xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c (revision d2825fa9)
1 /*
2  * Copyright 2019 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #include <linux/firmware.h>
25 #include "amdgpu.h"
26 #include "amdgpu_vcn.h"
27 #include "amdgpu_pm.h"
28 #include "amdgpu_cs.h"
29 #include "soc15.h"
30 #include "soc15d.h"
31 #include "vcn_v2_0.h"
32 #include "mmsch_v3_0.h"
33 
34 #include "vcn/vcn_3_0_0_offset.h"
35 #include "vcn/vcn_3_0_0_sh_mask.h"
36 #include "ivsrcid/vcn/irqsrcs_vcn_2_0.h"
37 
38 #include <drm/drm_drv.h>
39 
40 #define mmUVD_CONTEXT_ID_INTERNAL_OFFSET			0x27
41 #define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET			0x0f
42 #define mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET			0x10
43 #define mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET			0x11
44 #define mmUVD_NO_OP_INTERNAL_OFFSET				0x29
45 #define mmUVD_GP_SCRATCH8_INTERNAL_OFFSET			0x66
46 #define mmUVD_SCRATCH9_INTERNAL_OFFSET				0xc01d
47 
48 #define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET			0x431
49 #define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET		0x3b4
50 #define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET		0x3b5
51 #define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET			0x25c
52 
53 #define VCN_INSTANCES_SIENNA_CICHLID				2
54 #define DEC_SW_RING_ENABLED					FALSE
55 
56 #define RDECODE_MSG_CREATE					0x00000000
57 #define RDECODE_MESSAGE_CREATE					0x00000001
58 
59 static int amdgpu_ih_clientid_vcns[] = {
60 	SOC15_IH_CLIENTID_VCN,
61 	SOC15_IH_CLIENTID_VCN1
62 };
63 
64 static int vcn_v3_0_start_sriov(struct amdgpu_device *adev);
65 static void vcn_v3_0_set_dec_ring_funcs(struct amdgpu_device *adev);
66 static void vcn_v3_0_set_enc_ring_funcs(struct amdgpu_device *adev);
67 static void vcn_v3_0_set_irq_funcs(struct amdgpu_device *adev);
68 static int vcn_v3_0_set_powergating_state(void *handle,
69 			enum amd_powergating_state state);
70 static int vcn_v3_0_pause_dpg_mode(struct amdgpu_device *adev,
71 			int inst_idx, struct dpg_pause_state *new_state);
72 
73 static void vcn_v3_0_dec_ring_set_wptr(struct amdgpu_ring *ring);
74 static void vcn_v3_0_enc_ring_set_wptr(struct amdgpu_ring *ring);
75 
76 /**
77  * vcn_v3_0_early_init - set function pointers
78  *
79  * @handle: amdgpu_device pointer
80  *
81  * Set ring and irq function pointers
82  */
83 static int vcn_v3_0_early_init(void *handle)
84 {
85 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
86 
87 	if (amdgpu_sriov_vf(adev)) {
88 		adev->vcn.num_vcn_inst = VCN_INSTANCES_SIENNA_CICHLID;
89 		adev->vcn.harvest_config = 0;
90 		adev->vcn.num_enc_rings = 1;
91 
92 	} else {
93 		if (adev->vcn.harvest_config == (AMDGPU_VCN_HARVEST_VCN0 |
94 						 AMDGPU_VCN_HARVEST_VCN1))
95 			/* both instances are harvested, disable the block */
96 			return -ENOENT;
97 
98 		if (adev->ip_versions[UVD_HWIP][0] == IP_VERSION(3, 0, 33))
99 			adev->vcn.num_enc_rings = 0;
100 		else
101 			adev->vcn.num_enc_rings = 2;
102 	}
103 
104 	vcn_v3_0_set_dec_ring_funcs(adev);
105 	vcn_v3_0_set_enc_ring_funcs(adev);
106 	vcn_v3_0_set_irq_funcs(adev);
107 
108 	return 0;
109 }
110 
111 /**
112  * vcn_v3_0_sw_init - sw init for VCN block
113  *
114  * @handle: amdgpu_device pointer
115  *
116  * Load firmware and sw initialization
117  */
118 static int vcn_v3_0_sw_init(void *handle)
119 {
120 	struct amdgpu_ring *ring;
121 	int i, j, r;
122 	int vcn_doorbell_index = 0;
123 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
124 
125 	r = amdgpu_vcn_sw_init(adev);
126 	if (r)
127 		return r;
128 
129 	amdgpu_vcn_setup_ucode(adev);
130 
131 	r = amdgpu_vcn_resume(adev);
132 	if (r)
133 		return r;
134 
135 	/*
136 	 * Note: doorbell assignment is fixed for SRIOV multiple VCN engines
137 	 * Formula:
138 	 *   vcn_db_base  = adev->doorbell_index.vcn.vcn_ring0_1 << 1;
139 	 *   dec_ring_i   = vcn_db_base + i * (adev->vcn.num_enc_rings + 1)
140 	 *   enc_ring_i,j = vcn_db_base + i * (adev->vcn.num_enc_rings + 1) + 1 + j
141 	 */
142 	if (amdgpu_sriov_vf(adev)) {
143 		vcn_doorbell_index = adev->doorbell_index.vcn.vcn_ring0_1;
144 		/* get DWORD offset */
145 		vcn_doorbell_index = vcn_doorbell_index << 1;
146 	}
147 
148 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
149 		volatile struct amdgpu_fw_shared *fw_shared;
150 
151 		if (adev->vcn.harvest_config & (1 << i))
152 			continue;
153 
154 		adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
155 		adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
156 		adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
157 		adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
158 		adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
159 		adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
160 
161 		adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
162 		adev->vcn.inst[i].external.scratch9 = SOC15_REG_OFFSET(VCN, i, mmUVD_SCRATCH9);
163 		adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
164 		adev->vcn.inst[i].external.data0 = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_DATA0);
165 		adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
166 		adev->vcn.inst[i].external.data1 = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_DATA1);
167 		adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
168 		adev->vcn.inst[i].external.cmd = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_CMD);
169 		adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
170 		adev->vcn.inst[i].external.nop = SOC15_REG_OFFSET(VCN, i, mmUVD_NO_OP);
171 
172 		/* VCN DEC TRAP */
173 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
174 				VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst[i].irq);
175 		if (r)
176 			return r;
177 
178 		atomic_set(&adev->vcn.inst[i].sched_score, 0);
179 
180 		ring = &adev->vcn.inst[i].ring_dec;
181 		ring->use_doorbell = true;
182 		if (amdgpu_sriov_vf(adev)) {
183 			ring->doorbell_index = vcn_doorbell_index + i * (adev->vcn.num_enc_rings + 1);
184 		} else {
185 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i;
186 		}
187 		sprintf(ring->name, "vcn_dec_%d", i);
188 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
189 				     AMDGPU_RING_PRIO_DEFAULT,
190 				     &adev->vcn.inst[i].sched_score);
191 		if (r)
192 			return r;
193 
194 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
195 			enum amdgpu_ring_priority_level hw_prio = amdgpu_vcn_get_enc_ring_prio(j);
196 
197 			/* VCN ENC TRAP */
198 			r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
199 				j + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);
200 			if (r)
201 				return r;
202 
203 			ring = &adev->vcn.inst[i].ring_enc[j];
204 			ring->use_doorbell = true;
205 			if (amdgpu_sriov_vf(adev)) {
206 				ring->doorbell_index = vcn_doorbell_index + i * (adev->vcn.num_enc_rings + 1) + 1 + j;
207 			} else {
208 				ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + j + 8 * i;
209 			}
210 			sprintf(ring->name, "vcn_enc_%d.%d", i, j);
211 			r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
212 					     hw_prio, &adev->vcn.inst[i].sched_score);
213 			if (r)
214 				return r;
215 		}
216 
217 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
218 		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_SW_RING_FLAG) |
219 					     cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG) |
220 					     cpu_to_le32(AMDGPU_VCN_FW_SHARED_FLAG_0_RB);
221 		fw_shared->sw_ring.is_enabled = cpu_to_le32(DEC_SW_RING_ENABLED);
222 
223 		if (amdgpu_vcnfw_log)
224 			amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
225 	}
226 
227 	if (amdgpu_sriov_vf(adev)) {
228 		r = amdgpu_virt_alloc_mm_table(adev);
229 		if (r)
230 			return r;
231 	}
232 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
233 		adev->vcn.pause_dpg_mode = vcn_v3_0_pause_dpg_mode;
234 
235 	return 0;
236 }
237 
238 /**
239  * vcn_v3_0_sw_fini - sw fini for VCN block
240  *
241  * @handle: amdgpu_device pointer
242  *
243  * VCN suspend and free up sw allocation
244  */
245 static int vcn_v3_0_sw_fini(void *handle)
246 {
247 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
248 	int i, r, idx;
249 
250 	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
251 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
252 			volatile struct amdgpu_fw_shared *fw_shared;
253 
254 			if (adev->vcn.harvest_config & (1 << i))
255 				continue;
256 			fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
257 			fw_shared->present_flag_0 = 0;
258 			fw_shared->sw_ring.is_enabled = false;
259 		}
260 
261 		drm_dev_exit(idx);
262 	}
263 
264 	if (amdgpu_sriov_vf(adev))
265 		amdgpu_virt_free_mm_table(adev);
266 
267 	r = amdgpu_vcn_suspend(adev);
268 	if (r)
269 		return r;
270 
271 	r = amdgpu_vcn_sw_fini(adev);
272 
273 	return r;
274 }
275 
276 /**
277  * vcn_v3_0_hw_init - start and test VCN block
278  *
279  * @handle: amdgpu_device pointer
280  *
281  * Initialize the hardware, boot up the VCPU and do some testing
282  */
283 static int vcn_v3_0_hw_init(void *handle)
284 {
285 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
286 	struct amdgpu_ring *ring;
287 	int i, j, r;
288 
289 	if (amdgpu_sriov_vf(adev)) {
290 		r = vcn_v3_0_start_sriov(adev);
291 		if (r)
292 			goto done;
293 
294 		/* initialize VCN dec and enc ring buffers */
295 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
296 			if (adev->vcn.harvest_config & (1 << i))
297 				continue;
298 
299 			ring = &adev->vcn.inst[i].ring_dec;
300 			if (amdgpu_vcn_is_disabled_vcn(adev, VCN_DECODE_RING, i)) {
301 				ring->sched.ready = false;
302 				ring->no_scheduler = true;
303 				dev_info(adev->dev, "ring %s is disabled by hypervisor\n", ring->name);
304 			} else {
305 				ring->wptr = 0;
306 				ring->wptr_old = 0;
307 				vcn_v3_0_dec_ring_set_wptr(ring);
308 				ring->sched.ready = true;
309 			}
310 
311 			for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
312 				ring = &adev->vcn.inst[i].ring_enc[j];
313 				if (amdgpu_vcn_is_disabled_vcn(adev, VCN_ENCODE_RING, i)) {
314 					ring->sched.ready = false;
315 					ring->no_scheduler = true;
316 					dev_info(adev->dev, "ring %s is disabled by hypervisor\n", ring->name);
317 				} else {
318 					ring->wptr = 0;
319 					ring->wptr_old = 0;
320 					vcn_v3_0_enc_ring_set_wptr(ring);
321 					ring->sched.ready = true;
322 				}
323 			}
324 		}
325 	} else {
326 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
327 			if (adev->vcn.harvest_config & (1 << i))
328 				continue;
329 
330 			ring = &adev->vcn.inst[i].ring_dec;
331 
332 			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
333 						     ring->doorbell_index, i);
334 
335 			r = amdgpu_ring_test_helper(ring);
336 			if (r)
337 				goto done;
