xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c (revision 703e7713)
1 /*
2  * Copyright 2019 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #include <linux/firmware.h>
25 #include "amdgpu.h"
26 #include "amdgpu_vcn.h"
27 #include "amdgpu_pm.h"
28 #include "amdgpu_cs.h"
29 #include "soc15.h"
30 #include "soc15d.h"
31 #include "vcn_v2_0.h"
32 #include "mmsch_v3_0.h"
33 #include "vcn_sw_ring.h"
34 
35 #include "vcn/vcn_3_0_0_offset.h"
36 #include "vcn/vcn_3_0_0_sh_mask.h"
37 #include "ivsrcid/vcn/irqsrcs_vcn_2_0.h"
38 
39 #include <drm/drm_drv.h>
40 
41 #define VCN_VID_SOC_ADDRESS_2_0					0x1fa00
42 #define VCN1_VID_SOC_ADDRESS_3_0				0x48200
43 
44 #define mmUVD_CONTEXT_ID_INTERNAL_OFFSET			0x27
45 #define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET			0x0f
46 #define mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET			0x10
47 #define mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET			0x11
48 #define mmUVD_NO_OP_INTERNAL_OFFSET				0x29
49 #define mmUVD_GP_SCRATCH8_INTERNAL_OFFSET			0x66
50 #define mmUVD_SCRATCH9_INTERNAL_OFFSET				0xc01d
51 
52 #define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET			0x431
53 #define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET		0x3b4
54 #define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET		0x3b5
55 #define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET			0x25c
56 
57 #define VCN_INSTANCES_SIENNA_CICHLID				2
58 #define DEC_SW_RING_ENABLED					FALSE
59 
60 #define RDECODE_MSG_CREATE					0x00000000
61 #define RDECODE_MESSAGE_CREATE					0x00000001
62 
63 static int amdgpu_ih_clientid_vcns[] = {
64 	SOC15_IH_CLIENTID_VCN,
65 	SOC15_IH_CLIENTID_VCN1
66 };
67 
68 static int vcn_v3_0_start_sriov(struct amdgpu_device *adev);
69 static void vcn_v3_0_set_dec_ring_funcs(struct amdgpu_device *adev);
70 static void vcn_v3_0_set_enc_ring_funcs(struct amdgpu_device *adev);
71 static void vcn_v3_0_set_irq_funcs(struct amdgpu_device *adev);
72 static int vcn_v3_0_set_powergating_state(void *handle,
73 			enum amd_powergating_state state);
74 static int vcn_v3_0_pause_dpg_mode(struct amdgpu_device *adev,
75 			int inst_idx, struct dpg_pause_state *new_state);
76 
77 static void vcn_v3_0_dec_ring_set_wptr(struct amdgpu_ring *ring);
78 static void vcn_v3_0_enc_ring_set_wptr(struct amdgpu_ring *ring);
79 
80 /**
81  * vcn_v3_0_early_init - set function pointers and load microcode
82  *
83  * @handle: amdgpu_device pointer
84  *
85  * Set ring and irq function pointers
86  * Load microcode from filesystem
87  */
88 static int vcn_v3_0_early_init(void *handle)
89 {
90 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
91 
92 	if (amdgpu_sriov_vf(adev)) {
93 		adev->vcn.num_vcn_inst = VCN_INSTANCES_SIENNA_CICHLID;
94 		adev->vcn.harvest_config = 0;
95 		adev->vcn.num_enc_rings = 1;
96 
97 	} else {
98 		if (adev->vcn.harvest_config == (AMDGPU_VCN_HARVEST_VCN0 |
99 						 AMDGPU_VCN_HARVEST_VCN1))
100 			/* both instances are harvested, disable the block */
101 			return -ENOENT;
102 
103 		if (adev->ip_versions[UVD_HWIP][0] == IP_VERSION(3, 0, 33))
104 			adev->vcn.num_enc_rings = 0;
105 		else
106 			adev->vcn.num_enc_rings = 2;
107 	}
108 
109 	vcn_v3_0_set_dec_ring_funcs(adev);
110 	vcn_v3_0_set_enc_ring_funcs(adev);
111 	vcn_v3_0_set_irq_funcs(adev);
112 
113 	return amdgpu_vcn_early_init(adev);
114 }
115 
116 /**
117  * vcn_v3_0_sw_init - sw init for VCN block
118  *
119  * @handle: amdgpu_device pointer
120  *
121  * Load firmware and sw initialization
122  */
123 static int vcn_v3_0_sw_init(void *handle)
124 {
125 	struct amdgpu_ring *ring;
126 	int i, j, r;
127 	int vcn_doorbell_index = 0;
128 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
129 
130 	r = amdgpu_vcn_sw_init(adev);
131 	if (r)
132 		return r;
133 
134 	amdgpu_vcn_setup_ucode(adev);
135 
136 	r = amdgpu_vcn_resume(adev);
137 	if (r)
138 		return r;
139 
140 	/*
141 	 * Note: doorbell assignment is fixed for SRIOV multiple VCN engines
142 	 * Formula:
143 	 *   vcn_db_base  = adev->doorbell_index.vcn.vcn_ring0_1 << 1;
144 	 *   dec_ring_i   = vcn_db_base + i * (adev->vcn.num_enc_rings + 1)
145 	 *   enc_ring_i,j = vcn_db_base + i * (adev->vcn.num_enc_rings + 1) + 1 + j
146 	 */
147 	if (amdgpu_sriov_vf(adev)) {
148 		vcn_doorbell_index = adev->doorbell_index.vcn.vcn_ring0_1;
149 		/* get DWORD offset */
150 		vcn_doorbell_index = vcn_doorbell_index << 1;
151 	}
152 
153 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
154 		volatile struct amdgpu_fw_shared *fw_shared;
155 
156 		if (adev->vcn.harvest_config & (1 << i))
157 			continue;
158 
159 		adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
160 		adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
161 		adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
162 		adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
163 		adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
164 		adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
165 
166 		adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
167 		adev->vcn.inst[i].external.scratch9 = SOC15_REG_OFFSET(VCN, i, mmUVD_SCRATCH9);
168 		adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
169 		adev->vcn.inst[i].external.data0 = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_DATA0);
170 		adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
171 		adev->vcn.inst[i].external.data1 = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_DATA1);
172 		adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
173 		adev->vcn.inst[i].external.cmd = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_CMD);
174 		adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
175 		adev->vcn.inst[i].external.nop = SOC15_REG_OFFSET(VCN, i, mmUVD_NO_OP);
176 
177 		/* VCN DEC TRAP */
178 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
179 				VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst[i].irq);
180 		if (r)
181 			return r;
182 
183 		atomic_set(&adev->vcn.inst[i].sched_score, 0);
184 
185 		ring = &adev->vcn.inst[i].ring_dec;
186 		ring->use_doorbell = true;
187 		if (amdgpu_sriov_vf(adev)) {
188 			ring->doorbell_index = vcn_doorbell_index + i * (adev->vcn.num_enc_rings + 1);
189 		} else {
190 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i;
191 		}
192 		ring->vm_hub = AMDGPU_MMHUB0(0);
193 		sprintf(ring->name, "vcn_dec_%d", i);
194 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
195 				     AMDGPU_RING_PRIO_DEFAULT,
196 				     &adev->vcn.inst[i].sched_score);
197 		if (r)
198 			return r;
199 
200 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
201 			enum amdgpu_ring_priority_level hw_prio = amdgpu_vcn_get_enc_ring_prio(j);
202 
203 			/* VCN ENC TRAP */
204 			r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
205 				j + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);
206 			if (r)
207 				return r;
208 
209 			ring = &adev->vcn.inst[i].ring_enc[j];
210 			ring->use_doorbell = true;
211 			if (amdgpu_sriov_vf(adev)) {
212 				ring->doorbell_index = vcn_doorbell_index + i * (adev->vcn.num_enc_rings + 1) + 1 + j;
213 			} else {
214 				ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + j + 8 * i;
215 			}
216 			ring->vm_hub = AMDGPU_MMHUB0(0);
217 			sprintf(ring->name, "vcn_enc_%d.%d", i, j);
218 			r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
219 					     hw_prio, &adev->vcn.inst[i].sched_score);
220 			if (r)
221 				return r;
222 		}
223 
224 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
225 		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_SW_RING_FLAG) |
226 					     cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG) |
227 					     cpu_to_le32(AMDGPU_VCN_FW_SHARED_FLAG_0_RB);
228 		fw_shared->sw_ring.is_enabled = cpu_to_le32(DEC_SW_RING_ENABLED);
229 		fw_shared->present_flag_0 |= AMDGPU_VCN_SMU_VERSION_INFO_FLAG;
230 		if (adev->ip_versions[UVD_HWIP][0] == IP_VERSION(3, 1, 2))
231 			fw_shared->smu_interface_info.smu_interface_type = 2;
232 		else if (adev->ip_versions[UVD_HWIP][0] == IP_VERSION(3, 1, 1))
233 			fw_shared->smu_interface_info.smu_interface_type = 1;
234 
235 		if (amdgpu_vcnfw_log)
236 			amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
237 	}
238 
239 	if (amdgpu_sriov_vf(adev)) {
240 		r = amdgpu_virt_alloc_mm_table(adev);
241 		if (r)
242 			return r;
243 	}
244 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
245 		adev->vcn.pause_dpg_mode = vcn_v3_0_pause_dpg_mode;
246 
247 	return 0;
248 }
249 
250 /**
251  * vcn_v3_0_sw_fini - sw fini for VCN block
252  *
253  * @handle: amdgpu_device pointer
254  *
255  * VCN suspend and free up sw allocation
256  */
257 static int vcn_v3_0_sw_fini(void *handle)
258 {
259 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
260 	int i, r, idx;
261 
262 	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
263 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
264 			volatile struct amdgpu_fw_shared *fw_shared;
265 
266 			if (adev->vcn.harvest_config & (1 << i))
267 				continue;
268 			fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
269 			fw_shared->present_flag_0 = 0;
270 			fw_shared->sw_ring.is_enabled = false;
271 		}
272 
273 		drm_dev_exit(idx);
274 	}
275 
276 	if (amdgpu_sriov_vf(adev))
277 		amdgpu_virt_free_mm_table(adev);
278 
279 	r = amdgpu_vcn_suspend(adev);
280 	if (r)
281 		return r;
282 
283 	r = amdgpu_vcn_sw_fini(adev);
284 
285 	return r;
286 }
287 
288 /**
289  * vcn_v3_0_hw_init - start and test VCN block
290  *
291  * @handle: amdgpu_device pointer
292  *
293  * Initialize the hardware, boot up the VCPU and do some testing
294  */
295 static int vcn_v3_0_hw_init(void *handle)
296 {
297 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
298 	struct amdgpu_ring *ring;
299 	int i, j, r;
300 
301 	if (amdgpu_sriov_vf(adev)) {
302 		r = vcn_v3_0_start_sriov(adev);
303 		if (r)
304 			goto done;
305 
306 		/* initialize VCN dec and enc ring buffers */
307 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
308 			if (adev->vcn.harvest_config & (1 << i))
309 				continue;
310 
311 			ring = &adev->vcn.inst[i].ring_dec;
312 			if (amdgpu_vcn_is_disabled_vcn(adev, VCN_DECODE_RING, i)) {
313 				ring->sched.ready = false;
314 				ring->no_scheduler = true;
315 				dev_info(adev->dev, "ring %s is disabled by hypervisor\n", ring->name);
316 			} else {
317 				ring->wptr = 0;
318 				ring->wptr_old = 0;
319 				vcn_v3_0_dec_ring_set_wptr(ring);
320 				ring->sched.ready = true;
321 			}
322 
323 			for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
