xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c (revision 0cb4228f)
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 <drm/drm_drv.h>
26 
27 #include "amdgpu.h"
28 #include "amdgpu_vcn.h"
29 #include "amdgpu_pm.h"
30 #include "soc15.h"
31 #include "soc15d.h"
32 #include "vcn_v2_0.h"
33 #include "mmsch_v1_0.h"
34 #include "vcn_v2_5.h"
35 
36 #include "vcn/vcn_2_5_offset.h"
37 #include "vcn/vcn_2_5_sh_mask.h"
38 #include "ivsrcid/vcn/irqsrcs_vcn_2_0.h"
39 
40 #define VCN_VID_SOC_ADDRESS_2_0					0x1fa00
41 #define VCN1_VID_SOC_ADDRESS_3_0				0x48200
42 
43 #define mmUVD_CONTEXT_ID_INTERNAL_OFFSET			0x27
44 #define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET			0x0f
45 #define mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET			0x10
46 #define mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET			0x11
47 #define mmUVD_NO_OP_INTERNAL_OFFSET				0x29
48 #define mmUVD_GP_SCRATCH8_INTERNAL_OFFSET			0x66
49 #define mmUVD_SCRATCH9_INTERNAL_OFFSET				0xc01d
50 
51 #define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET			0x431
52 #define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET		0x3b4
53 #define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET		0x3b5
54 #define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET			0x25c
55 
56 #define VCN25_MAX_HW_INSTANCES_ARCTURUS			2
57 
58 static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev);
59 static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev);
60 static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev);
61 static int vcn_v2_5_set_powergating_state(void *handle,
62 				enum amd_powergating_state state);
63 static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
64 				int inst_idx, struct dpg_pause_state *new_state);
65 static int vcn_v2_5_sriov_start(struct amdgpu_device *adev);
66 static void vcn_v2_5_set_ras_funcs(struct amdgpu_device *adev);
67 
68 static int amdgpu_ih_clientid_vcns[] = {
69 	SOC15_IH_CLIENTID_VCN,
70 	SOC15_IH_CLIENTID_VCN1
71 };
72 
73 /**
74  * vcn_v2_5_early_init - set function pointers
75  *
76  * @handle: amdgpu_device pointer
77  *
78  * Set ring and irq function pointers
79  */
80 static int vcn_v2_5_early_init(void *handle)
81 {
82 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
83 
84 	if (amdgpu_sriov_vf(adev)) {
85 		adev->vcn.num_vcn_inst = 2;
86 		adev->vcn.harvest_config = 0;
87 		adev->vcn.num_enc_rings = 1;
88 	} else {
89 		u32 harvest;
90 		int i;
91 
92 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
93 			harvest = RREG32_SOC15(VCN, i, mmCC_UVD_HARVESTING);
94 			if (harvest & CC_UVD_HARVESTING__UVD_DISABLE_MASK)
95 				adev->vcn.harvest_config |= 1 << i;
96 		}
97 		if (adev->vcn.harvest_config == (AMDGPU_VCN_HARVEST_VCN0 |
98 					AMDGPU_VCN_HARVEST_VCN1))
99 			/* both instances are harvested, disable the block */
100 			return -ENOENT;
101 
102 		adev->vcn.num_enc_rings = 2;
103 	}
104 
105 	vcn_v2_5_set_dec_ring_funcs(adev);
106 	vcn_v2_5_set_enc_ring_funcs(adev);
107 	vcn_v2_5_set_irq_funcs(adev);
108 	vcn_v2_5_set_ras_funcs(adev);
109 
110 	return 0;
111 }
112 
113 /**
114  * vcn_v2_5_sw_init - sw init for VCN block
115  *
116  * @handle: amdgpu_device pointer
117  *
118  * Load firmware and sw initialization
119  */
120 static int vcn_v2_5_sw_init(void *handle)
121 {
122 	struct amdgpu_ring *ring;
123 	int i, j, r;
124 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
125 
126 	for (j = 0; j < adev->vcn.num_vcn_inst; j++) {
127 		if (adev->vcn.harvest_config & (1 << j))
128 			continue;
129 		/* VCN DEC TRAP */
130 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
131 				VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst[j].irq);
132 		if (r)
133 			return r;
134 
135 		/* VCN ENC TRAP */
136 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
137 			r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
138 				i + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[j].irq);
139 			if (r)
140 				return r;
141 		}
142 
143 		/* VCN POISON TRAP */
144 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
145 			VCN_2_6__SRCID_UVD_POISON, &adev->vcn.inst[j].irq);
146 		if (r)
147 			return r;
148 	}
149 
150 	r = amdgpu_vcn_sw_init(adev);
151 	if (r)
152 		return r;
153 
154 	amdgpu_vcn_setup_ucode(adev);
155 
156 	r = amdgpu_vcn_resume(adev);
157 	if (r)
158 		return r;
159 
160 	for (j = 0; j < adev->vcn.num_vcn_inst; j++) {
161 		volatile struct amdgpu_fw_shared *fw_shared;
162 
163 		if (adev->vcn.harvest_config & (1 << j))
164 			continue;
165 		adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
166 		adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
167 		adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
168 		adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
169 		adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
170 		adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
171 
172 		adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
173 		adev->vcn.inst[j].external.scratch9 = SOC15_REG_OFFSET(VCN, j, mmUVD_SCRATCH9);
174 		adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
175 		adev->vcn.inst[j].external.data0 = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_DATA0);
176 		adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
177 		adev->vcn.inst[j].external.data1 = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_DATA1);
178 		adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
179 		adev->vcn.inst[j].external.cmd = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_CMD);
180 		adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
181 		adev->vcn.inst[j].external.nop = SOC15_REG_OFFSET(VCN, j, mmUVD_NO_OP);
182 
183 		ring = &adev->vcn.inst[j].ring_dec;
184 		ring->use_doorbell = true;
185 
186 		ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
187 				(amdgpu_sriov_vf(adev) ? 2*j : 8*j);
188 		sprintf(ring->name, "vcn_dec_%d", j);
189 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[j].irq,
190 				     0, AMDGPU_RING_PRIO_DEFAULT, NULL);
191 		if (r)
192 			return r;
193 
194 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
195 			enum amdgpu_ring_priority_level hw_prio = amdgpu_vcn_get_enc_ring_prio(i);
196 
197 			ring = &adev->vcn.inst[j].ring_enc[i];
198 			ring->use_doorbell = true;
199 
200 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
201 					(amdgpu_sriov_vf(adev) ? (1 + i + 2*j) : (2 + i + 8*j));
202 
203 			sprintf(ring->name, "vcn_enc_%d.%d", j, i);
204 			r = amdgpu_ring_init(adev, ring, 512,
205 					     &adev->vcn.inst[j].irq, 0,
206 					     hw_prio, NULL);
207 			if (r)
208 				return r;
209 		}
210 
211 		fw_shared = adev->vcn.inst[j].fw_shared.cpu_addr;
212 		fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG);
213 
214 		if (amdgpu_vcnfw_log)
215 			amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
216 	}
217 
218 	if (amdgpu_sriov_vf(adev)) {
219 		r = amdgpu_virt_alloc_mm_table(adev);
220 		if (r)
221 			return r;
222 	}
223 
224 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
225 		adev->vcn.pause_dpg_mode = vcn_v2_5_pause_dpg_mode;
226 
227 	return 0;
228 }
229 
230 /**
231  * vcn_v2_5_sw_fini - sw fini for VCN block
232  *
233  * @handle: amdgpu_device pointer
234  *
235  * VCN suspend and free up sw allocation
236  */
237 static int vcn_v2_5_sw_fini(void *handle)
238 {
239 	int i, r, idx;
240 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
241 	volatile struct amdgpu_fw_shared *fw_shared;
242 
243 	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
244 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
245 			if (adev->vcn.harvest_config & (1 << i))
246 				continue;
247 			fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
248 			fw_shared->present_flag_0 = 0;
249 		}
250 		drm_dev_exit(idx);
251 	}
252 
253 
254 	if (amdgpu_sriov_vf(adev))
255 		amdgpu_virt_free_mm_table(adev);
256 
257 	r = amdgpu_vcn_suspend(adev);
258 	if (r)
259 		return r;
260 
261 	r = amdgpu_vcn_sw_fini(adev);
262 
263 	return r;
264 }
265 
266 /**
267  * vcn_v2_5_hw_init - start and test VCN block
268  *
269  * @handle: amdgpu_device pointer
270  *
271  * Initialize the hardware, boot up the VCPU and do some testing
272  */
273 static int vcn_v2_5_hw_init(void *handle)
274 {
275 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
276 	struct amdgpu_ring *ring;
277 	int i, j, r = 0;
278 
279 	if (amdgpu_sriov_vf(adev))
280 		r = vcn_v2_5_sriov_start(adev);
281 
282 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
283 		if (adev->vcn.harvest_config & (1 << j))
284 			continue;
285 
286 		if (amdgpu_sriov_vf(adev)) {
287 			adev->vcn.inst[j].ring_enc[0].sched.ready = true;
288 			adev->vcn.inst[j].ring_enc[1].sched.ready = false;
289 			adev->vcn.inst[j].ring_enc[2].sched.ready = false;
290 			adev->vcn.inst[j].ring_dec.sched.ready = true;
291 		} else {
292 
293 			ring = &adev->vcn.inst[j].ring_dec;
294 
295 			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
296 						     ring->doorbell_index, j);
297 
298 			r = amdgpu_ring_test_helper(ring);
299 			if (r)
300 				goto done;
301 
302 			for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
303 				ring = &adev->vcn.inst[j].ring_enc[i];
304 				r = amdgpu_ring_test_helper(ring);
305 				if (r)
306 					goto done;
307 			}
308 		}
309 	}
310 
311 done:
312 	if (!r)
313 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
