xref: /linux/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c (revision 79997eda0d31bc68203c95ecb978773ee6ce7a1f)
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 and load microcode
75  *
76  * @handle: amdgpu_device pointer
77  *
78  * Set ring and irq function pointers
79  * Load microcode from filesystem
80  */
81 static int vcn_v2_5_early_init(void *handle)
82 {
83 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
84 
85 	if (amdgpu_sriov_vf(adev)) {
86 		adev->vcn.num_vcn_inst = 2;
87 		adev->vcn.harvest_config = 0;
88 		adev->vcn.num_enc_rings = 1;
89 	} else {
90 		u32 harvest;
91 		int i;
92 
93 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
94 			harvest = RREG32_SOC15(VCN, i, mmCC_UVD_HARVESTING);
95 			if (harvest & CC_UVD_HARVESTING__UVD_DISABLE_MASK)
96 				adev->vcn.harvest_config |= 1 << i;
97 		}
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 		adev->vcn.num_enc_rings = 2;
104 	}
105 
106 	vcn_v2_5_set_dec_ring_funcs(adev);
107 	vcn_v2_5_set_enc_ring_funcs(adev);
108 	vcn_v2_5_set_irq_funcs(adev);
109 	vcn_v2_5_set_ras_funcs(adev);
110 
111 	return amdgpu_vcn_early_init(adev);
112 }
113 
114 /**
115  * vcn_v2_5_sw_init - sw init for VCN block
116  *
117  * @handle: amdgpu_device pointer
118  *
119  * Load firmware and sw initialization
120  */
121 static int vcn_v2_5_sw_init(void *handle)
122 {
123 	struct amdgpu_ring *ring;
124 	int i, j, r;
125 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
126 
127 	for (j = 0; j < adev->vcn.num_vcn_inst; j++) {
128 		if (adev->vcn.harvest_config & (1 << j))
129 			continue;
130 		/* VCN DEC TRAP */
131 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
132 				VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst[j].irq);
133 		if (r)
134 			return r;
135 
136 		/* VCN ENC TRAP */
137 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
138 			r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
139 				i + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[j].irq);
140 			if (r)
141 				return r;
142 		}
143 
144 		/* VCN POISON TRAP */
145 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
146 			VCN_2_6__SRCID_UVD_POISON, &adev->vcn.inst[j].ras_poison_irq);
147 		if (r)
148 			return r;
149 	}
150 
151 	r = amdgpu_vcn_sw_init(adev);
152 	if (r)
153 		return r;
154 
155 	amdgpu_vcn_setup_ucode(adev);
156 
157 	r = amdgpu_vcn_resume(adev);
158 	if (r)
159 		return r;
160 
161 	for (j = 0; j < adev->vcn.num_vcn_inst; j++) {
162 		volatile struct amdgpu_fw_shared *fw_shared;
163 
164 		if (adev->vcn.harvest_config & (1 << j))
165 			continue;
166 		adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
167 		adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
168 		adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
169 		adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
170 		adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
171 		adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
172 
173 		adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
174 		adev->vcn.inst[j].external.scratch9 = SOC15_REG_OFFSET(VCN, j, mmUVD_SCRATCH9);
175 		adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
176 		adev->vcn.inst[j].external.data0 = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_DATA0);
177 		adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
178 		adev->vcn.inst[j].external.data1 = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_DATA1);
179 		adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
180 		adev->vcn.inst[j].external.cmd = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_CMD);
181 		adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
182 		adev->vcn.inst[j].external.nop = SOC15_REG_OFFSET(VCN, j, mmUVD_NO_OP);
183 
184 		ring = &adev->vcn.inst[j].ring_dec;
185 		ring->use_doorbell = true;
186 
187 		ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
188 				(amdgpu_sriov_vf(adev) ? 2*j : 8*j);
189 
190 		if (amdgpu_ip_version(adev, UVD_HWIP, 0) == IP_VERSION(2, 5, 0))
191 			ring->vm_hub = AMDGPU_MMHUB1(0);
192 		else
193 			ring->vm_hub = AMDGPU_MMHUB0(0);
194 
195 		sprintf(ring->name, "vcn_dec_%d", j);
196 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[j].irq,
197 				     0, AMDGPU_RING_PRIO_DEFAULT, NULL);
198 		if (r)
199 			return r;
200 
201 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
202 			enum amdgpu_ring_priority_level hw_prio = amdgpu_vcn_get_enc_ring_prio(i);
203 
204 			ring = &adev->vcn.inst[j].ring_enc[i];
205 			ring->use_doorbell = true;
206 
207 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
208 					(amdgpu_sriov_vf(adev) ? (1 + i + 2*j) : (2 + i + 8*j));
209 
210 			if (amdgpu_ip_version(adev, UVD_HWIP, 0) ==
211 			    IP_VERSION(2, 5, 0))
212 				ring->vm_hub = AMDGPU_MMHUB1(0);
213 			else
214 				ring->vm_hub = AMDGPU_MMHUB0(0);
215 
216 			sprintf(ring->name, "vcn_enc_%d.%d", j, i);
217 			r = amdgpu_ring_init(adev, ring, 512,
218 					     &adev->vcn.inst[j].irq, 0,
219 					     hw_prio, NULL);
220 			if (r)
221 				return r;
222 		}
223 
224 		fw_shared = adev->vcn.inst[j].fw_shared.cpu_addr;
225 		fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG);
226 
227 		if (amdgpu_vcnfw_log)
228 			amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
229 	}
230 
231 	if (amdgpu_sriov_vf(adev)) {
232 		r = amdgpu_virt_alloc_mm_table(adev);
233 		if (r)
234 			return r;
235 	}
236 
237 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
238 		adev->vcn.pause_dpg_mode = vcn_v2_5_pause_dpg_mode;
239 
240 	r = amdgpu_vcn_ras_sw_init(adev);
241 	if (r)
242 		return r;
243 
244 	return 0;
245 }
246 
247 /**
248  * vcn_v2_5_sw_fini - sw fini for VCN block
249  *
250  * @handle: amdgpu_device pointer
251  *
252  * VCN suspend and free up sw allocation
253  */
254 static int vcn_v2_5_sw_fini(void *handle)
255 {
256 	int i, r, idx;
257 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
258 	volatile struct amdgpu_fw_shared *fw_shared;
259 
260 	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
261 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
262 			if (adev->vcn.harvest_config & (1 << i))
263 				continue;
264 			fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
265 			fw_shared->present_flag_0 = 0;
266 		}
267 		drm_dev_exit(idx);
268 	}
269 
270 
271 	if (amdgpu_sriov_vf(adev))
272 		amdgpu_virt_free_mm_table(adev);
273 
274 	r = amdgpu_vcn_suspend(adev);
275 	if (r)
276 		return r;
277 
278 	r = amdgpu_vcn_sw_fini(adev);
279 
280 	return r;
281 }
282 
283 /**
284  * vcn_v2_5_hw_init - start and test VCN block
285  *
286  * @handle: amdgpu_device pointer
287  *
288  * Initialize the hardware, boot up the VCPU and do some testing
289  */
290 static int vcn_v2_5_hw_init(void *handle)
291 {
292 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
293 	struct amdgpu_ring *ring;
294 	int i, j, r = 0;
295 
296 	if (amdgpu_sriov_vf(adev))
297 		r = vcn_v2_5_sriov_start(adev);
298 
299 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
300 		if (adev->vcn.harvest_config & (1 << j))
301 			continue;
302 
303 		if (amdgpu_sriov_vf(adev)) {
304 			adev->vcn.inst[j].ring_enc[0].sched.ready = true;
305 			adev->vcn.inst[j].ring_enc[1].sched.ready = false;
306 			adev->vcn.inst[j].ring_enc[2].sched.ready = false;
307 			adev->vcn.inst[j].ring_dec.sched.ready = true;
308 		} else {
309 
310 			ring = &adev->vcn.inst[j].ring_dec;
311 
312 			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
313 						     ring->doorbell_index, j);
314 
315 			r = amdgpu_ring_test_helper(ring);
316 			if (r)
317 				goto done;
318 
319 			for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
320 				ring = &adev->vcn.inst[j].ring_enc[i];
321 				r = amdgpu_ring_test_helper(ring);
322 				if (r)
323 					goto done;
324 			}
325 		}
326 	}
327 
328 done:
329 	if (!r)
330 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
331 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
332 
333 	return r;
334 }
335 
336 /**
337  * vcn_v2_5_hw_fini - stop the hardware block
338  *
339  * @handle: amdgpu_device pointer
340  *
341  * Stop the VCN block, mark ring as not ready any more
342  */
343 static int vcn_v2_5_hw_fini(void *handle)
344 {
345 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
346 	int i;
347 
348 	cancel_delayed_work_sync(&adev->vcn.idle_work);
349 
350 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
351 		if (adev->vcn.harvest_config & (1 << i))
352 			continue;
353 
354 		if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
355 		    (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
356 		     RREG32_SOC15(VCN, i, mmUVD_STATUS)))
357 			vcn_v2_5_set_powergating_state(adev, AMD_PG_STATE_GATE);
358 
359 		if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN))
360 			amdgpu_irq_put(adev, &adev->vcn.inst[i].ras_poison_irq, 0);
361 	}
