xref: /linux/drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c (revision 30dc5873967ecc0282a8283622156cba4be8daf4)
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 (amdgpu_ip_version(adev, UVD_HWIP, 0) ==
104 		    IP_VERSION(3, 0, 33))
105 			adev->vcn.num_enc_rings = 0;
106 		else
107 			adev->vcn.num_enc_rings = 2;
108 	}
109 
110 	vcn_v3_0_set_dec_ring_funcs(adev);
111 	vcn_v3_0_set_enc_ring_funcs(adev);
112 	vcn_v3_0_set_irq_funcs(adev);
113 
114 	return amdgpu_vcn_early_init(adev);
115 }
116 
117 /**
118  * vcn_v3_0_sw_init - sw init for VCN block
119  *
120  * @handle: amdgpu_device pointer
121  *
122  * Load firmware and sw initialization
123  */
124 static int vcn_v3_0_sw_init(void *handle)
125 {
126 	struct amdgpu_ring *ring;
127 	int i, j, r;
128 	int vcn_doorbell_index = 0;
129 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
130 
131 	r = amdgpu_vcn_sw_init(adev);
132 	if (r)
133 		return r;
134 
135 	amdgpu_vcn_setup_ucode(adev);
136 
137 	r = amdgpu_vcn_resume(adev);
138 	if (r)
139 		return r;
140 
141 	/*
142 	 * Note: doorbell assignment is fixed for SRIOV multiple VCN engines
143 	 * Formula:
144 	 *   vcn_db_base  = adev->doorbell_index.vcn.vcn_ring0_1 << 1;
145 	 *   dec_ring_i   = vcn_db_base + i * (adev->vcn.num_enc_rings + 1)
146 	 *   enc_ring_i,j = vcn_db_base + i * (adev->vcn.num_enc_rings + 1) + 1 + j
147 	 */
148 	if (amdgpu_sriov_vf(adev)) {
149 		vcn_doorbell_index = adev->doorbell_index.vcn.vcn_ring0_1;
150 		/* get DWORD offset */
151 		vcn_doorbell_index = vcn_doorbell_index << 1;
152 	}
153 
154 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
155 		volatile struct amdgpu_fw_shared *fw_shared;
156 
157 		if (adev->vcn.harvest_config & (1 << i))
158 			continue;
159 
160 		adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
161 		adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
162 		adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
163 		adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
164 		adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
165 		adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
166 
167 		adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
168 		adev->vcn.inst[i].external.scratch9 = SOC15_REG_OFFSET(VCN, i, mmUVD_SCRATCH9);
169 		adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
170 		adev->vcn.inst[i].external.data0 = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_DATA0);
171 		adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
172 		adev->vcn.inst[i].external.data1 = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_DATA1);
173 		adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
174 		adev->vcn.inst[i].external.cmd = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_CMD);
175 		adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
176 		adev->vcn.inst[i].external.nop = SOC15_REG_OFFSET(VCN, i, mmUVD_NO_OP);
177 
178 		/* VCN DEC TRAP */
179 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
180 				VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst[i].irq);
181 		if (r)
182 			return r;
183 
184 		atomic_set(&adev->vcn.inst[i].sched_score, 0);
185 
186 		ring = &adev->vcn.inst[i].ring_dec;
187 		ring->use_doorbell = true;
188 		if (amdgpu_sriov_vf(adev)) {
189 			ring->doorbell_index = vcn_doorbell_index + i * (adev->vcn.num_enc_rings + 1);
190 		} else {
191 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i;
192 		}
193 		ring->vm_hub = AMDGPU_MMHUB0(0);
194 		sprintf(ring->name, "vcn_dec_%d", i);
195 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
196 				     AMDGPU_RING_PRIO_DEFAULT,
197 				     &adev->vcn.inst[i].sched_score);
198 		if (r)
199 			return r;
200 
201 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
202 			enum amdgpu_ring_priority_level hw_prio = amdgpu_vcn_get_enc_ring_prio(j);
203 
204 			/* VCN ENC TRAP */
205 			r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
206 				j + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);
207 			if (r)
208 				return r;
209 
210 			ring = &adev->vcn.inst[i].ring_enc[j];
211 			ring->use_doorbell = true;
212 			if (amdgpu_sriov_vf(adev)) {
213 				ring->doorbell_index = vcn_doorbell_index + i * (adev->vcn.num_enc_rings + 1) + 1 + j;
214 			} else {
215 				ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + j + 8 * i;
216 			}
217 			ring->vm_hub = AMDGPU_MMHUB0(0);
218 			sprintf(ring->name, "vcn_enc_%d.%d", i, j);
219 			r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
220 					     hw_prio, &adev->vcn.inst[i].sched_score);
221 			if (r)
222 				return r;
223 		}
224 
225 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
226 		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_SW_RING_FLAG) |
227 					     cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG) |
228 					     cpu_to_le32(AMDGPU_VCN_FW_SHARED_FLAG_0_RB);
229 		fw_shared->sw_ring.is_enabled = cpu_to_le32(DEC_SW_RING_ENABLED);
230 		fw_shared->present_flag_0 |= AMDGPU_VCN_SMU_VERSION_INFO_FLAG;
231 		if (amdgpu_ip_version(adev, UVD_HWIP, 0) == IP_VERSION(3, 1, 2))
232 			fw_shared->smu_interface_info.smu_interface_type = 2;
233 		else if (amdgpu_ip_version(adev, UVD_HWIP, 0) ==
234 			 IP_VERSION(3, 1, 1))
235 			fw_shared->smu_interface_info.smu_interface_type = 1;
236 
237 		if (amdgpu_vcnfw_log)
238 			amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
239 	}
240 
241 	if (amdgpu_sriov_vf(adev)) {
242 		r = amdgpu_virt_alloc_mm_table(adev);
243 		if (r)
244 			return r;
245 	}
246 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
247 		adev->vcn.pause_dpg_mode = vcn_v3_0_pause_dpg_mode;
248 
249 	return 0;
250 }
251 
252 /**
253  * vcn_v3_0_sw_fini - sw fini for VCN block
254  *
255  * @handle: amdgpu_device pointer
256  *
257  * VCN suspend and free up sw allocation
258  */
259 static int vcn_v3_0_sw_fini(void *handle)
260 {
261 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
262 	int i, r, idx;
263 
264 	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
265 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
266 			volatile struct amdgpu_fw_shared *fw_shared;
267 
268 			if (adev->vcn.harvest_config & (1 << i))
269 				continue;
270 			fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
271 			fw_shared->present_flag_0 = 0;
272 			fw_shared->sw_ring.is_enabled = false;
273 		}
274 
275 		drm_dev_exit(idx);
276 	}
277 
278 	if (amdgpu_sriov_vf(adev))
279 		amdgpu_virt_free_mm_table(adev);
280 
281 	r = amdgpu_vcn_suspend(adev);
282 	if (r)
283 		return r;
284 
285 	r = amdgpu_vcn_sw_fini(adev);
286 
287 	return r;
288 }
289 
290 /**
291  * vcn_v3_0_hw_init - start and test VCN block
292  *
293  * @handle: amdgpu_device pointer
294  *
295  * Initialize the hardware, boot up the VCPU and do some testing
296  */
297 static int vcn_v3_0_hw_init(void *handle)
298 {
299 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
300 	struct amdgpu_ring *ring;
301 	int i, j, r;
302 
303 	if (amdgpu_sriov_vf(adev)) {
304 		r = vcn_v3_0_start_sriov(adev);
305 		if (r)
306 			goto done;
307 
308 		/* initialize VCN dec and enc ring buffers */
309 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
310 			if (adev->vcn.harvest_config & (1 << i))
311 				continue;
312 
313 			ring = &adev->vcn.inst[i].ring_dec;
314 			if (amdgpu_vcn_is_disabled_vcn(adev, VCN_DECODE_RING, i)) {
315 				ring->sched.ready = false;
316 				ring->no_scheduler = true;
317 				dev_info(adev->dev, "ring %s is disabled by hypervisor\n", ring->name);
318 			} else {
319 				ring->wptr = 0;
320 				ring->wptr_old = 0;
321 				vcn_v3_0_dec_ring_set_wptr(ring);
322 				ring->sched.ready = true;
323 			}
324 
325 			for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
326 				ring = &adev->vcn.inst[i].ring_enc[j];
327 				if (amdgpu_vcn_is_disabled_vcn(adev, VCN_ENCODE_RING, i)) {
328 					ring->sched.ready = false;
329 					ring->no_scheduler = true;
330 					dev_info(adev->dev, "ring %s is disabled by hypervisor\n", ring->name);
331 				} else {
332 					ring->wptr = 0;
333 					ring->wptr_old = 0;
334 					vcn_v3_0_enc_ring_set_wptr(ring);
335 					ring->sched.ready = true;
336 				}
337 			}
338 		}
339 	} else {
340 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
341 			if (adev->vcn.harvest_config & (1 << i))
342 				continue;
343 
344 			ring = &adev->vcn.inst[i].ring_dec;
345 
346 			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
347 						     ring->doorbell_index, i);
348 
349 			r = amdgpu_ring_test_helper(ring);
350 			if (r)
351 				goto done;
352 
353 			for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
354 				ring = &adev->vcn.inst[i].ring_enc[j];
355 				r = amdgpu_ring_test_helper(ring);
356 				if (r)
357 					goto done;
358 			}
359 		}
360 	}
361 
362 done:
363 	if (!r)
364 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
365 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
366 
367 	return r;
368 }
369 
370 /**
371  * vcn_v3_0_hw_fini - stop the hardware block
372  *
373  * @handle: amdgpu_device pointer
374  *
375  * Stop the VCN block, mark ring as not ready any more