314 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
315 
316 	return r;
317 }
318 
319 /**
320  * vcn_v2_5_hw_fini - stop the hardware block
321  *
322  * @handle: amdgpu_device pointer
323  *
324  * Stop the VCN block, mark ring as not ready any more
325  */
326 static int vcn_v2_5_hw_fini(void *handle)
327 {
328 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
329 	int i;
330 
331 	cancel_delayed_work_sync(&adev->vcn.idle_work);
332 
333 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
334 		if (adev->vcn.harvest_config & (1 << i))
335 			continue;
336 
337 		if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
338 		    (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
339 		     RREG32_SOC15(VCN, i, mmUVD_STATUS)))
340 			vcn_v2_5_set_powergating_state(adev, AMD_PG_STATE_GATE);
341 	}
342 
343 	return 0;
344 }
345 
346 /**
347  * vcn_v2_5_suspend - suspend VCN block
348  *
349  * @handle: amdgpu_device pointer
350  *
351  * HW fini and suspend VCN block
352  */
353 static int vcn_v2_5_suspend(void *handle)
354 {
355 	int r;
356 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
357 
358 	r = vcn_v2_5_hw_fini(adev);
359 	if (r)
360 		return r;
361 
362 	r = amdgpu_vcn_suspend(adev);
363 
364 	return r;
365 }
366 
367 /**
368  * vcn_v2_5_resume - resume VCN block
369  *
370  * @handle: amdgpu_device pointer
371  *
372  * Resume firmware and hw init VCN block
373  */
374 static int vcn_v2_5_resume(void *handle)
375 {
376 	int r;
377 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
378 
379 	r = amdgpu_vcn_resume(adev);
380 	if (r)
381 		return r;
382 
383 	r = vcn_v2_5_hw_init(adev);
384 
385 	return r;
386 }
387 
388 /**
389  * vcn_v2_5_mc_resume - memory controller programming
390  *
391  * @adev: amdgpu_device pointer
392  *
393  * Let the VCN memory controller know it's offsets
394  */
395 static void vcn_v2_5_mc_resume(struct amdgpu_device *adev)
396 {
397 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
398 	uint32_t offset;
399 	int i;
400 
401 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
402 		if (adev->vcn.harvest_config & (1 << i))
403 			continue;
404 		/* cache window 0: fw */
405 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
406 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
407 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo));
408 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
409 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi));
410 			WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET0, 0);
411 			offset = 0;
412 		} else {
413 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
414 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
415 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
416 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
417 			offset = size;
418 			WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET0,
419 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
420 		}
421 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE0, size);
422 
423 		/* cache window 1: stack */
424 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
425 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset));
426 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
427 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset));
428 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET1, 0);
429 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
430 
431 		/* cache window 2: context */
432 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
433 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
434 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
435 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
436 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET2, 0);
437 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
438 
439 		/* non-cache window */
440 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
441 			lower_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
442 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
443 			upper_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
444 		WREG32_SOC15(VCN, i, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
445 		WREG32_SOC15(VCN, i, mmUVD_VCPU_NONCACHE_SIZE0,
446 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
447 	}
448 }
449 
450 static void vcn_v2_5_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
451 {
452 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
453 	uint32_t offset;
454 
455 	/* cache window 0: fw */
456 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
457 		if (!indirect) {
458 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
459 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
460 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
461 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
462 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
463 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
464 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
465 				VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
466 		} else {
467 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
468 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
469 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
470 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
471 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
472 				VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
473 		}
474 		offset = 0;
475 	} else {
476 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
477 			VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
478 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
479 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
480 			VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
481 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
482 		offset = size;
483 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
484 			VCN, 0, mmUVD_VCPU_CACHE_OFFSET0),
485 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
486 	}
487 
488 	if (!indirect)
489 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
490 			VCN, 0, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
491 	else
492 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
493 			VCN, 0, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
494 
495 	/* cache window 1: stack */
496 	if (!indirect) {
497 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
498 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
499 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
500 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
501 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
502 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
503 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
504 			VCN, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
505 	} else {
506 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
507 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
508 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
509 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
510 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
511 			VCN, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
512 	}
513 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
514 		VCN, 0, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
515 
516 	/* cache window 2: context */
517 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
518 		VCN, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
519 		lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
520 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
521 		VCN, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
522 		upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
523 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
524 		VCN, 0, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
525 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
526 		VCN, 0, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
527 
528 	/* non-cache window */
529 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
530 		VCN, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
531 		lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
532 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
533 		VCN, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
534 		upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
535 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
536 		VCN, 0, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
537 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
538 		VCN, 0, mmUVD_VCPU_NONCACHE_SIZE0),
539 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
540 
541 	/* VCN global tiling registers */
542 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
543 		VCN, 0, mmUVD_GFX8_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
544 }
545 
546 /**
547  * vcn_v2_5_disable_clock_gating - disable VCN clock gating
548  *
549  * @adev: amdgpu_device pointer
550  *
551  * Disable clock gating for VCN block
552  */
553 static void vcn_v2_5_disable_clock_gating(struct amdgpu_device *adev)
554 {
555 	uint32_t data;
556 	int i;
557 
558 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
559 		if (adev->vcn.harvest_config & (1 << i))
560 			continue;
561 		/* UVD disable CGC */
562 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
563 		if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
