xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c (revision 2208f39c)
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 #define mmUVD_CONTEXT_ID_INTERNAL_OFFSET			0x27
38 #define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET			0x0f
39 #define mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET			0x10
40 #define mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET			0x11
41 #define mmUVD_NO_OP_INTERNAL_OFFSET				0x29
42 #define mmUVD_GP_SCRATCH8_INTERNAL_OFFSET			0x66
43 #define mmUVD_SCRATCH9_INTERNAL_OFFSET				0xc01d
44 
45 #define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET			0x431
46 #define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET		0x3b4
47 #define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET		0x3b5
48 #define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET			0x25c
49 
50 #define VCN_INSTANCES_SIENNA_CICHLID				2
51 #define DEC_SW_RING_ENABLED					FALSE
52 
53 static int amdgpu_ih_clientid_vcns[] = {
54 	SOC15_IH_CLIENTID_VCN,
55 	SOC15_IH_CLIENTID_VCN1
56 };
57 
58 static int amdgpu_ucode_id_vcns[] = {
59 	AMDGPU_UCODE_ID_VCN,
60 	AMDGPU_UCODE_ID_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->asic_type == CHIP_SIENNA_CICHLID) {
93 			u32 harvest;
94 			int i;
95 
96 			adev->vcn.num_vcn_inst = VCN_INSTANCES_SIENNA_CICHLID;
97 			for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
98 				harvest = RREG32_SOC15(VCN, i, mmCC_UVD_HARVESTING);
99 				if (harvest & CC_UVD_HARVESTING__UVD_DISABLE_MASK)
100 					adev->vcn.harvest_config |= 1 << i;
101 			}
102 
103 			if (adev->vcn.harvest_config == (AMDGPU_VCN_HARVEST_VCN0 |
104 						AMDGPU_VCN_HARVEST_VCN1))
105 				/* both instances are harvested, disable the block */
106 				return -ENOENT;
107 		} else
108 			adev->vcn.num_vcn_inst = 1;
109 
110 		adev->vcn.num_enc_rings = 2;
111 	}
112 
113 	vcn_v3_0_set_dec_ring_funcs(adev);
114 	vcn_v3_0_set_enc_ring_funcs(adev);
115 	vcn_v3_0_set_irq_funcs(adev);
116 
117 	return 0;
118 }
119 
120 /**
121  * vcn_v3_0_sw_init - sw init for VCN block
122  *
123  * @handle: amdgpu_device pointer
124  *
125  * Load firmware and sw initialization
126  */
127 static int vcn_v3_0_sw_init(void *handle)
128 {
129 	struct amdgpu_ring *ring;
130 	int i, j, r;
131 	int vcn_doorbell_index = 0;
132 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
133 
134 	r = amdgpu_vcn_sw_init(adev);
135 	if (r)
136 		return r;
137 
138 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
139 		const struct common_firmware_header *hdr;
140 		hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
141 		adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].ucode_id = AMDGPU_UCODE_ID_VCN;
142 		adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].fw = adev->vcn.fw;
143 		adev->firmware.fw_size +=
144 			ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
145 
146 		if (adev->vcn.num_vcn_inst == VCN_INSTANCES_SIENNA_CICHLID) {
147 			adev->firmware.ucode[AMDGPU_UCODE_ID_VCN1].ucode_id = AMDGPU_UCODE_ID_VCN1;
148 			adev->firmware.ucode[AMDGPU_UCODE_ID_VCN1].fw = adev->vcn.fw;
149 			adev->firmware.fw_size +=
150 				ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
151 		}
152 		DRM_INFO("PSP loading VCN firmware\n");
153 	}
154 
155 	r = amdgpu_vcn_resume(adev);
156 	if (r)
157 		return r;
158 
159 	/*
160 	 * Note: doorbell assignment is fixed for SRIOV multiple VCN engines
161 	 * Formula:
162 	 *   vcn_db_base  = adev->doorbell_index.vcn.vcn_ring0_1 << 1;
163 	 *   dec_ring_i   = vcn_db_base + i * (adev->vcn.num_enc_rings + 1)
164 	 *   enc_ring_i,j = vcn_db_base + i * (adev->vcn.num_enc_rings + 1) + 1 + j
165 	 */
166 	if (amdgpu_sriov_vf(adev)) {
167 		vcn_doorbell_index = adev->doorbell_index.vcn.vcn_ring0_1;
168 		/* get DWORD offset */
169 		vcn_doorbell_index = vcn_doorbell_index << 1;
170 	}
171 
172 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
173 		volatile struct amdgpu_fw_shared *fw_shared;
174 		if (adev->vcn.harvest_config & (1 << i))
175 			continue;
176 
177 		adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
178 		adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
179 		adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
180 		adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
181 		adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
182 		adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
183 
184 		adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
185 		adev->vcn.inst[i].external.scratch9 = SOC15_REG_OFFSET(VCN, i, mmUVD_SCRATCH9);
186 		adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
187 		adev->vcn.inst[i].external.data0 = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_DATA0);
188 		adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
189 		adev->vcn.inst[i].external.data1 = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_DATA1);
190 		adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
191 		adev->vcn.inst[i].external.cmd = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_CMD);
192 		adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
193 		adev->vcn.inst[i].external.nop = SOC15_REG_OFFSET(VCN, i, mmUVD_NO_OP);
194 
195 		/* VCN DEC TRAP */
196 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
197 				VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst[i].irq);
198 		if (r)
199 			return r;
200 
201 		ring = &adev->vcn.inst[i].ring_dec;
202 		ring->use_doorbell = true;
203 		if (amdgpu_sriov_vf(adev)) {
204 			ring->doorbell_index = vcn_doorbell_index + i * (adev->vcn.num_enc_rings + 1);
205 		} else {
206 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i;
207 		}
208 		if (adev->asic_type == CHIP_SIENNA_CICHLID && i != 0)
209 			ring->no_scheduler = true;
210 		sprintf(ring->name, "vcn_dec_%d", i);
211 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
212 				     AMDGPU_RING_PRIO_DEFAULT);
213 		if (r)
214 			return r;
215 
216 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
217 			/* VCN ENC TRAP */
218 			r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
219 				j + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);
220 			if (r)
221 				return r;
222 
223 			ring = &adev->vcn.inst[i].ring_enc[j];
224 			ring->use_doorbell = true;
225 			if (amdgpu_sriov_vf(adev)) {
226 				ring->doorbell_index = vcn_doorbell_index + i * (adev->vcn.num_enc_rings + 1) + 1 + j;
227 			} else {
228 				ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + j + 8 * i;
229 			}
230 			if (adev->asic_type == CHIP_SIENNA_CICHLID && i != 1)
231 				ring->no_scheduler = true;
232 			sprintf(ring->name, "vcn_enc_%d.%d", i, j);
233 			r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
234 					     AMDGPU_RING_PRIO_DEFAULT);
235 			if (r)
236 				return r;
237 		}
238 
239 		fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
240 		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_SW_RING_FLAG);
241 		fw_shared->sw_ring.is_enabled = cpu_to_le32(DEC_SW_RING_ENABLED);
242 	}
243 
244 	if (amdgpu_sriov_vf(adev)) {
245 		r = amdgpu_virt_alloc_mm_table(adev);
246 		if (r)
247 			return r;
248 	}
249 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
250 		adev->vcn.pause_dpg_mode = vcn_v3_0_pause_dpg_mode;
251 
252 	return 0;
253 }
254 
255 /**
256  * vcn_v3_0_sw_fini - sw fini for VCN block
257  *
258  * @handle: amdgpu_device pointer
259  *
260  * VCN suspend and free up sw allocation
261  */
262 static int vcn_v3_0_sw_fini(void *handle)
263 {
264 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
265 	int i, r;
266 
267 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
268 		volatile struct amdgpu_fw_shared *fw_shared;
269 
270 		if (adev->vcn.harvest_config & (1 << i))
271 			continue;
272 		fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
273 		fw_shared->present_flag_0 = 0;
274 		fw_shared->sw_ring.is_enabled = false;
275 	}
276 
277 	if (amdgpu_sriov_vf(adev))
278 		amdgpu_virt_free_mm_table(adev);
279 
280 	r = amdgpu_vcn_suspend(adev);
281 	if (r)
282 		return r;
283 
284 	r = amdgpu_vcn_sw_fini(adev);
285 
286 	return r;
287 }
288 
289 /**
290  * vcn_v3_0_hw_init - start and test VCN block
291  *
292  * @handle: amdgpu_device pointer
293  *
294  * Initialize the hardware, boot up the VCPU and do some testing
295  */
296 static int vcn_v3_0_hw_init(void *handle)
297 {
298 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
299 	struct amdgpu_ring *ring;
300 	int i, j, r;
301 
302 	if (amdgpu_sriov_vf(adev)) {
303 		r = vcn_v3_0_start_sriov(adev);
304 		if (r)
305 			goto done;
306 
307 		/* initialize VCN dec and enc ring buffers */
308 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
309 			if (adev->vcn.harvest_config & (1 << i))
310 				continue;
311 
312 			ring = &adev->vcn.inst[i].ring_dec;
313 			if (ring->sched.ready) {
314 				ring->wptr = 0;
315 				ring->wptr_old = 0;
316 				vcn_v3_0_dec_ring_set_wptr(ring);
317 			}
318 
319 			for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
320 				ring = &adev->vcn.inst[i].ring_enc[j];
321 				if (ring->sched.ready) {
322 					ring->wptr = 0;
323 					ring->wptr_old = 0;
324 					vcn_v3_0_enc_ring_set_wptr(ring);
325 				}
326 			}
327 		}
328 	} else {
329 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
330 			if (adev->vcn.harvest_config & (1 << i))
331 				continue;
332 
333 			ring = &adev->vcn.inst[i].ring_dec;
334 
335 			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
336 						     ring->doorbell_index, i);
337 
338 			r = amdgpu_ring_test_helper(ring);
339 			if (r)
340 				goto done;
341 
342 			for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
343 				ring = &adev->vcn.inst[i].ring_enc[j];
344 				r = amdgpu_ring_test_helper(ring);
345 				if (r)
346 					goto done;
347 			}
348 		}
349 	}
350 
351 done:
352 	if (!r)
353 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
