xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c (revision 6c33a6f4)
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 
26 #include "amdgpu.h"
27 #include "amdgpu_vcn.h"
28 #include "amdgpu_pm.h"
29 #include "soc15.h"
30 #include "soc15d.h"
31 #include "vcn_v2_0.h"
32 #include "mmsch_v1_0.h"
33 
34 #include "vcn/vcn_2_5_offset.h"
35 #include "vcn/vcn_2_5_sh_mask.h"
36 #include "ivsrcid/vcn/irqsrcs_vcn_2_0.h"
37 
38 #define mmUVD_CONTEXT_ID_INTERNAL_OFFSET			0x27
39 #define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET			0x0f
40 #define mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET			0x10
41 #define mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET			0x11
42 #define mmUVD_NO_OP_INTERNAL_OFFSET				0x29
43 #define mmUVD_GP_SCRATCH8_INTERNAL_OFFSET			0x66
44 #define mmUVD_SCRATCH9_INTERNAL_OFFSET				0xc01d
45 
46 #define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET			0x431
47 #define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET		0x3b4
48 #define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET 	0x3b5
49 #define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET			0x25c
50 
51 #define VCN25_MAX_HW_INSTANCES_ARCTURUS			2
52 
53 static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev);
54 static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev);
55 static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev);
56 static int vcn_v2_5_set_powergating_state(void *handle,
57 				enum amd_powergating_state state);
58 static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
59 				int inst_idx, struct dpg_pause_state *new_state);
60 static int vcn_v2_5_sriov_start(struct amdgpu_device *adev);
61 
62 static int amdgpu_ih_clientid_vcns[] = {
63 	SOC15_IH_CLIENTID_VCN,
64 	SOC15_IH_CLIENTID_VCN1
65 };
66 
67 /**
68  * vcn_v2_5_early_init - set function pointers
69  *
70  * @handle: amdgpu_device pointer
71  *
72  * Set ring and irq function pointers
73  */
74 static int vcn_v2_5_early_init(void *handle)
75 {
76 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
77 
78 	if (amdgpu_sriov_vf(adev)) {
79 		adev->vcn.num_vcn_inst = 2;
80 		adev->vcn.harvest_config = 0;
81 		adev->vcn.num_enc_rings = 1;
82 	} else {
83 		if (adev->asic_type == CHIP_ARCTURUS) {
84 			u32 harvest;
85 			int i;
86 
87 			adev->vcn.num_vcn_inst = VCN25_MAX_HW_INSTANCES_ARCTURUS;
88 			for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
89 				harvest = RREG32_SOC15(UVD, i, mmCC_UVD_HARVESTING);
90 				if (harvest & CC_UVD_HARVESTING__UVD_DISABLE_MASK)
91 					adev->vcn.harvest_config |= 1 << i;
92 			}
93 
94 			if (adev->vcn.harvest_config == (AMDGPU_VCN_HARVEST_VCN0 |
95 						AMDGPU_VCN_HARVEST_VCN1))
96 				/* both instances are harvested, disable the block */
97 				return -ENOENT;
98 		} else
99 			adev->vcn.num_vcn_inst = 1;
100 
101 		adev->vcn.num_enc_rings = 2;
102 	}
103 
104 	vcn_v2_5_set_dec_ring_funcs(adev);
105 	vcn_v2_5_set_enc_ring_funcs(adev);
106 	vcn_v2_5_set_irq_funcs(adev);
107 
108 	return 0;
109 }
110 
111 /**
112  * vcn_v2_5_sw_init - sw init for VCN block
113  *
114  * @handle: amdgpu_device pointer
115  *
116  * Load firmware and sw initialization
117  */
118 static int vcn_v2_5_sw_init(void *handle)
119 {
120 	struct amdgpu_ring *ring;
121 	int i, j, r;
122 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
123 
124 	for (j = 0; j < adev->vcn.num_vcn_inst; j++) {
125 		if (adev->vcn.harvest_config & (1 << j))
126 			continue;
127 		/* VCN DEC TRAP */
128 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
129 				VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst[j].irq);
130 		if (r)
131 			return r;
132 
133 		/* VCN ENC TRAP */
134 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
135 			r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
136 				i + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[j].irq);
137 			if (r)
138 				return r;
139 		}
140 	}
141 
142 	r = amdgpu_vcn_sw_init(adev);
143 	if (r)
144 		return r;
145 
146 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
147 		const struct common_firmware_header *hdr;
148 		hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
149 		adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].ucode_id = AMDGPU_UCODE_ID_VCN;
150 		adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].fw = adev->vcn.fw;
151 		adev->firmware.fw_size +=
152 			ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
153 
154 		if (adev->vcn.num_vcn_inst == VCN25_MAX_HW_INSTANCES_ARCTURUS) {
155 			adev->firmware.ucode[AMDGPU_UCODE_ID_VCN1].ucode_id = AMDGPU_UCODE_ID_VCN1;
156 			adev->firmware.ucode[AMDGPU_UCODE_ID_VCN1].fw = adev->vcn.fw;
157 			adev->firmware.fw_size +=
158 				ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
159 		}
160 		DRM_INFO("PSP loading VCN firmware\n");
161 	}
162 
163 	r = amdgpu_vcn_resume(adev);
164 	if (r)
165 		return r;
166 
167 	for (j = 0; j < adev->vcn.num_vcn_inst; j++) {
168 		if (adev->vcn.harvest_config & (1 << j))
169 			continue;
170 		adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
171 		adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
172 		adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
173 		adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
174 		adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
175 		adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
176 
177 		adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
178 		adev->vcn.inst[j].external.scratch9 = SOC15_REG_OFFSET(UVD, j, mmUVD_SCRATCH9);
179 		adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
180 		adev->vcn.inst[j].external.data0 = SOC15_REG_OFFSET(UVD, j, mmUVD_GPCOM_VCPU_DATA0);
181 		adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
182 		adev->vcn.inst[j].external.data1 = SOC15_REG_OFFSET(UVD, j, mmUVD_GPCOM_VCPU_DATA1);
183 		adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
184 		adev->vcn.inst[j].external.cmd = SOC15_REG_OFFSET(UVD, j, mmUVD_GPCOM_VCPU_CMD);
185 		adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
186 		adev->vcn.inst[j].external.nop = SOC15_REG_OFFSET(UVD, j, mmUVD_NO_OP);
187 
188 		ring = &adev->vcn.inst[j].ring_dec;
189 		ring->use_doorbell = true;
190 
191 		ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
192 				(amdgpu_sriov_vf(adev) ? 2*j : 8*j);
193 		sprintf(ring->name, "vcn_dec_%d", j);
194 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[j].irq, 0);
195 		if (r)
196 			return r;
197 
198 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
199 			ring = &adev->vcn.inst[j].ring_enc[i];
200 			ring->use_doorbell = true;
201 
202 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
203 					(amdgpu_sriov_vf(adev) ? (1 + i + 2*j) : (2 + i + 8*j));
204 
205 			sprintf(ring->name, "vcn_enc_%d.%d", j, i);
206 			r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[j].irq, 0);
207 			if (r)
208 				return r;
209 		}
210 	}
211 
212 	if (amdgpu_sriov_vf(adev)) {
213 		r = amdgpu_virt_alloc_mm_table(adev);
214 		if (r)
215 			return r;
216 	}
217 
218 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
219 		adev->vcn.pause_dpg_mode = vcn_v2_5_pause_dpg_mode;
220 
221 	return 0;
222 }
223 
224 /**
225  * vcn_v2_5_sw_fini - sw fini for VCN block
226  *
227  * @handle: amdgpu_device pointer
228  *
229  * VCN suspend and free up sw allocation
230  */
231 static int vcn_v2_5_sw_fini(void *handle)
232 {
233 	int r;
234 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
235 
236 	if (amdgpu_sriov_vf(adev))
237 		amdgpu_virt_free_mm_table(adev);
238 
239 	r = amdgpu_vcn_suspend(adev);
240 	if (r)
241 		return r;
242 
243 	r = amdgpu_vcn_sw_fini(adev);
244 
245 	return r;
246 }
247 
248 /**
249  * vcn_v2_5_hw_init - start and test VCN block
250  *
251  * @handle: amdgpu_device pointer
252  *
253  * Initialize the hardware, boot up the VCPU and do some testing
254  */
255 static int vcn_v2_5_hw_init(void *handle)
256 {
257 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
258 	struct amdgpu_ring *ring;
259 	int i, j, r = 0;
260 
261 	if (amdgpu_sriov_vf(adev))
262 		r = vcn_v2_5_sriov_start(adev);
263 
264 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
265 		if (adev->vcn.harvest_config & (1 << j))
266 			continue;
267 
268 		if (amdgpu_sriov_vf(adev)) {
269 			adev->vcn.inst[j].ring_enc[0].sched.ready = true;