376  */
377 static int vcn_v3_0_hw_fini(void *handle)
378 {
379 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
380 	int i;
381 
382 	cancel_delayed_work_sync(&adev->vcn.idle_work);
383 
384 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
385 		if (adev->vcn.harvest_config & (1 << i))
386 			continue;
387 
388 		if (!amdgpu_sriov_vf(adev)) {
389 			if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
390 					(adev->vcn.cur_state != AMD_PG_STATE_GATE &&
391 					 RREG32_SOC15(VCN, i, mmUVD_STATUS))) {
392 				vcn_v3_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
393 			}
394 		}
395 	}
396 
397 	return 0;
398 }
399 
400 /**
401  * vcn_v3_0_suspend - suspend VCN block
402  *
403  * @handle: amdgpu_device pointer
404  *
405  * HW fini and suspend VCN block
406  */
407 static int vcn_v3_0_suspend(void *handle)
408 {
409 	int r;
410 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
411 
412 	r = vcn_v3_0_hw_fini(adev);
413 	if (r)
414 		return r;
415 
416 	r = amdgpu_vcn_suspend(adev);
417 
418 	return r;
419 }
420 
421 /**
422  * vcn_v3_0_resume - resume VCN block
423  *
424  * @handle: amdgpu_device pointer
425  *
426  * Resume firmware and hw init VCN block
427  */
428 static int vcn_v3_0_resume(void *handle)
429 {
430 	int r;
431 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
432 
433 	r = amdgpu_vcn_resume(adev);
434 	if (r)
435 		return r;
436 
437 	r = vcn_v3_0_hw_init(adev);
438 
439 	return r;
440 }
441 
442 /**
443  * vcn_v3_0_mc_resume - memory controller programming
444  *
445  * @adev: amdgpu_device pointer
446  * @inst: instance number
447  *
448  * Let the VCN memory controller know it's offsets
449  */
450 static void vcn_v3_0_mc_resume(struct amdgpu_device *adev, int inst)
451 {
452 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw[inst]->size + 4);
453 	uint32_t offset;
454 
455 	/* cache window 0: fw */
456 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
457 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
458 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo));
459 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
460 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi));
461 		WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET0, 0);
462 		offset = 0;
463 	} else {
464 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
465 			lower_32_bits(adev->vcn.inst[inst].gpu_addr));
466 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
467 			upper_32_bits(adev->vcn.inst[inst].gpu_addr));
468 		offset = size;
469 		WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET0,
470 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
471 	}
472 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE0, size);
473 
474 	/* cache window 1: stack */
475 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
476 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
477 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
478 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
479 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET1, 0);
480 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
481 
482 	/* cache window 2: context */
483 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
484 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
485 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
486 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
487 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET2, 0);
488 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
489 
490 	/* non-cache window */
491 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
492 		lower_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
493 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
494 		upper_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
495 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
496 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_NONCACHE_SIZE0,
497 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
498 }
499 
500 static void vcn_v3_0_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
501 {
502 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw[inst_idx]->size + 4);
503 	uint32_t offset;
504 
505 	/* cache window 0: fw */
506 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
507 		if (!indirect) {
508 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
509 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
510 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
511 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
512 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
513 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
514 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
515 				VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
516 		} else {
517 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
518 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
519 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
520 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
521 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
522 				VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
523 		}
524 		offset = 0;
525 	} else {
526 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
527 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
528 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
529 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
530 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
531 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
532 		offset = size;
533 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
534 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0),
535 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
536 	}
537 
538 	if (!indirect)
539 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
540 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
541 	else
542 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
543 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
544 
545 	/* cache window 1: stack */
546 	if (!indirect) {
547 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
548 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
549 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
550 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
551 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
552 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
553 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
554 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
555 	} else {
556 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
557 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
558 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
559 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
560 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
561 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
562 	}
563 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
564 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
565 
566 	/* cache window 2: context */
567 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
568 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
569 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
570 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
571 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
572 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
573 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
574 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
575 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
576 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
577 
578 	/* non-cache window */
579 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
580 			VCN, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
581 			lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
582 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
583 			VCN, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
584 			upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
585 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
586 			VCN, inst_idx, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
587 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
588 			VCN, inst_idx, mmUVD_VCPU_NONCACHE_SIZE0),
589 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
590 
591 	/* VCN global tiling registers */
592 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
593 		UVD, inst_idx, mmUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
594 }
595 
596 static void vcn_v3_0_disable_static_power_gating(struct amdgpu_device *adev, int inst)
597 {
598 	uint32_t data = 0;
599 
600 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
601 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
602 			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
603 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
604 			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
605 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
606 			| 2 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
