM_MASK
707 		| UVD_CGC_GATE__UDEC_RE_MASK
708 		| UVD_CGC_GATE__UDEC_CM_MASK
709 		| UVD_CGC_GATE__UDEC_IT_MASK
710 		| UVD_CGC_GATE__UDEC_DB_MASK
711 		| UVD_CGC_GATE__UDEC_MP_MASK
712 		| UVD_CGC_GATE__WCB_MASK
713 		| UVD_CGC_GATE__VCPU_MASK
714 		| UVD_CGC_GATE__MMSCH_MASK);
715 
716 	WREG32_SOC15(VCN, inst, mmUVD_CGC_GATE, data);
717 
718 	SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_CGC_GATE, 0,  0xFFFFFFFF);
719 
720 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
721 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
722 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
723 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
724 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
725 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
726 		| UVD_CGC_CTRL__SYS_MODE_MASK
727 		| UVD_CGC_CTRL__UDEC_MODE_MASK
728 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
729 		| UVD_CGC_CTRL__REGS_MODE_MASK
730 		| UVD_CGC_CTRL__RBC_MODE_MASK
731 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
732 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
733 		| UVD_CGC_CTRL__IDCT_MODE_MASK
734 		| UVD_CGC_CTRL__MPRD_MODE_MASK
735 		| UVD_CGC_CTRL__MPC_MODE_MASK
736 		| UVD_CGC_CTRL__LBSI_MODE_MASK
737 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
738 		| UVD_CGC_CTRL__WCB_MODE_MASK
739 		| UVD_CGC_CTRL__VCPU_MODE_MASK
740 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
741 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
742 
743 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE);
744 	data |= (UVD_SUVD_CGC_GATE__SRE_MASK
745 		| UVD_SUVD_CGC_GATE__SIT_MASK
746 		| UVD_SUVD_CGC_GATE__SMP_MASK
747 		| UVD_SUVD_CGC_GATE__SCM_MASK
748 		| UVD_SUVD_CGC_GATE__SDB_MASK
749 		| UVD_SUVD_CGC_GATE__SRE_H264_MASK
750 		| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
751 		| UVD_SUVD_CGC_GATE__SIT_H264_MASK
752 		| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
753 		| UVD_SUVD_CGC_GATE__SCM_H264_MASK
754 		| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
755 		| UVD_SUVD_CGC_GATE__SDB_H264_MASK
756 		| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
757 		| UVD_SUVD_CGC_GATE__SCLR_MASK
758 		| UVD_SUVD_CGC_GATE__ENT_MASK
759 		| UVD_SUVD_CGC_GATE__IME_MASK
760 		| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
761 		| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
762 		| UVD_SUVD_CGC_GATE__SITE_MASK
763 		| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
764 		| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
765 		| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
766 		| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
767 		| UVD_SUVD_CGC_GATE__IME_HEVC_MASK
768 		| UVD_SUVD_CGC_GATE__EFC_MASK
769 		| UVD_SUVD_CGC_GATE__SAOE_MASK
770 		| UVD_SUVD_CGC_GATE__SRE_AV1_MASK
771 		| UVD_SUVD_CGC_GATE__FBC_PCLK_MASK
772 		| UVD_SUVD_CGC_GATE__FBC_CCLK_MASK
773 		| UVD_SUVD_CGC_GATE__SCM_AV1_MASK
774 		| UVD_SUVD_CGC_GATE__SMPA_MASK);
775 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE, data);
776 
777 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE2);
778 	data |= (UVD_SUVD_CGC_GATE2__MPBE0_MASK
779 		| UVD_SUVD_CGC_GATE2__MPBE1_MASK
780 		| UVD_SUVD_CGC_GATE2__SIT_AV1_MASK
781 		| UVD_SUVD_CGC_GATE2__SDB_AV1_MASK
782 		| UVD_SUVD_CGC_GATE2__MPC1_MASK);
783 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE2, data);
784 
785 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL);
786 	data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
787 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
788 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
789 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
790 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
791 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
792 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
793 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
794 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK
795 		| UVD_SUVD_CGC_CTRL__EFC_MODE_MASK
796 		| UVD_SUVD_CGC_CTRL__SAOE_MODE_MASK
797 		| UVD_SUVD_CGC_CTRL__SMPA_MODE_MASK
798 		| UVD_SUVD_CGC_CTRL__MPBE0_MODE_MASK
799 		| UVD_SUVD_CGC_CTRL__MPBE1_MODE_MASK
800 		| UVD_SUVD_CGC_CTRL__SIT_AV1_MODE_MASK
801 		| UVD_SUVD_CGC_CTRL__SDB_AV1_MODE_MASK
802 		| UVD_SUVD_CGC_CTRL__MPC1_MODE_MASK
803 		| UVD_SUVD_CGC_CTRL__FBC_PCLK_MASK
804 		| UVD_SUVD_CGC_CTRL__FBC_CCLK_MASK);
805 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL, data);
806 }
807 
808 static void vcn_v3_0_clock_gating_dpg_mode(struct amdgpu_device *adev,
809 		uint8_t sram_sel, int inst_idx, uint8_t indirect)
810 {
811 	uint32_t reg_data = 0;
812 
813 	/* enable sw clock gating control */
814 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
815 		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
816 	else
817 		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
818 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
819 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
820 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
821 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
822 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
823 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
824 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
825 		 UVD_CGC_CTRL__SYS_MODE_MASK |
826 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
827 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
828 		 UVD_CGC_CTRL__REGS_MODE_MASK |
829 		 UVD_CGC_CTRL__RBC_MODE_MASK |
830 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
831 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
832 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
833 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
834 		 UVD_CGC_CTRL__MPC_MODE_MASK |
835 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
836 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
837 		 UVD_CGC_CTRL__WCB_MODE_MASK |
838 		 UVD_CGC_CTRL__VCPU_MODE_MASK |
839 		 UVD_CGC_CTRL__MMSCH_MODE_MASK);
840 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
841 		VCN, inst_idx, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
842 
843 	/* turn off clock gating */
844 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
845 		VCN, inst_idx, mmUVD_CGC_GATE), 0, sram_sel, indirect);
846 
847 	/* turn on SUVD clock gating */
848 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
849 		VCN, inst_idx, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
850 
851 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
852 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
853 		VCN, inst_idx, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
854 }
855 
856 /**
857  * vcn_v3_0_enable_clock_gating - enable VCN clock gating
858  *
859  * @adev: amdgpu_device pointer
860  * @inst: instance number
861  *
862  * Enable clock gating for VCN block
863  */
864 static void vcn_v3_0_enable_clock_gating(struct amdgpu_device *adev, int inst)
865 {
866 	uint32_t data;
867 
868 	/* enable VCN CGC */
869 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
870 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
871 		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
872 	else
873 		data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
874 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
875 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
876 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
