xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c (revision 3ddc8b84)
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 (adev->ip_versions[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 (adev->ip_versions[UVD_HWIP][0] == IP_VERSION(2, 5, 0))
211 				ring->vm_hub = AMDGPU_MMHUB1(0);
212 			else
213 				ring->vm_hub = AMDGPU_MMHUB0(0);
214 
215 			sprintf(ring->name, "vcn_enc_%d.%d", j, i);
216 			r = amdgpu_ring_init(adev, ring, 512,
217 					     &adev->vcn.inst[j].irq, 0,
218 					     hw_prio, NULL);
219 			if (r)
220 				return r;
221 		}
222 
223 		fw_shared = adev->vcn.inst[j].fw_shared.cpu_addr;
224 		fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG);
225 
226 		if (amdgpu_vcnfw_log)
227 			amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
228 	}
229 
230 	if (amdgpu_sriov_vf(adev)) {
231 		r = amdgpu_virt_alloc_mm_table(adev);
232 		if (r)
233 			return r;
234 	}
235 
236 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
237 		adev->vcn.pause_dpg_mode = vcn_v2_5_pause_dpg_mode;
238 
239 	r = amdgpu_vcn_ras_sw_init(adev);
240 	if (r)
241 		return r;
242 
243 	return 0;
244 }
245 
246 /**
247  * vcn_v2_5_sw_fini - sw fini for VCN block
248  *
249  * @handle: amdgpu_device pointer
250  *
251  * VCN suspend and free up sw allocation
252  */
253 static int vcn_v2_5_sw_fini(void *handle)
254 {
255 	int i, r, idx;
256 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
257 	volatile struct amdgpu_fw_shared *fw_shared;
258 
259 	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
260 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
261 			if (adev->vcn.harvest_config & (1 << i))
262 				continue;
263 			fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
264 			fw_shared->present_flag_0 = 0;
265 		}
266 		drm_dev_exit(idx);
267 	}
268 
269 
270 	if (amdgpu_sriov_vf(adev))
271 		amdgpu_virt_free_mm_table(adev);
272 
273 	r = amdgpu_vcn_suspend(adev);
274 	if (r)
275 		return r;
276 
277 	r = amdgpu_vcn_sw_fini(adev);
278 
279 	return r;
280 }
281 
282 /**
283  * vcn_v2_5_hw_init - start and test VCN block
284  *
285  * @handle: amdgpu_device pointer
286  *
287  * Initialize the hardware, boot up the VCPU and do some testing
288  */
289 static int vcn_v2_5_hw_init(void *handle)
290 {
291 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
292 	struct amdgpu_ring *ring;
293 	int i, j, r = 0;
294 
295 	if (amdgpu_sriov_vf(adev))
296 		r = vcn_v2_5_sriov_start(adev);
297 
298 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
299 		if (adev->vcn.harvest_config & (1 << j))
300 			continue;
301 
302 		if (amdgpu_sriov_vf(adev)) {
303 			adev->vcn.inst[j].ring_enc[0].sched.ready = true;
304 			adev->vcn.inst[j].ring_enc[1].sched.ready = false;
305 			adev->vcn.inst[j].ring_enc[2].sched.ready = false;
306 			adev->vcn.inst[j].ring_dec.sched.ready = true;
307 		} else {
308 
309 			ring = &adev->vcn.inst[j].ring_dec;
310 
311 			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
312 						     ring->doorbell_index, j);
313 
314 			r = amdgpu_ring_test_helper(ring);
315 			if (r)
316 				goto done;
317 
318 			for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
319 				ring = &adev->vcn.inst[j].ring_enc[i];
320 				r = amdgpu_ring_test_helper(ring);
321 				if (r)
322 					goto done;
323 			}
324 		}
325 	}
326 
327 done:
328 	if (!r)
329 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
330 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
331 
332 	return r;
333 }
334 
335 /**
336  * vcn_v2_5_hw_fini - stop the hardware block
337  *
338  * @handle: amdgpu_device pointer
339  *
340  * Stop the VCN block, mark ring as not ready any more
341  */
342 static int vcn_v2_5_hw_fini(void *handle)
343 {
344 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
345 	int i;
346 
347 	cancel_delayed_work_sync(&adev->vcn.idle_work);
348 
349 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
350 		if (adev->vcn.harvest_config & (1 << i))
351 			continue;
352 
353 		if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
354 		    (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
355 		     RREG32_SOC15(VCN, i, mmUVD_STATUS)))
356 			vcn_v2_5_set_powergating_state(adev, AMD_PG_STATE_GATE);
357 
358 		if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN))
359 			amdgpu_irq_put(adev, &adev->vcn.inst[i].ras_poison_irq, 0);
360 	}
361 
362 	return 0;
363 }
364 
365 /**
366  * vcn_v2_5_suspend - suspend VCN block
367  *
368  * @handle: amdgpu_device pointer
369  *
370  * HW fini and suspend VCN block
371  */
372 static int vcn_v2_5_suspend(void *handle)
373 {
374 	int r;
375 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
376 
377 	r = vcn_v2_5_hw_fini(adev);
378 	if (r)
379 		return r;
380 
381 	r = amdgpu_vcn_suspend(adev);
382 
383 	return r;
384 }
385 
386 /**
387  * vcn_v2_5_resume - resume VCN block
388  *
389  * @handle: amdgpu_device pointer
390  *
391  * Resume firmware and hw init VCN block
392  */
393 static int vcn_v2_5_resume(void *handle)
394 {
395 	int r;
396 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
397 
398 	r = amdgpu_vcn_resume(adev);
399 	if (r)
400 		return r;
401 
402 	r = vcn_v2_5_hw_init(adev);
403 
404 	return r;
405 }
406 
407 /**
408  * vcn_v2_5_mc_resume - memory controller programming
409  *
410  * @adev: amdgpu_device pointer
411  *
412  * Let the VCN memory controller know it's offsets
413  */
414 static void vcn_v2_5_mc_resume(struct amdgpu_device *adev)
415 {
416 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
417 	uint32_t offset;
418 	int i;
419 
420 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
421 		if (adev->vcn.harvest_config & (1 << i))
422 			continue;
423 		/* cache window 0: fw */
424 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
425 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
426 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo));
427 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
428 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi));
429 			WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET0, 0);
430 			offset = 0;
431 		} else {
432 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
433 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
434 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
435 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
436 			offset = size;
437 			WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET0,
438 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
439 		}
440 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE0, size);
441 
442 		/* cache window 1: stack */
443 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
444 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset));
445 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
446 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset));
447 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET1, 0);
448 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
449 
450 		/* cache window 2: context */
451 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
452 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
453 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
454 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
455 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET2, 0);
456 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
457 
458 		/* non-cache window */
459 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
460 			lower_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
461 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
462 			upper_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
463 		WREG32_SOC15(VCN, i, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
464 		WREG32_SOC15(VCN, i, mmUVD_VCPU_NONCACHE_SIZE0,
465 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
466 	}
467 }
468 
469 static void vcn_v2_5_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
470 {
471 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
472 	uint32_t offset;
473 
474 	/* cache window 0: fw */
475 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
476 		if (!indirect) {
477 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
478 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