564 			data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
565 		else
566 			data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
567 		data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
568 		data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
569 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
570 
571 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_GATE);
572 		data &= ~(UVD_CGC_GATE__SYS_MASK
573 			| UVD_CGC_GATE__UDEC_MASK
574 			| UVD_CGC_GATE__MPEG2_MASK
575 			| UVD_CGC_GATE__REGS_MASK
576 			| UVD_CGC_GATE__RBC_MASK
577 			| UVD_CGC_GATE__LMI_MC_MASK
578 			| UVD_CGC_GATE__LMI_UMC_MASK
579 			| UVD_CGC_GATE__IDCT_MASK
580 			| UVD_CGC_GATE__MPRD_MASK
581 			| UVD_CGC_GATE__MPC_MASK
582 			| UVD_CGC_GATE__LBSI_MASK
583 			| UVD_CGC_GATE__LRBBM_MASK
584 			| UVD_CGC_GATE__UDEC_RE_MASK
585 			| UVD_CGC_GATE__UDEC_CM_MASK
586 			| UVD_CGC_GATE__UDEC_IT_MASK
587 			| UVD_CGC_GATE__UDEC_DB_MASK
588 			| UVD_CGC_GATE__UDEC_MP_MASK
589 			| UVD_CGC_GATE__WCB_MASK
590 			| UVD_CGC_GATE__VCPU_MASK
591 			| UVD_CGC_GATE__MMSCH_MASK);
592 
593 		WREG32_SOC15(VCN, i, mmUVD_CGC_GATE, data);
594 
595 		SOC15_WAIT_ON_RREG(VCN, i, mmUVD_CGC_GATE, 0,  0xFFFFFFFF);
596 
597 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
598 		data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
599 			| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
600 			| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
601 			| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
602 			| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
603 			| UVD_CGC_CTRL__SYS_MODE_MASK
604 			| UVD_CGC_CTRL__UDEC_MODE_MASK
605 			| UVD_CGC_CTRL__MPEG2_MODE_MASK
606 			| UVD_CGC_CTRL__REGS_MODE_MASK
607 			| UVD_CGC_CTRL__RBC_MODE_MASK
608 			| UVD_CGC_CTRL__LMI_MC_MODE_MASK
609 			| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
610 			| UVD_CGC_CTRL__IDCT_MODE_MASK
611 			| UVD_CGC_CTRL__MPRD_MODE_MASK
612 			| UVD_CGC_CTRL__MPC_MODE_MASK
613 			| UVD_CGC_CTRL__LBSI_MODE_MASK
614 			| UVD_CGC_CTRL__LRBBM_MODE_MASK
615 			| UVD_CGC_CTRL__WCB_MODE_MASK
616 			| UVD_CGC_CTRL__VCPU_MODE_MASK
617 			| UVD_CGC_CTRL__MMSCH_MODE_MASK);
618 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
619 
620 		/* turn on */
621 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE);
622 		data |= (UVD_SUVD_CGC_GATE__SRE_MASK
623 			| UVD_SUVD_CGC_GATE__SIT_MASK
624 			| UVD_SUVD_CGC_GATE__SMP_MASK
625 			| UVD_SUVD_CGC_GATE__SCM_MASK
626 			| UVD_SUVD_CGC_GATE__SDB_MASK
627 			| UVD_SUVD_CGC_GATE__SRE_H264_MASK
628 			| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
629 			| UVD_SUVD_CGC_GATE__SIT_H264_MASK
630 			| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
631 			| UVD_SUVD_CGC_GATE__SCM_H264_MASK
632 			| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
633 			| UVD_SUVD_CGC_GATE__SDB_H264_MASK
634 			| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
635 			| UVD_SUVD_CGC_GATE__SCLR_MASK
636 			| UVD_SUVD_CGC_GATE__UVD_SC_MASK
637 			| UVD_SUVD_CGC_GATE__ENT_MASK
638 			| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
639 			| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
640 			| UVD_SUVD_CGC_GATE__SITE_MASK
641 			| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
642 			| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
643 			| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
644 			| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
645 			| UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
646 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE, data);
647 
648 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL);
649 		data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
650 			| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
651 			| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
652 			| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
653 			| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
654 			| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
655 			| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
656 			| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
657 			| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
658 			| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
659 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data);
660 	}
661 }
662 
663 static void vcn_v2_5_clock_gating_dpg_mode(struct amdgpu_device *adev,
664 		uint8_t sram_sel, int inst_idx, uint8_t indirect)
665 {
666 	uint32_t reg_data = 0;
667 
668 	/* enable sw clock gating control */
669 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
670 		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
671 	else
672 		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
673 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
674 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
675 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
676 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
677 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
678 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
679 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
680 		 UVD_CGC_CTRL__SYS_MODE_MASK |
681 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
682 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
683 		 UVD_CGC_CTRL__REGS_MODE_MASK |
684 		 UVD_CGC_CTRL__RBC_MODE_MASK |
685 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
686 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
687 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
688 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
689 		 UVD_CGC_CTRL__MPC_MODE_MASK |
690 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
691 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
692 		 UVD_CGC_CTRL__WCB_MODE_MASK |
693 		 UVD_CGC_CTRL__VCPU_MODE_MASK |
694 		 UVD_CGC_CTRL__MMSCH_MODE_MASK);
695 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
696 		VCN, 0, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
697 
698 	/* turn off clock gating */
699 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
700 		VCN, 0, mmUVD_CGC_GATE), 0, sram_sel, indirect);
701 
702 	/* turn on SUVD clock gating */
703 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
704 		VCN, 0, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
705 
706 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
707 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
708 		VCN, 0, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
709 }
710 
711 /**
712  * vcn_v2_5_enable_clock_gating - enable VCN clock gating
713  *
714  * @adev: amdgpu_device pointer
715  *
716  * Enable clock gating for VCN block
717  */
718 static void vcn_v2_5_enable_clock_gating(struct amdgpu_device *adev)
719 {
720 	uint32_t data = 0;
721 	int i;
722 
723 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
724 		if (adev->vcn.harvest_config & (1 << i))
725 			continue;
726 		/* enable UVD CGC */
727 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
728 		if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
729 			data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
730 		else
731 			data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
732 		data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
733 		data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
734 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
735 
736 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
737 		data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
738 			| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
739 			| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
740 			| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
741 			| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
742 			| UVD_CGC_CTRL__SYS_MODE_MASK
743 			| UVD_CGC_CTRL__UDEC_MODE_MASK
744 			| UVD_CGC_CTRL__MPEG2_MODE_MASK
745 			| UVD_CGC_CTRL__REGS_MODE_MASK
746 			| UVD_CGC_CTRL__RBC_MODE_MASK
747 			| UVD_CGC_CTRL__LMI_MC_MODE_MASK
748 			| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
749 			| UVD_CGC_CTRL__IDCT_MODE_MASK
750 			| UVD_CGC_CTRL__MPRD_MODE_MASK
751 			| UVD_CGC_CTRL__MPC_MODE_MASK
752 			| UVD_CGC_CTRL__LBSI_MODE_MASK
753 			| UVD_CGC_CTRL__LRBBM_MODE_MASK
754 			| UVD_CGC_CTRL__WCB_MODE_MASK
755 			| UVD_CGC_CTRL__VCPU_MODE_MASK);
756 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
757 
758 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL);
759 		data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
760 			| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
761 			| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
762 			| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
763 			| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
764 			| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
765 			| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
766 			| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
767 			| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
768 			| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
769 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data);
770 	}
771 }
772 
773 static void vcn_v2_6_enable_ras(struct amdgpu_device *adev, int inst_idx,
774 				bool indirect)
775 {
776 	uint32_t tmp;
777 
778 	if (adev->ip_versions[UVD_HWIP][0] != IP_VERSION(2, 6, 0))
779 		return;
780 