877 
878 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
879 	data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
880 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
881 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
882 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
883 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
884 		| UVD_CGC_CTRL__SYS_MODE_MASK
885 		| UVD_CGC_CTRL__UDEC_MODE_MASK
886 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
887 		| UVD_CGC_CTRL__REGS_MODE_MASK
888 		| UVD_CGC_CTRL__RBC_MODE_MASK
889 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
890 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
891 		| UVD_CGC_CTRL__IDCT_MODE_MASK
892 		| UVD_CGC_CTRL__MPRD_MODE_MASK
893 		| UVD_CGC_CTRL__MPC_MODE_MASK
894 		| UVD_CGC_CTRL__LBSI_MODE_MASK
895 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
896 		| UVD_CGC_CTRL__WCB_MODE_MASK
897 		| UVD_CGC_CTRL__VCPU_MODE_MASK
898 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
899 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
900 
901 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL);
902 	data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
903 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
904 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
905 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
906 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
907 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
908 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
909 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
910 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK
911 		| UVD_SUVD_CGC_CTRL__EFC_MODE_MASK
912 		| UVD_SUVD_CGC_CTRL__SAOE_MODE_MASK
913 		| UVD_SUVD_CGC_CTRL__SMPA_MODE_MASK
914 		| UVD_SUVD_CGC_CTRL__MPBE0_MODE_MASK
915 		| UVD_SUVD_CGC_CTRL__MPBE1_MODE_MASK
916 		| UVD_SUVD_CGC_CTRL__SIT_AV1_MODE_MASK
917 		| UVD_SUVD_CGC_CTRL__SDB_AV1_MODE_MASK
918 		| UVD_SUVD_CGC_CTRL__MPC1_MODE_MASK
919 		| UVD_SUVD_CGC_CTRL__FBC_PCLK_MASK
920 		| UVD_SUVD_CGC_CTRL__FBC_CCLK_MASK);
921 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL, data);
922 }
923 
924 static int vcn_v3_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
925 {
926 	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared_cpu_addr;
927 	struct amdgpu_ring *ring;
928 	uint32_t rb_bufsz, tmp;
929 
930 	/* disable register anti-hang mechanism */
931 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1,
932 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
933 	/* enable dynamic power gating mode */
934 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS);
935 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
936 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
937 	WREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS, tmp);
938 
939 	if (indirect)
940 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
941 
942 	/* enable clock gating */
943 	vcn_v3_0_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
944 
945 	/* enable VCPU clock */
946 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
947 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
948 	tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
949 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
950 		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
951 
952 	/* disable master interupt */
953 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
954 		VCN, inst_idx, mmUVD_MASTINT_EN), 0, 0, indirect);
955 
956 	/* setup mmUVD_LMI_CTRL */
957 	tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
958 		UVD_LMI_CTRL__REQ_MODE_MASK |
959 		UVD_LMI_CTRL__CRC_RESET_MASK |
960 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
961 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
962 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
963 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
964 		0x00100000L);
965 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
966 		VCN, inst_idx, mmUVD_LMI_CTRL), tmp, 0, indirect);
967 
968 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
969 		VCN, inst_idx, mmUVD_MPC_CNTL),
970 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
971 
972 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
973 		VCN, inst_idx, mmUVD_MPC_SET_MUXA0),
974 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
975 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
976 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
977 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
978 
979 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
980 		VCN, inst_idx, mmUVD_MPC_SET_MUXB0),
981 		 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
982 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
983 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
984 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
985 
986 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
987 		VCN, inst_idx, mmUVD_MPC_SET_MUX),
988 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
989 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
990 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
991 
992 	vcn_v3_0_mc_resume_dpg_mode(adev, inst_idx, indirect);
993 
994 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
995 		VCN, inst_idx, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
996 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
997 		VCN, inst_idx, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
998 
999 	/* enable LMI MC and UMC channels */
1000 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1001 		VCN, inst_idx, mmUVD_LMI_CTRL2), 0, 0, indirect);
1002 
1003 	/* unblock VCPU register access */
1004 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1005 		VCN, inst_idx, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
1006 
1007 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
1008 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
1009 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1010 		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
1011 
1012 	/* enable master interrupt */
1013 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1014 		VCN, inst_idx, mmUVD_MASTINT_EN),
1015 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
1016 
1017 	/* add nop to workaround PSP size check */
1018 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1019 		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
1020 
1021 	if (indirect)
1022 		psp_update_vcn_sram(adev, inst_idx, adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
1023 			(uint32_t)((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
1024 				(uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr));
1025 
1026 	ring = &adev->vcn.inst[inst_idx].ring_dec;
1027 	/* force RBC into idle state */
1028 	rb_bufsz = order_base_2(ring->ring_size);
1029 	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1030 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1031 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1032 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1033 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1034 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
1035 
1036 	/* Stall DPG before WPTR/RPTR reset */
1037 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1038 		UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1039 		~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1040 	fw_shared->multi_queue.decode_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1041 
