xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c (revision c0ecca6604b80e438b032578634c6e133c7028f6)
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 "amdgpu.h"
26 #include "amdgpu_vcn.h"
27 #include "amdgpu_pm.h"
28 #include "soc15.h"
29 #include "soc15d.h"
30 #include "vcn_v2_0.h"
31 #include "mmsch_v3_0.h"
32 
33 #include "vcn/vcn_3_0_0_offset.h"
34 #include "vcn/vcn_3_0_0_sh_mask.h"
35 #include "ivsrcid/vcn/irqsrcs_vcn_2_0.h"
36 
37 #define mmUVD_CONTEXT_ID_INTERNAL_OFFSET			0x27
38 #define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET			0x0f
39 #define mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET			0x10
40 #define mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET			0x11
41 #define mmUVD_NO_OP_INTERNAL_OFFSET				0x29
42 #define mmUVD_GP_SCRATCH8_INTERNAL_OFFSET			0x66
43 #define mmUVD_SCRATCH9_INTERNAL_OFFSET				0xc01d
44 
45 #define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET			0x431
46 #define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET		0x3b4
47 #define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET		0x3b5
48 #define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET			0x25c
49 
50 #define VCN_INSTANCES_SIENNA_CICHLID				2
51 #define DEC_SW_RING_ENABLED					FALSE
52 
53 #define RDECODE_MSG_CREATE					0x00000000
54 #define RDECODE_MESSAGE_CREATE					0x00000001
55 
56 static int amdgpu_ih_clientid_vcns[] = {
57 	SOC15_IH_CLIENTID_VCN,
58 	SOC15_IH_CLIENTID_VCN1
59 };
60 
61 static int amdgpu_ucode_id_vcns[] = {
62 	AMDGPU_UCODE_ID_VCN,
63 	AMDGPU_UCODE_ID_VCN1
64 };
65 
66 static int vcn_v3_0_start_sriov(struct amdgpu_device *adev);
67 static void vcn_v3_0_set_dec_ring_funcs(struct amdgpu_device *adev);
68 static void vcn_v3_0_set_enc_ring_funcs(struct amdgpu_device *adev);
69 static void vcn_v3_0_set_irq_funcs(struct amdgpu_device *adev);
70 static int vcn_v3_0_set_powergating_state(void *handle,
71 			enum amd_powergating_state state);
72 static int vcn_v3_0_pause_dpg_mode(struct amdgpu_device *adev,
73 			int inst_idx, struct dpg_pause_state *new_state);
74 
75 static void vcn_v3_0_dec_ring_set_wptr(struct amdgpu_ring *ring);
76 static void vcn_v3_0_enc_ring_set_wptr(struct amdgpu_ring *ring);
77 
78 /**
79  * vcn_v3_0_early_init - set function pointers
80  *
81  * @handle: amdgpu_device pointer
82  *
83  * Set ring and irq function pointers
84  */
85 static int vcn_v3_0_early_init(void *handle)
86 {
87 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
88 
89 	if (amdgpu_sriov_vf(adev)) {
90 		adev->vcn.num_vcn_inst = VCN_INSTANCES_SIENNA_CICHLID;
91 		adev->vcn.harvest_config = 0;
92 		adev->vcn.num_enc_rings = 1;
93 
94 	} else {
95 		if (adev->asic_type == CHIP_SIENNA_CICHLID) {
96 			u32 harvest;
97 			int i;
98 
99 			adev->vcn.num_vcn_inst = VCN_INSTANCES_SIENNA_CICHLID;
100 			for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
101 				harvest = RREG32_SOC15(VCN, i, mmCC_UVD_HARVESTING);
102 				if (harvest & CC_UVD_HARVESTING__UVD_DISABLE_MASK)
103 					adev->vcn.harvest_config |= 1 << i;
104 			}
105 
106 			if (adev->vcn.harvest_config == (AMDGPU_VCN_HARVEST_VCN0 |
107 						AMDGPU_VCN_HARVEST_VCN1))
108 				/* both instances are harvested, disable the block */
109 				return -ENOENT;
110 		} else
111 			adev->vcn.num_vcn_inst = 1;
112 
113 		adev->vcn.num_enc_rings = 2;
114 	}
115 
116 	vcn_v3_0_set_dec_ring_funcs(adev);
117 	vcn_v3_0_set_enc_ring_funcs(adev);
118 	vcn_v3_0_set_irq_funcs(adev);
119 
120 	return 0;
121 }
122 
123 /**
124  * vcn_v3_0_sw_init - sw init for VCN block
125  *
126  * @handle: amdgpu_device pointer
127  *
128  * Load firmware and sw initialization
129  */
130 static int vcn_v3_0_sw_init(void *handle)
131 {
132 	struct amdgpu_ring *ring;
133 	int i, j, r;
134 	int vcn_doorbell_index = 0;
135 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
136 
137 	r = amdgpu_vcn_sw_init(adev);
138 	if (r)
139 		return r;
140 
141 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
142 		const struct common_firmware_header *hdr;
143 		hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
144 		adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].ucode_id = AMDGPU_UCODE_ID_VCN;
145 		adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].fw = adev->vcn.fw;
146 		adev->firmware.fw_size +=
147 			ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
148 
149 		if (adev->vcn.num_vcn_inst == VCN_INSTANCES_SIENNA_CICHLID) {
150 			adev->firmware.ucode[AMDGPU_UCODE_ID_VCN1].ucode_id = AMDGPU_UCODE_ID_VCN1;
151 			adev->firmware.ucode[AMDGPU_UCODE_ID_VCN1].fw = adev->vcn.fw;
152 			adev->firmware.fw_size +=
153 				ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
154 		}
155 		DRM_INFO("PSP loading VCN firmware\n");
156 	}
157 
158 	r = amdgpu_vcn_resume(adev);
159 	if (r)
160 		return r;
161 
162 	/*
163 	 * Note: doorbell assignment is fixed for SRIOV multiple VCN engines
164 	 * Formula:
165 	 *   vcn_db_base  = adev->doorbell_index.vcn.vcn_ring0_1 << 1;
166 	 *   dec_ring_i   = vcn_db_base + i * (adev->vcn.num_enc_rings + 1)
167 	 *   enc_ring_i,j = vcn_db_base + i * (adev->vcn.num_enc_rings + 1) + 1 + j
168 	 */
169 	if (amdgpu_sriov_vf(adev)) {
170 		vcn_doorbell_index = adev->doorbell_index.vcn.vcn_ring0_1;
171 		/* get DWORD offset */
172 		vcn_doorbell_index = vcn_doorbell_index << 1;
173 	}
174 
175 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
176 		volatile struct amdgpu_fw_shared *fw_shared;
177 
178 		if (adev->vcn.harvest_config & (1 << i))
179 			continue;
180 
181 		adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
182 		adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
183 		adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
184 		adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
185 		adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
186 		adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
187 
188 		adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
189 		adev->vcn.inst[i].external.scratch9 = SOC15_REG_OFFSET(VCN, i, mmUVD_SCRATCH9);
190 		adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
191 		adev->vcn.inst[i].external.data0 = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_DATA0);
192 		adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
193 		adev->vcn.inst[i].external.data1 = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_DATA1);
194 		adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
195 		adev->vcn.inst[i].external.cmd = SOC15_REG_OFFSET(VCN, i, mmUVD_GPCOM_VCPU_CMD);
196 		adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
197 		adev->vcn.inst[i].external.nop = SOC15_REG_OFFSET(VCN, i, mmUVD_NO_OP);
198 
199 		/* VCN DEC TRAP */
200 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
201 				VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst[i].irq);
202 		if (r)
203 			return r;
204 
205 		atomic_set(&adev->vcn.inst[i].sched_score, 0);
206 
207 		ring = &adev->vcn.inst[i].ring_dec;
208 		ring->use_doorbell = true;
209 		if (amdgpu_sriov_vf(adev)) {
210 			ring->doorbell_index = vcn_doorbell_index + i * (adev->vcn.num_enc_rings + 1);
211 		} else {
212 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i;
213 		}
214 		sprintf(ring->name, "vcn_dec_%d", i);
215 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
216 				     AMDGPU_RING_PRIO_DEFAULT,
217 				     &adev->vcn.inst[i].sched_score);
218 		if (r)
219 			return r;
220 
221 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
222 			/* VCN ENC TRAP */
223 			r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
224 				j + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);
225 			if (r)
226 				return r;
227 
228 			ring = &adev->vcn.inst[i].ring_enc[j];
229 			ring->use_doorbell = true;
230 			if (amdgpu_sriov_vf(adev)) {
231 				ring->doorbell_index = vcn_doorbell_index + i * (adev->vcn.num_enc_rings + 1) + 1 + j;
232 			} else {
233 				ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + j + 8 * i;
234 			}
235 			sprintf(ring->name, "vcn_enc_%d.%d", i, j);
236 			r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
237 					     AMDGPU_RING_PRIO_DEFAULT,
238 					     &adev->vcn.inst[i].sched_score);
239 			if (r)
240 				return r;
241 		}
242 
243 		fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
244 		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_SW_RING_FLAG) |
245 					     cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG) |
246 					     cpu_to_le32(AMDGPU_VCN_FW_SHARED_FLAG_0_RB);
247 		fw_shared->sw_ring.is_enabled = cpu_to_le32(DEC_SW_RING_ENABLED);
248 	}
249 
250 	if (amdgpu_sriov_vf(adev)) {
251 		r = amdgpu_virt_alloc_mm_table(adev);
252 		if (r)
253 			return r;
254 	}
255 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
256 		adev->vcn.pause_dpg_mode = vcn_v3_0_pause_dpg_mode;
257 
258 	return 0;
259 }
260 
261 /**
262  * vcn_v3_0_sw_fini - sw fini for VCN block
263  *
264  * @handle: amdgpu_device pointer
265  *
266  * VCN suspend and free up sw allocation
267  */
268 static int vcn_v3_0_sw_fini(void *handle)
269 {
270 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
271 	int i, r;
272 
273 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
274 		volatile struct amdgpu_fw_shared *fw_shared;
275 
276 		if (adev->vcn.harvest_config & (1 << i))
277 			continue;
278 		fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
279 		fw_shared->present_flag_0 = 0;
280 		fw_shared->sw_ring.is_enabled = false;
281 	}
282 
283 	if (amdgpu_sriov_vf(adev))
284 		amdgpu_virt_free_mm_table(adev);
285 
286 	r = amdgpu_vcn_suspend(adev);
287 	if (r)
288 		return r;
289 
290 	r = amdgpu_vcn_sw_fini(adev);
291 
292 	return r;
293 }
294 
295 /**
296  * vcn_v3_0_hw_init - start and test VCN block
297  *
298  * @handle: amdgpu_device pointer
299  *
300  * Initialize the hardware, boot up the VCPU and do some testing
301  */
302 static int vcn_v3_0_hw_init(void *handle)
303 {
304 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
305 	struct amdgpu_ring *ring;
306 	int i, j, r;
307 
308 	if (amdgpu_sriov_vf(adev)) {
309 		r = vcn_v3_0_start_sriov(adev);
310 		if (r)
311 			goto done;
312 
313 		/* initialize VCN dec and enc ring buffers */
314 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
315 			if (adev->vcn.harvest_config & (1 << i))
316 				continue;
317 
318 			ring = &adev->vcn.inst[i].ring_dec;
319 			if (ring->sched.ready) {
320 				ring->wptr = 0;
321 				ring->wptr_old = 0;
322 				vcn_v3_0_dec_ring_set_wptr(ring);
323 			}
324 
325 			for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
326 				ring = &adev->vcn.inst[i].ring_enc[j];
327 				if (ring->sched.ready) {
328 					ring->wptr = 0;
329 					ring->wptr_old = 0;
