xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c (revision 1f012283)
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 
35 #include "vcn/vcn_2_5_offset.h"
36 #include "vcn/vcn_2_5_sh_mask.h"
37 #include "ivsrcid/vcn/irqsrcs_vcn_2_0.h"
38 
39 #define mmUVD_CONTEXT_ID_INTERNAL_OFFSET			0x27
40 #define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET			0x0f
41 #define mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET			0x10
42 #define mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET			0x11
43 #define mmUVD_NO_OP_INTERNAL_OFFSET				0x29
44 #define mmUVD_GP_SCRATCH8_INTERNAL_OFFSET			0x66
45 #define mmUVD_SCRATCH9_INTERNAL_OFFSET				0xc01d
46 
47 #define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET			0x431
48 #define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET		0x3b4
49 #define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET		0x3b5
50 #define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET			0x25c
51 
52 #define VCN25_MAX_HW_INSTANCES_ARCTURUS			2
53 
54 static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev);
55 static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev);
56 static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev);
57 static int vcn_v2_5_set_powergating_state(void *handle,
58 				enum amd_powergating_state state);
59 static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
60 				int inst_idx, struct dpg_pause_state *new_state);
61 static int vcn_v2_5_sriov_start(struct amdgpu_device *adev);
62 
63 static int amdgpu_ih_clientid_vcns[] = {
64 	SOC15_IH_CLIENTID_VCN,
65 	SOC15_IH_CLIENTID_VCN1
66 };
67 
68 /**
69  * vcn_v2_5_early_init - set function pointers
70  *
71  * @handle: amdgpu_device pointer
72  *
73  * Set ring and irq function pointers
74  */
75 static int vcn_v2_5_early_init(void *handle)
76 {
77 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
78 
79 	if (amdgpu_sriov_vf(adev)) {
80 		adev->vcn.num_vcn_inst = 2;
81 		adev->vcn.harvest_config = 0;
82 		adev->vcn.num_enc_rings = 1;
83 	} else {
84 		u32 harvest;
85 		int i;
86 
87 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
88 			harvest = RREG32_SOC15(VCN, i, mmCC_UVD_HARVESTING);
89 			if (harvest & CC_UVD_HARVESTING__UVD_DISABLE_MASK)
90 				adev->vcn.harvest_config |= 1 << i;
91 		}
92 		if (adev->vcn.harvest_config == (AMDGPU_VCN_HARVEST_VCN0 |
93 					AMDGPU_VCN_HARVEST_VCN1))
94 			/* both instances are harvested, disable the block */
95 			return -ENOENT;
96 
97 		adev->vcn.num_enc_rings = 2;
98 	}
99 
100 	vcn_v2_5_set_dec_ring_funcs(adev);
101 	vcn_v2_5_set_enc_ring_funcs(adev);
102 	vcn_v2_5_set_irq_funcs(adev);
103 
104 	return 0;
105 }
106 
107 /**
108  * vcn_v2_5_sw_init - sw init for VCN block
109  *
110  * @handle: amdgpu_device pointer
111  *
112  * Load firmware and sw initialization
113  */
114 static int vcn_v2_5_sw_init(void *handle)
115 {
116 	struct amdgpu_ring *ring;
117 	int i, j, r;
118 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
119 
120 	for (j = 0; j < adev->vcn.num_vcn_inst; j++) {
121 		if (adev->vcn.harvest_config & (1 << j))
122 			continue;
123 		/* VCN DEC TRAP */
124 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
125 				VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst[j].irq);
126 		if (r)
127 			return r;
128 
129 		/* VCN ENC TRAP */
130 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
131 			r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
132 				i + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[j].irq);
133 			if (r)
134 				return r;
135 		}
136 	}
137 
138 	r = amdgpu_vcn_sw_init(adev);
139 	if (r)
140 		return r;
141 
142 	amdgpu_vcn_setup_ucode(adev);
143 
144 	r = amdgpu_vcn_resume(adev);
145 	if (r)
146 		return r;
147 
148 	for (j = 0; j < adev->vcn.num_vcn_inst; j++) {
149 		volatile struct amdgpu_fw_shared *fw_shared;
150 
151 		if (adev->vcn.harvest_config & (1 << j))
152 			continue;
153 		adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
154 		adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
155 		adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
156 		adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
157 		adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
158 		adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
159 
160 		adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
161 		adev->vcn.inst[j].external.scratch9 = SOC15_REG_OFFSET(VCN, j, mmUVD_SCRATCH9);
162 		adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
163 		adev->vcn.inst[j].external.data0 = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_DATA0);
164 		adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
165 		adev->vcn.inst[j].external.data1 = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_DATA1);
166 		adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
167 		adev->vcn.inst[j].external.cmd = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_CMD);
168 		adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
169 		adev->vcn.inst[j].external.nop = SOC15_REG_OFFSET(VCN, j, mmUVD_NO_OP);
170 
171 		ring = &adev->vcn.inst[j].ring_dec;
172 		ring->use_doorbell = true;
173 
174 		ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
175 				(amdgpu_sriov_vf(adev) ? 2*j : 8*j);
176 		sprintf(ring->name, "vcn_dec_%d", j);
177 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[j].irq,
178 				     0, AMDGPU_RING_PRIO_DEFAULT, NULL);
179 		if (r)
180 			return r;
181 
182 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
183 			enum amdgpu_ring_priority_level hw_prio = amdgpu_vcn_get_enc_ring_prio(i);
184 
185 			ring = &adev->vcn.inst[j].ring_enc[i];
186 			ring->use_doorbell = true;
187 
188 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
189 					(amdgpu_sriov_vf(adev) ? (1 + i + 2*j) : (2 + i + 8*j));
190 
191 			sprintf(ring->name, "vcn_enc_%d.%d", j, i);
192 			r = amdgpu_ring_init(adev, ring, 512,
193 					     &adev->vcn.inst[j].irq, 0,
194 					     hw_prio, NULL);
195 			if (r)
196 				return r;
197 		}
198 
199 		fw_shared = adev->vcn.inst[j].fw_shared_cpu_addr;
200 		fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG);
201 	}
202 
203 	if (amdgpu_sriov_vf(adev)) {
204 		r = amdgpu_virt_alloc_mm_table(adev);
205 		if (r)
206 			return r;
207 	}
208 
209 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
210 		adev->vcn.pause_dpg_mode = vcn_v2_5_pause_dpg_mode;
211 
212 	return 0;
213 }
214 
215 /**
216  * vcn_v2_5_sw_fini - sw fini for VCN block
217  *
218  * @handle: amdgpu_device pointer
219  *
220  * VCN suspend and free up sw allocation
221  */
222 static int vcn_v2_5_sw_fini(void *handle)
223 {
224 	int i, r, idx;
225 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
226 	volatile struct amdgpu_fw_shared *fw_shared;
227 
228 	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
229 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
230 			if (adev->vcn.harvest_config & (1 << i))
231 				continue;
232 			fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
233 			fw_shared->present_flag_0 = 0;
234 		}
235 		drm_dev_exit(idx);
236 	}
237 
238 
239 	if (amdgpu_sriov_vf(adev))
240 		amdgpu_virt_free_mm_table(adev);
241 
242 	r = amdgpu_vcn_suspend(adev);
243 	if (r)
244 		return r;
245 
246 	r = amdgpu_vcn_sw_fini(adev);
247 
248 	return r;
249 }
250 
251 /**
252  * vcn_v2_5_hw_init - start and test VCN block
253  *
254  * @handle: amdgpu_device pointer
255  *
256  * Initialize the hardware, boot up the VCPU and do some testing
257  */
258 static int vcn_v2_5_hw_init(void *handle)
259 {
260 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
261 	struct amdgpu_ring *ring;
262 	int i, j, r = 0;
263 
264 	if (amdgpu_sriov_vf(adev))
265 		r = vcn_v2_5_sriov_start(adev);
266 
267 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
268 		if (adev->vcn.harvest_config & (1 << j))
269 			continue;
270 
271 		if (amdgpu_sriov_vf(adev)) {
272 			adev->vcn.inst[j].ring_enc[0].sched.ready = true;
273 			adev->vcn.inst[j].ring_enc[1].sched.ready = false;
274 			adev->vcn.inst[j].ring_enc[2].sched.ready = false;
275 			adev->vcn.inst[j].ring_dec.sched.ready = true;
276 		} else {
277 
278 			ring = &adev->vcn.inst[j].ring_dec;
279 
280 			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
281 						     ring->doorbell_index, j);
282 
283 			r = amdgpu_ring_test_helper(ring);
284 			if (r)
285 				goto done;
286 
287 			for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
288 				ring = &adev->vcn.inst[j].ring_enc[i];
289 				r = amdgpu_ring_test_helper(ring);
290 				if (r)
291 					goto done;
292 			}
293 		}
294 	}
295 
296 done:
297 	if (!r)
298 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
299 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
300 
301 	return r;
302 }
303 