1042 	/* set the write pointer delay */
1043 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
1044 
1045 	/* set the wb address */
1046 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR_ADDR,
1047 		(upper_32_bits(ring->gpu_addr) >> 2));
1048 
1049 	/* programm the RB_BASE for ring buffer */
1050 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1051 		lower_32_bits(ring->gpu_addr));
1052 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1053 		upper_32_bits(ring->gpu_addr));
1054 
1055 	/* Initialize the ring buffer's read and write pointers */
1056 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, 0);
1057 
1058 	WREG32_SOC15(VCN, inst_idx, mmUVD_SCRATCH2, 0);
1059 
1060 	ring->wptr = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR);
1061 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR,
1062 		lower_32_bits(ring->wptr));
1063 
1064 	/* Reset FW shared memory RBC WPTR/RPTR */
1065 	fw_shared->rb.rptr = 0;
1066 	fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1067 
1068 	/*resetting done, fw can check RB ring */
1069 	fw_shared->multi_queue.decode_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1070 
1071 	/* Unstall DPG */
1072 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1073 		0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1074 
1075 	return 0;
1076 }
1077 
1078 static int vcn_v3_0_start(struct amdgpu_device *adev)
1079 {
1080 	volatile struct amdgpu_fw_shared *fw_shared;
1081 	struct amdgpu_ring *ring;
1082 	uint32_t rb_bufsz, tmp;
1083 	int i, j, k, r;
1084 
1085 	if (adev->pm.dpm_enabled)
1086 		amdgpu_dpm_enable_uvd(adev, true);
1087 
1088 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1089 		if (adev->vcn.harvest_config & (1 << i))
1090 			continue;
1091 
1092 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG){
1093 			r = vcn_v3_0_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
1094 			continue;
1095 		}
1096 
1097 		/* disable VCN power gating */
1098 		vcn_v3_0_disable_static_power_gating(adev, i);
1099 
1100 		/* set VCN status busy */
1101 		tmp = RREG32_SOC15(VCN, i, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY;
1102 		WREG32_SOC15(VCN, i, mmUVD_STATUS, tmp);
1103 
1104 		/*SW clock gating */
1105 		vcn_v3_0_disable_clock_gating(adev, i);
1106 
1107 		/* enable VCPU clock */
1108 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1109 			UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1110 
1111 		/* disable master interrupt */
1112 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN), 0,
1113 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1114 
1115 		/* enable LMI MC and UMC channels */
1116 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_LMI_CTRL2), 0,
1117 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1118 
1119 		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1120 		tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1121 		tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1122 		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1123 
1124 		/* setup mmUVD_LMI_CTRL */
1125 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL);
1126 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL, tmp |
1127 			UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK	|
1128 			UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1129 			UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1130 			UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1131 
1132 		/* setup mmUVD_MPC_CNTL */
1133 		tmp = RREG32_SOC15(VCN, i, mmUVD_MPC_CNTL);
1134 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1135 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1136 		WREG32_SOC15(VCN, i, mmUVD_MPC_CNTL, tmp);
1137 
1138 		/* setup UVD_MPC_SET_MUXA0 */
1139 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXA0,
1140 			((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1141 			(0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1142 			(0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1143 			(0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1144 
1145 		/* setup UVD_MPC_SET_MUXB0 */
1146 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXB0,
1147 			((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1148 			(0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1149 			(0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1150 			(0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1151 
1152 		/* setup mmUVD_MPC_SET_MUX */
1153 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUX,
1154 			((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1155 			(0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1156 			(0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1157 
1158 		vcn_v3_0_mc_resume(adev, i);
1159 
1160 		/* VCN global tiling registers */
1161 		WREG32_SOC15(VCN, i, mmUVD_GFX10_ADDR_CONFIG,
1162 			adev->gfx.config.gb_addr_config);
1163 
1164 		/* unblock VCPU register access */
1165 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL), 0,
1166 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1167 
1168 		/* release VCPU reset to boot */
1169 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1170 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1171 
1172 		for (j = 0; j < 10; ++j) {
1173 			uint32_t status;
1174 
1175 			for (k = 0; k < 100; ++k) {
1176 				status = RREG32_SOC15(VCN, i, mmUVD_STATUS);
1177 				if (status & 2)
1178 					break;
1179 				mdelay(10);
1180 			}
1181 			r = 0;
1182 			if (status & 2)
1183 				break;
1184 
1185 			DRM_ERROR("VCN[%d] decode not responding, trying to reset the VCPU!!!\n", i);
1186 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1187 				UVD_VCPU_CNTL__BLK_RST_MASK,
1188 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1189 			mdelay(10);
1190 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1191 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1192 
1193 			mdelay(10);
1194 			r = -1;
1195 		}
1196 
1197 		if (r) {
1198 			DRM_ERROR("VCN[%d] decode not responding, giving up!!!\n", i);
1199 			return r;
1200 		}
1201 
1202 		/* enable master interrupt */
1203 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN),
1204 			UVD_MASTINT_EN__VCPU_EN_MASK,
1205 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1206 
1207 		/* clear the busy bit of VCN_STATUS */
1208 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS), 0,
1209 			~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1210 
1211 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_VMID, 0);
1212 
1213 		ring = &adev->vcn.inst[i].ring_dec;
1214 		/* force RBC into idle state */
1215 		rb_bufsz = order_base_2(ring->ring_size);
1216 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1217 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1218 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1219 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1220 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1221 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_CNTL, tmp);
1222 
1223 		fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
1224 		fw_shared->multi_queue.decode_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1225 
1226 		/* programm the RB_BASE for ring buffer */
1227 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1228 			lower_32_bits(ring->gpu_addr));