330 					vcn_v3_0_enc_ring_set_wptr(ring);
331 				}
332 			}
333 		}
334 	} else {
335 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
336 			if (adev->vcn.harvest_config & (1 << i))
337 				continue;
338 
339 			ring = &adev->vcn.inst[i].ring_dec;
340 
341 			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
342 						     ring->doorbell_index, i);
343 
344 			r = amdgpu_ring_test_helper(ring);
345 			if (r)
346 				goto done;
347 
348 			for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
349 				ring = &adev->vcn.inst[i].ring_enc[j];
350 				r = amdgpu_ring_test_helper(ring);
351 				if (r)
352 					goto done;
353 			}
354 		}
355 	}
356 
357 done:
358 	if (!r)
359 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
360 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
361 
362 	return r;
363 }
364 
365 /**
366  * vcn_v3_0_hw_fini - stop the hardware block
367  *
368  * @handle: amdgpu_device pointer
369  *
370  * Stop the VCN block, mark ring as not ready any more
371  */
372 static int vcn_v3_0_hw_fini(void *handle)
373 {
374 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
375 	int i;
376 
377 	cancel_delayed_work_sync(&adev->vcn.idle_work);
378 
379 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
380 		if (adev->vcn.harvest_config & (1 << i))
381 			continue;
382 
383 		if (!amdgpu_sriov_vf(adev)) {
384 			if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
385 					(adev->vcn.cur_state != AMD_PG_STATE_GATE &&
386 					 RREG32_SOC15(VCN, i, mmUVD_STATUS))) {
387 				vcn_v3_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
388 			}
389 		}
390 	}
391 
392 	return 0;
393 }
394 
395 /**
396  * vcn_v3_0_suspend - suspend VCN block
397  *
398  * @handle: amdgpu_device pointer
399  *
400  * HW fini and suspend VCN block
401  */
402 static int vcn_v3_0_suspend(void *handle)
403 {
404 	int r;
405 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
406 
407 	r = vcn_v3_0_hw_fini(adev);
408 	if (r)
409 		return r;
410 
411 	r = amdgpu_vcn_suspend(adev);
412 
413 	return r;
414 }
415 
416 /**
417  * vcn_v3_0_resume - resume VCN block
418  *
419  * @handle: amdgpu_device pointer
420  *
421  * Resume firmware and hw init VCN block
422  */
423 static int vcn_v3_0_resume(void *handle)
424 {
425 	int r;
426 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
427 
428 	r = amdgpu_vcn_resume(adev);
429 	if (r)
430 		return r;
431 
432 	r = vcn_v3_0_hw_init(adev);
433 
434 	return r;
435 }
436 
437 /**
438  * vcn_v3_0_mc_resume - memory controller programming
439  *
440  * @adev: amdgpu_device pointer
441  * @inst: instance number
442  *
443  * Let the VCN memory controller know it's offsets
444  */
445 static void vcn_v3_0_mc_resume(struct amdgpu_device *adev, int inst)
446 {
447 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
448 	uint32_t offset;
449 
450 	/* cache window 0: fw */
451 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
452 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
453 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo));
454 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
455 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi));
456 		WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET0, 0);
457 		offset = 0;
458 	} else {
459 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
460 			lower_32_bits(adev->vcn.inst[inst].gpu_addr));
461 		WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
462 			upper_32_bits(adev->vcn.inst[inst].gpu_addr));
463 		offset = size;
464 		WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET0,
465 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
466 	}
467 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE0, size);
468 
469 	/* cache window 1: stack */
470 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
471 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
472 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
473 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
474 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET1, 0);
475 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
476 
477 	/* cache window 2: context */
478 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
479 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
480 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
481 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
482 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_OFFSET2, 0);
483 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
484 
485 	/* non-cache window */
486 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
487 		lower_32_bits(adev->vcn.inst[inst].fw_shared_gpu_addr));
488 	WREG32_SOC15(VCN, inst, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
489 		upper_32_bits(adev->vcn.inst[inst].fw_shared_gpu_addr));
490 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
491 	WREG32_SOC15(VCN, inst, mmUVD_VCPU_NONCACHE_SIZE0,
492 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
493 }
494 
495 static void vcn_v3_0_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
496 {
497 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
498 	uint32_t offset;
499 
500 	/* cache window 0: fw */
501 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
502 		if (!indirect) {
503 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
504 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
505 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
506 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
507 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
508 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
509 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
510 				VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
511 		} else {
512 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
513 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
514 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
515 				VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
516 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
517 				VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
518 		}
519 		offset = 0;
520 	} else {
521 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
522 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
523 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
524 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
525 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
526 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
527 		offset = size;
528 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
529 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET0),
530 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
531 	}
532 
533 	if (!indirect)
534 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
535 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
536 	else
537 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
538 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
539 
540 	/* cache window 1: stack */
541 	if (!indirect) {
542 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
543 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
544 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
545 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
546 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
547 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
548 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
549 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
550 	} else {
551 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
552 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
553 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
554 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
555 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
556 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
557 	}
558 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
559 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
560 
561 	/* cache window 2: context */
562 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
563 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
564 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
565 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
566 			VCN, inst_idx, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
567 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
568 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
569 			VCN, inst_idx, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
570 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
571 			VCN, inst_idx, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
572 
573 	/* non-cache window */
574 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
575 			VCN, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
576 			lower_32_bits(adev->vcn.inst[inst_idx].fw_shared_gpu_addr), 0, indirect);
577 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
578 			VCN, inst_idx, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
579 			upper_32_bits(adev->vcn.inst[inst_idx].fw_shared_gpu_addr), 0, indirect);
580 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
581 			VCN, inst_idx, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
582 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
583 			VCN, inst_idx, mmUVD_VCPU_NONCACHE_SIZE0),
584 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
585 
586 	/* VCN global tiling registers */
587 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
588 		UVD, 0, mmUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
589 }
590 
591 static void vcn_v3_0_disable_static_power_gating(struct amdgpu_device *adev, int inst)
592 {
593 	uint32_t data = 0;
594 
595 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
596 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
597 			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
598 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
599 			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
600 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
601 			| 2 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
602 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
603 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
604 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
605 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
606 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
607 			| 2 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
608 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
609 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
610 
611 		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