781 	tmp = VCN_RAS_CNTL__VCPU_VCODEC_REARM_MASK |
782 	      VCN_RAS_CNTL__VCPU_VCODEC_IH_EN_MASK |
783 	      VCN_RAS_CNTL__VCPU_VCODEC_PMI_EN_MASK |
784 	      VCN_RAS_CNTL__VCPU_VCODEC_STALL_EN_MASK;
785 	WREG32_SOC15_DPG_MODE(inst_idx,
786 			      SOC15_DPG_MODE_OFFSET(VCN, 0, mmVCN_RAS_CNTL),
787 			      tmp, 0, indirect);
788 
789 	tmp = UVD_VCPU_INT_EN__RASCNTL_VCPU_VCODEC_EN_MASK;
790 	WREG32_SOC15_DPG_MODE(inst_idx,
791 			      SOC15_DPG_MODE_OFFSET(VCN, 0, mmUVD_VCPU_INT_EN),
792 			      tmp, 0, indirect);
793 
794 	tmp = UVD_SYS_INT_EN__RASCNTL_VCPU_VCODEC_EN_MASK;
795 	WREG32_SOC15_DPG_MODE(inst_idx,
796 			      SOC15_DPG_MODE_OFFSET(VCN, 0, mmUVD_SYS_INT_EN),
797 			      tmp, 0, indirect);
798 }
799 
800 static int vcn_v2_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
801 {
802 	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
803 	struct amdgpu_ring *ring;
804 	uint32_t rb_bufsz, tmp;
805 
806 	/* disable register anti-hang mechanism */
807 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1,
808 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
809 	/* enable dynamic power gating mode */
810 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS);
811 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
812 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
813 	WREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS, tmp);
814 
815 	if (indirect)
816 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
817 
818 	/* enable clock gating */
819 	vcn_v2_5_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
820 
821 	/* enable VCPU clock */
822 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
823 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
824 	tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
825 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
826 		VCN, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
827 
828 	/* disable master interupt */
829 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
830 		VCN, 0, mmUVD_MASTINT_EN), 0, 0, indirect);
831 
832 	/* setup mmUVD_LMI_CTRL */
833 	tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
834 		UVD_LMI_CTRL__REQ_MODE_MASK |
835 		UVD_LMI_CTRL__CRC_RESET_MASK |
836 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
837 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
838 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
839 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
840 		0x00100000L);
841 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
842 		VCN, 0, mmUVD_LMI_CTRL), tmp, 0, indirect);
843 
844 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
845 		VCN, 0, mmUVD_MPC_CNTL),
846 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
847 
848 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
849 		VCN, 0, mmUVD_MPC_SET_MUXA0),
850 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
851 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
852 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
853 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
854 
855 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
856 		VCN, 0, mmUVD_MPC_SET_MUXB0),
857 		((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
858 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
859 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
860 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
861 
862 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
863 		VCN, 0, mmUVD_MPC_SET_MUX),
864 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
865 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
866 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
867 
868 	vcn_v2_5_mc_resume_dpg_mode(adev, inst_idx, indirect);
869 
870 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
871 		VCN, 0, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
872 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
873 		VCN, 0, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
874 
875 	/* enable LMI MC and UMC channels */
876 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
877 		VCN, 0, mmUVD_LMI_CTRL2), 0, 0, indirect);
878 
879 	vcn_v2_6_enable_ras(adev, inst_idx, indirect);
880 
881 	/* unblock VCPU register access */
882 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
883 		VCN, 0, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
884 
885 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
886 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
887 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
888 		VCN, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
889 
890 	/* enable master interrupt */
891 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
892 		VCN, 0, mmUVD_MASTINT_EN),
893 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
894 
895 	if (indirect)
896 		psp_update_vcn_sram(adev, inst_idx, adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
897 				    (uint32_t)((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
898 					       (uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr));
899 
900 	ring = &adev->vcn.inst[inst_idx].ring_dec;
901 	/* force RBC into idle state */
902 	rb_bufsz = order_base_2(ring->ring_size);
903 	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
904 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
905 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
906 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
907 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
908 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
909 
910 	/* Stall DPG before WPTR/RPTR reset */
911 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
912 		UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
913 		~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
914 	fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
915 
916 	/* set the write pointer delay */
917 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
918 
919 	/* set the wb address */
920 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR_ADDR,
921 		(upper_32_bits(ring->gpu_addr) >> 2));
922 
923 	/* program the RB_BASE for ring buffer */
924 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
925 		lower_32_bits(ring->gpu_addr));
926 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
927 		upper_32_bits(ring->gpu_addr));
928 
929 	/* Initialize the ring buffer's read and write pointers */
930 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, 0);
931 
932 	WREG32_SOC15(VCN, inst_idx, mmUVD_SCRATCH2, 0);
933 
934 	ring->wptr = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR);
935 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR,
936 		lower_32_bits(ring->wptr));
937 
938 	fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
939 	/* Unstall DPG */
940 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
941 		0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
942 
943 	return 0;
944 }
945 
946 static int vcn_v2_5_start(struct amdgpu_device *adev)
947 {
948 	struct amdgpu_ring *ring;
949 	uint32_t rb_bufsz, tmp;
950 	int i, j, k, r;
951 
952 	if (adev->pm.dpm_enabled)
953 		amdgpu_dpm_enable_uvd(adev, true);
954 
955 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
956 		if (adev->vcn.harvest_config & (1 << i))
957 			continue;
958 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
959 			r = vcn_v2_5_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
960 			continue;
961 		}
962 
963 		/* disable register anti-hang mechanism */
964 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_POWER_STATUS), 0,
965 			~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
966 
967 		/* set uvd status busy */
968 		tmp = RREG32_SOC15(VCN, i, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY;
969 		WREG32_SOC15(VCN, i, mmUVD_STATUS, tmp);
970 	}
971 
972 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
973 		return 0;
974 
975 	/*SW clock gating */
976 	vcn_v2_5_disable_clock_gating(adev);
977 
978 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
979 		if (adev->vcn.harvest_config & (1 << i))
980 			continue;
981 		/* enable VCPU clock */
982 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
983 			UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
984 
985 		/* disable master interrupt */
986 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN), 0,
987 			~UVD_MASTINT_EN__VCPU_EN_MASK);
988 
989 		/* setup mmUVD_LMI_CTRL */
990 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL);
991 		tmp &= ~0xff;
992 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL, tmp | 0x8|
993 			UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK	|
994 			UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
995 			UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
996 			UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
997 
998 		/* setup mmUVD_MPC_CNTL */
999 		tmp = RREG32_SOC15(VCN, i, mmUVD_MPC_CNTL);
1000 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1001 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1002 		WREG32_SOC15(VCN, i, mmUVD_MPC_CNTL, tmp);
1003 
1004 		/* setup UVD_MPC_SET_MUXA0 */
1005 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXA0,
1006 			((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1007 			(0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1008 			(0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1009 			(0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1010 