479 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
480 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
481 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
482 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
483 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
484 				VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
485 		} else {
486 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
487 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
488 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
489 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
490 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
491 				VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
492 		}
493 		offset = 0;
494 	} else {
495 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
496 			VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
497 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
498 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
499 			VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
500 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
501 		offset = size;
502 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
503 			VCN, 0, mmUVD_VCPU_CACHE_OFFSET0),
504 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
505 	}
506 
507 	if (!indirect)
508 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
509 			VCN, 0, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
510 	else
511 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
512 			VCN, 0, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
513 
514 	/* cache window 1: stack */
515 	if (!indirect) {
516 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
517 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
518 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
519 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
520 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
521 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
522 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
523 			VCN, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
524 	} else {
525 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
526 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
527 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
528 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
529 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
530 			VCN, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
531 	}
532 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
533 		VCN, 0, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
534 
535 	/* cache window 2: context */
536 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
537 		VCN, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
538 		lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
539 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
540 		VCN, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
541 		upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
542 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
543 		VCN, 0, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
544 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
545 		VCN, 0, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
546 
547 	/* non-cache window */
548 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
549 		VCN, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
550 		lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
551 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
552 		VCN, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
553 		upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
554 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
555 		VCN, 0, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
556 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
557 		VCN, 0, mmUVD_VCPU_NONCACHE_SIZE0),
558 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
559 
560 	/* VCN global tiling registers */
561 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
562 		VCN, 0, mmUVD_GFX8_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
563 }
564 
565 /**
566  * vcn_v2_5_disable_clock_gating - disable VCN clock gating
567  *
568  * @adev: amdgpu_device pointer
569  *
570  * Disable clock gating for VCN block
571  */
572 static void vcn_v2_5_disable_clock_gating(struct amdgpu_device *adev)
573 {
574 	uint32_t data;
575 	int i;
576 
577 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
578 		if (adev->vcn.harvest_config & (1 << i))
579 			continue;
580 		/* UVD disable CGC */
581 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
582 		if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
583 			data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
584 		else
585 			data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
586 		data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
587 		data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
588 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
589 
590 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_GATE);
591 		data &= ~(UVD_CGC_GATE__SYS_MASK
592 			| UVD_CGC_GATE__UDEC_MASK
593 			| UVD_CGC_GATE__MPEG2_MASK
594 			| UVD_CGC_GATE__REGS_MASK
595 			| UVD_CGC_GATE__RBC_MASK
596 			| UVD_CGC_GATE__LMI_MC_MASK
597 			| UVD_CGC_GATE__LMI_UMC_MASK
598 			| UVD_CGC_GATE__IDCT_MASK
599 			| UVD_CGC_GATE__MPRD_MASK
600 			| UVD_CGC_GATE__MPC_MASK
601 			| UVD_CGC_GATE__LBSI_MASK
602 			| UVD_CGC_GATE__LRBBM_MASK
603 			| UVD_CGC_GATE__UDEC_RE_MASK
604 			| UVD_CGC_GATE__UDEC_CM_MASK
605 			| UVD_CGC_GATE__UDEC_IT_MASK
606 			| UVD_CGC_GATE__UDEC_DB_MASK
607 			| UVD_CGC_GATE__UDEC_MP_MASK
608 			| UVD_CGC_GATE__WCB_MASK
609 			| UVD_CGC_GATE__VCPU_MASK
610 			| UVD_CGC_GATE__MMSCH_MASK);
611 
612 		WREG32_SOC15(VCN, i, mmUVD_CGC_GATE, data);
613 
614 		SOC15_WAIT_ON_RREG(VCN, i, mmUVD_CGC_GATE, 0,  0xFFFFFFFF);
615 
616 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
617 		data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
618 			| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
619 			| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
620 			| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
621 			| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
622 			| UVD_CGC_CTRL__SYS_MODE_MASK
623 			| UVD_CGC_CTRL__UDEC_MODE_MASK
624 			| UVD_CGC_CTRL__MPEG2_MODE_MASK
625 			| UVD_CGC_CTRL__REGS_MODE_MASK
626 			| UVD_CGC_CTRL__RBC_MODE_MASK
627 			| UVD_CGC_CTRL__LMI_MC_MODE_MASK
628 			| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
629 			| UVD_CGC_CTRL__IDCT_MODE_MASK
630 			| UVD_CGC_CTRL__MPRD_MODE_MASK
631 			| UVD_CGC_CTRL__MPC_MODE_MASK
632 			| UVD_CGC_CTRL__LBSI_MODE_MASK
633 			| UVD_CGC_CTRL__LRBBM_MODE_MASK
634 			| UVD_CGC_CTRL__WCB_MODE_MASK
635 			| UVD_CGC_CTRL__VCPU_MODE_MASK
636 			| UVD_CGC_CTRL__MMSCH_MODE_MASK);
637 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
638 
639 		/* turn on */
640 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE);
641 		data |= (UVD_SUVD_CGC_GATE__SRE_MASK
642 			| UVD_SUVD_CGC_GATE__SIT_MASK
643 			| UVD_SUVD_CGC_GATE__SMP_MASK
644 			| UVD_SUVD_CGC_GATE__SCM_MASK
645 			| UVD_SUVD_CGC_GATE__SDB_MASK
646 			| UVD_SUVD_CGC_GATE__SRE_H264_MASK
647 			| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
648 			| UVD_SUVD_CGC_GATE__SIT_H264_MASK
649 			| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
650 			| UVD_SUVD_CGC_GATE__SCM_H264_MASK
651 			| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
652 			| UVD_SUVD_CGC_GATE__SDB_H264_MASK
653 			| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
654 			| UVD_SUVD_CGC_GATE__SCLR_MASK
655 			| UVD_SUVD_CGC_GATE__UVD_SC_MASK
656 			| UVD_SUVD_CGC_GATE__ENT_MASK
657 			| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
658 			| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
659 			| UVD_SUVD_CGC_GATE__SITE_MASK
660 			| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
661 			| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
662 			| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
663 			| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
664 			| UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
665 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE, data);