270 			adev->vcn.inst[j].ring_enc[1].sched.ready = false;
271 			adev->vcn.inst[j].ring_enc[2].sched.ready = false;
272 			adev->vcn.inst[j].ring_dec.sched.ready = true;
273 		} else {
274 
275 			ring = &adev->vcn.inst[j].ring_dec;
276 
277 			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
278 						     ring->doorbell_index, j);
279 
280 			r = amdgpu_ring_test_helper(ring);
281 			if (r)
282 				goto done;
283 
284 			for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
285 				ring = &adev->vcn.inst[j].ring_enc[i];
286 				r = amdgpu_ring_test_helper(ring);
287 				if (r)
288 					goto done;
289 			}
290 		}
291 	}
292 
293 done:
294 	if (!r)
295 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
296 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
297 
298 	return r;
299 }
300 
301 /**
302  * vcn_v2_5_hw_fini - stop the hardware block
303  *
304  * @handle: amdgpu_device pointer
305  *
306  * Stop the VCN block, mark ring as not ready any more
307  */
308 static int vcn_v2_5_hw_fini(void *handle)
309 {
310 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
311 	struct amdgpu_ring *ring;
312 	int i, j;
313 
314 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
315 		if (adev->vcn.harvest_config & (1 << i))
316 			continue;
317 		ring = &adev->vcn.inst[i].ring_dec;
318 
319 		if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
320 		    (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
321 		     RREG32_SOC15(VCN, i, mmUVD_STATUS)))
322 			vcn_v2_5_set_powergating_state(adev, AMD_PG_STATE_GATE);
323 
324 		ring->sched.ready = false;
325 
326 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
327 			ring = &adev->vcn.inst[i].ring_enc[j];
328 			ring->sched.ready = false;
329 		}
330 	}
331 
332 	return 0;
333 }
334 
335 /**
336  * vcn_v2_5_suspend - suspend VCN block
337  *
338  * @handle: amdgpu_device pointer
339  *
340  * HW fini and suspend VCN block
341  */
342 static int vcn_v2_5_suspend(void *handle)
343 {
344 	int r;
345 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
346 
347 	r = vcn_v2_5_hw_fini(adev);
348 	if (r)
349 		return r;
350 
351 	r = amdgpu_vcn_suspend(adev);
352 
353 	return r;
354 }
355 
356 /**
357  * vcn_v2_5_resume - resume VCN block
358  *
359  * @handle: amdgpu_device pointer
360  *
361  * Resume firmware and hw init VCN block
362  */
363 static int vcn_v2_5_resume(void *handle)
364 {
365 	int r;
366 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
367 
368 	r = amdgpu_vcn_resume(adev);
369 	if (r)
370 		return r;
371 
372 	r = vcn_v2_5_hw_init(adev);
373 
374 	return r;
375 }
376 
377 /**
378  * vcn_v2_5_mc_resume - memory controller programming
379  *
380  * @adev: amdgpu_device pointer
381  *
382  * Let the VCN memory controller know it's offsets
383  */
384 static void vcn_v2_5_mc_resume(struct amdgpu_device *adev)
385 {
386 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
387 	uint32_t offset;
388 	int i;
389 
390 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
391 		if (adev->vcn.harvest_config & (1 << i))
392 			continue;
393 		/* cache window 0: fw */
394 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
395 			WREG32_SOC15(UVD, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
396 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo));
397 			WREG32_SOC15(UVD, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
398 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi));
399 			WREG32_SOC15(UVD, i, mmUVD_VCPU_CACHE_OFFSET0, 0);
400 			offset = 0;
401 		} else {
402 			WREG32_SOC15(UVD, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
403 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
404 			WREG32_SOC15(UVD, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
405 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
406 			offset = size;
407 			WREG32_SOC15(UVD, i, mmUVD_VCPU_CACHE_OFFSET0,
408 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
409 		}
410 		WREG32_SOC15(UVD, i, mmUVD_VCPU_CACHE_SIZE0, size);
411 
412 		/* cache window 1: stack */
413 		WREG32_SOC15(UVD, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
414 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset));
415 		WREG32_SOC15(UVD, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
416 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset));
417 		WREG32_SOC15(UVD, i, mmUVD_VCPU_CACHE_OFFSET1, 0);
418 		WREG32_SOC15(UVD, i, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
419 
420 		/* cache window 2: context */
421 		WREG32_SOC15(UVD, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
422 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
423 		WREG32_SOC15(UVD, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
424 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
425 		WREG32_SOC15(UVD, i, mmUVD_VCPU_CACHE_OFFSET2, 0);
426 		WREG32_SOC15(UVD, i, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
427 	}
428 }
429 
430 static void vcn_v2_5_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
431 {
432 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
433 	uint32_t offset;
434 
435 	/* cache window 0: fw */
436 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
437 		if (!indirect) {
438 			WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
439 				UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
440 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
441 			WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
442 				UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
443 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
444 			WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
445 				UVD, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
446 		} else {
447 			WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
448 				UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
449 			WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
450 				UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
451 			WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
452 				UVD, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
453 		}
454 		offset = 0;
455 	} else {
456 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
457 			UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
458 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
459 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
460 			UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
461 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
462 		offset = size;
463 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
464 			UVD, 0, mmUVD_VCPU_CACHE_OFFSET0),
465 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
466 	}
467 
468 	if (!indirect)
469 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
470 			UVD, 0, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
471 	else
472 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
473 			UVD, 0, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
474 
475 	/* cache window 1: stack */
476 	if (!indirect) {
477 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
478 			UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
479 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
480 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
481 			UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
482 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
483 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
484 			UVD, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
485 	} else {
486 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
487 			UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
488 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
489 			UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
490 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
491 			UVD, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
492 	}
493 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
494 		UVD, 0, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
495 
496 	/* cache window 2: context */
497 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
498 		UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
499 		lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
500 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