607 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
608 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
609 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
610 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
611 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
612 			| 2 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
613 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
614 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
615 
616 		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
617 		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS,
618 			UVD_PGFSM_STATUS__UVDM_UVDU_UVDLM_PWR_ON_3_0, 0x3F3FFFFF);
619 	} else {
620 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
621 			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
622 			| 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
623 			| 1 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
624 			| 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
625 			| 1 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
626 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
627 			| 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
628 			| 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
629 			| 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
630 			| 1 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
631 			| 1 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
632 			| 1 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
633 			| 1 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
634 		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
635 		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS, 0,  0x3F3FFFFF);
636 	}
637 
638 	data = RREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS);
639 	data &= ~0x103;
640 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
641 		data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
642 			UVD_POWER_STATUS__UVD_PG_EN_MASK;
643 
644 	WREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS, data);
645 }
646 
647 static void vcn_v3_0_enable_static_power_gating(struct amdgpu_device *adev, int inst)
648 {
649 	uint32_t data;
650 
651 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
652 		/* Before power off, this indicator has to be turned on */
653 		data = RREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS);
654 		data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
655 		data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
656 		WREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS, data);
657 
658 		data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
659 			| 2 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
660 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
661 			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
662 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
663 			| 2 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
664 			| 2 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
665 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
666 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
667 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
668 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
669 			| 2 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
670 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
671 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
672 		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
673 
674 		data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
675 			| 2 << UVD_PGFSM_STATUS__UVDU_PWR_STATUS__SHIFT
676 			| 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
677 			| 2 << UVD_PGFSM_STATUS__UVDC_PWR_STATUS__SHIFT
678 			| 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
679 			| 2 << UVD_PGFSM_STATUS__UVDIRL_PWR_STATUS__SHIFT
680 			| 2 << UVD_PGFSM_STATUS__UVDLM_PWR_STATUS__SHIFT
681 			| 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
682 			| 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
683 			| 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT
684 			| 2 << UVD_PGFSM_STATUS__UVDAB_PWR_STATUS__SHIFT
685 			| 2 << UVD_PGFSM_STATUS__UVDATD_PWR_STATUS__SHIFT
686 			| 2 << UVD_PGFSM_STATUS__UVDNA_PWR_STATUS__SHIFT
687 			| 2 << UVD_PGFSM_STATUS__UVDNB_PWR_STATUS__SHIFT);
688 		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS, data, 0x3F3FFFFF);
689 	}
690 }
691 
692 /**
693  * vcn_v3_0_disable_clock_gating - disable VCN clock gating
694  *
695  * @adev: amdgpu_device pointer
696  * @inst: instance number
697  *
698  * Disable clock gating for VCN block
699  */
700 static void vcn_v3_0_disable_clock_gating(struct amdgpu_device *adev, int inst)
701 {
702 	uint32_t data;
703 
704 	/* VCN disable CGC */
705 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
706 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
707 		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
708 	else
709 		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
710 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
711 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
712 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
713 
714 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_GATE);
715 	data &= ~(UVD_CGC_GATE__SYS_MASK
716 		| UVD_CGC_GATE__UDEC_MASK
717 		| UVD_CGC_GATE__MPEG2_MASK
718 		| UVD_CGC_GATE__REGS_MASK
719 		| UVD_CGC_GATE__RBC_MASK
720 		| UVD_CGC_GATE__LMI_MC_MASK
721 		| UVD_CGC_GATE__LMI_UMC_MASK
722 		| UVD_CGC_GATE__IDCT_MASK
723 		| UVD_CGC_GATE__MPRD_MASK
724 		| UVD_CGC_GATE__MPC_MASK
725 		| UVD_CGC_GATE__LBSI_MASK
726 		| UVD_CGC_GATE__LRBBM_MASK
727 		| UVD_CGC_GATE__UDEC_RE_MASK
728 		| UVD_CGC_GATE__UDEC_CM_MASK
729 		| UVD_CGC_GATE__UDEC_IT_MASK
730 		| UVD_CGC_GATE__UDEC_DB_MASK
731 		| UVD_CGC_GATE__UDEC_MP_MASK
732 		| UVD_CGC_GATE__WCB_MASK
733 		| UVD_CGC_GATE__VCPU_MASK
734 		| UVD_CGC_GATE__MMSCH_MASK);
735 
736 	WREG32_SOC15(VCN, inst, mmUVD_CGC_GATE, data);
737 
738 	SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_CGC_GATE, 0,  0xFFFFFFFF);
739 
740 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
741 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
742 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
743 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
744 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
745 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
746 		| UVD_CGC_CTRL__SYS_MODE_MASK
747 		| UVD_CGC_CTRL__UDEC_MODE_MASK
748 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
749 		| UVD_CGC_CTRL__REGS_MODE_MASK
750 		| UVD_CGC_CTRL__RBC_MODE_MASK
751 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
752 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
753 		| UVD_CGC_CTRL__IDCT_MODE_MASK
754 		| UVD_CGC_CTRL__MPRD_MODE_MASK
755 		| UVD_CGC_CTRL__MPC_MODE_MASK
756 		| UVD_CGC_CTRL__LBSI_MODE_MASK
757 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
758 		| UVD_CGC_CTRL__WCB_MODE_MASK
759 		| UVD_CGC_CTRL__VCPU_MODE_MASK
760 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
761 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
762 
763 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE);
764 	data |= (UVD_SUVD_CGC_GATE__SRE_MASK
765 		| UVD_SUVD_CGC_GATE__SIT_MASK
766 		| UVD_SUVD_CGC_GATE__SMP_MASK
767 		| UVD_SUVD_CGC_GATE__SCM_MASK
768 		| UVD_SUVD_CGC_GATE__SDB_MASK
769 		| UVD_SUVD_CGC_GATE__SRE_H264_MASK
770 		| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
771 		| UVD_SUVD_CGC_GATE__SIT_H264_MASK
772 		| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
773 		| UVD_SUVD_CGC_GATE__SCM_H264_MASK
774 		| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
775 		| UVD_SUVD_CGC_GATE__SDB_H264_MASK
776 		| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
777 		| UVD_SUVD_CGC_GATE__SCLR_MASK
778 		| UVD_SUVD_CGC_GATE__ENT_MASK
779 		| UVD_SUVD_CGC_GATE__IME_MASK
780 		| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
781 		| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
782 		| UVD_SUVD_CGC_GATE__SITE_MASK
783 		| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
784 		| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
785 		| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
786 		| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
787 		| UVD_SUVD_CGC_GATE__IME_HEVC_MASK
788 		| UVD_SUVD_CGC_GATE__EFC_MASK
789 		| UVD_SUVD_CGC_GATE__SAOE_MASK
790 		| UVD_SUVD_CGC_GATE__SRE_AV1_MASK
791 		| UVD_SUVD_CGC_GATE__FBC_PCLK_MASK
792 		| UVD_SUVD_CGC_GATE__FBC_CCLK_MASK
793 		| UVD_SUVD_CGC_GATE__SCM_AV1_MASK
794 		| UVD_SUVD_CGC_GATE__SMPA_MASK);
795 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE, data);
796 
797 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE2);
798 	data |= (UVD_SUVD_CGC_GATE2__MPBE0_MASK
799 		| UVD_SUVD_CGC_GATE2__MPBE1_MASK
800 		| UVD_SUVD_CGC_GATE2__SIT_AV1_MASK
801 		| UVD_SUVD_CGC_GATE2__SDB_AV1_MASK
802 		| UVD_SUVD_CGC_GATE2__MPC1_MASK);
803 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE2, data);
804 