1229 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1230 			upper_32_bits(ring->gpu_addr));
1231 
1232 		/* Initialize the ring buffer's read and write pointers */
1233 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR, 0);
1234 
1235 		WREG32_SOC15(VCN, i, mmUVD_SCRATCH2, 0);
1236 		ring->wptr = RREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR);
1237 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_WPTR,
1238 			lower_32_bits(ring->wptr));
1239 		fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1240 		fw_shared->multi_queue.decode_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1241 
1242 		if (adev->ip_versions[UVD_HWIP][0] != IP_VERSION(3, 0, 33)) {
1243 			fw_shared->multi_queue.encode_generalpurpose_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1244 			ring = &adev->vcn.inst[i].ring_enc[0];
1245 			WREG32_SOC15(VCN, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1246 			WREG32_SOC15(VCN, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1247 			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO, ring->gpu_addr);
1248 			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1249 			WREG32_SOC15(VCN, i, mmUVD_RB_SIZE, ring->ring_size / 4);
1250 			fw_shared->multi_queue.encode_generalpurpose_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1251 
1252 			fw_shared->multi_queue.encode_lowlatency_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1253 			ring = &adev->vcn.inst[i].ring_enc[1];
1254 			WREG32_SOC15(VCN, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1255 			WREG32_SOC15(VCN, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1256 			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1257 			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1258 			WREG32_SOC15(VCN, i, mmUVD_RB_SIZE2, ring->ring_size / 4);
1259 			fw_shared->multi_queue.encode_lowlatency_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1260 		}
1261 	}
1262 
1263 	return 0;
1264 }
1265 
1266 static int vcn_v3_0_start_sriov(struct amdgpu_device *adev)
1267 {
1268 	int i, j;
1269 	struct amdgpu_ring *ring;
1270 	uint64_t cache_addr;
1271 	uint64_t rb_addr;
1272 	uint64_t ctx_addr;
1273 	uint32_t param, resp, expected;
1274 	uint32_t offset, cache_size;
1275 	uint32_t tmp, timeout;
1276 
1277 	struct amdgpu_mm_table *table = &adev->virt.mm_table;
1278 	uint32_t *table_loc;
1279 	uint32_t table_size;
1280 	uint32_t size, size_dw;
1281 
1282 	struct mmsch_v3_0_cmd_direct_write
1283 		direct_wt = { {0} };
1284 	struct mmsch_v3_0_cmd_direct_read_modify_write
1285 		direct_rd_mod_wt = { {0} };
1286 	struct mmsch_v3_0_cmd_end end = { {0} };
1287 	struct mmsch_v3_0_init_header header;
1288 
1289 	direct_wt.cmd_header.command_type =
1290 		MMSCH_COMMAND__DIRECT_REG_WRITE;
1291 	direct_rd_mod_wt.cmd_header.command_type =
1292 		MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1293 	end.cmd_header.command_type =
1294 		MMSCH_COMMAND__END;
1295 
1296 	header.version = MMSCH_VERSION;
1297 	header.total_size = sizeof(struct mmsch_v3_0_init_header) >> 2;
1298 	for (i = 0; i < AMDGPU_MAX_VCN_INSTANCES; i++) {
1299 		header.inst[i].init_status = 0;
1300 		header.inst[i].table_offset = 0;
1301 		header.inst[i].table_size = 0;
1302 	}
1303 
1304 	table_loc = (uint32_t *)table->cpu_addr;
1305 	table_loc += header.total_size;
1306 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1307 		if (adev->vcn.harvest_config & (1 << i))
1308 			continue;
1309 
1310 		table_size = 0;
1311 
1312 		MMSCH_V3_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(VCN, i,
1313 			mmUVD_STATUS),
1314 			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1315 
1316 		cache_size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1317 
1318 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1319 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1320 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1321 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1322 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1323 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1324 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1325 			offset = 0;
1326 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1327 				mmUVD_VCPU_CACHE_OFFSET0),
1328 				0);
1329 		} else {
1330 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1331 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1332 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1333 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1334 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1335 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1336 			offset = cache_size;
1337 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1338 				mmUVD_VCPU_CACHE_OFFSET0),
1339 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1340 		}
1341 
1342 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1343 			mmUVD_VCPU_CACHE_SIZE0),
1344 			cache_size);
1345 
1346 		cache_addr = adev->vcn.inst[i].gpu_addr + offset;
1347 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1348 			mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1349 			lower_32_bits(cache_addr));
1350 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1351 			mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1352 			upper_32_bits(cache_addr));
1353 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1354 			mmUVD_VCPU_CACHE_OFFSET1),
1355 			0);
1356 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1357 			mmUVD_VCPU_CACHE_SIZE1),
1358 			AMDGPU_VCN_STACK_SIZE);
1359 
1360 		cache_addr = adev->vcn.inst[i].gpu_addr + offset +
1361 			AMDGPU_VCN_STACK_SIZE;
1362 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1363 			mmUVD_LMI_VCPU_CACHE2_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_CACHE2_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_OFFSET2),
1370 			0);
1371 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1372 			mmUVD_VCPU_CACHE_SIZE2),
1373 			AMDGPU_VCN_CONTEXT_SIZE);
1374 
1375 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
1376 			ring = &adev->vcn.inst[i].ring_enc[j];
1377 			ring->wptr = 0;
1378 			rb_addr = ring->gpu_addr;
1379 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1380 				mmUVD_RB_BASE_LO),
1381 				lower_32_bits(rb_addr));
1382 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1383 				mmUVD_RB_BASE_HI),
1384 				upper_32_bits(rb_addr));
1385 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1386 				mmUVD_RB_SIZE),
1387 				ring->ring_size / 4);
1388 		}
1389 
1390 		ring = &adev->vcn.inst[i].ring_dec;
1391 		ring->wptr = 0;
1392 		rb_addr = ring->gpu_addr;
1393 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1394 			mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1395 			lower_32_bits(rb_addr));
1396 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1397 			mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1398 			upper_32_bits(rb_addr));
1399 		/* force RBC into idle state */
1400 		tmp = order_base_2(ring->ring_size);
1401 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, tmp);
1402 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1403 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1404 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1405 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1406 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1407 			mmUVD_RBC_RB_CNTL),
1408 			tmp);
1409 
