xref: /linux/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c (revision a1a9b71ebd6582c94090ee5b70fb3eabf7088f22)
1 /*
2  * Copyright 2021 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 "soc15_hw_ip.h"
31 #include "vcn_v2_0.h"
32 #include "vcn_sw_ring.h"
33 
34 #include "vcn/vcn_4_0_0_offset.h"
35 #include "vcn/vcn_4_0_0_sh_mask.h"
36 #include "ivsrcid/vcn/irqsrcs_vcn_4_0.h"
37 
38 #include <drm/drm_drv.h>
39 
40 #define mmUVD_DPG_LMA_CTL							regUVD_DPG_LMA_CTL
41 #define mmUVD_DPG_LMA_CTL_BASE_IDX						regUVD_DPG_LMA_CTL_BASE_IDX
42 #define mmUVD_DPG_LMA_DATA							regUVD_DPG_LMA_DATA
43 #define mmUVD_DPG_LMA_DATA_BASE_IDX						regUVD_DPG_LMA_DATA_BASE_IDX
44 
45 #define VCN_VID_SOC_ADDRESS_2_0							0x1fb00
46 #define VCN1_VID_SOC_ADDRESS_3_0						0x48300
47 
48 bool unifiedQ_enabled = false;
49 
50 static int amdgpu_ih_clientid_vcns[] = {
51 	SOC15_IH_CLIENTID_VCN,
52 	SOC15_IH_CLIENTID_VCN1
53 };
54 
55 static void vcn_v4_0_set_dec_ring_funcs(struct amdgpu_device *adev);
56 static void vcn_v4_0_set_enc_ring_funcs(struct amdgpu_device *adev);
57 static void vcn_v4_0_set_irq_funcs(struct amdgpu_device *adev);
58 static int vcn_v4_0_set_powergating_state(void *handle,
59         enum amd_powergating_state state);
60 static int vcn_v4_0_pause_dpg_mode(struct amdgpu_device *adev,
61         int inst_idx, struct dpg_pause_state *new_state);
62 
63 /**
64  * vcn_v4_0_early_init - set function pointers
65  *
66  * @handle: amdgpu_device pointer
67  *
68  * Set ring and irq function pointers
69  */
70 static int vcn_v4_0_early_init(void *handle)
71 {
72 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
73 
74 	if (unifiedQ_enabled) {
75 		adev->vcn.num_vcn_inst = 1;
76 		adev->vcn.num_enc_rings = 1;
77 	} else {
78 		adev->vcn.num_enc_rings = 2;
79 	}
80 
81 	if (!unifiedQ_enabled)
82 		vcn_v4_0_set_dec_ring_funcs(adev);
83 
84 	vcn_v4_0_set_enc_ring_funcs(adev);
85 	vcn_v4_0_set_irq_funcs(adev);
86 
87 	return 0;
88 }
89 
90 static void amdgpu_vcn_setup_unified_queue_ucode(struct amdgpu_device *adev)
91 {
92 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
93 		const struct common_firmware_header *hdr;
94 
95 		hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
96 		adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].ucode_id = AMDGPU_UCODE_ID_VCN;
97 		adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].fw = adev->vcn.fw;
98 		adev->firmware.fw_size +=
99 			ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
100 		DRM_INFO("PSP loading VCN firmware\n");
101 	}
102 }
103 
104 /**
105  * vcn_v4_0_sw_init - sw init for VCN block
106  *
107  * @handle: amdgpu_device pointer
108  *
109  * Load firmware and sw initialization
110  */
111 static int vcn_v4_0_sw_init(void *handle)
112 {
113 	struct amdgpu_ring *ring;
114 	int i, j, r;
115 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
116 
117 	r = amdgpu_vcn_sw_init(adev);
118 	if (r)
119 		return r;
120 
121 	if (unifiedQ_enabled)
122 		amdgpu_vcn_setup_unified_queue_ucode(adev);
123 	else
124 		amdgpu_vcn_setup_ucode(adev);
125 
126 	r = amdgpu_vcn_resume(adev);
127 	if (r)
128 		return r;
129 
130 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
131 		volatile struct amdgpu_vcn4_fw_shared *fw_shared;
132 		if (adev->vcn.harvest_config & (1 << i))
133 			continue;
134 		/* VCN DEC TRAP */
135 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
136 				VCN_4_0__SRCID__UVD_TRAP, &adev->vcn.inst[i].irq);
137 		if (r)
138 			return r;
139 
140 		atomic_set(&adev->vcn.inst[i].sched_score, 0);
141 		if (!unifiedQ_enabled) {
142 			ring = &adev->vcn.inst[i].ring_dec;
143 			ring->use_doorbell = true;
144 
145 			/* VCN4 doorbell layout
146 			 * 1: VCN_JPEG_DB_CTRL UVD_JRBC_RB_WPTR; (jpeg)
147 			 * 2: VCN_RB1_DB_CTRL  UVD_RB_WPTR; (decode/encode for unified queue)
148 			 * 3: VCN_RB2_DB_CTRL  UVD_RB_WPTR2; (encode only for swqueue)
149 			 * 4: VCN_RB3_DB_CTRL  UVD_RB_WPTR3; (Reserved)
150 			 * 5: VCN_RB4_DB_CTRL  UVD_RB_WPTR4; (decode only for swqueue)
151 			 */
152 
153 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1)
154 						+ 5 + 8 * i;
155 
156 			sprintf(ring->name, "vcn_dec_%d", i);
157 			r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
158 					     AMDGPU_RING_PRIO_DEFAULT,
159 					     &adev->vcn.inst[i].sched_score);
160 			if (r)
161 				return r;
162 		}
163 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
164 			/* VCN ENC TRAP */
165 			r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
166 				j + VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);
167 			if (r)
168 				return r;
169 
170 			ring = &adev->vcn.inst[i].ring_enc[j];
171 			ring->use_doorbell = true;
172 
173 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + j + 8 * i;
174 
175 			if (unifiedQ_enabled) {
176 				sprintf(ring->name, "vcn_unified%d", i);
177 				r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0,
178 				     AMDGPU_RING_PRIO_DEFAULT, NULL);
179 			} else {
180 				enum amdgpu_ring_priority_level hw_prio;
181 
182 				hw_prio = amdgpu_vcn_get_enc_ring_prio(j);
183 				sprintf(ring->name, "vcn_enc_%d.%d", i, j);
184 				r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
185 					     hw_prio, &adev->vcn.inst[i].sched_score);
186 			}
187 			if (r)
188 				return r;
189 		}
190 
191 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
192 		fw_shared->present_flag_0 = 0;
193 
194 		if (unifiedQ_enabled) {
195 			fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_FW_SHARED_FLAG_0_UNIFIED_QUEUE);
196 			fw_shared->sq.is_enabled = 1;
197 		}
198 
199 		if (amdgpu_vcnfw_log)
200 			amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
201 	}
202 
203 	if (!unifiedQ_enabled) {
204 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
205 			adev->vcn.pause_dpg_mode = vcn_v4_0_pause_dpg_mode;
206 	}
207 	return 0;
208 }
209 
210 /**
211  * vcn_v4_0_sw_fini - sw fini for VCN block
212  *
213  * @handle: amdgpu_device pointer
214  *
215  * VCN suspend and free up sw allocation
216  */
217 static int vcn_v4_0_sw_fini(void *handle)
218 {
219 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
220 	int i, r, idx;
221 
222 	if (drm_dev_enter(&adev->ddev, &idx)) {
223                 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
224                         volatile struct amdgpu_vcn4_fw_shared *fw_shared;
225 
226                         if (adev->vcn.harvest_config & (1 << i))
227                                 continue;
228 
229                         fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
230                         fw_shared->present_flag_0 = 0;
231                         fw_shared->sq.is_enabled = 0;
232                 }
233 
234                 drm_dev_exit(idx);
235         }
236 
237 	r = amdgpu_vcn_suspend(adev);
238 	if (r)
239 		return r;
240 
241 	r = amdgpu_vcn_sw_fini(adev);
242 
243 	return r;
244 }
245 
246 /**
247  * vcn_v4_0_hw_init - start and test VCN block
248  *
249  * @handle: amdgpu_device pointer
250  *
251  * Initialize the hardware, boot up the VCPU and do some testing
252  */
253 static int vcn_v4_0_hw_init(void *handle)
254 {
255 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
256 	struct amdgpu_ring *ring;
257 	int i, j, r;
258 
259 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
260 		if (adev->vcn.harvest_config & (1 << i))
261 			continue;
262 		if (unifiedQ_enabled)
263 			ring = &adev->vcn.inst[i].ring_enc[0];
264 		else
265 			ring = &adev->vcn.inst[i].ring_dec;
266 
267 		adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
268 				((adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i), i);
269 
270 		r = amdgpu_ring_test_helper(ring);
271 		if (r)
272 			goto done;
273 
274 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
275 			ring = &adev->vcn.inst[i].ring_enc[j];
276 			r = amdgpu_ring_test_helper(ring);
277 			if (r)
278 				goto done;
279 		}
280 	}
281 
282 done:
283 	if (!r)
284 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
285 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
286 
287 	return r;
288 }
289 
290 /**
291  * vcn_v4_0_hw_fini - stop the hardware block
292  *
293  * @handle: amdgpu_device pointer
294  *
295  * Stop the VCN block, mark ring as not ready any more
296  */
297 static int vcn_v4_0_hw_fini(void *handle)
298 {
299 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
300 	int i;
301 
302 	cancel_delayed_work_sync(&adev->vcn.idle_work);
303 