324 				ring = &adev->vcn.inst[i].ring_enc[j];
325 				if (amdgpu_vcn_is_disabled_vcn(adev, VCN_ENCODE_RING, i)) {
326 					ring->sched.ready = false;
327 					ring->no_scheduler = true;
328 					dev_info(adev->dev, "ring %s is disabled by hypervisor\n", ring->name);
329 				} else {
330 					ring->wptr = 0;
331 					ring->wptr_old = 0;
332 					vcn_v3_0_enc_ring_set_wptr(ring);
333 					ring->sched.ready = true;
334 				}
335 			}
336 		}
337 	} else {
338 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
339 			if (adev->vcn.harvest_config & (1 << i))
340 				continue;
341 
342 			ring = &adev->vcn.inst[i].ring_dec;
343 
344 			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
345 						     ring->doorbell_index, i);
346 
347 			r = amdgpu_ring_test_helper(ring);
348 			if (r)
349 				goto done;
350 
351 			for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
352 				ring = &adev->vcn.inst[i].ring_enc[j];
353 				r = amdgpu_ring_test_helper(ring);
354 				if (r)
355 					goto done;
356 			}
357 		}
358 	}
359 
360 done:
361 	if (!r)
362 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
363 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
364 
365 	return r;
366 }
367 
368 /**
369  * vcn_v3_0_hw_fini - stop the hardware block
370  *
371  * @handle: amdgpu_device pointer
372  *
373  * Stop the VCN block, mark ring as not ready any more
374  */
375 static int vcn_v3_0_hw_fini(void *handle)
376 {
377 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
378 	int i;
379 
380 	cancel_delayed_work_sync(&adev->vcn.idle_work);
381 
382 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
383 		if (adev->vcn.harvest_config & (1 << i))
384 			continue;
385 
386 		if (!amdgpu_sriov_vf(adev)) {
387 			if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
388 					(adev->vcn.cur_state != AMD_PG_STATE_GATE &&
389 					 RREG32_SOC15(VCN, i, mmUVD_STATUS))) {
390 				vcn_v3_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
391 			}
392 		}
393 	}
394 
395 	return 0;
396 }
397 
398 /**
399  * vcn_v3_0_suspend - suspend VCN block
400  *
401  * @handle: amdgpu_device pointer
402  *
403  * HW fini and suspend VCN block
404  */
405 static int vcn_v3_0_suspend(void *handle)
406 {
407 	int r;
408 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
409 
410 	r = vcn_v3_0_hw_fini(adev);
411 	if (r)
412 		return r;
413 
414 	r = amdgpu_vcn_suspend(adev);
415 
416 	return r;
417 }
418 
419 /**
420  * vcn_v3_0_resume - resume VCN block
421  *
422  * @handle: amdgpu_device pointer
423  *
424  * Resume firmware and hw init VCN block
425  */
426 static int vcn_v3_0_resume(void *handle)
427 {
428 	int r;
429 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
430 
431 	r = amdgpu_vcn_resume(adev);
432 	if (r)
433 		return r;
434 
435 	r = vcn_v3_0_hw_init(adev);
436 
437 	return r;
438 }
439 
440 /**
441  * vcn_v3_0_mc_resume - memory controller programming
442  *
443  * @adev: amdgpu_device pointer
444  * @inst: instance number
445  *
446  * Let the VCN memory controller know it's offsets
447  */
448 static void vcn_v3_0_mc_resume(struct amdgpu_device *adev, int inst)
449 {
450 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
451 	uint32_t offset;
452 
453 	/* cache window 0: fw */
454 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
455 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
456 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo));
457 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
458 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi));
459 		WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET0, 0);
460 		offset = 0;
461 	} else {
462 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
463 			lower_32_bits(adev->vcn.inst[inst].gpu_addr));
464 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
465 			upper_32_bits(adev->vcn.inst[inst].gpu_addr));
466 		offset = size;
467 		WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET0,
468 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
469 	}
470 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE0, size);
471 
472 	/* cache window 1: stack */
473 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
474 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
475 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
476 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
477 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET1, 0);
478 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
479 
480 	/* cache window 2: context */
481 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
482 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
483 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
484 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
485 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET2, 0);
486 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
487 
488 	/* non-cache window */
489 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
490 		lower_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
491 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
492 		upper_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
493 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
494 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_NONCACHE_SIZE0,
495 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
496 }
497 
498 static void vcn_v3_0_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
499 {
500 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
501 	uint32_t offset;
502 
503 	/* cache window 0: fw */
504 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
505 		if (!indirect) {
506 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
507 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
508 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
509 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
510 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
511 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
512 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
513 				VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
514 		} else {
515 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
516 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
517 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
518 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
519 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
520 				VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
521 		}
522 		offset = 0;
523 	} else {
524 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
525 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
526 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
527 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
528 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
529 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
530 		offset = size;
531 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
532 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0),
533 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
534 	}
535 
536 	if (!indirect)
537 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
538 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
539 	else
540 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
541 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
542 
543 	/* cache window 1: stack */
544 	if (!indirect) {
545 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
546 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
547 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
548 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
549 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
550 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
551 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
552 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
553 	} else {
554 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
555 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
556 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
557 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
558 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
559 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
560 	}
561 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
562 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
563 
564 	/* cache window 2: context */
565 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
566 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
567 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
568 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
569 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
570 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
571 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
572 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
573 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
574 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
575 
576 	/* non-cache window */
577 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
578 			VCN, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
579 			lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
580 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
581 			VCN, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
582 			upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
583 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
584 			VCN, inst_idx, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
585 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
586 			VCN, inst_idx, mmUVD_VCPU_NONCACHE_SIZE0),
587 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
588 
589 	/* VCN global tiling registers */
590 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
591 		UVD, inst_idx, mmUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
592 }
593 
594 static void vcn_v3_0_disable_static_power_gating(struct amdgpu_device *adev, int inst)
595 {
596 	uint32_t data = 0;
597 
598 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
599 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
600 			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
601 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
602 			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
603 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
604 			| 2 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
605 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
606 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