1011 		/* setup UVD_MPC_SET_MUXB0 */
1012 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXB0,
1013 			((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1014 			(0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1015 			(0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1016 			(0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1017 
1018 		/* setup mmUVD_MPC_SET_MUX */
1019 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUX,
1020 			((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1021 			(0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1022 			(0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1023 	}
1024 
1025 	vcn_v2_5_mc_resume(adev);
1026 
1027 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1028 		volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1029 		if (adev->vcn.harvest_config & (1 << i))
1030 			continue;
1031 		/* VCN global tiling registers */
1032 		WREG32_SOC15(VCN, i, mmUVD_GFX8_ADDR_CONFIG,
1033 			adev->gfx.config.gb_addr_config);
1034 		WREG32_SOC15(VCN, i, mmUVD_GFX8_ADDR_CONFIG,
1035 			adev->gfx.config.gb_addr_config);
1036 
1037 		/* enable LMI MC and UMC channels */
1038 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_LMI_CTRL2), 0,
1039 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1040 
1041 		/* unblock VCPU register access */
1042 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL), 0,
1043 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1044 
1045 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1046 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1047 
1048 		for (k = 0; k < 10; ++k) {
1049 			uint32_t status;
1050 
1051 			for (j = 0; j < 100; ++j) {
1052 				status = RREG32_SOC15(VCN, i, mmUVD_STATUS);
1053 				if (status & 2)
1054 					break;
1055 				if (amdgpu_emu_mode == 1)
1056 					msleep(500);
1057 				else
1058 					mdelay(10);
1059 			}
1060 			r = 0;
1061 			if (status & 2)
1062 				break;
1063 
1064 			DRM_ERROR("VCN decode not responding, trying to reset the VCPU!!!\n");
1065 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1066 				UVD_VCPU_CNTL__BLK_RST_MASK,
1067 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1068 			mdelay(10);
1069 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1070 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1071 
1072 			mdelay(10);
1073 			r = -1;
1074 		}
1075 
1076 		if (r) {
1077 			DRM_ERROR("VCN decode not responding, giving up!!!\n");
1078 			return r;
1079 		}
1080 
1081 		/* enable master interrupt */
1082 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN),
1083 			UVD_MASTINT_EN__VCPU_EN_MASK,
1084 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1085 
1086 		/* clear the busy bit of VCN_STATUS */
1087 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS), 0,
1088 			~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1089 
1090 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_VMID, 0);
1091 
1092 		ring = &adev->vcn.inst[i].ring_dec;
1093 		/* force RBC into idle state */
1094 		rb_bufsz = order_base_2(ring->ring_size);
1095 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1096 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1097 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1098 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1099 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1100 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_CNTL, tmp);
1101 
1102 		fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
1103 		/* program the RB_BASE for ring buffer */
1104 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1105 			lower_32_bits(ring->gpu_addr));
1106 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1107 			upper_32_bits(ring->gpu_addr));
1108 
1109 		/* Initialize the ring buffer's read and write pointers */
1110 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR, 0);
1111 
1112 		ring->wptr = RREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR);
1113 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_WPTR,
1114 				lower_32_bits(ring->wptr));
1115 		fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
1116 
1117 		fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1118 		ring = &adev->vcn.inst[i].ring_enc[0];
1119 		WREG32_SOC15(VCN, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1120 		WREG32_SOC15(VCN, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1121 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO, ring->gpu_addr);
1122 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1123 		WREG32_SOC15(VCN, i, mmUVD_RB_SIZE, ring->ring_size / 4);
1124 		fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1125 
1126 		fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1127 		ring = &adev->vcn.inst[i].ring_enc[1];
1128 		WREG32_SOC15(VCN, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1129 		WREG32_SOC15(VCN, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1130 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1131 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1132 		WREG32_SOC15(VCN, i, mmUVD_RB_SIZE2, ring->ring_size / 4);
1133 		fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1134 	}
1135 
1136 	return 0;
1137 }
1138 
1139 static int vcn_v2_5_mmsch_start(struct amdgpu_device *adev,
1140 				struct amdgpu_mm_table *table)
1141 {
1142 	uint32_t data = 0, loop = 0, size = 0;
1143 	uint64_t addr = table->gpu_addr;
1144 	struct mmsch_v1_1_init_header *header = NULL;
1145 
1146 	header = (struct mmsch_v1_1_init_header *)table->cpu_addr;
1147 	size = header->total_size;
1148 
1149 	/*
1150 	 * 1, write to vce_mmsch_vf_ctx_addr_lo/hi register with GPU mc addr of
1151 	 *  memory descriptor location
1152 	 */
1153 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(addr));
1154 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(addr));
1155 
1156 	/* 2, update vmid of descriptor */
1157 	data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID);
1158 	data &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1159 	/* use domain0 for MM scheduler */
1160 	data |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1161 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID, data);
1162 
1163 	/* 3, notify mmsch about the size of this descriptor */
1164 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_SIZE, size);
1165 
1166 	/* 4, set resp to zero */
1167 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1168 
1169 	/*
1170 	 * 5, kick off the initialization and wait until
1171 	 * VCE_MMSCH_VF_MAILBOX_RESP becomes non-zero
1172 	 */
1173 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_HOST, 0x10000001);
1174 
1175 	data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1176 	loop = 10;
1177 	while ((data & 0x10000002) != 0x10000002) {
1178 		udelay(100);
1179 		data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1180 		loop--;
1181 		if (!loop)
1182 			break;
1183 	}
1184 
1185 	if (!loop) {
1186 		dev_err(adev->dev,
1187 			"failed to init MMSCH, mmMMSCH_VF_MAILBOX_RESP = %x\n",
1188 			data);
1189 		return -EBUSY;
1190 	}
1191 
1192 	return 0;
1193 }
1194 
1195 static int vcn_v2_5_sriov_start(struct amdgpu_device *adev)
1196 {
1197 	struct amdgpu_ring *ring;
1198 	uint32_t offset, size, tmp, i, rb_bufsz;
1199 	uint32_t table_size = 0;
1200 	struct mmsch_v1_0_cmd_direct_write direct_wt = { { 0 } };
1201 	struct mmsch_v1_0_cmd_direct_read_modify_write direct_rd_mod_wt = { { 0 } };
1202 	struct mmsch_v1_0_cmd_end end = { { 0 } };
1203 	uint32_t *init_table = adev->virt.mm_table.cpu_addr;
1204 	struct mmsch_v1_1_init_header *header = (struct mmsch_v1_1_init_header *)init_table;
1205 
1206 	direct_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_WRITE;
1207 	direct_rd_mod_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1208 	end.cmd_header.command_type = MMSCH_COMMAND__END;
1209 
1210 	header->version = MMSCH_VERSION;
1211 	header->total_size = sizeof(struct mmsch_v1_1_init_header) >> 2;
1212 	init_table += header->total_size;
1213 
1214 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1215 		header->eng[i].table_offset = header->total_size;
1216 		header->eng[i].init_status = 0;
1217 		header->eng[i].table_size = 0;
1218 
1219 		table_size = 0;
1220 
1221 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(
1222 			SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS),
1223 			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1224 
1225 		size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1226 		/* mc resume*/
1227 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1228 			MMSCH_V1_0_INSERT_DIRECT_WT(
1229 				SOC15_REG_OFFSET(VCN, i,
1230 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1231 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1232 			MMSCH_V1_0_INSERT_DIRECT_WT(
1233 				SOC15_REG_OFFSET(VCN, i,
1234 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1235 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1236 			offset = 0;
1237 			MMSCH_V1_0_INSERT_DIRECT_WT(
1238 				SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET0), 0);
1239 		} else {
1240 			MMSCH_V1_0_INSERT_DIRECT_WT(
1241 				SOC15_REG_OFFSET(VCN, i,
1242 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1243 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1244 			MMSCH_V1_0_INSERT_DIRECT_WT(
1245 				SOC15_REG_OFFSET(VCN, i,