304 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
305 		if (adev->vcn.harvest_config & (1 << i))
306 			continue;
307 
308 		if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
309                         (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
310                                 RREG32_SOC15(VCN, i, regUVD_STATUS))) {
311                         vcn_v4_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
312 		}
313 	}
314 
315 	return 0;
316 }
317 
318 /**
319  * vcn_v4_0_suspend - suspend VCN block
320  *
321  * @handle: amdgpu_device pointer
322  *
323  * HW fini and suspend VCN block
324  */
325 static int vcn_v4_0_suspend(void *handle)
326 {
327 	int r;
328 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
329 
330 	r = vcn_v4_0_hw_fini(adev);
331 	if (r)
332 		return r;
333 
334 	r = amdgpu_vcn_suspend(adev);
335 
336 	return r;
337 }
338 
339 /**
340  * vcn_v4_0_resume - resume VCN block
341  *
342  * @handle: amdgpu_device pointer
343  *
344  * Resume firmware and hw init VCN block
345  */
346 static int vcn_v4_0_resume(void *handle)
347 {
348 	int r;
349 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
350 
351 	r = amdgpu_vcn_resume(adev);
352 	if (r)
353 		return r;
354 
355 	r = vcn_v4_0_hw_init(adev);
356 
357 	return r;
358 }
359 
360 /**
361  * vcn_v4_0_mc_resume - memory controller programming
362  *
363  * @adev: amdgpu_device pointer
364  * @inst: instance number
365  *
366  * Let the VCN memory controller know it's offsets
367  */
368 static void vcn_v4_0_mc_resume(struct amdgpu_device *adev, int inst)
369 {
370 	uint32_t offset, size;
371 	const struct common_firmware_header *hdr;
372 
373 	hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
374 	size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
375 
376 	/* cache window 0: fw */
377 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
378 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
379 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo));
380 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
381 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi));
382 		WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, 0);
383 		offset = 0;
384 	} else {
385 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
386 			lower_32_bits(adev->vcn.inst[inst].gpu_addr));
387 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
388 			upper_32_bits(adev->vcn.inst[inst].gpu_addr));
389 		offset = size;
390                 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
391 	}
392 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE0, size);
393 
394 	/* cache window 1: stack */
395 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
396 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
397 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
398 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
399 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET1, 0);
400 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
401 
402 	/* cache window 2: context */
403 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
404 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
405 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
406 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
407 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET2, 0);
408 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
409 
410 	/* non-cache window */
411 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
412 		lower_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
413 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
414 		upper_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
415 	WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_OFFSET0, 0);
416 	WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_SIZE0,
417 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
418 }
419 
420 /**
421  * vcn_v4_0_mc_resume_dpg_mode - memory controller programming for dpg mode
422  *
423  * @adev: amdgpu_device pointer
424  * @inst_idx: instance number index
425  * @indirect: indirectly write sram
426  *
427  * Let the VCN memory controller know it's offsets with dpg mode
428  */
429 static void vcn_v4_0_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
430 {
431 	uint32_t offset, size;
432 	const struct common_firmware_header *hdr;
433 	hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
434 	size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
435 
436 	/* cache window 0: fw */
437 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
438 		if (!indirect) {
439 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
440 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
441 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
442 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
443 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
444 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
445 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
446 				VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
447 		} else {
448 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
449 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
450 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
451 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
452 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
453 				VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
454 		}
455 		offset = 0;
456 	} else {
457 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
458 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
459 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
460 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
461 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
462 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
463 		offset = size;
464 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
465 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0),
466 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
467 
468 	}
469 
470 	if (!indirect)
471 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
472 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
473 	else
474 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
475 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
476 
477 	/* cache window 1: stack */
478 	if (!indirect) {
479 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
480 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
481 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
482 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
483 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
484 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
485 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
486 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
487 	} else {
488 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
489 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
490 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
491 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
492 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
493 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
494 	}
495 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
496 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
497 
498 	/* cache window 2: context */
499 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
500 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
501 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
502 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
503 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
504 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
505 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
506 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
507 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
508 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
509 
510 	/* non-cache window */
511 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
512 			VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
513 			lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
514 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
515 			VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
516 			upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
517 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
518 			VCN, inst_idx, regUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