607 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
608 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
609 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
610 			| 2 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
611 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
612 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
613 
614 		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
615 		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS,
616 			UVD_PGFSM_STATUS__UVDM_UVDU_UVDLM_PWR_ON_3_0, 0x3F3FFFFF);
617 	} else {
618 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
619 			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
620 			| 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
621 			| 1 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
622 			| 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
623 			| 1 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
624 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
625 			| 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
626 			| 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
627 			| 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
628 			| 1 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
629 			| 1 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
630 			| 1 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
631 			| 1 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
632 		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
633 		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS, 0,  0x3F3FFFFF);
634 	}
635 
636 	data = RREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS);
637 	data &= ~0x103;
638 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
639 		data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
640 			UVD_POWER_STATUS__UVD_PG_EN_MASK;
641 
642 	WREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS, data);
643 }
644 
645 static void vcn_v3_0_enable_static_power_gating(struct amdgpu_device *adev, int inst)
646 {
647 	uint32_t data;
648 
649 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
650 		/* Before power off, this indicator has to be turned on */
651 		data = RREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS);
652 		data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
653 		data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
654 		WREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS, data);
655 
656 		data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
657 			| 2 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
658 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
659 			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
660 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
661 			| 2 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
662 			| 2 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
663 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
664 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
665 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
666 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
667 			| 2 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
668 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
669 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
670 		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
671 
672 		data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
673 			| 2 << UVD_PGFSM_STATUS__UVDU_PWR_STATUS__SHIFT
674 			| 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
675 			| 2 << UVD_PGFSM_STATUS__UVDC_PWR_STATUS__SHIFT
676 			| 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
677 			| 2 << UVD_PGFSM_STATUS__UVDIRL_PWR_STATUS__SHIFT
678 			| 2 << UVD_PGFSM_STATUS__UVDLM_PWR_STATUS__SHIFT
679 			| 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
680 			| 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
681 			| 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT
682 			| 2 << UVD_PGFSM_STATUS__UVDAB_PWR_STATUS__SHIFT
683 			| 2 << UVD_PGFSM_STATUS__UVDATD_PWR_STATUS__SHIFT
684 			| 2 << UVD_PGFSM_STATUS__UVDNA_PWR_STATUS__SHIFT
685 			| 2 << UVD_PGFSM_STATUS__UVDNB_PWR_STATUS__SHIFT);
686 		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS, data, 0x3F3FFFFF);
687 	}
688 }
689 
690 /**
691  * vcn_v3_0_disable_clock_gating - disable VCN clock gating
692  *
693  * @adev: amdgpu_device pointer
694  * @inst: instance number
695  *
696  * Disable clock gating for VCN block
697  */
698 static void vcn_v3_0_disable_clock_gating(struct amdgpu_device *adev, int inst)
699 {
700 	uint32_t data;
701 
702 	/* VCN disable CGC */
703 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
704 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
705 		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
706 	else
707 		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
708 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
709 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
710 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
711 
712 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_GATE);
713 	data &= ~(UVD_CGC_GATE__SYS_MASK
714 		| UVD_CGC_GATE__UDEC_MASK
715 		| UVD_CGC_GATE__MPEG2_MASK
716 		| UVD_CGC_GATE__REGS_MASK
717 		| UVD_CGC_GATE__RBC_MASK
718 		| UVD_CGC_GATE__LMI_MC_MASK
719 		| UVD_CGC_GATE__LMI_UMC_MASK
720 		| UVD_CGC_GATE__IDCT_MASK
721 		| UVD_CGC_GATE__MPRD_MASK
722 		| UVD_CGC_GATE__MPC_MASK
723 		| UVD_CGC_GATE__LBSI_MASK
724 		| UVD_CGC_GATE__LRBBM_MASK
725 		| UVD_CGC_GATE__UDEC_RE_MASK
726 		| UVD_CGC_GATE__UDEC_CM_MASK
727 		| UVD_CGC_GATE__UDEC_IT_MASK
728 		| UVD_CGC_GATE__UDEC_DB_MASK
729 		| UVD_CGC_GATE__UDEC_MP_MASK
730 		| UVD_CGC_GATE__WCB_MASK
731 		| UVD_CGC_GATE__VCPU_MASK
732 		| UVD_CGC_GATE__MMSCH_MASK);
733 
734 	WREG32_SOC15(VCN, inst, mmUVD_CGC_GATE, data);
735 
736 	SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_CGC_GATE, 0,  0xFFFFFFFF);
737 
738 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
739 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
740 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
741 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
742 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
743 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
744 		| UVD_CGC_CTRL__SYS_MODE_MASK
745 		| UVD_CGC_CTRL__UDEC_MODE_MASK
746 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
747 		| UVD_CGC_CTRL__REGS_MODE_MASK
748 		| UVD_CGC_CTRL__RBC_MODE_MASK
749 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
750 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
751 		| UVD_CGC_CTRL__IDCT_MODE_MASK
752 		| UVD_CGC_CTRL__MPRD_MODE_MASK
753 		| UVD_CGC_CTRL__MPC_MODE_MASK
754 		| UVD_CGC_CTRL__LBSI_MODE_MASK
755 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
756 		| UVD_CGC_CTRL__WCB_MODE_MASK
757 		| UVD_CGC_CTRL__VCPU_MODE_MASK
758 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
759 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
760 
761 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE);
762 	data |= (UVD_SUVD_CGC_GATE__SRE_MASK
763 		| UVD_SUVD_CGC_GATE__SIT_MASK
764 		| UVD_SUVD_CGC_GATE__SMP_MASK
765 		| UVD_SUVD_CGC_GATE__SCM_MASK
766 		| UVD_SUVD_CGC_GATE__SDB_MASK
767 		| UVD_SUVD_CGC_GATE__SRE_H264_MASK
768 		| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
769 		| UVD_SUVD_CGC_GATE__SIT_H264_MASK
770 		| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
771 		| UVD_SUVD_CGC_GATE__SCM_H264_MASK
772 		| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
773 		| UVD_SUVD_CGC_GATE__SDB_H264_MASK
774 		| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
775 		| UVD_SUVD_CGC_GATE__SCLR_MASK
776 		| UVD_SUVD_CGC_GATE__ENT_MASK
777 		| UVD_SUVD_CGC_GATE__IME_MASK
778 		| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
779 		| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
780 		| UVD_SUVD_CGC_GATE__SITE_MASK
781 		| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
782 		| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
783 		| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
784 		| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
785 		| UVD_SUVD_CGC_GATE__IME_HEVC_MASK
786 		| UVD_SUVD_CGC_GATE__EFC_MASK
787 		| UVD_SUVD_CGC_GATE__SAOE_MASK
788 		| UVD_SUVD_CGC_GATE__SRE_AV1_MASK
789 		| UVD_SUVD_CGC_GATE__FBC_PCLK_MASK
790 		| UVD_SUVD_CGC_GATE__FBC_CCLK_MASK
791 		| UVD_SUVD_CGC_GATE__SCM_AV1_MASK
792 		| UVD_SUVD_CGC_GATE__SMPA_MASK);
793 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE, data);
794 
795 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE2);
796 	data |= (UVD_SUVD_CGC_GATE2__MPBE0_MASK
797 		| UVD_SUVD_CGC_GATE2__MPBE1_MASK
798 		| UVD_SUVD_CGC_GATE2__SIT_AV1_MASK
799 		| UVD_SUVD_CGC_GATE2__SDB_AV1_MASK
800 		| UVD_SUVD_CGC_GATE2__MPC1_MASK);
801 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE2, data);
802 
803 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL);
804 	data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
805 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
806 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
807 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
808 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
809 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
810 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
811 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
812 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK
813 		| UVD_SUVD_CGC_CTRL__EFC_MODE_MASK
814 		| UVD_SUVD_CGC_CTRL__SAOE_MODE_MASK
815 		| UVD_SUVD_CGC_CTRL__SMPA_MODE_MASK
816 		| UVD_SUVD_CGC_CTRL__MPBE0_MODE_MASK
817 		| UVD_SUVD_CGC_CTRL__MPBE1_MODE_MASK
818 		| UVD_SUVD_CGC_CTRL__SIT_AV1_MODE_MASK
819 		| UVD_SUVD_CGC_CTRL__SDB_AV1_MODE_MASK
820 		| UVD_SUVD_CGC_CTRL__MPC1_MODE_MASK
821 		| UVD_SUVD_CGC_CTRL__FBC_PCLK_MASK
822 		| UVD_SUVD_CGC_CTRL__FBC_CCLK_MASK);
823 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL, data);
824 }
825 
826 static void vcn_v3_0_clock_gating_dpg_mode(struct amdgpu_device *adev,
827 		uint8_t sram_sel, int inst_idx, uint8_t indirect)
828 {
829 	uint32_t reg_data = 0;
830 
831 	/* enable sw clock gating control */
832 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
833 		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
834 	else
835 		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
836 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
837 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