501 		UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
502 		upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
503 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
504 		UVD, 0, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
505 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
506 		UVD, 0, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
507 
508 	/* non-cache window */
509 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
510 		UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW), 0, 0, indirect);
511 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
512 		UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH), 0, 0, indirect);
513 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
514 		UVD, 0, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
515 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
516 		UVD, 0, mmUVD_VCPU_NONCACHE_SIZE0), 0, 0, indirect);
517 
518 	/* VCN global tiling registers */
519 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
520 		UVD, 0, mmUVD_GFX8_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
521 }
522 
523 /**
524  * vcn_v2_5_disable_clock_gating - disable VCN clock gating
525  *
526  * @adev: amdgpu_device pointer
527  *
528  * Disable clock gating for VCN block
529  */
530 static void vcn_v2_5_disable_clock_gating(struct amdgpu_device *adev)
531 {
532 	uint32_t data;
533 	int ret = 0;
534 	int i;
535 
536 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
537 		if (adev->vcn.harvest_config & (1 << i))
538 			continue;
539 		/* UVD disable CGC */
540 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
541 		if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
542 			data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
543 		else
544 			data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
545 		data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
546 		data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
547 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
548 
549 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_GATE);
550 		data &= ~(UVD_CGC_GATE__SYS_MASK
551 			| UVD_CGC_GATE__UDEC_MASK
552 			| UVD_CGC_GATE__MPEG2_MASK
553 			| UVD_CGC_GATE__REGS_MASK
554 			| UVD_CGC_GATE__RBC_MASK
555 			| UVD_CGC_GATE__LMI_MC_MASK
556 			| UVD_CGC_GATE__LMI_UMC_MASK
557 			| UVD_CGC_GATE__IDCT_MASK
558 			| UVD_CGC_GATE__MPRD_MASK
559 			| UVD_CGC_GATE__MPC_MASK
560 			| UVD_CGC_GATE__LBSI_MASK
561 			| UVD_CGC_GATE__LRBBM_MASK
562 			| UVD_CGC_GATE__UDEC_RE_MASK
563 			| UVD_CGC_GATE__UDEC_CM_MASK
564 			| UVD_CGC_GATE__UDEC_IT_MASK
565 			| UVD_CGC_GATE__UDEC_DB_MASK
566 			| UVD_CGC_GATE__UDEC_MP_MASK
567 			| UVD_CGC_GATE__WCB_MASK
568 			| UVD_CGC_GATE__VCPU_MASK
569 			| UVD_CGC_GATE__MMSCH_MASK);
570 
571 		WREG32_SOC15(VCN, i, mmUVD_CGC_GATE, data);
572 
573 		SOC15_WAIT_ON_RREG(VCN, i, mmUVD_CGC_GATE, 0,  0xFFFFFFFF, ret);
574 
575 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
576 		data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
577 			| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
578 			| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
579 			| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
580 			| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
581 			| UVD_CGC_CTRL__SYS_MODE_MASK
582 			| UVD_CGC_CTRL__UDEC_MODE_MASK
583 			| UVD_CGC_CTRL__MPEG2_MODE_MASK
584 			| UVD_CGC_CTRL__REGS_MODE_MASK
585 			| UVD_CGC_CTRL__RBC_MODE_MASK
586 			| UVD_CGC_CTRL__LMI_MC_MODE_MASK
587 			| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
588 			| UVD_CGC_CTRL__IDCT_MODE_MASK
589 			| UVD_CGC_CTRL__MPRD_MODE_MASK
590 			| UVD_CGC_CTRL__MPC_MODE_MASK
591 			| UVD_CGC_CTRL__LBSI_MODE_MASK
592 			| UVD_CGC_CTRL__LRBBM_MODE_MASK
593 			| UVD_CGC_CTRL__WCB_MODE_MASK
594 			| UVD_CGC_CTRL__VCPU_MODE_MASK
595 			| UVD_CGC_CTRL__MMSCH_MODE_MASK);
596 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
597 
598 		/* turn on */
599 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE);
600 		data |= (UVD_SUVD_CGC_GATE__SRE_MASK
601 			| UVD_SUVD_CGC_GATE__SIT_MASK
602 			| UVD_SUVD_CGC_GATE__SMP_MASK
603 			| UVD_SUVD_CGC_GATE__SCM_MASK
604 			| UVD_SUVD_CGC_GATE__SDB_MASK
605 			| UVD_SUVD_CGC_GATE__SRE_H264_MASK
606 			| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
607 			| UVD_SUVD_CGC_GATE__SIT_H264_MASK
608 			| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
609 			| UVD_SUVD_CGC_GATE__SCM_H264_MASK
610 			| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
611 			| UVD_SUVD_CGC_GATE__SDB_H264_MASK
612 			| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
613 			| UVD_SUVD_CGC_GATE__SCLR_MASK
614 			| UVD_SUVD_CGC_GATE__UVD_SC_MASK
615 			| UVD_SUVD_CGC_GATE__ENT_MASK
616 			| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
617 			| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
618 			| UVD_SUVD_CGC_GATE__SITE_MASK
619 			| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
620 			| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
621 			| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
622 			| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
623 			| UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
624 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE, data);
625 
626 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL);
627 		data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
628 			| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
629 			| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
630 			| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
631 			| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
632 			| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
633 			| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
634 			| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
635 			| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
636 			| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
637 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data);
638 	}
639 }
640 
641 static void vcn_v2_5_clock_gating_dpg_mode(struct amdgpu_device *adev,
642 		uint8_t sram_sel, int inst_idx, uint8_t indirect)
643 {
644 	uint32_t reg_data = 0;
645 
646 	/* enable sw clock gating control */
647 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
648 		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
649 	else
650 		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
651 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
652 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
653 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
654 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
655 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
656 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
657 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
658 		 UVD_CGC_CTRL__SYS_MODE_MASK |
659 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
660 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
661 		 UVD_CGC_CTRL__REGS_MODE_MASK |
662 		 UVD_CGC_CTRL__RBC_MODE_MASK |
663 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
664 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
665 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
666 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
667 		 UVD_CGC_CTRL__MPC_MODE_MASK |
668 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
669 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
670 		 UVD_CGC_CTRL__WCB_MODE_MASK |
671 		 UVD_CGC_CTRL__VCPU_MODE_MASK |
672 		 UVD_CGC_CTRL__MMSCH_MODE_MASK);
673 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
674 		UVD, 0, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
675 
676 	/* turn off clock gating */
677 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
678 		UVD, 0, mmUVD_CGC_GATE), 0, sram_sel, indirect);
679 
680 	/* turn on SUVD clock gating */
681 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
682 		UVD, 0, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
683 
684 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
685 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
686 		UVD, 0, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
687 }
688 
689 /**
690  * vcn_v2_5_enable_clock_gating - enable VCN clock gating
691  *
692  * @adev: amdgpu_device pointer