519 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
520 			VCN, inst_idx, regUVD_VCPU_NONCACHE_SIZE0),
521 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)), 0, indirect);
522 
523 	/* VCN global tiling registers */
524 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
525 		VCN, 0, regUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
526 }
527 
528 /**
529  * vcn_v4_0_disable_static_power_gating - disable VCN static power gating
530  *
531  * @adev: amdgpu_device pointer
532  * @inst: instance number
533  *
534  * Disable static power gating for VCN block
535  */
536 static void vcn_v4_0_disable_static_power_gating(struct amdgpu_device *adev, int inst)
537 {
538 	uint32_t data = 0;
539 
540 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
541 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
542 			| 1 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
543 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
544 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
545 			| 2 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
546 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
547 			| 2 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
548 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
549 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
550 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
551 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
552 			| 2 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
553 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
554 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
555 
556 		WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
557 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS,
558 			UVD_PGFSM_STATUS__UVDM_UVDU_UVDLM_PWR_ON_3_0, 0x3F3FFFFF);
559 	} else {
560 		uint32_t value;
561 
562 		value = (inst) ? 0x2200800 : 0;
563 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
564 			| 1 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
565 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
566 			| 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
567 			| 1 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
568 			| 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
569 			| 1 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
570 			| 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
571 			| 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
572 			| 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
573 			| 1 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
574 			| 1 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
575 			| 1 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
576 			| 1 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
577 
578                 WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
579                 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS, value,  0x3F3FFFFF);
580         }
581 
582         data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
583         data &= ~0x103;
584         if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
585                 data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
586                         UVD_POWER_STATUS__UVD_PG_EN_MASK;
587 
588         WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
589 
590         return;
591 }
592 
593 /**
594  * vcn_v4_0_enable_static_power_gating - enable VCN static power gating
595  *
596  * @adev: amdgpu_device pointer
597  * @inst: instance number
598  *
599  * Enable static power gating for VCN block
600  */
601 static void vcn_v4_0_enable_static_power_gating(struct amdgpu_device *adev, int inst)
602 {
603 	uint32_t data;
604 
605 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
606 		/* Before power off, this indicator has to be turned on */
607 		data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
608 		data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
609 		data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
610 		WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
611 
612 		data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
613 			| 2 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
614 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
615 			| 2 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
616 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
617 			| 2 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
618 			| 2 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
619 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
620 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
621 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
622 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
623 			| 2 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
624 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
625 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
626 		WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
627 
628 		data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
629 			| 2 << UVD_PGFSM_STATUS__UVDS_PWR_STATUS__SHIFT
630 			| 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
631 			| 2 << UVD_PGFSM_STATUS__UVDTC_PWR_STATUS__SHIFT
632 			| 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
633 			| 2 << UVD_PGFSM_STATUS__UVDTA_PWR_STATUS__SHIFT
634 			| 2 << UVD_PGFSM_STATUS__UVDLM_PWR_STATUS__SHIFT
635 			| 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
636 			| 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
637 			| 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT
638 			| 2 << UVD_PGFSM_STATUS__UVDAB_PWR_STATUS__SHIFT
639 			| 2 << UVD_PGFSM_STATUS__UVDTB_PWR_STATUS__SHIFT
640 			| 2 << UVD_PGFSM_STATUS__UVDNA_PWR_STATUS__SHIFT
641 			| 2 << UVD_PGFSM_STATUS__UVDNB_PWR_STATUS__SHIFT);
642 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS, data, 0x3F3FFFFF);
643 	}
644 
645         return;
646 }
647 
648 /**
649  * vcn_v4_0_disable_clock_gating - disable VCN clock gating
650  *
651  * @adev: amdgpu_device pointer
652  * @inst: instance number
653  *
654  * Disable clock gating for VCN block
655  */
656 static void vcn_v4_0_disable_clock_gating(struct amdgpu_device *adev, int inst)
657 {
658 	uint32_t data;
659 
660 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
661 		return;
662 
663 	/* VCN disable CGC */
664 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
665 	data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
666 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
667 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
668 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
669 
670 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_GATE);
671 	data &= ~(UVD_CGC_GATE__SYS_MASK
672 		| UVD_CGC_GATE__UDEC_MASK
673 		| UVD_CGC_GATE__MPEG2_MASK
674 		| UVD_CGC_GATE__REGS_MASK
675 		| UVD_CGC_GATE__RBC_MASK
676 		| UVD_CGC_GATE__LMI_MC_MASK
677 		| UVD_CGC_GATE__LMI_UMC_MASK
678 		| UVD_CGC_GATE__IDCT_MASK
679 		| UVD_CGC_GATE__MPRD_MASK
680 		| UVD_CGC_GATE__MPC_MASK
681 		| UVD_CGC_GATE__LBSI_MASK
682 		| UVD_CGC_GATE__LRBBM_MASK
683 		| UVD_CGC_GATE__UDEC_RE_MASK
684 		| UVD_CGC_GATE__UDEC_CM_MASK
685 		| UVD_CGC_GATE__UDEC_IT_MASK
686 		| UVD_CGC_GATE__UDEC_DB_MASK
687 		| UVD_CGC_GATE__UDEC_MP_MASK
688 		| UVD_CGC_GATE__WCB_MASK
689 		| UVD_CGC_GATE__VCPU_MASK
690 		| UVD_CGC_GATE__MMSCH_MASK);
691 
692 	WREG32_SOC15(VCN, inst, regUVD_CGC_GATE, data);
693 	SOC15_WAIT_ON_RREG(VCN, inst, regUVD_CGC_GATE, 0,  0xFFFFFFFF);
694 
695 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
696 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
697 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
698 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
699 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
700 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
701 		| UVD_CGC_CTRL__SYS_MODE_MASK
702 		| UVD_CGC_CTRL__UDEC_MODE_MASK
703 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
704 		| UVD_CGC_CTRL__REGS_MODE_MASK
705 		| UVD_CGC_CTRL__RBC_MODE_MASK
706 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
707 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
708 		| UVD_CGC_CTRL__IDCT_MODE_MASK
709 		| UVD_CGC_CTRL__MPRD_MODE_MASK
710 		| UVD_CGC_CTRL__MPC_MODE_MASK
711 		| UVD_CGC_CTRL__LBSI_MODE_MASK
712 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
713 		| UVD_CGC_CTRL__WCB_MODE_MASK