362 
363 	return 0;
364 }
365 
366 /**
367  * vcn_v2_5_suspend - suspend VCN block
368  *
369  * @handle: amdgpu_device pointer
370  *
371  * HW fini and suspend VCN block
372  */
373 static int vcn_v2_5_suspend(void *handle)
374 {
375 	int r;
376 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
377 
378 	r = vcn_v2_5_hw_fini(adev);
379 	if (r)
380 		return r;
381 
382 	r = amdgpu_vcn_suspend(adev);
383 
384 	return r;
385 }
386 
387 /**
388  * vcn_v2_5_resume - resume VCN block
389  *
390  * @handle: amdgpu_device pointer
391  *
392  * Resume firmware and hw init VCN block
393  */
394 static int vcn_v2_5_resume(void *handle)
395 {
396 	int r;
397 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
398 
399 	r = amdgpu_vcn_resume(adev);
400 	if (r)
401 		return r;
402 
403 	r = vcn_v2_5_hw_init(adev);
404 
405 	return r;
406 }
407 
408 /**
409  * vcn_v2_5_mc_resume - memory controller programming
410  *
411  * @adev: amdgpu_device pointer
412  *
413  * Let the VCN memory controller know it's offsets
414  */
415 static void vcn_v2_5_mc_resume(struct amdgpu_device *adev)
416 {
417 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
418 	uint32_t offset;
419 	int i;
420 
421 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
422 		if (adev->vcn.harvest_config & (1 << i))
423 			continue;
424 		/* cache window 0: fw */
425 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
426 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
427 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo));
428 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
429 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi));
430 			WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET0, 0);
431 			offset = 0;
432 		} else {
433 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
434 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
435 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
436 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
437 			offset = size;
438 			WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET0,
439 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
440 		}
441 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE0, size);
442 
443 		/* cache window 1: stack */
444 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
445 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset));
446 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
447 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset));
448 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET1, 0);
449 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
450 
451 		/* cache window 2: context */
452 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
453 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
454 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
455 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
456 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET2, 0);
457 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
458 
459 		/* non-cache window */
460 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
461 			lower_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
462 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
463 			upper_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
464 		WREG32_SOC15(VCN, i, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
465 		WREG32_SOC15(VCN, i, mmUVD_VCPU_NONCACHE_SIZE0,
466 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
467 	}
468 }
469 
470 static void vcn_v2_5_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
471 {
472 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
473 	uint32_t offset;
474 
475 	/* cache window 0: fw */
476 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
477 		if (!indirect) {
478 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
479 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
480 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
481 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
482 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
483 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
484 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
485 				VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
486 		} else {
487 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
488 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
489 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
490 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
491 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
492 				VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
493 		}
494 		offset = 0;
495 	} else {
496 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
497 			VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
498 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
499 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
500 			VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
501 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
502 		offset = size;
503 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
504 			VCN, 0, mmUVD_VCPU_CACHE_OFFSET0),
505 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
506 	}
507 
508 	if (!indirect)
509 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
510 			VCN, 0, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
511 	else
512 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
513 			VCN, 0, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
514 
515 	/* cache window 1: stack */
516 	if (!indirect) {
517 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
518 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
519 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
520 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
521 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
522 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
523 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
524 			VCN, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
525 	} else {
526 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
527 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
528 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
529 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
530 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
531 			VCN, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
532 	}
533 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
534 		VCN, 0, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
535 
536 	/* cache window 2: context */
537 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
538 		VCN, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
539 		lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
540 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
541 		VCN, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
542 		upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
543 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
544 		VCN, 0, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
545 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
546 		VCN, 0, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
547 
548 	/* non-cache window */
549 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
550 		VCN, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
551 		lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
552 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
553 		VCN, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
554 		upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
555 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
556 		VCN, 0, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
557 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
558 		VCN, 0, mmUVD_VCPU_NONCACHE_SIZE0),
559 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
560 
561 	/* VCN global tiling registers */
562 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
563 		VCN, 0, mmUVD_GFX8_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
564 }
565 
566 /**
567  * vcn_v2_5_disable_clock_gating - disable VCN clock gating
568  *
569  * @adev: amdgpu_device pointer
570  *
571  * Disable clock gating for VCN block
572  */
573 static void vcn_v2_5_disable_clock_gating(struct amdgpu_device *adev)
574 {
575 	uint32_t data;
576 	int i;
577 
578 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
579 		if (adev->vcn.harvest_config & (1 << i))
580 			continue;
581 		/* UVD disable CGC */
582 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
583 		if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
584 			data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
585 		else
586 			data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
587 		data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