693  *
694  * Enable clock gating for VCN block
695  */
696 static void vcn_v2_5_enable_clock_gating(struct amdgpu_device *adev)
697 {
698 	uint32_t data = 0;
699 	int i;
700 
701 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
702 		if (adev->vcn.harvest_config & (1 << i))
703 			continue;
704 		/* enable UVD CGC */
705 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
706 		if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
707 			data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
708 		else
709 			data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
710 		data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
711 		data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
712 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
713 
714 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
715 		data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
716 			| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
717 			| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
718 			| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
719 			| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
720 			| UVD_CGC_CTRL__SYS_MODE_MASK
721 			| UVD_CGC_CTRL__UDEC_MODE_MASK
722 			| UVD_CGC_CTRL__MPEG2_MODE_MASK
723 			| UVD_CGC_CTRL__REGS_MODE_MASK
724 			| UVD_CGC_CTRL__RBC_MODE_MASK
725 			| UVD_CGC_CTRL__LMI_MC_MODE_MASK
726 			| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
727 			| UVD_CGC_CTRL__IDCT_MODE_MASK
728 			| UVD_CGC_CTRL__MPRD_MODE_MASK
729 			| UVD_CGC_CTRL__MPC_MODE_MASK
730 			| UVD_CGC_CTRL__LBSI_MODE_MASK
731 			| UVD_CGC_CTRL__LRBBM_MODE_MASK
732 			| UVD_CGC_CTRL__WCB_MODE_MASK
733 			| UVD_CGC_CTRL__VCPU_MODE_MASK);
734 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
735 
736 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL);
737 		data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
738 			| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
739 			| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
740 			| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
741 			| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
742 			| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
743 			| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
744 			| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
745 			| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
746 			| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
747 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data);
748 	}
749 }
750 
751 static int vcn_v2_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
752 {
753 	struct amdgpu_ring *ring;
754 	uint32_t rb_bufsz, tmp;
755 
756 	/* disable register anti-hang mechanism */
757 	WREG32_P(SOC15_REG_OFFSET(UVD, inst_idx, mmUVD_POWER_STATUS), 1,
758 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
759 	/* enable dynamic power gating mode */
760 	tmp = RREG32_SOC15(UVD, inst_idx, mmUVD_POWER_STATUS);
761 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
762 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
763 	WREG32_SOC15(UVD, inst_idx, mmUVD_POWER_STATUS, tmp);
764 
765 	if (indirect)
766 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t*)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
767 
768 	/* enable clock gating */
769 	vcn_v2_5_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
770 
771 	/* enable VCPU clock */
772 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
773 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
774 	tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
775 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
776 		UVD, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
777 
778 	/* disable master interupt */
779 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
780 		UVD, 0, mmUVD_MASTINT_EN), 0, 0, indirect);
781 
782 	/* setup mmUVD_LMI_CTRL */
783 	tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
784 		UVD_LMI_CTRL__REQ_MODE_MASK |
785 		UVD_LMI_CTRL__CRC_RESET_MASK |
786 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
787 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
788 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
789 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
790 		0x00100000L);
791 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
792 		UVD, 0, mmUVD_LMI_CTRL), tmp, 0, indirect);
793 
794 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
795 		UVD, 0, mmUVD_MPC_CNTL),
796 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
797 
798 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
799 		UVD, 0, mmUVD_MPC_SET_MUXA0),
800 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
801 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
802 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
803 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
804 
805 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
806 		UVD, 0, mmUVD_MPC_SET_MUXB0),
807 		((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
808 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
809 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
810 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
811 
812 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
813 		UVD, 0, mmUVD_MPC_SET_MUX),
814 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
815 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
816 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
817 
818 	vcn_v2_5_mc_resume_dpg_mode(adev, inst_idx, indirect);
819 
820 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
821 		UVD, 0, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
822 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
823 		UVD, 0, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
824 
825 	/* enable LMI MC and UMC channels */
826 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
827 		UVD, 0, mmUVD_LMI_CTRL2), 0, 0, indirect);
828 
829 	/* unblock VCPU register access */
830 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
831 		UVD, 0, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
832 
833 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
834 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
835 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
836 		UVD, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
837 
838 	/* enable master interrupt */
839 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
840 		UVD, 0, mmUVD_MASTINT_EN),
841 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
842 
843 	if (indirect)
844 		psp_update_vcn_sram(adev, inst_idx, adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
845 				    (uint32_t)((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
846 					       (uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr));
847 
848 	ring = &adev->vcn.inst[inst_idx].ring_dec;
849 	/* force RBC into idle state */
850 	rb_bufsz = order_base_2(ring->ring_size);
851 	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
852 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
853 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
854 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
855 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
856 	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
857 
858 	/* set the write pointer delay */
859 	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
860 
861 	/* set the wb address */
862 	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_RPTR_ADDR,
863 		(upper_32_bits(ring->gpu_addr) >> 2));
864 
865 	/* programm the RB_BASE for ring buffer */
866 	WREG32_SOC15(UVD, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
867 		lower_32_bits(ring->gpu_addr));
868 	WREG32_SOC15(UVD, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
869 		upper_32_bits(ring->gpu_addr));
870 
871 	/* Initialize the ring buffer's read and write pointers */
872 	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_RPTR, 0);
873 
874 	WREG32_SOC15(UVD, inst_idx, mmUVD_SCRATCH2, 0);
875 
876 	ring->wptr = RREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_RPTR);
877 	WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR,
878 		lower_32_bits(ring->wptr));
879 
880 	return 0;
881 }
882 
883 static int vcn_v2_5_start(struct amdgpu_device *adev)
884 {