666 
667 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL);
668 		data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
669 			| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
670 			| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
671 			| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
672 			| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
673 			| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
674 			| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
675 			| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
676 			| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
677 			| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
678 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data);
679 	}
680 }
681 
682 static void vcn_v2_5_clock_gating_dpg_mode(struct amdgpu_device *adev,
683 		uint8_t sram_sel, int inst_idx, uint8_t indirect)
684 {
685 	uint32_t reg_data = 0;
686 
687 	/* enable sw clock gating control */
688 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
689 		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
690 	else
691 		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
692 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
693 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
694 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
695 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
696 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
697 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
698 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
699 		 UVD_CGC_CTRL__SYS_MODE_MASK |
700 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
701 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
702 		 UVD_CGC_CTRL__REGS_MODE_MASK |
703 		 UVD_CGC_CTRL__RBC_MODE_MASK |
704 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
705 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
706 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
707 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
708 		 UVD_CGC_CTRL__MPC_MODE_MASK |
709 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
710 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
711 		 UVD_CGC_CTRL__WCB_MODE_MASK |
712 		 UVD_CGC_CTRL__VCPU_MODE_MASK |
713 		 UVD_CGC_CTRL__MMSCH_MODE_MASK);
714 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
715 		VCN, 0, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
716 
717 	/* turn off clock gating */
718 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
719 		VCN, 0, mmUVD_CGC_GATE), 0, sram_sel, indirect);
720 
721 	/* turn on SUVD clock gating */
722 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
723 		VCN, 0, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
724 
725 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
726 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
727 		VCN, 0, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
728 }
729 
730 /**
731  * vcn_v2_5_enable_clock_gating - enable VCN clock gating
732  *
733  * @adev: amdgpu_device pointer
734  *
735  * Enable clock gating for VCN block
736  */
737 static void vcn_v2_5_enable_clock_gating(struct amdgpu_device *adev)
738 {
739 	uint32_t data = 0;
740 	int i;
741 
742 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
743 		if (adev->vcn.harvest_config & (1 << i))
744 			continue;
745 		/* enable UVD CGC */
746 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
747 		if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
748 			data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
749 		else
750 			data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
751 		data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
752 		data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
753 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
754 
755 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
756 		data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
757 			| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
758 			| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
759 			| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
760 			| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
761 			| UVD_CGC_CTRL__SYS_MODE_MASK
762 			| UVD_CGC_CTRL__UDEC_MODE_MASK
763 			| UVD_CGC_CTRL__MPEG2_MODE_MASK
764 			| UVD_CGC_CTRL__REGS_MODE_MASK
765 			| UVD_CGC_CTRL__RBC_MODE_MASK
766 			| UVD_CGC_CTRL__LMI_MC_MODE_MASK
767 			| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
768 			| UVD_CGC_CTRL__IDCT_MODE_MASK
769 			| UVD_CGC_CTRL__MPRD_MODE_MASK
770 			| UVD_CGC_CTRL__MPC_MODE_MASK
771 			| UVD_CGC_CTRL__LBSI_MODE_MASK
772 			| UVD_CGC_CTRL__LRBBM_MODE_MASK
773 			| UVD_CGC_CTRL__WCB_MODE_MASK
774 			| UVD_CGC_CTRL__VCPU_MODE_MASK);
775 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
776 
777 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL);
778 		data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
779 			| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
780 			| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
781 			| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
782 			| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
783 			| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
784 			| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
785 			| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
786 			| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
787 			| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
788 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data);
789 	}
790 }
791 
792 static void vcn_v2_6_enable_ras(struct amdgpu_device *adev, int inst_idx,
793 				bool indirect)
794 {
795 	uint32_t tmp;
796 
797 	if (adev->ip_versions[UVD_HWIP][0] != IP_VERSION(2, 6, 0))
798 		return;
799 
800 	tmp = VCN_RAS_CNTL__VCPU_VCODEC_REARM_MASK |
801 	      VCN_RAS_CNTL__VCPU_VCODEC_IH_EN_MASK |
802 	      VCN_RAS_CNTL__VCPU_VCODEC_PMI_EN_MASK |
803 	      VCN_RAS_CNTL__VCPU_VCODEC_STALL_EN_MASK;
804 	WREG32_SOC15_DPG_MODE(inst_idx,
805 			      SOC15_DPG_MODE_OFFSET(VCN, 0, mmVCN_RAS_CNTL),
806 			      tmp, 0, indirect);
807 
808 	tmp = UVD_VCPU_INT_EN__RASCNTL_VCPU_VCODEC_EN_MASK;
809 	WREG32_SOC15_DPG_MODE(inst_idx,
810 			      SOC15_DPG_MODE_OFFSET(VCN, 0, mmUVD_VCPU_INT_EN),
811 			      tmp, 0, indirect);
812 
813 	tmp = UVD_SYS_INT_EN__RASCNTL_VCPU_VCODEC_EN_MASK;
814 	WREG32_SOC15_DPG_MODE(inst_idx,
815 			      SOC15_DPG_MODE_OFFSET(VCN, 0, mmUVD_SYS_INT_EN),
816 			      tmp, 0, indirect);
817 }
818 
819 static int vcn_v2_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
820 {
821 	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
822 	struct amdgpu_ring *ring;
823 	uint32_t rb_bufsz, tmp;
824 
825 	/* disable register anti-hang mechanism */
826 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1,
827 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
828 	/* enable dynamic power gating mode */
829 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS);
830 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
831 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
832 	WREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS, tmp);
833 
834 	if (indirect)
835 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
836 
837 	/* enable clock gating */
838 	vcn_v2_5_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
839 
840 	/* enable VCPU clock */
841 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
842 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
843 	tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
844 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
845 		VCN, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
846 
847 	/* disable master interupt */
848 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
849 		VCN, 0, mmUVD_MASTINT_EN), 0, 0, indirect);
850 
851 	/* setup mmUVD_LMI_CTRL */
852 	tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
853 		UVD_LMI_CTRL__REQ_MODE_MASK |
854 		UVD_LMI_CTRL__CRC_RESET_MASK |
855 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
856 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
857 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
858 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