1410 		/* add end packet */
1411 		MMSCH_V3_0_INSERT_END();
1412 
1413 		/* refine header */
1414 		header.inst[i].init_status = 0;
1415 		header.inst[i].table_offset = header.total_size;
1416 		header.inst[i].table_size = table_size;
1417 		header.total_size += table_size;
1418 	}
1419 
1420 	/* Update init table header in memory */
1421 	size = sizeof(struct mmsch_v3_0_init_header);
1422 	table_loc = (uint32_t *)table->cpu_addr;
1423 	memcpy((void *)table_loc, &header, size);
1424 
1425 	/* message MMSCH (in VCN[0]) to initialize this client
1426 	 * 1, write to mmsch_vf_ctx_addr_lo/hi register with GPU mc addr
1427 	 * of memory descriptor location
1428 	 */
1429 	ctx_addr = table->gpu_addr;
1430 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(ctx_addr));
1431 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(ctx_addr));
1432 
1433 	/* 2, update vmid of descriptor */
1434 	tmp = RREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID);
1435 	tmp &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1436 	/* use domain0 for MM scheduler */
1437 	tmp |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1438 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID, tmp);
1439 
1440 	/* 3, notify mmsch about the size of this descriptor */
1441 	size = header.total_size;
1442 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_SIZE, size);
1443 
1444 	/* 4, set resp to zero */
1445 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1446 
1447 	/* 5, kick off the initialization and wait until
1448 	 * MMSCH_VF_MAILBOX_RESP becomes non-zero
1449 	 */
1450 	param = 0x10000001;
1451 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_HOST, param);
1452 	tmp = 0;
1453 	timeout = 1000;
1454 	resp = 0;
1455 	expected = param + 1;
1456 	while (resp != expected) {
1457 		resp = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1458 		if (resp == expected)
1459 			break;
1460 
1461 		udelay(10);
1462 		tmp = tmp + 10;
1463 		if (tmp >= timeout) {
1464 			DRM_ERROR("failed to init MMSCH. TIME-OUT after %d usec"\
1465 				" waiting for mmMMSCH_VF_MAILBOX_RESP "\
1466 				"(expected=0x%08x, readback=0x%08x)\n",
1467 				tmp, expected, resp);
1468 			return -EBUSY;
1469 		}
1470 	}
1471 
1472 	return 0;
1473 }
1474 
1475 static int vcn_v3_0_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1476 {
1477 	uint32_t tmp;
1478 
1479 	/* Wait for power status to be 1 */
1480 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1481 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1482 
1483 	/* wait for read ptr to be equal to write ptr */
1484 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR);
1485 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1486 
1487 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2);
1488 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
1489 
1490 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1491 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF);
1492 
1493 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1494 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1495 
1496 	/* disable dynamic power gating mode */
1497 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 0,
1498 		~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1499 
1500 	return 0;
1501 }
1502 
1503 static int vcn_v3_0_stop(struct amdgpu_device *adev)
1504 {
1505 	uint32_t tmp;
1506 	int i, r = 0;
1507 
1508 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1509 		if (adev->vcn.harvest_config & (1 << i))
1510 			continue;
1511 
1512 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1513 			r = vcn_v3_0_stop_dpg_mode(adev, i);
1514 			continue;
1515 		}
1516 
1517 		/* wait for vcn idle */
1518 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1519 		if (r)
1520 			return r;
1521 
1522 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1523 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1524 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1525 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1526 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1527 		if (r)
1528 			return r;
1529 
1530 		/* disable LMI UMC channel */
1531 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2);
1532 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1533 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2, tmp);
1534 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1535 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1536 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1537 		if (r)
1538 			return r;
1539 
1540 		/* block VCPU register access */
1541 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL),
1542 			UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1543 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1544 
1545 		/* reset VCPU */
1546 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1547 			UVD_VCPU_CNTL__BLK_RST_MASK,
1548 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1549 
1550 		/* disable VCPU clock */
1551 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1552 			~(UVD_VCPU_CNTL__CLK_EN_MASK));
1553 
1554 		/* apply soft reset */
1555 		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1556 		tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1557 		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1558 		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1559 		tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1560 		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1561 
1562 		/* clear status */
1563 		WREG32_SOC15(VCN, i, mmUVD_STATUS, 0);
1564 
1565 		/* apply HW clock gating */
1566 		vcn_v3_0_enable_clock_gating(adev, i);
1567 
1568 		/* enable VCN power gating */
1569 		vcn_v3_0_enable_static_power_gating(adev, i);
1570 	}
1571 
1572 	if (adev->pm.dpm_enabled)
1573 		amdgpu_dpm_enable_uvd(adev, false);
1574 
1575 	return 0;
1576 }
1577 
1578 static int vcn_v3_0_pause_dpg_mode(struct amdgpu_device *adev,
1579 		   int inst_idx, struct dpg_pause_state *new_state)
1580 {
1581 	volatile struct amdgpu_fw_shared *fw_shared;
1582 	struct amdgpu_ring *ring;
1583 	uint32_t reg_data = 0;
1584 	int ret_code;
1585 
1586 	/* pause/unpause if state is changed */
1587 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1588 		DRM_DEBUG("dpg pause state changed %d -> %d",
1589 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1590 		reg_data = RREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE) &
1591 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1592 
1593 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1594 			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1595 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1596 
1597 			if (!ret_code) {
1598 				/* pause DPG */
1599 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1600 				WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1601 
1602 				/* wait for ACK */
1603 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_DPG_PAUSE,
1604 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1605 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1606 
1607 				/* Stall DPG before WPTR/RPTR reset */
1608 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1609 					UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1610 					~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1611 
1612 				if (adev->ip_versions[UVD_HWIP][0] != IP_VERSION(3, 0, 33)) {
1613 					/* Restore */