885 	struct amdgpu_ring *ring;
886 	uint32_t rb_bufsz, tmp;
887 	int i, j, k, r;
888 
889 	if (adev->pm.dpm_enabled)
890 		amdgpu_dpm_enable_uvd(adev, true);
891 
892 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
893 		if (adev->vcn.harvest_config & (1 << i))
894 			continue;
895 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
896 			r = vcn_v2_5_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
897 			continue;
898 		}
899 
900 		/* disable register anti-hang mechanism */
901 		WREG32_P(SOC15_REG_OFFSET(UVD, i, mmUVD_POWER_STATUS), 0,
902 			~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
903 
904 		/* set uvd status busy */
905 		tmp = RREG32_SOC15(UVD, i, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY;
906 		WREG32_SOC15(UVD, i, mmUVD_STATUS, tmp);
907 	}
908 
909 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
910 		return 0;
911 
912 	/*SW clock gating */
913 	vcn_v2_5_disable_clock_gating(adev);
914 
915 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
916 		if (adev->vcn.harvest_config & (1 << i))
917 			continue;
918 		/* enable VCPU clock */
919 		WREG32_P(SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CNTL),
920 			UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
921 
922 		/* disable master interrupt */
923 		WREG32_P(SOC15_REG_OFFSET(UVD, i, mmUVD_MASTINT_EN), 0,
924 			~UVD_MASTINT_EN__VCPU_EN_MASK);
925 
926 		/* setup mmUVD_LMI_CTRL */
927 		tmp = RREG32_SOC15(UVD, i, mmUVD_LMI_CTRL);
928 		tmp &= ~0xff;
929 		WREG32_SOC15(UVD, i, mmUVD_LMI_CTRL, tmp | 0x8|
930 			UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK	|
931 			UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
932 			UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
933 			UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
934 
935 		/* setup mmUVD_MPC_CNTL */
936 		tmp = RREG32_SOC15(UVD, i, mmUVD_MPC_CNTL);
937 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
938 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
939 		WREG32_SOC15(VCN, i, mmUVD_MPC_CNTL, tmp);
940 
941 		/* setup UVD_MPC_SET_MUXA0 */
942 		WREG32_SOC15(UVD, i, mmUVD_MPC_SET_MUXA0,
943 			((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
944 			(0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
945 			(0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
946 			(0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
947 
948 		/* setup UVD_MPC_SET_MUXB0 */
949 		WREG32_SOC15(UVD, i, mmUVD_MPC_SET_MUXB0,
950 			((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
951 			(0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
952 			(0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
953 			(0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
954 
955 		/* setup mmUVD_MPC_SET_MUX */
956 		WREG32_SOC15(UVD, i, mmUVD_MPC_SET_MUX,
957 			((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
958 			(0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
959 			(0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
960 	}
961 
962 	vcn_v2_5_mc_resume(adev);
963 
964 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
965 		if (adev->vcn.harvest_config & (1 << i))
966 			continue;
967 		/* VCN global tiling registers */
968 		WREG32_SOC15(UVD, i, mmUVD_GFX8_ADDR_CONFIG,
969 			adev->gfx.config.gb_addr_config);
970 		WREG32_SOC15(UVD, i, mmUVD_GFX8_ADDR_CONFIG,
971 			adev->gfx.config.gb_addr_config);
972 
973 		/* enable LMI MC and UMC channels */
974 		WREG32_P(SOC15_REG_OFFSET(UVD, i, mmUVD_LMI_CTRL2), 0,
975 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
976 
977 		/* unblock VCPU register access */
978 		WREG32_P(SOC15_REG_OFFSET(UVD, i, mmUVD_RB_ARB_CTRL), 0,
979 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
980 
981 		WREG32_P(SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CNTL), 0,
982 			~UVD_VCPU_CNTL__BLK_RST_MASK);
983 
984 		for (k = 0; k < 10; ++k) {
985 			uint32_t status;
986 
987 			for (j = 0; j < 100; ++j) {
988 				status = RREG32_SOC15(UVD, i, mmUVD_STATUS);
989 				if (status & 2)
990 					break;
991 				if (amdgpu_emu_mode == 1)
992 					msleep(500);
993 				else
994 					mdelay(10);
995 			}
996 			r = 0;
997 			if (status & 2)
998 				break;
999 
1000 			DRM_ERROR("VCN decode not responding, trying to reset the VCPU!!!\n");
1001 			WREG32_P(SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CNTL),
1002 				UVD_VCPU_CNTL__BLK_RST_MASK,
1003 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1004 			mdelay(10);
1005 			WREG32_P(SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CNTL), 0,
1006 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1007 
1008 			mdelay(10);
1009 			r = -1;
1010 		}
1011 
1012 		if (r) {
1013 			DRM_ERROR("VCN decode not responding, giving up!!!\n");
1014 			return r;
1015 		}
1016 
1017 		/* enable master interrupt */
1018 		WREG32_P(SOC15_REG_OFFSET(UVD, i, mmUVD_MASTINT_EN),
1019 			UVD_MASTINT_EN__VCPU_EN_MASK,
1020 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1021 
1022 		/* clear the busy bit of VCN_STATUS */
1023 		WREG32_P(SOC15_REG_OFFSET(UVD, i, mmUVD_STATUS), 0,
1024 			~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1025 
1026 		WREG32_SOC15(UVD, i, mmUVD_LMI_RBC_RB_VMID, 0);
1027 
1028 		ring = &adev->vcn.inst[i].ring_dec;
1029 		/* force RBC into idle state */
1030 		rb_bufsz = order_base_2(ring->ring_size);
1031 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1032 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1033 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1034 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1035 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1036 		WREG32_SOC15(UVD, i, mmUVD_RBC_RB_CNTL, tmp);
1037 
1038 		/* programm the RB_BASE for ring buffer */
1039 		WREG32_SOC15(UVD, i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1040 			lower_32_bits(ring->gpu_addr));
1041 		WREG32_SOC15(UVD, i, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1042 			upper_32_bits(ring->gpu_addr));
1043 
1044 		/* Initialize the ring buffer's read and write pointers */
1045 		WREG32_SOC15(UVD, i, mmUVD_RBC_RB_RPTR, 0);
1046 
1047 		ring->wptr = RREG32_SOC15(UVD, i, mmUVD_RBC_RB_RPTR);
1048 		WREG32_SOC15(UVD, i, mmUVD_RBC_RB_WPTR,
1049 				lower_32_bits(ring->wptr));
1050 		ring = &adev->vcn.inst[i].ring_enc[0];
1051 		WREG32_SOC15(UVD, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1052 		WREG32_SOC15(UVD, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1053 		WREG32_SOC15(UVD, i, mmUVD_RB_BASE_LO, ring->gpu_addr);
1054 		WREG32_SOC15(UVD, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1055 		WREG32_SOC15(UVD, i, mmUVD_RB_SIZE, ring->ring_size / 4);
1056 
1057 		ring = &adev->vcn.inst[i].ring_enc[1];
1058 		WREG32_SOC15(UVD, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1059 		WREG32_SOC15(UVD, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1060 		WREG32_SOC15(UVD, i, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1061 		WREG32_SOC15(UVD, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1062 		WREG32_SOC15(UVD, i, mmUVD_RB_SIZE2, ring->ring_size / 4);
1063 	}
1064 
1065 	return 0;
1066 }
1067 
1068 static int vcn_v2_5_mmsch_start(struct amdgpu_device *adev,
1069 				struct amdgpu_mm_table *table)
1070 {
1071 	uint32_t data = 0, loop = 0, size = 0;
1072 	uint64_t addr = table->gpu_addr;
1073 	struct mmsch_v1_1_init_header *header = NULL;;
1074 
1075 	header = (struct mmsch_v1_1_init_header *)table->cpu_addr;
1076 	size = header->total_size;
1077 
1078 	/*
1079 	 * 1, write to vce_mmsch_vf_ctx_addr_lo/hi register with GPU mc addr of
1080 	 *  memory descriptor location
1081 	 */
1082 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(addr));
1083 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(addr));
1084 
1085 	/* 2, update vmid of descriptor */
1086 	data = RREG32_SOC15(UVD, 0, mmMMSCH_VF_VMID);
1087 	data &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1088 	/* use domain0 for MM scheduler */
1089 	data |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1090 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_VMID, data);
1091 
1092 	/* 3, notify mmsch about the size of this descriptor */
1093 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_CTX_SIZE, size);
1094 
1095 	/* 4, set resp to zero */