338 
339 			for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
340 				ring = &adev->vcn.inst[i].ring_enc[j];
341 				r = amdgpu_ring_test_helper(ring);
342 				if (r)
343 					goto done;
344 			}
345 		}
346 	}
347 
348 done:
349 	if (!r)
350 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
351 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
352 
353 	return r;
354 }
355 
356 /**
357  * vcn_v3_0_hw_fini - stop the hardware block
358  *
359  * @handle: amdgpu_device pointer
360  *
361  * Stop the VCN block, mark ring as not ready any more
362  */
363 static int vcn_v3_0_hw_fini(void *handle)
364 {
365 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
366 	int i;
367 
368 	cancel_delayed_work_sync(&adev->vcn.idle_work);
369 
370 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
371 		if (adev->vcn.harvest_config & (1 << i))
372 			continue;
373 
374 		if (!amdgpu_sriov_vf(adev)) {
375 			if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
376 					(adev->vcn.cur_state != AMD_PG_STATE_GATE &&
377 					 RREG32_SOC15(VCN, i, mmUVD_STATUS))) {
378 				vcn_v3_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
379 			}
380 		}
381 	}
382 
383 	return 0;
384 }
385 
386 /**
387  * vcn_v3_0_suspend - suspend VCN block
388  *
389  * @handle: amdgpu_device pointer
390  *
391  * HW fini and suspend VCN block
392  */
393 static int vcn_v3_0_suspend(void *handle)
394 {
395 	int r;
396 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
397 
398 	r = vcn_v3_0_hw_fini(adev);
399 	if (r)
400 		return r;
401 
402 	r = amdgpu_vcn_suspend(adev);
403 
404 	return r;
405 }
406 
407 /**
408  * vcn_v3_0_resume - resume VCN block
409  *
410  * @handle: amdgpu_device pointer
411  *
412  * Resume firmware and hw init VCN block
413  */
414 static int vcn_v3_0_resume(void *handle)
415 {
416 	int r;
417 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
418 
419 	r = amdgpu_vcn_resume(adev);
420 	if (r)
421 		return r;
422 
423 	r = vcn_v3_0_hw_init(adev);
424 
425 	return r;
426 }
427 
428 /**
429  * vcn_v3_0_mc_resume - memory controller programming
430  *
431  * @adev: amdgpu_device pointer
432  * @inst: instance number
433  *
434  * Let the VCN memory controller know it's offsets
435  */
436 static void vcn_v3_0_mc_resume(struct amdgpu_device *adev, int inst)
437 {
438 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
439 	uint32_t offset;
440 
441 	/* cache window 0: fw */
442 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
443 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
444 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo));
445 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
446 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi));
447 		WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET0, 0);
448 		offset = 0;
449 	} else {
450 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
451 			lower_32_bits(adev->vcn.inst[inst].gpu_addr));
452 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
453 			upper_32_bits(adev->vcn.inst[inst].gpu_addr));
454 		offset = size;
455 		WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET0,
456 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
457 	}
458 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE0, size);
459 
460 	/* cache window 1: stack */
461 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
462 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
463 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
464 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
465 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET1, 0);
466 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
467 
468 	/* cache window 2: context */
469 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
470 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
471 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
472 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
473 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET2, 0);
474 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
475 
476 	/* non-cache window */
477 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
478 		lower_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
479 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
480 		upper_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
481 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
482 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_NONCACHE_SIZE0,
483 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
484 }
485 
486 static void vcn_v3_0_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
487 {
488 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
489 	uint32_t offset;
490 
491 	/* cache window 0: fw */
492 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
493 		if (!indirect) {
494 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
495 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
496 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
497 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
498 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
499 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
500 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
501 				VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
502 		} else {
503 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
504 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
505 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
506 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
507 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
508 				VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
509 		}
510 		offset = 0;
511 	} else {
512 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
513 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
514 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
515 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
516 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
517 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
518 		offset = size;
519 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
520 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0),
521 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
522 	}
523 
524 	if (!indirect)
525 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
526 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
527 	else
528 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
529 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
530 
531 	/* cache window 1: stack */
532 	if (!indirect) {
533 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
534 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
535 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
536 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
537 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
538 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
539 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
540 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
541 	} else {
542 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
543 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
544 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
545 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
546 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
547 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
548 	}
549 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
550 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
551 
552 	/* cache window 2: context */
553 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
554 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
555 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
556 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
557 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
558 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
559 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
560 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
561 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
562 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
563 
564 	/* non-cache window */
565 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
566 			VCN, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
567 			lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
568 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
569 			VCN, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
570 			upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
571 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
572 			VCN, inst_idx, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
573 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
574 			VCN, inst_idx, mmUVD_VCPU_NONCACHE_SIZE0),
575 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
576 
577 	/* VCN global tiling registers */
578 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
579 		UVD, 0, mmUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
580 }
581 
582 static void vcn_v3_0_disable_static_power_gating(struct amdgpu_device *adev, int inst)
583 {
584 	uint32_t data = 0;
585 
586 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
587 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
588 			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
589 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
590 			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
591 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
592 			| 2 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
593 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
594 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
595 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
596 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
597 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
598 			| 2 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
599 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
600 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
601 
602 		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
603 		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS,
604 			UVD_PGFSM_STATUS__UVDM_UVDU_UVDLM_PWR_ON_3_0, 0x3F3FFFFF);
605 	} else {
606 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
607 			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
608 			| 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
609 			| 1 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
610 			| 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
611 			| 1 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
612 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
613 			| 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