838 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
839 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
840 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
841 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
842 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
843 		 UVD_CGC_CTRL__SYS_MODE_MASK |
844 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
845 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
846 		 UVD_CGC_CTRL__REGS_MODE_MASK |
847 		 UVD_CGC_CTRL__RBC_MODE_MASK |
848 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
849 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
850 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
851 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
852 		 UVD_CGC_CTRL__MPC_MODE_MASK |
853 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
854 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
855 		 UVD_CGC_CTRL__WCB_MODE_MASK |
856 		 UVD_CGC_CTRL__VCPU_MODE_MASK |
857 		 UVD_CGC_CTRL__MMSCH_MODE_MASK);
858 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
859 		VCN, inst_idx, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
860 
861 	/* turn off clock gating */
862 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
863 		VCN, inst_idx, mmUVD_CGC_GATE), 0, sram_sel, indirect);
864 
865 	/* turn on SUVD clock gating */
866 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
867 		VCN, inst_idx, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
868 
869 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
870 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
871 		VCN, inst_idx, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
872 }
873 
874 /**
875  * vcn_v3_0_enable_clock_gating - enable VCN clock gating
876  *
877  * @adev: amdgpu_device pointer
878  * @inst: instance number
879  *
880  * Enable clock gating for VCN block
881  */
882 static void vcn_v3_0_enable_clock_gating(struct amdgpu_device *adev, int inst)
883 {
884 	uint32_t data;
885 
886 	/* enable VCN CGC */
887 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
888 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
889 		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
890 	else
891 		data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
892 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
893 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
894 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
895 
896 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
897 	data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
898 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
899 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
900 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
901 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
902 		| UVD_CGC_CTRL__SYS_MODE_MASK
903 		| UVD_CGC_CTRL__UDEC_MODE_MASK
904 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
905 		| UVD_CGC_CTRL__REGS_MODE_MASK
906 		| UVD_CGC_CTRL__RBC_MODE_MASK
907 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
908 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
909 		| UVD_CGC_CTRL__IDCT_MODE_MASK
910 		| UVD_CGC_CTRL__MPRD_MODE_MASK
911 		| UVD_CGC_CTRL__MPC_MODE_MASK
912 		| UVD_CGC_CTRL__LBSI_MODE_MASK
913 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
914 		| UVD_CGC_CTRL__WCB_MODE_MASK
915 		| UVD_CGC_CTRL__VCPU_MODE_MASK
916 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
917 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
918 
919 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL);
920 	data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
921 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
922 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
923 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
924 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
925 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
926 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
927 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
928 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK
929 		| UVD_SUVD_CGC_CTRL__EFC_MODE_MASK
930 		| UVD_SUVD_CGC_CTRL__SAOE_MODE_MASK
931 		| UVD_SUVD_CGC_CTRL__SMPA_MODE_MASK
932 		| UVD_SUVD_CGC_CTRL__MPBE0_MODE_MASK
933 		| UVD_SUVD_CGC_CTRL__MPBE1_MODE_MASK
934 		| UVD_SUVD_CGC_CTRL__SIT_AV1_MODE_MASK
935 		| UVD_SUVD_CGC_CTRL__SDB_AV1_MODE_MASK
936 		| UVD_SUVD_CGC_CTRL__MPC1_MODE_MASK
937 		| UVD_SUVD_CGC_CTRL__FBC_PCLK_MASK
938 		| UVD_SUVD_CGC_CTRL__FBC_CCLK_MASK);
939 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL, data);
940 }
941 
942 static int vcn_v3_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
943 {
944 	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
945 	struct amdgpu_ring *ring;
946 	uint32_t rb_bufsz, tmp;
947 
948 	/* disable register anti-hang mechanism */
949 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1,
950 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
951 	/* enable dynamic power gating mode */
952 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS);
953 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
954 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
955 	WREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS, tmp);
956 
957 	if (indirect)
958 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
959 
960 	/* enable clock gating */
961 	vcn_v3_0_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
962 
963 	/* enable VCPU clock */
964 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
965 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
966 	tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
967 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
968 		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
969 
970 	/* disable master interupt */
971 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
972 		VCN, inst_idx, mmUVD_MASTINT_EN), 0, 0, indirect);
973 
974 	/* setup mmUVD_LMI_CTRL */
975 	tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
976 		UVD_LMI_CTRL__REQ_MODE_MASK |
977 		UVD_LMI_CTRL__CRC_RESET_MASK |
978 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
979 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
980 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
981 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
982 		0x00100000L);
983 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
984 		VCN, inst_idx, mmUVD_LMI_CTRL), tmp, 0, indirect);
985 
986 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
987 		VCN, inst_idx, mmUVD_MPC_CNTL),
988 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
989 
990 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
991 		VCN, inst_idx, mmUVD_MPC_SET_MUXA0),
992 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
993 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
994 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
995 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
996 
997 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
998 		VCN, inst_idx, mmUVD_MPC_SET_MUXB0),
999 		 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1000 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1001 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1002 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
1003 
1004 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1005 		VCN, inst_idx, mmUVD_MPC_SET_MUX),
1006 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1007 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1008 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
1009 
1010 	vcn_v3_0_mc_resume_dpg_mode(adev, inst_idx, indirect);
1011 
1012 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1013 		VCN, inst_idx, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
1014 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1015 		VCN, inst_idx, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
1016 
1017 	/* enable LMI MC and UMC channels */
1018 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1019 		VCN, inst_idx, mmUVD_LMI_CTRL2), 0, 0, indirect);
1020 
1021 	/* unblock VCPU register access */
1022 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1023 		VCN, inst_idx, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
1024 
1025 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
1026 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
1027 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1028 		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
1029 
1030 	/* enable master interrupt */
1031 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1032 		VCN, inst_idx, mmUVD_MASTINT_EN),
1033 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
1034 
1035 	/* add nop to workaround PSP size check */
1036 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1037 		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
1038 
1039 	if (indirect)
1040 		psp_update_vcn_sram(adev, inst_idx, adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
1041 			(uint32_t)((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
1042 				(uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr));
1043 
1044 	ring = &adev->vcn.inst[inst_idx].ring_dec;
1045 	/* force RBC into idle state */
1046 	rb_bufsz = order_base_2(ring->ring_size);
1047 	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1048 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1049 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1050 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1051 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1052 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
1053 
1054 	/* Stall DPG before WPTR/RPTR reset */
1055 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1056 		UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1057 		~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1058 	fw_shared->multi_queue.decode_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1059 
1060 	/* set the write pointer delay */
1061 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
1062 
1063 	/* set the wb address */
1064 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR_ADDR,
1065 		(upper_32_bits(ring->gpu_addr) >> 2));
1066 
1067 	/* programm the RB_BASE for ring buffer */
1068 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1069 		lower_32_bits(ring->gpu_addr));
1070 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1071 		upper_32_bits(ring->gpu_addr));
1072 