1614 					fw_shared = adev->vcn.inst[inst_idx].fw_shared_cpu_addr;
1615 					fw_shared->multi_queue.encode_generalpurpose_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1616 					ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1617 					ring->wptr = 0;
1618 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
1619 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1620 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
1621 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1622 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1623 					fw_shared->multi_queue.encode_generalpurpose_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1624 
1625 					fw_shared->multi_queue.encode_lowlatency_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1626 					ring = &adev->vcn.inst[inst_idx].ring_enc[1];
1627 					ring->wptr = 0;
1628 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1629 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1630 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
1631 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1632 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1633 					fw_shared->multi_queue.encode_lowlatency_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1634 
1635 					/* restore wptr/rptr with pointers saved in FW shared memory*/
1636 					WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, fw_shared->rb.rptr);
1637 					WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR, fw_shared->rb.wptr);
1638 				}
1639 
1640 				/* Unstall DPG */
1641 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1642 					0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1643 
1644 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS,
1645 					UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1646 			}
1647 		} else {
1648 			/* unpause dpg, no need to wait */
1649 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1650 			WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1651 		}
1652 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1653 	}
1654 
1655 	return 0;
1656 }
1657 
1658 /**
1659  * vcn_v3_0_dec_ring_get_rptr - get read pointer
1660  *
1661  * @ring: amdgpu_ring pointer
1662  *
1663  * Returns the current hardware read pointer
1664  */
1665 static uint64_t vcn_v3_0_dec_ring_get_rptr(struct amdgpu_ring *ring)
1666 {
1667 	struct amdgpu_device *adev = ring->adev;
1668 
1669 	return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_RPTR);
1670 }
1671 
1672 /**
1673  * vcn_v3_0_dec_ring_get_wptr - get write pointer
1674  *
1675  * @ring: amdgpu_ring pointer
1676  *
1677  * Returns the current hardware write pointer
1678  */
1679 static uint64_t vcn_v3_0_dec_ring_get_wptr(struct amdgpu_ring *ring)
1680 {
1681 	struct amdgpu_device *adev = ring->adev;
1682 
1683 	if (ring->use_doorbell)
1684 		return adev->wb.wb[ring->wptr_offs];
1685 	else
1686 		return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR);
1687 }
1688 
1689 /**
1690  * vcn_v3_0_dec_ring_set_wptr - set write pointer
1691  *
1692  * @ring: amdgpu_ring pointer
1693  *
1694  * Commits the write pointer to the hardware
1695  */
1696 static void vcn_v3_0_dec_ring_set_wptr(struct amdgpu_ring *ring)
1697 {
1698 	struct amdgpu_device *adev = ring->adev;
1699 	volatile struct amdgpu_fw_shared *fw_shared;
1700 
1701 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1702 		/*whenever update RBC_RB_WPTR, we save the wptr in shared rb.wptr and scratch2 */
1703 		fw_shared = adev->vcn.inst[ring->me].fw_shared_cpu_addr;
1704 		fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1705 		WREG32_SOC15(VCN, ring->me, mmUVD_SCRATCH2,
1706 			lower_32_bits(ring->wptr));
1707 	}
1708 
1709 	if (ring->use_doorbell) {
1710 		adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1711 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1712 	} else {
1713 		WREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1714 	}
1715 }
1716 
1717 static void vcn_v3_0_dec_sw_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
1718 				u64 seq, uint32_t flags)
1719 {
1720 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
1721 
1722 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_FENCE);
1723 	amdgpu_ring_write(ring, addr);
1724 	amdgpu_ring_write(ring, upper_32_bits(addr));
1725 	amdgpu_ring_write(ring, seq);
1726 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_TRAP);
1727 }
1728 
1729 static void vcn_v3_0_dec_sw_ring_insert_end(struct amdgpu_ring *ring)
1730 {
1731 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_END);
1732 }
1733 
1734 static void vcn_v3_0_dec_sw_ring_emit_ib(struct amdgpu_ring *ring,
1735 			       struct amdgpu_job *job,
1736 			       struct amdgpu_ib *ib,
1737 			       uint32_t flags)
1738 {
1739 	uint32_t vmid = AMDGPU_JOB_GET_VMID(job);
1740 
1741 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_IB);
1742 	amdgpu_ring_write(ring, vmid);
1743 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
1744 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
1745 	amdgpu_ring_write(ring, ib->length_dw);
1746 }
1747 
1748 static void vcn_v3_0_dec_sw_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
1749 				uint32_t val, uint32_t mask)
1750 {
1751 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_REG_WAIT);
1752 	amdgpu_ring_write(ring, reg << 2);
1753 	amdgpu_ring_write(ring, mask);
1754 	amdgpu_ring_write(ring, val);
1755 }
1756 
1757 static void vcn_v3_0_dec_sw_ring_emit_vm_flush(struct amdgpu_ring *ring,
1758 				uint32_t vmid, uint64_t pd_addr)
1759 {
1760 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
1761 	uint32_t data0, data1, mask;
1762 
1763 	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1764 
1765 	/* wait for register write */
1766 	data0 = hub->ctx0_ptb_addr_lo32 + vmid * hub->ctx_addr_distance;
1767 	data1 = lower_32_bits(pd_addr);
1768 	mask = 0xffffffff;
1769 	vcn_v3_0_dec_sw_ring_emit_reg_wait(ring, data0, data1, mask);
1770 }
1771 
1772 static void vcn_v3_0_dec_sw_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg, uint32_t val)
1773 {
1774 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_REG_WRITE);
1775 	amdgpu_ring_write(ring,	reg << 2);
1776 	amdgpu_ring_write(ring, val);
1777 }
1778 
1779 static const struct amdgpu_ring_funcs vcn_v3_0_dec_sw_ring_vm_funcs = {
1780 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1781 	.align_mask = 0x3f,
1782 	.nop = VCN_DEC_SW_CMD_NO_OP,
1783 	.vmhub = AMDGPU_MMHUB_0,
1784 	.get_rptr = vcn_v3_0_dec_ring_get_rptr,
1785 	.get_wptr = vcn_v3_0_dec_ring_get_wptr,
1786 	.set_wptr = vcn_v3_0_dec_ring_set_wptr,
1787 	.emit_frame_size =
1788 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1789 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1790 		4 + /* vcn_v3_0_dec_sw_ring_emit_vm_flush */
1791 		5 + 5 + /* vcn_v3_0_dec_sw_ring_emit_fdec_swe x2 vm fdec_swe */
1792 		1, /* vcn_v3_0_dec_sw_ring_insert_end */
1793 	.emit_ib_size = 5, /* vcn_v3_0_dec_sw_ring_emit_ib */
1794 	.emit_ib = vcn_v3_0_dec_sw_ring_emit_ib,
1795 	.emit_fence = vcn_v3_0_dec_sw_ring_emit_fence,
1796 	.emit_vm_flush = vcn_v3_0_dec_sw_ring_emit_vm_flush,
1797 	.test_ring = amdgpu_vcn_dec_sw_ring_test_ring,
1798 	.test_ib = NULL,//amdgpu_vcn_dec_sw_ring_test_ib,
1799 	.insert_nop = amdgpu_ring_insert_nop,
1800 	.insert_end = vcn_v3_0_dec_sw_ring_insert_end,
1801 	.pad_ib = amdgpu_ring_generic_pad_ib,
1802 	.begin_use = amdgpu_vcn_ring_begin_use,