805 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL);
806 	data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
807 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
808 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
809 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
810 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
811 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
812 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
813 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
814 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK
815 		| UVD_SUVD_CGC_CTRL__EFC_MODE_MASK
816 		| UVD_SUVD_CGC_CTRL__SAOE_MODE_MASK
817 		| UVD_SUVD_CGC_CTRL__SMPA_MODE_MASK
818 		| UVD_SUVD_CGC_CTRL__MPBE0_MODE_MASK
819 		| UVD_SUVD_CGC_CTRL__MPBE1_MODE_MASK
820 		| UVD_SUVD_CGC_CTRL__SIT_AV1_MODE_MASK
821 		| UVD_SUVD_CGC_CTRL__SDB_AV1_MODE_MASK
822 		| UVD_SUVD_CGC_CTRL__MPC1_MODE_MASK
823 		| UVD_SUVD_CGC_CTRL__FBC_PCLK_MASK
824 		| UVD_SUVD_CGC_CTRL__FBC_CCLK_MASK);
825 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL, data);
826 }
827 
828 static void vcn_v3_0_clock_gating_dpg_mode(struct amdgpu_device *adev,
829 		uint8_t sram_sel, int inst_idx, uint8_t indirect)
830 {
831 	uint32_t reg_data = 0;
832 
833 	/* enable sw clock gating control */
834 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
835 		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
836 	else
837 		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
838 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
839 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
840 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
841 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
842 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
843 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
844 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
845 		 UVD_CGC_CTRL__SYS_MODE_MASK |
846 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
847 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
848 		 UVD_CGC_CTRL__REGS_MODE_MASK |
849 		 UVD_CGC_CTRL__RBC_MODE_MASK |
850 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
851 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
852 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
853 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
854 		 UVD_CGC_CTRL__MPC_MODE_MASK |
855 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
856 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
857 		 UVD_CGC_CTRL__WCB_MODE_MASK |
858 		 UVD_CGC_CTRL__VCPU_MODE_MASK |
859 		 UVD_CGC_CTRL__MMSCH_MODE_MASK);
860 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
861 		VCN, inst_idx, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
862 
863 	/* turn off clock gating */
864 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
865 		VCN, inst_idx, mmUVD_CGC_GATE), 0, sram_sel, indirect);
866 
867 	/* turn on SUVD clock gating */
868 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
869 		VCN, inst_idx, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
870 
871 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
872 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
873 		VCN, inst_idx, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
874 }
875 
876 /**
877  * vcn_v3_0_enable_clock_gating - enable VCN clock gating
878  *
879  * @adev: amdgpu_device pointer
880  * @inst: instance number
881  *
882  * Enable clock gating for VCN block
883  */
884 static void vcn_v3_0_enable_clock_gating(struct amdgpu_device *adev, int inst)
885 {
886 	uint32_t data;
887 
888 	/* enable VCN CGC */
889 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
890 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
891 		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
892 	else
893 		data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
894 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
895 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
896 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
897 
898 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
899 	data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
900 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
901 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
902 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
903 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
904 		| UVD_CGC_CTRL__SYS_MODE_MASK
905 		| UVD_CGC_CTRL__UDEC_MODE_MASK
906 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
907 		| UVD_CGC_CTRL__REGS_MODE_MASK
908 		| UVD_CGC_CTRL__RBC_MODE_MASK
909 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
910 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
911 		| UVD_CGC_CTRL__IDCT_MODE_MASK
912 		| UVD_CGC_CTRL__MPRD_MODE_MASK
913 		| UVD_CGC_CTRL__MPC_MODE_MASK
914 		| UVD_CGC_CTRL__LBSI_MODE_MASK
915 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
916 		| UVD_CGC_CTRL__WCB_MODE_MASK
917 		| UVD_CGC_CTRL__VCPU_MODE_MASK
918 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
919 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
920 
921 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL);
922 	data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
923 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
924 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
925 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
926 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
927 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
928 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
929 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
930 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK
931 		| UVD_SUVD_CGC_CTRL__EFC_MODE_MASK
932 		| UVD_SUVD_CGC_CTRL__SAOE_MODE_MASK
933 		| UVD_SUVD_CGC_CTRL__SMPA_MODE_MASK
934 		| UVD_SUVD_CGC_CTRL__MPBE0_MODE_MASK
935 		| UVD_SUVD_CGC_CTRL__MPBE1_MODE_MASK
936 		| UVD_SUVD_CGC_CTRL__SIT_AV1_MODE_MASK
937 		| UVD_SUVD_CGC_CTRL__SDB_AV1_MODE_MASK
938 		| UVD_SUVD_CGC_CTRL__MPC1_MODE_MASK
939 		| UVD_SUVD_CGC_CTRL__FBC_PCLK_MASK
940 		| UVD_SUVD_CGC_CTRL__FBC_CCLK_MASK);
941 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL, data);
942 }
943 
944 static int vcn_v3_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
945 {
946 	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
947 	struct amdgpu_ring *ring;
948 	uint32_t rb_bufsz, tmp;
949 
950 	/* disable register anti-hang mechanism */
951 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1,
952 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
953 	/* enable dynamic power gating mode */
954 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS);
955 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
956 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
957 	WREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS, tmp);
958 
959 	if (indirect)
960 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
961 
962 	/* enable clock gating */
963 	vcn_v3_0_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
964 
965 	/* enable VCPU clock */
966 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
967 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
968 	tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
969 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
970 		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
971 
972 	/* disable master interupt */
973 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
974 		VCN, inst_idx, mmUVD_MASTINT_EN), 0, 0, indirect);
975 
976 	/* setup mmUVD_LMI_CTRL */
977 	tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
978 		UVD_LMI_CTRL__REQ_MODE_MASK |
979 		UVD_LMI_CTRL__CRC_RESET_MASK |
980 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
981 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
982 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
983 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
984 		0x00100000L);
985 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
986 		VCN, inst_idx, mmUVD_LMI_CTRL), tmp, 0, indirect);
987 
988 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
989 		VCN, inst_idx, mmUVD_MPC_CNTL),
990 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
991 
992 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
993 		VCN, inst_idx, mmUVD_MPC_SET_MUXA0),
994 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
995 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
996 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
997 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
998 
999 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1000 		VCN, inst_idx, mmUVD_MPC_SET_MUXB0),
1001 		 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1002 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1003 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1004 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
1005 
1006 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1007 		VCN, inst_idx, mmUVD_MPC_SET_MUX),