1073 	/* Initialize the ring buffer's read and write pointers */
1074 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, 0);
1075 
1076 	WREG32_SOC15(VCN, inst_idx, mmUVD_SCRATCH2, 0);
1077 
1078 	ring->wptr = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR);
1079 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR,
1080 		lower_32_bits(ring->wptr));
1081 
1082 	/* Reset FW shared memory RBC WPTR/RPTR */
1083 	fw_shared->rb.rptr = 0;
1084 	fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1085 
1086 	/*resetting done, fw can check RB ring */
1087 	fw_shared->multi_queue.decode_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1088 
1089 	/* Unstall DPG */
1090 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1091 		0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1092 
1093 	return 0;
1094 }
1095 
1096 static int vcn_v3_0_start(struct amdgpu_device *adev)
1097 {
1098 	volatile struct amdgpu_fw_shared *fw_shared;
1099 	struct amdgpu_ring *ring;
1100 	uint32_t rb_bufsz, tmp;
1101 	int i, j, k, r;
1102 
1103 	if (adev->pm.dpm_enabled)
1104 		amdgpu_dpm_enable_uvd(adev, true);
1105 
1106 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1107 		if (adev->vcn.harvest_config & (1 << i))
1108 			continue;
1109 
1110 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG){
1111 			r = vcn_v3_0_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
1112 			continue;
1113 		}
1114 
1115 		/* disable VCN power gating */
1116 		vcn_v3_0_disable_static_power_gating(adev, i);
1117 
1118 		/* set VCN status busy */
1119 		tmp = RREG32_SOC15(VCN, i, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY;
1120 		WREG32_SOC15(VCN, i, mmUVD_STATUS, tmp);
1121 
1122 		/*SW clock gating */
1123 		vcn_v3_0_disable_clock_gating(adev, i);
1124 
1125 		/* enable VCPU clock */
1126 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1127 			UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1128 
1129 		/* disable master interrupt */
1130 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN), 0,
1131 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1132 
1133 		/* enable LMI MC and UMC channels */
1134 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_LMI_CTRL2), 0,
1135 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1136 
1137 		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1138 		tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1139 		tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1140 		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1141 
1142 		/* setup mmUVD_LMI_CTRL */
1143 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL);
1144 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL, tmp |
1145 			UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK	|
1146 			UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1147 			UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1148 			UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1149 
1150 		/* setup mmUVD_MPC_CNTL */
1151 		tmp = RREG32_SOC15(VCN, i, mmUVD_MPC_CNTL);
1152 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1153 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1154 		WREG32_SOC15(VCN, i, mmUVD_MPC_CNTL, tmp);
1155 
1156 		/* setup UVD_MPC_SET_MUXA0 */
1157 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXA0,
1158 			((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1159 			(0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1160 			(0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1161 			(0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1162 
1163 		/* setup UVD_MPC_SET_MUXB0 */
1164 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXB0,
1165 			((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1166 			(0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1167 			(0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1168 			(0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1169 
1170 		/* setup mmUVD_MPC_SET_MUX */
1171 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUX,
1172 			((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1173 			(0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1174 			(0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1175 
1176 		vcn_v3_0_mc_resume(adev, i);
1177 
1178 		/* VCN global tiling registers */
1179 		WREG32_SOC15(VCN, i, mmUVD_GFX10_ADDR_CONFIG,
1180 			adev->gfx.config.gb_addr_config);
1181 
1182 		/* unblock VCPU register access */
1183 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL), 0,
1184 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1185 
1186 		/* release VCPU reset to boot */
1187 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1188 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1189 
1190 		for (j = 0; j < 10; ++j) {
1191 			uint32_t status;
1192 
1193 			for (k = 0; k < 100; ++k) {
1194 				status = RREG32_SOC15(VCN, i, mmUVD_STATUS);
1195 				if (status & 2)
1196 					break;
1197 				mdelay(10);
1198 			}
1199 			r = 0;
1200 			if (status & 2)
1201 				break;
1202 
1203 			DRM_ERROR("VCN[%d] decode not responding, trying to reset the VCPU!!!\n", i);
1204 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1205 				UVD_VCPU_CNTL__BLK_RST_MASK,
1206 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1207 			mdelay(10);
1208 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1209 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1210 
1211 			mdelay(10);
1212 			r = -1;
1213 		}
1214 
1215 		if (r) {
1216 			DRM_ERROR("VCN[%d] decode not responding, giving up!!!\n", i);
1217 			return r;
1218 		}
1219 
1220 		/* enable master interrupt */
1221 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN),
1222 			UVD_MASTINT_EN__VCPU_EN_MASK,
1223 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1224 
1225 		/* clear the busy bit of VCN_STATUS */
1226 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS), 0,
1227 			~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1228 
1229 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_VMID, 0);
1230 
1231 		ring = &adev->vcn.inst[i].ring_dec;
1232 		/* force RBC into idle state */
1233 		rb_bufsz = order_base_2(ring->ring_size);
1234 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1235 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1236 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1237 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1238 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1239 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_CNTL, tmp);
1240 
1241 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1242 		fw_shared->multi_queue.decode_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1243 
1244 		/* programm the RB_BASE for ring buffer */
1245 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1246 			lower_32_bits(ring->gpu_addr));
1247 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1248 			upper_32_bits(ring->gpu_addr));
1249 
1250 		/* Initialize the ring buffer's read and write pointers */
1251 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR, 0);
1252 
1253 		WREG32_SOC15(VCN, i, mmUVD_SCRATCH2, 0);
1254 		ring->wptr = RREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR);
1255 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_WPTR,
1256 			lower_32_bits(ring->wptr));
1257 		fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1258 		fw_shared->multi_queue.decode_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1259 
1260 		if (adev->ip_versions[UVD_HWIP][0] != IP_VERSION(3, 0, 33)) {
1261 			fw_shared->multi_queue.encode_generalpurpose_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1262 			ring = &adev->vcn.inst[i].ring_enc[0];
1263 			WREG32_SOC15(VCN, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1264 			WREG32_SOC15(VCN, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1265 			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO, ring->gpu_addr);
1266 			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1267 			WREG32_SOC15(VCN, i, mmUVD_RB_SIZE, ring->ring_size / 4);
1268 			fw_shared->multi_queue.encode_generalpurpose_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1269 
1270 			fw_shared->multi_queue.encode_lowlatency_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1271 			ring = &adev->vcn.inst[i].ring_enc[1];
1272 			WREG32_SOC15(VCN, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1273 			WREG32_SOC15(VCN, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1274 			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1275 			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1276 			WREG32_SOC15(VCN, i, mmUVD_RB_SIZE2, ring->ring_size / 4);
1277 			fw_shared->multi_queue.encode_lowlatency_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1278 		}
1279 	}
1280 
1281 	return 0;
1282 }
1283 
1284 static int vcn_v3_0_start_sriov(struct amdgpu_device *adev)
1285 {
1286 	int i, j;
1287 	struct amdgpu_ring *ring;
1288 	uint64_t cache_addr;
1289 	uint64_t rb_addr;
1290 	uint64_t ctx_addr;
1291 	uint32_t param, resp, expected;
1292 	uint32_t offset, cache_size;
1293 	uint32_t tmp, timeout;
1294 
1295 	struct amdgpu_mm_table *table = &adev->virt.mm_table;
1296 	uint32_t *table_loc;
1297 	uint32_t table_size;
1298 	uint32_t size, size_dw;
1299 
1300 	struct mmsch_v3_0_cmd_direct_write
1301 		direct_wt = { {0} };
1302 	struct mmsch_v3_0_cmd_direct_read_modify_write
1303 		direct_rd_mod_wt = { {0} };
1304 	struct mmsch_v3_0_cmd_end end = { {0} };
1305 	struct mmsch_v3_0_init_header header;
1306 
1307 	direct_wt.cmd_header.command_type =
1308 		MMSCH_COMMAND__DIRECT_REG_WRITE;
1309 	direct_rd_mod_wt.cmd_header.command_type =
1310 		MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1311 	end.cmd_header.command_type =
1312 		MMSCH_COMMAND__END;
1313 
1314 	header.version = MMSCH_VERSION;
1315 	header.total_size = sizeof(struct mmsch_v3_0_init_header) >> 2;
1316 	for (i = 0; i < MMSCH_V3_0_VCN_INSTANCES; i++) {
1317 		header.inst[i].init_status = 0;
1318 		header.inst[i].table_offset = 0;
1319 		header.inst[i].table_size = 0;
1320 	}
1321 
1322 	table_loc = (uint32_t *)table->cpu_addr;
1323 	table_loc += header.total_size;
1324 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1325 		if (adev->vcn.harvest_config & (1 << i))
1326 			continue;
1327 
1328 		table_size = 0;
1329 
1330 		MMSCH_V3_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(VCN, i,
1331 			mmUVD_STATUS),
1332 			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1333 
1334 		cache_size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1335 