304 /**
305  * vcn_v2_5_hw_fini - stop the hardware block
306  *
307  * @handle: amdgpu_device pointer
308  *
309  * Stop the VCN block, mark ring as not ready any more
310  */
311 static int vcn_v2_5_hw_fini(void *handle)
312 {
313 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
314 	int i;
315 
316 	cancel_delayed_work_sync(&adev->vcn.idle_work);
317 
318 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
319 		if (adev->vcn.harvest_config & (1 << i))
320 			continue;
321 
322 		if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
323 		    (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
324 		     RREG32_SOC15(VCN, i, mmUVD_STATUS)))
325 			vcn_v2_5_set_powergating_state(adev, AMD_PG_STATE_GATE);
326 	}
327 
328 	return 0;
329 }
330 
331 /**
332  * vcn_v2_5_suspend - suspend VCN block
333  *
334  * @handle: amdgpu_device pointer
335  *
336  * HW fini and suspend VCN block
337  */
338 static int vcn_v2_5_suspend(void *handle)
339 {
340 	int r;
341 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
342 
343 	r = vcn_v2_5_hw_fini(adev);
344 	if (r)
345 		return r;
346 
347 	r = amdgpu_vcn_suspend(adev);
348 
349 	return r;
350 }
351 
352 /**
353  * vcn_v2_5_resume - resume VCN block
354  *
355  * @handle: amdgpu_device pointer
356  *
357  * Resume firmware and hw init VCN block
358  */
359 static int vcn_v2_5_resume(void *handle)
360 {
361 	int r;
362 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
363 
364 	r = amdgpu_vcn_resume(adev);
365 	if (r)
366 		return r;
367 
368 	r = vcn_v2_5_hw_init(adev);
369 
370 	return r;
371 }
372 
373 /**
374  * vcn_v2_5_mc_resume - memory controller programming
375  *
376  * @adev: amdgpu_device pointer
377  *
378  * Let the VCN memory controller know it's offsets
379  */
380 static void vcn_v2_5_mc_resume(struct amdgpu_device *adev)
381 {
382 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
383 	uint32_t offset;
384 	int i;
385 
386 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
387 		if (adev->vcn.harvest_config & (1 << i))
388 			continue;
389 		/* cache window 0: fw */
390 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
391 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
392 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo));
393 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
394 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi));
395 			WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET0, 0);
396 			offset = 0;
397 		} else {
398 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
399 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
400 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
401 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
402 			offset = size;
403 			WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET0,
404 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
405 		}
406 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE0, size);
407 
408 		/* cache window 1: stack */
409 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
410 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset));
411 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
412 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset));
413 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET1, 0);
414 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
415 
416 		/* cache window 2: context */
417 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
418 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
419 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
420 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
421 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET2, 0);
422 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
423 
424 		/* non-cache window */
425 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
426 			lower_32_bits(adev->vcn.inst[i].fw_shared_gpu_addr));
427 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
428 			upper_32_bits(adev->vcn.inst[i].fw_shared_gpu_addr));
429 		WREG32_SOC15(VCN, i, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
430 		WREG32_SOC15(VCN, i, mmUVD_VCPU_NONCACHE_SIZE0,
431 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
432 	}
433 }
434 
435 static void vcn_v2_5_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
436 {
437 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
438 	uint32_t offset;
439 
440 	/* cache window 0: fw */
441 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
442 		if (!indirect) {
443 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
444 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
445 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
446 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
447 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
448 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
449 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
450 				VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
451 		} else {
452 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
453 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
454 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
455 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
456 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
457 				VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
458 		}
459 		offset = 0;
460 	} else {
461 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
462 			VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
463 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
464 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
465 			VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
466 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
467 		offset = size;
468 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
469 			VCN, 0, mmUVD_VCPU_CACHE_OFFSET0),
470 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
471 	}
472 
473 	if (!indirect)
474 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
475 			VCN, 0, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
476 	else
477 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
478 			VCN, 0, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
479 
480 	/* cache window 1: stack */
481 	if (!indirect) {
482 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
483 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
484 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
485 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
486 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
487 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
488 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
489 			VCN, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
490 	} else {
491 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
492 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
493 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
494 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
495 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
496 			VCN, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
497 	}
498 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
499 		VCN, 0, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
500 
501 	/* cache window 2: context */
502 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
503 		VCN, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
504 		lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
505 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
506 		VCN, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
507 		upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
508 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
509 		VCN, 0, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
510 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
511 		VCN, 0, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
512 
513 	/* non-cache window */
514 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
515 		VCN, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
516 		lower_32_bits(adev->vcn.inst[inst_idx].fw_shared_gpu_addr), 0, indirect);
517 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
518 		VCN, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
519 		upper_32_bits(adev->vcn.inst[inst_idx].fw_shared_gpu_addr), 0, indirect);
520 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
521 		VCN, 0, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
522 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
523 		VCN, 0, mmUVD_VCPU_NONCACHE_SIZE0),
524 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
525 
526 	/* VCN global tiling registers */
527 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
528 		VCN, 0, mmUVD_GFX8_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