714 		| UVD_CGC_CTRL__VCPU_MODE_MASK
715 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
716 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
717 
718 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE);
719 	data |= (UVD_SUVD_CGC_GATE__SRE_MASK
720 		| UVD_SUVD_CGC_GATE__SIT_MASK
721 		| UVD_SUVD_CGC_GATE__SMP_MASK
722 		| UVD_SUVD_CGC_GATE__SCM_MASK
723 		| UVD_SUVD_CGC_GATE__SDB_MASK
724 		| UVD_SUVD_CGC_GATE__SRE_H264_MASK
725 		| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
726 		| UVD_SUVD_CGC_GATE__SIT_H264_MASK
727 		| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
728 		| UVD_SUVD_CGC_GATE__SCM_H264_MASK
729 		| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
730 		| UVD_SUVD_CGC_GATE__SDB_H264_MASK
731 		| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
732 		| UVD_SUVD_CGC_GATE__SCLR_MASK
733 		| UVD_SUVD_CGC_GATE__UVD_SC_MASK
734 		| UVD_SUVD_CGC_GATE__ENT_MASK
735 		| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
736 		| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
737 		| UVD_SUVD_CGC_GATE__SITE_MASK
738 		| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
739 		| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
740 		| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
741 		| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
742 		| UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
743 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE, data);
744 
745 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
746 	data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
747 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
748 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
749 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
750 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
751 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
752 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
753 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
754 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
755 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
756 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
757 }
758 
759 /**
760  * vcn_v4_0_disable_clock_gating_dpg_mode - disable VCN clock gating dpg mode
761  *
762  * @adev: amdgpu_device pointer
763  * @sram_sel: sram select
764  * @inst_idx: instance number index
765  * @indirect: indirectly write sram
766  *
767  * Disable clock gating for VCN block with dpg mode
768  */
769 static void vcn_v4_0_disable_clock_gating_dpg_mode(struct amdgpu_device *adev, uint8_t sram_sel,
770       int inst_idx, uint8_t indirect)
771 {
772 	uint32_t reg_data = 0;
773 
774 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
775 		return;
776 
777 	/* enable sw clock gating control */
778 	reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
779 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
780 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
781 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
782 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
783 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
784 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
785 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
786 		 UVD_CGC_CTRL__SYS_MODE_MASK |
787 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
788 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
789 		 UVD_CGC_CTRL__REGS_MODE_MASK |
790 		 UVD_CGC_CTRL__RBC_MODE_MASK |
791 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
792 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
793 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
794 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
795 		 UVD_CGC_CTRL__MPC_MODE_MASK |
796 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
797 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
798 		 UVD_CGC_CTRL__WCB_MODE_MASK |
799 		 UVD_CGC_CTRL__VCPU_MODE_MASK);
800 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
801 		VCN, inst_idx, regUVD_CGC_CTRL), reg_data, sram_sel, indirect);
802 
803 	/* turn off clock gating */
804 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
805 		VCN, inst_idx, regUVD_CGC_GATE), 0, sram_sel, indirect);
806 
807 	/* turn on SUVD clock gating */
808 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
809 		VCN, inst_idx, regUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
810 
811 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
812 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
813 		VCN, inst_idx, regUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
814 }
815 
816 /**
817  * vcn_v4_0_enable_clock_gating - enable VCN clock gating
818  *
819  * @adev: amdgpu_device pointer
820  * @inst: instance number
821  *
822  * Enable clock gating for VCN block
823  */
824 static void vcn_v4_0_enable_clock_gating(struct amdgpu_device *adev, int inst)
825 {
826 	uint32_t data;
827 
828 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
829 		return;
830 
831 	/* enable VCN CGC */
832 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
833 	data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
834 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
835 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
836 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
837 
838 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
839 	data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
840 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
841 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
842 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
843 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
844 		| UVD_CGC_CTRL__SYS_MODE_MASK
845 		| UVD_CGC_CTRL__UDEC_MODE_MASK
846 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
847 		| UVD_CGC_CTRL__REGS_MODE_MASK
848 		| UVD_CGC_CTRL__RBC_MODE_MASK
849 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
850 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
851 		| UVD_CGC_CTRL__IDCT_MODE_MASK
852 		| UVD_CGC_CTRL__MPRD_MODE_MASK
853 		| UVD_CGC_CTRL__MPC_MODE_MASK
854 		| UVD_CGC_CTRL__LBSI_MODE_MASK
855 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
856 		| UVD_CGC_CTRL__WCB_MODE_MASK
857 		| UVD_CGC_CTRL__VCPU_MODE_MASK
858 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
859 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
860 
861 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
862 	data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
863 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
864 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
865 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
866 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
867 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
868 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
869 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
870 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
871 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
872 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
873 
874 	return;
875 }
876 
877 /**
878  * vcn_v4_0_start_dpg_mode - VCN start with dpg mode
879  *
880  * @adev: amdgpu_device pointer
881  * @inst_idx: instance number index
882  * @indirect: indirectly write sram
883  *
884  * Start VCN block with dpg mode
885  */
886 static int vcn_v4_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
887 {
888 	volatile struct amdgpu_vcn4_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
889 	struct amdgpu_ring *ring;
890 	uint32_t tmp;
891 	int i;
892 
893 	/* disable register anti-hang mechanism */
894 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 1,
895 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
896 	/* enable dynamic power gating mode */
897 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS);
898 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
899 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
900 	WREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS, tmp);
901 
902 	if (indirect)
903 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
904 
905 	/* enable clock gating */
906 	vcn_v4_0_disable_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
907 
908 	/* enable VCPU clock */
909 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
910 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK | UVD_VCPU_CNTL__BLK_RST_MASK;