614 			| 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
615 			| 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
616 			| 1 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
617 			| 1 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
618 			| 1 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
619 			| 1 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
620 		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
621 		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS, 0,  0x3F3FFFFF);
622 	}
623 
624 	data = RREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS);
625 	data &= ~0x103;
626 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
627 		data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
628 			UVD_POWER_STATUS__UVD_PG_EN_MASK;
629 
630 	WREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS, data);
631 }
632 
633 static void vcn_v3_0_enable_static_power_gating(struct amdgpu_device *adev, int inst)
634 {
635 	uint32_t data;
636 
637 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
638 		/* Before power off, this indicator has to be turned on */
639 		data = RREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS);
640 		data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
641 		data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
642 		WREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS, data);
643 
644 		data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
645 			| 2 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
646 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
647 			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
648 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
649 			| 2 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
650 			| 2 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
651 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
652 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
653 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
654 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
655 			| 2 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
656 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
657 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
658 		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
659 
660 		data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
661 			| 2 << UVD_PGFSM_STATUS__UVDU_PWR_STATUS__SHIFT
662 			| 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
663 			| 2 << UVD_PGFSM_STATUS__UVDC_PWR_STATUS__SHIFT
664 			| 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
665 			| 2 << UVD_PGFSM_STATUS__UVDIRL_PWR_STATUS__SHIFT
666 			| 2 << UVD_PGFSM_STATUS__UVDLM_PWR_STATUS__SHIFT
667 			| 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
668 			| 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
669 			| 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT
670 			| 2 << UVD_PGFSM_STATUS__UVDAB_PWR_STATUS__SHIFT
671 			| 2 << UVD_PGFSM_STATUS__UVDATD_PWR_STATUS__SHIFT
672 			| 2 << UVD_PGFSM_STATUS__UVDNA_PWR_STATUS__SHIFT
673 			| 2 << UVD_PGFSM_STATUS__UVDNB_PWR_STATUS__SHIFT);
674 		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS, data, 0x3F3FFFFF);
675 	}
676 }
677 
678 /**
679  * vcn_v3_0_disable_clock_gating - disable VCN clock gating
680  *
681  * @adev: amdgpu_device pointer
682  * @inst: instance number
683  *
684  * Disable clock gating for VCN block
685  */
686 static void vcn_v3_0_disable_clock_gating(struct amdgpu_device *adev, int inst)
687 {
688 	uint32_t data;
689 
690 	/* VCN disable CGC */
691 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
692 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
693 		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
694 	else
695 		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
696 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
697 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
698 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
699 
700 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_GATE);
701 	data &= ~(UVD_CGC_GATE__SYS_MASK
702 		| UVD_CGC_GATE__UDEC_MASK
703 		| UVD_CGC_GATE__MPEG2_MASK
704 		| UVD_CGC_GATE__REGS_MASK
705 		| UVD_CGC_GATE__RBC_MASK
706 		| UVD_CGC_GATE__LMI_MC_MASK
707 		| UVD_CGC_GATE__LMI_UMC_MASK
708 		| UVD_CGC_GATE__IDCT_MASK
709 		| UVD_CGC_GATE__MPRD_MASK
710 		| UVD_CGC_GATE__MPC_MASK
711 		| UVD_CGC_GATE__LBSI_MASK
712 		| UVD_CGC_GATE__LRBBM_MASK
713 		| UVD_CGC_GATE__UDEC_RE_MASK
714 		| UVD_CGC_GATE__UDEC_CM_MASK
715 		| UVD_CGC_GATE__UDEC_IT_MASK
716 		| UVD_CGC_GATE__UDEC_DB_MASK
717 		| UVD_CGC_GATE__UDEC_MP_MASK
718 		| UVD_CGC_GATE__WCB_MASK
719 		| UVD_CGC_GATE__VCPU_MASK
720 		| UVD_CGC_GATE__MMSCH_MASK);
721 
722 	WREG32_SOC15(VCN, inst, mmUVD_CGC_GATE, data);
723 
724 	SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_CGC_GATE, 0,  0xFFFFFFFF);
725 
726 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
727 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
728 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
729 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
730 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
731 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
732 		| UVD_CGC_CTRL__SYS_MODE_MASK
733 		| UVD_CGC_CTRL__UDEC_MODE_MASK
734 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
735 		| UVD_CGC_CTRL__REGS_MODE_MASK
736 		| UVD_CGC_CTRL__RBC_MODE_MASK
737 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
738 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
739 		| UVD_CGC_CTRL__IDCT_MODE_MASK
740 		| UVD_CGC_CTRL__MPRD_MODE_MASK
741 		| UVD_CGC_CTRL__MPC_MODE_MASK
742 		| UVD_CGC_CTRL__LBSI_MODE_MASK
743 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
744 		| UVD_CGC_CTRL__WCB_MODE_MASK
745 		| UVD_CGC_CTRL__VCPU_MODE_MASK
746 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
747 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
748 
749 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE);
750 	data |= (UVD_SUVD_CGC_GATE__SRE_MASK
751 		| UVD_SUVD_CGC_GATE__SIT_MASK
752 		| UVD_SUVD_CGC_GATE__SMP_MASK
753 		| UVD_SUVD_CGC_GATE__SCM_MASK
754 		| UVD_SUVD_CGC_GATE__SDB_MASK
755 		| UVD_SUVD_CGC_GATE__SRE_H264_MASK
756 		| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
757 		| UVD_SUVD_CGC_GATE__SIT_H264_MASK
758 		| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
759 		| UVD_SUVD_CGC_GATE__SCM_H264_MASK
760 		| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
761 		| UVD_SUVD_CGC_GATE__SDB_H264_MASK
762 		| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
763 		| UVD_SUVD_CGC_GATE__SCLR_MASK
764 		| UVD_SUVD_CGC_GATE__ENT_MASK
765 		| UVD_SUVD_CGC_GATE__IME_MASK
766 		| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
767 		| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
768 		| UVD_SUVD_CGC_GATE__SITE_MASK
769 		| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
770 		| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
771 		| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
772 		| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
773 		| UVD_SUVD_CGC_GATE__IME_HEVC_MASK
774 		| UVD_SUVD_CGC_GATE__EFC_MASK
775 		| UVD_SUVD_CGC_GATE__SAOE_MASK
776 		| UVD_SUVD_CGC_GATE__SRE_AV1_MASK
777 		| UVD_SUVD_CGC_GATE__FBC_PCLK_MASK
778 		| UVD_SUVD_CGC_GATE__FBC_CCLK_MASK
779 		| UVD_SUVD_CGC_GATE__SCM_AV1_MASK
780 		| UVD_SUVD_CGC_GATE__SMPA_MASK);
781 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE, data);
782 
783 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE2);
784 	data |= (UVD_SUVD_CGC_GATE2__MPBE0_MASK
785 		| UVD_SUVD_CGC_GATE2__MPBE1_MASK
786 		| UVD_SUVD_CGC_GATE2__SIT_AV1_MASK
787 		| UVD_SUVD_CGC_GATE2__SDB_AV1_MASK
788 		| UVD_SUVD_CGC_GATE2__MPC1_MASK);
789 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE2, data);
790 
791 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL);
792 	data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
793 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
794 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
795 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
796 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
797 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
798 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
799 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
800 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK
801 		| UVD_SUVD_CGC_CTRL__EFC_MODE_MASK
802 		| UVD_SUVD_CGC_CTRL__SAOE_MODE_MASK
803 		| UVD_SUVD_CGC_CTRL__SMPA_MODE_MASK
804 		| UVD_SUVD_CGC_CTRL__MPBE0_MODE_MASK
805 		| UVD_SUVD_CGC_CTRL__MPBE1_MODE_MASK
806 		| UVD_SUVD_CGC_CTRL__SIT_AV1_MODE_MASK
807 		| UVD_SUVD_CGC_CTRL__SDB_AV1_MODE_MASK
808 		| UVD_SUVD_CGC_CTRL__MPC1_MODE_MASK
809 		| UVD_SUVD_CGC_CTRL__FBC_PCLK_MASK
810 		| UVD_SUVD_CGC_CTRL__FBC_CCLK_MASK);
811 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL, data);
812 }
813 
814 static void vcn_v3_0_clock_gating_dpg_mode(struct amdgpu_device *adev,
815 		uint8_t sram_sel, int inst_idx, uint8_t indirect)
816 {
817 	uint32_t reg_data = 0;
818 
819 	/* enable sw clock gating control */
820 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
821 		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
822 	else
823 		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
824 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
825 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
826 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
827 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
828 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
829 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
830 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
831 		 UVD_CGC_CTRL__SYS_MODE_MASK |
832 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
833 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
834 		 UVD_CGC_CTRL__REGS_MODE_MASK |
835 		 UVD_CGC_CTRL__RBC_MODE_MASK |
836 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
837 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
838 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
839 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
840 		 UVD_CGC_CTRL__MPC_MODE_MASK |
841 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
842 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
843 		 UVD_CGC_CTRL__WCB_MODE_MASK |
844 		 UVD_CGC_CTRL__VCPU_MODE_MASK |
845 		 UVD_CGC_CTRL__MMSCH_MODE_MASK);
846 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
847 		VCN, inst_idx, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
848 
849 	/* turn off clock gating */
850 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