612 		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS,
613 			UVD_PGFSM_STATUS__UVDM_UVDU_UVDLM_PWR_ON_3_0, 0x3F3FFFFF);
614 	} else {
615 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
616 			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
617 			| 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
618 			| 1 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
619 			| 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
620 			| 1 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
621 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
622 			| 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
623 			| 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
624 			| 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
625 			| 1 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
626 			| 1 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
627 			| 1 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
628 			| 1 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
629 		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
630 		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS, 0,  0x3F3FFFFF);
631 	}
632 
633 	data = RREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS);
634 	data &= ~0x103;
635 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
636 		data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
637 			UVD_POWER_STATUS__UVD_PG_EN_MASK;
638 
639 	WREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS, data);
640 }
641 
642 static void vcn_v3_0_enable_static_power_gating(struct amdgpu_device *adev, int inst)
643 {
644 	uint32_t data;
645 
646 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
647 		/* Before power off, this indicator has to be turned on */
648 		data = RREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS);
649 		data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
650 		data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
651 		WREG32_SOC15(VCN, inst, mmUVD_POWER_STATUS, data);
652 
653 		data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
654 			| 2 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
655 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
656 			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
657 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
658 			| 2 << UVD_PGFSM_CONFIG__UVDIRL_PWR_CONFIG__SHIFT
659 			| 2 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
660 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
661 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
662 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
663 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
664 			| 2 << UVD_PGFSM_CONFIG__UVDATD_PWR_CONFIG__SHIFT
665 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
666 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
667 		WREG32_SOC15(VCN, inst, mmUVD_PGFSM_CONFIG, data);
668 
669 		data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
670 			| 2 << UVD_PGFSM_STATUS__UVDU_PWR_STATUS__SHIFT
671 			| 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
672 			| 2 << UVD_PGFSM_STATUS__UVDC_PWR_STATUS__SHIFT
673 			| 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
674 			| 2 << UVD_PGFSM_STATUS__UVDIRL_PWR_STATUS__SHIFT
675 			| 2 << UVD_PGFSM_STATUS__UVDLM_PWR_STATUS__SHIFT
676 			| 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
677 			| 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
678 			| 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT
679 			| 2 << UVD_PGFSM_STATUS__UVDAB_PWR_STATUS__SHIFT
680 			| 2 << UVD_PGFSM_STATUS__UVDATD_PWR_STATUS__SHIFT
681 			| 2 << UVD_PGFSM_STATUS__UVDNA_PWR_STATUS__SHIFT
682 			| 2 << UVD_PGFSM_STATUS__UVDNB_PWR_STATUS__SHIFT);
683 		SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_PGFSM_STATUS, data, 0x3F3FFFFF);
684 	}
685 }
686 
687 /**
688  * vcn_v3_0_disable_clock_gating - disable VCN clock gating
689  *
690  * @adev: amdgpu_device pointer
691  * @inst: instance number
692  *
693  * Disable clock gating for VCN block
694  */
695 static void vcn_v3_0_disable_clock_gating(struct amdgpu_device *adev, int inst)
696 {
697 	uint32_t data;
698 
699 	/* VCN disable CGC */
700 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
701 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
702 		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
703 	else
704 		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
705 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
706 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
707 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
708 
709 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_GATE);
710 	data &= ~(UVD_CGC_GATE__SYS_MASK
711 		| UVD_CGC_GATE__UDEC_MASK
712 		| UVD_CGC_GATE__MPEG2_MASK
713 		| UVD_CGC_GATE__REGS_MASK
714 		| UVD_CGC_GATE__RBC_MASK
715 		| UVD_CGC_GATE__LMI_MC_MASK
716 		| UVD_CGC_GATE__LMI_UMC_MASK
717 		| UVD_CGC_GATE__IDCT_MASK
718 		| UVD_CGC_GATE__MPRD_MASK
719 		| UVD_CGC_GATE__MPC_MASK
720 		| UVD_CGC_GATE__LBSI_MASK
721 		| UVD_CGC_GATE__LRBBM_MASK
722 		| UVD_CGC_GATE__UDEC_RE_MASK
723 		| UVD_CGC_GATE__UDEC_CM_MASK
724 		| UVD_CGC_GATE__UDEC_IT_MASK
725 		| UVD_CGC_GATE__UDEC_DB_MASK
726 		| UVD_CGC_GATE__UDEC_MP_MASK
727 		| UVD_CGC_GATE__WCB_MASK
728 		| UVD_CGC_GATE__VCPU_MASK
729 		| UVD_CGC_GATE__MMSCH_MASK);
730 
731 	WREG32_SOC15(VCN, inst, mmUVD_CGC_GATE, data);
732 
733 	SOC15_WAIT_ON_RREG(VCN, inst, mmUVD_CGC_GATE, 0,  0xFFFFFFFF);
734 
735 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
736 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
737 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
738 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
739 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
740 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
741 		| UVD_CGC_CTRL__SYS_MODE_MASK
742 		| UVD_CGC_CTRL__UDEC_MODE_MASK
743 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
744 		| UVD_CGC_CTRL__REGS_MODE_MASK
745 		| UVD_CGC_CTRL__RBC_MODE_MASK
746 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
747 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
748 		| UVD_CGC_CTRL__IDCT_MODE_MASK
749 		| UVD_CGC_CTRL__MPRD_MODE_MASK
750 		| UVD_CGC_CTRL__MPC_MODE_MASK
751 		| UVD_CGC_CTRL__LBSI_MODE_MASK
752 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
753 		| UVD_CGC_CTRL__WCB_MODE_MASK
754 		| UVD_CGC_CTRL__VCPU_MODE_MASK
755 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
756 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
757 
758 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE);
759 	data |= (UVD_SUVD_CGC_GATE__SRE_MASK
760 		| UVD_SUVD_CGC_GATE__SIT_MASK
761 		| UVD_SUVD_CGC_GATE__SMP_MASK
762 		| UVD_SUVD_CGC_GATE__SCM_MASK
763 		| UVD_SUVD_CGC_GATE__SDB_MASK
764 		| UVD_SUVD_CGC_GATE__SRE_H264_MASK
765 		| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
766 		| UVD_SUVD_CGC_GATE__SIT_H264_MASK
767 		| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
768 		| UVD_SUVD_CGC_GATE__SCM_H264_MASK
769 		| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
770 		| UVD_SUVD_CGC_GATE__SDB_H264_MASK
771 		| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
772 		| UVD_SUVD_CGC_GATE__SCLR_MASK
773 		| UVD_SUVD_CGC_GATE__ENT_MASK
774 		| UVD_SUVD_CGC_GATE__IME_MASK
775 		| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
776 		| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
777 		| UVD_SUVD_CGC_GATE__SITE_MASK
778 		| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
779 		| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
780 		| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
781 		| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
782 		| UVD_SUVD_CGC_GATE__IME_HEVC_MASK
783 		| UVD_SUVD_CGC_GATE__EFC_MASK
784 		| UVD_SUVD_CGC_GATE__SAOE_MASK
785 		| UVD_SUVD_CGC_GATE__SRE_AV1_MASK
786 		| UVD_SUVD_CGC_GATE__FBC_PCLK_MASK
787 		| UVD_SUVD_CGC_GATE__FBC_CCLK_MASK
788 		| UVD_SUVD_CGC_GATE__SCM_AV1_MASK
789 		| UVD_SUVD_CGC_GATE__SMPA_MASK);
790 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE, data);
791 
792 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE2);
793 	data |= (UVD_SUVD_CGC_GATE2__MPBE0_MASK
794 		| UVD_SUVD_CGC_GATE2__MPBE1_MASK
795 		| UVD_SUVD_CGC_GATE2__SIT_AV1_MASK
796 		| UVD_SUVD_CGC_GATE2__SDB_AV1_MASK
797 		| UVD_SUVD_CGC_GATE2__MPC1_MASK);
798 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_GATE2, data);
799 
800 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL);
801 	data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
802 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
803 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
804 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
805 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
806 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
807 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
808 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
809 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK
810 		| UVD_SUVD_CGC_CTRL__EFC_MODE_MASK
811 		| UVD_SUVD_CGC_CTRL__SAOE_MODE_MASK
812 		| UVD_SUVD_CGC_CTRL__SMPA_MODE_MASK
813 		| UVD_SUVD_CGC_CTRL__MPBE0_MODE_MASK
814 		| UVD_SUVD_CGC_CTRL__MPBE1_MODE_MASK
815 		| UVD_SUVD_CGC_CTRL__SIT_AV1_MODE_MASK
816 		| UVD_SUVD_CGC_CTRL__SDB_AV1_MODE_MASK
817 		| UVD_SUVD_CGC_CTRL__MPC1_MODE_MASK
818 		| UVD_SUVD_CGC_CTRL__FBC_PCLK_MASK
819 		| UVD_SUVD_CGC_CTRL__FBC_CCLK_MASK);
820 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL, data);
821 }
822 
823 static void vcn_v3_0_clock_gating_dpg_mode(struct amdgpu_device *adev,
824 		uint8_t sram_sel, int inst_idx, uint8_t indirect)
825 {
826 	uint32_t reg_data = 0;
827 
828 	/* enable sw clock gating control */
829 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
830 		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
831 	else
832 		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
833 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
834 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
835 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
836 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
837 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
838 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
839 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
840 		 UVD_CGC_CTRL__SYS_MODE_MASK |
841 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
842 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
843 		 UVD_CGC_CTRL__REGS_MODE_MASK |
844 		 UVD_CGC_CTRL__RBC_MODE_MASK |
845 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