354 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
355 
356 	return r;
357 }
358 
359 /**
360  * vcn_v3_0_hw_fini - stop the hardware block
361  *
362  * @handle: amdgpu_device pointer
363  *
364  * Stop the VCN block, mark ring as not ready any more
365  */
366 static int vcn_v3_0_hw_fini(void *handle)
367 {
368 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
369 	struct amdgpu_ring *ring;
370 	int i, j;
371 
372 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
373 		if (adev->vcn.harvest_config & (1 << i))
374 			continue;
375 
376 		ring = &adev->vcn.inst[i].ring_dec;
377 
378 		if (!amdgpu_sriov_vf(adev)) {
379 			if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
380 					(adev->vcn.cur_state != AMD_PG_STATE_GATE &&
381 					 RREG32_SOC15(VCN, i, mmUVD_STATUS))) {
382 				vcn_v3_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
383 			}
384 		}
385 		ring->sched.ready = false;
386 
387 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
388 			ring = &adev->vcn.inst[i].ring_enc[j];
389 			ring->sched.ready = false;
390 		}
391 	}
392 
393 	return 0;
394 }
395 
396 /**
397  * vcn_v3_0_suspend - suspend VCN block
398  *
399  * @handle: amdgpu_device pointer
400  *
401  * HW fini and suspend VCN block
402  */
403 static int vcn_v3_0_suspend(void *handle)
404 {
405 	int r;
406 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
407 
408 	r = vcn_v3_0_hw_fini(adev);
409 	if (r)
410 		return r;
411 
412 	r = amdgpu_vcn_suspend(adev);
413 
414 	return r;
415 }
416 
417 /**
418  * vcn_v3_0_resume - resume VCN block
419  *
420  * @handle: amdgpu_device pointer
421  *
422  * Resume firmware and hw init VCN block
423  */
424 static int vcn_v3_0_resume(void *handle)
425 {
426 	int r;
427 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
428 
429 	r = amdgpu_vcn_resume(adev);
430 	if (r)
431 		return r;
432 
433 	r = vcn_v3_0_hw_init(adev);
434 
435 	return r;
436 }
437 
438 /**
439  * vcn_v3_0_mc_resume - memory controller programming
440  *
441  * @adev: amdgpu_device pointer
442  * @inst: instance number
443  *
444  * Let the VCN memory controller know it's offsets
445  */
446 static void vcn_v3_0_mc_resume(struct amdgpu_device *adev, int inst)
447 {
448 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
449 	uint32_t offset;
450 
451 	/* cache window 0: fw */
452 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
453 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
454 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo));
455 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
456 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi));
457 		WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET0, 0);
458 		offset = 0;
459 	} else {
460 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
461 			lower_32_bits(adev->vcn.inst[inst].gpu_addr));
462 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
463 			upper_32_bits(adev->vcn.inst[inst].gpu_addr));
464 		offset = size;
465 		WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET0,
466 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
467 	}
468 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE0, size);
469 
470 	/* cache window 1: stack */
471 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
472 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
473 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
474 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
475 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET1, 0);
476 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
477 
478 	/* cache window 2: context */
479 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
480 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
481 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
482 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
483 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET2, 0);
484 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
485 
486 	/* non-cache window */
487 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
488 		lower_32_bits(adev->vcn.inst[inst].fw_shared_gpu_addr));
489 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
490 		upper_32_bits(adev->vcn.inst[inst].fw_shared_gpu_addr));
491 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
492 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_NONCACHE_SIZE0,
493 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
494 }
495 
496 static void vcn_v3_0_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
497 {
498 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
499 	uint32_t offset;
500 
501 	/* cache window 0: fw */
502 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
503 		if (!indirect) {
504 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
505 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
506 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
507 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
508 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
509 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
510 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
511 				VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
512 		} else {
513 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
514 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
515 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
516 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
517 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
518 				VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
519 		}
520 		offset = 0;
521 	} else {
522 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
523 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
524 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
525 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
526 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
527 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
528 		offset = size;
529 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
530 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0),
531 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
532 	}
533 
534 	if (!indirect)
535 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
536 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
537 	else
538 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
539 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
540 
541 	/* cache window 1: stack */
542 	if (!indirect) {
543 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
544 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
545 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
546 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
547 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
548 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
549 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
550 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
551 	} else {
552 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
553 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
554 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
555 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
556 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
557 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
558 	}
559 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
560 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
561 
562 	/* cache window 2: context */
563 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
564 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
565 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
566 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
567 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
568 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
569 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
570 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
571 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
572 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
573 
574 	/* non-cache window */
575 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
576 			VCN, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
577 			lower_32_bits(adev->vcn.inst[inst_idx].fw_shared_gpu_addr), 0, indirect);
578 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
579 			VCN, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
580 			upper_32_bits(adev->vcn.inst[inst_idx].fw_shared_gpu_addr), 0, indirect);
581 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
582 			VCN, inst_idx, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
583 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
584 			VCN, inst_idx, mmUVD_VCPU_NONCACHE_SIZE0),
585 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
586 }
587 
588 static void vcn_v3_0_disable_static_power_gating(struct amdgpu_device *adev, int inst)
589 {
590 	uint32_t data = 0;
591 
592 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
593 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
594 			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
595 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
596 			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
597 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
598 			| 2 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
599 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
600 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
601 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
602 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
603 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
604 			| 2 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
605 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
606 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
607 