911 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
912 		VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
913 
914 	/* disable master interupt */
915 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
916 		VCN, inst_idx, regUVD_MASTINT_EN), 0, 0, indirect);
917 
918 	/* setup regUVD_LMI_CTRL */
919 	tmp = (UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
920 		UVD_LMI_CTRL__REQ_MODE_MASK |
921 		UVD_LMI_CTRL__CRC_RESET_MASK |
922 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
923 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
924 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
925 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
926 		0x00100000L);
927 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
928 		VCN, inst_idx, regUVD_LMI_CTRL), tmp, 0, indirect);
929 
930 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
931 		VCN, inst_idx, regUVD_MPC_CNTL),
932 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
933 
934 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
935 		VCN, inst_idx, regUVD_MPC_SET_MUXA0),
936 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
937 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
938 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
939 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
940 
941 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
942 		VCN, inst_idx, regUVD_MPC_SET_MUXB0),
943 		 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
944 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
945 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
946 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
947 
948 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
949 		VCN, inst_idx, regUVD_MPC_SET_MUX),
950 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
951 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
952 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
953 
954 	vcn_v4_0_mc_resume_dpg_mode(adev, inst_idx, indirect);
955 
956 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
957 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
958 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
959 		VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
960 
961 	/* enable LMI MC and UMC channels */
962 	tmp = 0x1f << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT;
963 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
964 		VCN, inst_idx, regUVD_LMI_CTRL2), tmp, 0, indirect);
965 
966 	/* enable master interrupt */
967 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
968 		VCN, inst_idx, regUVD_MASTINT_EN),
969 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
970 
971 
972 	if (indirect)
973 		psp_update_vcn_sram(adev, inst_idx, adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
974 			(uint32_t)((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
975 				(uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr));
976 
977 	if (unifiedQ_enabled) {
978 		ring = &adev->vcn.inst[inst_idx].ring_enc[0];
979 		fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
980 	} else
981 		ring = &adev->vcn.inst[inst_idx].ring_dec;
982 
983 	WREG32_SOC15(VCN, inst_idx, regVCN_RB4_DB_CTRL,
984 		ring->doorbell_index << VCN_RB4_DB_CTRL__OFFSET__SHIFT |
985 		VCN_RB4_DB_CTRL__EN_MASK);
986 
987 	/* program the RB_BASE for ring buffer */
988 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_LO4,
989 		lower_32_bits(ring->gpu_addr));
990 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_HI4,
991 		upper_32_bits(ring->gpu_addr));
992 
993 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_SIZE4, ring->ring_size / sizeof(uint32_t));
994 
995 	/* reseting ring, fw should not check RB ring */
996 	tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
997 	tmp &= ~(VCN_RB_ENABLE__RB4_EN_MASK);
998 	WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
999 
1000 	/* Initialize the ring buffer's read and write pointers */
1001 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR4);
1002 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR4, tmp);
1003 	ring->wptr = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR4);
1004 
1005 	tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
1006 	tmp |= VCN_RB_ENABLE__RB4_EN_MASK;
1007 	WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
1008 
1009 	WREG32_SOC15(VCN, inst_idx, regUVD_SCRATCH2, 0);
1010 
1011 	if (unifiedQ_enabled)
1012 		fw_shared->sq.queue_mode &= ~FW_QUEUE_RING_RESET;
1013 
1014 	for (i = 0; i < adev->vcn.num_enc_rings; i++) {
1015 		ring = &adev->vcn.inst[inst_idx].ring_enc[i];
1016 
1017 		if (i) {
1018 			ring = &adev->vcn.inst[inst_idx].ring_enc[1];
1019 
1020 			WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_LO2, ring->gpu_addr);
1021 			WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1022 			WREG32_SOC15(VCN, inst_idx, regUVD_RB_SIZE2, ring->ring_size / 4);
1023 			tmp= RREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR2);
1024 			WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR2, tmp);
1025 			ring->wptr = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR2);
1026 
1027 			WREG32_SOC15(VCN, inst_idx, regVCN_RB2_DB_CTRL,
1028 				ring->doorbell_index << VCN_RB2_DB_CTRL__OFFSET__SHIFT |
1029 				VCN_RB2_DB_CTRL__EN_MASK);
1030 		} else {
1031 			ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1032 
1033 			WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_LO, ring->gpu_addr);
1034 			WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1035 			WREG32_SOC15(VCN, inst_idx, regUVD_RB_SIZE, ring->ring_size / 4);
1036 			tmp= RREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR);
1037 			WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, tmp);
1038 			ring->wptr = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
1039 
1040 			WREG32_SOC15(VCN, inst_idx, regVCN_RB1_DB_CTRL,
1041 				ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
1042 				VCN_RB1_DB_CTRL__EN_MASK);
1043 		}
1044 	}
1045 	return 0;
1046 }
1047 
1048 
1049 /**
1050  * vcn_v4_0_start - VCN start
1051  *
1052  * @adev: amdgpu_device pointer
1053  *
1054  * Start VCN block
1055  */
1056 static int vcn_v4_0_start(struct amdgpu_device *adev)
1057 {
1058 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1059 	struct amdgpu_ring *ring;
1060 	uint32_t tmp;
1061 	int i, j, k, r;
1062 
1063 	if (adev->pm.dpm_enabled)
1064 		amdgpu_dpm_enable_uvd(adev, true);
1065 
1066 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1067 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1068 			r = vcn_v4_0_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
1069 			continue;
1070 		}
1071 
1072 		/* disable VCN power gating */
1073 		vcn_v4_0_disable_static_power_gating(adev, i);
1074 
1075 		/* set VCN status busy */
1076 		tmp = RREG32_SOC15(VCN, i, regUVD_STATUS) | UVD_STATUS__UVD_BUSY;
1077 		WREG32_SOC15(VCN, i, regUVD_STATUS, tmp);
1078 
1079 		/*SW clock gating */
1080 		vcn_v4_0_disable_clock_gating(adev, i);
1081 
1082 		/* enable VCPU clock */
1083 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1084 			UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1085 
1086 		/* disable master interrupt */
1087 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN), 0,
1088 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1089 
1090 		/* enable LMI MC and UMC channels */
1091 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_LMI_CTRL2), 0,
1092 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1093 
1094 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1095 		tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1096 		tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1097 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1098 
1099 		/* setup regUVD_LMI_CTRL */
1100 		tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL);
1101 		WREG32_SOC15(VCN, i, regUVD_LMI_CTRL, tmp |
1102 		UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
1103 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1104 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1105 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1106 
1107 		/* setup regUVD_MPC_CNTL */
1108 		tmp = RREG32_SOC15(VCN, i, regUVD_MPC_CNTL);
1109 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1110 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1111 		WREG32_SOC15(VCN, i, regUVD_MPC_CNTL, tmp);
1112 
1113 		/* setup UVD_MPC_SET_MUXA0 */
1114 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXA0,