846 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
847 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
848 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
849 		 UVD_CGC_CTRL__MPC_MODE_MASK |
850 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
851 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
852 		 UVD_CGC_CTRL__WCB_MODE_MASK |
853 		 UVD_CGC_CTRL__VCPU_MODE_MASK |
854 		 UVD_CGC_CTRL__MMSCH_MODE_MASK);
855 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
856 		VCN, inst_idx, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
857 
858 	/* turn off clock gating */
859 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
860 		VCN, inst_idx, mmUVD_CGC_GATE), 0, sram_sel, indirect);
861 
862 	/* turn on SUVD clock gating */
863 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
864 		VCN, inst_idx, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
865 
866 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
867 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
868 		VCN, inst_idx, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
869 }
870 
871 /**
872  * vcn_v3_0_enable_clock_gating - enable VCN clock gating
873  *
874  * @adev: amdgpu_device pointer
875  * @inst: instance number
876  *
877  * Enable clock gating for VCN block
878  */
879 static void vcn_v3_0_enable_clock_gating(struct amdgpu_device *adev, int inst)
880 {
881 	uint32_t data;
882 
883 	/* enable VCN CGC */
884 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
885 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
886 		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
887 	else
888 		data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
889 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
890 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
891 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
892 
893 	data = RREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL);
894 	data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
895 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
896 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
897 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
898 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
899 		| UVD_CGC_CTRL__SYS_MODE_MASK
900 		| UVD_CGC_CTRL__UDEC_MODE_MASK
901 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
902 		| UVD_CGC_CTRL__REGS_MODE_MASK
903 		| UVD_CGC_CTRL__RBC_MODE_MASK
904 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
905 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
906 		| UVD_CGC_CTRL__IDCT_MODE_MASK
907 		| UVD_CGC_CTRL__MPRD_MODE_MASK
908 		| UVD_CGC_CTRL__MPC_MODE_MASK
909 		| UVD_CGC_CTRL__LBSI_MODE_MASK
910 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
911 		| UVD_CGC_CTRL__WCB_MODE_MASK
912 		| UVD_CGC_CTRL__VCPU_MODE_MASK
913 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
914 	WREG32_SOC15(VCN, inst, mmUVD_CGC_CTRL, data);
915 
916 	data = RREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL);
917 	data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
918 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
919 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
920 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
921 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
922 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
923 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
924 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
925 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK
926 		| UVD_SUVD_CGC_CTRL__EFC_MODE_MASK
927 		| UVD_SUVD_CGC_CTRL__SAOE_MODE_MASK
928 		| UVD_SUVD_CGC_CTRL__SMPA_MODE_MASK
929 		| UVD_SUVD_CGC_CTRL__MPBE0_MODE_MASK
930 		| UVD_SUVD_CGC_CTRL__MPBE1_MODE_MASK
931 		| UVD_SUVD_CGC_CTRL__SIT_AV1_MODE_MASK
932 		| UVD_SUVD_CGC_CTRL__SDB_AV1_MODE_MASK
933 		| UVD_SUVD_CGC_CTRL__MPC1_MODE_MASK
934 		| UVD_SUVD_CGC_CTRL__FBC_PCLK_MASK
935 		| UVD_SUVD_CGC_CTRL__FBC_CCLK_MASK);
936 	WREG32_SOC15(VCN, inst, mmUVD_SUVD_CGC_CTRL, data);
937 }
938 
939 static int vcn_v3_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
940 {
941 	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared_cpu_addr;
942 	struct amdgpu_ring *ring;
943 	uint32_t rb_bufsz, tmp;
944 
945 	/* disable register anti-hang mechanism */
946 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1,
947 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
948 	/* enable dynamic power gating mode */
949 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS);
950 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
951 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
952 	WREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS, tmp);
953 
954 	if (indirect)
955 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
956 
957 	/* enable clock gating */
958 	vcn_v3_0_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
959 
960 	/* enable VCPU clock */
961 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
962 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
963 	tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
964 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
965 		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
966 
967 	/* disable master interupt */
968 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
969 		VCN, inst_idx, mmUVD_MASTINT_EN), 0, 0, indirect);
970 
971 	/* setup mmUVD_LMI_CTRL */
972 	tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
973 		UVD_LMI_CTRL__REQ_MODE_MASK |
974 		UVD_LMI_CTRL__CRC_RESET_MASK |
975 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
976 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
977 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
978 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
979 		0x00100000L);
980 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
981 		VCN, inst_idx, mmUVD_LMI_CTRL), tmp, 0, indirect);
982 
983 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
984 		VCN, inst_idx, mmUVD_MPC_CNTL),
985 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
986 
987 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
988 		VCN, inst_idx, mmUVD_MPC_SET_MUXA0),
989 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
990 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
991 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
992 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
993 
994 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
995 		VCN, inst_idx, mmUVD_MPC_SET_MUXB0),
996 		 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
997 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
998 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
999 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
1000 
1001 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1002 		VCN, inst_idx, mmUVD_MPC_SET_MUX),
1003 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1004 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1005 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
1006 
1007 	vcn_v3_0_mc_resume_dpg_mode(adev, inst_idx, indirect);
1008 
1009 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1010 		VCN, inst_idx, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
1011 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1012 		VCN, inst_idx, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
1013 
1014 	/* enable LMI MC and UMC channels */
1015 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1016 		VCN, inst_idx, mmUVD_LMI_CTRL2), 0, 0, indirect);
1017 
1018 	/* unblock VCPU register access */
1019 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1020 		VCN, inst_idx, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
1021 
1022 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
1023 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
1024 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1025 		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
1026 
1027 	/* enable master interrupt */
1028 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1029 		VCN, inst_idx, mmUVD_MASTINT_EN),
1030 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
1031 
1032 	/* add nop to workaround PSP size check */
1033 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1034 		VCN, inst_idx, mmUVD_VCPU_CNTL), tmp, 0, indirect);
1035 
1036 	if (indirect)
1037 		psp_update_vcn_sram(adev, inst_idx, adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
1038 			(uint32_t)((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
1039 				(uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr));
1040 
1041 	ring = &adev->vcn.inst[inst_idx].ring_dec;
1042 	/* force RBC into idle state */
1043 	rb_bufsz = order_base_2(ring->ring_size);
1044 	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1045 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1046 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1047 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1048 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1049 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
1050 
1051 	/* Stall DPG before WPTR/RPTR reset */
1052 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1053 		UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1054 		~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1055 	fw_shared->multi_queue.decode_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1056 
1057 	/* set the write pointer delay */
1058 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
1059 
1060 	/* set the wb address */
1061 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR_ADDR,
1062 		(upper_32_bits(ring->gpu_addr) >> 2));
1063 
1064 	/* programm the RB_BASE for ring buffer */
1065 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1066 		lower_32_bits(ring->gpu_addr));
1067 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1068 		upper_32_bits(ring->gpu_addr));
1069 
1070 	/* Initialize the ring buffer's read and write pointers */
1071 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, 0);
1072 
1073 	WREG32_SOC15(VCN, inst_idx, mmUVD_SCRATCH2, 0);
1074 
1075 	ring->wptr = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR);
1076 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR,
1077 		lower_32_bits(ring->wptr));
1078 
1079 	/* Reset FW shared memory RBC WPTR/RPTR */
1080 	fw_shared->rb.rptr = 0;
1081 	fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1082 
1083 	/*resetting done, fw can check RB ring */
1084 	fw_shared->multi_queue.decode_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1085 
1086 	/* Unstall DPG */