588 		data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
589 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
590 
591 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_GATE);
592 		data &= ~(UVD_CGC_GATE__SYS_MASK
593 			| UVD_CGC_GATE__UDEC_MASK
594 			| UVD_CGC_GATE__MPEG2_MASK
595 			| UVD_CGC_GATE__REGS_MASK
596 			| UVD_CGC_GATE__RBC_MASK
597 			| UVD_CGC_GATE__LMI_MC_MASK
598 			| UVD_CGC_GATE__LMI_UMC_MASK
599 			| UVD_CGC_GATE__IDCT_MASK
600 			| UVD_CGC_GATE__MPRD_MASK
601 			| UVD_CGC_GATE__MPC_MASK
602 			| UVD_CGC_GATE__LBSI_MASK
603 			| UVD_CGC_GATE__LRBBM_MASK
604 			| UVD_CGC_GATE__UDEC_RE_MASK
605 			| UVD_CGC_GATE__UDEC_CM_MASK
606 			| UVD_CGC_GATE__UDEC_IT_MASK
607 			| UVD_CGC_GATE__UDEC_DB_MASK
608 			| UVD_CGC_GATE__UDEC_MP_MASK
609 			| UVD_CGC_GATE__WCB_MASK
610 			| UVD_CGC_GATE__VCPU_MASK
611 			| UVD_CGC_GATE__MMSCH_MASK);
612 
613 		WREG32_SOC15(VCN, i, mmUVD_CGC_GATE, data);
614 
615 		SOC15_WAIT_ON_RREG(VCN, i, mmUVD_CGC_GATE, 0,  0xFFFFFFFF);
616 
617 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
618 		data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
619 			| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
620 			| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
621 			| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
622 			| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
623 			| UVD_CGC_CTRL__SYS_MODE_MASK
624 			| UVD_CGC_CTRL__UDEC_MODE_MASK
625 			| UVD_CGC_CTRL__MPEG2_MODE_MASK
626 			| UVD_CGC_CTRL__REGS_MODE_MASK
627 			| UVD_CGC_CTRL__RBC_MODE_MASK
628 			| UVD_CGC_CTRL__LMI_MC_MODE_MASK
629 			| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
630 			| UVD_CGC_CTRL__IDCT_MODE_MASK
631 			| UVD_CGC_CTRL__MPRD_MODE_MASK
632 			| UVD_CGC_CTRL__MPC_MODE_MASK
633 			| UVD_CGC_CTRL__LBSI_MODE_MASK
634 			| UVD_CGC_CTRL__LRBBM_MODE_MASK
635 			| UVD_CGC_CTRL__WCB_MODE_MASK
636 			| UVD_CGC_CTRL__VCPU_MODE_MASK
637 			| UVD_CGC_CTRL__MMSCH_MODE_MASK);
638 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
639 
640 		/* turn on */
641 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE);
642 		data |= (UVD_SUVD_CGC_GATE__SRE_MASK
643 			| UVD_SUVD_CGC_GATE__SIT_MASK
644 			| UVD_SUVD_CGC_GATE__SMP_MASK
645 			| UVD_SUVD_CGC_GATE__SCM_MASK
646 			| UVD_SUVD_CGC_GATE__SDB_MASK
647 			| UVD_SUVD_CGC_GATE__SRE_H264_MASK
648 			| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
649 			| UVD_SUVD_CGC_GATE__SIT_H264_MASK
650 			| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
651 			| UVD_SUVD_CGC_GATE__SCM_H264_MASK
652 			| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
653 			| UVD_SUVD_CGC_GATE__SDB_H264_MASK
654 			| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
655 			| UVD_SUVD_CGC_GATE__SCLR_MASK
656 			| UVD_SUVD_CGC_GATE__UVD_SC_MASK
657 			| UVD_SUVD_CGC_GATE__ENT_MASK
658 			| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
659 			| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
660 			| UVD_SUVD_CGC_GATE__SITE_MASK
661 			| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
662 			| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
663 			| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
664 			| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
665 			| UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
666 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE, data);
667 
668 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL);
669 		data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
670 			| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
671 			| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
672 			| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
673 			| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
674 			| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
675 			| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
676 			| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
677 			| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
678 			| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
679 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data);
680 	}
681 }
682 
683 static void vcn_v2_5_clock_gating_dpg_mode(struct amdgpu_device *adev,
684 		uint8_t sram_sel, int inst_idx, uint8_t indirect)
685 {
686 	uint32_t reg_data = 0;
687 
688 	/* enable sw clock gating control */
689 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
690 		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
691 	else
692 		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
693 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
694 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
695 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
696 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
697 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
698 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
699 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
700 		 UVD_CGC_CTRL__SYS_MODE_MASK |
701 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
702 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
703 		 UVD_CGC_CTRL__REGS_MODE_MASK |
704 		 UVD_CGC_CTRL__RBC_MODE_MASK |
705 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
706 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
707 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
708 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
709 		 UVD_CGC_CTRL__MPC_MODE_MASK |
710 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
711 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
712 		 UVD_CGC_CTRL__WCB_MODE_MASK |
713 		 UVD_CGC_CTRL__VCPU_MODE_MASK |
714 		 UVD_CGC_CTRL__MMSCH_MODE_MASK);
715 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
716 		VCN, 0, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
717 
718 	/* turn off clock gating */
719 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
720 		VCN, 0, mmUVD_CGC_GATE), 0, sram_sel, indirect);
721 
722 	/* turn on SUVD clock gating */
723 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
724 		VCN, 0, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
725 
726 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
727 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
728 		VCN, 0, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
729 }
730 
731 /**
732  * vcn_v2_5_enable_clock_gating - enable VCN clock gating
733  *
734  * @adev: amdgpu_device pointer
735  *
736  * Enable clock gating for VCN block
737  */
738 static void vcn_v2_5_enable_clock_gating(struct amdgpu_device *adev)
739 {
740 	uint32_t data = 0;
741 	int i;
742 
743 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
744 		if (adev->vcn.harvest_config & (1 << i))
745 			continue;
746 		/* enable UVD CGC */
747 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
748 		if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
749 			data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
750 		else
751 			data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
752 		data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
753 		data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
754 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
755 
756 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
757 		data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
758 			| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
759 			| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
760 			| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
761 			| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
762 			| UVD_CGC_CTRL__SYS_MODE_MASK
763 			| UVD_CGC_CTRL__UDEC_MODE_MASK
764 			| UVD_CGC_CTRL__MPEG2_MODE_MASK
765 			| UVD_CGC_CTRL__REGS_MODE_MASK
766 			| UVD_CGC_CTRL__RBC_MODE_MASK
767 			| UVD_CGC_CTRL__LMI_MC_MODE_MASK
768 			| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
769 			| UVD_CGC_CTRL__IDCT_MODE_MASK
770 			| UVD_CGC_CTRL__MPRD_MODE_MASK
771 			| UVD_CGC_CTRL__MPC_MODE_MASK
772 			| UVD_CGC_CTRL__LBSI_MODE_MASK
773 			| UVD_CGC_CTRL__LRBBM_MODE_MASK
774 			| UVD_CGC_CTRL__WCB_MODE_MASK
775 			| UVD_CGC_CTRL__VCPU_MODE_MASK);
776 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
777 
778 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL);
779 		data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
780 			| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
781 			| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
782 			| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
783 			| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
784 			| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
785 			| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
786 			| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
787 			| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