859 		0x00100000L);
860 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
861 		VCN, 0, mmUVD_LMI_CTRL), tmp, 0, indirect);
862 
863 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
864 		VCN, 0, mmUVD_MPC_CNTL),
865 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
866 
867 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
868 		VCN, 0, mmUVD_MPC_SET_MUXA0),
869 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
870 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
871 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
872 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
873 
874 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
875 		VCN, 0, mmUVD_MPC_SET_MUXB0),
876 		((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
877 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
878 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
879 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
880 
881 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
882 		VCN, 0, mmUVD_MPC_SET_MUX),
883 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
884 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
885 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
886 
887 	vcn_v2_5_mc_resume_dpg_mode(adev, inst_idx, indirect);
888 
889 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
890 		VCN, 0, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
891 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
892 		VCN, 0, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
893 
894 	/* enable LMI MC and UMC channels */
895 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
896 		VCN, 0, mmUVD_LMI_CTRL2), 0, 0, indirect);
897 
898 	vcn_v2_6_enable_ras(adev, inst_idx, indirect);
899 
900 	/* unblock VCPU register access */
901 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
902 		VCN, 0, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
903 
904 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
905 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
906 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
907 		VCN, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
908 
909 	/* enable master interrupt */
910 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
911 		VCN, 0, mmUVD_MASTINT_EN),
912 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
913 
914 	if (indirect)
915 		amdgpu_vcn_psp_update_sram(adev, inst_idx, 0);
916 
917 	ring = &adev->vcn.inst[inst_idx].ring_dec;
918 	/* force RBC into idle state */
919 	rb_bufsz = order_base_2(ring->ring_size);
920 	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
921 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
922 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
923 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
924 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
925 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
926 
927 	/* Stall DPG before WPTR/RPTR reset */
928 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
929 		UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
930 		~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
931 	fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
932 
933 	/* set the write pointer delay */
934 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
935 
936 	/* set the wb address */
937 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR_ADDR,
938 		(upper_32_bits(ring->gpu_addr) >> 2));
939 
940 	/* program the RB_BASE for ring buffer */
941 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
942 		lower_32_bits(ring->gpu_addr));
943 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
944 		upper_32_bits(ring->gpu_addr));
945 
946 	/* Initialize the ring buffer's read and write pointers */
947 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, 0);
948 
949 	WREG32_SOC15(VCN, inst_idx, mmUVD_SCRATCH2, 0);
950 
951 	ring->wptr = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR);
952 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR,
953 		lower_32_bits(ring->wptr));
954 
955 	fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
956 	/* Unstall DPG */
957 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
958 		0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
959 
960 	return 0;
961 }
962 
963 static int vcn_v2_5_start(struct amdgpu_device *adev)
964 {
965 	struct amdgpu_ring *ring;
966 	uint32_t rb_bufsz, tmp;
967 	int i, j, k, r;
968 
969 	if (adev->pm.dpm_enabled)
970 		amdgpu_dpm_enable_uvd(adev, true);
971 
972 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
973 		if (adev->vcn.harvest_config & (1 << i))
974 			continue;
975 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
976 			r = vcn_v2_5_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
977 			continue;
978 		}
979 
980 		/* disable register anti-hang mechanism */
981 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_POWER_STATUS), 0,
982 			~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
983 
984 		/* set uvd status busy */
985 		tmp = RREG32_SOC15(VCN, i, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY;
986 		WREG32_SOC15(VCN, i, mmUVD_STATUS, tmp);
987 	}
988 
989 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
990 		return 0;
991 
992 	/*SW clock gating */
993 	vcn_v2_5_disable_clock_gating(adev);
994 
995 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
996 		if (adev->vcn.harvest_config & (1 << i))
997 			continue;
998 		/* enable VCPU clock */
999 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1000 			UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1001 
1002 		/* disable master interrupt */
1003 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN), 0,
1004 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1005 
1006 		/* setup mmUVD_LMI_CTRL */
1007 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL);
1008 		tmp &= ~0xff;
1009 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL, tmp | 0x8|
1010 			UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK	|
1011 			UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1012 			UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1013 			UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1014 
1015 		/* setup mmUVD_MPC_CNTL */
1016 		tmp = RREG32_SOC15(VCN, i, mmUVD_MPC_CNTL);
1017 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1018 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1019 		WREG32_SOC15(VCN, i, mmUVD_MPC_CNTL, tmp);
1020 
1021 		/* setup UVD_MPC_SET_MUXA0 */
1022 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXA0,
1023 			((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1024 			(0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1025 			(0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1026 			(0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1027 
1028 		/* setup UVD_MPC_SET_MUXB0 */
1029 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXB0,
1030 			((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1031 			(0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1032 			(0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1033 			(0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1034 
1035 		/* setup mmUVD_MPC_SET_MUX */
1036 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUX,
1037 			((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1038 			(0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1039 			(0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1040 	}
1041 
1042 	vcn_v2_5_mc_resume(adev);
1043 
1044 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1045 		volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1046 		if (adev->vcn.harvest_config & (1 << i))
1047 			continue;
1048 		/* VCN global tiling registers */
1049 		WREG32_SOC15(VCN, i, mmUVD_GFX8_ADDR_CONFIG,
1050 			adev->gfx.config.gb_addr_config);
1051 		WREG32_SOC15(VCN, i, mmUVD_GFX8_ADDR_CONFIG,
1052 			adev->gfx.config.gb_addr_config);
1053 
1054 		/* enable LMI MC and UMC channels */
1055 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_LMI_CTRL2), 0,
1056 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1057 
1058 		/* unblock VCPU register access */
1059 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL), 0,
1060 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1061 