1087 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1088 		0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1089 
1090 	return 0;
1091 }
1092 
1093 static int vcn_v3_0_start(struct amdgpu_device *adev)
1094 {
1095 	volatile struct amdgpu_fw_shared *fw_shared;
1096 	struct amdgpu_ring *ring;
1097 	uint32_t rb_bufsz, tmp;
1098 	int i, j, k, r;
1099 
1100 	if (adev->pm.dpm_enabled)
1101 		amdgpu_dpm_enable_uvd(adev, true);
1102 
1103 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1104 		if (adev->vcn.harvest_config & (1 << i))
1105 			continue;
1106 
1107 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG){
1108 			r = vcn_v3_0_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
1109 			continue;
1110 		}
1111 
1112 		/* disable VCN power gating */
1113 		vcn_v3_0_disable_static_power_gating(adev, i);
1114 
1115 		/* set VCN status busy */
1116 		tmp = RREG32_SOC15(VCN, i, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY;
1117 		WREG32_SOC15(VCN, i, mmUVD_STATUS, tmp);
1118 
1119 		/*SW clock gating */
1120 		vcn_v3_0_disable_clock_gating(adev, i);
1121 
1122 		/* enable VCPU clock */
1123 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1124 			UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1125 
1126 		/* disable master interrupt */
1127 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN), 0,
1128 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1129 
1130 		/* enable LMI MC and UMC channels */
1131 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_LMI_CTRL2), 0,
1132 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1133 
1134 		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1135 		tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1136 		tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1137 		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1138 
1139 		/* setup mmUVD_LMI_CTRL */
1140 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL);
1141 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL, tmp |
1142 			UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK	|
1143 			UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1144 			UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1145 			UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1146 
1147 		/* setup mmUVD_MPC_CNTL */
1148 		tmp = RREG32_SOC15(VCN, i, mmUVD_MPC_CNTL);
1149 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1150 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1151 		WREG32_SOC15(VCN, i, mmUVD_MPC_CNTL, tmp);
1152 
1153 		/* setup UVD_MPC_SET_MUXA0 */
1154 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXA0,
1155 			((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1156 			(0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1157 			(0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1158 			(0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1159 
1160 		/* setup UVD_MPC_SET_MUXB0 */
1161 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXB0,
1162 			((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1163 			(0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1164 			(0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1165 			(0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1166 
1167 		/* setup mmUVD_MPC_SET_MUX */
1168 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUX,
1169 			((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1170 			(0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1171 			(0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1172 
1173 		vcn_v3_0_mc_resume(adev, i);
1174 
1175 		/* VCN global tiling registers */
1176 		WREG32_SOC15(VCN, i, mmUVD_GFX10_ADDR_CONFIG,
1177 			adev->gfx.config.gb_addr_config);
1178 
1179 		/* unblock VCPU register access */
1180 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL), 0,
1181 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1182 
1183 		/* release VCPU reset to boot */
1184 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1185 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1186 
1187 		for (j = 0; j < 10; ++j) {
1188 			uint32_t status;
1189 
1190 			for (k = 0; k < 100; ++k) {
1191 				status = RREG32_SOC15(VCN, i, mmUVD_STATUS);
1192 				if (status & 2)
1193 					break;
1194 				mdelay(10);
1195 			}
1196 			r = 0;
1197 			if (status & 2)
1198 				break;
1199 
1200 			DRM_ERROR("VCN[%d] decode not responding, trying to reset the VCPU!!!\n", i);
1201 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1202 				UVD_VCPU_CNTL__BLK_RST_MASK,
1203 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1204 			mdelay(10);
1205 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1206 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1207 
1208 			mdelay(10);
1209 			r = -1;
1210 		}
1211 
1212 		if (r) {
1213 			DRM_ERROR("VCN[%d] decode not responding, giving up!!!\n", i);
1214 			return r;
1215 		}
1216 
1217 		/* enable master interrupt */
1218 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN),
1219 			UVD_MASTINT_EN__VCPU_EN_MASK,
1220 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1221 
1222 		/* clear the busy bit of VCN_STATUS */
1223 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS), 0,
1224 			~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1225 
1226 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_VMID, 0);
1227 
1228 		ring = &adev->vcn.inst[i].ring_dec;
1229 		/* force RBC into idle state */
1230 		rb_bufsz = order_base_2(ring->ring_size);
1231 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1232 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1233 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1234 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1235 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1236 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_CNTL, tmp);
1237 
1238 		fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
1239 		fw_shared->multi_queue.decode_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1240 
1241 		/* programm the RB_BASE for ring buffer */
1242 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1243 			lower_32_bits(ring->gpu_addr));
1244 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1245 			upper_32_bits(ring->gpu_addr));
1246 
1247 		/* Initialize the ring buffer's read and write pointers */
1248 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR, 0);
1249 
1250 		WREG32_SOC15(VCN, i, mmUVD_SCRATCH2, 0);
1251 		ring->wptr = RREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR);
1252 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_WPTR,
1253 			lower_32_bits(ring->wptr));
1254 		fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1255 		fw_shared->multi_queue.decode_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1256 
1257 		fw_shared->multi_queue.encode_generalpurpose_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1258 		ring = &adev->vcn.inst[i].ring_enc[0];
1259 		WREG32_SOC15(VCN, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1260 		WREG32_SOC15(VCN, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1261 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO, ring->gpu_addr);
1262 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1263 		WREG32_SOC15(VCN, i, mmUVD_RB_SIZE, ring->ring_size / 4);
1264 		fw_shared->multi_queue.encode_generalpurpose_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1265 
1266 		fw_shared->multi_queue.encode_lowlatency_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1267 		ring = &adev->vcn.inst[i].ring_enc[1];
1268 		WREG32_SOC15(VCN, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1269 		WREG32_SOC15(VCN, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1270 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1271 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1272 		WREG32_SOC15(VCN, i, mmUVD_RB_SIZE2, ring->ring_size / 4);
1273 		fw_shared->multi_queue.encode_lowlatency_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1274 	}
1275 
1276 	return 0;
1277 }
1278 
1279 static int vcn_v3_0_start_sriov(struct amdgpu_device *adev)
1280 {
1281 	int i, j;
1282 	struct amdgpu_ring *ring;
1283 	uint64_t cache_addr;
1284 	uint64_t rb_addr;
1285 	uint64_t ctx_addr;
1286 	uint32_t param, resp, expected;
1287 	uint32_t offset, cache_size;
1288 	uint32_t tmp, timeout;
1289 	uint32_t id;
1290 
1291 	struct amdgpu_mm_table *table = &adev->virt.mm_table;
1292 	uint32_t *table_loc;
1293 	uint32_t table_size;
1294 	uint32_t size, size_dw;
1295 
1296 	bool is_vcn_ready;
1297 
1298 	struct mmsch_v3_0_cmd_direct_write
1299 		direct_wt = { {0} };
1300 	struct mmsch_v3_0_cmd_direct_read_modify_write
1301 		direct_rd_mod_wt = { {0} };
1302 	struct mmsch_v3_0_cmd_end end = { {0} };
1303 	struct mmsch_v3_0_init_header header;
1304 
1305 	direct_wt.cmd_header.command_type =
1306 		MMSCH_COMMAND__DIRECT_REG_WRITE;
1307 	direct_rd_mod_wt.cmd_header.command_type =
1308 		MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1309 	end.cmd_header.command_type =
1310 		MMSCH_COMMAND__END;
1311 
1312 	header.version = MMSCH_VERSION;
1313 	header.total_size = sizeof(struct mmsch_v3_0_init_header) >> 2;
1314 	for (i = 0; i < AMDGPU_MAX_VCN_INSTANCES; i++) {
1315 		header.inst[i].init_status = 0;
1316 		header.inst[i].table_offset = 0;
1317 		header.inst[i].table_size = 0;
1318 	}
1319 
1320 	table_loc = (uint32_t *)table->cpu_addr;
1321 	table_loc += header.total_size;
1322 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1323 		if (adev->vcn.harvest_config & (1 << i))
1324 			continue;
1325 
1326 		table_size = 0;
1327 
1328 		MMSCH_V3_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(VCN, i,
1329 			mmUVD_STATUS),
1330 			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1331 
1332 		cache_size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1333 
1334 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1335 			id = amdgpu_ucode_id_vcns[i];
1336 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1337 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1338 				adev->firmware.ucode[id].tmr_mc_addr_lo);
1339 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1340 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1341 				adev->firmware.ucode[id].tmr_mc_addr_hi);
1342 			offset = 0;
1343 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1344 				mmUVD_VCPU_CACHE_OFFSET0),
1345 				0);
1346 		} else {
1347 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1348 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1349 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1350 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1351 				mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1352 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1353 			offset = cache_size;