851 		VCN, inst_idx, mmUVD_CGC_GATE), 0, sram_sel, indirect);
852 
853 	/* turn on SUVD clock gating */
854 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
855 		VCN, inst_idx, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
856 
857 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
858 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
859 		VCN, inst_idx, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
860 }
861 
862 /**
863  * vcn_v3_0_enable_clock_gating - enable VCN clock gating
864  *
865  * @adev: amdgpu_device pointer
866  * @inst: instance number
867  *
868  * Enable clock gating for VCN block
869  */
870 static void vcn_v3_0_enable_clock_gating(struct amdgpu_device *adev, int inst)
871 {
872 	uint32_t data;
873 
874 	/* enable VCN CGC */
875 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
876 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
877 		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
878 	else
879 		data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
880 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
881 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
882 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
883 
884 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
885 	data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
886 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
887 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
888 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
889 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
890 		| UVD_CGC_CTRL__SYS_MODE_MASK
891 		| UVD_CGC_CTRL__UDEC_MODE_MASK
892 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
893 		| UVD_CGC_CTRL__REGS_MODE_MASK
894 		| UVD_CGC_CTRL__RBC_MODE_MASK
895 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
896 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
897 		| UVD_CGC_CTRL__IDCT_MODE_MASK
898 		| UVD_CGC_CTRL__MPRD_MODE_MASK
899 		| UVD_CGC_CTRL__MPC_MODE_MASK
900 		| UVD_CGC_CTRL__LBSI_MODE_MASK
901 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
902 		| UVD_CGC_CTRL__WCB_MODE_MASK
903 		| UVD_CGC_CTRL__VCPU_MODE_MASK
904 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
905 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
906 
907 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL);
908 	data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
909 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
910 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
911 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
912 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
913 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
914 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
915 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
916 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK
917 		| UVD_SUVD_CGC_CTRL__EFC_MODE_MASK
918 		| UVD_SUVD_CGC_CTRL__SAOE_MODE_MASK
919 		| UVD_SUVD_CGC_CTRL__SMPA_MODE_MASK
920 		| UVD_SUVD_CGC_CTRL__MPBE0_MODE_MASK
921 		| UVD_SUVD_CGC_CTRL__MPBE1_MODE_MASK
922 		| UVD_SUVD_CGC_CTRL__SIT_AV1_MODE_MASK
923 		| UVD_SUVD_CGC_CTRL__SDB_AV1_MODE_MASK
924 		| UVD_SUVD_CGC_CTRL__MPC1_MODE_MASK
925 		| UVD_SUVD_CGC_CTRL__FBC_PCLK_MASK
926 		| UVD_SUVD_CGC_CTRL__FBC_CCLK_MASK);
927 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL, data);
928 }
929 
930 static int vcn_v3_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
931 {
932 	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
933 	struct amdgpu_ring *ring;
934 	uint32_t rb_bufsz, tmp;
935 
936 	/* disable register anti-hang mechanism */
937 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1,
938 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
939 	/* enable dynamic power gating mode */
940 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS);
941 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
942 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
943 	WREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS, tmp);
944 
945 	if (indirect)
946 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
947 
948 	/* enable clock gating */
949 	vcn_v3_0_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
950 
951 	/* enable VCPU clock */
952 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
953 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
954 	tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
955 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
956 		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
957 
958 	/* disable master interupt */
959 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
960 		VCN, inst_idx, mmUVD_MASTINT_EN), 0, 0, indirect);
961 
962 	/* setup mmUVD_LMI_CTRL */
963 	tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
964 		UVD_LMI_CTRL__REQ_MODE_MASK |
965 		UVD_LMI_CTRL__CRC_RESET_MASK |
966 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
967 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
968 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
969 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
970 		0x00100000L);
971 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
972 		VCN, inst_idx, mmUVD_LMI_CTRL), tmp, 0, indirect);
973 
974 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
975 		VCN, inst_idx, mmUVD_MPC_CNTL),
976 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
977 
978 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
979 		VCN, inst_idx, mmUVD_MPC_SET_MUXA0),
980 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
981 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
982 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
983 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
984 
985 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
986 		VCN, inst_idx, mmUVD_MPC_SET_MUXB0),
987 		 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
988 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
989 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
990 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
991 
992 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
993 		VCN, inst_idx, mmUVD_MPC_SET_MUX),
994 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
995 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
996 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
997 
998 	vcn_v3_0_mc_resume_dpg_mode(adev, inst_idx, indirect);
999 
1000 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1001 		VCN, inst_idx, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
1002 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1003 		VCN, inst_idx, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
1004 
1005 	/* enable LMI MC and UMC channels */
1006 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1007 		VCN, inst_idx, mmUVD_LMI_CTRL2), 0, 0, indirect);
1008 
1009 	/* unblock VCPU register access */
1010 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1011 		VCN, inst_idx, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
1012 
1013 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
1014 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
1015 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1016 		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
1017 
1018 	/* enable master interrupt */
1019 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1020 		VCN, inst_idx, mmUVD_MASTINT_EN),
1021 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
1022 
1023 	/* add nop to workaround PSP size check */
1024 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1025 		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
1026 
1027 	if (indirect)
1028 		psp_update_vcn_sram(adev, inst_idx, adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
1029 			(uint32_t)((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
1030 				(uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr));
1031 
1032 	ring = &adev->vcn.inst[inst_idx].ring_dec;
1033 	/* force RBC into idle state */
1034 	rb_bufsz = order_base_2(ring->ring_size);
1035 	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1036 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1037 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1038 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1039 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1040 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
1041 
1042 	/* Stall DPG before WPTR/RPTR reset */
1043 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1044 		UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1045 		~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1046 	fw_shared->multi_queue.decode_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1047 
1048 	/* set the write pointer delay */
1049 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
1050 
1051 	/* set the wb address */
1052 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR_ADDR,
1053 		(upper_32_bits(ring->gpu_addr) >> 2));
1054 
1055 	/* programm the RB_BASE for ring buffer */
1056 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1057 		lower_32_bits(ring->gpu_addr));
1058 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1059 		upper_32_bits(ring->gpu_addr));
1060 
1061 	/* Initialize the ring buffer's read and write pointers */
1062 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, 0);
1063 
1064 	WREG32_SOC15(VCN, inst_idx, mmUVD_SCRATCH2, 0);
1065 
1066 	ring->wptr = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR);
1067 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR,
1068 		lower_32_bits(ring->wptr));
1069 
1070 	/* Reset FW shared memory RBC WPTR/RPTR */
1071 	fw_shared->rb.rptr = 0;
1072 	fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1073 
1074 	/*resetting done, fw can check RB ring */
1075 	fw_shared->multi_queue.decode_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1076 
1077 	/* Unstall DPG */
1078 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1079 		0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1080 
1081 	return 0;
1082 }
1083 
1084 static int vcn_v3_0_start(struct amdgpu_device *adev)
1085 {
1086 	volatile struct amdgpu_fw_shared *fw_shared;
1087 	struct amdgpu_ring *ring;
1088 	uint32_t rb_bufsz, tmp;
1089 	int i, j, k, r;
1090 
1091 	if (adev->pm.dpm_enabled)
1092 		amdgpu_dpm_enable_uvd(adev, true);
1093 
1094 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1095 		if (adev->vcn.harvest_config & (1 << i))
1096 			continue;
1097 
1098 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG){
1099 			r = vcn_v3_0_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
1100 			continue;
1101 		}
1102 
1103 		/* disable VCN power gating */
1104 		vcn_v3_0_disable_static_power_gating(adev, i);