1062 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1063 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1064 
1065 		for (k = 0; k < 10; ++k) {
1066 			uint32_t status;
1067 
1068 			for (j = 0; j < 100; ++j) {
1069 				status = RREG32_SOC15(VCN, i, mmUVD_STATUS);
1070 				if (status & 2)
1071 					break;
1072 				if (amdgpu_emu_mode == 1)
1073 					msleep(500);
1074 				else
1075 					mdelay(10);
1076 			}
1077 			r = 0;
1078 			if (status & 2)
1079 				break;
1080 
1081 			DRM_ERROR("VCN decode not responding, trying to reset the VCPU!!!\n");
1082 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1083 				UVD_VCPU_CNTL__BLK_RST_MASK,
1084 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1085 			mdelay(10);
1086 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1087 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1088 
1089 			mdelay(10);
1090 			r = -1;
1091 		}
1092 
1093 		if (r) {
1094 			DRM_ERROR("VCN decode not responding, giving up!!!\n");
1095 			return r;
1096 		}
1097 
1098 		/* enable master interrupt */
1099 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN),
1100 			UVD_MASTINT_EN__VCPU_EN_MASK,
1101 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1102 
1103 		/* clear the busy bit of VCN_STATUS */
1104 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS), 0,
1105 			~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1106 
1107 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_VMID, 0);
1108 
1109 		ring = &adev->vcn.inst[i].ring_dec;
1110 		/* force RBC into idle state */
1111 		rb_bufsz = order_base_2(ring->ring_size);
1112 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1113 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1114 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1115 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1116 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1117 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_CNTL, tmp);
1118 
1119 		fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
1120 		/* program the RB_BASE for ring buffer */
1121 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1122 			lower_32_bits(ring->gpu_addr));
1123 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1124 			upper_32_bits(ring->gpu_addr));
1125 
1126 		/* Initialize the ring buffer's read and write pointers */
1127 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR, 0);
1128 
1129 		ring->wptr = RREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR);
1130 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_WPTR,
1131 				lower_32_bits(ring->wptr));
1132 		fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
1133 
1134 		fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1135 		ring = &adev->vcn.inst[i].ring_enc[0];
1136 		WREG32_SOC15(VCN, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1137 		WREG32_SOC15(VCN, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1138 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO, ring->gpu_addr);
1139 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1140 		WREG32_SOC15(VCN, i, mmUVD_RB_SIZE, ring->ring_size / 4);
1141 		fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1142 
1143 		fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1144 		ring = &adev->vcn.inst[i].ring_enc[1];
1145 		WREG32_SOC15(VCN, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1146 		WREG32_SOC15(VCN, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1147 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1148 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1149 		WREG32_SOC15(VCN, i, mmUVD_RB_SIZE2, ring->ring_size / 4);
1150 		fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1151 	}
1152 
1153 	return 0;
1154 }
1155 
1156 static int vcn_v2_5_mmsch_start(struct amdgpu_device *adev,
1157 				struct amdgpu_mm_table *table)
1158 {
1159 	uint32_t data = 0, loop = 0, size = 0;
1160 	uint64_t addr = table->gpu_addr;
1161 	struct mmsch_v1_1_init_header *header = NULL;
1162 
1163 	header = (struct mmsch_v1_1_init_header *)table->cpu_addr;
1164 	size = header->total_size;
1165 
1166 	/*
1167 	 * 1, write to vce_mmsch_vf_ctx_addr_lo/hi register with GPU mc addr of
1168 	 *  memory descriptor location
1169 	 */
1170 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(addr));
1171 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(addr));
1172 
1173 	/* 2, update vmid of descriptor */
1174 	data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID);
1175 	data &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1176 	/* use domain0 for MM scheduler */
1177 	data |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1178 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID, data);
1179 
1180 	/* 3, notify mmsch about the size of this descriptor */
1181 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_SIZE, size);
1182 
1183 	/* 4, set resp to zero */
1184 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1185 
1186 	/*
1187 	 * 5, kick off the initialization and wait until
1188 	 * VCE_MMSCH_VF_MAILBOX_RESP becomes non-zero
1189 	 */
1190 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_HOST, 0x10000001);
1191 
1192 	data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1193 	loop = 10;
1194 	while ((data & 0x10000002) != 0x10000002) {
1195 		udelay(100);
1196 		data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1197 		loop--;
1198 		if (!loop)
1199 			break;
1200 	}
1201 
1202 	if (!loop) {
1203 		dev_err(adev->dev,
1204 			"failed to init MMSCH, mmMMSCH_VF_MAILBOX_RESP = %x\n",
1205 			data);
1206 		return -EBUSY;
1207 	}
1208 
1209 	return 0;
1210 }
1211 
1212 static int vcn_v2_5_sriov_start(struct amdgpu_device *adev)
1213 {
1214 	struct amdgpu_ring *ring;
1215 	uint32_t offset, size, tmp, i, rb_bufsz;
1216 	uint32_t table_size = 0;
1217 	struct mmsch_v1_0_cmd_direct_write direct_wt = { { 0 } };
1218 	struct mmsch_v1_0_cmd_direct_read_modify_write direct_rd_mod_wt = { { 0 } };
1219 	struct mmsch_v1_0_cmd_end end = { { 0 } };
1220 	uint32_t *init_table = adev->virt.mm_table.cpu_addr;
1221 	struct mmsch_v1_1_init_header *header = (struct mmsch_v1_1_init_header *)init_table;
1222 
1223 	direct_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_WRITE;
1224 	direct_rd_mod_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1225 	end.cmd_header.command_type = MMSCH_COMMAND__END;
1226 
1227 	header->version = MMSCH_VERSION;
1228 	header->total_size = sizeof(struct mmsch_v1_1_init_header) >> 2;
1229 	init_table += header->total_size;
1230 
1231 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1232 		header->eng[i].table_offset = header->total_size;
1233 		header->eng[i].init_status = 0;
1234 		header->eng[i].table_size = 0;
1235 
1236 		table_size = 0;
1237 
1238 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(
1239 			SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS),
1240 			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1241 
1242 		size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1243 		/* mc resume*/
1244 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1245 			MMSCH_V1_0_INSERT_DIRECT_WT(
1246 				SOC15_REG_OFFSET(VCN, i,
1247 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1248 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1249 			MMSCH_V1_0_INSERT_DIRECT_WT(
1250 				SOC15_REG_OFFSET(VCN, i,
1251 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1252 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1253 			offset = 0;
1254 			MMSCH_V1_0_INSERT_DIRECT_WT(
1255 				SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET0), 0);
1256 		} else {
1257 			MMSCH_V1_0_INSERT_DIRECT_WT(
1258 				SOC15_REG_OFFSET(VCN, i,
1259 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1260 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1261 			MMSCH_V1_0_INSERT_DIRECT_WT(
1262 				SOC15_REG_OFFSET(VCN, i,
1263 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1264 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1265 			offset = size;
1266 			MMSCH_V1_0_INSERT_DIRECT_WT(
1267 				SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET0),
1268 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1269 		}
1270 
1271 		MMSCH_V1_0_INSERT_DIRECT_WT(