1336 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1337 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1338 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1339 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1340 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1341 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1342 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1343 			offset = 0;
1344 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1345 				mmUVD_VCPU_CACHE_OFFSET0),
1346 				0);
1347 		} else {
1348 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1349 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1350 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1351 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1352 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1353 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1354 			offset = cache_size;
1355 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1356 				mmUVD_VCPU_CACHE_OFFSET0),
1357 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1358 		}
1359 
1360 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1361 			mmUVD_VCPU_CACHE_SIZE0),
1362 			cache_size);
1363 
1364 		cache_addr = adev->vcn.inst[i].gpu_addr + offset;
1365 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1366 			mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1367 			lower_32_bits(cache_addr));
1368 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1369 			mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1370 			upper_32_bits(cache_addr));
1371 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1372 			mmUVD_VCPU_CACHE_OFFSET1),
1373 			0);
1374 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1375 			mmUVD_VCPU_CACHE_SIZE1),
1376 			AMDGPU_VCN_STACK_SIZE);
1377 
1378 		cache_addr = adev->vcn.inst[i].gpu_addr + offset +
1379 			AMDGPU_VCN_STACK_SIZE;
1380 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1381 			mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1382 			lower_32_bits(cache_addr));
1383 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1384 			mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1385 			upper_32_bits(cache_addr));
1386 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1387 			mmUVD_VCPU_CACHE_OFFSET2),
1388 			0);
1389 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1390 			mmUVD_VCPU_CACHE_SIZE2),
1391 			AMDGPU_VCN_CONTEXT_SIZE);
1392 
1393 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
1394 			ring = &adev->vcn.inst[i].ring_enc[j];
1395 			ring->wptr = 0;
1396 			rb_addr = ring->gpu_addr;
1397 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1398 				mmUVD_RB_BASE_LO),
1399 				lower_32_bits(rb_addr));
1400 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1401 				mmUVD_RB_BASE_HI),
1402 				upper_32_bits(rb_addr));
1403 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1404 				mmUVD_RB_SIZE),
1405 				ring->ring_size / 4);
1406 		}
1407 
1408 		ring = &adev->vcn.inst[i].ring_dec;
1409 		ring->wptr = 0;
1410 		rb_addr = ring->gpu_addr;
1411 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1412 			mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1413 			lower_32_bits(rb_addr));
1414 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1415 			mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1416 			upper_32_bits(rb_addr));
1417 		/* force RBC into idle state */
1418 		tmp = order_base_2(ring->ring_size);
1419 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, tmp);
1420 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1421 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1422 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1423 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1424 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1425 			mmUVD_RBC_RB_CNTL),
1426 			tmp);
1427 
1428 		/* add end packet */
1429 		MMSCH_V3_0_INSERT_END();
1430 
1431 		/* refine header */
1432 		header.inst[i].init_status = 0;
1433 		header.inst[i].table_offset = header.total_size;
1434 		header.inst[i].table_size = table_size;
1435 		header.total_size += table_size;
1436 	}
1437 
1438 	/* Update init table header in memory */
1439 	size = sizeof(struct mmsch_v3_0_init_header);
1440 	table_loc = (uint32_t *)table->cpu_addr;
1441 	memcpy((void *)table_loc, &header, size);
1442 
1443 	/* message MMSCH (in VCN[0]) to initialize this client
1444 	 * 1, write to mmsch_vf_ctx_addr_lo/hi register with GPU mc addr
1445 	 * of memory descriptor location
1446 	 */
1447 	ctx_addr = table->gpu_addr;
1448 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(ctx_addr));
1449 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(ctx_addr));
1450 
1451 	/* 2, update vmid of descriptor */
1452 	tmp = RREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID);
1453 	tmp &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1454 	/* use domain0 for MM scheduler */
1455 	tmp |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1456 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID, tmp);
1457 
1458 	/* 3, notify mmsch about the size of this descriptor */
1459 	size = header.total_size;
1460 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_SIZE, size);
1461 
1462 	/* 4, set resp to zero */
1463 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1464 
1465 	/* 5, kick off the initialization and wait until
1466 	 * MMSCH_VF_MAILBOX_RESP becomes non-zero
1467 	 */
1468 	param = 0x10000001;
1469 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_HOST, param);
1470 	tmp = 0;
1471 	timeout = 1000;
1472 	resp = 0;
1473 	expected = param + 1;
1474 	while (resp != expected) {
1475 		resp = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1476 		if (resp == expected)
1477 			break;
1478 
1479 		udelay(10);
1480 		tmp = tmp + 10;
1481 		if (tmp >= timeout) {
1482 			DRM_ERROR("failed to init MMSCH. TIME-OUT after %d usec"\
1483 				" waiting for mmMMSCH_VF_MAILBOX_RESP "\
1484 				"(expected=0x%08x, readback=0x%08x)\n",
1485 				tmp, expected, resp);
1486 			return -EBUSY;
1487 		}
1488 	}
1489 
1490 	return 0;
1491 }
1492 
1493 static int vcn_v3_0_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1494 {
1495 	struct dpg_pause_state state = {.fw_based = VCN_DPG_STATE__UNPAUSE};
1496 	uint32_t tmp;
1497 
1498 	vcn_v3_0_pause_dpg_mode(adev, inst_idx, &state);
1499 
1500 	/* Wait for power status to be 1 */
1501 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1502 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1503 
1504 	/* wait for read ptr to be equal to write ptr */
1505 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR);
1506 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1507 
1508 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2);
1509 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
1510 
1511 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1512 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF);
1513 
1514 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1515 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1516 
1517 	/* disable dynamic power gating mode */
1518 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 0,
1519 		~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1520 
1521 	return 0;
1522 }
1523 
1524 static int vcn_v3_0_stop(struct amdgpu_device *adev)
1525 {
1526 	uint32_t tmp;
1527 	int i, r = 0;
1528 
1529 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1530 		if (adev->vcn.harvest_config & (1 << i))
1531 			continue;
1532 
1533 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1534 			r = vcn_v3_0_stop_dpg_mode(adev, i);
1535 			continue;
1536 		}
1537 
1538 		/* wait for vcn idle */
1539 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1540 		if (r)
1541 			return r;
1542 
1543 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1544 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1545 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1546 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1547 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1548 		if (r)
1549 			return r;
1550 
1551 		/* disable LMI UMC channel */
1552 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2);
1553 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1554 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2, tmp);
1555 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1556 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1557 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1558 		if (r)
1559 			return r;
1560 
1561 		/* block VCPU register access */
1562 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL),
1563 			UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1564 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1565 
1566 		/* reset VCPU */
1567 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1568 			UVD_VCPU_CNTL__BLK_RST_MASK,
1569 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1570 
1571 		/* disable VCPU clock */
1572 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1573 			~(UVD_VCPU_CNTL__CLK_EN_MASK));
1574 
1575 		/* apply soft reset */
1576 		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1577 		tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1578 		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1579 		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1580 		tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1581 		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1582 
1583 		/* clear status */
1584 		WREG32_SOC15(VCN, i, mmUVD_STATUS, 0);
1585 
1586 		/* apply HW clock gating */
1587 		vcn_v3_0_enable_clock_gating(adev, i);
1588 
1589 		/* enable VCN power gating */
1590 		vcn_v3_0_enable_static_power_gating(adev, i);
1591 	}
1592 
1593 	if (adev->pm.dpm_enabled)
1594 		amdgpu_dpm_enable_uvd(adev, false);
1595 
1596 	return 0;
1597 }
1598 
1599 static int vcn_v3_0_pause_dpg_mode(struct amdgpu_device *adev,
1600 		   int inst_idx, struct dpg_pause_state *new_state)
1601 {
1602 	volatile struct amdgpu_fw_shared *fw_shared;
1603 	struct amdgpu_ring *ring;
1604 	uint32_t reg_data = 0;
1605 	int ret_code;
1606 
1607 	/* pause/unpause if state is changed */
1608 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1609 		DRM_DEBUG("dpg pause state changed %d -> %d",
1610 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1611 		reg_data = RREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE) &
1612 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1613 