1008 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1009 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1010 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
1011 
1012 	vcn_v3_0_mc_resume_dpg_mode(adev, inst_idx, indirect);
1013 
1014 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1015 		VCN, inst_idx, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
1016 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1017 		VCN, inst_idx, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
1018 
1019 	/* enable LMI MC and UMC channels */
1020 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1021 		VCN, inst_idx, mmUVD_LMI_CTRL2), 0, 0, indirect);
1022 
1023 	/* unblock VCPU register access */
1024 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1025 		VCN, inst_idx, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
1026 
1027 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
1028 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
1029 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1030 		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
1031 
1032 	/* enable master interrupt */
1033 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1034 		VCN, inst_idx, mmUVD_MASTINT_EN),
1035 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
1036 
1037 	/* add nop to workaround PSP size check */
1038 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1039 		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
1040 
1041 	if (indirect)
1042 		amdgpu_vcn_psp_update_sram(adev, inst_idx, 0);
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 (amdgpu_ip_version(adev, UVD_HWIP, 0) !=
1261 		    IP_VERSION(3, 0, 33)) {
1262 			fw_shared->multi_queue.encode_generalpurpose_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1263 			ring = &adev->vcn.inst[i].ring_enc[0];
1264 			WREG32_SOC15(VCN, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1265 			WREG32_SOC15(VCN, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1266 			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO, ring->gpu_addr);
1267 			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1268 			WREG32_SOC15(VCN, i, mmUVD_RB_SIZE, ring->ring_size / 4);
1269 			fw_shared->multi_queue.encode_generalpurpose_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1270 
1271 			fw_shared->multi_queue.encode_lowlatency_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1272 			ring = &adev->vcn.inst[i].ring_enc[1];
1273 			WREG32_SOC15(VCN, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1274 			WREG32_SOC15(VCN, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1275 			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1276 			WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1277 			WREG32_SOC15(VCN, i, mmUVD_RB_SIZE2, ring->ring_size / 4);
1278 			fw_shared->multi_queue.encode_lowlatency_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1279 		}
1280 	}
1281 
1282 	return 0;
1283 }
1284 
1285 static int vcn_v3_0_start_sriov(struct amdgpu_device *adev)
1286 {
1287 	int i, j;
1288 	struct amdgpu_ring *ring;
1289 	uint64_t cache_addr;
1290 	uint64_t rb_addr;
1291 	uint64_t ctx_addr;
1292 	uint32_t param, resp, expected;
1293 	uint32_t offset, cache_size;
1294 	uint32_t tmp, timeout;
1295 
1296 	struct amdgpu_mm_table *table = &adev->virt.mm_table;
1297 	uint32_t *table_loc;
1298 	uint32_t table_size;
1299 	uint32_t size, size_dw;
1300 
1301 	struct mmsch_v3_0_cmd_direct_write
1302 		direct_wt = { {0} };
1303 	struct mmsch_v3_0_cmd_direct_read_modify_write
1304 		direct_rd_mod_wt = { {0} };
1305 	struct mmsch_v3_0_cmd_end end = { {0} };
1306 	struct mmsch_v3_0_init_header header;
1307 
1308 	direct_wt.cmd_header.command_type =
1309 		MMSCH_COMMAND__DIRECT_REG_WRITE;
1310 	direct_rd_mod_wt.cmd_header.command_type =
1311 		MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1312 	end.cmd_header.command_type =
1313 		MMSCH_COMMAND__END;
1314 
1315 	header.version = MMSCH_VERSION;
1316 	header.total_size = sizeof(struct mmsch_v3_0_init_header) >> 2;
1317 	for (i = 0; i < MMSCH_V3_0_VCN_INSTANCES; i++) {
1318 		header.inst[i].init_status = 0;
1319 		header.inst[i].table_offset = 0;
1320 		header.inst[i].table_size = 0;
1321 	}
1322 
1323 	table_loc = (uint32_t *)table->cpu_addr;
1324 	table_loc += header.total_size;
1325 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1326 		if (adev->vcn.harvest_config & (1 << i))
1327 			continue;
1328 
1329 		table_size = 0;
1330 
1331 		MMSCH_V3_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(VCN, i,
1332 			mmUVD_STATUS),
1333 			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1334 
1335 		cache_size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw[i]->size + 4);
1336 
1337 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1338 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1339 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1340 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1341 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1342 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1343 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1344 			offset = 0;
1345 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1346 				mmUVD_VCPU_CACHE_OFFSET0),
1347 				0);
1348 		} else {
1349 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1350 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1351 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1352 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1353 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1354 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1355 			offset = cache_size;
1356 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1357 				mmUVD_VCPU_CACHE_OFFSET0),
1358 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1359 		}
1360 
1361 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1362 			mmUVD_VCPU_CACHE_SIZE0),
1363 			cache_size);
1364 
1365 		cache_addr = adev->vcn.inst[i].gpu_addr + offset;
1366 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1367 			mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1368 			lower_32_bits(cache_addr));
1369 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1370 			mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1371 			upper_32_bits(cache_addr));
1372 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1373 			mmUVD_VCPU_CACHE_OFFSET1),
1374 			0);
1375 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1376 			mmUVD_VCPU_CACHE_SIZE1),
1377 			AMDGPU_VCN_STACK_SIZE);
1378 
1379 		cache_addr = adev->vcn.inst[i].gpu_addr + offset +
1380 			AMDGPU_VCN_STACK_SIZE;
1381 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1382 			mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1383 			lower_32_bits(cache_addr));
1384 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1385 			mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1386 			upper_32_bits(cache_addr));
1387 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1388 			mmUVD_VCPU_CACHE_OFFSET2),
1389 			0);
1390 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1391 			mmUVD_VCPU_CACHE_SIZE2),
1392 			AMDGPU_VCN_CONTEXT_SIZE);
1393 
1394 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
1395 			ring = &adev->vcn.inst[i].ring_enc[j];
1396 			ring->wptr = 0;
1397 			rb_addr = ring->gpu_addr;
1398 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1399 				mmUVD_RB_BASE_LO),
1400 				lower_32_bits(rb_addr));
1401 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1402 				mmUVD_RB_BASE_HI),
1403 				upper_32_bits(rb_addr));
1404 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1405 				mmUVD_RB_SIZE),
1406 				ring->ring_size / 4);
1407 		}
1408 
1409 		ring = &adev->vcn.inst[i].ring_dec;
1410 		ring->wptr = 0;
1411 		rb_addr = ring->gpu_addr;
1412 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1413 			mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1414 			lower_32_bits(rb_addr));
1415 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1416 			mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1417 			upper_32_bits(rb_addr));
1418 		/* force RBC into idle state */
1419 		tmp = order_base_2(ring->ring_size);
1420 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, tmp);
1421 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1422 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1423 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1424 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1425 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1426 			mmUVD_RBC_RB_CNTL),
1427 			tmp);
1428 
1429 		/* add end packet */
1430 		MMSCH_V3_0_INSERT_END();
1431 
1432 		/* refine header */
1433 		header.inst[i].init_status = 0;
1434 		header.inst[i].table_offset = header.total_size;
1435 		header.inst[i].table_size = table_size;
1436 		header.total_size += table_size;
1437 	}
1438 
1439 	/* Update init table header in memory */
1440 	size = sizeof(struct mmsch_v3_0_init_header);
1441 	table_loc = (uint32_t *)table->cpu_addr;
1442 	memcpy((void *)table_loc, &header, size);
1443 