608 		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
609 		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS,
610 			UVD_PGFSM_STATUS__UVDM_UVDU_UVDLM_PWR_ON_3_0, 0x3F3FFFFF);
611 	} else {
612 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
613 			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
614 			| 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
615 			| 1 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
616 			| 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
617 			| 1 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
618 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
619 			| 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
620 			| 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
621 			| 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
622 			| 1 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
623 			| 1 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
624 			| 1 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
625 			| 1 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
626 		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
627 		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS, 0,  0x3F3FFFFF);
628 	}
629 
630 	data = RREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS);
631 	data &= ~0x103;
632 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
633 		data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
634 			UVD_POWER_STATUS__UVD_PG_EN_MASK;
635 
636 	WREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS, data);
637 }
638 
639 static void vcn_v3_0_enable_static_power_gating(struct amdgpu_device *adev, int inst)
640 {
641 	uint32_t data;
642 
643 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
644 		/* Before power off, this indicator has to be turned on */
645 		data = RREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS);
646 		data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
647 		data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
648 		WREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS, data);
649 
650 		data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
651 			| 2 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
652 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
653 			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
654 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
655 			| 2 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
656 			| 2 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
657 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
658 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
659 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
660 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
661 			| 2 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
662 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
663 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
664 		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
665 
666 		data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
667 			| 2 << UVD_PGFSM_STATUS__UVDU_PWR_STATUS__SHIFT
668 			| 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
669 			| 2 << UVD_PGFSM_STATUS__UVDC_PWR_STATUS__SHIFT
670 			| 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
671 			| 2 << UVD_PGFSM_STATUS__UVDIRL_PWR_STATUS__SHIFT
672 			| 2 << UVD_PGFSM_STATUS__UVDLM_PWR_STATUS__SHIFT
673 			| 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
674 			| 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
675 			| 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT
676 			| 2 << UVD_PGFSM_STATUS__UVDAB_PWR_STATUS__SHIFT
677 			| 2 << UVD_PGFSM_STATUS__UVDATD_PWR_STATUS__SHIFT
678 			| 2 << UVD_PGFSM_STATUS__UVDNA_PWR_STATUS__SHIFT
679 			| 2 << UVD_PGFSM_STATUS__UVDNB_PWR_STATUS__SHIFT);
680 		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS, data, 0x3F3FFFFF);
681 	}
682 }
683 
684 /**
685  * vcn_v3_0_disable_clock_gating - disable VCN clock gating
686  *
687  * @adev: amdgpu_device pointer
688  * @inst: instance number
689  *
690  * Disable clock gating for VCN block
691  */
692 static void vcn_v3_0_disable_clock_gating(struct amdgpu_device *adev, int inst)
693 {
694 	uint32_t data;
695 
696 	/* VCN disable CGC */
697 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
698 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
699 		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
700 	else
701 		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
702 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
703 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
704 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
705 
706 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_GATE);
707 	data &= ~(UVD_CGC_GATE__SYS_MASK
708 		| UVD_CGC_GATE__UDEC_MASK
709 		| UVD_CGC_GATE__MPEG2_MASK
710 		| UVD_CGC_GATE__REGS_MASK
711 		| UVD_CGC_GATE__RBC_MASK
712 		| UVD_CGC_GATE__LMI_MC_MASK
713 		| UVD_CGC_GATE__LMI_UMC_MASK
714 		| UVD_CGC_GATE__IDCT_MASK
715 		| UVD_CGC_GATE__MPRD_MASK
716 		| UVD_CGC_GATE__MPC_MASK
717 		| UVD_CGC_GATE__LBSI_MASK
718 		| UVD_CGC_GATE__LRBBM_MASK
719 		| UVD_CGC_GATE__UDEC_RE_MASK
720 		| UVD_CGC_GATE__UDEC_CM_MASK
721 		| UVD_CGC_GATE__UDEC_IT_MASK
722 		| UVD_CGC_GATE__UDEC_DB_MASK
723 		| UVD_CGC_GATE__UDEC_MP_MASK
724 		| UVD_CGC_GATE__WCB_MASK
725 		| UVD_CGC_GATE__VCPU_MASK
726 		| UVD_CGC_GATE__MMSCH_MASK);
727 
728 	WREG32_SOC15(VCN, inst, mmUVD_CGC_GATE, data);
729 
730 	SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_CGC_GATE, 0,  0xFFFFFFFF);
731 
732 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
733 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
734 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
735 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
736 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
737 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
738 		| UVD_CGC_CTRL__SYS_MODE_MASK
739 		| UVD_CGC_CTRL__UDEC_MODE_MASK
740 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
741 		| UVD_CGC_CTRL__REGS_MODE_MASK
742 		| UVD_CGC_CTRL__RBC_MODE_MASK
743 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
744 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
745 		| UVD_CGC_CTRL__IDCT_MODE_MASK
746 		| UVD_CGC_CTRL__MPRD_MODE_MASK
747 		| UVD_CGC_CTRL__MPC_MODE_MASK
748 		| UVD_CGC_CTRL__LBSI_MODE_MASK
749 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
750 		| UVD_CGC_CTRL__WCB_MODE_MASK
751 		| UVD_CGC_CTRL__VCPU_MODE_MASK
752 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
753 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
754 
755 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE);
756 	data |= (UVD_SUVD_CGC_GATE__SRE_MASK
757 		| UVD_SUVD_CGC_GATE__SIT_MASK
758 		| UVD_SUVD_CGC_GATE__SMP_MASK
759 		| UVD_SUVD_CGC_GATE__SCM_MASK
760 		| UVD_SUVD_CGC_GATE__SDB_MASK
761 		| UVD_SUVD_CGC_GATE__SRE_H264_MASK
762 		| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
763 		| UVD_SUVD_CGC_GATE__SIT_H264_MASK
764 		| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
765 		| UVD_SUVD_CGC_GATE__SCM_H264_MASK
766 		| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
767 		| UVD_SUVD_CGC_GATE__SDB_H264_MASK
768 		| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
769 		| UVD_SUVD_CGC_GATE__SCLR_MASK
770 		| UVD_SUVD_CGC_GATE__ENT_MASK
771 		| UVD_SUVD_CGC_GATE__IME_MASK
772 		| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
773 		| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
774 		| UVD_SUVD_CGC_GATE__SITE_MASK
775 		| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
776 		| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
777 		| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
778 		| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
779 		| UVD_SUVD_CGC_GATE__IME_HEVC_MASK
780 		| UVD_SUVD_CGC_GATE__EFC_MASK
781 		| UVD_SUVD_CGC_GATE__SAOE_MASK
782 		| UVD_SUVD_CGC_GATE__SRE_AV1_MASK
783 		| UVD_SUVD_CGC_GATE__FBC_PCLK_MASK
784 		| UVD_SUVD_CGC_GATE__FBC_CCLK_MASK
785 		| UVD_SUVD_CGC_GATE__SCM_AV1_MASK
786 		| UVD_SUVD_CGC_GATE__SMPA_MASK);
787 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE, data);
788 
789 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE2);
790 	data |= (UVD_SUVD_CGC_GATE2__MPBE0_MASK
791 		| UVD_SUVD_CGC_GATE2__MPBE1_MASK
792 		| UVD_SUVD_CGC_GATE2__SIT_AV1_MASK
793 		| UVD_SUVD_CGC_GATE2__SDB_AV1_MASK
794 		| UVD_SUVD_CGC_GATE2__MPC1_MASK);
795 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE2, data);
796 
797 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL);
798 	data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
799 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
800 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
801 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
802 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
803 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
804 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
805 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
806 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK
807 		| UVD_SUVD_CGC_CTRL__EFC_MODE_MASK
808 		| UVD_SUVD_CGC_CTRL__SAOE_MODE_MASK
809 		| UVD_SUVD_CGC_CTRL__SMPA_MODE_MASK
810 		| UVD_SUVD_CGC_CTRL__MPBE0_MODE_MASK
811 		| UVD_SUVD_CGC_CTRL__MPBE1_MODE_MASK
812 		| UVD_SUVD_CGC_CTRL__SIT_AV1_MODE_MASK
813 		| UVD_SUVD_CGC_CTRL__SDB_AV1_MODE_MASK
814 		| UVD_SUVD_CGC_CTRL__MPC1_MODE_MASK
815 		| UVD_SUVD_CGC_CTRL__FBC_PCLK_MASK
816 		| UVD_SUVD_CGC_CTRL__FBC_CCLK_MASK);
817 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL, data);
818 }
819 
820 static void vcn_v3_0_clock_gating_dpg_mode(struct amdgpu_device *adev,
821 		uint8_t sram_sel, int inst_idx, uint8_t indirect)
822 {
823 	uint32_t reg_data = 0;
824 