1614 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1615 			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1616 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1617 
1618 			if (!ret_code) {
1619 				/* pause DPG */
1620 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1621 				WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1622 
1623 				/* wait for ACK */
1624 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_DPG_PAUSE,
1625 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1626 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1627 
1628 				/* Stall DPG before WPTR/RPTR reset */
1629 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1630 					UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1631 					~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1632 
1633 				if (adev->ip_versions[UVD_HWIP][0] != IP_VERSION(3, 0, 33)) {
1634 					/* Restore */
1635 					fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
1636 					fw_shared->multi_queue.encode_generalpurpose_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1637 					ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1638 					ring->wptr = 0;
1639 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
1640 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1641 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
1642 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1643 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1644 					fw_shared->multi_queue.encode_generalpurpose_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1645 
1646 					fw_shared->multi_queue.encode_lowlatency_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1647 					ring = &adev->vcn.inst[inst_idx].ring_enc[1];
1648 					ring->wptr = 0;
1649 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1650 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1651 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
1652 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1653 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1654 					fw_shared->multi_queue.encode_lowlatency_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1655 
1656 					/* restore wptr/rptr with pointers saved in FW shared memory*/
1657 					WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, fw_shared->rb.rptr);
1658 					WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR, fw_shared->rb.wptr);
1659 				}
1660 
1661 				/* Unstall DPG */
1662 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1663 					0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1664 
1665 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS,
1666 					UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1667 			}
1668 		} else {
1669 			/* unpause dpg, no need to wait */
1670 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1671 			WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1672 		}
1673 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1674 	}
1675 
1676 	return 0;
1677 }
1678 
1679 /**
1680  * vcn_v3_0_dec_ring_get_rptr - get read pointer
1681  *
1682  * @ring: amdgpu_ring pointer
1683  *
1684  * Returns the current hardware read pointer
1685  */
1686 static uint64_t vcn_v3_0_dec_ring_get_rptr(struct amdgpu_ring *ring)
1687 {
1688 	struct amdgpu_device *adev = ring->adev;
1689 
1690 	return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_RPTR);
1691 }
1692 
1693 /**
1694  * vcn_v3_0_dec_ring_get_wptr - get write pointer
1695  *
1696  * @ring: amdgpu_ring pointer
1697  *
1698  * Returns the current hardware write pointer
1699  */
1700 static uint64_t vcn_v3_0_dec_ring_get_wptr(struct amdgpu_ring *ring)
1701 {
1702 	struct amdgpu_device *adev = ring->adev;
1703 
1704 	if (ring->use_doorbell)
1705 		return *ring->wptr_cpu_addr;
1706 	else
1707 		return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR);
1708 }
1709 
1710 /**
1711  * vcn_v3_0_dec_ring_set_wptr - set write pointer
1712  *
1713  * @ring: amdgpu_ring pointer
1714  *
1715  * Commits the write pointer to the hardware
1716  */
1717 static void vcn_v3_0_dec_ring_set_wptr(struct amdgpu_ring *ring)
1718 {
1719 	struct amdgpu_device *adev = ring->adev;
1720 	volatile struct amdgpu_fw_shared *fw_shared;
1721 
1722 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1723 		/*whenever update RBC_RB_WPTR, we save the wptr in shared rb.wptr and scratch2 */
1724 		fw_shared = adev->vcn.inst[ring->me].fw_shared.cpu_addr;
1725 		fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1726 		WREG32_SOC15(VCN, ring->me, mmUVD_SCRATCH2,
1727 			lower_32_bits(ring->wptr));
1728 	}
1729 
1730 	if (ring->use_doorbell) {
1731 		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1732 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1733 	} else {
1734 		WREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1735 	}
1736 }
1737 
1738 static const struct amdgpu_ring_funcs vcn_v3_0_dec_sw_ring_vm_funcs = {
1739 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1740 	.align_mask = 0x3f,
1741 	.nop = VCN_DEC_SW_CMD_NO_OP,
1742 	.secure_submission_supported = true,
1743 	.get_rptr = vcn_v3_0_dec_ring_get_rptr,
1744 	.get_wptr = vcn_v3_0_dec_ring_get_wptr,
1745 	.set_wptr = vcn_v3_0_dec_ring_set_wptr,
1746 	.emit_frame_size =
1747 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1748 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1749 		VCN_SW_RING_EMIT_FRAME_SIZE,
1750 	.emit_ib_size = 5, /* vcn_dec_sw_ring_emit_ib */
1751 	.emit_ib = vcn_dec_sw_ring_emit_ib,
1752 	.emit_fence = vcn_dec_sw_ring_emit_fence,
1753 	.emit_vm_flush = vcn_dec_sw_ring_emit_vm_flush,
1754 	.test_ring = amdgpu_vcn_dec_sw_ring_test_ring,
1755 	.test_ib = NULL,//amdgpu_vcn_dec_sw_ring_test_ib,
1756 	.insert_nop = amdgpu_ring_insert_nop,
1757 	.insert_end = vcn_dec_sw_ring_insert_end,
1758 	.pad_ib = amdgpu_ring_generic_pad_ib,
1759 	.begin_use = amdgpu_vcn_ring_begin_use,
1760 	.end_use = amdgpu_vcn_ring_end_use,
1761 	.emit_wreg = vcn_dec_sw_ring_emit_wreg,
1762 	.emit_reg_wait = vcn_dec_sw_ring_emit_reg_wait,
1763 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1764 };
1765 
1766 static int vcn_v3_0_limit_sched(struct amdgpu_cs_parser *p,
1767 				struct amdgpu_job *job)
1768 {
1769 	struct drm_gpu_scheduler **scheds;
1770 
1771 	/* The create msg must be in the first IB submitted */
1772 	if (atomic_read(&job->base.entity->fence_seq))
1773 		return -EINVAL;
1774 
1775 	/* if VCN0 is harvested, we can't support AV1 */
1776 	if (p->adev->vcn.harvest_config & AMDGPU_VCN_HARVEST_VCN0)
1777 		return -EINVAL;
1778 
1779 	scheds = p->adev->gpu_sched[AMDGPU_HW_IP_VCN_DEC]
1780 		[AMDGPU_RING_PRIO_DEFAULT].sched;
1781 	drm_sched_entity_modify_sched(job->base.entity, scheds, 1);
1782 	return 0;
1783 }
1784 
1785 static int vcn_v3_0_dec_msg(struct amdgpu_cs_parser *p, struct amdgpu_job *job,
1786 			    uint64_t addr)
1787 {
1788 	struct ttm_operation_ctx ctx = { false, false };
1789 	struct amdgpu_bo_va_mapping *map;
1790 	uint32_t *msg, num_buffers;
1791 	struct amdgpu_bo *bo;
1792 	uint64_t start, end;
1793 	unsigned int i;
1794 	void * ptr;
1795 	int r;
1796 
1797 	addr &= AMDGPU_GMC_HOLE_MASK;
1798 	r = amdgpu_cs_find_mapping(p, addr, &bo, &map);
1799 	if (r) {
1800 		DRM_ERROR("Can't find BO for addr 0x%08Lx\n", addr);
1801 		return r;
1802 	}
1803 
1804 	start = map->start * AMDGPU_GPU_PAGE_SIZE;
1805 	end = (map->last + 1) * AMDGPU_GPU_PAGE_SIZE;
1806 	if (addr & 0x7) {
1807 		DRM_ERROR("VCN messages must be 8 byte aligned!\n");
1808 		return -EINVAL;
1809 	}
1810 
1811 	bo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
1812 	amdgpu_bo_placement_from_domain(bo, bo->allowed_domains);
1813 	r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
1814 	if (r) {
1815 		DRM_ERROR("Failed validating the VCN message BO (%d)!\n", r);
1816 		return r;
1817 	}
1818 
1819 	r = amdgpu_bo_kmap(bo, &ptr);
1820 	if (r) {
1821 		DRM_ERROR("Failed mapping the VCN message (%d)!\n", r);
1822 		return r;
1823 	}
1824 
1825 	msg = ptr + addr - start;
1826 
1827 	/* Check length */
1828 	if (msg[1] > end - addr) {
1829 		r = -EINVAL;
1830 		goto out;
1831 	}
1832 
1833 	if (msg[3] != RDECODE_MSG_CREATE)
1834 		goto out;
1835 
1836 	num_buffers = msg[2];
1837 	for (i = 0, msg = &msg[6]; i < num_buffers; ++i, msg += 4) {
1838 		uint32_t offset, size, *create;
1839 
1840 		if (msg[0] != RDECODE_MESSAGE_CREATE)
1841 			continue;
1842 
1843 		offset = msg[1];
1844 		size = msg[2];
1845 
1846 		if (offset + size > end) {
1847 			r = -EINVAL;
1848 			goto out;
1849 		}
1850 
1851 		create = ptr + addr + offset - start;
1852 
1853 		/* H246, HEVC and VP9 can run on any instance */
1854 		if (create[0] == 0x7 || create[0] == 0x10 || create[0] == 0x11)
1855 			continue;
1856 
1857 		r = vcn_v3_0_limit_sched(p, job);
1858 		if (r)
1859 			goto out;
1860 	}
1861 
1862 out:
1863 	amdgpu_bo_kunmap(bo);
1864 	return r;
1865 }
1866 
1867 static int vcn_v3_0_ring_patch_cs_in_place(struct amdgpu_cs_parser *p,
1868 					   struct amdgpu_job *job,
1869 					   struct amdgpu_ib *ib)
1870 {
1871 	struct amdgpu_ring *ring = amdgpu_job_ring(job);
1872 	uint32_t msg_lo = 0, msg_hi = 0;
1873 	unsigned i;
1874 	int r;
1875 
1876 	/* The first instance can decode anything */
1877 	if (!ring->me)
1878 		return 0;
1879 
1880 	for (i = 0; i < ib->length_dw; i += 2) {
1881 		uint32_t reg = amdgpu_ib_get_value(ib, i);
1882 		uint32_t val = amdgpu_ib_get_value(ib, i + 1);
1883 
1884 		if (reg == PACKET0(p->adev->vcn.internal.data0, 0)) {
1885 			msg_lo = val;
1886 		} else if (reg == PACKET0(p->adev->vcn.internal.data1, 0)) {
1887 			msg_hi = val;
1888 		} else if (reg == PACKET0(p->adev->vcn.internal.cmd, 0) &&
1889 			   val == 0) {
1890 			r = vcn_v3_0_dec_msg(p, job,
1891 					     ((u64)msg_hi) << 32 | msg_lo);
1892 			if (r)
1893 				return r;
1894 		}
1895 	}
1896 	return 0;
1897 }
1898 
1899 static const struct amdgpu_ring_funcs vcn_v3_0_dec_ring_vm_funcs = {
1900 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1901 	.align_mask = 0xf,
1902 	.secure_submission_supported = true,
1903 	.get_rptr = vcn_v3_0_dec_ring_get_rptr,
1904 	.get_wptr = vcn_v3_0_dec_ring_get_wptr,
1905 	.set_wptr = vcn_v3_0_dec_ring_set_wptr,
1906 	.patch_cs_in_place = vcn_v3_0_ring_patch_cs_in_place,
1907 	.emit_frame_size =
1908 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1909 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1910 		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1911 		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1912 		6,