1444 	/* message MMSCH (in VCN[0]) to initialize this client
1445 	 * 1, write to mmsch_vf_ctx_addr_lo/hi register with GPU mc addr
1446 	 * of memory descriptor location
1447 	 */
1448 	ctx_addr = table->gpu_addr;
1449 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(ctx_addr));
1450 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(ctx_addr));
1451 
1452 	/* 2, update vmid of descriptor */
1453 	tmp = RREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID);
1454 	tmp &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1455 	/* use domain0 for MM scheduler */
1456 	tmp |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1457 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID, tmp);
1458 
1459 	/* 3, notify mmsch about the size of this descriptor */
1460 	size = header.total_size;
1461 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_SIZE, size);
1462 
1463 	/* 4, set resp to zero */
1464 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1465 
1466 	/* 5, kick off the initialization and wait until
1467 	 * MMSCH_VF_MAILBOX_RESP becomes non-zero
1468 	 */
1469 	param = 0x10000001;
1470 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_HOST, param);
1471 	tmp = 0;
1472 	timeout = 1000;
1473 	resp = 0;
1474 	expected = param + 1;
1475 	while (resp != expected) {
1476 		resp = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1477 		if (resp == expected)
1478 			break;
1479 
1480 		udelay(10);
1481 		tmp = tmp + 10;
1482 		if (tmp >= timeout) {
1483 			DRM_ERROR("failed to init MMSCH. TIME-OUT after %d usec"\
1484 				" waiting for mmMMSCH_VF_MAILBOX_RESP "\
1485 				"(expected=0x%08x, readback=0x%08x)\n",
1486 				tmp, expected, resp);
1487 			return -EBUSY;
1488 		}
1489 	}
1490 
1491 	return 0;
1492 }
1493 
1494 static int vcn_v3_0_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1495 {
1496 	struct dpg_pause_state state = {.fw_based = VCN_DPG_STATE__UNPAUSE};
1497 	uint32_t tmp;
1498 
1499 	vcn_v3_0_pause_dpg_mode(adev, inst_idx, &state);
1500 
1501 	/* Wait for power status to be 1 */
1502 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1503 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1504 
1505 	/* wait for read ptr to be equal to write ptr */
1506 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR);
1507 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1508 
1509 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2);
1510 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
1511 
1512 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1513 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF);
1514 
1515 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1516 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1517 
1518 	/* disable dynamic power gating mode */
1519 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 0,
1520 		~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1521 
1522 	return 0;
1523 }
1524 
1525 static int vcn_v3_0_stop(struct amdgpu_device *adev)
1526 {
1527 	uint32_t tmp;
1528 	int i, r = 0;
1529 
1530 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1531 		if (adev->vcn.harvest_config & (1 << i))
1532 			continue;
1533 
1534 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1535 			r = vcn_v3_0_stop_dpg_mode(adev, i);
1536 			continue;
1537 		}
1538 
1539 		/* wait for vcn idle */
1540 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1541 		if (r)
1542 			return r;
1543 
1544 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1545 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1546 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1547 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1548 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1549 		if (r)
1550 			return r;
1551 
1552 		/* disable LMI UMC channel */
1553 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2);
1554 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1555 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2, tmp);
1556 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1557 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1558 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1559 		if (r)
1560 			return r;
1561 
1562 		/* block VCPU register access */
1563 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL),
1564 			UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1565 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1566 
1567 		/* reset VCPU */
1568 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1569 			UVD_VCPU_CNTL__BLK_RST_MASK,
1570 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1571 
1572 		/* disable VCPU clock */
1573 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1574 			~(UVD_VCPU_CNTL__CLK_EN_MASK));
1575 
1576 		/* apply soft reset */
1577 		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1578 		tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1579 		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1580 		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1581 		tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1582 		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1583 
1584 		/* clear status */
1585 		WREG32_SOC15(VCN, i, mmUVD_STATUS, 0);
1586 
1587 		/* apply HW clock gating */
1588 		vcn_v3_0_enable_clock_gating(adev, i);
1589 
1590 		/* enable VCN power gating */
1591 		vcn_v3_0_enable_static_power_gating(adev, i);
1592 	}
1593 
1594 	if (adev->pm.dpm_enabled)
1595 		amdgpu_dpm_enable_uvd(adev, false);
1596 
1597 	return 0;
1598 }
1599 
1600 static int vcn_v3_0_pause_dpg_mode(struct amdgpu_device *adev,
1601 		   int inst_idx, struct dpg_pause_state *new_state)
1602 {
1603 	volatile struct amdgpu_fw_shared *fw_shared;
1604 	struct amdgpu_ring *ring;
1605 	uint32_t reg_data = 0;
1606 	int ret_code;
1607 
1608 	/* pause/unpause if state is changed */
1609 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1610 		DRM_DEBUG("dpg pause state changed %d -> %d",
1611 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1612 		reg_data = RREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE) &
1613 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1614 
1615 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1616 			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1617 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1618 
1619 			if (!ret_code) {
1620 				/* pause DPG */
1621 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1622 				WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1623 
1624 				/* wait for ACK */
1625 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_DPG_PAUSE,
1626 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1627 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1628 
1629 				/* Stall DPG before WPTR/RPTR reset */
1630 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1631 					UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1632 					~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1633 
1634 				if (amdgpu_ip_version(adev, UVD_HWIP, 0) !=
1635 				    IP_VERSION(3, 0, 33)) {
1636 					/* Restore */
1637 					fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
1638 					fw_shared->multi_queue.encode_generalpurpose_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1639 					ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1640 					ring->wptr = 0;
1641 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
1642 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1643 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
1644 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1645 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1646 					fw_shared->multi_queue.encode_generalpurpose_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1647 
1648 					fw_shared->multi_queue.encode_lowlatency_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1649 					ring = &adev->vcn.inst[inst_idx].ring_enc[1];
1650 					ring->wptr = 0;
1651 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1652 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1653 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
1654 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1655 					WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1656 					fw_shared->multi_queue.encode_lowlatency_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1657 
1658 					/* restore wptr/rptr with pointers saved in FW shared memory*/
1659 					WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, fw_shared->rb.rptr);
1660 					WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR, fw_shared->rb.wptr);
1661 				}
1662 
1663 				/* Unstall DPG */
1664 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1665 					0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1666 
1667 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS,
1668 					UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1669 			}
1670 		} else {
1671 			/* unpause dpg, no need to wait */
1672 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1673 			WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1674 		}