825 	/* enable sw clock gating control */
826 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
827 		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
828 	else
829 		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
830 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
831 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
832 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
833 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
834 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
835 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
836 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
837 		 UVD_CGC_CTRL__SYS_MODE_MASK |
838 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
839 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
840 		 UVD_CGC_CTRL__REGS_MODE_MASK |
841 		 UVD_CGC_CTRL__RBC_MODE_MASK |
842 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
843 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
844 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
845 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
846 		 UVD_CGC_CTRL__MPC_MODE_MASK |
847 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
848 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
849 		 UVD_CGC_CTRL__WCB_MODE_MASK |
850 		 UVD_CGC_CTRL__VCPU_MODE_MASK |
851 		 UVD_CGC_CTRL__MMSCH_MODE_MASK);
852 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
853 		VCN, inst_idx, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
854 
855 	/* turn off clock gating */
856 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
857 		VCN, inst_idx, mmUVD_CGC_GATE), 0, sram_sel, indirect);
858 
859 	/* turn on SUVD clock gating */
860 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
861 		VCN, inst_idx, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
862 
863 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
864 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
865 		VCN, inst_idx, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
866 }
867 
868 /**
869  * vcn_v3_0_enable_clock_gating - enable VCN clock gating
870  *
871  * @adev: amdgpu_device pointer
872  * @inst: instance number
873  *
874  * Enable clock gating for VCN block
875  */
876 static void vcn_v3_0_enable_clock_gating(struct amdgpu_device *adev, int inst)
877 {
878 	uint32_t data;
879 
880 	/* enable VCN CGC */
881 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
882 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
883 		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
884 	else
885 		data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
886 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
887 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
888 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
889 
890 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
891 	data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
892 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
893 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
894 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
895 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
896 		| UVD_CGC_CTRL__SYS_MODE_MASK
897 		| UVD_CGC_CTRL__UDEC_MODE_MASK
898 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
899 		| UVD_CGC_CTRL__REGS_MODE_MASK
900 		| UVD_CGC_CTRL__RBC_MODE_MASK
901 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
902 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
903 		| UVD_CGC_CTRL__IDCT_MODE_MASK
904 		| UVD_CGC_CTRL__MPRD_MODE_MASK
905 		| UVD_CGC_CTRL__MPC_MODE_MASK
906 		| UVD_CGC_CTRL__LBSI_MODE_MASK
907 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
908 		| UVD_CGC_CTRL__WCB_MODE_MASK
909 		| UVD_CGC_CTRL__VCPU_MODE_MASK
910 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
911 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
912 
913 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL);
914 	data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
915 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
916 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
917 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
918 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
919 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
920 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
921 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
922 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK
923 		| UVD_SUVD_CGC_CTRL__EFC_MODE_MASK
924 		| UVD_SUVD_CGC_CTRL__SAOE_MODE_MASK
925 		| UVD_SUVD_CGC_CTRL__SMPA_MODE_MASK
926 		| UVD_SUVD_CGC_CTRL__MPBE0_MODE_MASK
927 		| UVD_SUVD_CGC_CTRL__MPBE1_MODE_MASK
928 		| UVD_SUVD_CGC_CTRL__SIT_AV1_MODE_MASK
929 		| UVD_SUVD_CGC_CTRL__SDB_AV1_MODE_MASK
930 		| UVD_SUVD_CGC_CTRL__MPC1_MODE_MASK
931 		| UVD_SUVD_CGC_CTRL__FBC_PCLK_MASK
932 		| UVD_SUVD_CGC_CTRL__FBC_CCLK_MASK);
933 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL, data);
934 }
935 
936 static int vcn_v3_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
937 {
938 	struct amdgpu_ring *ring;
939 	uint32_t rb_bufsz, tmp;
940 
941 	/* disable register anti-hang mechanism */
942 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1,
943 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
944 	/* enable dynamic power gating mode */
945 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS);
946 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
947 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
948 	WREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS, tmp);
949 
950 	if (indirect)
951 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
952 
953 	/* enable clock gating */
954 	vcn_v3_0_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
955 
956 	/* enable VCPU clock */
957 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
958 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
959 	tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
960 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
961 		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
962 
963 	/* disable master interupt */
964 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
965 		VCN, inst_idx, mmUVD_MASTINT_EN), 0, 0, indirect);
966 
967 	/* setup mmUVD_LMI_CTRL */
968 	tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
969 		UVD_LMI_CTRL__REQ_MODE_MASK |
970 		UVD_LMI_CTRL__CRC_RESET_MASK |
971 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
972 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
973 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
974 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
975 		0x00100000L);
976 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
977 		VCN, inst_idx, mmUVD_LMI_CTRL), tmp, 0, indirect);
978 
979 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
980 		VCN, inst_idx, mmUVD_MPC_CNTL),
981 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
982 
983 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
984 		VCN, inst_idx, mmUVD_MPC_SET_MUXA0),
985 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
986 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
987 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
988 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
989 
990 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
991 		VCN, inst_idx, mmUVD_MPC_SET_MUXB0),
992 		 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
993 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
994 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
995 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
996 
997 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
998 		VCN, inst_idx, mmUVD_MPC_SET_MUX),
999 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1000 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1001 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
1002 
1003 	vcn_v3_0_mc_resume_dpg_mode(adev, inst_idx, indirect);
1004 
1005 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1006 		VCN, inst_idx, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
1007 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1008 		VCN, inst_idx, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
1009 
1010 	/* enable LMI MC and UMC channels */
1011 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1012 		VCN, inst_idx, mmUVD_LMI_CTRL2), 0, 0, indirect);
1013 
1014 	/* unblock VCPU register access */
1015 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1016 		VCN, inst_idx, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
1017 
1018 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
1019 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
1020 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1021 		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
1022 
1023 	/* enable master interrupt */
1024 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1025 		VCN, inst_idx, mmUVD_MASTINT_EN),
1026 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
1027 
1028 	/* add nop to workaround PSP size check */
1029 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1030 		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
1031 
1032 	if (indirect)
1033 		psp_update_vcn_sram(adev, inst_idx, adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
1034 			(uint32_t)((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
1035 				(uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr));
1036 
1037 	ring = &adev->vcn.inst[inst_idx].ring_dec;
1038 	/* force RBC into idle state */
1039 	rb_bufsz = order_base_2(ring->ring_size);
1040 	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1041 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1042 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1043 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1044 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1045 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