788 			| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
789 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data);
790 	}
791 }
792 
793 static void vcn_v2_6_enable_ras(struct amdgpu_device *adev, int inst_idx,
794 				bool indirect)
795 {
796 	uint32_t tmp;
797 
798 	if (amdgpu_ip_version(adev, UVD_HWIP, 0) != IP_VERSION(2, 6, 0))
799 		return;
800 
801 	tmp = VCN_RAS_CNTL__VCPU_VCODEC_REARM_MASK |
802 	      VCN_RAS_CNTL__VCPU_VCODEC_IH_EN_MASK |
803 	      VCN_RAS_CNTL__VCPU_VCODEC_PMI_EN_MASK |
804 	      VCN_RAS_CNTL__VCPU_VCODEC_STALL_EN_MASK;
805 	WREG32_SOC15_DPG_MODE(inst_idx,
806 			      SOC15_DPG_MODE_OFFSET(VCN, 0, mmVCN_RAS_CNTL),
807 			      tmp, 0, indirect);
808 
809 	tmp = UVD_VCPU_INT_EN__RASCNTL_VCPU_VCODEC_EN_MASK;
810 	WREG32_SOC15_DPG_MODE(inst_idx,
811 			      SOC15_DPG_MODE_OFFSET(VCN, 0, mmUVD_VCPU_INT_EN),
812 			      tmp, 0, indirect);
813 
814 	tmp = UVD_SYS_INT_EN__RASCNTL_VCPU_VCODEC_EN_MASK;
815 	WREG32_SOC15_DPG_MODE(inst_idx,
816 			      SOC15_DPG_MODE_OFFSET(VCN, 0, mmUVD_SYS_INT_EN),
817 			      tmp, 0, indirect);
818 }
819 
820 static int vcn_v2_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
821 {
822 	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
823 	struct amdgpu_ring *ring;
824 	uint32_t rb_bufsz, tmp;
825 
826 	/* disable register anti-hang mechanism */
827 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1,
828 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
829 	/* enable dynamic power gating mode */
830 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS);
831 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
832 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
833 	WREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS, tmp);
834 
835 	if (indirect)
836 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
837 
838 	/* enable clock gating */
839 	vcn_v2_5_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
840 
841 	/* enable VCPU clock */
842 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
843 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
844 	tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
845 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
846 		VCN, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
847 
848 	/* disable master interupt */
849 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
850 		VCN, 0, mmUVD_MASTINT_EN), 0, 0, indirect);
851 
852 	/* setup mmUVD_LMI_CTRL */
853 	tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
854 		UVD_LMI_CTRL__REQ_MODE_MASK |
855 		UVD_LMI_CTRL__CRC_RESET_MASK |
856 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
857 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
858 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
859 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
860 		0x00100000L);
861 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
862 		VCN, 0, mmUVD_LMI_CTRL), tmp, 0, indirect);
863 
864 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
865 		VCN, 0, mmUVD_MPC_CNTL),
866 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
867 
868 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
869 		VCN, 0, mmUVD_MPC_SET_MUXA0),
870 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
871 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
872 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
873 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
874 
875 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
876 		VCN, 0, mmUVD_MPC_SET_MUXB0),
877 		((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
878 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
879 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
880 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
881 
882 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
883 		VCN, 0, mmUVD_MPC_SET_MUX),
884 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
885 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
886 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
887 
888 	vcn_v2_5_mc_resume_dpg_mode(adev, inst_idx, indirect);
889 
890 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
891 		VCN, 0, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
892 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
893 		VCN, 0, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
894 
895 	/* enable LMI MC and UMC channels */
896 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
897 		VCN, 0, mmUVD_LMI_CTRL2), 0, 0, indirect);
898 
899 	vcn_v2_6_enable_ras(adev, inst_idx, indirect);
900 
901 	/* unblock VCPU register access */
902 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
903 		VCN, 0, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
904 
905 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
906 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
907 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
908 		VCN, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
909 
910 	/* enable master interrupt */
911 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
912 		VCN, 0, mmUVD_MASTINT_EN),
913 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
914 
915 	if (indirect)
916 		amdgpu_vcn_psp_update_sram(adev, inst_idx, 0);
917 
918 	ring = &adev->vcn.inst[inst_idx].ring_dec;
919 	/* force RBC into idle state */
920 	rb_bufsz = order_base_2(ring->ring_size);
921 	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
922 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
923 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
924 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
925 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
926 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
927 
928 	/* Stall DPG before WPTR/RPTR reset */
929 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
930 		UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
931 		~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
932 	fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
933 
934 	/* set the write pointer delay */
935 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
936 
937 	/* set the wb address */
938 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR_ADDR,
939 		(upper_32_bits(ring->gpu_addr) >> 2));
940 
941 	/* program the RB_BASE for ring buffer */
942 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
943 		lower_32_bits(ring->gpu_addr));
944 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
945 		upper_32_bits(ring->gpu_addr));
946 
947 	/* Initialize the ring buffer's read and write pointers */
948 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, 0);
949 
950 	WREG32_SOC15(VCN, inst_idx, mmUVD_SCRATCH2, 0);
951 
952 	ring->wptr = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR);
953 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR,
954 		lower_32_bits(ring->wptr));
955 
956 	fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
957 	/* Unstall DPG */
958 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
959 		0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
960 
961 	return 0;
962 }
963 
964 static int vcn_v2_5_start(struct amdgpu_device *adev)
965 {
966 	struct amdgpu_ring *ring;
967 	uint32_t rb_bufsz, tmp;
968 	int i, j, k, r;
969 
970 	if (adev->pm.dpm_enabled)
971 		amdgpu_dpm_enable_uvd(adev, true);
972 
973 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
974 		if (adev->vcn.harvest_config & (1 << i))
975 			continue;
976 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
977 			r = vcn_v2_5_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
978 			continue;
979 		}
980 
981 		/* disable register anti-hang mechanism */
982 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_POWER_STATUS), 0,
983 			~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
984 
985 		/* set uvd status busy */
986 		tmp = RREG32_SOC15(VCN, i, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY;
987 		WREG32_SOC15(VCN, i, mmUVD_STATUS, tmp);
988 	}
989 
990 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
991 		return 0;
992 
993 	/*SW clock gating */
994 	vcn_v2_5_disable_clock_gating(adev);
995 
996 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
997 		if (adev->vcn.harvest_config & (1 << i))
998 			continue;
999 		/* enable VCPU clock */
1000 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1001 			UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1002 
1003 		/* disable master interrupt */
1004 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN), 0,
1005 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1006 
1007 		/* setup mmUVD_LMI_CTRL */
1008 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL);
1009 		tmp &= ~0xff;
1010 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL, tmp | 0x8|
1011 			UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK	|