1803 	.end_use = amdgpu_vcn_ring_end_use,
1804 	.emit_wreg = vcn_v3_0_dec_sw_ring_emit_wreg,
1805 	.emit_reg_wait = vcn_v3_0_dec_sw_ring_emit_reg_wait,
1806 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1807 };
1808 
1809 static int vcn_v3_0_limit_sched(struct amdgpu_cs_parser *p)
1810 {
1811 	struct drm_gpu_scheduler **scheds;
1812 
1813 	/* The create msg must be in the first IB submitted */
1814 	if (atomic_read(&p->entity->fence_seq))
1815 		return -EINVAL;
1816 
1817 	scheds = p->adev->gpu_sched[AMDGPU_HW_IP_VCN_DEC]
1818 		[AMDGPU_RING_PRIO_DEFAULT].sched;
1819 	drm_sched_entity_modify_sched(p->entity, scheds, 1);
1820 	return 0;
1821 }
1822 
1823 static int vcn_v3_0_dec_msg(struct amdgpu_cs_parser *p, uint64_t addr)
1824 {
1825 	struct ttm_operation_ctx ctx = { false, false };
1826 	struct amdgpu_bo_va_mapping *map;
1827 	uint32_t *msg, num_buffers;
1828 	struct amdgpu_bo *bo;
1829 	uint64_t start, end;
1830 	unsigned int i;
1831 	void * ptr;
1832 	int r;
1833 
1834 	addr &= AMDGPU_GMC_HOLE_MASK;
1835 	r = amdgpu_cs_find_mapping(p, addr, &bo, &map);
1836 	if (r) {
1837 		DRM_ERROR("Can't find BO for addr 0x%08Lx\n", addr);
1838 		return r;
1839 	}
1840 
1841 	start = map->start * AMDGPU_GPU_PAGE_SIZE;
1842 	end = (map->last + 1) * AMDGPU_GPU_PAGE_SIZE;
1843 	if (addr & 0x7) {
1844 		DRM_ERROR("VCN messages must be 8 byte aligned!\n");
1845 		return -EINVAL;
1846 	}
1847 
1848 	bo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
1849 	amdgpu_bo_placement_from_domain(bo, bo->allowed_domains);
1850 	r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
1851 	if (r) {
1852 		DRM_ERROR("Failed validating the VCN message BO (%d)!\n", r);
1853 		return r;
1854 	}
1855 
1856 	r = amdgpu_bo_kmap(bo, &ptr);
1857 	if (r) {
1858 		DRM_ERROR("Failed mapping the VCN message (%d)!\n", r);
1859 		return r;
1860 	}
1861 
1862 	msg = ptr + addr - start;
1863 
1864 	/* Check length */
1865 	if (msg[1] > end - addr) {
1866 		r = -EINVAL;
1867 		goto out;
1868 	}
1869 
1870 	if (msg[3] != RDECODE_MSG_CREATE)
1871 		goto out;
1872 
1873 	num_buffers = msg[2];
1874 	for (i = 0, msg = &msg[6]; i < num_buffers; ++i, msg += 4) {
1875 		uint32_t offset, size, *create;
1876 
1877 		if (msg[0] != RDECODE_MESSAGE_CREATE)
1878 			continue;
1879 
1880 		offset = msg[1];
1881 		size = msg[2];
1882 
1883 		if (offset + size > end) {
1884 			r = -EINVAL;
1885 			goto out;
1886 		}
1887 
1888 		create = ptr + addr + offset - start;
1889 
1890 		/* H246, HEVC and VP9 can run on any instance */
1891 		if (create[0] == 0x7 || create[0] == 0x10 || create[0] == 0x11)
1892 			continue;
1893 
1894 		r = vcn_v3_0_limit_sched(p);
1895 		if (r)
1896 			goto out;
1897 	}
1898 
1899 out:
1900 	amdgpu_bo_kunmap(bo);
1901 	return r;
1902 }
1903 
1904 static int vcn_v3_0_ring_patch_cs_in_place(struct amdgpu_cs_parser *p,
1905 					   uint32_t ib_idx)
1906 {
1907 	struct amdgpu_ring *ring = to_amdgpu_ring(p->entity->rq->sched);
1908 	struct amdgpu_ib *ib = &p->job->ibs[ib_idx];
1909 	uint32_t msg_lo = 0, msg_hi = 0;
1910 	unsigned i;
1911 	int r;
1912 
1913 	/* The first instance can decode anything */
1914 	if (!ring->me)
1915 		return 0;
1916 
1917 	for (i = 0; i < ib->length_dw; i += 2) {
1918 		uint32_t reg = amdgpu_get_ib_value(p, ib_idx, i);
1919 		uint32_t val = amdgpu_get_ib_value(p, ib_idx, i + 1);
1920 
1921 		if (reg == PACKET0(p->adev->vcn.internal.data0, 0)) {
1922 			msg_lo = val;
1923 		} else if (reg == PACKET0(p->adev->vcn.internal.data1, 0)) {
1924 			msg_hi = val;
1925 		} else if (reg == PACKET0(p->adev->vcn.internal.cmd, 0) &&
1926 			   val == 0) {
1927 			r = vcn_v3_0_dec_msg(p, ((u64)msg_hi) << 32 | msg_lo);
1928 			if (r)
1929 				return r;
1930 		}
1931 	}
1932 	return 0;
1933 }
1934 
1935 static const struct amdgpu_ring_funcs vcn_v3_0_dec_ring_vm_funcs = {
1936 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1937 	.align_mask = 0xf,
1938 	.vmhub = AMDGPU_MMHUB_0,
1939 	.get_rptr = vcn_v3_0_dec_ring_get_rptr,
1940 	.get_wptr = vcn_v3_0_dec_ring_get_wptr,
1941 	.set_wptr = vcn_v3_0_dec_ring_set_wptr,
1942 	.patch_cs_in_place = vcn_v3_0_ring_patch_cs_in_place,
1943 	.emit_frame_size =
1944 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1945 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1946 		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1947 		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1948 		6,
1949 	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1950 	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
1951 	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
1952 	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1953 	.test_ring = vcn_v2_0_dec_ring_test_ring,
1954 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
1955 	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
1956 	.insert_start = vcn_v2_0_dec_ring_insert_start,
1957 	.insert_end = vcn_v2_0_dec_ring_insert_end,
1958 	.pad_ib = amdgpu_ring_generic_pad_ib,
1959 	.begin_use = amdgpu_vcn_ring_begin_use,
1960 	.end_use = amdgpu_vcn_ring_end_use,
1961 	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1962 	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1963 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1964 };
1965 
1966 /**
1967  * vcn_v3_0_enc_ring_get_rptr - get enc read pointer
1968  *
1969  * @ring: amdgpu_ring pointer
1970  *
1971  * Returns the current hardware enc read pointer
1972  */
1973 static uint64_t vcn_v3_0_enc_ring_get_rptr(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 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR);
1979 	else
1980 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR2);
1981 }
1982 
1983 /**
1984  * vcn_v3_0_enc_ring_get_wptr - get enc write pointer
1985  *
1986  * @ring: amdgpu_ring pointer
1987  *
1988  * Returns the current hardware enc write pointer
1989  */
1990 static uint64_t vcn_v3_0_enc_ring_get_wptr(struct amdgpu_ring *ring)
1991 {
1992 	struct amdgpu_device *adev = ring->adev;
1993 
1994 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1995 		if (ring->use_doorbell)
1996 			return adev->wb.wb[ring->wptr_offs];
1997 		else
1998 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR);
1999 	} else {
2000 		if (ring->use_doorbell)
2001 			return adev->wb.wb[ring->wptr_offs];
2002 		else
2003 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2);
2004 	}
2005 }
2006 
2007 /**
2008  * vcn_v3_0_enc_ring_set_wptr - set enc write pointer
2009  *
2010  * @ring: amdgpu_ring pointer
2011  *
2012  * Commits the enc write pointer to the hardware
2013  */
2014 static void vcn_v3_0_enc_ring_set_wptr(struct amdgpu_ring *ring)
2015 {
2016 	struct amdgpu_device *adev = ring->adev;
2017 
2018 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
2019 		if (ring->use_doorbell) {
2020 			adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
2021 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
2022 		} else {
2023 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
2024 		}
2025 	} else {
2026 		if (ring->use_doorbell) {
2027 			adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
2028 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
2029 		} else {
2030 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
2031 		}
2032 	}
2033 }
2034 
2035 static const struct amdgpu_ring_funcs vcn_v3_0_enc_ring_vm_funcs = {
2036 	.type = AMDGPU_RING_TYPE_VCN_ENC,
2037 	.align_mask = 0x3f,
2038 	.nop = VCN_ENC_CMD_NO_OP,
2039 	.vmhub = AMDGPU_MMHUB_0,