1115 			((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1116 			(0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1117 			(0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1118 			(0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1119 
1120 		/* setup UVD_MPC_SET_MUXB0 */
1121 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXB0,
1122 			((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1123 			(0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1124 			(0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1125 			(0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1126 
1127 		/* setup UVD_MPC_SET_MUX */
1128 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUX,
1129 			((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1130 			(0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1131 			(0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1132 
1133 		vcn_v4_0_mc_resume(adev, i);
1134 
1135 		/* VCN global tiling registers */
1136 		WREG32_SOC15(VCN, i, regUVD_GFX10_ADDR_CONFIG,
1137 			adev->gfx.config.gb_addr_config);
1138 
1139 		/* unblock VCPU register access */
1140 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL), 0,
1141 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1142 
1143 		/* release VCPU reset to boot */
1144 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1145 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1146 
1147 		for (j = 0; j < 10; ++j) {
1148 			uint32_t status;
1149 
1150 			for (k = 0; k < 100; ++k) {
1151 				status = RREG32_SOC15(VCN, i, regUVD_STATUS);
1152 				if (status & 2)
1153 					break;
1154 				mdelay(10);
1155 				if (amdgpu_emu_mode==1)
1156 					msleep(1);
1157 			}
1158 
1159 			if (amdgpu_emu_mode==1) {
1160 				if (status & 2) {
1161 					r = 0;
1162 					break;
1163 				}
1164 			} else {
1165 				r = 0;
1166 				if (status & 2)
1167 					break;
1168 
1169 				dev_err(adev->dev, "VCN[%d] decode not responding, trying to reset the VCPU!!!\n", i);
1170 					WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1171 					UVD_VCPU_CNTL__BLK_RST_MASK,
1172 					~UVD_VCPU_CNTL__BLK_RST_MASK);
1173 				mdelay(10);
1174 				WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1175 					~UVD_VCPU_CNTL__BLK_RST_MASK);
1176 
1177 				mdelay(10);
1178 				r = -1;
1179 			}
1180 		}
1181 
1182 		if (r) {
1183 			dev_err(adev->dev, "VCN[%d] decode not responding, giving up!!!\n", i);
1184 			return r;
1185 		}
1186 
1187 		/* enable master interrupt */
1188 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN),
1189 			UVD_MASTINT_EN__VCPU_EN_MASK,
1190 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1191 
1192 		/* clear the busy bit of VCN_STATUS */
1193 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_STATUS), 0,
1194 			~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1195 
1196 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1197 		if (unifiedQ_enabled) {
1198 			ring = &adev->vcn.inst[i].ring_enc[0];
1199 			fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
1200 		} else {
1201 			ring = &adev->vcn.inst[i].ring_dec;
1202 
1203 			WREG32_SOC15(VCN, i, regVCN_RB4_DB_CTRL,
1204 				ring->doorbell_index << VCN_RB4_DB_CTRL__OFFSET__SHIFT |
1205 				VCN_RB4_DB_CTRL__EN_MASK);
1206 
1207 			/* program the RB_BASE for ring buffer */
1208 			WREG32_SOC15(VCN, i, regUVD_RB_BASE_LO4,
1209 				lower_32_bits(ring->gpu_addr));
1210 			WREG32_SOC15(VCN, i, regUVD_RB_BASE_HI4,
1211 				upper_32_bits(ring->gpu_addr));
1212 
1213 			WREG32_SOC15(VCN, i, regUVD_RB_SIZE4, ring->ring_size / sizeof(uint32_t));
1214 
1215 			/* resetting ring, fw should not check RB ring */
1216 			tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1217 			tmp &= ~(VCN_RB_ENABLE__RB4_EN_MASK);
1218 			WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1219 
1220 			/* Initialize the ring buffer's read and write pointers */
1221 			tmp = RREG32_SOC15(VCN, i, regUVD_RB_RPTR4);
1222 			WREG32_SOC15(VCN, i, regUVD_RB_WPTR4, tmp);
1223 			ring->wptr = RREG32_SOC15(VCN, i, regUVD_RB_WPTR4);
1224 
1225 			tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1226 			tmp |= VCN_RB_ENABLE__RB4_EN_MASK;
1227 			WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1228 
1229 			ring->wptr = RREG32_SOC15(VCN, i, regUVD_RB_RPTR4);
1230 		}
1231 		ring = &adev->vcn.inst[i].ring_enc[0];
1232 		WREG32_SOC15(VCN, i, regVCN_RB1_DB_CTRL,
1233 			ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
1234 			VCN_RB1_DB_CTRL__EN_MASK);
1235 		tmp = RREG32_SOC15(VCN, i, regUVD_RB_RPTR);
1236 		WREG32_SOC15(VCN, i, regUVD_RB_WPTR, tmp);
1237 		ring->wptr = RREG32_SOC15(VCN, i, regUVD_RB_WPTR);
1238 		WREG32_SOC15(VCN, i, regUVD_RB_BASE_LO, ring->gpu_addr);
1239 		WREG32_SOC15(VCN, i, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1240 		WREG32_SOC15(VCN, i, regUVD_RB_SIZE, ring->ring_size / 4);
1241 		if (unifiedQ_enabled)
1242 			fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
1243 		else {
1244 			ring = &adev->vcn.inst[i].ring_enc[1];
1245 			WREG32_SOC15(VCN, i, regVCN_RB2_DB_CTRL,
1246 				ring->doorbell_index << VCN_RB2_DB_CTRL__OFFSET__SHIFT |
1247 				VCN_RB2_DB_CTRL__EN_MASK);
1248 			tmp = RREG32_SOC15(VCN, i, regUVD_RB_RPTR2);
1249 			WREG32_SOC15(VCN, i, regUVD_RB_WPTR2, tmp);
1250 			ring->wptr = RREG32_SOC15(VCN, i, regUVD_RB_WPTR2);
1251 			WREG32_SOC15(VCN, i, regUVD_RB_BASE_LO2, ring->gpu_addr);
1252 			WREG32_SOC15(VCN, i, regUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1253 			WREG32_SOC15(VCN, i, regUVD_RB_SIZE2, ring->ring_size / 4);
1254 		}
1255 	}
1256 
1257 	return 0;
1258 }
1259 
1260 /**
1261  * vcn_v4_0_stop_dpg_mode - VCN stop with dpg mode
1262  *
1263  * @adev: amdgpu_device pointer
1264  * @inst_idx: instance number index
1265  *
1266  * Stop VCN block with dpg mode
1267  */
1268 static int vcn_v4_0_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1269 {
1270 	uint32_t tmp;
1271 
1272 	/* Wait for power status to be 1 */
1273 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1274 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1275 
1276 	/* wait for read ptr to be equal to write ptr */
1277 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
1278 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1279 
1280 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR2);
1281 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
1282 
1283 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR4);
1284 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_RB_RPTR4, tmp, 0xFFFFFFFF);
1285 
1286 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1287 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1288 
1289 	/* disable dynamic power gating mode */
1290 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 0,
1291 		~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1292 	return 0;
1293 }
1294 
1295 /**
1296  * vcn_v4_0_stop - VCN stop
1297  *
1298  * @adev: amdgpu_device pointer
1299  *
1300  * Stop VCN block
1301  */
1302 static int vcn_v4_0_stop(struct amdgpu_device *adev)
1303 {
1304 	uint32_t tmp;
1305 	int i, r = 0;
1306 
1307 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1308 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1309 			r = vcn_v4_0_stop_dpg_mode(adev, i);
1310 			continue;
1311 		}
1312 
1313 		/* wait for vcn idle */
1314 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1315 		if (r)
1316 			return r;
1317 
1318 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1319 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1320 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1321 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1322 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1323 		if (r)
1324 			return r;
1325 
1326 		/* disable LMI UMC channel */
1327 		tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL2);
1328 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1329 		WREG32_SOC15(VCN, i, regUVD_LMI_CTRL2, tmp);
1330 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK |
1331 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1332 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1333 		if (r)
1334 			return r;
1335 
1336 		/* block VCPU register access */
1337 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL),
1338 				UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1339 				~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1340 
1341 		/* reset VCPU */