1354 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1355 				mmUVD_VCPU_CACHE_OFFSET0),
1356 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1357 		}
1358 
1359 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1360 			mmUVD_VCPU_CACHE_SIZE0),
1361 			cache_size);
1362 
1363 		cache_addr = adev->vcn.inst[i].gpu_addr + offset;
1364 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1365 			mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1366 			lower_32_bits(cache_addr));
1367 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1368 			mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1369 			upper_32_bits(cache_addr));
1370 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1371 			mmUVD_VCPU_CACHE_OFFSET1),
1372 			0);
1373 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1374 			mmUVD_VCPU_CACHE_SIZE1),
1375 			AMDGPU_VCN_STACK_SIZE);
1376 
1377 		cache_addr = adev->vcn.inst[i].gpu_addr + offset +
1378 			AMDGPU_VCN_STACK_SIZE;
1379 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1380 			mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1381 			lower_32_bits(cache_addr));
1382 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1383 			mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1384 			upper_32_bits(cache_addr));
1385 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1386 			mmUVD_VCPU_CACHE_OFFSET2),
1387 			0);
1388 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1389 			mmUVD_VCPU_CACHE_SIZE2),
1390 			AMDGPU_VCN_CONTEXT_SIZE);
1391 
1392 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
1393 			ring = &adev->vcn.inst[i].ring_enc[j];
1394 			ring->wptr = 0;
1395 			rb_addr = ring->gpu_addr;
1396 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1397 				mmUVD_RB_BASE_LO),
1398 				lower_32_bits(rb_addr));
1399 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1400 				mmUVD_RB_BASE_HI),
1401 				upper_32_bits(rb_addr));
1402 			MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1403 				mmUVD_RB_SIZE),
1404 				ring->ring_size / 4);
1405 		}
1406 
1407 		ring = &adev->vcn.inst[i].ring_dec;
1408 		ring->wptr = 0;
1409 		rb_addr = ring->gpu_addr;
1410 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1411 			mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1412 			lower_32_bits(rb_addr));
1413 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1414 			mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1415 			upper_32_bits(rb_addr));
1416 		/* force RBC into idle state */
1417 		tmp = order_base_2(ring->ring_size);
1418 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, tmp);
1419 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1420 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1421 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1422 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1423 		MMSCH_V3_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1424 			mmUVD_RBC_RB_CNTL),
1425 			tmp);
1426 
1427 		/* add end packet */
1428 		MMSCH_V3_0_INSERT_END();
1429 
1430 		/* refine header */
1431 		header.inst[i].init_status = 0;
1432 		header.inst[i].table_offset = header.total_size;
1433 		header.inst[i].table_size = table_size;
1434 		header.total_size += table_size;
1435 	}
1436 
1437 	/* Update init table header in memory */
1438 	size = sizeof(struct mmsch_v3_0_init_header);
1439 	table_loc = (uint32_t *)table->cpu_addr;
1440 	memcpy((void *)table_loc, &header, size);
1441 
1442 	/* message MMSCH (in VCN[0]) to initialize this client
1443 	 * 1, write to mmsch_vf_ctx_addr_lo/hi register with GPU mc addr
1444 	 * of memory descriptor location
1445 	 */
1446 	ctx_addr = table->gpu_addr;
1447 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(ctx_addr));
1448 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(ctx_addr));
1449 
1450 	/* 2, update vmid of descriptor */
1451 	tmp = RREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID);
1452 	tmp &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1453 	/* use domain0 for MM scheduler */
1454 	tmp |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1455 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID, tmp);
1456 
1457 	/* 3, notify mmsch about the size of this descriptor */
1458 	size = header.total_size;
1459 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_SIZE, size);
1460 
1461 	/* 4, set resp to zero */
1462 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1463 
1464 	/* 5, kick off the initialization and wait until
1465 	 * MMSCH_VF_MAILBOX_RESP becomes non-zero
1466 	 */
1467 	param = 0x10000001;
1468 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_HOST, param);
1469 	tmp = 0;
1470 	timeout = 1000;
1471 	resp = 0;
1472 	expected = param + 1;
1473 	while (resp != expected) {
1474 		resp = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1475 		if (resp == expected)
1476 			break;
1477 
1478 		udelay(10);
1479 		tmp = tmp + 10;
1480 		if (tmp >= timeout) {
1481 			DRM_ERROR("failed to init MMSCH. TIME-OUT after %d usec"\
1482 				" waiting for mmMMSCH_VF_MAILBOX_RESP "\
1483 				"(expected=0x%08x, readback=0x%08x)\n",
1484 				tmp, expected, resp);
1485 			return -EBUSY;
1486 		}
1487 	}
1488 
1489 	/* 6, check each VCN's init_status
1490 	 * if it remains as 0, then this VCN is not assigned to current VF
1491 	 * do not start ring for this VCN
1492 	 */
1493 	size = sizeof(struct mmsch_v3_0_init_header);
1494 	table_loc = (uint32_t *)table->cpu_addr;
1495 	memcpy(&header, (void *)table_loc, size);
1496 
1497 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1498 		if (adev->vcn.harvest_config & (1 << i))
1499 			continue;
1500 
1501 		is_vcn_ready = (header.inst[i].init_status == 1);
1502 		if (!is_vcn_ready)
1503 			DRM_INFO("VCN(%d) engine is disabled by hypervisor\n", i);
1504 
1505 		ring = &adev->vcn.inst[i].ring_dec;
1506 		ring->sched.ready = is_vcn_ready;
1507 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
1508 			ring = &adev->vcn.inst[i].ring_enc[j];
1509 			ring->sched.ready = is_vcn_ready;
1510 		}
1511 	}
1512 
1513 	return 0;
1514 }
1515 
1516 static int vcn_v3_0_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1517 {
1518 	uint32_t tmp;
1519 
1520 	/* Wait for power status to be 1 */
1521 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1522 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1523 
1524 	/* wait for read ptr to be equal to write ptr */
1525 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR);
1526 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1527 
1528 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2);
1529 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
1530 
1531 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1532 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF);
1533 
1534 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1535 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1536 
1537 	/* disable dynamic power gating mode */
1538 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 0,
1539 		~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1540 
1541 	return 0;
1542 }
1543 
1544 static int vcn_v3_0_stop(struct amdgpu_device *adev)
1545 {
1546 	uint32_t tmp;
1547 	int i, r = 0;
1548 
1549 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1550 		if (adev->vcn.harvest_config & (1 << i))
1551 			continue;
1552 
1553 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1554 			r = vcn_v3_0_stop_dpg_mode(adev, i);
1555 			continue;
1556 		}
1557 
1558 		/* wait for vcn idle */
1559 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1560 		if (r)
1561 			return r;
1562 
1563 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1564 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1565 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1566 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1567 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1568 		if (r)
1569 			return r;
1570 
1571 		/* disable LMI UMC channel */
1572 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2);
1573 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1574 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2, tmp);
1575 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1576 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1577 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1578 		if (r)
1579 			return r;
1580 
1581 		/* block VCPU register access */
1582 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL),
1583 			UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1584 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1585 
1586 		/* reset VCPU */
1587 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1588 			UVD_VCPU_CNTL__BLK_RST_MASK,
1589 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1590 
1591 		/* disable VCPU clock */
1592 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1593 			~(UVD_VCPU_CNTL__CLK_EN_MASK));
1594 
1595 		/* apply soft reset */
1596 		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1597 		tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1598 		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1599 		tmp = RREG32_SOC15(VCN, i, mmUVD_SOFT_RESET);
1600 		tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1601 		WREG32_SOC15(VCN, i, mmUVD_SOFT_RESET, tmp);
1602 
1603 		/* clear status */
1604 		WREG32_SOC15(VCN, i, mmUVD_STATUS, 0);
1605 
1606 		/* apply HW clock gating */
1607 		vcn_v3_0_enable_clock_gating(adev, i);
1608 
1609 		/* enable VCN power gating */
1610 		vcn_v3_0_enable_static_power_gating(adev, i);
1611 	}
1612 
1613 	if (adev->pm.dpm_enabled)
1614 		amdgpu_dpm_enable_uvd(adev, false);
1615 
1616 	return 0;
1617 }
1618 
1619 static int vcn_v3_0_pause_dpg_mode(struct amdgpu_device *adev,
1620 		   int inst_idx, struct dpg_pause_state *new_state)
1621 {
1622 	volatile struct amdgpu_fw_shared *fw_shared;
1623 	struct amdgpu_ring *ring;
1624 	uint32_t reg_data = 0;
1625 	int ret_code;
1626 
1627 	/* pause/unpause if state is changed */
1628 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1629 		DRM_DEBUG("dpg pause state changed %d -> %d",
1630 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1631 		reg_data = RREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE) &
1632 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1633 
1634 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1635 			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1636 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1637 
1638 			if (!ret_code) {
1639 				/* pause DPG */