1046 
1047 	/* set the write pointer delay */
1048 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
1049 
1050 	/* set the wb address */
1051 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR_ADDR,
1052 		(upper_32_bits(ring->gpu_addr) >> 2));
1053 
1054 	/* programm the RB_BASE for ring buffer */
1055 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1056 		lower_32_bits(ring->gpu_addr));
1057 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1058 		upper_32_bits(ring->gpu_addr));
1059 
1060 	/* Initialize the ring buffer's read and write pointers */
1061 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, 0);
1062 
1063 	WREG32_SOC15(VCN, inst_idx, mmUVD_SCRATCH2, 0);
1064 
1065 	ring->wptr = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR);
1066 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR,
1067 		lower_32_bits(ring->wptr));
1068 
1069 	return 0;
1070 }
1071 
1072 static int vcn_v3_0_start(struct amdgpu_device *adev)
1073 {
1074 	struct amdgpu_ring *ring;
1075 	uint32_t rb_bufsz, tmp;
1076 	int i, j, k, r;
1077 
1078 	if (adev->pm.dpm_enabled)
1079 		amdgpu_dpm_enable_uvd(adev, true);
1080 
1081 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1082 		if (adev->vcn.harvest_config & (1 << i))
1083 			continue;
1084 
1085 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG){
1086 			r = vcn_v3_0_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
1087 			continue;
1088 		}
1089 
1090 		/* disable VCN power gating */
1091 		vcn_v3_0_disable_static_power_gating(adev, i);
1092 
1093 		/* set VCN status busy */
1094 		tmp = RREG32_SOC15(VCN, i, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY;
1095 		WREG32_SOC15(VCN, i, mmUVD_STATUS, tmp);
1096 
1097 		/*SW clock gating */
1098 		vcn_v3_0_disable_clock_gating(adev, i);
1099 
1100 		/* enable VCPU clock */
1101 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1102 			UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1103 
1104 		/* disable master interrupt */
1105 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN), 0,
1106 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1107 
1108 		/* enable LMI MC and UMC channels */
1109 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_LMI_CTRL2), 0,
1110 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1111 
1112 		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1113 		tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1114 		tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1115 		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1116 
1117 		/* setup mmUVD_LMI_CTRL */
1118 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL);
1119 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL, tmp |
1120 			UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK	|
1121 			UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1122 			UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1123 			UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1124 
1125 		/* setup mmUVD_MPC_CNTL */
1126 		tmp = RREG32_SOC15(VCN, i, mmUVD_MPC_CNTL);
1127 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1128 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1129 		WREG32_SOC15(VCN, i, mmUVD_MPC_CNTL, tmp);
1130 
1131 		/* setup UVD_MPC_SET_MUXA0 */
1132 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXA0,
1133 			((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1134 			(0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1135 			(0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1136 			(0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1137 
1138 		/* setup UVD_MPC_SET_MUXB0 */
1139 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXB0,
1140 			((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1141 			(0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1142 			(0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1143 			(0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1144 
1145 		/* setup mmUVD_MPC_SET_MUX */
1146 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUX,
1147 			((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1148 			(0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1149 			(0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1150 
1151 		vcn_v3_0_mc_resume(adev, i);
1152 
1153 		/* VCN global tiling registers */
1154 		WREG32_SOC15(VCN, i, mmUVD_GFX10_ADDR_CONFIG,
1155 			adev->gfx.config.gb_addr_config);
1156 
1157 		/* unblock VCPU register access */
1158 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL), 0,
1159 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1160 
1161 		/* release VCPU reset to boot */
1162 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1163 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1164 
1165 		for (j = 0; j < 10; ++j) {
1166 			uint32_t status;
1167 
1168 			for (k = 0; k < 100; ++k) {
1169 				status = RREG32_SOC15(VCN, i, mmUVD_STATUS);
1170 				if (status & 2)
1171 					break;
1172 				mdelay(10);
1173 			}
1174 			r = 0;
1175 			if (status & 2)
1176 				break;
1177 
1178 			DRM_ERROR("VCN[%d] decode not responding, trying to reset the VCPU!!!\n", i);
1179 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1180 				UVD_VCPU_CNTL__BLK_RST_MASK,
1181 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1182 			mdelay(10);
1183 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1184 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1185 
1186 			mdelay(10);
1187 			r = -1;
1188 		}
1189 
1190 		if (r) {
1191 			DRM_ERROR("VCN[%d] decode not responding, giving up!!!\n", i);
1192 			return r;
1193 		}
1194 
1195 		/* enable master interrupt */
1196 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN),
1197 			UVD_MASTINT_EN__VCPU_EN_MASK,
1198 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1199 
1200 		/* clear the busy bit of VCN_STATUS */
1201 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS), 0,
1202 			~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1203 
1204 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_VMID, 0);
1205 
1206 		ring = &adev->vcn.inst[i].ring_dec;
1207 		/* force RBC into idle state */
1208 		rb_bufsz = order_base_2(ring->ring_size);
1209 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1210 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1211 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1212 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1213 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1214 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_CNTL, tmp);
1215 
1216 		/* programm the RB_BASE for ring buffer */
1217 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1218 			lower_32_bits(ring->gpu_addr));
1219 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1220 			upper_32_bits(ring->gpu_addr));
1221 
1222 		/* Initialize the ring buffer's read and write pointers */
1223 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR, 0);
1224 
1225 		ring->wptr = RREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR);
1226 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_WPTR,
1227 			lower_32_bits(ring->wptr));
1228 		ring = &adev->vcn.inst[i].ring_enc[0];
1229 		WREG32_SOC15(VCN, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1230 		WREG32_SOC15(VCN, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1231 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO, ring->gpu_addr);
1232 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1233 		WREG32_SOC15(VCN, i, mmUVD_RB_SIZE, ring->ring_size / 4);
1234 
1235 		ring = &adev->vcn.inst[i].ring_enc[1];
1236 		WREG32_SOC15(VCN, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1237 		WREG32_SOC15(VCN, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1238 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1239 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1240 		WREG32_SOC15(VCN, i, mmUVD_RB_SIZE2, ring->ring_size / 4);
1241 	}
1242 
1243 	return 0;
1244 }
1245 
1246 static int vcn_v3_0_start_sriov(struct amdgpu_device *adev)
1247 {
1248 	int i, j;
1249 	struct amdgpu_ring *ring;
1250 	uint64_t cache_addr;
1251 	uint64_t rb_addr;
1252 	uint64_t ctx_addr;
1253 	uint32_t param, resp, expected;
1254 	uint32_t offset, cache_size;
1255 	uint32_t tmp, timeout;
1256 	uint32_t id;
1257 
1258 	struct amdgpu_mm_table *table = &adev->virt.mm_table;
1259 	uint32_t *table_loc;
1260 	uint32_t table_size;
1261 	uint32_t size, size_dw;
1262 
1263 	bool is_vcn_ready;
1264 
1265 	struct mmsch_v3_0_cmd_direct_write
1266 		direct_wt = { {0} };
1267 	struct mmsch_v3_0_cmd_direct_read_modify_write
1268 		direct_rd_mod_wt = { {0} };
1269 	struct mmsch_v3_0_cmd_direct_polling
1270 		direct_poll = { {0} };
1271 	struct mmsch_v3_0_cmd_end end = { {0} };
1272 	struct mmsch_v3_0_init_header header;
1273 
1274 	direct_wt.cmd_header.command_type =
1275 		MMSCH_COMMAND__DIRECT_REG_WRITE;
1276 	direct_rd_mod_wt.cmd_header.command_type =
1277 		MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1278 	direct_poll.cmd_header.command_type =
1279 		MMSCH_COMMAND__DIRECT_REG_POLLING;
1280 	end.cmd_header.command_type =
1281 		MMSCH_COMMAND__END;
1282 
1283 	header.version = MMSCH_VERSION;
1284 	header.total_size = sizeof(struct mmsch_v3_0_init_header) >> 2;
1285 	for (i = 0; i < AMDGPU_MAX_VCN_INSTANCES; i++) {
1286 		header.inst[i].init_status = 0;
1287 		header.inst[i].table_offset = 0;
1288 		header.inst[i].table_size = 0;
1289 	}
1290 
1291 	table_loc = (uint32_t *)table->cpu_addr;
1292 	table_loc += header.total_size;
1293 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1294 		if (adev->vcn.harvest_config & (1 << i))
1295 			continue;
1296 
1297 		table_size = 0;
1298 
1299 		MMSCH_V3_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(VCN, i,