1105 
1106 		/* set VCN status busy */
1107 		tmp = RREG32_SOC15(VCN, i, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY;
1108 		WREG32_SOC15(VCN, i, mmUVD_STATUS, tmp);
1109 
1110 		/*SW clock gating */
1111 		vcn_v3_0_disable_clock_gating(adev, i);
1112 
1113 		/* enable VCPU clock */
1114 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1115 			UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1116 
1117 		/* disable master interrupt */
1118 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN), 0,
1119 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1120 
1121 		/* enable LMI MC and UMC channels */
1122 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_LMI_CTRL2), 0,
1123 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1124 
1125 		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1126 		tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1127 		tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1128 		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1129 
1130 		/* setup mmUVD_LMI_CTRL */
1131 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL);
1132 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL, tmp |
1133 			UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK	|
1134 			UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1135 			UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1136 			UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1137 
1138 		/* setup mmUVD_MPC_CNTL */
1139 		tmp = RREG32_SOC15(VCN, i, mmUVD_MPC_CNTL);
1140 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1141 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1142 		WREG32_SOC15(VCN, i, mmUVD_MPC_CNTL, tmp);
1143 
1144 		/* setup UVD_MPC_SET_MUXA0 */
1145 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXA0,
1146 			((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1147 			(0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1148 			(0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1149 			(0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1150 
1151 		/* setup UVD_MPC_SET_MUXB0 */
1152 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXB0,
1153 			((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1154 			(0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1155 			(0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1156 			(0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1157 
1158 		/* setup mmUVD_MPC_SET_MUX */
1159 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUX,
1160 			((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1161 			(0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1162 			(0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1163 
1164 		vcn_v3_0_mc_resume(adev, i);
1165 
1166 		/* VCN global tiling registers */
1167 		WREG32_SOC15(VCN, i, mmUVD_GFX10_ADDR_CONFIG,
1168 			adev->gfx.config.gb_addr_config);
1169 
1170 		/* unblock VCPU register access */
1171 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL), 0,
1172 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1173 
1174 		/* release VCPU reset to boot */
1175 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1176 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1177 
1178 		for (j = 0; j < 10; ++j) {
1179 			uint32_t status;
1180 
1181 			for (k = 0; k < 100; ++k) {
1182 				status = RREG32_SOC15(VCN, i, mmUVD_STATUS);
1183 				if (status & 2)
1184 					break;
1185 				mdelay(10);
1186 			}
1187 			r = 0;
1188 			if (status & 2)
1189 				break;
1190 
1191 			DRM_ERROR("VCN[%d] decode not responding, trying to reset the VCPU!!!\n", i);
1192 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1193 				UVD_VCPU_CNTL__BLK_RST_MASK,
1194 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1195 			mdelay(10);
1196 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1197 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1198 
1199 			mdelay(10);
1200 			r = -1;
1201 		}
1202 
1203 		if (r) {
1204 			DRM_ERROR("VCN[%d] decode not responding, giving up!!!\n", i);
1205 			return r;
1206 		}
1207 
1208 		/* enable master interrupt */
1209 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN),
1210 			UVD_MASTINT_EN__VCPU_EN_MASK,
1211 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1212 
1213 		/* clear the busy bit of VCN_STATUS */
1214 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS), 0,
1215 			~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1216 
1217 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_VMID, 0);
1218 
1219 		ring = &adev->vcn.inst[i].ring_dec;
1220 		/* force RBC into idle state */
1221 		rb_bufsz = order_base_2(ring->ring_size);
1222 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1223 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1224 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1225 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1226 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1227 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_CNTL, tmp);
1228 
1229 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1230 		fw_shared->multi_queue.decode_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1231 
1232 		/* programm the RB_BASE for ring buffer */
1233 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1234 			lower_32_bits(ring->gpu_addr));
1235 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1236 			upper_32_bits(ring->gpu_addr));
1237 
1238 		/* Initialize the ring buffer's read and write pointers */
1239 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR, 0);
1240 
1241 		WREG32_SOC15(VCN, i, mmUVD_SCRATCH2, 0);
1242 		ring->wptr = RREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR);
1243 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_WPTR,
1244 			lower_32_bits(ring->wptr));
1245 		fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1246 		fw_shared->multi_queue.decode_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1247 
1248 		if (adev->ip_versions[UVD_HWIP][0] != IP_VERSION(3, 0, 33)) {
1249 			fw_shared->multi_queue.encode_generalpurpose_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1250 			ring = &adev->vcn.inst[i].ring_enc[0];
1251 			WREG32_SOC15(VCN, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1252 			WREG32_SOC15(VCN, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1253 			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO, ring->gpu_addr);
1254 			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1255 			WREG32_SOC15(VCN, i, mmUVD_RB_SIZE, ring->ring_size / 4);
1256 			fw_shared->multi_queue.encode_generalpurpose_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1257 
1258 			fw_shared->multi_queue.encode_lowlatency_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1259 			ring = &adev->vcn.inst[i].ring_enc[1];
1260 			WREG32_SOC15(VCN, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1261 			WREG32_SOC15(VCN, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1262 			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1263 			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1264 			WREG32_SOC15(VCN, i, mmUVD_RB_SIZE2, ring->ring_size / 4);
1265 			fw_shared->multi_queue.encode_lowlatency_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1266 		}
1267 	}
1268 
1269 	return 0;
1270 }
1271 
1272 static int vcn_v3_0_start_sriov(struct amdgpu_device *adev)
1273 {
1274 	int i, j;
1275 	struct amdgpu_ring *ring;
1276 	uint64_t cache_addr;
1277 	uint64_t rb_addr;
1278 	uint64_t ctx_addr;
1279 	uint32_t param, resp, expected;
1280 	uint32_t offset, cache_size;
1281 	uint32_t tmp, timeout;
1282 
1283 	struct amdgpu_mm_table *table = &adev->virt.mm_table;
1284 	uint32_t *table_loc;
1285 	uint32_t table_size;
1286 	uint32_t size, size_dw;
1287 
1288 	struct mmsch_v3_0_cmd_direct_write
1289 		direct_wt = { {0} };
1290 	struct mmsch_v3_0_cmd_direct_read_modify_write
1291 		direct_rd_mod_wt = { {0} };
1292 	struct mmsch_v3_0_cmd_end end = { {0} };
1293 	struct mmsch_v3_0_init_header header;
1294 
1295 	direct_wt.cmd_header.command_type =
1296 		MMSCH_COMMAND__DIRECT_REG_WRITE;
1297 	direct_rd_mod_wt.cmd_header.command_type =
1298 		MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1299 	end.cmd_header.command_type =
1300 		MMSCH_COMMAND__END;
1301 
1302 	header.version = MMSCH_VERSION;
1303 	header.total_size = sizeof(struct mmsch_v3_0_init_header) >> 2;
1304 	for (i = 0; i < AMDGPU_MAX_VCN_INSTANCES; i++) {
1305 		header.inst[i].init_status = 0;
1306 		header.inst[i].table_offset = 0;
1307 		header.inst[i].table_size = 0;
1308 	}
1309 
1310 	table_loc = (uint32_t *)table->cpu_addr;
1311 	table_loc += header.total_size;
1312 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1313 		if (adev->vcn.harvest_config & (1 << i))
1314 			continue;
1315 
1316 		table_size = 0;
1317 
1318 		MMSCH_V3_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(VCN, i,
1319 			mmUVD_STATUS),
1320 			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1321 
1322 		cache_size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1323 
1324 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1325 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1326 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1327 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1328 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1329 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1330 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1331 			offset = 0;
1332 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1333 				mmUVD_VCPU_CACHE_OFFSET0),
1334 				0);
1335 		} else {
1336 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1337 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1338 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1339 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1340 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1341 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1342 			offset = cache_size;
1343 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1344 				mmUVD_VCPU_CACHE_OFFSET0),
1345 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1346 		}
1347 
1348 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1349 			mmUVD_VCPU_CACHE_SIZE0),
1350 			cache_size);
1351 
1352 		cache_addr = adev->vcn.inst[i].gpu_addr + offset;
1353 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1354 			mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1355 			lower_32_bits(cache_addr));
1356 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1357 			mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1358 			upper_32_bits(cache_addr));
1359 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1360 			mmUVD_VCPU_CACHE_OFFSET1),