1246 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1247 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1248 			offset = size;
1249 			MMSCH_V1_0_INSERT_DIRECT_WT(
1250 				SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET0),
1251 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1252 		}
1253 
1254 		MMSCH_V1_0_INSERT_DIRECT_WT(
1255 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE0),
1256 			size);
1257 		MMSCH_V1_0_INSERT_DIRECT_WT(
1258 			SOC15_REG_OFFSET(VCN, i,
1259 				mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1260 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset));
1261 		MMSCH_V1_0_INSERT_DIRECT_WT(
1262 			SOC15_REG_OFFSET(VCN, i,
1263 				mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1264 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset));
1265 		MMSCH_V1_0_INSERT_DIRECT_WT(
1266 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET1),
1267 			0);
1268 		MMSCH_V1_0_INSERT_DIRECT_WT(
1269 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE1),
1270 			AMDGPU_VCN_STACK_SIZE);
1271 		MMSCH_V1_0_INSERT_DIRECT_WT(
1272 			SOC15_REG_OFFSET(VCN, i,
1273 				mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1274 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset +
1275 				AMDGPU_VCN_STACK_SIZE));
1276 		MMSCH_V1_0_INSERT_DIRECT_WT(
1277 			SOC15_REG_OFFSET(VCN, i,
1278 				mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1279 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset +
1280 				AMDGPU_VCN_STACK_SIZE));
1281 		MMSCH_V1_0_INSERT_DIRECT_WT(
1282 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET2),
1283 			0);
1284 		MMSCH_V1_0_INSERT_DIRECT_WT(
1285 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE2),
1286 			AMDGPU_VCN_CONTEXT_SIZE);
1287 
1288 		ring = &adev->vcn.inst[i].ring_enc[0];
1289 		ring->wptr = 0;
1290 
1291 		MMSCH_V1_0_INSERT_DIRECT_WT(
1292 			SOC15_REG_OFFSET(VCN, i, mmUVD_RB_BASE_LO),
1293 			lower_32_bits(ring->gpu_addr));
1294 		MMSCH_V1_0_INSERT_DIRECT_WT(
1295 			SOC15_REG_OFFSET(VCN, i, mmUVD_RB_BASE_HI),
1296 			upper_32_bits(ring->gpu_addr));
1297 		MMSCH_V1_0_INSERT_DIRECT_WT(
1298 			SOC15_REG_OFFSET(VCN, i, mmUVD_RB_SIZE),
1299 			ring->ring_size / 4);
1300 
1301 		ring = &adev->vcn.inst[i].ring_dec;
1302 		ring->wptr = 0;
1303 		MMSCH_V1_0_INSERT_DIRECT_WT(
1304 			SOC15_REG_OFFSET(VCN, i,
1305 				mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1306 			lower_32_bits(ring->gpu_addr));
1307 		MMSCH_V1_0_INSERT_DIRECT_WT(
1308 			SOC15_REG_OFFSET(VCN, i,
1309 				mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1310 			upper_32_bits(ring->gpu_addr));
1311 
1312 		/* force RBC into idle state */
1313 		rb_bufsz = order_base_2(ring->ring_size);
1314 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1315 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1316 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1317 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1318 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1319 		MMSCH_V1_0_INSERT_DIRECT_WT(
1320 			SOC15_REG_OFFSET(VCN, i, mmUVD_RBC_RB_CNTL), tmp);
1321 
1322 		/* add end packet */
1323 		memcpy((void *)init_table, &end, sizeof(struct mmsch_v1_0_cmd_end));
1324 		table_size += sizeof(struct mmsch_v1_0_cmd_end) / 4;
1325 		init_table += sizeof(struct mmsch_v1_0_cmd_end) / 4;
1326 
1327 		/* refine header */
1328 		header->eng[i].table_size = table_size;
1329 		header->total_size += table_size;
1330 	}
1331 
1332 	return vcn_v2_5_mmsch_start(adev, &adev->virt.mm_table);
1333 }
1334 
1335 static int vcn_v2_5_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1336 {
1337 	uint32_t tmp;
1338 
1339 	/* Wait for power status to be 1 */
1340 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1341 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1342 
1343 	/* wait for read ptr to be equal to write ptr */
1344 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR);
1345 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1346 
1347 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2);
1348 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
1349 
1350 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1351 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF);
1352 
1353 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1354 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1355 
1356 	/* disable dynamic power gating mode */
1357 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 0,
1358 			~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1359 
1360 	return 0;
1361 }
1362 
1363 static int vcn_v2_5_stop(struct amdgpu_device *adev)
1364 {
1365 	uint32_t tmp;
1366 	int i, r = 0;
1367 
1368 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1369 		if (adev->vcn.harvest_config & (1 << i))
1370 			continue;
1371 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1372 			r = vcn_v2_5_stop_dpg_mode(adev, i);
1373 			continue;
1374 		}
1375 
1376 		/* wait for vcn idle */
1377 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1378 		if (r)
1379 			return r;
1380 
1381 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1382 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1383 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1384 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1385 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1386 		if (r)
1387 			return r;
1388 
1389 		/* block LMI UMC channel */
1390 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2);
1391 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1392 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2, tmp);
1393 
1394 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1395 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1396 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1397 		if (r)
1398 			return r;
1399 
1400 		/* block VCPU register access */
1401 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL),
1402 			UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1403 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1404 
1405 		/* reset VCPU */
1406 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1407 			UVD_VCPU_CNTL__BLK_RST_MASK,
1408 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1409 
1410 		/* disable VCPU clock */
1411 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1412 			~(UVD_VCPU_CNTL__CLK_EN_MASK));
1413 
1414 		/* clear status */
1415 		WREG32_SOC15(VCN, i, mmUVD_STATUS, 0);
1416 
1417 		vcn_v2_5_enable_clock_gating(adev);
1418 
1419 		/* enable register anti-hang mechanism */
1420 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_POWER_STATUS),
1421 			UVD_POWER_STATUS__UVD_POWER_STATUS_MASK,
1422 			~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1423 	}
1424 
1425 	if (adev->pm.dpm_enabled)
1426 		amdgpu_dpm_enable_uvd(adev, false);
1427 
1428 	return 0;
1429 }
1430 
1431 static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
1432 				int inst_idx, struct dpg_pause_state *new_state)
1433 {
1434 	struct amdgpu_ring *ring;
1435 	uint32_t reg_data = 0;
1436 	int ret_code = 0;
1437 
1438 	/* pause/unpause if state is changed */
1439 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1440 		DRM_DEBUG("dpg pause state changed %d -> %d",
1441 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1442 		reg_data = RREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE) &
1443 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1444 
1445 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1446 			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1447 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1448 
1449 			if (!ret_code) {
1450 				volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
1451 
1452 				/* pause DPG */
1453 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1454 				WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1455 
1456 				/* wait for ACK */
1457 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_DPG_PAUSE,
1458 					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1459 					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1460 
1461 				/* Stall DPG before WPTR/RPTR reset */
1462 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1463 					   UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1464 					   ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1465 
1466 				/* Restore */
1467 				fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1468 				ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1469 				ring->wptr = 0;
1470 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
1471 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1472 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
1473 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1474 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1475 				fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1476 
1477 				fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1478 				ring = &adev->vcn.inst[inst_idx].ring_enc[1];
1479 				ring->wptr = 0;
1480 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1481 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1482 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
1483 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1484 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1485 				fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1486 