1096 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1097 
1098 	/*
1099 	 * 5, kick off the initialization and wait until
1100 	 * VCE_MMSCH_VF_MAILBOX_RESP becomes non-zero
1101 	 */
1102 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_HOST, 0x10000001);
1103 
1104 	data = RREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_RESP);
1105 	loop = 10;
1106 	while ((data & 0x10000002) != 0x10000002) {
1107 		udelay(100);
1108 		data = RREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_RESP);
1109 		loop--;
1110 		if (!loop)
1111 			break;
1112 	}
1113 
1114 	if (!loop) {
1115 		dev_err(adev->dev,
1116 			"failed to init MMSCH, mmMMSCH_VF_MAILBOX_RESP = %x\n",
1117 			data);
1118 		return -EBUSY;
1119 	}
1120 
1121 	return 0;
1122 }
1123 
1124 static int vcn_v2_5_sriov_start(struct amdgpu_device *adev)
1125 {
1126 	struct amdgpu_ring *ring;
1127 	uint32_t offset, size, tmp, i, rb_bufsz;
1128 	uint32_t table_size = 0;
1129 	struct mmsch_v1_0_cmd_direct_write direct_wt = { { 0 } };
1130 	struct mmsch_v1_0_cmd_direct_read_modify_write direct_rd_mod_wt = { { 0 } };
1131 	struct mmsch_v1_0_cmd_direct_polling direct_poll = { { 0 } };
1132 	struct mmsch_v1_0_cmd_end end = { { 0 } };
1133 	uint32_t *init_table = adev->virt.mm_table.cpu_addr;
1134 	struct mmsch_v1_1_init_header *header = (struct mmsch_v1_1_init_header *)init_table;
1135 
1136 	direct_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_WRITE;
1137 	direct_rd_mod_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1138 	direct_poll.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_POLLING;
1139 	end.cmd_header.command_type = MMSCH_COMMAND__END;
1140 
1141 	header->version = MMSCH_VERSION;
1142 	header->total_size = sizeof(struct mmsch_v1_1_init_header) >> 2;
1143 	init_table += header->total_size;
1144 
1145 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1146 		header->eng[i].table_offset = header->total_size;
1147 		header->eng[i].init_status = 0;
1148 		header->eng[i].table_size = 0;
1149 
1150 		table_size = 0;
1151 
1152 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(
1153 			SOC15_REG_OFFSET(UVD, i, mmUVD_STATUS),
1154 			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1155 
1156 		size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1157 		/* mc resume*/
1158 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1159 			MMSCH_V1_0_INSERT_DIRECT_WT(
1160 				SOC15_REG_OFFSET(UVD, i,
1161 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1162 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1163 			MMSCH_V1_0_INSERT_DIRECT_WT(
1164 				SOC15_REG_OFFSET(UVD, i,
1165 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1166 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1167 			offset = 0;
1168 			MMSCH_V1_0_INSERT_DIRECT_WT(
1169 				SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_OFFSET0), 0);
1170 		} else {
1171 			MMSCH_V1_0_INSERT_DIRECT_WT(
1172 				SOC15_REG_OFFSET(UVD, i,
1173 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1174 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1175 			MMSCH_V1_0_INSERT_DIRECT_WT(
1176 				SOC15_REG_OFFSET(UVD, i,
1177 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1178 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1179 			offset = size;
1180 			MMSCH_V1_0_INSERT_DIRECT_WT(
1181 				SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_OFFSET0),
1182 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1183 		}
1184 
1185 		MMSCH_V1_0_INSERT_DIRECT_WT(
1186 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_SIZE0),
1187 			size);
1188 		MMSCH_V1_0_INSERT_DIRECT_WT(
1189 			SOC15_REG_OFFSET(UVD, i,
1190 				mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1191 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset));
1192 		MMSCH_V1_0_INSERT_DIRECT_WT(
1193 			SOC15_REG_OFFSET(UVD, i,
1194 				mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1195 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset));
1196 		MMSCH_V1_0_INSERT_DIRECT_WT(
1197 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_OFFSET1),
1198 			0);
1199 		MMSCH_V1_0_INSERT_DIRECT_WT(
1200 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_SIZE1),
1201 			AMDGPU_VCN_STACK_SIZE);
1202 		MMSCH_V1_0_INSERT_DIRECT_WT(
1203 			SOC15_REG_OFFSET(UVD, i,
1204 				mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1205 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset +
1206 				AMDGPU_VCN_STACK_SIZE));
1207 		MMSCH_V1_0_INSERT_DIRECT_WT(
1208 			SOC15_REG_OFFSET(UVD, i,
1209 				mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1210 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset +
1211 				AMDGPU_VCN_STACK_SIZE));
1212 		MMSCH_V1_0_INSERT_DIRECT_WT(
1213 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_OFFSET2),
1214 			0);
1215 		MMSCH_V1_0_INSERT_DIRECT_WT(
1216 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_SIZE2),
1217 			AMDGPU_VCN_CONTEXT_SIZE);
1218 
1219 		ring = &adev->vcn.inst[i].ring_enc[0];
1220 		ring->wptr = 0;
1221 
1222 		MMSCH_V1_0_INSERT_DIRECT_WT(
1223 			SOC15_REG_OFFSET(UVD, i, mmUVD_RB_BASE_LO),
1224 			lower_32_bits(ring->gpu_addr));
1225 		MMSCH_V1_0_INSERT_DIRECT_WT(
1226 			SOC15_REG_OFFSET(UVD, i, mmUVD_RB_BASE_HI),
1227 			upper_32_bits(ring->gpu_addr));
1228 		MMSCH_V1_0_INSERT_DIRECT_WT(
1229 			SOC15_REG_OFFSET(UVD, i, mmUVD_RB_SIZE),
1230 			ring->ring_size / 4);
1231 
1232 		ring = &adev->vcn.inst[i].ring_dec;
1233 		ring->wptr = 0;
1234 		MMSCH_V1_0_INSERT_DIRECT_WT(
1235 			SOC15_REG_OFFSET(UVD, i,
1236 				mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1237 			lower_32_bits(ring->gpu_addr));
1238 		MMSCH_V1_0_INSERT_DIRECT_WT(
1239 			SOC15_REG_OFFSET(UVD, i,
1240 				mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1241 			upper_32_bits(ring->gpu_addr));
1242 
1243 		/* force RBC into idle state */
1244 		rb_bufsz = order_base_2(ring->ring_size);
1245 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1246 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1247 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1248 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1249 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1250 		MMSCH_V1_0_INSERT_DIRECT_WT(
1251 			SOC15_REG_OFFSET(UVD, i, mmUVD_RBC_RB_CNTL), tmp);
1252 
1253 		/* add end packet */
1254 		memcpy((void *)init_table, &end, sizeof(struct mmsch_v1_0_cmd_end));
1255 		table_size += sizeof(struct mmsch_v1_0_cmd_end) / 4;
1256 		init_table += sizeof(struct mmsch_v1_0_cmd_end) / 4;
1257 
1258 		/* refine header */
1259 		header->eng[i].table_size = table_size;
1260 		header->total_size += table_size;
1261 	}
1262 
1263 	return vcn_v2_5_mmsch_start(adev, &adev->virt.mm_table);
1264 }
1265 
1266 static int vcn_v2_5_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1267 {
1268 	int ret_code = 0;
1269 	uint32_t tmp;
1270 
1271 	/* Wait for power status to be 1 */
1272 	SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_POWER_STATUS, 1,
1273 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
1274 
1275 	/* wait for read ptr to be equal to write ptr */
1276 	tmp = RREG32_SOC15(UVD, inst_idx, mmUVD_RB_WPTR);
1277 	SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF, ret_code);
1278 
1279 	tmp = RREG32_SOC15(UVD, inst_idx, mmUVD_RB_WPTR2);
1280 	SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF, ret_code);
1281 
1282 	tmp = RREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1283 	SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF, ret_code);
1284 
1285 	SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_POWER_STATUS, 1,
1286 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
1287 
1288 	/* disable dynamic power gating mode */
1289 	WREG32_P(SOC15_REG_OFFSET(UVD, inst_idx, mmUVD_POWER_STATUS), 0,
1290 			~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1291 
1292 	return 0;
1293 }
1294 
1295 static int vcn_v2_5_stop(struct amdgpu_device *adev)
1296 {
1297 	uint32_t tmp;
1298 	int i, r = 0;
1299 
1300 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1301 		if (adev->vcn.harvest_config & (1 << i))
1302 			continue;
1303 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1304 			r = vcn_v2_5_stop_dpg_mode(adev, i);
1305 			continue;
1306 		}
1307 
1308 		/* wait for vcn idle */
1309 		SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7, r);