1012 			UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1013 			UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1014 			UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1015 
1016 		/* setup mmUVD_MPC_CNTL */
1017 		tmp = RREG32_SOC15(VCN, i, mmUVD_MPC_CNTL);
1018 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1019 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1020 		WREG32_SOC15(VCN, i, mmUVD_MPC_CNTL, tmp);
1021 
1022 		/* setup UVD_MPC_SET_MUXA0 */
1023 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXA0,
1024 			((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1025 			(0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1026 			(0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1027 			(0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1028 
1029 		/* setup UVD_MPC_SET_MUXB0 */
1030 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXB0,
1031 			((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1032 			(0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1033 			(0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1034 			(0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1035 
1036 		/* setup mmUVD_MPC_SET_MUX */
1037 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUX,
1038 			((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1039 			(0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1040 			(0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1041 	}
1042 
1043 	vcn_v2_5_mc_resume(adev);
1044 
1045 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1046 		volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1047 		if (adev->vcn.harvest_config & (1 << i))
1048 			continue;
1049 		/* VCN global tiling registers */
1050 		WREG32_SOC15(VCN, i, mmUVD_GFX8_ADDR_CONFIG,
1051 			adev->gfx.config.gb_addr_config);
1052 		WREG32_SOC15(VCN, i, mmUVD_GFX8_ADDR_CONFIG,
1053 			adev->gfx.config.gb_addr_config);
1054 
1055 		/* enable LMI MC and UMC channels */
1056 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_LMI_CTRL2), 0,
1057 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1058 
1059 		/* unblock VCPU register access */
1060 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL), 0,
1061 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1062 
1063 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1064 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1065 
1066 		for (k = 0; k < 10; ++k) {
1067 			uint32_t status;
1068 
1069 			for (j = 0; j < 100; ++j) {
1070 				status = RREG32_SOC15(VCN, i, mmUVD_STATUS);
1071 				if (status & 2)
1072 					break;
1073 				if (amdgpu_emu_mode == 1)
1074 					msleep(500);
1075 				else
1076 					mdelay(10);
1077 			}
1078 			r = 0;
1079 			if (status & 2)
1080 				break;
1081 
1082 			DRM_ERROR("VCN decode not responding, trying to reset the VCPU!!!\n");
1083 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1084 				UVD_VCPU_CNTL__BLK_RST_MASK,
1085 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1086 			mdelay(10);
1087 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1088 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1089 
1090 			mdelay(10);
1091 			r = -1;
1092 		}
1093 
1094 		if (r) {
1095 			DRM_ERROR("VCN decode not responding, giving up!!!\n");
1096 			return r;
1097 		}
1098 
1099 		/* enable master interrupt */
1100 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN),
1101 			UVD_MASTINT_EN__VCPU_EN_MASK,
1102 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1103 
1104 		/* clear the busy bit of VCN_STATUS */
1105 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS), 0,
1106 			~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1107 
1108 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_VMID, 0);
1109 
1110 		ring = &adev->vcn.inst[i].ring_dec;
1111 		/* force RBC into idle state */
1112 		rb_bufsz = order_base_2(ring->ring_size);
1113 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1114 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1115 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1116 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1117 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1118 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_CNTL, tmp);
1119 
1120 		fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
1121 		/* program the RB_BASE for ring buffer */
1122 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1123 			lower_32_bits(ring->gpu_addr));
1124 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1125 			upper_32_bits(ring->gpu_addr));
1126 
1127 		/* Initialize the ring buffer's read and write pointers */
1128 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR, 0);
1129 
1130 		ring->wptr = RREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR);
1131 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_WPTR,
1132 				lower_32_bits(ring->wptr));
1133 		fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
1134 
1135 		fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1136 		ring = &adev->vcn.inst[i].ring_enc[0];
1137 		WREG32_SOC15(VCN, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1138 		WREG32_SOC15(VCN, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1139 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO, ring->gpu_addr);
1140 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1141 		WREG32_SOC15(VCN, i, mmUVD_RB_SIZE, ring->ring_size / 4);
1142 		fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1143 
1144 		fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1145 		ring = &adev->vcn.inst[i].ring_enc[1];
1146 		WREG32_SOC15(VCN, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1147 		WREG32_SOC15(VCN, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1148 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1149 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1150 		WREG32_SOC15(VCN, i, mmUVD_RB_SIZE2, ring->ring_size / 4);
1151 		fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1152 	}
1153 
1154 	return 0;
1155 }
1156 
1157 static int vcn_v2_5_mmsch_start(struct amdgpu_device *adev,
1158 				struct amdgpu_mm_table *table)
1159 {
1160 	uint32_t data = 0, loop = 0, size = 0;
1161 	uint64_t addr = table->gpu_addr;
1162 	struct mmsch_v1_1_init_header *header = NULL;
1163 
1164 	header = (struct mmsch_v1_1_init_header *)table->cpu_addr;
1165 	size = header->total_size;
1166 
1167 	/*
1168 	 * 1, write to vce_mmsch_vf_ctx_addr_lo/hi register with GPU mc addr of
1169 	 *  memory descriptor location
1170 	 */
1171 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(addr));
1172 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(addr));
1173 
1174 	/* 2, update vmid of descriptor */
1175 	data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID);
1176 	data &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1177 	/* use domain0 for MM scheduler */
1178 	data |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1179 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID, data);
1180 
1181 	/* 3, notify mmsch about the size of this descriptor */
1182 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_SIZE, size);
1183 
1184 	/* 4, set resp to zero */
1185 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1186 
1187 	/*
1188 	 * 5, kick off the initialization and wait until
1189 	 * VCE_MMSCH_VF_MAILBOX_RESP becomes non-zero
1190 	 */
1191 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_HOST, 0x10000001);
1192 
1193 	data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1194 	loop = 10;
1195 	while ((data & 0x10000002) != 0x10000002) {
1196 		udelay(100);
1197 		data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1198 		loop--;
1199 		if (!loop)
1200 			break;
1201 	}
1202 
1203 	if (!loop) {
1204 		dev_err(adev->dev,
1205 			"failed to init MMSCH, mmMMSCH_VF_MAILBOX_RESP = %x\n",
1206 			data);
1207 		return -EBUSY;
1208 	}
1209 
1210 	return 0;
1211 }
1212 
1213 static int vcn_v2_5_sriov_start(struct amdgpu_device *adev)
1214 {
1215 	struct amdgpu_ring *ring;
1216 	uint32_t offset, size, tmp, i, rb_bufsz;
1217 	uint32_t table_size = 0;
1218 	struct mmsch_v1_0_cmd_direct_write direct_wt = { { 0 } };
1219 	struct mmsch_v1_0_cmd_direct_read_modify_write direct_rd_mod_wt = { { 0 } };
1220 	struct mmsch_v1_0_cmd_end end = { { 0 } };
1221 	uint32_t *init_table = adev->virt.mm_table.cpu_addr;
1222 	struct mmsch_v1_1_init_header *header = (struct mmsch_v1_1_init_header *)init_table;
1223 
1224 	direct_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_WRITE;
1225 	direct_rd_mod_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1226 	end.cmd_header.command_type = MMSCH_COMMAND__END;
1227 
1228 	header->version = MMSCH_VERSION;
1229 	header->total_size = sizeof(struct mmsch_v1_1_init_header) >> 2;
1230 	init_table += header->total_size;
1231 