1487 				/* Unstall DPG */
1488 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1489 					   0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1490 
1491 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS,
1492 					   UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1493 			}
1494 		} else {
1495 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1496 			WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1497 			SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1498 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1499 		}
1500 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1501 	}
1502 
1503 	return 0;
1504 }
1505 
1506 /**
1507  * vcn_v2_5_dec_ring_get_rptr - get read pointer
1508  *
1509  * @ring: amdgpu_ring pointer
1510  *
1511  * Returns the current hardware read pointer
1512  */
1513 static uint64_t vcn_v2_5_dec_ring_get_rptr(struct amdgpu_ring *ring)
1514 {
1515 	struct amdgpu_device *adev = ring->adev;
1516 
1517 	return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_RPTR);
1518 }
1519 
1520 /**
1521  * vcn_v2_5_dec_ring_get_wptr - get write pointer
1522  *
1523  * @ring: amdgpu_ring pointer
1524  *
1525  * Returns the current hardware write pointer
1526  */
1527 static uint64_t vcn_v2_5_dec_ring_get_wptr(struct amdgpu_ring *ring)
1528 {
1529 	struct amdgpu_device *adev = ring->adev;
1530 
1531 	if (ring->use_doorbell)
1532 		return *ring->wptr_cpu_addr;
1533 	else
1534 		return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR);
1535 }
1536 
1537 /**
1538  * vcn_v2_5_dec_ring_set_wptr - set write pointer
1539  *
1540  * @ring: amdgpu_ring pointer
1541  *
1542  * Commits the write pointer to the hardware
1543  */
1544 static void vcn_v2_5_dec_ring_set_wptr(struct amdgpu_ring *ring)
1545 {
1546 	struct amdgpu_device *adev = ring->adev;
1547 
1548 	if (ring->use_doorbell) {
1549 		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1550 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1551 	} else {
1552 		WREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1553 	}
1554 }
1555 
1556 static const struct amdgpu_ring_funcs vcn_v2_5_dec_ring_vm_funcs = {
1557 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1558 	.align_mask = 0xf,
1559 	.secure_submission_supported = true,
1560 	.vmhub = AMDGPU_MMHUB_1,
1561 	.get_rptr = vcn_v2_5_dec_ring_get_rptr,
1562 	.get_wptr = vcn_v2_5_dec_ring_get_wptr,
1563 	.set_wptr = vcn_v2_5_dec_ring_set_wptr,
1564 	.emit_frame_size =
1565 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1566 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1567 		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1568 		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1569 		6,
1570 	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1571 	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
1572 	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
1573 	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1574 	.test_ring = vcn_v2_0_dec_ring_test_ring,
1575 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
1576 	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
1577 	.insert_start = vcn_v2_0_dec_ring_insert_start,
1578 	.insert_end = vcn_v2_0_dec_ring_insert_end,
1579 	.pad_ib = amdgpu_ring_generic_pad_ib,
1580 	.begin_use = amdgpu_vcn_ring_begin_use,
1581 	.end_use = amdgpu_vcn_ring_end_use,
1582 	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1583 	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1584 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1585 };
1586 
1587 static const struct amdgpu_ring_funcs vcn_v2_6_dec_ring_vm_funcs = {
1588 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1589 	.align_mask = 0xf,
1590 	.secure_submission_supported = true,
1591 	.vmhub = AMDGPU_MMHUB_0,
1592 	.get_rptr = vcn_v2_5_dec_ring_get_rptr,
1593 	.get_wptr = vcn_v2_5_dec_ring_get_wptr,
1594 	.set_wptr = vcn_v2_5_dec_ring_set_wptr,
1595 	.emit_frame_size =
1596 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1597 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1598 		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1599 		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1600 		6,
1601 	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1602 	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
1603 	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
1604 	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1605 	.test_ring = vcn_v2_0_dec_ring_test_ring,
1606 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
1607 	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
1608 	.insert_start = vcn_v2_0_dec_ring_insert_start,
1609 	.insert_end = vcn_v2_0_dec_ring_insert_end,
1610 	.pad_ib = amdgpu_ring_generic_pad_ib,
1611 	.begin_use = amdgpu_vcn_ring_begin_use,
1612 	.end_use = amdgpu_vcn_ring_end_use,
1613 	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1614 	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1615 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1616 };
1617 
1618 /**
1619  * vcn_v2_5_enc_ring_get_rptr - get enc read pointer
1620  *
1621  * @ring: amdgpu_ring pointer
1622  *
1623  * Returns the current hardware enc read pointer
1624  */
1625 static uint64_t vcn_v2_5_enc_ring_get_rptr(struct amdgpu_ring *ring)
1626 {
1627 	struct amdgpu_device *adev = ring->adev;
1628 
1629 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0])
1630 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR);
1631 	else
1632 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR2);
1633 }
1634 
1635 /**
1636  * vcn_v2_5_enc_ring_get_wptr - get enc write pointer
1637  *
1638  * @ring: amdgpu_ring pointer
1639  *
1640  * Returns the current hardware enc write pointer
1641  */
1642 static uint64_t vcn_v2_5_enc_ring_get_wptr(struct amdgpu_ring *ring)
1643 {
1644 	struct amdgpu_device *adev = ring->adev;
1645 
1646 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1647 		if (ring->use_doorbell)
1648 			return *ring->wptr_cpu_addr;
1649 		else
1650 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR);
1651 	} else {
1652 		if (ring->use_doorbell)
1653 			return *ring->wptr_cpu_addr;
1654 		else
1655 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2);
1656 	}
1657 }
1658 
1659 /**
1660  * vcn_v2_5_enc_ring_set_wptr - set enc write pointer
1661  *
1662  * @ring: amdgpu_ring pointer
1663  *
1664  * Commits the enc write pointer to the hardware
1665  */
1666 static void vcn_v2_5_enc_ring_set_wptr(struct amdgpu_ring *ring)
1667 {
1668 	struct amdgpu_device *adev = ring->adev;
1669 
1670 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1671 		if (ring->use_doorbell) {
1672 			*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1673 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1674 		} else {
1675 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1676 		}
1677 	} else {
1678 		if (ring->use_doorbell) {
1679 			*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1680 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1681 		} else {
1682 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1683 		}
1684 	}
1685 }
1686 
1687 static const struct amdgpu_ring_funcs vcn_v2_5_enc_ring_vm_funcs = {
1688 	.type = AMDGPU_RING_TYPE_VCN_ENC,
1689 	.align_mask = 0x3f,
1690 	.nop = VCN_ENC_CMD_NO_OP,
1691 	.vmhub = AMDGPU_MMHUB_1,
1692 	.get_rptr = vcn_v2_5_enc_ring_get_rptr,
1693 	.get_wptr = vcn_v2_5_enc_ring_get_wptr,
1694 	.set_wptr = vcn_v2_5_enc_ring_set_wptr,
1695 	.emit_frame_size =
1696 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1697 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1698 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1699 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1700 		1, /* vcn_v2_0_enc_ring_insert_end */
1701 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1702 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
1703 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
1704 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1705 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
1706 	.test_ib = amdgpu_vcn_enc_ring_test_ib,
1707 	.insert_nop = amdgpu_ring_insert_nop,
1708 	.insert_end = vcn_v2_0_enc_ring_insert_end,
1709 	.pad_ib = amdgpu_ring_generic_pad_ib,
1710 	.begin_use = amdgpu_vcn_ring_begin_use,
1711 	.end_use = amdgpu_vcn_ring_end_use,
1712 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1713 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1714 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1715 };
1716 
1717 static const struct amdgpu_ring_funcs vcn_v2_6_enc_ring_vm_funcs = {
1718         .type = AMDGPU_RING_TYPE_VCN_ENC,
1719         .align_mask = 0x3f,
1720         .nop = VCN_ENC_CMD_NO_OP,
1721         .vmhub = AMDGPU_MMHUB_0,
1722         .get_rptr = vcn_v2_5_enc_ring_get_rptr,
1723         .get_wptr = vcn_v2_5_enc_ring_get_wptr,
1724         .set_wptr = vcn_v2_5_enc_ring_set_wptr,
1725         .emit_frame_size =
1726                 SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1727                 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1728                 4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1729                 5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1730                 1, /* vcn_v2_0_enc_ring_insert_end */
1731         .emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1732         .emit_ib = vcn_v2_0_enc_ring_emit_ib,