529 }
530 
531 /**
532  * vcn_v2_5_disable_clock_gating - disable VCN clock gating
533  *
534  * @adev: amdgpu_device pointer
535  *
536  * Disable clock gating for VCN block
537  */
538 static void vcn_v2_5_disable_clock_gating(struct amdgpu_device *adev)
539 {
540 	uint32_t data;
541 	int i;
542 
543 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
544 		if (adev->vcn.harvest_config & (1 << i))
545 			continue;
546 		/* UVD disable CGC */
547 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
548 		if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
549 			data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
550 		else
551 			data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
552 		data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
553 		data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
554 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
555 
556 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_GATE);
557 		data &= ~(UVD_CGC_GATE__SYS_MASK
558 			| UVD_CGC_GATE__UDEC_MASK
559 			| UVD_CGC_GATE__MPEG2_MASK
560 			| UVD_CGC_GATE__REGS_MASK
561 			| UVD_CGC_GATE__RBC_MASK
562 			| UVD_CGC_GATE__LMI_MC_MASK
563 			| UVD_CGC_GATE__LMI_UMC_MASK
564 			| UVD_CGC_GATE__IDCT_MASK
565 			| UVD_CGC_GATE__MPRD_MASK
566 			| UVD_CGC_GATE__MPC_MASK
567 			| UVD_CGC_GATE__LBSI_MASK
568 			| UVD_CGC_GATE__LRBBM_MASK
569 			| UVD_CGC_GATE__UDEC_RE_MASK
570 			| UVD_CGC_GATE__UDEC_CM_MASK
571 			| UVD_CGC_GATE__UDEC_IT_MASK
572 			| UVD_CGC_GATE__UDEC_DB_MASK
573 			| UVD_CGC_GATE__UDEC_MP_MASK
574 			| UVD_CGC_GATE__WCB_MASK
575 			| UVD_CGC_GATE__VCPU_MASK
576 			| UVD_CGC_GATE__MMSCH_MASK);
577 
578 		WREG32_SOC15(VCN, i, mmUVD_CGC_GATE, data);
579 
580 		SOC15_WAIT_ON_RREG(VCN, i, mmUVD_CGC_GATE, 0,  0xFFFFFFFF);
581 
582 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
583 		data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
584 			| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
585 			| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
586 			| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
587 			| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
588 			| UVD_CGC_CTRL__SYS_MODE_MASK
589 			| UVD_CGC_CTRL__UDEC_MODE_MASK
590 			| UVD_CGC_CTRL__MPEG2_MODE_MASK
591 			| UVD_CGC_CTRL__REGS_MODE_MASK
592 			| UVD_CGC_CTRL__RBC_MODE_MASK
593 			| UVD_CGC_CTRL__LMI_MC_MODE_MASK
594 			| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
595 			| UVD_CGC_CTRL__IDCT_MODE_MASK
596 			| UVD_CGC_CTRL__MPRD_MODE_MASK
597 			| UVD_CGC_CTRL__MPC_MODE_MASK
598 			| UVD_CGC_CTRL__LBSI_MODE_MASK
599 			| UVD_CGC_CTRL__LRBBM_MODE_MASK
600 			| UVD_CGC_CTRL__WCB_MODE_MASK
601 			| UVD_CGC_CTRL__VCPU_MODE_MASK
602 			| UVD_CGC_CTRL__MMSCH_MODE_MASK);
603 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
604 
605 		/* turn on */
606 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE);
607 		data |= (UVD_SUVD_CGC_GATE__SRE_MASK
608 			| UVD_SUVD_CGC_GATE__SIT_MASK
609 			| UVD_SUVD_CGC_GATE__SMP_MASK
610 			| UVD_SUVD_CGC_GATE__SCM_MASK
611 			| UVD_SUVD_CGC_GATE__SDB_MASK
612 			| UVD_SUVD_CGC_GATE__SRE_H264_MASK
613 			| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
614 			| UVD_SUVD_CGC_GATE__SIT_H264_MASK
615 			| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
616 			| UVD_SUVD_CGC_GATE__SCM_H264_MASK
617 			| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
618 			| UVD_SUVD_CGC_GATE__SDB_H264_MASK
619 			| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
620 			| UVD_SUVD_CGC_GATE__SCLR_MASK
621 			| UVD_SUVD_CGC_GATE__UVD_SC_MASK
622 			| UVD_SUVD_CGC_GATE__ENT_MASK
623 			| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
624 			| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
625 			| UVD_SUVD_CGC_GATE__SITE_MASK
626 			| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
627 			| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
628 			| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
629 			| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
630 			| UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
631 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE, data);
632 
633 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL);
634 		data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
635 			| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
636 			| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
637 			| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
638 			| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
639 			| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
640 			| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
641 			| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
642 			| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
643 			| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
644 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data);
645 	}
646 }
647 
648 static void vcn_v2_5_clock_gating_dpg_mode(struct amdgpu_device *adev,
649 		uint8_t sram_sel, int inst_idx, uint8_t indirect)
650 {
651 	uint32_t reg_data = 0;
652 
653 	/* enable sw clock gating control */
654 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
655 		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
656 	else
657 		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
658 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
659 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
660 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
661 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
662 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
663 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
664 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
665 		 UVD_CGC_CTRL__SYS_MODE_MASK |
666 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
667 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
668 		 UVD_CGC_CTRL__REGS_MODE_MASK |
669 		 UVD_CGC_CTRL__RBC_MODE_MASK |
670 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
671 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
672 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
673 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
674 		 UVD_CGC_CTRL__MPC_MODE_MASK |
675 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
676 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
677 		 UVD_CGC_CTRL__WCB_MODE_MASK |
678 		 UVD_CGC_CTRL__VCPU_MODE_MASK |
679 		 UVD_CGC_CTRL__MMSCH_MODE_MASK);
680 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
681 		VCN, 0, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
682 
683 	/* turn off clock gating */
684 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
685 		VCN, 0, mmUVD_CGC_GATE), 0, sram_sel, indirect);
686 
687 	/* turn on SUVD clock gating */
688 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
689 		VCN, 0, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
690 
691 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
692 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
693 		VCN, 0, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
694 }
695 
696 /**
697  * vcn_v2_5_enable_clock_gating - enable VCN clock gating
698  *
699  * @adev: amdgpu_device pointer
700  *
701  * Enable clock gating for VCN block
702  */
703 static void vcn_v2_5_enable_clock_gating(struct amdgpu_device *adev)
704 {
705 	uint32_t data = 0;
706 	int i;
707 
708 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
709 		if (adev->vcn.harvest_config & (1 << i))
710 			continue;
711 		/* enable UVD CGC */
712 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
713 		if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
714 			data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
715 		else
716 			data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
717 		data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
718 		data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
719 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
720 
721 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
722 		data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
723 			| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
724 			| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
725 			| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
726 			| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
727 			| UVD_CGC_CTRL__SYS_MODE_MASK
728 			| UVD_CGC_CTRL__UDEC_MODE_MASK
729 			| UVD_CGC_CTRL__MPEG2_MODE_MASK
730 			| UVD_CGC_CTRL__REGS_MODE_MASK
731 			| UVD_CGC_CTRL__RBC_MODE_MASK
732 			| UVD_CGC_CTRL__LMI_MC_MODE_MASK
733 			| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
734 			| UVD_CGC_CTRL__IDCT_MODE_MASK
735 			| UVD_CGC_CTRL__MPRD_MODE_MASK
736 			| UVD_CGC_CTRL__MPC_MODE_MASK