1361 			0);
1362 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1363 			mmUVD_VCPU_CACHE_SIZE1),
1364 			AMDGPU_VCN_STACK_SIZE);
1365 
1366 		cache_addr = adev->vcn.inst[i].gpu_addr + offset +
1367 			AMDGPU_VCN_STACK_SIZE;
1368 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1369 			mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1370 			lower_32_bits(cache_addr));
1371 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1372 			mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1373 			upper_32_bits(cache_addr));
1374 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1375 			mmUVD_VCPU_CACHE_OFFSET2),
1376 			0);
1377 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1378 			mmUVD_VCPU_CACHE_SIZE2),
1379 			AMDGPU_VCN_CONTEXT_SIZE);
1380 
1381 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
1382 			ring = &adev->vcn.inst[i].ring_enc[j];
1383 			ring->wptr = 0;
1384 			rb_addr = ring->gpu_addr;
1385 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1386 				mmUVD_RB_BASE_LO),
1387 				lower_32_bits(rb_addr));
1388 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1389 				mmUVD_RB_BASE_HI),
1390 				upper_32_bits(rb_addr));
1391 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1392 				mmUVD_RB_SIZE),
1393 				ring->ring_size / 4);
1394 		}
1395 
1396 		ring = &adev->vcn.inst[i].ring_dec;
1397 		ring->wptr = 0;
1398 		rb_addr = ring->gpu_addr;
1399 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1400 			mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1401 			lower_32_bits(rb_addr));
1402 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1403 			mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1404 			upper_32_bits(rb_addr));
1405 		/* force RBC into idle state */
1406 		tmp = order_base_2(ring->ring_size);
1407 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, tmp);
1408 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1409 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1410 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1411 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1412 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1413 			mmUVD_RBC_RB_CNTL),
1414 			tmp);
1415 
1416 		/* add end packet */
1417 		MMSCH_V3_0_INSERT_END();
1418 
1419 		/* refine header */
1420 		header.inst[i].init_status = 0;
1421 		header.inst[i].table_offset = header.total_size;
1422 		header.inst[i].table_size = table_size;
1423 		header.total_size += table_size;
1424 	}
1425 
1426 	/* Update init table header in memory */
1427 	size = sizeof(struct mmsch_v3_0_init_header);
1428 	table_loc = (uint32_t *)table->cpu_addr;
1429 	memcpy((void *)table_loc, &header, size);
1430 
1431 	/* message MMSCH (in VCN[0]) to initialize this client
1432 	 * 1, write to mmsch_vf_ctx_addr_lo/hi register with GPU mc addr
1433 	 * of memory descriptor location
1434 	 */
1435 	ctx_addr = table->gpu_addr;
1436 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(ctx_addr));
1437 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(ctx_addr));
1438 
1439 	/* 2, update vmid of descriptor */
1440 	tmp = RREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID);
1441 	tmp &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1442 	/* use domain0 for MM scheduler */
1443 	tmp |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1444 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID, tmp);
1445 
1446 	/* 3, notify mmsch about the size of this descriptor */
1447 	size = header.total_size;
1448 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_SIZE, size);
1449 
1450 	/* 4, set resp to zero */
1451 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1452 
1453 	/* 5, kick off the initialization and wait until
1454 	 * MMSCH_VF_MAILBOX_RESP becomes non-zero
1455 	 */
1456 	param = 0x10000001;
1457 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_HOST, param);
1458 	tmp = 0;
1459 	timeout = 1000;
1460 	resp = 0;
1461 	expected = param + 1;
1462 	while (resp != expected) {
1463 		resp = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1464 		if (resp == expected)
1465 			break;
1466 
1467 		udelay(10);
1468 		tmp = tmp + 10;
1469 		if (tmp >= timeout) {
1470 			DRM_ERROR("failed to init MMSCH. TIME-OUT after %d usec"\
1471 				" waiting for mmMMSCH_VF_MAILBOX_RESP "\
1472 				"(expected=0x%08x, readback=0x%08x)\n",
1473 				tmp, expected, resp);
1474 			return -EBUSY;
1475 		}
1476 	}
1477 
1478 	return 0;
1479 }
1480 
1481 static int vcn_v3_0_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1482 {
1483 	uint32_t tmp;
1484 
1485 	/* Wait for power status to be 1 */
1486 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1487 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1488 
1489 	/* wait for read ptr to be equal to write ptr */
1490 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR);
1491 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1492 
1493 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2);
1494 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
1495 
1496 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1497 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF);
1498 
1499 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1500 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1501 
1502 	/* disable dynamic power gating mode */
1503 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 0,
1504 		~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1505 
1506 	return 0;
1507 }
1508 
1509 static int vcn_v3_0_stop(struct amdgpu_device *adev)
1510 {
1511 	uint32_t tmp;
1512 	int i, r = 0;
1513 
1514 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1515 		if (adev->vcn.harvest_config & (1 << i))
1516 			continue;
1517 
1518 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1519 			r = vcn_v3_0_stop_dpg_mode(adev, i);
1520 			continue;
1521 		}
1522 
1523 		/* wait for vcn idle */
1524 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1525 		if (r)
1526 			return r;
1527 
1528 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1529 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1530 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1531 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1532 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1533 		if (r)
1534 			return r;
1535 
1536 		/* disable LMI UMC channel */
1537 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2);
1538 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1539 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2, tmp);
1540 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1541 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1542 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1543 		if (r)
1544 			return r;
1545 
1546 		/* block VCPU register access */
1547 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL),
1548 			UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1549 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1550 
1551 		/* reset VCPU */
1552 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1553 			UVD_VCPU_CNTL__BLK_RST_MASK,
1554 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1555 
1556 		/* disable VCPU clock */
1557 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1558 			~(UVD_VCPU_CNTL__CLK_EN_MASK));
1559 
1560 		/* apply soft reset */
1561 		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1562 		tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1563 		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1564 		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1565 		tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1566 		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1567 
1568 		/* clear status */
1569 		WREG32_SOC15(VCN, i, mmUVD_STATUS, 0);
1570 
1571 		/* apply HW clock gating */
1572 		vcn_v3_0_enable_clock_gating(adev, i);
1573 
1574 		/* enable VCN power gating */
1575 		vcn_v3_0_enable_static_power_gating(adev, i);
1576 	}
1577 
1578 	if (adev->pm.dpm_enabled)
1579 		amdgpu_dpm_enable_uvd(adev, false);
1580 
1581 	return 0;
1582 }
1583 
1584 static int vcn_v3_0_pause_dpg_mode(struct amdgpu_device *adev,
1585 		   int inst_idx, struct dpg_pause_state *new_state)
1586 {
1587 	volatile struct amdgpu_fw_shared *fw_shared;
1588 	struct amdgpu_ring *ring;
1589 	uint32_t reg_data = 0;
1590 	int ret_code;
1591 
1592 	/* pause/unpause if state is changed */
1593 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1594 		DRM_DEBUG("dpg pause state changed %d -> %d",
1595 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1596 		reg_data = RREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE) &
1597 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1598 
1599 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1600 			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1601 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1602 
1603 			if (!ret_code) {
1604 				/* pause DPG */
1605 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1606 				WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1607 
1608 				/* wait for ACK */
1609 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_DPG_PAUSE,
1610 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1611 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1612 
1613 				/* Stall DPG before WPTR/RPTR reset */
1614 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1615 					UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1616 					~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1617 
1618 				if (adev->ip_versions[UVD_HWIP][0] != IP_VERSION(3, 0, 33)) {
1619 					/* Restore */
1620 					fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
1621 					fw_shared->multi_queue.encode_generalpurpose_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1622 					ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1623 					ring->wptr = 0;
1624 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
1625 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1626 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
1627 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1628 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1629 					fw_shared->multi_queue.encode_generalpurpose_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1630 
1631 					fw_shared->multi_queue.encode_lowlatency_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1632 					ring = &adev->vcn.inst[inst_idx].ring_enc[1];
1633 					ring->wptr = 0;
1634 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1635 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1636 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