1310 		if (r)
1311 			return r;
1312 
1313 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1314 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1315 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1316 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1317 		SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp, r);
1318 		if (r)
1319 			return r;
1320 
1321 		/* block LMI UMC channel */
1322 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2);
1323 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1324 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2, tmp);
1325 
1326 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1327 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1328 		SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp, r);
1329 		if (r)
1330 			return r;
1331 
1332 		/* block VCPU register access */
1333 		WREG32_P(SOC15_REG_OFFSET(UVD, i, mmUVD_RB_ARB_CTRL),
1334 			UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1335 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1336 
1337 		/* reset VCPU */
1338 		WREG32_P(SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CNTL),
1339 			UVD_VCPU_CNTL__BLK_RST_MASK,
1340 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1341 
1342 		/* disable VCPU clock */
1343 		WREG32_P(SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CNTL), 0,
1344 			~(UVD_VCPU_CNTL__CLK_EN_MASK));
1345 
1346 		/* clear status */
1347 		WREG32_SOC15(VCN, i, mmUVD_STATUS, 0);
1348 
1349 		vcn_v2_5_enable_clock_gating(adev);
1350 
1351 		/* enable register anti-hang mechanism */
1352 		WREG32_P(SOC15_REG_OFFSET(UVD, i, mmUVD_POWER_STATUS),
1353 			UVD_POWER_STATUS__UVD_POWER_STATUS_MASK,
1354 			~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1355 	}
1356 
1357 	if (adev->pm.dpm_enabled)
1358 		amdgpu_dpm_enable_uvd(adev, false);
1359 
1360 	return 0;
1361 }
1362 
1363 static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
1364 				int inst_idx, struct dpg_pause_state *new_state)
1365 {
1366 	struct amdgpu_ring *ring;
1367 	uint32_t reg_data = 0;
1368 	int ret_code;
1369 
1370 	/* pause/unpause if state is changed */
1371 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1372 		DRM_DEBUG("dpg pause state changed %d -> %d",
1373 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1374 		reg_data = RREG32_SOC15(UVD, inst_idx, mmUVD_DPG_PAUSE) &
1375 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1376 
1377 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1378 			ret_code = 0;
1379 			SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_POWER_STATUS, 0x1,
1380 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
1381 
1382 			if (!ret_code) {
1383 				/* pause DPG */
1384 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1385 				WREG32_SOC15(UVD, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1386 
1387 				/* wait for ACK */
1388 				SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_DPG_PAUSE,
1389 					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1390 					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK, ret_code);
1391 
1392 				/* Restore */
1393 				ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1394 				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
1395 				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1396 				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
1397 				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1398 				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1399 
1400 				ring = &adev->vcn.inst[inst_idx].ring_enc[1];
1401 				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1402 				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1403 				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
1404 				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1405 				WREG32_SOC15(UVD, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1406 
1407 				WREG32_SOC15(UVD, inst_idx, mmUVD_RBC_RB_WPTR,
1408 					   RREG32_SOC15(UVD, inst_idx, mmUVD_SCRATCH2) & 0x7FFFFFFF);
1409 
1410 				SOC15_WAIT_ON_RREG(UVD, inst_idx, mmUVD_POWER_STATUS,
1411 					   UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
1412 			}
1413 		} else {
1414 			/* unpause dpg, no need to wait */
1415 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1416 			WREG32_SOC15(UVD, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1417 		}
1418 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1419 	}
1420 
1421 	return 0;
1422 }
1423 
1424 /**
1425  * vcn_v2_5_dec_ring_get_rptr - get read pointer
1426  *
1427  * @ring: amdgpu_ring pointer
1428  *
1429  * Returns the current hardware read pointer
1430  */
1431 static uint64_t vcn_v2_5_dec_ring_get_rptr(struct amdgpu_ring *ring)
1432 {
1433 	struct amdgpu_device *adev = ring->adev;
1434 
1435 	return RREG32_SOC15(UVD, ring->me, mmUVD_RBC_RB_RPTR);
1436 }
1437 
1438 /**
1439  * vcn_v2_5_dec_ring_get_wptr - get write pointer
1440  *
1441  * @ring: amdgpu_ring pointer
1442  *
1443  * Returns the current hardware write pointer
1444  */
1445 static uint64_t vcn_v2_5_dec_ring_get_wptr(struct amdgpu_ring *ring)
1446 {
1447 	struct amdgpu_device *adev = ring->adev;
1448 
1449 	if (ring->use_doorbell)
1450 		return adev->wb.wb[ring->wptr_offs];
1451 	else
1452 		return RREG32_SOC15(UVD, ring->me, mmUVD_RBC_RB_WPTR);
1453 }
1454 
1455 /**
1456  * vcn_v2_5_dec_ring_set_wptr - set write pointer
1457  *
1458  * @ring: amdgpu_ring pointer
1459  *
1460  * Commits the write pointer to the hardware
1461  */
1462 static void vcn_v2_5_dec_ring_set_wptr(struct amdgpu_ring *ring)
1463 {
1464 	struct amdgpu_device *adev = ring->adev;
1465 
1466 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
1467 		WREG32_SOC15(UVD, ring->me, mmUVD_SCRATCH2,
1468 			lower_32_bits(ring->wptr) | 0x80000000);
1469 
1470 	if (ring->use_doorbell) {
1471 		adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1472 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1473 	} else {
1474 		WREG32_SOC15(UVD, ring->me, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1475 	}
1476 }
1477 
1478 static const struct amdgpu_ring_funcs vcn_v2_5_dec_ring_vm_funcs = {
1479 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1480 	.align_mask = 0xf,
1481 	.vmhub = AMDGPU_MMHUB_1,
1482 	.get_rptr = vcn_v2_5_dec_ring_get_rptr,
1483 	.get_wptr = vcn_v2_5_dec_ring_get_wptr,
1484 	.set_wptr = vcn_v2_5_dec_ring_set_wptr,
1485 	.emit_frame_size =
1486 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1487 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1488 		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1489 		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1490 		6,
1491 	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1492 	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
1493 	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
1494 	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1495 	.test_ring = vcn_v2_0_dec_ring_test_ring,
1496 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
1497 	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
1498 	.insert_start = vcn_v2_0_dec_ring_insert_start,
1499 	.insert_end = vcn_v2_0_dec_ring_insert_end,
1500 	.pad_ib = amdgpu_ring_generic_pad_ib,
1501 	.begin_use = amdgpu_vcn_ring_begin_use,
1502 	.end_use = amdgpu_vcn_ring_end_use,
1503 	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1504 	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1505 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1506 };
1507 
1508 /**
1509  * vcn_v2_5_enc_ring_get_rptr - get enc read pointer
1510  *
1511  * @ring: amdgpu_ring pointer
1512  *
1513  * Returns the current hardware enc read pointer
1514  */
1515 static uint64_t vcn_v2_5_enc_ring_get_rptr(struct amdgpu_ring *ring)
1516 {
1517 	struct amdgpu_device *adev = ring->adev;
1518 
1519 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0])
1520 		return RREG32_SOC15(UVD, ring->me, mmUVD_RB_RPTR);
1521 	else
1522 		return RREG32_SOC15(UVD, ring->me, mmUVD_RB_RPTR2);
1523 }
1524 
1525 /**
1526  * vcn_v2_5_enc_ring_get_wptr - get enc write pointer
1527  *
1528  * @ring: amdgpu_ring pointer
1529  *
1530  * Returns the current hardware enc write pointer
1531  */
1532 static uint64_t vcn_v2_5_enc_ring_get_wptr(struct amdgpu_ring *ring)
1533 {
1534 	struct amdgpu_device *adev = ring->adev;