737 			| UVD_CGC_CTRL__LBSI_MODE_MASK
738 			| UVD_CGC_CTRL__LRBBM_MODE_MASK
739 			| UVD_CGC_CTRL__WCB_MODE_MASK
740 			| UVD_CGC_CTRL__VCPU_MODE_MASK);
741 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
742 
743 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL);
744 		data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
745 			| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
746 			| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
747 			| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
748 			| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
749 			| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
750 			| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
751 			| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
752 			| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
753 			| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
754 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data);
755 	}
756 }
757 
758 static int vcn_v2_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
759 {
760 	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared_cpu_addr;
761 	struct amdgpu_ring *ring;
762 	uint32_t rb_bufsz, tmp;
763 
764 	/* disable register anti-hang mechanism */
765 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1,
766 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
767 	/* enable dynamic power gating mode */
768 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS);
769 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
770 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
771 	WREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS, tmp);
772 
773 	if (indirect)
774 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
775 
776 	/* enable clock gating */
777 	vcn_v2_5_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
778 
779 	/* enable VCPU clock */
780 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
781 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
782 	tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
783 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
784 		VCN, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
785 
786 	/* disable master interupt */
787 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
788 		VCN, 0, mmUVD_MASTINT_EN), 0, 0, indirect);
789 
790 	/* setup mmUVD_LMI_CTRL */
791 	tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
792 		UVD_LMI_CTRL__REQ_MODE_MASK |
793 		UVD_LMI_CTRL__CRC_RESET_MASK |
794 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
795 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
796 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
797 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
798 		0x00100000L);
799 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
800 		VCN, 0, mmUVD_LMI_CTRL), tmp, 0, indirect);
801 
802 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
803 		VCN, 0, mmUVD_MPC_CNTL),
804 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
805 
806 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
807 		VCN, 0, mmUVD_MPC_SET_MUXA0),
808 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
809 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
810 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
811 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
812 
813 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
814 		VCN, 0, mmUVD_MPC_SET_MUXB0),
815 		((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
816 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
817 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
818 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
819 
820 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
821 		VCN, 0, mmUVD_MPC_SET_MUX),
822 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
823 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
824 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
825 
826 	vcn_v2_5_mc_resume_dpg_mode(adev, inst_idx, indirect);
827 
828 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
829 		VCN, 0, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
830 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
831 		VCN, 0, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
832 
833 	/* enable LMI MC and UMC channels */
834 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
835 		VCN, 0, mmUVD_LMI_CTRL2), 0, 0, indirect);
836 
837 	/* unblock VCPU register access */
838 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
839 		VCN, 0, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
840 
841 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
842 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
843 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
844 		VCN, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
845 
846 	/* enable master interrupt */
847 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
848 		VCN, 0, mmUVD_MASTINT_EN),
849 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
850 
851 	if (indirect)
852 		psp_update_vcn_sram(adev, inst_idx, adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
853 				    (uint32_t)((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
854 					       (uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr));
855 
856 	ring = &adev->vcn.inst[inst_idx].ring_dec;
857 	/* force RBC into idle state */
858 	rb_bufsz = order_base_2(ring->ring_size);
859 	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
860 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
861 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
862 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
863 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
864 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
865 
866 	/* Stall DPG before WPTR/RPTR reset */
867 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
868 		UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
869 		~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
870 	fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
871 
872 	/* set the write pointer delay */
873 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
874 
875 	/* set the wb address */
876 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR_ADDR,
877 		(upper_32_bits(ring->gpu_addr) >> 2));
878 
879 	/* program the RB_BASE for ring buffer */
880 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
881 		lower_32_bits(ring->gpu_addr));
882 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
883 		upper_32_bits(ring->gpu_addr));
884 
885 	/* Initialize the ring buffer's read and write pointers */
886 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, 0);
887 
888 	WREG32_SOC15(VCN, inst_idx, mmUVD_SCRATCH2, 0);
889 
890 	ring->wptr = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR);
891 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR,
892 		lower_32_bits(ring->wptr));
893 
894 	fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
895 	/* Unstall DPG */
896 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
897 		0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
898 
899 	return 0;
900 }
901 
902 static int vcn_v2_5_start(struct amdgpu_device *adev)
903 {
904 	struct amdgpu_ring *ring;
905 	uint32_t rb_bufsz, tmp;
906 	int i, j, k, r;
907 
908 	if (adev->pm.dpm_enabled)
909 		amdgpu_dpm_enable_uvd(adev, true);
910 
911 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
912 		if (adev->vcn.harvest_config & (1 << i))
913 			continue;
914 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
915 			r = vcn_v2_5_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
916 			continue;
917 		}
918 
919 		/* disable register anti-hang mechanism */
920 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_POWER_STATUS), 0,
921 			~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
922 
923 		/* set uvd status busy */
924 		tmp = RREG32_SOC15(VCN, i, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY;
925 		WREG32_SOC15(VCN, i, mmUVD_STATUS, tmp);
926 	}
927 
928 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
929 		return 0;
930 
931 	/*SW clock gating */
932 	vcn_v2_5_disable_clock_gating(adev);
933 
934 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
935 		if (adev->vcn.harvest_config & (1 << i))
936 			continue;
937 		/* enable VCPU clock */
938 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
939 			UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
940 
941 		/* disable master interrupt */
942 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN), 0,
943 			~UVD_MASTINT_EN__VCPU_EN_MASK);
944 
945 		/* setup mmUVD_LMI_CTRL */
946 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL);
947 		tmp &= ~0xff;
948 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL, tmp | 0x8|