1300 			mmUVD_STATUS),
1301 			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1302 
1303 		cache_size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1304 
1305 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1306 			id = amdgpu_ucode_id_vcns[i];
1307 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1308 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1309 				adev->firmware.ucode[id].tmr_mc_addr_lo);
1310 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1311 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1312 				adev->firmware.ucode[id].tmr_mc_addr_hi);
1313 			offset = 0;
1314 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1315 				mmUVD_VCPU_CACHE_OFFSET0),
1316 				0);
1317 		} else {
1318 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1319 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1320 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1321 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1322 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1323 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1324 			offset = cache_size;
1325 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1326 				mmUVD_VCPU_CACHE_OFFSET0),
1327 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1328 		}
1329 
1330 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1331 			mmUVD_VCPU_CACHE_SIZE0),
1332 			cache_size);
1333 
1334 		cache_addr = adev->vcn.inst[i].gpu_addr + offset;
1335 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1336 			mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1337 			lower_32_bits(cache_addr));
1338 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1339 			mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1340 			upper_32_bits(cache_addr));
1341 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1342 			mmUVD_VCPU_CACHE_OFFSET1),
1343 			0);
1344 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1345 			mmUVD_VCPU_CACHE_SIZE1),
1346 			AMDGPU_VCN_STACK_SIZE);
1347 
1348 		cache_addr = adev->vcn.inst[i].gpu_addr + offset +
1349 			AMDGPU_VCN_STACK_SIZE;
1350 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1351 			mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1352 			lower_32_bits(cache_addr));
1353 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1354 			mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1355 			upper_32_bits(cache_addr));
1356 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1357 			mmUVD_VCPU_CACHE_OFFSET2),
1358 			0);
1359 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1360 			mmUVD_VCPU_CACHE_SIZE2),
1361 			AMDGPU_VCN_CONTEXT_SIZE);
1362 
1363 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
1364 			ring = &adev->vcn.inst[i].ring_enc[j];
1365 			ring->wptr = 0;
1366 			rb_addr = ring->gpu_addr;
1367 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1368 				mmUVD_RB_BASE_LO),
1369 				lower_32_bits(rb_addr));
1370 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1371 				mmUVD_RB_BASE_HI),
1372 				upper_32_bits(rb_addr));
1373 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1374 				mmUVD_RB_SIZE),
1375 				ring->ring_size / 4);
1376 		}
1377 
1378 		ring = &adev->vcn.inst[i].ring_dec;
1379 		ring->wptr = 0;
1380 		rb_addr = ring->gpu_addr;
1381 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1382 			mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1383 			lower_32_bits(rb_addr));
1384 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1385 			mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1386 			upper_32_bits(rb_addr));
1387 		/* force RBC into idle state */
1388 		tmp = order_base_2(ring->ring_size);
1389 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, tmp);
1390 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1391 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1392 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1393 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1394 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1395 			mmUVD_RBC_RB_CNTL),
1396 			tmp);
1397 
1398 		/* add end packet */
1399 		MMSCH_V3_0_INSERT_END();
1400 
1401 		/* refine header */
1402 		header.inst[i].init_status = 0;
1403 		header.inst[i].table_offset = header.total_size;
1404 		header.inst[i].table_size = table_size;
1405 		header.total_size += table_size;
1406 	}
1407 
1408 	/* Update init table header in memory */
1409 	size = sizeof(struct mmsch_v3_0_init_header);
1410 	table_loc = (uint32_t *)table->cpu_addr;
1411 	memcpy((void *)table_loc, &header, size);
1412 
1413 	/* message MMSCH (in VCN[0]) to initialize this client
1414 	 * 1, write to mmsch_vf_ctx_addr_lo/hi register with GPU mc addr
1415 	 * of memory descriptor location
1416 	 */
1417 	ctx_addr = table->gpu_addr;
1418 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(ctx_addr));
1419 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(ctx_addr));
1420 
1421 	/* 2, update vmid of descriptor */
1422 	tmp = RREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID);
1423 	tmp &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1424 	/* use domain0 for MM scheduler */
1425 	tmp |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1426 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID, tmp);
1427 
1428 	/* 3, notify mmsch about the size of this descriptor */
1429 	size = header.total_size;
1430 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_SIZE, size);
1431 
1432 	/* 4, set resp to zero */
1433 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1434 
1435 	/* 5, kick off the initialization and wait until
1436 	 * MMSCH_VF_MAILBOX_RESP becomes non-zero
1437 	 */
1438 	param = 0x10000001;
1439 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_HOST, param);
1440 	tmp = 0;
1441 	timeout = 1000;
1442 	resp = 0;
1443 	expected = param + 1;
1444 	while (resp != expected) {
1445 		resp = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1446 		if (resp == expected)
1447 			break;
1448 
1449 		udelay(10);
1450 		tmp = tmp + 10;
1451 		if (tmp >= timeout) {
1452 			DRM_ERROR("failed to init MMSCH. TIME-OUT after %d usec"\
1453 				" waiting for mmMMSCH_VF_MAILBOX_RESP "\
1454 				"(expected=0x%08x, readback=0x%08x)\n",
1455 				tmp, expected, resp);
1456 			return -EBUSY;
1457 		}
1458 	}
1459 
1460 	/* 6, check each VCN's init_status
1461 	 * if it remains as 0, then this VCN is not assigned to current VF
1462 	 * do not start ring for this VCN
1463 	 */
1464 	size = sizeof(struct mmsch_v3_0_init_header);
1465 	table_loc = (uint32_t *)table->cpu_addr;
1466 	memcpy(&header, (void *)table_loc, size);
1467 
1468 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1469 		if (adev->vcn.harvest_config & (1 << i))
1470 			continue;
1471 
1472 		is_vcn_ready = (header.inst[i].init_status == 1);
1473 		if (!is_vcn_ready)
1474 			DRM_INFO("VCN(%d) engine is disabled by hypervisor\n", i);
1475 
1476 		ring = &adev->vcn.inst[i].ring_dec;
1477 		ring->sched.ready = is_vcn_ready;
1478 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
1479 			ring = &adev->vcn.inst[i].ring_enc[j];
1480 			ring->sched.ready = is_vcn_ready;
1481 		}
1482 	}
1483 
1484 	return 0;
1485 }
1486 
1487 static int vcn_v3_0_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1488 {
1489 	uint32_t tmp;
1490 
1491 	/* Wait for power status to be 1 */
1492 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1493 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1494 
1495 	/* wait for read ptr to be equal to write ptr */
1496 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR);
1497 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1498 
1499 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2);
1500 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
1501 
1502 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1503 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF);
1504 
1505 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1506 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1507 
1508 	/* disable dynamic power gating mode */
1509 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 0,
1510 		~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1511 
1512 	return 0;
1513 }
1514 
1515 static int vcn_v3_0_stop(struct amdgpu_device *adev)
1516 {
1517 	uint32_t tmp;
1518 	int i, r = 0;
1519 
1520 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1521 		if (adev->vcn.harvest_config & (1 << i))
1522 			continue;
1523 
1524 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1525 			r = vcn_v3_0_stop_dpg_mode(adev, i);
1526 			continue;
1527 		}
1528 
1529 		/* wait for vcn idle */
1530 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1531 		if (r)
1532 			return r;
1533 
1534 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1535 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1536 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1537 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1538 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1539 		if (r)
1540 			return r;
1541 
1542 		/* disable LMI UMC channel */
1543 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2);
1544 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1545 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2, tmp);
1546 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1547 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1548 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1549 		if (r)
1550 			return r;
1551 
1552 		/* block VCPU register access */
1553 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL),
1554 			UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1555 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1556 
1557 		/* reset VCPU */
1558 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1559 			UVD_VCPU_CNTL__BLK_RST_MASK,