1640 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1641 				WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1642 
1643 				/* wait for ACK */
1644 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_DPG_PAUSE,
1645 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1646 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1647 
1648 				/* Stall DPG before WPTR/RPTR reset */
1649 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1650 					UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1651 					~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1652 
1653 				/* Restore */
1654 				fw_shared = adev->vcn.inst[inst_idx].fw_shared_cpu_addr;
1655 				fw_shared->multi_queue.encode_generalpurpose_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1656 				ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1657 				ring->wptr = 0;
1658 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
1659 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1660 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
1661 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1662 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1663 				fw_shared->multi_queue.encode_generalpurpose_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1664 
1665 				fw_shared->multi_queue.encode_lowlatency_queue_mode |= cpu_to_le32(FW_QUEUE_RING_RESET);
1666 				ring = &adev->vcn.inst[inst_idx].ring_enc[1];
1667 				ring->wptr = 0;
1668 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1669 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1670 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
1671 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1672 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1673 				fw_shared->multi_queue.encode_lowlatency_queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
1674 
1675 				/* restore wptr/rptr with pointers saved in FW shared memory*/
1676 				WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, fw_shared->rb.rptr);
1677 				WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR, fw_shared->rb.wptr);
1678 
1679 				/* Unstall DPG */
1680 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1681 					0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1682 
1683 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS,
1684 					UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1685 			}
1686 		} else {
1687 			/* unpause dpg, no need to wait */
1688 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1689 			WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1690 		}
1691 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1692 	}
1693 
1694 	return 0;
1695 }
1696 
1697 /**
1698  * vcn_v3_0_dec_ring_get_rptr - get read pointer
1699  *
1700  * @ring: amdgpu_ring pointer
1701  *
1702  * Returns the current hardware read pointer
1703  */
1704 static uint64_t vcn_v3_0_dec_ring_get_rptr(struct amdgpu_ring *ring)
1705 {
1706 	struct amdgpu_device *adev = ring->adev;
1707 
1708 	return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_RPTR);
1709 }
1710 
1711 /**
1712  * vcn_v3_0_dec_ring_get_wptr - get write pointer
1713  *
1714  * @ring: amdgpu_ring pointer
1715  *
1716  * Returns the current hardware write pointer
1717  */
1718 static uint64_t vcn_v3_0_dec_ring_get_wptr(struct amdgpu_ring *ring)
1719 {
1720 	struct amdgpu_device *adev = ring->adev;
1721 
1722 	if (ring->use_doorbell)
1723 		return adev->wb.wb[ring->wptr_offs];
1724 	else
1725 		return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR);
1726 }
1727 
1728 /**
1729  * vcn_v3_0_dec_ring_set_wptr - set write pointer
1730  *
1731  * @ring: amdgpu_ring pointer
1732  *
1733  * Commits the write pointer to the hardware
1734  */
1735 static void vcn_v3_0_dec_ring_set_wptr(struct amdgpu_ring *ring)
1736 {
1737 	struct amdgpu_device *adev = ring->adev;
1738 	volatile struct amdgpu_fw_shared *fw_shared;
1739 
1740 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1741 		/*whenever update RBC_RB_WPTR, we save the wptr in shared rb.wptr and scratch2 */
1742 		fw_shared = adev->vcn.inst[ring->me].fw_shared_cpu_addr;
1743 		fw_shared->rb.wptr = lower_32_bits(ring->wptr);
1744 		WREG32_SOC15(VCN, ring->me, mmUVD_SCRATCH2,
1745 			lower_32_bits(ring->wptr));
1746 	}
1747 
1748 	if (ring->use_doorbell) {
1749 		adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1750 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1751 	} else {
1752 		WREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1753 	}
1754 }
1755 
1756 static void vcn_v3_0_dec_sw_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
1757 				u64 seq, uint32_t flags)
1758 {
1759 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
1760 
1761 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_FENCE);
1762 	amdgpu_ring_write(ring, addr);
1763 	amdgpu_ring_write(ring, upper_32_bits(addr));
1764 	amdgpu_ring_write(ring, seq);
1765 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_TRAP);
1766 }
1767 
1768 static void vcn_v3_0_dec_sw_ring_insert_end(struct amdgpu_ring *ring)
1769 {
1770 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_END);
1771 }
1772 
1773 static void vcn_v3_0_dec_sw_ring_emit_ib(struct amdgpu_ring *ring,
1774 			       struct amdgpu_job *job,
1775 			       struct amdgpu_ib *ib,
1776 			       uint32_t flags)
1777 {
1778 	uint32_t vmid = AMDGPU_JOB_GET_VMID(job);
1779 
1780 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_IB);
1781 	amdgpu_ring_write(ring, vmid);
1782 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
1783 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
1784 	amdgpu_ring_write(ring, ib->length_dw);
1785 }
1786 
1787 static void vcn_v3_0_dec_sw_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
1788 				uint32_t val, uint32_t mask)
1789 {
1790 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_REG_WAIT);
1791 	amdgpu_ring_write(ring, reg << 2);
1792 	amdgpu_ring_write(ring, mask);
1793 	amdgpu_ring_write(ring, val);
1794 }
1795 
1796 static void vcn_v3_0_dec_sw_ring_emit_vm_flush(struct amdgpu_ring *ring,
1797 				uint32_t vmid, uint64_t pd_addr)
1798 {
1799 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
1800 	uint32_t data0, data1, mask;
1801 
1802 	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1803 
1804 	/* wait for register write */
1805 	data0 = hub->ctx0_ptb_addr_lo32 + vmid * hub->ctx_addr_distance;
1806 	data1 = lower_32_bits(pd_addr);
1807 	mask = 0xffffffff;
1808 	vcn_v3_0_dec_sw_ring_emit_reg_wait(ring, data0, data1, mask);
1809 }
1810 
1811 static void vcn_v3_0_dec_sw_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg, uint32_t val)
1812 {
1813 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_REG_WRITE);
1814 	amdgpu_ring_write(ring,	reg << 2);
1815 	amdgpu_ring_write(ring, val);
1816 }
1817 
1818 static const struct amdgpu_ring_funcs vcn_v3_0_dec_sw_ring_vm_funcs = {
1819 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1820 	.align_mask = 0x3f,
1821 	.nop = VCN_DEC_SW_CMD_NO_OP,
1822 	.vmhub = AMDGPU_MMHUB_0,
1823 	.get_rptr = vcn_v3_0_dec_ring_get_rptr,
1824 	.get_wptr = vcn_v3_0_dec_ring_get_wptr,
1825 	.set_wptr = vcn_v3_0_dec_ring_set_wptr,
1826 	.emit_frame_size =
1827 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1828 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1829 		4 + /* vcn_v3_0_dec_sw_ring_emit_vm_flush */
1830 		5 + 5 + /* vcn_v3_0_dec_sw_ring_emit_fdec_swe x2 vm fdec_swe */
1831 		1, /* vcn_v3_0_dec_sw_ring_insert_end */
1832 	.emit_ib_size = 5, /* vcn_v3_0_dec_sw_ring_emit_ib */
1833 	.emit_ib = vcn_v3_0_dec_sw_ring_emit_ib,
1834 	.emit_fence = vcn_v3_0_dec_sw_ring_emit_fence,
1835 	.emit_vm_flush = vcn_v3_0_dec_sw_ring_emit_vm_flush,
1836 	.test_ring = amdgpu_vcn_dec_sw_ring_test_ring,
1837 	.test_ib = NULL,//amdgpu_vcn_dec_sw_ring_test_ib,
1838 	.insert_nop = amdgpu_ring_insert_nop,
1839 	.insert_end = vcn_v3_0_dec_sw_ring_insert_end,
1840 	.pad_ib = amdgpu_ring_generic_pad_ib,
1841 	.begin_use = amdgpu_vcn_ring_begin_use,
1842 	.end_use = amdgpu_vcn_ring_end_use,
1843 	.emit_wreg = vcn_v3_0_dec_sw_ring_emit_wreg,
1844 	.emit_reg_wait = vcn_v3_0_dec_sw_ring_emit_reg_wait,
1845 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1846 };
1847 
1848 static int vcn_v3_0_limit_sched(struct amdgpu_cs_parser *p)
1849 {
1850 	struct drm_gpu_scheduler **scheds;
1851 
1852 	/* The create msg must be in the first IB submitted */
1853 	if (atomic_read(&p->entity->fence_seq))
1854 		return -EINVAL;
1855 
1856 	scheds = p->adev->gpu_sched[AMDGPU_HW_IP_VCN_DEC]
1857 		[AMDGPU_RING_PRIO_DEFAULT].sched;
1858 	drm_sched_entity_modify_sched(p->entity, scheds, 1);
1859 	return 0;
1860 }
1861 
1862 static int vcn_v3_0_dec_msg(struct amdgpu_cs_parser *p, uint64_t addr)
1863 {
1864 	struct ttm_operation_ctx ctx = { false, false };
1865 	struct amdgpu_bo_va_mapping *map;
1866 	uint32_t *msg, num_buffers;
1867 	struct amdgpu_bo *bo;
1868 	uint64_t start, end;
1869 	unsigned int i;
1870 	void * ptr;
1871 	int r;
1872 
1873 	addr &= AMDGPU_GMC_HOLE_MASK;
1874 	r = amdgpu_cs_find_mapping(p, addr, &bo, &map);
1875 	if (r) {
1876 		DRM_ERROR("Can't find BO for addr 0x%08Lx\n", addr);
1877 		return r;
1878 	}
1879 
1880 	start = map->start * AMDGPU_GPU_PAGE_SIZE;
1881 	end = (map->last + 1) * AMDGPU_GPU_PAGE_SIZE;
1882 	if (addr & 0x7) {
1883 		DRM_ERROR("VCN messages must be 8 byte aligned!\n");
1884 		return -EINVAL;
1885 	}
1886 
1887 	bo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
1888 	amdgpu_bo_placement_from_domain(bo, bo->allowed_domains);
1889 	r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
1890 	if (r) {
1891 		DRM_ERROR("Failed validating the VCN message BO (%d)!\n", r);
1892 		return r;
1893 	}
1894 
1895 	r = amdgpu_bo_kmap(bo, &ptr);
1896 	if (r) {
1897 		DRM_ERROR("Failed mapping the VCN message (%d)!\n", r);
1898 		return r;
1899 	}
1900 
1901 	msg = ptr + addr - start;
1902 
1903 	/* Check length */
1904 	if (msg[1] > end - addr) {
1905 		r = -EINVAL;
1906 		goto out;
1907 	}
1908 
1909 	if (msg[3] != RDECODE_MSG_CREATE)
1910 		goto out;
1911 
1912 	num_buffers = msg[2];
1913 	for (i = 0, msg = &msg[6]; i < num_buffers; ++i, msg += 4) {
1914 		uint32_t offset, size, *create;
1915 
1916 		if (msg[0] != RDECODE_MESSAGE_CREATE)
1917 			continue;
1918 
1919 		offset = msg[1];
1920 		size = msg[2];
1921 
1922 		if (offset + size > end) {
1923 			r = -EINVAL;
1924 			goto out;
1925 		}
1926 
1927 		create = ptr + addr + offset - start;
1928 
1929 		/* H246, HEVC and VP9 can run on any instance */
1930 		if (create[0] == 0x7 || create[0] == 0x10 || create[0] == 0x11)
1931 			continue;
1932 
1933 		r = vcn_v3_0_limit_sched(p);
1934 		if (r)
1935 			goto out;
1936 	}