949 			UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK	|
950 			UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
951 			UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
952 			UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
953 
954 		/* setup mmUVD_MPC_CNTL */
955 		tmp = RREG32_SOC15(VCN, i, mmUVD_MPC_CNTL);
956 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
957 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
958 		WREG32_SOC15(VCN, i, mmUVD_MPC_CNTL, tmp);
959 
960 		/* setup UVD_MPC_SET_MUXA0 */
961 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXA0,
962 			((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
963 			(0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
964 			(0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
965 			(0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
966 
967 		/* setup UVD_MPC_SET_MUXB0 */
968 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXB0,
969 			((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
970 			(0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
971 			(0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
972 			(0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
973 
974 		/* setup mmUVD_MPC_SET_MUX */
975 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUX,
976 			((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
977 			(0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
978 			(0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
979 	}
980 
981 	vcn_v2_5_mc_resume(adev);
982 
983 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
984 		volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
985 		if (adev->vcn.harvest_config & (1 << i))
986 			continue;
987 		/* VCN global tiling registers */
988 		WREG32_SOC15(VCN, i, mmUVD_GFX8_ADDR_CONFIG,
989 			adev->gfx.config.gb_addr_config);
990 		WREG32_SOC15(VCN, i, mmUVD_GFX8_ADDR_CONFIG,
991 			adev->gfx.config.gb_addr_config);
992 
993 		/* enable LMI MC and UMC channels */
994 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_LMI_CTRL2), 0,
995 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
996 
997 		/* unblock VCPU register access */
998 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL), 0,
999 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1000 
1001 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1002 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1003 
1004 		for (k = 0; k < 10; ++k) {
1005 			uint32_t status;
1006 
1007 			for (j = 0; j < 100; ++j) {
1008 				status = RREG32_SOC15(VCN, i, mmUVD_STATUS);
1009 				if (status & 2)
1010 					break;
1011 				if (amdgpu_emu_mode == 1)
1012 					msleep(500);
1013 				else
1014 					mdelay(10);
1015 			}
1016 			r = 0;
1017 			if (status & 2)
1018 				break;
1019 
1020 			DRM_ERROR("VCN decode not responding, trying to reset the VCPU!!!\n");
1021 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1022 				UVD_VCPU_CNTL__BLK_RST_MASK,
1023 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1024 			mdelay(10);
1025 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1026 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1027 
1028 			mdelay(10);
1029 			r = -1;
1030 		}
1031 
1032 		if (r) {
1033 			DRM_ERROR("VCN decode not responding, giving up!!!\n");
1034 			return r;
1035 		}
1036 
1037 		/* enable master interrupt */
1038 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN),
1039 			UVD_MASTINT_EN__VCPU_EN_MASK,
1040 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1041 
1042 		/* clear the busy bit of VCN_STATUS */
1043 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS), 0,
1044 			~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1045 
1046 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_VMID, 0);
1047 
1048 		ring = &adev->vcn.inst[i].ring_dec;
1049 		/* force RBC into idle state */
1050 		rb_bufsz = order_base_2(ring->ring_size);
1051 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1052 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1053 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1054 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1055 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1056 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_CNTL, tmp);
1057 
1058 		fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
1059 		/* program the RB_BASE for ring buffer */
1060 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1061 			lower_32_bits(ring->gpu_addr));
1062 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1063 			upper_32_bits(ring->gpu_addr));
1064 
1065 		/* Initialize the ring buffer's read and write pointers */
1066 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR, 0);
1067 
1068 		ring->wptr = RREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR);
1069 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_WPTR,
1070 				lower_32_bits(ring->wptr));
1071 		fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
1072 
1073 		fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1074 		ring = &adev->vcn.inst[i].ring_enc[0];
1075 		WREG32_SOC15(VCN, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1076 		WREG32_SOC15(VCN, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1077 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO, ring->gpu_addr);
1078 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1079 		WREG32_SOC15(VCN, i, mmUVD_RB_SIZE, ring->ring_size / 4);
1080 		fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1081 
1082 		fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1083 		ring = &adev->vcn.inst[i].ring_enc[1];
1084 		WREG32_SOC15(VCN, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1085 		WREG32_SOC15(VCN, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1086 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1087 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1088 		WREG32_SOC15(VCN, i, mmUVD_RB_SIZE2, ring->ring_size / 4);
1089 		fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1090 	}
1091 
1092 	return 0;
1093 }
1094 
1095 static int vcn_v2_5_mmsch_start(struct amdgpu_device *adev,
1096 				struct amdgpu_mm_table *table)
1097 {
1098 	uint32_t data = 0, loop = 0, size = 0;
1099 	uint64_t addr = table->gpu_addr;
1100 	struct mmsch_v1_1_init_header *header = NULL;
1101 
1102 	header = (struct mmsch_v1_1_init_header *)table->cpu_addr;
1103 	size = header->total_size;
1104 
1105 	/*
1106 	 * 1, write to vce_mmsch_vf_ctx_addr_lo/hi register with GPU mc addr of
1107 	 *  memory descriptor location
1108 	 */
1109 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(addr));
1110 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(addr));
1111 
1112 	/* 2, update vmid of descriptor */
1113 	data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID);
1114 	data &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1115 	/* use domain0 for MM scheduler */
1116 	data |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1117 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID, data);
1118 
1119 	/* 3, notify mmsch about the size of this descriptor */
1120 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_SIZE, size);
1121 
1122 	/* 4, set resp to zero */
1123 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1124 
1125 	/*
1126 	 * 5, kick off the initialization and wait until
1127 	 * VCE_MMSCH_VF_MAILBOX_RESP becomes non-zero
1128 	 */
1129 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_HOST, 0x10000001);
1130 
1131 	data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1132 	loop = 10;
1133 	while ((data & 0x10000002) != 0x10000002) {
1134 		udelay(100);
1135 		data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1136 		loop--;
1137 		if (!loop)
1138 			break;
1139 	}
1140 
1141 	if (!loop) {
1142 		dev_err(adev->dev,
1143 			"failed to init MMSCH, mmMMSCH_VF_MAILBOX_RESP = %x\n",
1144 			data);
1145 		return -EBUSY;
1146 	}
1147 
1148 	return 0;
1149 }
1150 
1151 static int vcn_v2_5_sriov_start(struct amdgpu_device *adev)
1152 {
1153 	struct amdgpu_ring *ring;
1154 	uint32_t offset, size, tmp, i, rb_bufsz;
1155 	uint32_t table_size = 0;
1156 	struct mmsch_v1_0_cmd_direct_write direct_wt = { { 0 } };
1157 	struct mmsch_v1_0_cmd_direct_read_modify_write direct_rd_mod_wt = { { 0 } };
1158 	struct mmsch_v1_0_cmd_end end = { { 0 } };
1159 	uint32_t *init_table = adev->virt.mm_table.cpu_addr;
1160 	struct mmsch_v1_1_init_header *header = (struct mmsch_v1_1_init_header *)init_table;
1161 
1162 	direct_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_WRITE;
1163 	direct_rd_mod_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1164 	end.cmd_header.command_type = MMSCH_COMMAND__END;
1165 
1166 	header->version = MMSCH_VERSION;