1637 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1638 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1639 					fw_shared->multi_queue.encode_lowlatency_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1640 
1641 					/* restore wptr/rptr with pointers saved in FW shared memory*/
1642 					WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, fw_shared->rb.rptr);
1643 					WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR, fw_shared->rb.wptr);
1644 				}
1645 
1646 				/* Unstall DPG */
1647 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1648 					0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1649 
1650 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS,
1651 					UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1652 			}
1653 		} else {
1654 			/* unpause dpg, no need to wait */
1655 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1656 			WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1657 		}
1658 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1659 	}
1660 
1661 	return 0;
1662 }
1663 
1664 /**
1665  * vcn_v3_0_dec_ring_get_rptr - get read pointer
1666  *
1667  * @ring: amdgpu_ring pointer
1668  *
1669  * Returns the current hardware read pointer
1670  */
1671 static uint64_t vcn_v3_0_dec_ring_get_rptr(struct amdgpu_ring *ring)
1672 {
1673 	struct amdgpu_device *adev = ring->adev;
1674 
1675 	return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_RPTR);
1676 }
1677 
1678 /**
1679  * vcn_v3_0_dec_ring_get_wptr - get write pointer
1680  *
1681  * @ring: amdgpu_ring pointer
1682  *
1683  * Returns the current hardware write pointer
1684  */
1685 static uint64_t vcn_v3_0_dec_ring_get_wptr(struct amdgpu_ring *ring)
1686 {
1687 	struct amdgpu_device *adev = ring->adev;
1688 
1689 	if (ring->use_doorbell)
1690 		return adev->wb.wb[ring->wptr_offs];
1691 	else
1692 		return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR);
1693 }
1694 
1695 /**
1696  * vcn_v3_0_dec_ring_set_wptr - set write pointer
1697  *
1698  * @ring: amdgpu_ring pointer
1699  *
1700  * Commits the write pointer to the hardware
1701  */
1702 static void vcn_v3_0_dec_ring_set_wptr(struct amdgpu_ring *ring)
1703 {
1704 	struct amdgpu_device *adev = ring->adev;
1705 	volatile struct amdgpu_fw_shared *fw_shared;
1706 
1707 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1708 		/*whenever update RBC_RB_WPTR, we save the wptr in shared rb.wptr and scratch2 */
1709 		fw_shared = adev->vcn.inst[ring->me].fw_shared.cpu_addr;
1710 		fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1711 		WREG32_SOC15(VCN, ring->me, mmUVD_SCRATCH2,
1712 			lower_32_bits(ring->wptr));
1713 	}
1714 
1715 	if (ring->use_doorbell) {
1716 		adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1717 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1718 	} else {
1719 		WREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1720 	}
1721 }
1722 
1723 static void vcn_v3_0_dec_sw_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
1724 				u64 seq, uint32_t flags)
1725 {
1726 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
1727 
1728 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_FENCE);
1729 	amdgpu_ring_write(ring, addr);
1730 	amdgpu_ring_write(ring, upper_32_bits(addr));
1731 	amdgpu_ring_write(ring, seq);
1732 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_TRAP);
1733 }
1734 
1735 static void vcn_v3_0_dec_sw_ring_insert_end(struct amdgpu_ring *ring)
1736 {
1737 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_END);
1738 }
1739 
1740 static void vcn_v3_0_dec_sw_ring_emit_ib(struct amdgpu_ring *ring,
1741 			       struct amdgpu_job *job,
1742 			       struct amdgpu_ib *ib,
1743 			       uint32_t flags)
1744 {
1745 	uint32_t vmid = AMDGPU_JOB_GET_VMID(job);
1746 
1747 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_IB);
1748 	amdgpu_ring_write(ring, vmid);
1749 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
1750 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
1751 	amdgpu_ring_write(ring, ib->length_dw);
1752 }
1753 
1754 static void vcn_v3_0_dec_sw_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
1755 				uint32_t val, uint32_t mask)
1756 {
1757 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_REG_WAIT);
1758 	amdgpu_ring_write(ring, reg << 2);
1759 	amdgpu_ring_write(ring, mask);
1760 	amdgpu_ring_write(ring, val);
1761 }
1762 
1763 static void vcn_v3_0_dec_sw_ring_emit_vm_flush(struct amdgpu_ring *ring,
1764 				uint32_t vmid, uint64_t pd_addr)
1765 {
1766 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
1767 	uint32_t data0, data1, mask;
1768 
1769 	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1770 
1771 	/* wait for register write */
1772 	data0 = hub->ctx0_ptb_addr_lo32 + vmid * hub->ctx_addr_distance;
1773 	data1 = lower_32_bits(pd_addr);
1774 	mask = 0xffffffff;
1775 	vcn_v3_0_dec_sw_ring_emit_reg_wait(ring, data0, data1, mask);
1776 }
1777 
1778 static void vcn_v3_0_dec_sw_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg, uint32_t val)
1779 {
1780 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_REG_WRITE);
1781 	amdgpu_ring_write(ring,	reg << 2);
1782 	amdgpu_ring_write(ring, val);
1783 }
1784 
1785 static const struct amdgpu_ring_funcs vcn_v3_0_dec_sw_ring_vm_funcs = {
1786 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1787 	.align_mask = 0x3f,
1788 	.nop = VCN_DEC_SW_CMD_NO_OP,
1789 	.secure_submission_supported = true,
1790 	.vmhub = AMDGPU_MMHUB_0,
1791 	.get_rptr = vcn_v3_0_dec_ring_get_rptr,
1792 	.get_wptr = vcn_v3_0_dec_ring_get_wptr,
1793 	.set_wptr = vcn_v3_0_dec_ring_set_wptr,
1794 	.emit_frame_size =
1795 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1796 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1797 		4 + /* vcn_v3_0_dec_sw_ring_emit_vm_flush */
1798 		5 + 5 + /* vcn_v3_0_dec_sw_ring_emit_fdec_swe x2 vm fdec_swe */
1799 		1, /* vcn_v3_0_dec_sw_ring_insert_end */
1800 	.emit_ib_size = 5, /* vcn_v3_0_dec_sw_ring_emit_ib */
1801 	.emit_ib = vcn_v3_0_dec_sw_ring_emit_ib,
1802 	.emit_fence = vcn_v3_0_dec_sw_ring_emit_fence,
1803 	.emit_vm_flush = vcn_v3_0_dec_sw_ring_emit_vm_flush,
1804 	.test_ring = amdgpu_vcn_dec_sw_ring_test_ring,
1805 	.test_ib = NULL,//amdgpu_vcn_dec_sw_ring_test_ib,
1806 	.insert_nop = amdgpu_ring_insert_nop,
1807 	.insert_end = vcn_v3_0_dec_sw_ring_insert_end,
1808 	.pad_ib = amdgpu_ring_generic_pad_ib,
1809 	.begin_use = amdgpu_vcn_ring_begin_use,
1810 	.end_use = amdgpu_vcn_ring_end_use,
1811 	.emit_wreg = vcn_v3_0_dec_sw_ring_emit_wreg,
1812 	.emit_reg_wait = vcn_v3_0_dec_sw_ring_emit_reg_wait,
1813 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1814 };
1815 
1816 static int vcn_v3_0_limit_sched(struct amdgpu_cs_parser *p,
1817 				struct amdgpu_job *job)
1818 {
1819 	struct drm_gpu_scheduler **scheds;
1820 
1821 	/* The create msg must be in the first IB submitted */
1822 	if (atomic_read(&job->base.entity->fence_seq))
1823 		return -EINVAL;
1824 
1825 	scheds = p->adev->gpu_sched[AMDGPU_HW_IP_VCN_DEC]
1826 		[AMDGPU_RING_PRIO_DEFAULT].sched;
1827 	drm_sched_entity_modify_sched(job->base.entity, scheds, 1);
1828 	return 0;
1829 }
1830 
1831 static int vcn_v3_0_dec_msg(struct amdgpu_cs_parser *p, struct amdgpu_job *job,
1832 			    uint64_t addr)
1833 {
1834 	struct ttm_operation_ctx ctx = { false, false };
1835 	struct amdgpu_bo_va_mapping *map;
1836 	uint32_t *msg, num_buffers;
1837 	struct amdgpu_bo *bo;
1838 	uint64_t start, end;
1839 	unsigned int i;
1840 	void * ptr;
1841 	int r;
1842 
1843 	addr &= AMDGPU_GMC_HOLE_MASK;
1844 	r = amdgpu_cs_find_mapping(p, addr, &bo, &map);
1845 	if (r) {
1846 		DRM_ERROR("Can't find BO for addr 0x%08Lx\n", addr);
1847 		return r;
1848 	}
1849 
1850 	start = map->start * AMDGPU_GPU_PAGE_SIZE;
1851 	end = (map->last + 1) * AMDGPU_GPU_PAGE_SIZE;
1852 	if (addr & 0x7) {
1853 		DRM_ERROR("VCN messages must be 8 byte aligned!\n");
1854 		return -EINVAL;
1855 	}
1856 
1857 	bo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
1858 	amdgpu_bo_placement_from_domain(bo, bo->allowed_domains);
1859 	r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
1860 	if (r) {
1861 		DRM_ERROR("Failed validating the VCN message BO (%d)!\n", r);
1862 		return r;
1863 	}
1864 
1865 	r = amdgpu_bo_kmap(bo, &ptr);
1866 	if (r) {
1867 		DRM_ERROR("Failed mapping the VCN message (%d)!\n", r);
1868 		return r;
1869 	}
1870 
1871 	msg = ptr + addr - start;
1872 
1873 	/* Check length */
1874 	if (msg[1] > end - addr) {
1875 		r = -EINVAL;
1876 		goto out;
1877 	}
1878 
1879 	if (msg[3] != RDECODE_MSG_CREATE)
1880 		goto out;
1881 
1882 	num_buffers = msg[2];
1883 	for (i = 0, msg = &msg[6]; i < num_buffers; ++i, msg += 4) {
1884 		uint32_t offset, size, *create;
1885 
1886 		if (msg[0] != RDECODE_MESSAGE_CREATE)
1887 			continue;
1888 
1889 		offset = msg[1];
1890 		size = msg[2];
1891 
1892 		if (offset + size > end) {
1893 			r = -EINVAL;
1894 			goto out;
1895 		}
1896 
1897 		create = ptr + addr + offset - start;
1898 
1899 		/* H246, HEVC and VP9 can run on any instance */
1900 		if (create[0] == 0x7 || create[0] == 0x10 || create[0] == 0x11)
1901 			continue;
1902 
1903 		r = vcn_v3_0_limit_sched(p, job);
1904 		if (r)
1905 			goto out;
1906 	}
1907 
1908 out:
1909 	amdgpu_bo_kunmap(bo);
1910 	return r;
1911 }
1912 
1913 static int vcn_v3_0_ring_patch_cs_in_place(struct amdgpu_cs_parser *p,
1914 					   struct amdgpu_job *job,
1915 					   struct amdgpu_ib *ib)
1916 {
1917 	struct amdgpu_ring *ring = to_amdgpu_ring(job->base.sched);
1918 	uint32_t msg_lo = 0, msg_hi = 0;
1919 	unsigned i;
1920 	int r;
1921 
1922 	/* The first instance can decode anything */
1923 	if (!ring->me)
1924 		return 0;
1925 
1926 	for (i = 0; i < ib->length_dw; i += 2) {
1927 		uint32_t reg = amdgpu_ib_get_value(ib, i);
1928 		uint32_t val = amdgpu_ib_get_value(ib, i + 1);
1929 
1930 		if (reg == PACKET0(p->adev->vcn.internal.data0, 0)) {
1931 			msg_lo = val;
1932 		} else if (reg == PACKET0(p->adev->vcn.internal.data1, 0)) {
1933 			msg_hi = val;
1934 		} else if (reg == PACKET0(p->adev->vcn.internal.cmd, 0) &&
1935 			   val == 0) {
1936 			r = vcn_v3_0_dec_msg(p, job,
1937 					     ((u64)msg_hi) << 32 | msg_lo);
1938 			if (r)
1939 				return r;
1940 		}
1941 	}
1942 	return 0;
1943 }
1944 
1945 static const struct amdgpu_ring_funcs vcn_v3_0_dec_ring_vm_funcs = {
1946 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1947 	.align_mask = 0xf,
1948 	.secure_submission_supported = true,
1949 	.vmhub = AMDGPU_MMHUB_0,
1950 	.get_rptr = vcn_v3_0_dec_ring_get_rptr,
1951 	.get_wptr = vcn_v3_0_dec_ring_get_wptr,
1952 	.set_wptr = vcn_v3_0_dec_ring_set_wptr,
1953 	.patch_cs_in_place = vcn_v3_0_ring_patch_cs_in_place,
1954 	.emit_frame_size =