1913 	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1914 	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
1915 	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
1916 	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1917 	.test_ring = vcn_v2_0_dec_ring_test_ring,
1918 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
1919 	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
1920 	.insert_start = vcn_v2_0_dec_ring_insert_start,
1921 	.insert_end = vcn_v2_0_dec_ring_insert_end,
1922 	.pad_ib = amdgpu_ring_generic_pad_ib,
1923 	.begin_use = amdgpu_vcn_ring_begin_use,
1924 	.end_use = amdgpu_vcn_ring_end_use,
1925 	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1926 	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1927 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1928 };
1929 
1930 /**
1931  * vcn_v3_0_enc_ring_get_rptr - get enc read pointer
1932  *
1933  * @ring: amdgpu_ring pointer
1934  *
1935  * Returns the current hardware enc read pointer
1936  */
1937 static uint64_t vcn_v3_0_enc_ring_get_rptr(struct amdgpu_ring *ring)
1938 {
1939 	struct amdgpu_device *adev = ring->adev;
1940 
1941 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0])
1942 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR);
1943 	else
1944 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR2);
1945 }
1946 
1947 /**
1948  * vcn_v3_0_enc_ring_get_wptr - get enc write pointer
1949  *
1950  * @ring: amdgpu_ring pointer
1951  *
1952  * Returns the current hardware enc write pointer
1953  */
1954 static uint64_t vcn_v3_0_enc_ring_get_wptr(struct amdgpu_ring *ring)
1955 {
1956 	struct amdgpu_device *adev = ring->adev;
1957 
1958 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1959 		if (ring->use_doorbell)
1960 			return *ring->wptr_cpu_addr;
1961 		else
1962 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR);
1963 	} else {
1964 		if (ring->use_doorbell)
1965 			return *ring->wptr_cpu_addr;
1966 		else
1967 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2);
1968 	}
1969 }
1970 
1971 /**
1972  * vcn_v3_0_enc_ring_set_wptr - set enc write pointer
1973  *
1974  * @ring: amdgpu_ring pointer
1975  *
1976  * Commits the enc write pointer to the hardware
1977  */
1978 static void vcn_v3_0_enc_ring_set_wptr(struct amdgpu_ring *ring)
1979 {
1980 	struct amdgpu_device *adev = ring->adev;
1981 
1982 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1983 		if (ring->use_doorbell) {
1984 			*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1985 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1986 		} else {
1987 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1988 		}
1989 	} else {
1990 		if (ring->use_doorbell) {
1991 			*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1992 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1993 		} else {
1994 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1995 		}
1996 	}
1997 }
1998 
1999 static const struct amdgpu_ring_funcs vcn_v3_0_enc_ring_vm_funcs = {
2000 	.type = AMDGPU_RING_TYPE_VCN_ENC,
2001 	.align_mask = 0x3f,
2002 	.nop = VCN_ENC_CMD_NO_OP,
2003 	.get_rptr = vcn_v3_0_enc_ring_get_rptr,
2004 	.get_wptr = vcn_v3_0_enc_ring_get_wptr,
2005 	.set_wptr = vcn_v3_0_enc_ring_set_wptr,
2006 	.emit_frame_size =
2007 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2008 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
2009 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
2010 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
2011 		1, /* vcn_v2_0_enc_ring_insert_end */
2012 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
2013 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
2014 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
2015 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
2016 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
2017 	.test_ib = amdgpu_vcn_enc_ring_test_ib,
2018 	.insert_nop = amdgpu_ring_insert_nop,
2019 	.insert_end = vcn_v2_0_enc_ring_insert_end,
2020 	.pad_ib = amdgpu_ring_generic_pad_ib,
2021 	.begin_use = amdgpu_vcn_ring_begin_use,
2022 	.end_use = amdgpu_vcn_ring_end_use,
2023 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
2024 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
2025 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2026 };
2027 
2028 static void vcn_v3_0_set_dec_ring_funcs(struct amdgpu_device *adev)
2029 {
2030 	int i;
2031 
2032 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2033 		if (adev->vcn.harvest_config & (1 << i))
2034 			continue;
2035 
2036 		if (!DEC_SW_RING_ENABLED)
2037 			adev->vcn.inst[i].ring_dec.funcs = &vcn_v3_0_dec_ring_vm_funcs;
2038 		else
2039 			adev->vcn.inst[i].ring_dec.funcs = &vcn_v3_0_dec_sw_ring_vm_funcs;
2040 		adev->vcn.inst[i].ring_dec.me = i;
2041 		DRM_INFO("VCN(%d) decode%s is enabled in VM mode\n", i,
2042 			  DEC_SW_RING_ENABLED?"(Software Ring)":"");
2043 	}
2044 }
2045 
2046 static void vcn_v3_0_set_enc_ring_funcs(struct amdgpu_device *adev)
2047 {
2048 	int i, j;
2049 
2050 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2051 		if (adev->vcn.harvest_config & (1 << i))
2052 			continue;
2053 
2054 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
2055 			adev->vcn.inst[i].ring_enc[j].funcs = &vcn_v3_0_enc_ring_vm_funcs;
2056 			adev->vcn.inst[i].ring_enc[j].me = i;
2057 		}
2058 		if (adev->vcn.num_enc_rings > 0)
2059 			DRM_INFO("VCN(%d) encode is enabled in VM mode\n", i);
2060 	}
2061 }
2062 
2063 static bool vcn_v3_0_is_idle(void *handle)
2064 {
2065 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2066 	int i, ret = 1;
2067 
2068 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2069 		if (adev->vcn.harvest_config & (1 << i))
2070 			continue;
2071 
2072 		ret &= (RREG32_SOC15(VCN, i, mmUVD_STATUS) == UVD_STATUS__IDLE);
2073 	}
2074 
2075 	return ret;
2076 }
2077 
2078 static int vcn_v3_0_wait_for_idle(void *handle)
2079 {
2080 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2081 	int i, ret = 0;
2082 
2083 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2084 		if (adev->vcn.harvest_config & (1 << i))
2085 			continue;
2086 
2087 		ret = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE,
2088 			UVD_STATUS__IDLE);
2089 		if (ret)
2090 			return ret;
2091 	}
2092 
2093 	return ret;
2094 }
2095 
2096 static int vcn_v3_0_set_clockgating_state(void *handle,
2097 					  enum amd_clockgating_state state)
2098 {
2099 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2100 	bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
2101 	int i;
2102 
2103 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2104 		if (adev->vcn.harvest_config & (1 << i))
2105 			continue;
2106 
2107 		if (enable) {
2108 			if (RREG32_SOC15(VCN, i, mmUVD_STATUS) != UVD_STATUS__IDLE)
2109 				return -EBUSY;
2110 			vcn_v3_0_enable_clock_gating(adev, i);
2111 		} else {
2112 			vcn_v3_0_disable_clock_gating(adev, i);
2113 		}
2114 	}
2115 
2116 	return 0;
2117 }
2118 
2119 static int vcn_v3_0_set_powergating_state(void *handle,
2120 					  enum amd_powergating_state state)
2121 {
2122 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2123 	int ret;
2124 
2125 	/* for SRIOV, guest should not control VCN Power-gating
2126 	 * MMSCH FW should control Power-gating and clock-gating
2127 	 * guest should avoid touching CGC and PG
2128 	 */
2129 	if (amdgpu_sriov_vf(adev)) {
2130 		adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
2131 		return 0;
2132 	}
2133 
2134 	if(state == adev->vcn.cur_state)
2135 		return 0;
2136 
2137 	if (state == AMD_PG_STATE_GATE)
2138 		ret = vcn_v3_0_stop(adev);
2139 	else
2140 		ret = vcn_v3_0_start(adev);
2141 
2142 	if(!ret)
2143 		adev->vcn.cur_state = state;
2144 
2145 	return ret;
2146 }
2147 
2148 static int vcn_v3_0_set_interrupt_state(struct amdgpu_device *adev,
2149 					struct amdgpu_irq_src *source,
2150 					unsigned type,
2151 					enum amdgpu_interrupt_state state)
2152 {
2153 	return 0;
2154 }
2155 
2156 static int vcn_v3_0_process_interrupt(struct amdgpu_device *adev,
2157 				      struct amdgpu_irq_src *source,
2158 				      struct amdgpu_iv_entry *entry)
2159 {
2160 	uint32_t ip_instance;
2161 
2162 	switch (entry->client_id) {
2163 	case SOC15_IH_CLIENTID_VCN:
2164 		ip_instance = 0;
2165 		break;
2166 	case SOC15_IH_CLIENTID_VCN1:
2167 		ip_instance = 1;
2168 		break;
2169 	default:
2170 		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
2171 		return 0;
2172 	}
2173 
2174 	DRM_DEBUG("IH: VCN TRAP\n");
2175 
2176 	switch (entry->src_id) {
2177 	case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
2178 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec);
2179 		break;
2180 	case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
2181 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
2182 		break;
2183 	case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
2184 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]);
2185 		break;
2186 	default:
2187 		DRM_ERROR("Unhandled interrupt: %d %d\n",
2188 			  entry->src_id, entry->src_data[0]);
2189 		break;
2190 	}
2191 
2192 	return 0;
2193 }
2194 
2195 static const struct amdgpu_irq_src_funcs vcn_v3_0_irq_funcs = {
2196 	.set = vcn_v3_0_set_interrupt_state,
2197 	.process = vcn_v3_0_process_interrupt,
2198 };
2199 
2200 static void vcn_v3_0_set_irq_funcs(struct amdgpu_device *adev)
2201 {
2202 	int i;
2203 
2204 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2205 		if (adev->vcn.harvest_config & (1 << i))
2206 			continue;
2207 
2208 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
2209 		adev->vcn.inst[i].irq.funcs = &vcn_v3_0_irq_funcs;
2210 	}
2211 }
2212 
2213 static const struct amd_ip_funcs vcn_v3_0_ip_funcs = {
2214 	.name = "vcn_v3_0",
2215 	.early_init = vcn_v3_0_early_init,
2216 	.late_init = NULL,
2217 	.sw_init = vcn_v3_0_sw_init,
2218 	.sw_fini = vcn_v3_0_sw_fini,
2219 	.hw_init = vcn_v3_0_hw_init,
2220 	.hw_fini = vcn_v3_0_hw_fini,
2221 	.suspend = vcn_v3_0_suspend,
2222 	.resume = vcn_v3_0_resume,
2223 	.is_idle = vcn_v3_0_is_idle,
2224 	.wait_for_idle = vcn_v3_0_wait_for_idle,
2225 	.check_soft_reset = NULL,
2226 	.pre_soft_reset = NULL,
2227 	.soft_reset = NULL,
2228 	.post_soft_reset = NULL,
2229 	.set_clockgating_state = vcn_v3_0_set_clockgating_state,
2230 	.set_powergating_state = vcn_v3_0_set_powergating_state,
2231 };
2232 
2233 const struct amdgpu_ip_block_version vcn_v3_0_ip_block =
2234 {
2235 	.type = AMD_IP_BLOCK_TYPE_VCN,
2236 	.major = 3,
2237 	.minor = 0,
2238 	.rev = 0,
2239 	.funcs = &vcn_v3_0_ip_funcs,
2240 };
2241