1535 
1536 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1537 		if (ring->use_doorbell)
1538 			return adev->wb.wb[ring->wptr_offs];
1539 		else
1540 			return RREG32_SOC15(UVD, ring->me, mmUVD_RB_WPTR);
1541 	} else {
1542 		if (ring->use_doorbell)
1543 			return adev->wb.wb[ring->wptr_offs];
1544 		else
1545 			return RREG32_SOC15(UVD, ring->me, mmUVD_RB_WPTR2);
1546 	}
1547 }
1548 
1549 /**
1550  * vcn_v2_5_enc_ring_set_wptr - set enc write pointer
1551  *
1552  * @ring: amdgpu_ring pointer
1553  *
1554  * Commits the enc write pointer to the hardware
1555  */
1556 static void vcn_v2_5_enc_ring_set_wptr(struct amdgpu_ring *ring)
1557 {
1558 	struct amdgpu_device *adev = ring->adev;
1559 
1560 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1561 		if (ring->use_doorbell) {
1562 			adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1563 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1564 		} else {
1565 			WREG32_SOC15(UVD, ring->me, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1566 		}
1567 	} else {
1568 		if (ring->use_doorbell) {
1569 			adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1570 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1571 		} else {
1572 			WREG32_SOC15(UVD, ring->me, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1573 		}
1574 	}
1575 }
1576 
1577 static const struct amdgpu_ring_funcs vcn_v2_5_enc_ring_vm_funcs = {
1578 	.type = AMDGPU_RING_TYPE_VCN_ENC,
1579 	.align_mask = 0x3f,
1580 	.nop = VCN_ENC_CMD_NO_OP,
1581 	.vmhub = AMDGPU_MMHUB_1,
1582 	.get_rptr = vcn_v2_5_enc_ring_get_rptr,
1583 	.get_wptr = vcn_v2_5_enc_ring_get_wptr,
1584 	.set_wptr = vcn_v2_5_enc_ring_set_wptr,
1585 	.emit_frame_size =
1586 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1587 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1588 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1589 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1590 		1, /* vcn_v2_0_enc_ring_insert_end */
1591 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1592 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
1593 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
1594 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1595 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
1596 	.test_ib = amdgpu_vcn_enc_ring_test_ib,
1597 	.insert_nop = amdgpu_ring_insert_nop,
1598 	.insert_end = vcn_v2_0_enc_ring_insert_end,
1599 	.pad_ib = amdgpu_ring_generic_pad_ib,
1600 	.begin_use = amdgpu_vcn_ring_begin_use,
1601 	.end_use = amdgpu_vcn_ring_end_use,
1602 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1603 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1604 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1605 };
1606 
1607 static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev)
1608 {
1609 	int i;
1610 
1611 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1612 		if (adev->vcn.harvest_config & (1 << i))
1613 			continue;
1614 		adev->vcn.inst[i].ring_dec.funcs = &vcn_v2_5_dec_ring_vm_funcs;
1615 		adev->vcn.inst[i].ring_dec.me = i;
1616 		DRM_INFO("VCN(%d) decode is enabled in VM mode\n", i);
1617 	}
1618 }
1619 
1620 static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev)
1621 {
1622 	int i, j;
1623 
1624 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
1625 		if (adev->vcn.harvest_config & (1 << j))
1626 			continue;
1627 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
1628 			adev->vcn.inst[j].ring_enc[i].funcs = &vcn_v2_5_enc_ring_vm_funcs;
1629 			adev->vcn.inst[j].ring_enc[i].me = j;
1630 		}
1631 		DRM_INFO("VCN(%d) encode is enabled in VM mode\n", j);
1632 	}
1633 }
1634 
1635 static bool vcn_v2_5_is_idle(void *handle)
1636 {
1637 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1638 	int i, ret = 1;
1639 
1640 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1641 		if (adev->vcn.harvest_config & (1 << i))
1642 			continue;
1643 		ret &= (RREG32_SOC15(VCN, i, mmUVD_STATUS) == UVD_STATUS__IDLE);
1644 	}
1645 
1646 	return ret;
1647 }
1648 
1649 static int vcn_v2_5_wait_for_idle(void *handle)
1650 {
1651 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1652 	int i, ret = 0;
1653 
1654 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1655 		if (adev->vcn.harvest_config & (1 << i))
1656 			continue;
1657 		SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE,
1658 			UVD_STATUS__IDLE, ret);
1659 		if (ret)
1660 			return ret;
1661 	}
1662 
1663 	return ret;
1664 }
1665 
1666 static int vcn_v2_5_set_clockgating_state(void *handle,
1667 					  enum amd_clockgating_state state)
1668 {
1669 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1670 	bool enable = (state == AMD_CG_STATE_GATE);
1671 
1672 	if (amdgpu_sriov_vf(adev))
1673 		return 0;
1674 
1675 	if (enable) {
1676 		if (!vcn_v2_5_is_idle(handle))
1677 			return -EBUSY;
1678 		vcn_v2_5_enable_clock_gating(adev);
1679 	} else {
1680 		vcn_v2_5_disable_clock_gating(adev);
1681 	}
1682 
1683 	return 0;
1684 }
1685 
1686 static int vcn_v2_5_set_powergating_state(void *handle,
1687 					  enum amd_powergating_state state)
1688 {
1689 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1690 	int ret;
1691 
1692 	if (amdgpu_sriov_vf(adev))
1693 		return 0;
1694 
1695 	if(state == adev->vcn.cur_state)
1696 		return 0;
1697 
1698 	if (state == AMD_PG_STATE_GATE)
1699 		ret = vcn_v2_5_stop(adev);
1700 	else
1701 		ret = vcn_v2_5_start(adev);
1702 
1703 	if(!ret)
1704 		adev->vcn.cur_state = state;
1705 
1706 	return ret;
1707 }
1708 
1709 static int vcn_v2_5_set_interrupt_state(struct amdgpu_device *adev,
1710 					struct amdgpu_irq_src *source,
1711 					unsigned type,
1712 					enum amdgpu_interrupt_state state)
1713 {
1714 	return 0;
1715 }
1716 
1717 static int vcn_v2_5_process_interrupt(struct amdgpu_device *adev,
1718 				      struct amdgpu_irq_src *source,
1719 				      struct amdgpu_iv_entry *entry)
1720 {
1721 	uint32_t ip_instance;
1722 
1723 	switch (entry->client_id) {
1724 	case SOC15_IH_CLIENTID_VCN:
1725 		ip_instance = 0;
1726 		break;
1727 	case SOC15_IH_CLIENTID_VCN1:
1728 		ip_instance = 1;
1729 		break;
1730 	default:
1731 		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
1732 		return 0;
1733 	}
1734 
1735 	DRM_DEBUG("IH: VCN TRAP\n");
1736 
1737 	switch (entry->src_id) {
1738 	case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
1739 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec);
1740 		break;
1741 	case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
1742 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
1743 		break;
1744 	case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
1745 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]);
1746 		break;
1747 	default:
1748 		DRM_ERROR("Unhandled interrupt: %d %d\n",
1749 			  entry->src_id, entry->src_data[0]);
1750 		break;
1751 	}
1752 
1753 	return 0;
1754 }
1755 
1756 static const struct amdgpu_irq_src_funcs vcn_v2_5_irq_funcs = {
1757 	.set = vcn_v2_5_set_interrupt_state,
1758 	.process = vcn_v2_5_process_interrupt,
1759 };
1760 
1761 static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev)
1762 {
1763 	int i;
1764 
1765 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1766 		if (adev->vcn.harvest_config & (1 << i))
1767 			continue;
1768 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
1769 		adev->vcn.inst[i].irq.funcs = &vcn_v2_5_irq_funcs;
1770 	}
1771 }
1772 
1773 static const struct amd_ip_funcs vcn_v2_5_ip_funcs = {
1774 	.name = "vcn_v2_5",
1775 	.early_init = vcn_v2_5_early_init,
1776 	.late_init = NULL,
1777 	.sw_init = vcn_v2_5_sw_init,
1778 	.sw_fini = vcn_v2_5_sw_fini,
1779 	.hw_init = vcn_v2_5_hw_init,
1780 	.hw_fini = vcn_v2_5_hw_fini,
1781 	.suspend = vcn_v2_5_suspend,
1782 	.resume = vcn_v2_5_resume,
1783 	.is_idle = vcn_v2_5_is_idle,
1784 	.wait_for_idle = vcn_v2_5_wait_for_idle,
1785 	.check_soft_reset = NULL,
1786 	.pre_soft_reset = NULL,
1787 	.soft_reset = NULL,
1788 	.post_soft_reset = NULL,
1789 	.set_clockgating_state = vcn_v2_5_set_clockgating_state,
1790 	.set_powergating_state = vcn_v2_5_set_powergating_state,
1791 };
1792 
1793 const struct amdgpu_ip_block_version vcn_v2_5_ip_block =
1794 {
1795 		.type = AMD_IP_BLOCK_TYPE_VCN,
1796 		.major = 2,
1797 		.minor = 5,
1798 		.rev = 0,
1799 		.funcs = &vcn_v2_5_ip_funcs,
1800 };
1801