1342 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1343 				UVD_VCPU_CNTL__BLK_RST_MASK,
1344 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1345 
1346 		/* disable VCPU clock */
1347 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1348 				~(UVD_VCPU_CNTL__CLK_EN_MASK));
1349 
1350 		/* apply soft reset */
1351 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1352 		tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1353 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1354 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1355 		tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1356 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1357 
1358 		/* clear status */
1359 		WREG32_SOC15(VCN, i, regUVD_STATUS, 0);
1360 
1361 		/* apply HW clock gating */
1362 		vcn_v4_0_enable_clock_gating(adev, i);
1363 
1364 		/* enable VCN power gating */
1365 		vcn_v4_0_enable_static_power_gating(adev, i);
1366 	}
1367 
1368 	if (adev->pm.dpm_enabled)
1369 		amdgpu_dpm_enable_uvd(adev, false);
1370 
1371 	return 0;
1372 }
1373 
1374 /**
1375  * vcn_v4_0_pause_dpg_mode - VCN pause with dpg mode
1376  *
1377  * @adev: amdgpu_device pointer
1378  * @inst_idx: instance number index
1379  * @new_state: pause state
1380  *
1381  * Pause dpg mode for VCN block
1382  */
1383 static int vcn_v4_0_pause_dpg_mode(struct amdgpu_device *adev, int inst_idx,
1384       struct dpg_pause_state *new_state)
1385 {
1386 	uint32_t reg_data = 0;
1387 	int ret_code;
1388 
1389 	/* pause/unpause if state is changed */
1390 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1391 		DRM_DEV_DEBUG(adev->dev, "dpg pause state changed %d -> %d",
1392 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1393 		reg_data = RREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE) &
1394 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1395 
1396 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1397 			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 0x1,
1398 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1399 
1400 			if (!ret_code) {
1401 				/* pause DPG */
1402 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1403 				WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1404 
1405 				/* wait for ACK */
1406 				SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_DPG_PAUSE,
1407 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1408 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1409 
1410 				SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS,
1411 					UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1412 			}
1413 		} else {
1414 			/* unpause dpg, no need to wait */
1415 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1416 			WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1417 			SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 0x1,
1418 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1419 		}
1420 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1421 	}
1422 
1423 	return 0;
1424 }
1425 
1426 /**
1427  * vcn_v4_0_dec_ring_get_rptr - get read pointer
1428  *
1429  * @ring: amdgpu_ring pointer
1430  *
1431  * Returns the current hardware read pointer
1432  */
1433 static uint64_t vcn_v4_0_dec_ring_get_rptr(struct amdgpu_ring *ring)
1434 {
1435 	struct amdgpu_device *adev = ring->adev;
1436 
1437 	return RREG32_SOC15(VCN, ring->me, regUVD_RB_RPTR4);
1438 }
1439 
1440 /**
1441  * vcn_v4_0_dec_ring_get_wptr - get write pointer
1442  *
1443  * @ring: amdgpu_ring pointer
1444  *
1445  * Returns the current hardware write pointer
1446  */
1447 static uint64_t vcn_v4_0_dec_ring_get_wptr(struct amdgpu_ring *ring)
1448 {
1449 	struct amdgpu_device *adev = ring->adev;
1450 
1451 	if (ring->use_doorbell)
1452 		return *ring->wptr_cpu_addr;
1453 	else
1454 		return RREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR4);
1455 }
1456 
1457 /**
1458  * vcn_v4_0_dec_ring_set_wptr - set write pointer
1459  *
1460  * @ring: amdgpu_ring pointer
1461  *
1462  * Commits the write pointer to the hardware
1463  */
1464 static void vcn_v4_0_dec_ring_set_wptr(struct amdgpu_ring *ring)
1465 {
1466 	struct amdgpu_device *adev = ring->adev;
1467 
1468 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1469 		WREG32_SOC15(VCN, ring->me, regUVD_SCRATCH2,
1470 			lower_32_bits(ring->wptr));
1471 	}
1472 
1473 	if (ring->use_doorbell) {
1474 		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1475 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1476 	} else {
1477 		WREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR4, lower_32_bits(ring->wptr));
1478 	}
1479 }
1480 
1481 static const struct amdgpu_ring_funcs vcn_v4_0_dec_sw_ring_vm_funcs = {
1482 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1483 	.align_mask = 0x3f,
1484 	.nop = VCN_DEC_SW_CMD_NO_OP,
1485 	.vmhub = AMDGPU_MMHUB_0,
1486 	.get_rptr = vcn_v4_0_dec_ring_get_rptr,
1487 	.get_wptr = vcn_v4_0_dec_ring_get_wptr,
1488 	.set_wptr = vcn_v4_0_dec_ring_set_wptr,
1489 	.emit_frame_size =
1490 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1491 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1492 		VCN_SW_RING_EMIT_FRAME_SIZE,
1493 	.emit_ib_size = 5, /* vcn_dec_sw_ring_emit_ib */
1494 	.emit_ib = vcn_dec_sw_ring_emit_ib,
1495 	.emit_fence = vcn_dec_sw_ring_emit_fence,
1496 	.emit_vm_flush = vcn_dec_sw_ring_emit_vm_flush,
1497 	.test_ring = amdgpu_vcn_dec_sw_ring_test_ring,
1498 	.test_ib = amdgpu_vcn_dec_sw_ring_test_ib,
1499 	.insert_nop = amdgpu_ring_insert_nop,
1500 	.insert_end = vcn_dec_sw_ring_insert_end,
1501 	.pad_ib = amdgpu_ring_generic_pad_ib,
1502 	.begin_use = amdgpu_vcn_ring_begin_use,
1503 	.end_use = amdgpu_vcn_ring_end_use,
1504 	.emit_wreg = vcn_dec_sw_ring_emit_wreg,
1505 	.emit_reg_wait = vcn_dec_sw_ring_emit_reg_wait,
1506 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1507 };
1508 
1509 /**
1510  * vcn_v4_0_enc_ring_get_rptr - get enc read pointer
1511  *
1512  * @ring: amdgpu_ring pointer
1513  *
1514  * Returns the current hardware enc read pointer
1515  */
1516 static uint64_t vcn_v4_0_enc_ring_get_rptr(struct amdgpu_ring *ring)
1517 {
1518 	struct amdgpu_device *adev = ring->adev;
1519 
1520 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0])
1521 		return RREG32_SOC15(VCN, ring->me, regUVD_RB_RPTR);
1522 	else
1523 		return RREG32_SOC15(VCN, ring->me, regUVD_RB_RPTR2);
1524 }
1525 
1526 /**
1527  * vcn_v4_0_enc_ring_get_wptr - get enc write pointer
1528  *
1529  * @ring: amdgpu_ring pointer
1530  *
1531  * Returns the current hardware enc write pointer
1532  */
1533 static uint64_t vcn_v4_0_enc_ring_get_wptr(struct amdgpu_ring *ring)
1534 {
1535 	struct amdgpu_device *adev = ring->adev;
1536 
1537 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1538 		if (ring->use_doorbell)
1539 			return *ring->wptr_cpu_addr;
1540 		else
1541 			return RREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR);
1542 	} else {
1543 		if (ring->use_doorbell)
1544 			return *ring->wptr_cpu_addr;
1545 		else
1546 			return RREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR2);
1547 	}
1548 }
1549 
1550 /**
1551  * vcn_v4_0_enc_ring_set_wptr - set enc write pointer
1552  *
1553  * @ring: amdgpu_ring pointer
1554  *
1555  * Commits the enc write pointer to the hardware
1556  */
1557 static void vcn_v4_0_enc_ring_set_wptr(struct amdgpu_ring *ring)
1558 {
1559 	struct amdgpu_device *adev = ring->adev;
1560 
1561 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1562 		if (ring->use_doorbell) {
1563 			*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1564 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1565 		} else {
1566 			WREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR, lower_32_bits(ring->wptr));
1567 		}
1568 	} else {
1569 		if (ring->use_doorbell) {
1570 			*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1571 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1572 		} else {
1573 			WREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1574 		}
1575 	}
1576 }
1577 
1578 static const struct amdgpu_ring_funcs vcn_v4_0_enc_ring_vm_funcs = {
1579 	.type = AMDGPU_RING_TYPE_VCN_ENC,
1580 	.align_mask = 0x3f,
1581 	.nop = VCN_ENC_CMD_NO_OP,
1582 	.vmhub = AMDGPU_MMHUB_0,
1583 	.get_rptr = vcn_v4_0_enc_ring_get_rptr,
1584 	.get_wptr = vcn_v4_0_enc_ring_get_wptr,
1585 	.set_wptr = vcn_v4_0_enc_ring_set_wptr,
1586 	.emit_frame_size =