1955 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1956 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1957 		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1958 		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1959 		6,
1960 	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1961 	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
1962 	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
1963 	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1964 	.test_ring = vcn_v2_0_dec_ring_test_ring,
1965 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
1966 	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
1967 	.insert_start = vcn_v2_0_dec_ring_insert_start,
1968 	.insert_end = vcn_v2_0_dec_ring_insert_end,
1969 	.pad_ib = amdgpu_ring_generic_pad_ib,
1970 	.begin_use = amdgpu_vcn_ring_begin_use,
1971 	.end_use = amdgpu_vcn_ring_end_use,
1972 	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1973 	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1974 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1975 };
1976 
1977 /**
1978  * vcn_v3_0_enc_ring_get_rptr - get enc read pointer
1979  *
1980  * @ring: amdgpu_ring pointer
1981  *
1982  * Returns the current hardware enc read pointer
1983  */
1984 static uint64_t vcn_v3_0_enc_ring_get_rptr(struct amdgpu_ring *ring)
1985 {
1986 	struct amdgpu_device *adev = ring->adev;
1987 
1988 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0])
1989 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR);
1990 	else
1991 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR2);
1992 }
1993 
1994 /**
1995  * vcn_v3_0_enc_ring_get_wptr - get enc write pointer
1996  *
1997  * @ring: amdgpu_ring pointer
1998  *
1999  * Returns the current hardware enc write pointer
2000  */
2001 static uint64_t vcn_v3_0_enc_ring_get_wptr(struct amdgpu_ring *ring)
2002 {
2003 	struct amdgpu_device *adev = ring->adev;
2004 
2005 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
2006 		if (ring->use_doorbell)
2007 			return adev->wb.wb[ring->wptr_offs];
2008 		else
2009 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR);
2010 	} else {
2011 		if (ring->use_doorbell)
2012 			return adev->wb.wb[ring->wptr_offs];
2013 		else
2014 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2);
2015 	}
2016 }
2017 
2018 /**
2019  * vcn_v3_0_enc_ring_set_wptr - set enc write pointer
2020  *
2021  * @ring: amdgpu_ring pointer
2022  *
2023  * Commits the enc write pointer to the hardware
2024  */
2025 static void vcn_v3_0_enc_ring_set_wptr(struct amdgpu_ring *ring)
2026 {
2027 	struct amdgpu_device *adev = ring->adev;
2028 
2029 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
2030 		if (ring->use_doorbell) {
2031 			adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
2032 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
2033 		} else {
2034 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
2035 		}
2036 	} else {
2037 		if (ring->use_doorbell) {
2038 			adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
2039 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
2040 		} else {
2041 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
2042 		}
2043 	}
2044 }
2045 
2046 static const struct amdgpu_ring_funcs vcn_v3_0_enc_ring_vm_funcs = {
2047 	.type = AMDGPU_RING_TYPE_VCN_ENC,
2048 	.align_mask = 0x3f,
2049 	.nop = VCN_ENC_CMD_NO_OP,
2050 	.vmhub = AMDGPU_MMHUB_0,
2051 	.get_rptr = vcn_v3_0_enc_ring_get_rptr,
2052 	.get_wptr = vcn_v3_0_enc_ring_get_wptr,
2053 	.set_wptr = vcn_v3_0_enc_ring_set_wptr,
2054 	.emit_frame_size =
2055 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2056 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
2057 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
2058 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
2059 		1, /* vcn_v2_0_enc_ring_insert_end */
2060 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
2061 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
2062 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
2063 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
2064 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
2065 	.test_ib = amdgpu_vcn_enc_ring_test_ib,
2066 	.insert_nop = amdgpu_ring_insert_nop,
2067 	.insert_end = vcn_v2_0_enc_ring_insert_end,
2068 	.pad_ib = amdgpu_ring_generic_pad_ib,
2069 	.begin_use = amdgpu_vcn_ring_begin_use,
2070 	.end_use = amdgpu_vcn_ring_end_use,
2071 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
2072 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
2073 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2074 };
2075 
2076 static void vcn_v3_0_set_dec_ring_funcs(struct amdgpu_device *adev)
2077 {
2078 	int i;
2079 
2080 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2081 		if (adev->vcn.harvest_config & (1 << i))
2082 			continue;
2083 
2084 		if (!DEC_SW_RING_ENABLED)
2085 			adev->vcn.inst[i].ring_dec.funcs = &vcn_v3_0_dec_ring_vm_funcs;
2086 		else
2087 			adev->vcn.inst[i].ring_dec.funcs = &vcn_v3_0_dec_sw_ring_vm_funcs;
2088 		adev->vcn.inst[i].ring_dec.me = i;
2089 		DRM_INFO("VCN(%d) decode%s is enabled in VM mode\n", i,
2090 			  DEC_SW_RING_ENABLED?"(Software Ring)":"");
2091 	}
2092 }
2093 
2094 static void vcn_v3_0_set_enc_ring_funcs(struct amdgpu_device *adev)
2095 {
2096 	int i, j;
2097 
2098 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2099 		if (adev->vcn.harvest_config & (1 << i))
2100 			continue;
2101 
2102 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
2103 			adev->vcn.inst[i].ring_enc[j].funcs = &vcn_v3_0_enc_ring_vm_funcs;
2104 			adev->vcn.inst[i].ring_enc[j].me = i;
2105 		}
2106 		if (adev->vcn.num_enc_rings > 0)
2107 			DRM_INFO("VCN(%d) encode is enabled in VM mode\n", i);
2108 	}
2109 }
2110 
2111 static bool vcn_v3_0_is_idle(void *handle)
2112 {
2113 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2114 	int i, ret = 1;
2115 
2116 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2117 		if (adev->vcn.harvest_config & (1 << i))
2118 			continue;
2119 
2120 		ret &= (RREG32_SOC15(VCN, i, mmUVD_STATUS) == UVD_STATUS__IDLE);
2121 	}
2122 
2123 	return ret;
2124 }
2125 
2126 static int vcn_v3_0_wait_for_idle(void *handle)
2127 {
2128 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2129 	int i, ret = 0;
2130 
2131 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2132 		if (adev->vcn.harvest_config & (1 << i))
2133 			continue;
2134 
2135 		ret = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE,
2136 			UVD_STATUS__IDLE);
2137 		if (ret)
2138 			return ret;
2139 	}
2140 
2141 	return ret;
2142 }
2143 
2144 static int vcn_v3_0_set_clockgating_state(void *handle,
2145 					  enum amd_clockgating_state state)
2146 {
2147 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2148 	bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
2149 	int i;
2150 
2151 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2152 		if (adev->vcn.harvest_config & (1 << i))
2153 			continue;
2154 
2155 		if (enable) {
2156 			if (RREG32_SOC15(VCN, i, mmUVD_STATUS) != UVD_STATUS__IDLE)
2157 				return -EBUSY;
2158 			vcn_v3_0_enable_clock_gating(adev, i);
2159 		} else {
2160 			vcn_v3_0_disable_clock_gating(adev, i);
2161 		}
2162 	}
2163 
2164 	return 0;
2165 }
2166 
2167 static int vcn_v3_0_set_powergating_state(void *handle,
2168 					  enum amd_powergating_state state)
2169 {
2170 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2171 	int ret;
2172 
2173 	/* for SRIOV, guest should not control VCN Power-gating
2174 	 * MMSCH FW should control Power-gating and clock-gating
2175 	 * guest should avoid touching CGC and PG
2176 	 */
2177 	if (amdgpu_sriov_vf(adev)) {
2178 		adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
2179 		return 0;
2180 	}
2181 
2182 	if(state == adev->vcn.cur_state)
2183 		return 0;
2184 
2185 	if (state == AMD_PG_STATE_GATE)
2186 		ret = vcn_v3_0_stop(adev);
2187 	else
2188 		ret = vcn_v3_0_start(adev);
2189 
2190 	if(!ret)
2191 		adev->vcn.cur_state = state;
2192 
2193 	return ret;
2194 }
2195 
2196 static int vcn_v3_0_set_interrupt_state(struct amdgpu_device *adev,
2197 					struct amdgpu_irq_src *source,
2198 					unsigned type,
2199 					enum amdgpu_interrupt_state state)
2200 {
2201 	return 0;
2202 }
2203 
2204 static int vcn_v3_0_process_interrupt(struct amdgpu_device *adev,
2205 				      struct amdgpu_irq_src *source,
2206 				      struct amdgpu_iv_entry *entry)
2207 {
2208 	uint32_t ip_instance;
2209 
2210 	switch (entry->client_id) {
2211 	case SOC15_IH_CLIENTID_VCN:
2212 		ip_instance = 0;
2213 		break;
2214 	case SOC15_IH_CLIENTID_VCN1:
2215 		ip_instance = 1;
2216 		break;
2217 	default:
2218 		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
2219 		return 0;
2220 	}
2221 
2222 	DRM_DEBUG("IH: VCN TRAP\n");
2223 
2224 	switch (entry->src_id) {
2225 	case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
2226 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec);
2227 		break;
2228 	case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
2229 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
2230 		break;
2231 	case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
2232 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]);
2233 		break;
2234 	default:
2235 		DRM_ERROR("Unhandled interrupt: %d %d\n",
2236 			  entry->src_id, entry->src_data[0]);
2237 		break;
2238 	}
2239 
2240 	return 0;
2241 }
2242 
2243 static const struct amdgpu_irq_src_funcs vcn_v3_0_irq_funcs = {
2244 	.set = vcn_v3_0_set_interrupt_state,
2245 	.process = vcn_v3_0_process_interrupt,
2246 };
2247 
2248 static void vcn_v3_0_set_irq_funcs(struct amdgpu_device *adev)
2249 {
2250 	int i;
2251 
2252 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2253 		if (adev->vcn.harvest_config & (1 << i))
2254 			continue;
2255 
2256 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
2257 		adev->vcn.inst[i].irq.funcs = &vcn_v3_0_irq_funcs;
2258 	}
2259 }
2260 
2261 static const struct amd_ip_funcs vcn_v3_0_ip_funcs = {
2262 	.name = "vcn_v3_0",
2263 	.early_init = vcn_v3_0_early_init,
2264 	.late_init = NULL,
2265 	.sw_init = vcn_v3_0_sw_init,
2266 	.sw_fini = vcn_v3_0_sw_fini,
2267 	.hw_init = vcn_v3_0_hw_init,
2268 	.hw_fini = vcn_v3_0_hw_fini,
2269 	.suspend = vcn_v3_0_suspend,
2270 	.resume = vcn_v3_0_resume,
2271 	.is_idle = vcn_v3_0_is_idle,
2272 	.wait_for_idle = vcn_v3_0_wait_for_idle,
2273 	.check_soft_reset = NULL,
2274 	.pre_soft_reset = NULL,
2275 	.soft_reset = NULL,
2276 	.post_soft_reset = NULL,
2277 	.set_clockgating_state = vcn_v3_0_set_clockgating_state,
2278 	.set_powergating_state = vcn_v3_0_set_powergating_state,
2279 };
2280 
2281 const struct amdgpu_ip_block_version vcn_v3_0_ip_block =
2282 {
2283 	.type = AMD_IP_BLOCK_TYPE_VCN,
2284 	.major = 3,
2285 	.minor = 0,
2286 	.rev = 0,
2287 	.funcs = &vcn_v3_0_ip_funcs,
2288 };
2289