1675 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1676 	}
1677 
1678 	return 0;
1679 }
1680 
1681 /**
1682  * vcn_v3_0_dec_ring_get_rptr - get read pointer
1683  *
1684  * @ring: amdgpu_ring pointer
1685  *
1686  * Returns the current hardware read pointer
1687  */
1688 static uint64_t vcn_v3_0_dec_ring_get_rptr(struct amdgpu_ring *ring)
1689 {
1690 	struct amdgpu_device *adev = ring->adev;
1691 
1692 	return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_RPTR);
1693 }
1694 
1695 /**
1696  * vcn_v3_0_dec_ring_get_wptr - get write pointer
1697  *
1698  * @ring: amdgpu_ring pointer
1699  *
1700  * Returns the current hardware write pointer
1701  */
1702 static uint64_t vcn_v3_0_dec_ring_get_wptr(struct amdgpu_ring *ring)
1703 {
1704 	struct amdgpu_device *adev = ring->adev;
1705 
1706 	if (ring->use_doorbell)
1707 		return *ring->wptr_cpu_addr;
1708 	else
1709 		return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR);
1710 }
1711 
1712 /**
1713  * vcn_v3_0_dec_ring_set_wptr - set write pointer
1714  *
1715  * @ring: amdgpu_ring pointer
1716  *
1717  * Commits the write pointer to the hardware
1718  */
1719 static void vcn_v3_0_dec_ring_set_wptr(struct amdgpu_ring *ring)
1720 {
1721 	struct amdgpu_device *adev = ring->adev;
1722 	volatile struct amdgpu_fw_shared *fw_shared;
1723 
1724 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1725 		/*whenever update RBC_RB_WPTR, we save the wptr in shared rb.wptr and scratch2 */
1726 		fw_shared = adev->vcn.inst[ring->me].fw_shared.cpu_addr;
1727 		fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1728 		WREG32_SOC15(VCN, ring->me, mmUVD_SCRATCH2,
1729 			lower_32_bits(ring->wptr));
1730 	}
1731 
1732 	if (ring->use_doorbell) {
1733 		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1734 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1735 	} else {
1736 		WREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1737 	}
1738 }
1739 
1740 static const struct amdgpu_ring_funcs vcn_v3_0_dec_sw_ring_vm_funcs = {
1741 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1742 	.align_mask = 0x3f,
1743 	.nop = VCN_DEC_SW_CMD_NO_OP,
1744 	.secure_submission_supported = true,
1745 	.get_rptr = vcn_v3_0_dec_ring_get_rptr,
1746 	.get_wptr = vcn_v3_0_dec_ring_get_wptr,
1747 	.set_wptr = vcn_v3_0_dec_ring_set_wptr,
1748 	.emit_frame_size =
1749 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1750 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1751 		VCN_SW_RING_EMIT_FRAME_SIZE,
1752 	.emit_ib_size = 5, /* vcn_dec_sw_ring_emit_ib */
1753 	.emit_ib = vcn_dec_sw_ring_emit_ib,
1754 	.emit_fence = vcn_dec_sw_ring_emit_fence,
1755 	.emit_vm_flush = vcn_dec_sw_ring_emit_vm_flush,
1756 	.test_ring = amdgpu_vcn_dec_sw_ring_test_ring,
1757 	.test_ib = NULL,//amdgpu_vcn_dec_sw_ring_test_ib,
1758 	.insert_nop = amdgpu_ring_insert_nop,
1759 	.insert_end = vcn_dec_sw_ring_insert_end,
1760 	.pad_ib = amdgpu_ring_generic_pad_ib,
1761 	.begin_use = amdgpu_vcn_ring_begin_use,
1762 	.end_use = amdgpu_vcn_ring_end_use,
1763 	.emit_wreg = vcn_dec_sw_ring_emit_wreg,
1764 	.emit_reg_wait = vcn_dec_sw_ring_emit_reg_wait,
1765 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1766 };
1767 
1768 static int vcn_v3_0_limit_sched(struct amdgpu_cs_parser *p,
1769 				struct amdgpu_job *job)
1770 {
1771 	struct drm_gpu_scheduler **scheds;
1772 
1773 	/* The create msg must be in the first IB submitted */
1774 	if (atomic_read(&job->base.entity->fence_seq))
1775 		return -EINVAL;
1776 
1777 	/* if VCN0 is harvested, we can't support AV1 */
1778 	if (p->adev->vcn.harvest_config & AMDGPU_VCN_HARVEST_VCN0)
1779 		return -EINVAL;
1780 
1781 	scheds = p->adev->gpu_sched[AMDGPU_HW_IP_VCN_DEC]
1782 		[AMDGPU_RING_PRIO_DEFAULT].sched;
1783 	drm_sched_entity_modify_sched(job->base.entity, scheds, 1);
1784 	return 0;
1785 }
1786 
1787 static int vcn_v3_0_dec_msg(struct amdgpu_cs_parser *p, struct amdgpu_job *job,
1788 			    uint64_t addr)
1789 {
1790 	struct ttm_operation_ctx ctx = { false, false };
1791 	struct amdgpu_bo_va_mapping *map;
1792 	uint32_t *msg, num_buffers;
1793 	struct amdgpu_bo *bo;
1794 	uint64_t start, end;
1795 	unsigned int i;
1796 	void *ptr;
1797 	int r;
1798 
1799 	addr &= AMDGPU_GMC_HOLE_MASK;
1800 	r = amdgpu_cs_find_mapping(p, addr, &bo, &map);
1801 	if (r) {
1802 		DRM_ERROR("Can't find BO for addr 0x%08Lx\n", addr);
1803 		return r;
1804 	}
1805 
1806 	start = map->start * AMDGPU_GPU_PAGE_SIZE;
1807 	end = (map->last + 1) * AMDGPU_GPU_PAGE_SIZE;
1808 	if (addr & 0x7) {
1809 		DRM_ERROR("VCN messages must be 8 byte aligned!\n");
1810 		return -EINVAL;
1811 	}
1812 
1813 	bo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
1814 	amdgpu_bo_placement_from_domain(bo, bo->allowed_domains);
1815 	r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
1816 	if (r) {
1817 		DRM_ERROR("Failed validating the VCN message BO (%d)!\n", r);
1818 		return r;
1819 	}
1820 
1821 	r = amdgpu_bo_kmap(bo, &ptr);
1822 	if (r) {
1823 		DRM_ERROR("Failed mapping the VCN message (%d)!\n", r);
1824 		return r;
1825 	}
1826 
1827 	msg = ptr + addr - start;
1828 
1829 	/* Check length */
1830 	if (msg[1] > end - addr) {
1831 		r = -EINVAL;
1832 		goto out;
1833 	}
1834 
1835 	if (msg[3] != RDECODE_MSG_CREATE)
1836 		goto out;
1837 
1838 	num_buffers = msg[2];
1839 	for (i = 0, msg = &msg[6]; i < num_buffers; ++i, msg += 4) {
1840 		uint32_t offset, size, *create;
1841 
1842 		if (msg[0] != RDECODE_MESSAGE_CREATE)
1843 			continue;
1844 
1845 		offset = msg[1];
1846 		size = msg[2];
1847 
1848 		if (offset + size > end) {
1849 			r = -EINVAL;
1850 			goto out;
1851 		}
1852 
1853 		create = ptr + addr + offset - start;
1854 
1855 		/* H246, HEVC and VP9 can run on any instance */
1856 		if (create[0] == 0x7 || create[0] == 0x10 || create[0] == 0x11)
1857 			continue;
1858 
1859 		r = vcn_v3_0_limit_sched(p, job);
1860 		if (r)
1861 			goto out;
1862 	}
1863 
1864 out:
1865 	amdgpu_bo_kunmap(bo);
1866 	return r;
1867 }
1868 
1869 static int vcn_v3_0_ring_patch_cs_in_place(struct amdgpu_cs_parser *p,
1870 					   struct amdgpu_job *job,
1871 					   struct amdgpu_ib *ib)
1872 {
1873 	struct amdgpu_ring *ring = amdgpu_job_ring(job);
1874 	uint32_t msg_lo = 0, msg_hi = 0;
1875 	unsigned i;
1876 	int r;
1877 
1878 	/* The first instance can decode anything */
1879 	if (!ring->me)
1880 		return 0;
1881 
1882 	for (i = 0; i < ib->length_dw; i += 2) {
1883 		uint32_t reg = amdgpu_ib_get_value(ib, i);
1884 		uint32_t val = amdgpu_ib_get_value(ib, i + 1);
1885 
1886 		if (reg == PACKET0(p->adev->vcn.internal.data0, 0)) {
1887 			msg_lo = val;
1888 		} else if (reg == PACKET0(p->adev->vcn.internal.data1, 0)) {
1889 			msg_hi = val;
1890 		} else if (reg == PACKET0(p->adev->vcn.internal.cmd, 0) &&
1891 			   val == 0) {
1892 			r = vcn_v3_0_dec_msg(p, job,
1893 					     ((u64)msg_hi) << 32 | msg_lo);
1894 			if (r)
1895 				return r;
1896 		}
1897 	}
1898 	return 0;
1899 }
1900 
1901 static const struct amdgpu_ring_funcs vcn_v3_0_dec_ring_vm_funcs = {
1902 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1903 	.align_mask = 0xf,
1904 	.secure_submission_supported = true,
1905 	.get_rptr = vcn_v3_0_dec_ring_get_rptr,
1906 	.get_wptr = vcn_v3_0_dec_ring_get_wptr,
1907 	.set_wptr = vcn_v3_0_dec_ring_set_wptr,
1908 	.patch_cs_in_place = vcn_v3_0_ring_patch_cs_in_place,
1909 	.emit_frame_size =
1910 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1911 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1912 		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1913 		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1914 		6,
1915 	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1916 	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
1917 	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
1918 	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1919 	.test_ring = vcn_v2_0_dec_ring_test_ring,
1920 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
1921 	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
1922 	.insert_start = vcn_v2_0_dec_ring_insert_start,
1923 	.insert_end = vcn_v2_0_dec_ring_insert_end,
1924 	.pad_ib = amdgpu_ring_generic_pad_ib,
1925 	.begin_use = amdgpu_vcn_ring_begin_use,
1926 	.end_use = amdgpu_vcn_ring_end_use,
1927 	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1928 	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1929 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1930 };
1931 
1932 /**
1933  * vcn_v3_0_enc_ring_get_rptr - get enc read pointer
1934  *
1935  * @ring: amdgpu_ring pointer
1936  *
1937  * Returns the current hardware enc read pointer
1938  */
1939 static uint64_t vcn_v3_0_enc_ring_get_rptr(struct amdgpu_ring *ring)
1940 {
1941 	struct amdgpu_device *adev = ring->adev;
1942 
1943 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0])
1944 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR);
1945 	else
1946 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR2);
1947 }
1948 
1949 /**
1950  * vcn_v3_0_enc_ring_get_wptr - get enc write pointer
1951  *