1587 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1588 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1589 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1590 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1591 		1, /* vcn_v2_0_enc_ring_insert_end */
1592 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1593 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
1594 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
1595 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1596 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
1597 	.test_ib = amdgpu_vcn_enc_ring_test_ib,
1598 	.insert_nop = amdgpu_ring_insert_nop,
1599 	.insert_end = vcn_v2_0_enc_ring_insert_end,
1600 	.pad_ib = amdgpu_ring_generic_pad_ib,
1601 	.begin_use = amdgpu_vcn_ring_begin_use,
1602 	.end_use = amdgpu_vcn_ring_end_use,
1603 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1604 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1605 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1606 };
1607 
1608 /**
1609  * vcn_v4_0_set_dec_ring_funcs - set dec ring functions
1610  *
1611  * @adev: amdgpu_device pointer
1612  *
1613  * Set decode ring functions
1614  */
1615 static void vcn_v4_0_set_dec_ring_funcs(struct amdgpu_device *adev)
1616 {
1617 	int i;
1618 
1619 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1620 		if (adev->vcn.harvest_config & (1 << i))
1621 			continue;
1622 
1623 		adev->vcn.inst[i].ring_dec.funcs = &vcn_v4_0_dec_sw_ring_vm_funcs;
1624 		adev->vcn.inst[i].ring_dec.me = i;
1625 		DRM_INFO("VCN(%d) decode software ring is enabled in VM mode\n", i);
1626 	}
1627 }
1628 
1629 /**
1630  * vcn_v4_0_set_enc_ring_funcs - set enc ring functions
1631  *
1632  * @adev: amdgpu_device pointer
1633  *
1634  * Set encode ring functions
1635  */
1636 static void vcn_v4_0_set_enc_ring_funcs(struct amdgpu_device *adev)
1637 {
1638 	int i, j;
1639 
1640 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1641 		if (adev->vcn.harvest_config & (1 << i))
1642 			continue;
1643 
1644 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
1645 			adev->vcn.inst[i].ring_enc[j].funcs = &vcn_v4_0_enc_ring_vm_funcs;
1646 			adev->vcn.inst[i].ring_enc[j].me = i;
1647 		}
1648 		DRM_INFO("VCN(%d) encode is enabled in VM mode\n", i);
1649 	}
1650 }
1651 
1652 /**
1653  * vcn_v4_0_is_idle - check VCN block is idle
1654  *
1655  * @handle: amdgpu_device pointer
1656  *
1657  * Check whether VCN block is idle
1658  */
1659 static bool vcn_v4_0_is_idle(void *handle)
1660 {
1661 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1662 	int i, ret = 1;
1663 
1664 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1665 		if (adev->vcn.harvest_config & (1 << i))
1666 			continue;
1667 
1668 		ret &= (RREG32_SOC15(VCN, i, regUVD_STATUS) == UVD_STATUS__IDLE);
1669 	}
1670 
1671 	return ret;
1672 }
1673 
1674 /**
1675  * vcn_v4_0_wait_for_idle - wait for VCN block idle
1676  *
1677  * @handle: amdgpu_device pointer
1678  *
1679  * Wait for VCN block idle
1680  */
1681 static int vcn_v4_0_wait_for_idle(void *handle)
1682 {
1683 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1684 	int i, ret = 0;
1685 
1686 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1687 		if (adev->vcn.harvest_config & (1 << i))
1688 			continue;
1689 
1690 		ret = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE,
1691 			UVD_STATUS__IDLE);
1692 		if (ret)
1693 			return ret;
1694 	}
1695 
1696 	return ret;
1697 }
1698 
1699 /**
1700  * vcn_v4_0_set_clockgating_state - set VCN block clockgating state
1701  *
1702  * @handle: amdgpu_device pointer
1703  * @state: clock gating state
1704  *
1705  * Set VCN block clockgating state
1706  */
1707 static int vcn_v4_0_set_clockgating_state(void *handle, enum amd_clockgating_state state)
1708 {
1709 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1710 	bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
1711 	int i;
1712 
1713 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1714 		if (adev->vcn.harvest_config & (1 << i))
1715 			continue;
1716 
1717 		if (enable) {
1718 			if (RREG32_SOC15(VCN, i, regUVD_STATUS) != UVD_STATUS__IDLE)
1719 				return -EBUSY;
1720 			vcn_v4_0_enable_clock_gating(adev, i);
1721 		} else {
1722 			vcn_v4_0_disable_clock_gating(adev, i);
1723 		}
1724 	}
1725 
1726 	return 0;
1727 }
1728 
1729 /**
1730  * vcn_v4_0_set_powergating_state - set VCN block powergating state
1731  *
1732  * @handle: amdgpu_device pointer
1733  * @state: power gating state
1734  *
1735  * Set VCN block powergating state
1736  */
1737 static int vcn_v4_0_set_powergating_state(void *handle, enum amd_powergating_state state)
1738 {
1739 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1740 	int ret;
1741 
1742 	if(state == adev->vcn.cur_state)
1743 		return 0;
1744 
1745 	if (state == AMD_PG_STATE_GATE)
1746 		ret = vcn_v4_0_stop(adev);
1747 	else
1748 		ret = vcn_v4_0_start(adev);
1749 
1750 	if(!ret)
1751 		adev->vcn.cur_state = state;
1752 
1753 	return ret;
1754 }
1755 
1756 /**
1757  * vcn_v4_0_set_interrupt_state - set VCN block interrupt state
1758  *
1759  * @adev: amdgpu_device pointer
1760  * @source: interrupt sources
1761  * @type: interrupt types
1762  * @state: interrupt states
1763  *
1764  * Set VCN block interrupt state
1765  */
1766 static int vcn_v4_0_set_interrupt_state(struct amdgpu_device *adev, struct amdgpu_irq_src *source,
1767       unsigned type, enum amdgpu_interrupt_state state)
1768 {
1769 	return 0;
1770 }
1771 
1772 /**
1773  * vcn_v4_0_process_interrupt - process VCN block interrupt
1774  *
1775  * @adev: amdgpu_device pointer
1776  * @source: interrupt sources
1777  * @entry: interrupt entry from clients and sources
1778  *
1779  * Process VCN block interrupt
1780  */
1781 static int vcn_v4_0_process_interrupt(struct amdgpu_device *adev, struct amdgpu_irq_src *source,
1782       struct amdgpu_iv_entry *entry)
1783 {
1784 	uint32_t ip_instance;
1785 
1786 	switch (entry->client_id) {
1787 	case SOC15_IH_CLIENTID_VCN:
1788 		ip_instance = 0;
1789 		break;
1790 	case SOC15_IH_CLIENTID_VCN1:
1791 		ip_instance = 1;
1792 		break;
1793 	default:
1794 		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
1795 		return 0;
1796 	}
1797 
1798 	DRM_DEBUG("IH: VCN TRAP\n");
1799 
1800 	switch (entry->src_id) {
1801 	case VCN_4_0__SRCID__UVD_TRAP:
1802 		if (!unifiedQ_enabled) {
1803 			amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec);
1804 			break;
1805 		}
1806 		break;
1807 	case VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
1808 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
1809 		break;
1810 	case VCN_4_0__SRCID__UVD_ENC_LOW_LATENCY:
1811 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]);
1812 		break;
1813 	default:
1814 		DRM_ERROR("Unhandled interrupt: %d %d\n",
1815 			  entry->src_id, entry->src_data[0]);
1816 		break;
1817 	}
1818 
1819 	return 0;
1820 }
1821 
1822 static const struct amdgpu_irq_src_funcs vcn_v4_0_irq_funcs = {
1823 	.set = vcn_v4_0_set_interrupt_state,
1824 	.process = vcn_v4_0_process_interrupt,
1825 };
1826 
1827 /**
1828  * vcn_v4_0_set_irq_funcs - set VCN block interrupt irq functions
1829  *
1830  * @adev: amdgpu_device pointer
1831  *
1832  * Set VCN block interrupt irq functions
1833  */
1834 static void vcn_v4_0_set_irq_funcs(struct amdgpu_device *adev)
1835 {
1836 	int i;
1837 
1838 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1839 		if (adev->vcn.harvest_config & (1 << i))
1840 			continue;
1841 
1842 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
1843 		adev->vcn.inst[i].irq.funcs = &vcn_v4_0_irq_funcs;
1844 	}
1845 }
1846 
1847 static const struct amd_ip_funcs vcn_v4_0_ip_funcs = {
1848 	.name = "vcn_v4_0",
1849 	.early_init = vcn_v4_0_early_init,
1850 	.late_init = NULL,
1851 	.sw_init = vcn_v4_0_sw_init,
1852 	.sw_fini = vcn_v4_0_sw_fini,
1853 	.hw_init = vcn_v4_0_hw_init,
1854 	.hw_fini = vcn_v4_0_hw_fini,
1855 	.suspend = vcn_v4_0_suspend,
1856 	.resume = vcn_v4_0_resume,
1857 	.is_idle = vcn_v4_0_is_idle,
1858 	.wait_for_idle = vcn_v4_0_wait_for_idle,
1859 	.check_soft_reset = NULL,
1860 	.pre_soft_reset = NULL,
1861 	.soft_reset = NULL,
1862 	.post_soft_reset = NULL,
1863 	.set_clockgating_state = vcn_v4_0_set_clockgating_state,
1864 	.set_powergating_state = vcn_v4_0_set_powergating_state,
1865 };
1866 
1867 const struct amdgpu_ip_block_version vcn_v4_0_ip_block =
1868 {
1869 	.type = AMD_IP_BLOCK_TYPE_VCN,
1870 	.major = 4,
1871 	.minor = 0,
1872 	.rev = 0,
1873 	.funcs = &vcn_v4_0_ip_funcs,
1874 };
1875