1167 	header->total_size = sizeof(struct mmsch_v1_1_init_header) >> 2;
1168 	init_table += header->total_size;
1169 
1170 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1171 		header->eng[i].table_offset = header->total_size;
1172 		header->eng[i].init_status = 0;
1173 		header->eng[i].table_size = 0;
1174 
1175 		table_size = 0;
1176 
1177 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(
1178 			SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS),
1179 			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1180 
1181 		size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1182 		/* mc resume*/
1183 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1184 			MMSCH_V1_0_INSERT_DIRECT_WT(
1185 				SOC15_REG_OFFSET(VCN, i,
1186 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1187 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1188 			MMSCH_V1_0_INSERT_DIRECT_WT(
1189 				SOC15_REG_OFFSET(VCN, i,
1190 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1191 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1192 			offset = 0;
1193 			MMSCH_V1_0_INSERT_DIRECT_WT(
1194 				SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET0), 0);
1195 		} else {
1196 			MMSCH_V1_0_INSERT_DIRECT_WT(
1197 				SOC15_REG_OFFSET(VCN, i,
1198 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1199 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1200 			MMSCH_V1_0_INSERT_DIRECT_WT(
1201 				SOC15_REG_OFFSET(VCN, i,
1202 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1203 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1204 			offset = size;
1205 			MMSCH_V1_0_INSERT_DIRECT_WT(
1206 				SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET0),
1207 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1208 		}
1209 
1210 		MMSCH_V1_0_INSERT_DIRECT_WT(
1211 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE0),
1212 			size);
1213 		MMSCH_V1_0_INSERT_DIRECT_WT(
1214 			SOC15_REG_OFFSET(VCN, i,
1215 				mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1216 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset));
1217 		MMSCH_V1_0_INSERT_DIRECT_WT(
1218 			SOC15_REG_OFFSET(VCN, i,
1219 				mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1220 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset));
1221 		MMSCH_V1_0_INSERT_DIRECT_WT(
1222 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET1),
1223 			0);
1224 		MMSCH_V1_0_INSERT_DIRECT_WT(
1225 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE1),
1226 			AMDGPU_VCN_STACK_SIZE);
1227 		MMSCH_V1_0_INSERT_DIRECT_WT(
1228 			SOC15_REG_OFFSET(VCN, i,
1229 				mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1230 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset +
1231 				AMDGPU_VCN_STACK_SIZE));
1232 		MMSCH_V1_0_INSERT_DIRECT_WT(
1233 			SOC15_REG_OFFSET(VCN, i,
1234 				mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1235 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset +
1236 				AMDGPU_VCN_STACK_SIZE));
1237 		MMSCH_V1_0_INSERT_DIRECT_WT(
1238 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET2),
1239 			0);
1240 		MMSCH_V1_0_INSERT_DIRECT_WT(
1241 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE2),
1242 			AMDGPU_VCN_CONTEXT_SIZE);
1243 
1244 		ring = &adev->vcn.inst[i].ring_enc[0];
1245 		ring->wptr = 0;
1246 
1247 		MMSCH_V1_0_INSERT_DIRECT_WT(
1248 			SOC15_REG_OFFSET(VCN, i, mmUVD_RB_BASE_LO),
1249 			lower_32_bits(ring->gpu_addr));
1250 		MMSCH_V1_0_INSERT_DIRECT_WT(
1251 			SOC15_REG_OFFSET(VCN, i, mmUVD_RB_BASE_HI),
1252 			upper_32_bits(ring->gpu_addr));
1253 		MMSCH_V1_0_INSERT_DIRECT_WT(
1254 			SOC15_REG_OFFSET(VCN, i, mmUVD_RB_SIZE),
1255 			ring->ring_size / 4);
1256 
1257 		ring = &adev->vcn.inst[i].ring_dec;
1258 		ring->wptr = 0;
1259 		MMSCH_V1_0_INSERT_DIRECT_WT(
1260 			SOC15_REG_OFFSET(VCN, i,
1261 				mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1262 			lower_32_bits(ring->gpu_addr));
1263 		MMSCH_V1_0_INSERT_DIRECT_WT(
1264 			SOC15_REG_OFFSET(VCN, i,
1265 				mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1266 			upper_32_bits(ring->gpu_addr));
1267 
1268 		/* force RBC into idle state */
1269 		rb_bufsz = order_base_2(ring->ring_size);
1270 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1271 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1272 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1273 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1274 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1275 		MMSCH_V1_0_INSERT_DIRECT_WT(
1276 			SOC15_REG_OFFSET(VCN, i, mmUVD_RBC_RB_CNTL), tmp);
1277 
1278 		/* add end packet */
1279 		memcpy((void *)init_table, &end, sizeof(struct mmsch_v1_0_cmd_end));
1280 		table_size += sizeof(struct mmsch_v1_0_cmd_end) / 4;
1281 		init_table += sizeof(struct mmsch_v1_0_cmd_end) / 4;
1282 
1283 		/* refine header */
1284 		header->eng[i].table_size = table_size;
1285 		header->total_size += table_size;
1286 	}
1287 
1288 	return vcn_v2_5_mmsch_start(adev, &adev->virt.mm_table);
1289 }
1290 
1291 static int vcn_v2_5_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1292 {
1293 	uint32_t tmp;
1294 
1295 	/* Wait for power status to be 1 */
1296 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1297 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1298 
1299 	/* wait for read ptr to be equal to write ptr */
1300 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR);
1301 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1302 
1303 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2);
1304 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
1305 
1306 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1307 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF);
1308 
1309 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1310 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1311 
1312 	/* disable dynamic power gating mode */
1313 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 0,
1314 			~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1315 
1316 	return 0;
1317 }
1318 
1319 static int vcn_v2_5_stop(struct amdgpu_device *adev)
1320 {
1321 	uint32_t tmp;
1322 	int i, r = 0;
1323 
1324 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1325 		if (adev->vcn.harvest_config & (1 << i))
1326 			continue;
1327 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1328 			r = vcn_v2_5_stop_dpg_mode(adev, i);
1329 			continue;
1330 		}
1331 
1332 		/* wait for vcn idle */
1333 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1334 		if (r)
1335 			return r;
1336 
1337 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1338 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1339 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1340 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1341 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1342 		if (r)
1343 			return r;
1344 
1345 		/* block LMI UMC channel */
1346 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2);
1347 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1348 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2, tmp);
1349 
1350 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1351 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1352 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1353 		if (r)
1354 			return r;
1355 
1356 		/* block VCPU register access */
1357 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL),
1358 			UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1359 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1360 
1361 		/* reset VCPU */
1362 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1363 			UVD_VCPU_CNTL__BLK_RST_MASK,
1364 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1365 
1366 		/* disable VCPU clock */
1367 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1368 			~(UVD_VCPU_CNTL__CLK_EN_MASK));
1369 
1370 		/* clear status */
1371 		WREG32_SOC15(VCN, i, mmUVD_STATUS, 0);
1372 
1373 		vcn_v2_5_enable_clock_gating(adev);
1374 
1375 		/* enable register anti-hang mechanism */
1376 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_POWER_STATUS),
1377 			UVD_POWER_STATUS__UVD_POWER_STATUS_MASK,
1378 			~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1379 	}
1380 
1381 	if (adev->pm.dpm_enabled)
1382 		amdgpu_dpm_enable_uvd(adev, false);
1383 
1384 	return 0;
1385 }
1386 
1387 static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
1388 				int inst_idx, struct dpg_pause_state *new_state)
1389 {
1390 	struct amdgpu_ring *ring;
1391 	uint32_t reg_data = 0;
1392 	int ret_code = 0;
1393 
1394 	/* pause/unpause if state is changed */
1395 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1396 		DRM_DEBUG("dpg pause state changed %d -> %d",