2040 	.get_rptr = vcn_v3_0_enc_ring_get_rptr,
2041 	.get_wptr = vcn_v3_0_enc_ring_get_wptr,
2042 	.set_wptr = vcn_v3_0_enc_ring_set_wptr,
2043 	.emit_frame_size =
2044 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2045 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
2046 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
2047 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
2048 		1, /* vcn_v2_0_enc_ring_insert_end */
2049 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
2050 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
2051 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
2052 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
2053 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
2054 	.test_ib = amdgpu_vcn_enc_ring_test_ib,
2055 	.insert_nop = amdgpu_ring_insert_nop,
2056 	.insert_end = vcn_v2_0_enc_ring_insert_end,
2057 	.pad_ib = amdgpu_ring_generic_pad_ib,
2058 	.begin_use = amdgpu_vcn_ring_begin_use,
2059 	.end_use = amdgpu_vcn_ring_end_use,
2060 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
2061 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
2062 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2063 };
2064 
2065 static void vcn_v3_0_set_dec_ring_funcs(struct amdgpu_device *adev)
2066 {
2067 	int i;
2068 
2069 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2070 		if (adev->vcn.harvest_config & (1 << i))
2071 			continue;
2072 
2073 		if (!DEC_SW_RING_ENABLED)
2074 			adev->vcn.inst[i].ring_dec.funcs = &vcn_v3_0_dec_ring_vm_funcs;
2075 		else
2076 			adev->vcn.inst[i].ring_dec.funcs = &vcn_v3_0_dec_sw_ring_vm_funcs;
2077 		adev->vcn.inst[i].ring_dec.me = i;
2078 		DRM_INFO("VCN(%d) decode%s is enabled in VM mode\n", i,
2079 			  DEC_SW_RING_ENABLED?"(Software Ring)":"");
2080 	}
2081 }
2082 
2083 static void vcn_v3_0_set_enc_ring_funcs(struct amdgpu_device *adev)
2084 {
2085 	int i, j;
2086 
2087 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2088 		if (adev->vcn.harvest_config & (1 << i))
2089 			continue;
2090 
2091 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
2092 			adev->vcn.inst[i].ring_enc[j].funcs = &vcn_v3_0_enc_ring_vm_funcs;
2093 			adev->vcn.inst[i].ring_enc[j].me = i;
2094 		}
2095 		if (adev->vcn.num_enc_rings > 0)
2096 			DRM_INFO("VCN(%d) encode is enabled in VM mode\n", i);
2097 	}
2098 }
2099 
2100 static bool vcn_v3_0_is_idle(void *handle)
2101 {
2102 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2103 	int i, ret = 1;
2104 
2105 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2106 		if (adev->vcn.harvest_config & (1 << i))
2107 			continue;
2108 
2109 		ret &= (RREG32_SOC15(VCN, i, mmUVD_STATUS) == UVD_STATUS__IDLE);
2110 	}
2111 
2112 	return ret;
2113 }
2114 
2115 static int vcn_v3_0_wait_for_idle(void *handle)
2116 {
2117 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2118 	int i, ret = 0;
2119 
2120 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2121 		if (adev->vcn.harvest_config & (1 << i))
2122 			continue;
2123 
2124 		ret = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE,
2125 			UVD_STATUS__IDLE);
2126 		if (ret)
2127 			return ret;
2128 	}
2129 
2130 	return ret;
2131 }
2132 
2133 static int vcn_v3_0_set_clockgating_state(void *handle,
2134 					  enum amd_clockgating_state state)
2135 {
2136 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2137 	bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
2138 	int i;
2139 
2140 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2141 		if (adev->vcn.harvest_config & (1 << i))
2142 			continue;
2143 
2144 		if (enable) {
2145 			if (RREG32_SOC15(VCN, i, mmUVD_STATUS) != UVD_STATUS__IDLE)
2146 				return -EBUSY;
2147 			vcn_v3_0_enable_clock_gating(adev, i);
2148 		} else {
2149 			vcn_v3_0_disable_clock_gating(adev, i);
2150 		}
2151 	}
2152 
2153 	return 0;
2154 }
2155 
2156 static int vcn_v3_0_set_powergating_state(void *handle,
2157 					  enum amd_powergating_state state)
2158 {
2159 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2160 	int ret;
2161 
2162 	/* for SRIOV, guest should not control VCN Power-gating
2163 	 * MMSCH FW should control Power-gating and clock-gating
2164 	 * guest should avoid touching CGC and PG
2165 	 */
2166 	if (amdgpu_sriov_vf(adev)) {
2167 		adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
2168 		return 0;
2169 	}
2170 
2171 	if(state == adev->vcn.cur_state)
2172 		return 0;
2173 
2174 	if (state == AMD_PG_STATE_GATE)
2175 		ret = vcn_v3_0_stop(adev);
2176 	else
2177 		ret = vcn_v3_0_start(adev);
2178 
2179 	if(!ret)
2180 		adev->vcn.cur_state = state;
2181 
2182 	return ret;
2183 }
2184 
2185 static int vcn_v3_0_set_interrupt_state(struct amdgpu_device *adev,
2186 					struct amdgpu_irq_src *source,
2187 					unsigned type,
2188 					enum amdgpu_interrupt_state state)
2189 {
2190 	return 0;
2191 }
2192 
2193 static int vcn_v3_0_process_interrupt(struct amdgpu_device *adev,
2194 				      struct amdgpu_irq_src *source,
2195 				      struct amdgpu_iv_entry *entry)
2196 {
2197 	uint32_t ip_instance;
2198 
2199 	switch (entry->client_id) {
2200 	case SOC15_IH_CLIENTID_VCN:
2201 		ip_instance = 0;
2202 		break;
2203 	case SOC15_IH_CLIENTID_VCN1:
2204 		ip_instance = 1;
2205 		break;
2206 	default:
2207 		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
2208 		return 0;
2209 	}
2210 
2211 	DRM_DEBUG("IH: VCN TRAP\n");
2212 
2213 	switch (entry->src_id) {
2214 	case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
2215 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec);
2216 		break;
2217 	case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
2218 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
2219 		break;
2220 	case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
2221 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]);
2222 		break;
2223 	default:
2224 		DRM_ERROR("Unhandled interrupt: %d %d\n",
2225 			  entry->src_id, entry->src_data[0]);
2226 		break;
2227 	}
2228 
2229 	return 0;
2230 }
2231 
2232 static const struct amdgpu_irq_src_funcs vcn_v3_0_irq_funcs = {
2233 	.set = vcn_v3_0_set_interrupt_state,
2234 	.process = vcn_v3_0_process_interrupt,
2235 };
2236 
2237 static void vcn_v3_0_set_irq_funcs(struct amdgpu_device *adev)
2238 {
2239 	int i;
2240 
2241 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2242 		if (adev->vcn.harvest_config & (1 << i))
2243 			continue;
2244 
2245 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
2246 		adev->vcn.inst[i].irq.funcs = &vcn_v3_0_irq_funcs;
2247 	}
2248 }
2249 
2250 static const struct amd_ip_funcs vcn_v3_0_ip_funcs = {
2251 	.name = "vcn_v3_0",
2252 	.early_init = vcn_v3_0_early_init,
2253 	.late_init = NULL,
2254 	.sw_init = vcn_v3_0_sw_init,
2255 	.sw_fini = vcn_v3_0_sw_fini,
2256 	.hw_init = vcn_v3_0_hw_init,
2257 	.hw_fini = vcn_v3_0_hw_fini,
2258 	.suspend = vcn_v3_0_suspend,
2259 	.resume = vcn_v3_0_resume,
2260 	.is_idle = vcn_v3_0_is_idle,
2261 	.wait_for_idle = vcn_v3_0_wait_for_idle,
2262 	.check_soft_reset = NULL,
2263 	.pre_soft_reset = NULL,
2264 	.soft_reset = NULL,
2265 	.post_soft_reset = NULL,
2266 	.set_clockgating_state = vcn_v3_0_set_clockgating_state,
2267 	.set_powergating_state = vcn_v3_0_set_powergating_state,
2268 };
2269 
2270 const struct amdgpu_ip_block_version vcn_v3_0_ip_block =
2271 {
2272 	.type = AMD_IP_BLOCK_TYPE_VCN,
2273 	.major = 3,
2274 	.minor = 0,
2275 	.rev = 0,
2276 	.funcs = &vcn_v3_0_ip_funcs,
2277 };
2278