1272 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE0),
1273 			size);
1274 		MMSCH_V1_0_INSERT_DIRECT_WT(
1275 			SOC15_REG_OFFSET(VCN, i,
1276 				mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1277 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset));
1278 		MMSCH_V1_0_INSERT_DIRECT_WT(
1279 			SOC15_REG_OFFSET(VCN, i,
1280 				mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1281 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset));
1282 		MMSCH_V1_0_INSERT_DIRECT_WT(
1283 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET1),
1284 			0);
1285 		MMSCH_V1_0_INSERT_DIRECT_WT(
1286 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE1),
1287 			AMDGPU_VCN_STACK_SIZE);
1288 		MMSCH_V1_0_INSERT_DIRECT_WT(
1289 			SOC15_REG_OFFSET(VCN, i,
1290 				mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1291 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset +
1292 				AMDGPU_VCN_STACK_SIZE));
1293 		MMSCH_V1_0_INSERT_DIRECT_WT(
1294 			SOC15_REG_OFFSET(VCN, i,
1295 				mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1296 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset +
1297 				AMDGPU_VCN_STACK_SIZE));
1298 		MMSCH_V1_0_INSERT_DIRECT_WT(
1299 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET2),
1300 			0);
1301 		MMSCH_V1_0_INSERT_DIRECT_WT(
1302 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE2),
1303 			AMDGPU_VCN_CONTEXT_SIZE);
1304 
1305 		ring = &adev->vcn.inst[i].ring_enc[0];
1306 		ring->wptr = 0;
1307 
1308 		MMSCH_V1_0_INSERT_DIRECT_WT(
1309 			SOC15_REG_OFFSET(VCN, i, mmUVD_RB_BASE_LO),
1310 			lower_32_bits(ring->gpu_addr));
1311 		MMSCH_V1_0_INSERT_DIRECT_WT(
1312 			SOC15_REG_OFFSET(VCN, i, mmUVD_RB_BASE_HI),
1313 			upper_32_bits(ring->gpu_addr));
1314 		MMSCH_V1_0_INSERT_DIRECT_WT(
1315 			SOC15_REG_OFFSET(VCN, i, mmUVD_RB_SIZE),
1316 			ring->ring_size / 4);
1317 
1318 		ring = &adev->vcn.inst[i].ring_dec;
1319 		ring->wptr = 0;
1320 		MMSCH_V1_0_INSERT_DIRECT_WT(
1321 			SOC15_REG_OFFSET(VCN, i,
1322 				mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1323 			lower_32_bits(ring->gpu_addr));
1324 		MMSCH_V1_0_INSERT_DIRECT_WT(
1325 			SOC15_REG_OFFSET(VCN, i,
1326 				mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1327 			upper_32_bits(ring->gpu_addr));
1328 
1329 		/* force RBC into idle state */
1330 		rb_bufsz = order_base_2(ring->ring_size);
1331 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1332 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1333 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1334 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1335 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1336 		MMSCH_V1_0_INSERT_DIRECT_WT(
1337 			SOC15_REG_OFFSET(VCN, i, mmUVD_RBC_RB_CNTL), tmp);
1338 
1339 		/* add end packet */
1340 		memcpy((void *)init_table, &end, sizeof(struct mmsch_v1_0_cmd_end));
1341 		table_size += sizeof(struct mmsch_v1_0_cmd_end) / 4;
1342 		init_table += sizeof(struct mmsch_v1_0_cmd_end) / 4;
1343 
1344 		/* refine header */
1345 		header->eng[i].table_size = table_size;
1346 		header->total_size += table_size;
1347 	}
1348 
1349 	return vcn_v2_5_mmsch_start(adev, &adev->virt.mm_table);
1350 }
1351 
1352 static int vcn_v2_5_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1353 {
1354 	uint32_t tmp;
1355 
1356 	/* Wait for power status to be 1 */
1357 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1358 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1359 
1360 	/* wait for read ptr to be equal to write ptr */
1361 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR);
1362 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1363 
1364 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2);
1365 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
1366 
1367 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1368 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF);
1369 
1370 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1371 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1372 
1373 	/* disable dynamic power gating mode */
1374 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 0,
1375 			~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1376 
1377 	return 0;
1378 }
1379 
1380 static int vcn_v2_5_stop(struct amdgpu_device *adev)
1381 {
1382 	uint32_t tmp;
1383 	int i, r = 0;
1384 
1385 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1386 		if (adev->vcn.harvest_config & (1 << i))
1387 			continue;
1388 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1389 			r = vcn_v2_5_stop_dpg_mode(adev, i);
1390 			continue;
1391 		}
1392 
1393 		/* wait for vcn idle */
1394 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1395 		if (r)
1396 			return r;
1397 
1398 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1399 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1400 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1401 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1402 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1403 		if (r)
1404 			return r;
1405 
1406 		/* block LMI UMC channel */
1407 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2);
1408 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1409 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2, tmp);
1410 
1411 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1412 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1413 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1414 		if (r)
1415 			return r;
1416 
1417 		/* block VCPU register access */
1418 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL),
1419 			UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1420 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1421 
1422 		/* reset VCPU */
1423 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1424 			UVD_VCPU_CNTL__BLK_RST_MASK,
1425 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1426 
1427 		/* disable VCPU clock */
1428 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1429 			~(UVD_VCPU_CNTL__CLK_EN_MASK));
1430 
1431 		/* clear status */
1432 		WREG32_SOC15(VCN, i, mmUVD_STATUS, 0);
1433 
1434 		vcn_v2_5_enable_clock_gating(adev);
1435 
1436 		/* enable register anti-hang mechanism */
1437 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_POWER_STATUS),
1438 			UVD_POWER_STATUS__UVD_POWER_STATUS_MASK,
1439 			~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1440 	}
1441 
1442 	if (adev->pm.dpm_enabled)
1443 		amdgpu_dpm_enable_uvd(adev, false);
1444 
1445 	return 0;
1446 }
1447 
1448 static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
1449 				int inst_idx, struct dpg_pause_state *new_state)
1450 {
1451 	struct amdgpu_ring *ring;
1452 	uint32_t reg_data = 0;
1453 	int ret_code = 0;
1454 
1455 	/* pause/unpause if state is changed */
1456 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1457 		DRM_DEBUG("dpg pause state changed %d -> %d",
1458 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1459 		reg_data = RREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE) &
1460 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1461 
1462 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1463 			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1464 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1465 
1466 			if (!ret_code) {
1467 				volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
1468 
1469 				/* pause DPG */
1470 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1471 				WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1472 
1473 				/* wait for ACK */
1474 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_DPG_PAUSE,
1475 					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1476 					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1477 
1478 				/* Stall DPG before WPTR/RPTR reset */
1479 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1480 					   UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1481 					   ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1482 
1483 				/* Restore */
1484 				fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1485 				ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1486 				ring->wptr = 0;
1487 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