1560 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1561 
1562 		/* disable VCPU clock */
1563 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1564 			~(UVD_VCPU_CNTL__CLK_EN_MASK));
1565 
1566 		/* apply soft reset */
1567 		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1568 		tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1569 		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1570 		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1571 		tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1572 		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1573 
1574 		/* clear status */
1575 		WREG32_SOC15(VCN, i, mmUVD_STATUS, 0);
1576 
1577 		/* apply HW clock gating */
1578 		vcn_v3_0_enable_clock_gating(adev, i);
1579 
1580 		/* enable VCN power gating */
1581 		vcn_v3_0_enable_static_power_gating(adev, i);
1582 	}
1583 
1584 	if (adev->pm.dpm_enabled)
1585 		amdgpu_dpm_enable_uvd(adev, false);
1586 
1587 	return 0;
1588 }
1589 
1590 static int vcn_v3_0_pause_dpg_mode(struct amdgpu_device *adev,
1591 		   int inst_idx, struct dpg_pause_state *new_state)
1592 {
1593 	struct amdgpu_ring *ring;
1594 	uint32_t reg_data = 0;
1595 	int ret_code;
1596 
1597 	/* pause/unpause if state is changed */
1598 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1599 		DRM_DEBUG("dpg pause state changed %d -> %d",
1600 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1601 		reg_data = RREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE) &
1602 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1603 
1604 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1605 			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1606 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1607 
1608 			if (!ret_code) {
1609 				/* pause DPG */
1610 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1611 				WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1612 
1613 				/* wait for ACK */
1614 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_DPG_PAUSE,
1615 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1616 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1617 
1618 				/* Restore */
1619 				ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1620 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
1621 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1622 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
1623 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1624 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1625 
1626 				ring = &adev->vcn.inst[inst_idx].ring_enc[1];
1627 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1628 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1629 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
1630 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1631 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1632 
1633 				WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR,
1634 					RREG32_SOC15(VCN, inst_idx, mmUVD_SCRATCH2) & 0x7FFFFFFF);
1635 
1636 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS,
1637 					UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1638 			}
1639 		} else {
1640 			/* unpause dpg, no need to wait */
1641 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1642 			WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1643 		}
1644 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1645 	}
1646 
1647 	return 0;
1648 }
1649 
1650 /**
1651  * vcn_v3_0_dec_ring_get_rptr - get read pointer
1652  *
1653  * @ring: amdgpu_ring pointer
1654  *
1655  * Returns the current hardware read pointer
1656  */
1657 static uint64_t vcn_v3_0_dec_ring_get_rptr(struct amdgpu_ring *ring)
1658 {
1659 	struct amdgpu_device *adev = ring->adev;
1660 
1661 	return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_RPTR);
1662 }
1663 
1664 /**
1665  * vcn_v3_0_dec_ring_get_wptr - get write pointer
1666  *
1667  * @ring: amdgpu_ring pointer
1668  *
1669  * Returns the current hardware write pointer
1670  */
1671 static uint64_t vcn_v3_0_dec_ring_get_wptr(struct amdgpu_ring *ring)
1672 {
1673 	struct amdgpu_device *adev = ring->adev;
1674 
1675 	if (ring->use_doorbell)
1676 		return adev->wb.wb[ring->wptr_offs];
1677 	else
1678 		return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR);
1679 }
1680 
1681 /**
1682  * vcn_v3_0_dec_ring_set_wptr - set write pointer
1683  *
1684  * @ring: amdgpu_ring pointer
1685  *
1686  * Commits the write pointer to the hardware
1687  */
1688 static void vcn_v3_0_dec_ring_set_wptr(struct amdgpu_ring *ring)
1689 {
1690 	struct amdgpu_device *adev = ring->adev;
1691 
1692 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
1693 		WREG32_SOC15(VCN, ring->me, mmUVD_SCRATCH2,
1694 			lower_32_bits(ring->wptr) | 0x80000000);
1695 
1696 	if (ring->use_doorbell) {
1697 		adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1698 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1699 	} else {
1700 		WREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1701 	}
1702 }
1703 
1704 void vcn_v3_0_dec_sw_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
1705 				u64 seq, uint32_t flags)
1706 {
1707 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
1708 
1709 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_FENCE);
1710 	amdgpu_ring_write(ring, addr);
1711 	amdgpu_ring_write(ring, upper_32_bits(addr));
1712 	amdgpu_ring_write(ring, seq);
1713 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_TRAP);
1714 }
1715 
1716 void vcn_v3_0_dec_sw_ring_insert_end(struct amdgpu_ring *ring)
1717 {
1718 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_END);
1719 }
1720 
1721 void vcn_v3_0_dec_sw_ring_emit_ib(struct amdgpu_ring *ring,
1722 			       struct amdgpu_job *job,
1723 			       struct amdgpu_ib *ib,
1724 			       uint32_t flags)
1725 {
1726 	uint32_t vmid = AMDGPU_JOB_GET_VMID(job);
1727 
1728 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_IB);
1729 	amdgpu_ring_write(ring, vmid);
1730 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
1731 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
1732 	amdgpu_ring_write(ring, ib->length_dw);
1733 }
1734 
1735 void vcn_v3_0_dec_sw_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
1736 				uint32_t val, uint32_t mask)
1737 {
1738 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_REG_WAIT);
1739 	amdgpu_ring_write(ring, reg << 2);
1740 	amdgpu_ring_write(ring, mask);
1741 	amdgpu_ring_write(ring, val);
1742 }
1743 
1744 void vcn_v3_0_dec_sw_ring_emit_vm_flush(struct amdgpu_ring *ring,
1745 				uint32_t vmid, uint64_t pd_addr)
1746 {
1747 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
1748 	uint32_t data0, data1, mask;
1749 
1750 	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1751 
1752 	/* wait for register write */
1753 	data0 = hub->ctx0_ptb_addr_lo32 + vmid * hub->ctx_addr_distance;
1754 	data1 = lower_32_bits(pd_addr);
1755 	mask = 0xffffffff;
1756 	vcn_v3_0_dec_sw_ring_emit_reg_wait(ring, data0, data1, mask);
1757 }
1758 
1759 void vcn_v3_0_dec_sw_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg, uint32_t val)
1760 {
1761 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_REG_WRITE);
1762 	amdgpu_ring_write(ring,	reg << 2);
1763 	amdgpu_ring_write(ring, val);
1764 }
1765 
1766 static const struct amdgpu_ring_funcs vcn_v3_0_dec_sw_ring_vm_funcs = {
1767 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1768 	.align_mask = 0x3f,
1769 	.nop = VCN_DEC_SW_CMD_NO_OP,
1770 	.vmhub = AMDGPU_MMHUB_0,
1771 	.get_rptr = vcn_v3_0_dec_ring_get_rptr,
1772 	.get_wptr = vcn_v3_0_dec_ring_get_wptr,
1773 	.set_wptr = vcn_v3_0_dec_ring_set_wptr,
1774 	.emit_frame_size =
1775 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1776 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1777 		4 + /* vcn_v3_0_dec_sw_ring_emit_vm_flush */
1778 		5 + 5 + /* vcn_v3_0_dec_sw_ring_emit_fdec_swe x2 vm fdec_swe */
1779 		1, /* vcn_v3_0_dec_sw_ring_insert_end */
1780 	.emit_ib_size = 5, /* vcn_v3_0_dec_sw_ring_emit_ib */
1781 	.emit_ib = vcn_v3_0_dec_sw_ring_emit_ib,
1782 	.emit_fence = vcn_v3_0_dec_sw_ring_emit_fence,
1783 	.emit_vm_flush = vcn_v3_0_dec_sw_ring_emit_vm_flush,
1784 	.test_ring = amdgpu_vcn_dec_sw_ring_test_ring,
1785 	.test_ib = NULL,//amdgpu_vcn_dec_sw_ring_test_ib,
1786 	.insert_nop = amdgpu_ring_insert_nop,
1787 	.insert_end = vcn_v3_0_dec_sw_ring_insert_end,
1788 	.pad_ib = amdgpu_ring_generic_pad_ib,
1789 	.begin_use = amdgpu_vcn_ring_begin_use,
1790 	.end_use = amdgpu_vcn_ring_end_use,
1791 	.emit_wreg = vcn_v3_0_dec_sw_ring_emit_wreg,
1792 	.emit_reg_wait = vcn_v3_0_dec_sw_ring_emit_reg_wait,
1793 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1794 };
1795 
1796 static const struct amdgpu_ring_funcs vcn_v3_0_dec_ring_vm_funcs = {
1797 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1798 	.align_mask = 0xf,
1799 	.vmhub = AMDGPU_MMHUB_0,
1800 	.get_rptr = vcn_v3_0_dec_ring_get_rptr,
1801 	.get_wptr = vcn_v3_0_dec_ring_get_wptr,
1802 	.set_wptr = vcn_v3_0_dec_ring_set_wptr,
1803 	.emit_frame_size =
1804 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1805 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1806 		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1807 		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1808 		6,
1809 	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1810 	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
1811 	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
1812 	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1813 	.test_ring = vcn_v2_0_dec_ring_test_ring,
1814 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
1815 	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
1816 	.insert_start = vcn_v2_0_dec_ring_insert_start,
1817 	.insert_end = vcn_v2_0_dec_ring_insert_end,
1818 	.pad_ib = amdgpu_ring_generic_pad_ib,