1397 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1398 		reg_data = RREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE) &
1399 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1400 
1401 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1402 			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1403 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1404 
1405 			if (!ret_code) {
1406 				volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared_cpu_addr;
1407 
1408 				/* pause DPG */
1409 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1410 				WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1411 
1412 				/* wait for ACK */
1413 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_DPG_PAUSE,
1414 					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1415 					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1416 
1417 				/* Stall DPG before WPTR/RPTR reset */
1418 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1419 					   UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1420 					   ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1421 
1422 				/* Restore */
1423 				fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1424 				ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1425 				ring->wptr = 0;
1426 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
1427 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1428 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
1429 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1430 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1431 				fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1432 
1433 				fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1434 				ring = &adev->vcn.inst[inst_idx].ring_enc[1];
1435 				ring->wptr = 0;
1436 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1437 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1438 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
1439 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1440 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1441 				fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1442 
1443 				/* Unstall DPG */
1444 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1445 					   0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1446 
1447 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS,
1448 					   UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1449 			}
1450 		} else {
1451 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1452 			WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1453 			SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1454 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1455 		}
1456 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1457 	}
1458 
1459 	return 0;
1460 }
1461 
1462 /**
1463  * vcn_v2_5_dec_ring_get_rptr - get read pointer
1464  *
1465  * @ring: amdgpu_ring pointer
1466  *
1467  * Returns the current hardware read pointer
1468  */
1469 static uint64_t vcn_v2_5_dec_ring_get_rptr(struct amdgpu_ring *ring)
1470 {
1471 	struct amdgpu_device *adev = ring->adev;
1472 
1473 	return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_RPTR);
1474 }
1475 
1476 /**
1477  * vcn_v2_5_dec_ring_get_wptr - get write pointer
1478  *
1479  * @ring: amdgpu_ring pointer
1480  *
1481  * Returns the current hardware write pointer
1482  */
1483 static uint64_t vcn_v2_5_dec_ring_get_wptr(struct amdgpu_ring *ring)
1484 {
1485 	struct amdgpu_device *adev = ring->adev;
1486 
1487 	if (ring->use_doorbell)
1488 		return adev->wb.wb[ring->wptr_offs];
1489 	else
1490 		return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR);
1491 }
1492 
1493 /**
1494  * vcn_v2_5_dec_ring_set_wptr - set write pointer
1495  *
1496  * @ring: amdgpu_ring pointer
1497  *
1498  * Commits the write pointer to the hardware
1499  */
1500 static void vcn_v2_5_dec_ring_set_wptr(struct amdgpu_ring *ring)
1501 {
1502 	struct amdgpu_device *adev = ring->adev;
1503 
1504 	if (ring->use_doorbell) {
1505 		adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1506 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1507 	} else {
1508 		WREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1509 	}
1510 }
1511 
1512 static const struct amdgpu_ring_funcs vcn_v2_5_dec_ring_vm_funcs = {
1513 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1514 	.align_mask = 0xf,
1515 	.vmhub = AMDGPU_MMHUB_1,
1516 	.get_rptr = vcn_v2_5_dec_ring_get_rptr,
1517 	.get_wptr = vcn_v2_5_dec_ring_get_wptr,
1518 	.set_wptr = vcn_v2_5_dec_ring_set_wptr,
1519 	.emit_frame_size =
1520 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1521 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1522 		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1523 		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1524 		6,
1525 	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1526 	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
1527 	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
1528 	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1529 	.test_ring = vcn_v2_0_dec_ring_test_ring,
1530 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
1531 	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
1532 	.insert_start = vcn_v2_0_dec_ring_insert_start,
1533 	.insert_end = vcn_v2_0_dec_ring_insert_end,
1534 	.pad_ib = amdgpu_ring_generic_pad_ib,
1535 	.begin_use = amdgpu_vcn_ring_begin_use,
1536 	.end_use = amdgpu_vcn_ring_end_use,
1537 	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1538 	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1539 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1540 };
1541 
1542 static const struct amdgpu_ring_funcs vcn_v2_6_dec_ring_vm_funcs = {
1543 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1544 	.align_mask = 0xf,
1545 	.vmhub = AMDGPU_MMHUB_0,
1546 	.get_rptr = vcn_v2_5_dec_ring_get_rptr,
1547 	.get_wptr = vcn_v2_5_dec_ring_get_wptr,
1548 	.set_wptr = vcn_v2_5_dec_ring_set_wptr,
1549 	.emit_frame_size =
1550 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1551 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1552 		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1553 		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1554 		6,
1555 	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1556 	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
1557 	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
1558 	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1559 	.test_ring = vcn_v2_0_dec_ring_test_ring,
1560 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
1561 	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
1562 	.insert_start = vcn_v2_0_dec_ring_insert_start,
1563 	.insert_end = vcn_v2_0_dec_ring_insert_end,
1564 	.pad_ib = amdgpu_ring_generic_pad_ib,
1565 	.begin_use = amdgpu_vcn_ring_begin_use,
1566 	.end_use = amdgpu_vcn_ring_end_use,
1567 	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1568 	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1569 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1570 };
1571 
1572 /**
1573  * vcn_v2_5_enc_ring_get_rptr - get enc read pointer
1574  *
1575  * @ring: amdgpu_ring pointer
1576  *
1577  * Returns the current hardware enc read pointer
1578  */
1579 static uint64_t vcn_v2_5_enc_ring_get_rptr(struct amdgpu_ring *ring)
1580 {
1581 	struct amdgpu_device *adev = ring->adev;
1582 
1583 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0])
1584 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR);
1585 	else
1586 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR2);
1587 }
1588 
1589 /**
1590  * vcn_v2_5_enc_ring_get_wptr - get enc write pointer
1591  *
1592  * @ring: amdgpu_ring pointer
1593  *
1594  * Returns the current hardware enc write pointer
1595  */
1596 static uint64_t vcn_v2_5_enc_ring_get_wptr(struct amdgpu_ring *ring)
1597 {
1598 	struct amdgpu_device *adev = ring->adev;
1599 
1600 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1601 		if (ring->use_doorbell)
1602 			return adev->wb.wb[ring->wptr_offs];
1603 		else
1604 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR);
1605 	} else {
1606 		if (ring->use_doorbell)
1607 			return adev->wb.wb[ring->wptr_offs];
1608 		else
1609 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2);
1610 	}
1611 }
1612 
1613 /**
1614  * vcn_v2_5_enc_ring_set_wptr - set enc write pointer
1615  *
1616  * @ring: amdgpu_ring pointer
1617  *
1618  * Commits the enc write pointer to the hardware
1619  */
1620 static void vcn_v2_5_enc_ring_set_wptr(struct amdgpu_ring *ring)
1621 {
1622 	struct amdgpu_device *adev = ring->adev;
1623 
1624 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1625 		if (ring->use_doorbell) {
1626 			adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1627 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1628 		} else {