1488 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1489 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
1490 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1491 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1492 				fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1493 
1494 				fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1495 				ring = &adev->vcn.inst[inst_idx].ring_enc[1];
1496 				ring->wptr = 0;
1497 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1498 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1499 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
1500 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1501 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1502 				fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1503 
1504 				/* Unstall DPG */
1505 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1506 					   0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1507 
1508 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS,
1509 					   UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1510 			}
1511 		} else {
1512 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1513 			WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1514 			SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1515 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1516 		}
1517 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1518 	}
1519 
1520 	return 0;
1521 }
1522 
1523 /**
1524  * vcn_v2_5_dec_ring_get_rptr - get read pointer
1525  *
1526  * @ring: amdgpu_ring pointer
1527  *
1528  * Returns the current hardware read pointer
1529  */
1530 static uint64_t vcn_v2_5_dec_ring_get_rptr(struct amdgpu_ring *ring)
1531 {
1532 	struct amdgpu_device *adev = ring->adev;
1533 
1534 	return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_RPTR);
1535 }
1536 
1537 /**
1538  * vcn_v2_5_dec_ring_get_wptr - get write pointer
1539  *
1540  * @ring: amdgpu_ring pointer
1541  *
1542  * Returns the current hardware write pointer
1543  */
1544 static uint64_t vcn_v2_5_dec_ring_get_wptr(struct amdgpu_ring *ring)
1545 {
1546 	struct amdgpu_device *adev = ring->adev;
1547 
1548 	if (ring->use_doorbell)
1549 		return *ring->wptr_cpu_addr;
1550 	else
1551 		return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR);
1552 }
1553 
1554 /**
1555  * vcn_v2_5_dec_ring_set_wptr - set write pointer
1556  *
1557  * @ring: amdgpu_ring pointer
1558  *
1559  * Commits the write pointer to the hardware
1560  */
1561 static void vcn_v2_5_dec_ring_set_wptr(struct amdgpu_ring *ring)
1562 {
1563 	struct amdgpu_device *adev = ring->adev;
1564 
1565 	if (ring->use_doorbell) {
1566 		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1567 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1568 	} else {
1569 		WREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1570 	}
1571 }
1572 
1573 static const struct amdgpu_ring_funcs vcn_v2_5_dec_ring_vm_funcs = {
1574 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1575 	.align_mask = 0xf,
1576 	.secure_submission_supported = true,
1577 	.get_rptr = vcn_v2_5_dec_ring_get_rptr,
1578 	.get_wptr = vcn_v2_5_dec_ring_get_wptr,
1579 	.set_wptr = vcn_v2_5_dec_ring_set_wptr,
1580 	.emit_frame_size =
1581 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1582 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1583 		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1584 		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1585 		6,
1586 	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1587 	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
1588 	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
1589 	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1590 	.test_ring = vcn_v2_0_dec_ring_test_ring,
1591 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
1592 	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
1593 	.insert_start = vcn_v2_0_dec_ring_insert_start,
1594 	.insert_end = vcn_v2_0_dec_ring_insert_end,
1595 	.pad_ib = amdgpu_ring_generic_pad_ib,
1596 	.begin_use = amdgpu_vcn_ring_begin_use,
1597 	.end_use = amdgpu_vcn_ring_end_use,
1598 	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1599 	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1600 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1601 };
1602 
1603 /**
1604  * vcn_v2_5_enc_ring_get_rptr - get enc read pointer
1605  *
1606  * @ring: amdgpu_ring pointer
1607  *
1608  * Returns the current hardware enc read pointer
1609  */
1610 static uint64_t vcn_v2_5_enc_ring_get_rptr(struct amdgpu_ring *ring)
1611 {
1612 	struct amdgpu_device *adev = ring->adev;
1613 
1614 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0])
1615 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR);
1616 	else
1617 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR2);
1618 }
1619 
1620 /**
1621  * vcn_v2_5_enc_ring_get_wptr - get enc write pointer
1622  *
1623  * @ring: amdgpu_ring pointer
1624  *
1625  * Returns the current hardware enc write pointer
1626  */
1627 static uint64_t vcn_v2_5_enc_ring_get_wptr(struct amdgpu_ring *ring)
1628 {
1629 	struct amdgpu_device *adev = ring->adev;
1630 
1631 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1632 		if (ring->use_doorbell)
1633 			return *ring->wptr_cpu_addr;
1634 		else
1635 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR);
1636 	} else {
1637 		if (ring->use_doorbell)
1638 			return *ring->wptr_cpu_addr;
1639 		else
1640 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2);
1641 	}
1642 }
1643 
1644 /**
1645  * vcn_v2_5_enc_ring_set_wptr - set enc write pointer
1646  *
1647  * @ring: amdgpu_ring pointer
1648  *
1649  * Commits the enc write pointer to the hardware
1650  */
1651 static void vcn_v2_5_enc_ring_set_wptr(struct amdgpu_ring *ring)
1652 {
1653 	struct amdgpu_device *adev = ring->adev;
1654 
1655 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1656 		if (ring->use_doorbell) {
1657 			*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1658 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1659 		} else {
1660 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1661 		}
1662 	} else {
1663 		if (ring->use_doorbell) {
1664 			*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1665 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1666 		} else {
1667 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1668 		}
1669 	}
1670 }
1671 
1672 static const struct amdgpu_ring_funcs vcn_v2_5_enc_ring_vm_funcs = {
1673 	.type = AMDGPU_RING_TYPE_VCN_ENC,
1674 	.align_mask = 0x3f,
1675 	.nop = VCN_ENC_CMD_NO_OP,
1676 	.get_rptr = vcn_v2_5_enc_ring_get_rptr,
1677 	.get_wptr = vcn_v2_5_enc_ring_get_wptr,
1678 	.set_wptr = vcn_v2_5_enc_ring_set_wptr,
1679 	.emit_frame_size =
1680 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1681 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1682 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1683 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1684 		1, /* vcn_v2_0_enc_ring_insert_end */
1685 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1686 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
1687 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
1688 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1689 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
1690 	.test_ib = amdgpu_vcn_enc_ring_test_ib,
1691 	.insert_nop = amdgpu_ring_insert_nop,
1692 	.insert_end = vcn_v2_0_enc_ring_insert_end,
1693 	.pad_ib = amdgpu_ring_generic_pad_ib,
1694 	.begin_use = amdgpu_vcn_ring_begin_use,
1695 	.end_use = amdgpu_vcn_ring_end_use,
1696 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1697 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1698 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1699 };
1700 
1701 static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev)
1702 {
1703 	int i;
1704 
1705 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1706 		if (adev->vcn.harvest_config & (1 << i))
1707 			continue;
1708 		adev->vcn.inst[i].ring_dec.funcs = &vcn_v2_5_dec_ring_vm_funcs;
1709 		adev->vcn.inst[i].ring_dec.me = i;
1710 		DRM_INFO("VCN(%d) decode is enabled in VM mode\n", i);
1711 	}
1712 }
1713 
1714 static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev)
1715 {
1716 	int i, j;
1717 