1232 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1233 		header->eng[i].table_offset = header->total_size;
1234 		header->eng[i].init_status = 0;
1235 		header->eng[i].table_size = 0;
1236 
1237 		table_size = 0;
1238 
1239 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(
1240 			SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS),
1241 			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1242 
1243 		size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1244 		/* mc resume*/
1245 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1246 			MMSCH_V1_0_INSERT_DIRECT_WT(
1247 				SOC15_REG_OFFSET(VCN, i,
1248 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1249 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1250 			MMSCH_V1_0_INSERT_DIRECT_WT(
1251 				SOC15_REG_OFFSET(VCN, i,
1252 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1253 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1254 			offset = 0;
1255 			MMSCH_V1_0_INSERT_DIRECT_WT(
1256 				SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET0), 0);
1257 		} else {
1258 			MMSCH_V1_0_INSERT_DIRECT_WT(
1259 				SOC15_REG_OFFSET(VCN, i,
1260 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1261 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1262 			MMSCH_V1_0_INSERT_DIRECT_WT(
1263 				SOC15_REG_OFFSET(VCN, i,
1264 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1265 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1266 			offset = size;
1267 			MMSCH_V1_0_INSERT_DIRECT_WT(
1268 				SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET0),
1269 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1270 		}
1271 
1272 		MMSCH_V1_0_INSERT_DIRECT_WT(
1273 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE0),
1274 			size);
1275 		MMSCH_V1_0_INSERT_DIRECT_WT(
1276 			SOC15_REG_OFFSET(VCN, i,
1277 				mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1278 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset));
1279 		MMSCH_V1_0_INSERT_DIRECT_WT(
1280 			SOC15_REG_OFFSET(VCN, i,
1281 				mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1282 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset));
1283 		MMSCH_V1_0_INSERT_DIRECT_WT(
1284 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET1),
1285 			0);
1286 		MMSCH_V1_0_INSERT_DIRECT_WT(
1287 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE1),
1288 			AMDGPU_VCN_STACK_SIZE);
1289 		MMSCH_V1_0_INSERT_DIRECT_WT(
1290 			SOC15_REG_OFFSET(VCN, i,
1291 				mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1292 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset +
1293 				AMDGPU_VCN_STACK_SIZE));
1294 		MMSCH_V1_0_INSERT_DIRECT_WT(
1295 			SOC15_REG_OFFSET(VCN, i,
1296 				mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1297 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset +
1298 				AMDGPU_VCN_STACK_SIZE));
1299 		MMSCH_V1_0_INSERT_DIRECT_WT(
1300 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET2),
1301 			0);
1302 		MMSCH_V1_0_INSERT_DIRECT_WT(
1303 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE2),
1304 			AMDGPU_VCN_CONTEXT_SIZE);
1305 
1306 		ring = &adev->vcn.inst[i].ring_enc[0];
1307 		ring->wptr = 0;
1308 
1309 		MMSCH_V1_0_INSERT_DIRECT_WT(
1310 			SOC15_REG_OFFSET(VCN, i, mmUVD_RB_BASE_LO),
1311 			lower_32_bits(ring->gpu_addr));
1312 		MMSCH_V1_0_INSERT_DIRECT_WT(
1313 			SOC15_REG_OFFSET(VCN, i, mmUVD_RB_BASE_HI),
1314 			upper_32_bits(ring->gpu_addr));
1315 		MMSCH_V1_0_INSERT_DIRECT_WT(
1316 			SOC15_REG_OFFSET(VCN, i, mmUVD_RB_SIZE),
1317 			ring->ring_size / 4);
1318 
1319 		ring = &adev->vcn.inst[i].ring_dec;
1320 		ring->wptr = 0;
1321 		MMSCH_V1_0_INSERT_DIRECT_WT(
1322 			SOC15_REG_OFFSET(VCN, i,
1323 				mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1324 			lower_32_bits(ring->gpu_addr));
1325 		MMSCH_V1_0_INSERT_DIRECT_WT(
1326 			SOC15_REG_OFFSET(VCN, i,
1327 				mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1328 			upper_32_bits(ring->gpu_addr));
1329 
1330 		/* force RBC into idle state */
1331 		rb_bufsz = order_base_2(ring->ring_size);
1332 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1333 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1334 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1335 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1336 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1337 		MMSCH_V1_0_INSERT_DIRECT_WT(
1338 			SOC15_REG_OFFSET(VCN, i, mmUVD_RBC_RB_CNTL), tmp);
1339 
1340 		/* add end packet */
1341 		memcpy((void *)init_table, &end, sizeof(struct mmsch_v1_0_cmd_end));
1342 		table_size += sizeof(struct mmsch_v1_0_cmd_end) / 4;
1343 		init_table += sizeof(struct mmsch_v1_0_cmd_end) / 4;
1344 
1345 		/* refine header */
1346 		header->eng[i].table_size = table_size;
1347 		header->total_size += table_size;
1348 	}
1349 
1350 	return vcn_v2_5_mmsch_start(adev, &adev->virt.mm_table);
1351 }
1352 
1353 static int vcn_v2_5_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1354 {
1355 	uint32_t tmp;
1356 
1357 	/* Wait for power status to be 1 */
1358 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1359 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1360 
1361 	/* wait for read ptr to be equal to write ptr */
1362 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR);
1363 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1364 
1365 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2);
1366 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
1367 
1368 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1369 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF);
1370 
1371 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1372 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1373 
1374 	/* disable dynamic power gating mode */
1375 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 0,
1376 			~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1377 
1378 	return 0;
1379 }
1380 
1381 static int vcn_v2_5_stop(struct amdgpu_device *adev)
1382 {
1383 	uint32_t tmp;
1384 	int i, r = 0;
1385 
1386 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1387 		if (adev->vcn.harvest_config & (1 << i))
1388 			continue;
1389 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1390 			r = vcn_v2_5_stop_dpg_mode(adev, i);
1391 			continue;
1392 		}
1393 
1394 		/* wait for vcn idle */
1395 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1396 		if (r)
1397 			return r;
1398 
1399 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1400 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1401 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1402 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1403 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1404 		if (r)
1405 			return r;
1406 
1407 		/* block LMI UMC channel */
1408 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2);
1409 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1410 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2, tmp);
1411 
1412 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1413 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1414 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1415 		if (r)
1416 			return r;
1417 
1418 		/* block VCPU register access */
1419 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL),
1420 			UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1421 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1422 
1423 		/* reset VCPU */
1424 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1425 			UVD_VCPU_CNTL__BLK_RST_MASK,
1426 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1427 
1428 		/* disable VCPU clock */
1429 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1430 			~(UVD_VCPU_CNTL__CLK_EN_MASK));
1431 
1432 		/* clear status */
1433 		WREG32_SOC15(VCN, i, mmUVD_STATUS, 0);
1434 
1435 		vcn_v2_5_enable_clock_gating(adev);
1436 
1437 		/* enable register anti-hang mechanism */
1438 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_POWER_STATUS),
1439 			UVD_POWER_STATUS__UVD_POWER_STATUS_MASK,
1440 			~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1441 	}
1442 
1443 	if (adev->pm.dpm_enabled)
1444 		amdgpu_dpm_enable_uvd(adev, false);
1445 
1446 	return 0;
1447 }
1448 
1449 static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
1450 				int inst_idx, struct dpg_pause_state *new_state)
1451 {
1452 	struct amdgpu_ring *ring;
1453 	uint32_t reg_data = 0;
1454 	int ret_code = 0;
1455 
1456 	/* pause/unpause if state is changed */
1457 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1458 		DRM_DEBUG("dpg pause state changed %d -> %d",
1459 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1460 		reg_data = RREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE) &
1461 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1462 