1629 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1630 		}
1631 	} else {
1632 		if (ring->use_doorbell) {
1633 			adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1634 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1635 		} else {
1636 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1637 		}
1638 	}
1639 }
1640 
1641 static const struct amdgpu_ring_funcs vcn_v2_5_enc_ring_vm_funcs = {
1642 	.type = AMDGPU_RING_TYPE_VCN_ENC,
1643 	.align_mask = 0x3f,
1644 	.nop = VCN_ENC_CMD_NO_OP,
1645 	.vmhub = AMDGPU_MMHUB_1,
1646 	.get_rptr = vcn_v2_5_enc_ring_get_rptr,
1647 	.get_wptr = vcn_v2_5_enc_ring_get_wptr,
1648 	.set_wptr = vcn_v2_5_enc_ring_set_wptr,
1649 	.emit_frame_size =
1650 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1651 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1652 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1653 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1654 		1, /* vcn_v2_0_enc_ring_insert_end */
1655 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1656 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
1657 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
1658 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1659 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
1660 	.test_ib = amdgpu_vcn_enc_ring_test_ib,
1661 	.insert_nop = amdgpu_ring_insert_nop,
1662 	.insert_end = vcn_v2_0_enc_ring_insert_end,
1663 	.pad_ib = amdgpu_ring_generic_pad_ib,
1664 	.begin_use = amdgpu_vcn_ring_begin_use,
1665 	.end_use = amdgpu_vcn_ring_end_use,
1666 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1667 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1668 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1669 };
1670 
1671 static const struct amdgpu_ring_funcs vcn_v2_6_enc_ring_vm_funcs = {
1672         .type = AMDGPU_RING_TYPE_VCN_ENC,
1673         .align_mask = 0x3f,
1674         .nop = VCN_ENC_CMD_NO_OP,
1675         .vmhub = AMDGPU_MMHUB_0,
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 		if (adev->ip_versions[UVD_HWIP][0] == IP_VERSION(2, 5, 0))
1709 			adev->vcn.inst[i].ring_dec.funcs = &vcn_v2_5_dec_ring_vm_funcs;
1710 		else /* CHIP_ALDEBARAN */
1711 			adev->vcn.inst[i].ring_dec.funcs = &vcn_v2_6_dec_ring_vm_funcs;
1712 		adev->vcn.inst[i].ring_dec.me = i;
1713 		DRM_INFO("VCN(%d) decode is enabled in VM mode\n", i);
1714 	}
1715 }
1716 
1717 static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev)
1718 {
1719 	int i, j;
1720 
1721 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
1722 		if (adev->vcn.harvest_config & (1 << j))
1723 			continue;
1724 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
1725 			if (adev->ip_versions[UVD_HWIP][0] == IP_VERSION(2, 5, 0))
1726 				adev->vcn.inst[j].ring_enc[i].funcs = &vcn_v2_5_enc_ring_vm_funcs;
1727 			else /* CHIP_ALDEBARAN */
1728 				adev->vcn.inst[j].ring_enc[i].funcs = &vcn_v2_6_enc_ring_vm_funcs;
1729 			adev->vcn.inst[j].ring_enc[i].me = j;
1730 		}
1731 		DRM_INFO("VCN(%d) encode is enabled in VM mode\n", j);
1732 	}
1733 }
1734 
1735 static bool vcn_v2_5_is_idle(void *handle)
1736 {
1737 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1738 	int i, ret = 1;
1739 
1740 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1741 		if (adev->vcn.harvest_config & (1 << i))
1742 			continue;
1743 		ret &= (RREG32_SOC15(VCN, i, mmUVD_STATUS) == UVD_STATUS__IDLE);
1744 	}
1745 
1746 	return ret;
1747 }
1748 
1749 static int vcn_v2_5_wait_for_idle(void *handle)
1750 {
1751 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1752 	int i, ret = 0;
1753 
1754 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1755 		if (adev->vcn.harvest_config & (1 << i))
1756 			continue;
1757 		ret = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE,
1758 			UVD_STATUS__IDLE);
1759 		if (ret)
1760 			return ret;
1761 	}
1762 
1763 	return ret;
1764 }
1765 
1766 static int vcn_v2_5_set_clockgating_state(void *handle,
1767 					  enum amd_clockgating_state state)
1768 {
1769 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1770 	bool enable = (state == AMD_CG_STATE_GATE);
1771 
1772 	if (amdgpu_sriov_vf(adev))
1773 		return 0;
1774 
1775 	if (enable) {
1776 		if (!vcn_v2_5_is_idle(handle))
1777 			return -EBUSY;
1778 		vcn_v2_5_enable_clock_gating(adev);
1779 	} else {
1780 		vcn_v2_5_disable_clock_gating(adev);
1781 	}
1782 
1783 	return 0;
1784 }
1785 
1786 static int vcn_v2_5_set_powergating_state(void *handle,
1787 					  enum amd_powergating_state state)
1788 {
1789 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1790 	int ret;
1791 
1792 	if (amdgpu_sriov_vf(adev))
1793 		return 0;
1794 
1795 	if(state == adev->vcn.cur_state)
1796 		return 0;
1797 
1798 	if (state == AMD_PG_STATE_GATE)
1799 		ret = vcn_v2_5_stop(adev);
1800 	else
1801 		ret = vcn_v2_5_start(adev);
1802 
1803 	if(!ret)
1804 		adev->vcn.cur_state = state;
1805 
1806 	return ret;
1807 }
1808 
1809 static int vcn_v2_5_set_interrupt_state(struct amdgpu_device *adev,
1810 					struct amdgpu_irq_src *source,
1811 					unsigned type,
1812 					enum amdgpu_interrupt_state state)
1813 {
1814 	return 0;
1815 }
1816 
1817 static int vcn_v2_5_process_interrupt(struct amdgpu_device *adev,
1818 				      struct amdgpu_irq_src *source,
1819 				      struct amdgpu_iv_entry *entry)
1820 {
1821 	uint32_t ip_instance;
1822 
1823 	switch (entry->client_id) {
1824 	case SOC15_IH_CLIENTID_VCN:
1825 		ip_instance = 0;
1826 		break;
1827 	case SOC15_IH_CLIENTID_VCN1:
1828 		ip_instance = 1;
1829 		break;
1830 	default:
1831 		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
1832 		return 0;
1833 	}
1834 
1835 	DRM_DEBUG("IH: VCN TRAP\n");
1836 
1837 	switch (entry->src_id) {
1838 	case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
1839 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec);
1840 		break;
1841 	case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
1842 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
1843 		break;
1844 	case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
1845 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]);
1846 		break;
1847 	default:
1848 		DRM_ERROR("Unhandled interrupt: %d %d\n",
1849 			  entry->src_id, entry->src_data[0]);
1850 		break;
1851 	}
1852 
1853 	return 0;
1854 }
1855 
1856 static const struct amdgpu_irq_src_funcs vcn_v2_5_irq_funcs = {
1857 	.set = vcn_v2_5_set_interrupt_state,
1858 	.process = vcn_v2_5_process_interrupt,
1859 };
1860 
1861 static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev)
1862 {
1863 	int i;
1864 
1865 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1866 		if (adev->vcn.harvest_config & (1 << i))
1867 			continue;
1868 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
1869 		adev->vcn.inst[i].irq.funcs = &vcn_v2_5_irq_funcs;
1870 	}
1871 }
1872 
1873 static const struct amd_ip_funcs vcn_v2_5_ip_funcs = {
1874 	.name = "vcn_v2_5",
1875 	.early_init = vcn_v2_5_early_init,
1876 	.late_init = NULL,
1877 	.sw_init = vcn_v2_5_sw_init,
1878 	.sw_fini = vcn_v2_5_sw_fini,
1879 	.hw_init = vcn_v2_5_hw_init,
1880 	.hw_fini = vcn_v2_5_hw_fini,
1881 	.suspend = vcn_v2_5_suspend,
1882 	.resume = vcn_v2_5_resume,
1883 	.is_idle = vcn_v2_5_is_idle,
1884 	.wait_for_idle = vcn_v2_5_wait_for_idle,
1885 	.check_soft_reset = NULL,
1886 	.pre_soft_reset = NULL,
1887 	.soft_reset = NULL,
1888 	.post_soft_reset = NULL,
1889 	.set_clockgating_state = vcn_v2_5_set_clockgating_state,
1890 	.set_powergating_state = vcn_v2_5_set_powergating_state,
1891 };
1892 
1893 static const struct amd_ip_funcs vcn_v2_6_ip_funcs = {
1894         .name = "vcn_v2_6",
1895         .early_init = vcn_v2_5_early_init,
1896         .late_init = NULL,
1897         .sw_init = vcn_v2_5_sw_init,
1898         .sw_fini = vcn_v2_5_sw_fini,
1899         .hw_init = vcn_v2_5_hw_init,
1900         .hw_fini = vcn_v2_5_hw_fini,
1901         .suspend = vcn_v2_5_suspend,
1902         .resume = vcn_v2_5_resume,
1903         .is_idle = vcn_v2_5_is_idle,
1904         .wait_for_idle = vcn_v2_5_wait_for_idle,
1905         .check_soft_reset = NULL,
1906         .pre_soft_reset = NULL,
1907         .soft_reset = NULL,
1908         .post_soft_reset = NULL,
1909         .set_clockgating_state = vcn_v2_5_set_clockgating_state,
1910         .set_powergating_state = vcn_v2_5_set_powergating_state,
1911 };
1912 
1913 const struct amdgpu_ip_block_version vcn_v2_5_ip_block =
1914 {
1915 		.type = AMD_IP_BLOCK_TYPE_VCN,
1916 		.major = 2,
1917 		.minor = 5,
1918 		.rev = 0,
1919 		.funcs = &vcn_v2_5_ip_funcs,
1920 };
1921 
1922 const struct amdgpu_ip_block_version vcn_v2_6_ip_block =
1923 {
1924 		.type = AMD_IP_BLOCK_TYPE_VCN,
1925 		.major = 2,
1926 		.minor = 6,
1927 		.rev = 0,
1928 		.funcs = &vcn_v2_6_ip_funcs,
1929 };
1930