1733         .emit_fence = vcn_v2_0_enc_ring_emit_fence,
1734         .emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1735         .test_ring = amdgpu_vcn_enc_ring_test_ring,
1736         .test_ib = amdgpu_vcn_enc_ring_test_ib,
1737         .insert_nop = amdgpu_ring_insert_nop,
1738         .insert_end = vcn_v2_0_enc_ring_insert_end,
1739         .pad_ib = amdgpu_ring_generic_pad_ib,
1740         .begin_use = amdgpu_vcn_ring_begin_use,
1741         .end_use = amdgpu_vcn_ring_end_use,
1742         .emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1743         .emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1744         .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1745 };
1746 
1747 static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev)
1748 {
1749 	int i;
1750 
1751 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1752 		if (adev->vcn.harvest_config & (1 << i))
1753 			continue;
1754 		if (adev->ip_versions[UVD_HWIP][0] == IP_VERSION(2, 5, 0))
1755 			adev->vcn.inst[i].ring_dec.funcs = &vcn_v2_5_dec_ring_vm_funcs;
1756 		else /* CHIP_ALDEBARAN */
1757 			adev->vcn.inst[i].ring_dec.funcs = &vcn_v2_6_dec_ring_vm_funcs;
1758 		adev->vcn.inst[i].ring_dec.me = i;
1759 		DRM_INFO("VCN(%d) decode is enabled in VM mode\n", i);
1760 	}
1761 }
1762 
1763 static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev)
1764 {
1765 	int i, j;
1766 
1767 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
1768 		if (adev->vcn.harvest_config & (1 << j))
1769 			continue;
1770 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
1771 			if (adev->ip_versions[UVD_HWIP][0] == IP_VERSION(2, 5, 0))
1772 				adev->vcn.inst[j].ring_enc[i].funcs = &vcn_v2_5_enc_ring_vm_funcs;
1773 			else /* CHIP_ALDEBARAN */
1774 				adev->vcn.inst[j].ring_enc[i].funcs = &vcn_v2_6_enc_ring_vm_funcs;
1775 			adev->vcn.inst[j].ring_enc[i].me = j;
1776 		}
1777 		DRM_INFO("VCN(%d) encode is enabled in VM mode\n", j);
1778 	}
1779 }
1780 
1781 static bool vcn_v2_5_is_idle(void *handle)
1782 {
1783 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1784 	int i, ret = 1;
1785 
1786 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1787 		if (adev->vcn.harvest_config & (1 << i))
1788 			continue;
1789 		ret &= (RREG32_SOC15(VCN, i, mmUVD_STATUS) == UVD_STATUS__IDLE);
1790 	}
1791 
1792 	return ret;
1793 }
1794 
1795 static int vcn_v2_5_wait_for_idle(void *handle)
1796 {
1797 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1798 	int i, ret = 0;
1799 
1800 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1801 		if (adev->vcn.harvest_config & (1 << i))
1802 			continue;
1803 		ret = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE,
1804 			UVD_STATUS__IDLE);
1805 		if (ret)
1806 			return ret;
1807 	}
1808 
1809 	return ret;
1810 }
1811 
1812 static int vcn_v2_5_set_clockgating_state(void *handle,
1813 					  enum amd_clockgating_state state)
1814 {
1815 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1816 	bool enable = (state == AMD_CG_STATE_GATE);
1817 
1818 	if (amdgpu_sriov_vf(adev))
1819 		return 0;
1820 
1821 	if (enable) {
1822 		if (!vcn_v2_5_is_idle(handle))
1823 			return -EBUSY;
1824 		vcn_v2_5_enable_clock_gating(adev);
1825 	} else {
1826 		vcn_v2_5_disable_clock_gating(adev);
1827 	}
1828 
1829 	return 0;
1830 }
1831 
1832 static int vcn_v2_5_set_powergating_state(void *handle,
1833 					  enum amd_powergating_state state)
1834 {
1835 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1836 	int ret;
1837 
1838 	if (amdgpu_sriov_vf(adev))
1839 		return 0;
1840 
1841 	if(state == adev->vcn.cur_state)
1842 		return 0;
1843 
1844 	if (state == AMD_PG_STATE_GATE)
1845 		ret = vcn_v2_5_stop(adev);
1846 	else
1847 		ret = vcn_v2_5_start(adev);
1848 
1849 	if(!ret)
1850 		adev->vcn.cur_state = state;
1851 
1852 	return ret;
1853 }
1854 
1855 static int vcn_v2_5_set_interrupt_state(struct amdgpu_device *adev,
1856 					struct amdgpu_irq_src *source,
1857 					unsigned type,
1858 					enum amdgpu_interrupt_state state)
1859 {
1860 	return 0;
1861 }
1862 
1863 static int vcn_v2_5_process_interrupt(struct amdgpu_device *adev,
1864 				      struct amdgpu_irq_src *source,
1865 				      struct amdgpu_iv_entry *entry)
1866 {
1867 	uint32_t ip_instance;
1868 
1869 	switch (entry->client_id) {
1870 	case SOC15_IH_CLIENTID_VCN:
1871 		ip_instance = 0;
1872 		break;
1873 	case SOC15_IH_CLIENTID_VCN1:
1874 		ip_instance = 1;
1875 		break;
1876 	default:
1877 		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
1878 		return 0;
1879 	}
1880 
1881 	DRM_DEBUG("IH: VCN TRAP\n");
1882 
1883 	switch (entry->src_id) {
1884 	case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
1885 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec);
1886 		break;
1887 	case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
1888 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
1889 		break;
1890 	case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
1891 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]);
1892 		break;
1893 	case VCN_2_6__SRCID_UVD_POISON:
1894 		amdgpu_vcn_process_poison_irq(adev, source, entry);
1895 		break;
1896 	default:
1897 		DRM_ERROR("Unhandled interrupt: %d %d\n",
1898 			  entry->src_id, entry->src_data[0]);
1899 		break;
1900 	}
1901 
1902 	return 0;
1903 }
1904 
1905 static const struct amdgpu_irq_src_funcs vcn_v2_5_irq_funcs = {
1906 	.set = vcn_v2_5_set_interrupt_state,
1907 	.process = vcn_v2_5_process_interrupt,
1908 };
1909 
1910 static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev)
1911 {
1912 	int i;
1913 
1914 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1915 		if (adev->vcn.harvest_config & (1 << i))
1916 			continue;
1917 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
1918 		adev->vcn.inst[i].irq.funcs = &vcn_v2_5_irq_funcs;
1919 	}
1920 }
1921 
1922 static const struct amd_ip_funcs vcn_v2_5_ip_funcs = {
1923 	.name = "vcn_v2_5",
1924 	.early_init = vcn_v2_5_early_init,
1925 	.late_init = NULL,
1926 	.sw_init = vcn_v2_5_sw_init,
1927 	.sw_fini = vcn_v2_5_sw_fini,
1928 	.hw_init = vcn_v2_5_hw_init,
1929 	.hw_fini = vcn_v2_5_hw_fini,
1930 	.suspend = vcn_v2_5_suspend,
1931 	.resume = vcn_v2_5_resume,
1932 	.is_idle = vcn_v2_5_is_idle,
1933 	.wait_for_idle = vcn_v2_5_wait_for_idle,
1934 	.check_soft_reset = NULL,
1935 	.pre_soft_reset = NULL,
1936 	.soft_reset = NULL,
1937 	.post_soft_reset = NULL,
1938 	.set_clockgating_state = vcn_v2_5_set_clockgating_state,
1939 	.set_powergating_state = vcn_v2_5_set_powergating_state,
1940 };
1941 
1942 static const struct amd_ip_funcs vcn_v2_6_ip_funcs = {
1943         .name = "vcn_v2_6",
1944         .early_init = vcn_v2_5_early_init,
1945         .late_init = NULL,
1946         .sw_init = vcn_v2_5_sw_init,
1947         .sw_fini = vcn_v2_5_sw_fini,
1948         .hw_init = vcn_v2_5_hw_init,
1949         .hw_fini = vcn_v2_5_hw_fini,
1950         .suspend = vcn_v2_5_suspend,
1951         .resume = vcn_v2_5_resume,
1952         .is_idle = vcn_v2_5_is_idle,
1953         .wait_for_idle = vcn_v2_5_wait_for_idle,
1954         .check_soft_reset = NULL,
1955         .pre_soft_reset = NULL,
1956         .soft_reset = NULL,
1957         .post_soft_reset = NULL,
1958         .set_clockgating_state = vcn_v2_5_set_clockgating_state,
1959         .set_powergating_state = vcn_v2_5_set_powergating_state,
1960 };
1961 
1962 const struct amdgpu_ip_block_version vcn_v2_5_ip_block =
1963 {
1964 		.type = AMD_IP_BLOCK_TYPE_VCN,
1965 		.major = 2,
1966 		.minor = 5,
1967 		.rev = 0,
1968 		.funcs = &vcn_v2_5_ip_funcs,
1969 };
1970 
1971 const struct amdgpu_ip_block_version vcn_v2_6_ip_block =
1972 {
1973 		.type = AMD_IP_BLOCK_TYPE_VCN,
1974 		.major = 2,
1975 		.minor = 6,
1976 		.rev = 0,
1977 		.funcs = &vcn_v2_6_ip_funcs,
1978 };
1979 
1980 static uint32_t vcn_v2_6_query_poison_by_instance(struct amdgpu_device *adev,
1981 			uint32_t instance, uint32_t sub_block)
1982 {
1983 	uint32_t poison_stat = 0, reg_value = 0;
1984 
1985 	switch (sub_block) {
1986 	case AMDGPU_VCN_V2_6_VCPU_VCODEC:
1987 		reg_value = RREG32_SOC15(VCN, instance, mmUVD_RAS_VCPU_VCODEC_STATUS);
1988 		poison_stat = REG_GET_FIELD(reg_value, UVD_RAS_VCPU_VCODEC_STATUS, POISONED_PF);
1989 		break;
1990 	default:
1991 		break;
1992 	}
1993 
1994 	if (poison_stat)
1995 		dev_info(adev->dev, "Poison detected in VCN%d, sub_block%d\n",
1996 			instance, sub_block);
1997 
1998 	return poison_stat;
1999 }
2000 
2001 static bool vcn_v2_6_query_poison_status(struct amdgpu_device *adev)
2002 {
2003 	uint32_t inst, sub;
2004 	uint32_t poison_stat = 0;
2005 
2006 	for (inst = 0; inst < adev->vcn.num_vcn_inst; inst++)
2007 		for (sub = 0; sub < AMDGPU_VCN_V2_6_MAX_SUB_BLOCK; sub++)
2008 			poison_stat +=
2009 			vcn_v2_6_query_poison_by_instance(adev, inst, sub);
2010 
2011 	return !!poison_stat;
2012 }
2013 
2014 const struct amdgpu_ras_block_hw_ops vcn_v2_6_ras_hw_ops = {
2015 	.query_poison_status = vcn_v2_6_query_poison_status,
2016 };
2017 
2018 static struct amdgpu_vcn_ras vcn_v2_6_ras = {
2019 	.ras_block = {
2020 		.hw_ops = &vcn_v2_6_ras_hw_ops,
2021 	},
2022 };
2023 
2024 static void vcn_v2_5_set_ras_funcs(struct amdgpu_device *adev)
2025 {
2026 	switch (adev->ip_versions[VCN_HWIP][0]) {
2027 	case IP_VERSION(2, 6, 0):
2028 		adev->vcn.ras = &vcn_v2_6_ras;
2029 		break;
2030 	default:
2031 		break;
2032 	}
2033 
2034 	amdgpu_vcn_set_ras_funcs(adev);
2035 }
2036