1463 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1464 			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1465 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1466 
1467 			if (!ret_code) {
1468 				volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
1469 
1470 				/* pause DPG */
1471 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1472 				WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1473 
1474 				/* wait for ACK */
1475 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_DPG_PAUSE,
1476 					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1477 					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1478 
1479 				/* Stall DPG before WPTR/RPTR reset */
1480 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1481 					   UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1482 					   ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1483 
1484 				/* Restore */
1485 				fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1486 				ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1487 				ring->wptr = 0;
1488 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
1489 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1490 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
1491 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1492 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1493 				fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1494 
1495 				fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1496 				ring = &adev->vcn.inst[inst_idx].ring_enc[1];
1497 				ring->wptr = 0;
1498 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1499 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1500 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
1501 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1502 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1503 				fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1504 
1505 				/* Unstall DPG */
1506 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1507 					   0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1508 
1509 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS,
1510 					   UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1511 			}
1512 		} else {
1513 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1514 			WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1515 			SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1516 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1517 		}
1518 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1519 	}
1520 
1521 	return 0;
1522 }
1523 
1524 /**
1525  * vcn_v2_5_dec_ring_get_rptr - get read pointer
1526  *
1527  * @ring: amdgpu_ring pointer
1528  *
1529  * Returns the current hardware read pointer
1530  */
1531 static uint64_t vcn_v2_5_dec_ring_get_rptr(struct amdgpu_ring *ring)
1532 {
1533 	struct amdgpu_device *adev = ring->adev;
1534 
1535 	return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_RPTR);
1536 }
1537 
1538 /**
1539  * vcn_v2_5_dec_ring_get_wptr - get write pointer
1540  *
1541  * @ring: amdgpu_ring pointer
1542  *
1543  * Returns the current hardware write pointer
1544  */
1545 static uint64_t vcn_v2_5_dec_ring_get_wptr(struct amdgpu_ring *ring)
1546 {
1547 	struct amdgpu_device *adev = ring->adev;
1548 
1549 	if (ring->use_doorbell)
1550 		return *ring->wptr_cpu_addr;
1551 	else
1552 		return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR);
1553 }
1554 
1555 /**
1556  * vcn_v2_5_dec_ring_set_wptr - set write pointer
1557  *
1558  * @ring: amdgpu_ring pointer
1559  *
1560  * Commits the write pointer to the hardware
1561  */
1562 static void vcn_v2_5_dec_ring_set_wptr(struct amdgpu_ring *ring)
1563 {
1564 	struct amdgpu_device *adev = ring->adev;
1565 
1566 	if (ring->use_doorbell) {
1567 		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1568 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1569 	} else {
1570 		WREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1571 	}
1572 }
1573 
1574 static const struct amdgpu_ring_funcs vcn_v2_5_dec_ring_vm_funcs = {
1575 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1576 	.align_mask = 0xf,
1577 	.secure_submission_supported = true,
1578 	.get_rptr = vcn_v2_5_dec_ring_get_rptr,
1579 	.get_wptr = vcn_v2_5_dec_ring_get_wptr,
1580 	.set_wptr = vcn_v2_5_dec_ring_set_wptr,
1581 	.emit_frame_size =
1582 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1583 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1584 		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1585 		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1586 		6,
1587 	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1588 	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
1589 	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
1590 	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1591 	.test_ring = vcn_v2_0_dec_ring_test_ring,
1592 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
1593 	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
1594 	.insert_start = vcn_v2_0_dec_ring_insert_start,
1595 	.insert_end = vcn_v2_0_dec_ring_insert_end,
1596 	.pad_ib = amdgpu_ring_generic_pad_ib,
1597 	.begin_use = amdgpu_vcn_ring_begin_use,
1598 	.end_use = amdgpu_vcn_ring_end_use,
1599 	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1600 	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1601 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1602 };
1603 
1604 /**
1605  * vcn_v2_5_enc_ring_get_rptr - get enc read pointer
1606  *
1607  * @ring: amdgpu_ring pointer
1608  *
1609  * Returns the current hardware enc read pointer
1610  */
1611 static uint64_t vcn_v2_5_enc_ring_get_rptr(struct amdgpu_ring *ring)
1612 {
1613 	struct amdgpu_device *adev = ring->adev;
1614 
1615 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0])
1616 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR);
1617 	else
1618 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR2);
1619 }
1620 
1621 /**
1622  * vcn_v2_5_enc_ring_get_wptr - get enc write pointer
1623  *
1624  * @ring: amdgpu_ring pointer
1625  *
1626  * Returns the current hardware enc write pointer
1627  */
1628 static uint64_t vcn_v2_5_enc_ring_get_wptr(struct amdgpu_ring *ring)
1629 {
1630 	struct amdgpu_device *adev = ring->adev;
1631 
1632 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1633 		if (ring->use_doorbell)
1634 			return *ring->wptr_cpu_addr;
1635 		else
1636 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR);
1637 	} else {
1638 		if (ring->use_doorbell)
1639 			return *ring->wptr_cpu_addr;
1640 		else
1641 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2);
1642 	}
1643 }
1644 
1645 /**
1646  * vcn_v2_5_enc_ring_set_wptr - set enc write pointer
1647  *
1648  * @ring: amdgpu_ring pointer
1649  *
1650  * Commits the enc write pointer to the hardware
1651  */
1652 static void vcn_v2_5_enc_ring_set_wptr(struct amdgpu_ring *ring)
1653 {
1654 	struct amdgpu_device *adev = ring->adev;
1655 
1656 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1657 		if (ring->use_doorbell) {
1658 			*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1659 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1660 		} else {
1661 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1662 		}
1663 	} else {
1664 		if (ring->use_doorbell) {
1665 			*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1666 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1667 		} else {
1668 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1669 		}
1670 	}
1671 }
1672 
1673 static const struct amdgpu_ring_funcs vcn_v2_5_enc_ring_vm_funcs = {
1674 	.type = AMDGPU_RING_TYPE_VCN_ENC,
1675 	.align_mask = 0x3f,
1676 	.nop = VCN_ENC_CMD_NO_OP,
1677 	.get_rptr = vcn_v2_5_enc_ring_get_rptr,
1678 	.get_wptr = vcn_v2_5_enc_ring_get_wptr,
1679 	.set_wptr = vcn_v2_5_enc_ring_set_wptr,
1680 	.emit_frame_size =
1681 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1682 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1683 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1684 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1685 		1, /* vcn_v2_0_enc_ring_insert_end */
1686 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1687 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
1688 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
1689 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1690 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
1691 	.test_ib = amdgpu_vcn_enc_ring_test_ib,
1692 	.insert_nop = amdgpu_ring_insert_nop,
1693 	.insert_end = vcn_v2_0_enc_ring_insert_end,
1694 	.pad_ib = amdgpu_ring_generic_pad_ib,
1695 	.begin_use = amdgpu_vcn_ring_begin_use,
1696 	.end_use = amdgpu_vcn_ring_end_use,
1697 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1698 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1699 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1700 };