1718 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
1719 		if (adev->vcn.harvest_config & (1 << j))
1720 			continue;
1721 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
1722 			adev->vcn.inst[j].ring_enc[i].funcs = &vcn_v2_5_enc_ring_vm_funcs;
1723 			adev->vcn.inst[j].ring_enc[i].me = j;
1724 		}
1725 		DRM_INFO("VCN(%d) encode is enabled in VM mode\n", j);
1726 	}
1727 }
1728 
1729 static bool vcn_v2_5_is_idle(void *handle)
1730 {
1731 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1732 	int i, ret = 1;
1733 
1734 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1735 		if (adev->vcn.harvest_config & (1 << i))
1736 			continue;
1737 		ret &= (RREG32_SOC15(VCN, i, mmUVD_STATUS) == UVD_STATUS__IDLE);
1738 	}
1739 
1740 	return ret;
1741 }
1742 
1743 static int vcn_v2_5_wait_for_idle(void *handle)
1744 {
1745 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1746 	int i, ret = 0;
1747 
1748 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1749 		if (adev->vcn.harvest_config & (1 << i))
1750 			continue;
1751 		ret = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE,
1752 			UVD_STATUS__IDLE);
1753 		if (ret)
1754 			return ret;
1755 	}
1756 
1757 	return ret;
1758 }
1759 
1760 static int vcn_v2_5_set_clockgating_state(void *handle,
1761 					  enum amd_clockgating_state state)
1762 {
1763 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1764 	bool enable = (state == AMD_CG_STATE_GATE);
1765 
1766 	if (amdgpu_sriov_vf(adev))
1767 		return 0;
1768 
1769 	if (enable) {
1770 		if (!vcn_v2_5_is_idle(handle))
1771 			return -EBUSY;
1772 		vcn_v2_5_enable_clock_gating(adev);
1773 	} else {
1774 		vcn_v2_5_disable_clock_gating(adev);
1775 	}
1776 
1777 	return 0;
1778 }
1779 
1780 static int vcn_v2_5_set_powergating_state(void *handle,
1781 					  enum amd_powergating_state state)
1782 {
1783 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1784 	int ret;
1785 
1786 	if (amdgpu_sriov_vf(adev))
1787 		return 0;
1788 
1789 	if(state == adev->vcn.cur_state)
1790 		return 0;
1791 
1792 	if (state == AMD_PG_STATE_GATE)
1793 		ret = vcn_v2_5_stop(adev);
1794 	else
1795 		ret = vcn_v2_5_start(adev);
1796 
1797 	if(!ret)
1798 		adev->vcn.cur_state = state;
1799 
1800 	return ret;
1801 }
1802 
1803 static int vcn_v2_5_set_interrupt_state(struct amdgpu_device *adev,
1804 					struct amdgpu_irq_src *source,
1805 					unsigned type,
1806 					enum amdgpu_interrupt_state state)
1807 {
1808 	return 0;
1809 }
1810 
1811 static int vcn_v2_6_set_ras_interrupt_state(struct amdgpu_device *adev,
1812 					struct amdgpu_irq_src *source,
1813 					unsigned int type,
1814 					enum amdgpu_interrupt_state state)
1815 {
1816 	return 0;
1817 }
1818 
1819 static int vcn_v2_5_process_interrupt(struct amdgpu_device *adev,
1820 				      struct amdgpu_irq_src *source,
1821 				      struct amdgpu_iv_entry *entry)
1822 {
1823 	uint32_t ip_instance;
1824 
1825 	switch (entry->client_id) {
1826 	case SOC15_IH_CLIENTID_VCN:
1827 		ip_instance = 0;
1828 		break;
1829 	case SOC15_IH_CLIENTID_VCN1:
1830 		ip_instance = 1;
1831 		break;
1832 	default:
1833 		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
1834 		return 0;
1835 	}
1836 
1837 	DRM_DEBUG("IH: VCN TRAP\n");
1838 
1839 	switch (entry->src_id) {
1840 	case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
1841 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec);
1842 		break;
1843 	case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
1844 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
1845 		break;
1846 	case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
1847 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]);
1848 		break;
1849 	default:
1850 		DRM_ERROR("Unhandled interrupt: %d %d\n",
1851 			  entry->src_id, entry->src_data[0]);
1852 		break;
1853 	}
1854 
1855 	return 0;
1856 }
1857 
1858 static const struct amdgpu_irq_src_funcs vcn_v2_5_irq_funcs = {
1859 	.set = vcn_v2_5_set_interrupt_state,
1860 	.process = vcn_v2_5_process_interrupt,
1861 };
1862 
1863 static const struct amdgpu_irq_src_funcs vcn_v2_6_ras_irq_funcs = {
1864 	.set = vcn_v2_6_set_ras_interrupt_state,
1865 	.process = amdgpu_vcn_process_poison_irq,
1866 };
1867 
1868 static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev)
1869 {
1870 	int i;
1871 
1872 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1873 		if (adev->vcn.harvest_config & (1 << i))
1874 			continue;
1875 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
1876 		adev->vcn.inst[i].irq.funcs = &vcn_v2_5_irq_funcs;
1877 
1878 		adev->vcn.inst[i].ras_poison_irq.num_types = adev->vcn.num_enc_rings + 1;
1879 		adev->vcn.inst[i].ras_poison_irq.funcs = &vcn_v2_6_ras_irq_funcs;
1880 	}
1881 }
1882 
1883 static const struct amd_ip_funcs vcn_v2_5_ip_funcs = {
1884 	.name = "vcn_v2_5",
1885 	.early_init = vcn_v2_5_early_init,
1886 	.late_init = NULL,
1887 	.sw_init = vcn_v2_5_sw_init,
1888 	.sw_fini = vcn_v2_5_sw_fini,
1889 	.hw_init = vcn_v2_5_hw_init,
1890 	.hw_fini = vcn_v2_5_hw_fini,
1891 	.suspend = vcn_v2_5_suspend,
1892 	.resume = vcn_v2_5_resume,
1893 	.is_idle = vcn_v2_5_is_idle,
1894 	.wait_for_idle = vcn_v2_5_wait_for_idle,
1895 	.check_soft_reset = NULL,
1896 	.pre_soft_reset = NULL,
1897 	.soft_reset = NULL,
1898 	.post_soft_reset = NULL,
1899 	.set_clockgating_state = vcn_v2_5_set_clockgating_state,
1900 	.set_powergating_state = vcn_v2_5_set_powergating_state,
1901 };
1902 
1903 static const struct amd_ip_funcs vcn_v2_6_ip_funcs = {
1904         .name = "vcn_v2_6",
1905         .early_init = vcn_v2_5_early_init,
1906         .late_init = NULL,
1907         .sw_init = vcn_v2_5_sw_init,
1908         .sw_fini = vcn_v2_5_sw_fini,
1909         .hw_init = vcn_v2_5_hw_init,
1910         .hw_fini = vcn_v2_5_hw_fini,
1911         .suspend = vcn_v2_5_suspend,
1912         .resume = vcn_v2_5_resume,
1913         .is_idle = vcn_v2_5_is_idle,
1914         .wait_for_idle = vcn_v2_5_wait_for_idle,
1915         .check_soft_reset = NULL,
1916         .pre_soft_reset = NULL,
1917         .soft_reset = NULL,
1918         .post_soft_reset = NULL,
1919         .set_clockgating_state = vcn_v2_5_set_clockgating_state,
1920         .set_powergating_state = vcn_v2_5_set_powergating_state,
1921 };
1922 
1923 const struct amdgpu_ip_block_version vcn_v2_5_ip_block =
1924 {
1925 		.type = AMD_IP_BLOCK_TYPE_VCN,
1926 		.major = 2,
1927 		.minor = 5,
1928 		.rev = 0,
1929 		.funcs = &vcn_v2_5_ip_funcs,
1930 };
1931 
1932 const struct amdgpu_ip_block_version vcn_v2_6_ip_block =
1933 {
1934 		.type = AMD_IP_BLOCK_TYPE_VCN,
1935 		.major = 2,
1936 		.minor = 6,
1937 		.rev = 0,
1938 		.funcs = &vcn_v2_6_ip_funcs,
1939 };
1940 
1941 static uint32_t vcn_v2_6_query_poison_by_instance(struct amdgpu_device *adev,
1942 			uint32_t instance, uint32_t sub_block)
1943 {
1944 	uint32_t poison_stat = 0, reg_value = 0;
1945 
1946 	switch (sub_block) {
1947 	case AMDGPU_VCN_V2_6_VCPU_VCODEC:
1948 		reg_value = RREG32_SOC15(VCN, instance, mmUVD_RAS_VCPU_VCODEC_STATUS);
1949 		poison_stat = REG_GET_FIELD(reg_value, UVD_RAS_VCPU_VCODEC_STATUS, POISONED_PF);
1950 		break;
1951 	default:
1952 		break;
1953 	}
1954 
1955 	if (poison_stat)
1956 		dev_info(adev->dev, "Poison detected in VCN%d, sub_block%d\n",
1957 			instance, sub_block);
1958 
1959 	return poison_stat;
1960 }
1961 
1962 static bool vcn_v2_6_query_poison_status(struct amdgpu_device *adev)
1963 {
1964 	uint32_t inst, sub;
1965 	uint32_t poison_stat = 0;
1966 
1967 	for (inst = 0; inst < adev->vcn.num_vcn_inst; inst++)
1968 		for (sub = 0; sub < AMDGPU_VCN_V2_6_MAX_SUB_BLOCK; sub++)
1969 			poison_stat +=
1970 			vcn_v2_6_query_poison_by_instance(adev, inst, sub);
1971 
1972 	return !!poison_stat;
1973 }
1974 
1975 const struct amdgpu_ras_block_hw_ops vcn_v2_6_ras_hw_ops = {
1976 	.query_poison_status = vcn_v2_6_query_poison_status,
1977 };
1978 
1979 static struct amdgpu_vcn_ras vcn_v2_6_ras = {
1980 	.ras_block = {
1981 		.hw_ops = &vcn_v2_6_ras_hw_ops,
1982 		.ras_late_init = amdgpu_vcn_ras_late_init,
1983 	},
1984 };
1985 
1986 static void vcn_v2_5_set_ras_funcs(struct amdgpu_device *adev)
1987 {
1988 	switch (adev->ip_versions[VCN_HWIP][0]) {
1989 	case IP_VERSION(2, 6, 0):
1990 		adev->vcn.ras = &vcn_v2_6_ras;
1991 		break;
1992 	default:
1993 		break;
1994 	}
1995 }
1996