1952  * @ring: amdgpu_ring pointer
1953  *
1954  * Returns the current hardware enc write pointer
1955  */
1956 static uint64_t vcn_v3_0_enc_ring_get_wptr(struct amdgpu_ring *ring)
1957 {
1958 	struct amdgpu_device *adev = ring->adev;
1959 
1960 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1961 		if (ring->use_doorbell)
1962 			return *ring->wptr_cpu_addr;
1963 		else
1964 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR);
1965 	} else {
1966 		if (ring->use_doorbell)
1967 			return *ring->wptr_cpu_addr;
1968 		else
1969 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2);
1970 	}
1971 }
1972 
1973 /**
1974  * vcn_v3_0_enc_ring_set_wptr - set enc write pointer
1975  *
1976  * @ring: amdgpu_ring pointer
1977  *
1978  * Commits the enc write pointer to the hardware
1979  */
1980 static void vcn_v3_0_enc_ring_set_wptr(struct amdgpu_ring *ring)
1981 {
1982 	struct amdgpu_device *adev = ring->adev;
1983 
1984 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1985 		if (ring->use_doorbell) {
1986 			*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1987 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1988 		} else {
1989 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1990 		}
1991 	} else {
1992 		if (ring->use_doorbell) {
1993 			*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1994 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1995 		} else {
1996 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1997 		}
1998 	}
1999 }
2000 
2001 static const struct amdgpu_ring_funcs vcn_v3_0_enc_ring_vm_funcs = {
2002 	.type = AMDGPU_RING_TYPE_VCN_ENC,
2003 	.align_mask = 0x3f,
2004 	.nop = VCN_ENC_CMD_NO_OP,
2005 	.get_rptr = vcn_v3_0_enc_ring_get_rptr,
2006 	.get_wptr = vcn_v3_0_enc_ring_get_wptr,
2007 	.set_wptr = vcn_v3_0_enc_ring_set_wptr,
2008 	.emit_frame_size =
2009 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2010 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
2011 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
2012 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
2013 		1, /* vcn_v2_0_enc_ring_insert_end */
2014 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
2015 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
2016 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
2017 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
2018 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
2019 	.test_ib = amdgpu_vcn_enc_ring_test_ib,
2020 	.insert_nop = amdgpu_ring_insert_nop,
2021 	.insert_end = vcn_v2_0_enc_ring_insert_end,
2022 	.pad_ib = amdgpu_ring_generic_pad_ib,
2023 	.begin_use = amdgpu_vcn_ring_begin_use,
2024 	.end_use = amdgpu_vcn_ring_end_use,
2025 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
2026 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
2027 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2028 };
2029 
2030 static void vcn_v3_0_set_dec_ring_funcs(struct amdgpu_device *adev)
2031 {
2032 	int i;
2033 
2034 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2035 		if (adev->vcn.harvest_config & (1 << i))
2036 			continue;
2037 
2038 		if (!DEC_SW_RING_ENABLED)
2039 			adev->vcn.inst[i].ring_dec.funcs = &vcn_v3_0_dec_ring_vm_funcs;
2040 		else
2041 			adev->vcn.inst[i].ring_dec.funcs = &vcn_v3_0_dec_sw_ring_vm_funcs;
2042 		adev->vcn.inst[i].ring_dec.me = i;
2043 		DRM_INFO("VCN(%d) decode%s is enabled in VM mode\n", i,
2044 			  DEC_SW_RING_ENABLED?"(Software Ring)":"");
2045 	}
2046 }
2047 
2048 static void vcn_v3_0_set_enc_ring_funcs(struct amdgpu_device *adev)
2049 {
2050 	int i, j;
2051 
2052 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2053 		if (adev->vcn.harvest_config & (1 << i))
2054 			continue;
2055 
2056 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
2057 			adev->vcn.inst[i].ring_enc[j].funcs = &vcn_v3_0_enc_ring_vm_funcs;
2058 			adev->vcn.inst[i].ring_enc[j].me = i;
2059 		}
2060 		if (adev->vcn.num_enc_rings > 0)
2061 			DRM_INFO("VCN(%d) encode is enabled in VM mode\n", i);
2062 	}
2063 }
2064 
2065 static bool vcn_v3_0_is_idle(void *handle)
2066 {
2067 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2068 	int i, ret = 1;
2069 
2070 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2071 		if (adev->vcn.harvest_config & (1 << i))
2072 			continue;
2073 
2074 		ret &= (RREG32_SOC15(VCN, i, mmUVD_STATUS) == UVD_STATUS__IDLE);
2075 	}
2076 
2077 	return ret;
2078 }
2079 
2080 static int vcn_v3_0_wait_for_idle(void *handle)
2081 {
2082 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2083 	int i, ret = 0;
2084 
2085 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2086 		if (adev->vcn.harvest_config & (1 << i))
2087 			continue;
2088 
2089 		ret = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE,
2090 			UVD_STATUS__IDLE);
2091 		if (ret)
2092 			return ret;
2093 	}
2094 
2095 	return ret;
2096 }
2097 
2098 static int vcn_v3_0_set_clockgating_state(void *handle,
2099 					  enum amd_clockgating_state state)
2100 {
2101 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2102 	bool enable = state == AMD_CG_STATE_GATE;
2103 	int i;
2104 
2105 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2106 		if (adev->vcn.harvest_config & (1 << i))
2107 			continue;
2108 
2109 		if (enable) {
2110 			if (RREG32_SOC15(VCN, i, mmUVD_STATUS) != UVD_STATUS__IDLE)
2111 				return -EBUSY;
2112 			vcn_v3_0_enable_clock_gating(adev, i);
2113 		} else {
2114 			vcn_v3_0_disable_clock_gating(adev, i);
2115 		}
2116 	}
2117 
2118 	return 0;
2119 }
2120 
2121 static int vcn_v3_0_set_powergating_state(void *handle,
2122 					  enum amd_powergating_state state)
2123 {
2124 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2125 	int ret;
2126 
2127 	/* for SRIOV, guest should not control VCN Power-gating
2128 	 * MMSCH FW should control Power-gating and clock-gating
2129 	 * guest should avoid touching CGC and PG
2130 	 */
2131 	if (amdgpu_sriov_vf(adev)) {
2132 		adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
2133 		return 0;
2134 	}
2135 
2136 	if (state == adev->vcn.cur_state)
2137 		return 0;
2138 
2139 	if (state == AMD_PG_STATE_GATE)
2140 		ret = vcn_v3_0_stop(adev);
2141 	else
2142 		ret = vcn_v3_0_start(adev);
2143 
2144 	if (!ret)
2145 		adev->vcn.cur_state = state;
2146 
2147 	return ret;
2148 }
2149 
2150 static int vcn_v3_0_set_interrupt_state(struct amdgpu_device *adev,
2151 					struct amdgpu_irq_src *source,
2152 					unsigned type,
2153 					enum amdgpu_interrupt_state state)
2154 {
2155 	return 0;
2156 }
2157 
2158 static int vcn_v3_0_process_interrupt(struct amdgpu_device *adev,
2159 				      struct amdgpu_irq_src *source,
2160 				      struct amdgpu_iv_entry *entry)
2161 {
2162 	uint32_t ip_instance;
2163 
2164 	switch (entry->client_id) {
2165 	case SOC15_IH_CLIENTID_VCN:
2166 		ip_instance = 0;
2167 		break;
2168 	case SOC15_IH_CLIENTID_VCN1:
2169 		ip_instance = 1;
2170 		break;
2171 	default:
2172 		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
2173 		return 0;
2174 	}
2175 
2176 	DRM_DEBUG("IH: VCN TRAP\n");
2177 
2178 	switch (entry->src_id) {
2179 	case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
2180 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec);
2181 		break;
2182 	case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
2183 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
2184 		break;
2185 	case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
2186 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]);
2187 		break;
2188 	default:
2189 		DRM_ERROR("Unhandled interrupt: %d %d\n",
2190 			  entry->src_id, entry->src_data[0]);
2191 		break;
2192 	}
2193 
2194 	return 0;
2195 }
2196 
2197 static const struct amdgpu_irq_src_funcs vcn_v3_0_irq_funcs = {
2198 	.set = vcn_v3_0_set_interrupt_state,
2199 	.process = vcn_v3_0_process_interrupt,
2200 };
2201 
2202 static void vcn_v3_0_set_irq_funcs(struct amdgpu_device *adev)
2203 {
2204 	int i;
2205 
2206 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2207 		if (adev->vcn.harvest_config & (1 << i))
2208 			continue;
2209 
2210 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
2211 		adev->vcn.inst[i].irq.funcs = &vcn_v3_0_irq_funcs;
2212 	}
2213 }
2214 
2215 static const struct amd_ip_funcs vcn_v3_0_ip_funcs = {
2216 	.name = "vcn_v3_0",
2217 	.early_init = vcn_v3_0_early_init,
2218 	.late_init = NULL,
2219 	.sw_init = vcn_v3_0_sw_init,
2220 	.sw_fini = vcn_v3_0_sw_fini,
2221 	.hw_init = vcn_v3_0_hw_init,
2222 	.hw_fini = vcn_v3_0_hw_fini,
2223 	.suspend = vcn_v3_0_suspend,
2224 	.resume = vcn_v3_0_resume,
2225 	.is_idle = vcn_v3_0_is_idle,
2226 	.wait_for_idle = vcn_v3_0_wait_for_idle,
2227 	.check_soft_reset = NULL,
2228 	.pre_soft_reset = NULL,
2229 	.soft_reset = NULL,
2230 	.post_soft_reset = NULL,
2231 	.set_clockgating_state = vcn_v3_0_set_clockgating_state,
2232 	.set_powergating_state = vcn_v3_0_set_powergating_state,
2233 };
2234 
2235 const struct amdgpu_ip_block_version vcn_v3_0_ip_block = {
2236 	.type = AMD_IP_BLOCK_TYPE_VCN,
2237 	.major = 3,
2238 	.minor = 0,
2239 	.rev = 0,
2240 	.funcs = &vcn_v3_0_ip_funcs,
2241 };
2242