1701 
1702 static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev)
1703 {
1704 	int i;
1705 
1706 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1707 		if (adev->vcn.harvest_config & (1 << i))
1708 			continue;
1709 		adev->vcn.inst[i].ring_dec.funcs = &vcn_v2_5_dec_ring_vm_funcs;
1710 		adev->vcn.inst[i].ring_dec.me = i;
1711 		DRM_INFO("VCN(%d) decode is enabled in VM mode\n", i);
1712 	}
1713 }
1714 
1715 static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev)
1716 {
1717 	int i, j;
1718 
1719 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
1720 		if (adev->vcn.harvest_config & (1 << j))
1721 			continue;
1722 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
1723 			adev->vcn.inst[j].ring_enc[i].funcs = &vcn_v2_5_enc_ring_vm_funcs;
1724 			adev->vcn.inst[j].ring_enc[i].me = j;
1725 		}
1726 		DRM_INFO("VCN(%d) encode is enabled in VM mode\n", j);
1727 	}
1728 }
1729 
1730 static bool vcn_v2_5_is_idle(void *handle)
1731 {
1732 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1733 	int i, ret = 1;
1734 
1735 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1736 		if (adev->vcn.harvest_config & (1 << i))
1737 			continue;
1738 		ret &= (RREG32_SOC15(VCN, i, mmUVD_STATUS) == UVD_STATUS__IDLE);
1739 	}
1740 
1741 	return ret;
1742 }
1743 
1744 static int vcn_v2_5_wait_for_idle(void *handle)
1745 {
1746 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1747 	int i, ret = 0;
1748 
1749 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1750 		if (adev->vcn.harvest_config & (1 << i))
1751 			continue;
1752 		ret = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE,
1753 			UVD_STATUS__IDLE);
1754 		if (ret)
1755 			return ret;
1756 	}
1757 
1758 	return ret;
1759 }
1760 
1761 static int vcn_v2_5_set_clockgating_state(void *handle,
1762 					  enum amd_clockgating_state state)
1763 {
1764 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1765 	bool enable = (state == AMD_CG_STATE_GATE);
1766 
1767 	if (amdgpu_sriov_vf(adev))
1768 		return 0;
1769 
1770 	if (enable) {
1771 		if (!vcn_v2_5_is_idle(handle))
1772 			return -EBUSY;
1773 		vcn_v2_5_enable_clock_gating(adev);
1774 	} else {
1775 		vcn_v2_5_disable_clock_gating(adev);
1776 	}
1777 
1778 	return 0;
1779 }
1780 
1781 static int vcn_v2_5_set_powergating_state(void *handle,
1782 					  enum amd_powergating_state state)
1783 {
1784 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1785 	int ret;
1786 
1787 	if (amdgpu_sriov_vf(adev))
1788 		return 0;
1789 
1790 	if(state == adev->vcn.cur_state)
1791 		return 0;
1792 
1793 	if (state == AMD_PG_STATE_GATE)
1794 		ret = vcn_v2_5_stop(adev);
1795 	else
1796 		ret = vcn_v2_5_start(adev);
1797 
1798 	if(!ret)
1799 		adev->vcn.cur_state = state;
1800 
1801 	return ret;
1802 }
1803 
1804 static int vcn_v2_5_set_interrupt_state(struct amdgpu_device *adev,
1805 					struct amdgpu_irq_src *source,
1806 					unsigned type,
1807 					enum amdgpu_interrupt_state state)
1808 {
1809 	return 0;
1810 }
1811 
1812 static int vcn_v2_6_set_ras_interrupt_state(struct amdgpu_device *adev,
1813 					struct amdgpu_irq_src *source,
1814 					unsigned int type,
1815 					enum amdgpu_interrupt_state state)
1816 {
1817 	return 0;
1818 }
1819 
1820 static int vcn_v2_5_process_interrupt(struct amdgpu_device *adev,
1821 				      struct amdgpu_irq_src *source,
1822 				      struct amdgpu_iv_entry *entry)
1823 {
1824 	uint32_t ip_instance;
1825 
1826 	switch (entry->client_id) {
1827 	case SOC15_IH_CLIENTID_VCN:
1828 		ip_instance = 0;
1829 		break;
1830 	case SOC15_IH_CLIENTID_VCN1:
1831 		ip_instance = 1;
1832 		break;
1833 	default:
1834 		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
1835 		return 0;
1836 	}
1837 
1838 	DRM_DEBUG("IH: VCN TRAP\n");
1839 
1840 	switch (entry->src_id) {
1841 	case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
1842 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec);
1843 		break;
1844 	case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
1845 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
1846 		break;
1847 	case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
1848 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]);
1849 		break;
1850 	default:
1851 		DRM_ERROR("Unhandled interrupt: %d %d\n",
1852 			  entry->src_id, entry->src_data[0]);
1853 		break;
1854 	}
1855 
1856 	return 0;
1857 }
1858 
1859 static const struct amdgpu_irq_src_funcs vcn_v2_5_irq_funcs = {
1860 	.set = vcn_v2_5_set_interrupt_state,
1861 	.process = vcn_v2_5_process_interrupt,
1862 };
1863 
1864 static const struct amdgpu_irq_src_funcs vcn_v2_6_ras_irq_funcs = {
1865 	.set = vcn_v2_6_set_ras_interrupt_state,
1866 	.process = amdgpu_vcn_process_poison_irq,
1867 };
1868 
1869 static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev)
1870 {
1871 	int i;
1872 
1873 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1874 		if (adev->vcn.harvest_config & (1 << i))
1875 			continue;
1876 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
1877 		adev->vcn.inst[i].irq.funcs = &vcn_v2_5_irq_funcs;
1878 
1879 		adev->vcn.inst[i].ras_poison_irq.num_types = adev->vcn.num_enc_rings + 1;
1880 		adev->vcn.inst[i].ras_poison_irq.funcs = &vcn_v2_6_ras_irq_funcs;
1881 	}
1882 }
1883 
1884 static const struct amd_ip_funcs vcn_v2_5_ip_funcs = {
1885 	.name = "vcn_v2_5",
1886 	.early_init = vcn_v2_5_early_init,
1887 	.late_init = NULL,
1888 	.sw_init = vcn_v2_5_sw_init,
1889 	.sw_fini = vcn_v2_5_sw_fini,
1890 	.hw_init = vcn_v2_5_hw_init,
1891 	.hw_fini = vcn_v2_5_hw_fini,
1892 	.suspend = vcn_v2_5_suspend,
1893 	.resume = vcn_v2_5_resume,
1894 	.is_idle = vcn_v2_5_is_idle,
1895 	.wait_for_idle = vcn_v2_5_wait_for_idle,
1896 	.check_soft_reset = NULL,
1897 	.pre_soft_reset = NULL,
1898 	.soft_reset = NULL,
1899 	.post_soft_reset = NULL,
1900 	.set_clockgating_state = vcn_v2_5_set_clockgating_state,
1901 	.set_powergating_state = vcn_v2_5_set_powergating_state,
1902 };
1903 
1904 static const struct amd_ip_funcs vcn_v2_6_ip_funcs = {
1905         .name = "vcn_v2_6",
1906         .early_init = vcn_v2_5_early_init,
1907         .late_init = NULL,
1908         .sw_init = vcn_v2_5_sw_init,
1909         .sw_fini = vcn_v2_5_sw_fini,
1910         .hw_init = vcn_v2_5_hw_init,
1911         .hw_fini = vcn_v2_5_hw_fini,
1912         .suspend = vcn_v2_5_suspend,
1913         .resume = vcn_v2_5_resume,
1914         .is_idle = vcn_v2_5_is_idle,
1915         .wait_for_idle = vcn_v2_5_wait_for_idle,
1916         .check_soft_reset = NULL,
1917         .pre_soft_reset = NULL,
1918         .soft_reset = NULL,
1919         .post_soft_reset = NULL,
1920         .set_clockgating_state = vcn_v2_5_set_clockgating_state,
1921         .set_powergating_state = vcn_v2_5_set_powergating_state,
1922 };
1923 
1924 const struct amdgpu_ip_block_version vcn_v2_5_ip_block =
1925 {
1926 		.type = AMD_IP_BLOCK_TYPE_VCN,
1927 		.major = 2,
1928 		.minor = 5,
1929 		.rev = 0,
1930 		.funcs = &vcn_v2_5_ip_funcs,
1931 };
1932 
1933 const struct amdgpu_ip_block_version vcn_v2_6_ip_block =
1934 {
1935 		.type = AMD_IP_BLOCK_TYPE_VCN,
1936 		.major = 2,
1937 		.minor = 6,
1938 		.rev = 0,
1939 		.funcs = &vcn_v2_6_ip_funcs,
1940 };
1941 
1942 static uint32_t vcn_v2_6_query_poison_by_instance(struct amdgpu_device *adev,
1943 			uint32_t instance, uint32_t sub_block)
1944 {
1945 	uint32_t poison_stat = 0, reg_value = 0;
1946 
1947 	switch (sub_block) {
1948 	case AMDGPU_VCN_V2_6_VCPU_VCODEC:
1949 		reg_value = RREG32_SOC15(VCN, instance, mmUVD_RAS_VCPU_VCODEC_STATUS);
1950 		poison_stat = REG_GET_FIELD(reg_value, UVD_RAS_VCPU_VCODEC_STATUS, POISONED_PF);
1951 		break;
1952 	default:
1953 		break;
1954 	}
1955 
1956 	if (poison_stat)
1957 		dev_info(adev->dev, "Poison detected in VCN%d, sub_block%d\n",
1958 			instance, sub_block);
1959 
1960 	return poison_stat;
1961 }
1962 
1963 static bool vcn_v2_6_query_poison_status(struct amdgpu_device *adev)
1964 {
1965 	uint32_t inst, sub;
1966 	uint32_t poison_stat = 0;
1967 
1968 	for (inst = 0; inst < adev->vcn.num_vcn_inst; inst++)
1969 		for (sub = 0; sub < AMDGPU_VCN_V2_6_MAX_SUB_BLOCK; sub++)
1970 			poison_stat +=
1971 			vcn_v2_6_query_poison_by_instance(adev, inst, sub);
1972 
1973 	return !!poison_stat;
1974 }
1975 
1976 const struct amdgpu_ras_block_hw_ops vcn_v2_6_ras_hw_ops = {
1977 	.query_poison_status = vcn_v2_6_query_poison_status,
1978 };
1979 
1980 static struct amdgpu_vcn_ras vcn_v2_6_ras = {
1981 	.ras_block = {
1982 		.hw_ops = &vcn_v2_6_ras_hw_ops,
1983 		.ras_late_init = amdgpu_vcn_ras_late_init,
1984 	},
1985 };
1986 
1987 static void vcn_v2_5_set_ras_funcs(struct amdgpu_device *adev)
1988 {
1989 	switch (amdgpu_ip_version(adev, VCN_HWIP, 0)) {
1990 	case IP_VERSION(2, 6, 0):
1991 		adev->vcn.ras = &vcn_v2_6_ras;
1992 		break;
1993 	default:
1994 		break;
1995 	}
1996 }
1997