1819 	.begin_use = amdgpu_vcn_ring_begin_use,
1820 	.end_use = amdgpu_vcn_ring_end_use,
1821 	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1822 	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1823 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1824 };
1825 
1826 /**
1827  * vcn_v3_0_enc_ring_get_rptr - get enc read pointer
1828  *
1829  * @ring: amdgpu_ring pointer
1830  *
1831  * Returns the current hardware enc read pointer
1832  */
1833 static uint64_t vcn_v3_0_enc_ring_get_rptr(struct amdgpu_ring *ring)
1834 {
1835 	struct amdgpu_device *adev = ring->adev;
1836 
1837 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0])
1838 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR);
1839 	else
1840 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR2);
1841 }
1842 
1843 /**
1844  * vcn_v3_0_enc_ring_get_wptr - get enc write pointer
1845  *
1846  * @ring: amdgpu_ring pointer
1847  *
1848  * Returns the current hardware enc write pointer
1849  */
1850 static uint64_t vcn_v3_0_enc_ring_get_wptr(struct amdgpu_ring *ring)
1851 {
1852 	struct amdgpu_device *adev = ring->adev;
1853 
1854 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1855 		if (ring->use_doorbell)
1856 			return adev->wb.wb[ring->wptr_offs];
1857 		else
1858 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR);
1859 	} else {
1860 		if (ring->use_doorbell)
1861 			return adev->wb.wb[ring->wptr_offs];
1862 		else
1863 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2);
1864 	}
1865 }
1866 
1867 /**
1868  * vcn_v3_0_enc_ring_set_wptr - set enc write pointer
1869  *
1870  * @ring: amdgpu_ring pointer
1871  *
1872  * Commits the enc write pointer to the hardware
1873  */
1874 static void vcn_v3_0_enc_ring_set_wptr(struct amdgpu_ring *ring)
1875 {
1876 	struct amdgpu_device *adev = ring->adev;
1877 
1878 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1879 		if (ring->use_doorbell) {
1880 			adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1881 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1882 		} else {
1883 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1884 		}
1885 	} else {
1886 		if (ring->use_doorbell) {
1887 			adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1888 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1889 		} else {
1890 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1891 		}
1892 	}
1893 }
1894 
1895 static const struct amdgpu_ring_funcs vcn_v3_0_enc_ring_vm_funcs = {
1896 	.type = AMDGPU_RING_TYPE_VCN_ENC,
1897 	.align_mask = 0x3f,
1898 	.nop = VCN_ENC_CMD_NO_OP,
1899 	.vmhub = AMDGPU_MMHUB_0,
1900 	.get_rptr = vcn_v3_0_enc_ring_get_rptr,
1901 	.get_wptr = vcn_v3_0_enc_ring_get_wptr,
1902 	.set_wptr = vcn_v3_0_enc_ring_set_wptr,
1903 	.emit_frame_size =
1904 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1905 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1906 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1907 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1908 		1, /* vcn_v2_0_enc_ring_insert_end */
1909 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1910 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
1911 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
1912 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1913 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
1914 	.test_ib = amdgpu_vcn_enc_ring_test_ib,
1915 	.insert_nop = amdgpu_ring_insert_nop,
1916 	.insert_end = vcn_v2_0_enc_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_enc_ring_emit_wreg,
1921 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1922 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1923 };
1924 
1925 static void vcn_v3_0_set_dec_ring_funcs(struct amdgpu_device *adev)
1926 {
1927 	int i;
1928 
1929 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1930 		if (adev->vcn.harvest_config & (1 << i))
1931 			continue;
1932 
1933 		if (!DEC_SW_RING_ENABLED)
1934 			adev->vcn.inst[i].ring_dec.funcs = &vcn_v3_0_dec_ring_vm_funcs;
1935 		else
1936 			adev->vcn.inst[i].ring_dec.funcs = &vcn_v3_0_dec_sw_ring_vm_funcs;
1937 		adev->vcn.inst[i].ring_dec.me = i;
1938 		DRM_INFO("VCN(%d) decode%s is enabled in VM mode\n", i,
1939 			  DEC_SW_RING_ENABLED?"(Software Ring)":"");
1940 	}
1941 }
1942 
1943 static void vcn_v3_0_set_enc_ring_funcs(struct amdgpu_device *adev)
1944 {
1945 	int i, j;
1946 
1947 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1948 		if (adev->vcn.harvest_config & (1 << i))
1949 			continue;
1950 
1951 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
1952 			adev->vcn.inst[i].ring_enc[j].funcs = &vcn_v3_0_enc_ring_vm_funcs;
1953 			adev->vcn.inst[i].ring_enc[j].me = i;
1954 		}
1955 		DRM_INFO("VCN(%d) encode is enabled in VM mode\n", i);
1956 	}
1957 }
1958 
1959 static bool vcn_v3_0_is_idle(void *handle)
1960 {
1961 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1962 	int i, ret = 1;
1963 
1964 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1965 		if (adev->vcn.harvest_config & (1 << i))
1966 			continue;
1967 
1968 		ret &= (RREG32_SOC15(VCN, i, mmUVD_STATUS) == UVD_STATUS__IDLE);
1969 	}
1970 
1971 	return ret;
1972 }
1973 
1974 static int vcn_v3_0_wait_for_idle(void *handle)
1975 {
1976 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1977 	int i, ret = 0;
1978 
1979 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1980 		if (adev->vcn.harvest_config & (1 << i))
1981 			continue;
1982 
1983 		ret = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE,
1984 			UVD_STATUS__IDLE);
1985 		if (ret)
1986 			return ret;
1987 	}
1988 
1989 	return ret;
1990 }
1991 
1992 static int vcn_v3_0_set_clockgating_state(void *handle,
1993 					  enum amd_clockgating_state state)
1994 {
1995 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1996 	bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
1997 	int i;
1998 
1999 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2000 		if (adev->vcn.harvest_config & (1 << i))
2001 			continue;
2002 
2003 		if (enable) {
2004 			if (RREG32_SOC15(VCN, i, mmUVD_STATUS) != UVD_STATUS__IDLE)
2005 				return -EBUSY;
2006 			vcn_v3_0_enable_clock_gating(adev, i);
2007 		} else {
2008 			vcn_v3_0_disable_clock_gating(adev, i);
2009 		}
2010 	}
2011 
2012 	return 0;
2013 }
2014 
2015 static int vcn_v3_0_set_powergating_state(void *handle,
2016 					  enum amd_powergating_state state)
2017 {
2018 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2019 	int ret;
2020 
2021 	/* for SRIOV, guest should not control VCN Power-gating
2022 	 * MMSCH FW should control Power-gating and clock-gating
2023 	 * guest should avoid touching CGC and PG
2024 	 */
2025 	if (amdgpu_sriov_vf(adev)) {
2026 		adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
2027 		return 0;
2028 	}
2029 
2030 	if(state == adev->vcn.cur_state)
2031 		return 0;
2032 
2033 	if (state == AMD_PG_STATE_GATE)
2034 		ret = vcn_v3_0_stop(adev);
2035 	else
2036 		ret = vcn_v3_0_start(adev);
2037 
2038 	if(!ret)
2039 		adev->vcn.cur_state = state;
2040 
2041 	return ret;
2042 }
2043 
2044 static int vcn_v3_0_set_interrupt_state(struct amdgpu_device *adev,
2045 					struct amdgpu_irq_src *source,
2046 					unsigned type,
2047 					enum amdgpu_interrupt_state state)
2048 {
2049 	return 0;
2050 }
2051 
2052 static int vcn_v3_0_process_interrupt(struct amdgpu_device *adev,
2053 				      struct amdgpu_irq_src *source,
2054 				      struct amdgpu_iv_entry *entry)
2055 {
2056 	uint32_t ip_instance;
2057 
2058 	switch (entry->client_id) {
2059 	case SOC15_IH_CLIENTID_VCN:
2060 		ip_instance = 0;
2061 		break;
2062 	case SOC15_IH_CLIENTID_VCN1:
2063 		ip_instance = 1;
2064 		break;
2065 	default:
2066 		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
2067 		return 0;
2068 	}
2069 
2070 	DRM_DEBUG("IH: VCN TRAP\n");
2071 
2072 	switch (entry->src_id) {
2073 	case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
2074 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec);
2075 		break;
2076 	case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
2077 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
2078 		break;
2079 	case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
2080 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]);
2081 		break;
2082 	default:
2083 		DRM_ERROR("Unhandled interrupt: %d %d\n",
2084 			  entry->src_id, entry->src_data[0]);
2085 		break;
2086 	}
2087 
2088 	return 0;
2089 }
2090 
2091 static const struct amdgpu_irq_src_funcs vcn_v3_0_irq_funcs = {
2092 	.set = vcn_v3_0_set_interrupt_state,
2093 	.process = vcn_v3_0_process_interrupt,
2094 };
2095 
2096 static void vcn_v3_0_set_irq_funcs(struct amdgpu_device *adev)
2097 {
2098 	int i;
2099 
2100 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2101 		if (adev->vcn.harvest_config & (1 << i))
2102 			continue;
2103 
2104 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
2105 		adev->vcn.inst[i].irq.funcs = &vcn_v3_0_irq_funcs;
2106 	}
2107 }
2108 
2109 static const struct amd_ip_funcs vcn_v3_0_ip_funcs = {
2110 	.name = "vcn_v3_0",
2111 	.early_init = vcn_v3_0_early_init,
2112 	.late_init = NULL,
2113 	.sw_init = vcn_v3_0_sw_init,
2114 	.sw_fini = vcn_v3_0_sw_fini,
2115 	.hw_init = vcn_v3_0_hw_init,
2116 	.hw_fini = vcn_v3_0_hw_fini,
2117 	.suspend = vcn_v3_0_suspend,
2118 	.resume = vcn_v3_0_resume,
2119 	.is_idle = vcn_v3_0_is_idle,
2120 	.wait_for_idle = vcn_v3_0_wait_for_idle,
2121 	.check_soft_reset = NULL,
2122 	.pre_soft_reset = NULL,
2123 	.soft_reset = NULL,
2124 	.post_soft_reset = NULL,
2125 	.set_clockgating_state = vcn_v3_0_set_clockgating_state,
2126 	.set_powergating_state = vcn_v3_0_set_powergating_state,
2127 };
2128 
2129 const struct amdgpu_ip_block_version vcn_v3_0_ip_block =
2130 {
2131 	.type = AMD_IP_BLOCK_TYPE_VCN,
2132 	.major = 3,
2133 	.minor = 0,
2134 	.rev = 0,
2135 	.funcs = &vcn_v3_0_ip_funcs,
2136 };
2137