1937 
1938 out:
1939 	amdgpu_bo_kunmap(bo);
1940 	return r;
1941 }
1942 
1943 static int vcn_v3_0_ring_patch_cs_in_place(struct amdgpu_cs_parser *p,
1944 					   uint32_t ib_idx)
1945 {
1946 	struct amdgpu_ring *ring = to_amdgpu_ring(p->entity->rq->sched);
1947 	struct amdgpu_ib *ib = &p->job->ibs[ib_idx];
1948 	uint32_t msg_lo = 0, msg_hi = 0;
1949 	unsigned i;
1950 	int r;
1951 
1952 	/* The first instance can decode anything */
1953 	if (!ring->me)
1954 		return 0;
1955 
1956 	for (i = 0; i < ib->length_dw; i += 2) {
1957 		uint32_t reg = amdgpu_get_ib_value(p, ib_idx, i);
1958 		uint32_t val = amdgpu_get_ib_value(p, ib_idx, i + 1);
1959 
1960 		if (reg == PACKET0(p->adev->vcn.internal.data0, 0)) {
1961 			msg_lo = val;
1962 		} else if (reg == PACKET0(p->adev->vcn.internal.data1, 0)) {
1963 			msg_hi = val;
1964 		} else if (reg == PACKET0(p->adev->vcn.internal.cmd, 0) &&
1965 			   val == 0) {
1966 			r = vcn_v3_0_dec_msg(p, ((u64)msg_hi) << 32 | msg_lo);
1967 			if (r)
1968 				return r;
1969 		}
1970 	}
1971 	return 0;
1972 }
1973 
1974 static const struct amdgpu_ring_funcs vcn_v3_0_dec_ring_vm_funcs = {
1975 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1976 	.align_mask = 0xf,
1977 	.vmhub = AMDGPU_MMHUB_0,
1978 	.get_rptr = vcn_v3_0_dec_ring_get_rptr,
1979 	.get_wptr = vcn_v3_0_dec_ring_get_wptr,
1980 	.set_wptr = vcn_v3_0_dec_ring_set_wptr,
1981 	.patch_cs_in_place = vcn_v3_0_ring_patch_cs_in_place,
1982 	.emit_frame_size =
1983 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1984 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1985 		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1986 		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1987 		6,
1988 	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1989 	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
1990 	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
1991 	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1992 	.test_ring = vcn_v2_0_dec_ring_test_ring,
1993 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
1994 	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
1995 	.insert_start = vcn_v2_0_dec_ring_insert_start,
1996 	.insert_end = vcn_v2_0_dec_ring_insert_end,
1997 	.pad_ib = amdgpu_ring_generic_pad_ib,
1998 	.begin_use = amdgpu_vcn_ring_begin_use,
1999 	.end_use = amdgpu_vcn_ring_end_use,
2000 	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
2001 	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
2002 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2003 };
2004 
2005 /**
2006  * vcn_v3_0_enc_ring_get_rptr - get enc read pointer
2007  *
2008  * @ring: amdgpu_ring pointer
2009  *
2010  * Returns the current hardware enc read pointer
2011  */
2012 static uint64_t vcn_v3_0_enc_ring_get_rptr(struct amdgpu_ring *ring)
2013 {
2014 	struct amdgpu_device *adev = ring->adev;
2015 
2016 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0])
2017 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR);
2018 	else
2019 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR2);
2020 }
2021 
2022 /**
2023  * vcn_v3_0_enc_ring_get_wptr - get enc write pointer
2024  *
2025  * @ring: amdgpu_ring pointer
2026  *
2027  * Returns the current hardware enc write pointer
2028  */
2029 static uint64_t vcn_v3_0_enc_ring_get_wptr(struct amdgpu_ring *ring)
2030 {
2031 	struct amdgpu_device *adev = ring->adev;
2032 
2033 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
2034 		if (ring->use_doorbell)
2035 			return adev->wb.wb[ring->wptr_offs];
2036 		else
2037 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR);
2038 	} else {
2039 		if (ring->use_doorbell)
2040 			return adev->wb.wb[ring->wptr_offs];
2041 		else
2042 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2);
2043 	}
2044 }
2045 
2046 /**
2047  * vcn_v3_0_enc_ring_set_wptr - set enc write pointer
2048  *
2049  * @ring: amdgpu_ring pointer
2050  *
2051  * Commits the enc write pointer to the hardware
2052  */
2053 static void vcn_v3_0_enc_ring_set_wptr(struct amdgpu_ring *ring)
2054 {
2055 	struct amdgpu_device *adev = ring->adev;
2056 
2057 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
2058 		if (ring->use_doorbell) {
2059 			adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
2060 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
2061 		} else {
2062 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
2063 		}
2064 	} else {
2065 		if (ring->use_doorbell) {
2066 			adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
2067 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
2068 		} else {
2069 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
2070 		}
2071 	}
2072 }
2073 
2074 static const struct amdgpu_ring_funcs vcn_v3_0_enc_ring_vm_funcs = {
2075 	.type = AMDGPU_RING_TYPE_VCN_ENC,
2076 	.align_mask = 0x3f,
2077 	.nop = VCN_ENC_CMD_NO_OP,
2078 	.vmhub = AMDGPU_MMHUB_0,
2079 	.get_rptr = vcn_v3_0_enc_ring_get_rptr,
2080 	.get_wptr = vcn_v3_0_enc_ring_get_wptr,
2081 	.set_wptr = vcn_v3_0_enc_ring_set_wptr,
2082 	.emit_frame_size =
2083 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2084 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
2085 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
2086 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
2087 		1, /* vcn_v2_0_enc_ring_insert_end */
2088 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
2089 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
2090 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
2091 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
2092 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
2093 	.test_ib = amdgpu_vcn_enc_ring_test_ib,
2094 	.insert_nop = amdgpu_ring_insert_nop,
2095 	.insert_end = vcn_v2_0_enc_ring_insert_end,
2096 	.pad_ib = amdgpu_ring_generic_pad_ib,
2097 	.begin_use = amdgpu_vcn_ring_begin_use,
2098 	.end_use = amdgpu_vcn_ring_end_use,
2099 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
2100 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
2101 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2102 };
2103 
2104 static void vcn_v3_0_set_dec_ring_funcs(struct amdgpu_device *adev)
2105 {
2106 	int i;
2107 
2108 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2109 		if (adev->vcn.harvest_config & (1 << i))
2110 			continue;
2111 
2112 		if (!DEC_SW_RING_ENABLED)
2113 			adev->vcn.inst[i].ring_dec.funcs = &vcn_v3_0_dec_ring_vm_funcs;
2114 		else
2115 			adev->vcn.inst[i].ring_dec.funcs = &vcn_v3_0_dec_sw_ring_vm_funcs;
2116 		adev->vcn.inst[i].ring_dec.me = i;
2117 		DRM_INFO("VCN(%d) decode%s is enabled in VM mode\n", i,
2118 			  DEC_SW_RING_ENABLED?"(Software Ring)":"");
2119 	}
2120 }
2121 
2122 static void vcn_v3_0_set_enc_ring_funcs(struct amdgpu_device *adev)
2123 {
2124 	int i, j;
2125 
2126 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2127 		if (adev->vcn.harvest_config & (1 << i))
2128 			continue;
2129 
2130 		for (j = 0; j < adev->vcn.num_enc_rings; ++j) {
2131 			adev->vcn.inst[i].ring_enc[j].funcs = &vcn_v3_0_enc_ring_vm_funcs;
2132 			adev->vcn.inst[i].ring_enc[j].me = i;
2133 		}
2134 		DRM_INFO("VCN(%d) encode is enabled in VM mode\n", i);
2135 	}
2136 }
2137 
2138 static bool vcn_v3_0_is_idle(void *handle)
2139 {
2140 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2141 	int i, ret = 1;
2142 
2143 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2144 		if (adev->vcn.harvest_config & (1 << i))
2145 			continue;
2146 
2147 		ret &= (RREG32_SOC15(VCN, i, mmUVD_STATUS) == UVD_STATUS__IDLE);
2148 	}
2149 
2150 	return ret;
2151 }
2152 
2153 static int vcn_v3_0_wait_for_idle(void *handle)
2154 {
2155 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2156 	int i, ret = 0;
2157 
2158 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2159 		if (adev->vcn.harvest_config & (1 << i))
2160 			continue;
2161 
2162 		ret = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE,
2163 			UVD_STATUS__IDLE);
2164 		if (ret)
2165 			return ret;
2166 	}
2167 
2168 	return ret;
2169 }
2170 
2171 static int vcn_v3_0_set_clockgating_state(void *handle,
2172 					  enum amd_clockgating_state state)
2173 {
2174 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2175 	bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
2176 	int i;
2177 
2178 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2179 		if (adev->vcn.harvest_config & (1 << i))
2180 			continue;
2181 
2182 		if (enable) {
2183 			if (RREG32_SOC15(VCN, i, mmUVD_STATUS) != UVD_STATUS__IDLE)
2184 				return -EBUSY;
2185 			vcn_v3_0_enable_clock_gating(adev, i);
2186 		} else {
2187 			vcn_v3_0_disable_clock_gating(adev, i);
2188 		}
2189 	}
2190 
2191 	return 0;
2192 }
2193 
2194 static int vcn_v3_0_set_powergating_state(void *handle,
2195 					  enum amd_powergating_state state)
2196 {
2197 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2198 	int ret;
2199 
2200 	/* for SRIOV, guest should not control VCN Power-gating
2201 	 * MMSCH FW should control Power-gating and clock-gating
2202 	 * guest should avoid touching CGC and PG
2203 	 */
2204 	if (amdgpu_sriov_vf(adev)) {
2205 		adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
2206 		return 0;
2207 	}
2208 
2209 	if(state == adev->vcn.cur_state)
2210 		return 0;
2211 
2212 	if (state == AMD_PG_STATE_GATE)
2213 		ret = vcn_v3_0_stop(adev);
2214 	else
2215 		ret = vcn_v3_0_start(adev);
2216 
2217 	if(!ret)
2218 		adev->vcn.cur_state = state;
2219 
2220 	return ret;
2221 }
2222 
2223 static int vcn_v3_0_set_interrupt_state(struct amdgpu_device *adev,
2224 					struct amdgpu_irq_src *source,
2225 					unsigned type,
2226 					enum amdgpu_interrupt_state state)
2227 {
2228 	return 0;
2229 }
2230 
2231 static int vcn_v3_0_process_interrupt(struct amdgpu_device *adev,
2232 				      struct amdgpu_irq_src *source,
2233 				      struct amdgpu_iv_entry *entry)
2234 {
2235 	uint32_t ip_instance;
2236 
2237 	switch (entry->client_id) {
2238 	case SOC15_IH_CLIENTID_VCN:
2239 		ip_instance = 0;
2240 		break;
2241 	case SOC15_IH_CLIENTID_VCN1:
2242 		ip_instance = 1;
2243 		break;
2244 	default:
2245 		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
2246 		return 0;
2247 	}
2248 
2249 	DRM_DEBUG("IH: VCN TRAP\n");
2250 
2251 	switch (entry->src_id) {
2252 	case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
2253 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec);
2254 		break;
2255 	case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
2256 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
2257 		break;
2258 	case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
2259 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]);
2260 		break;
2261 	default:
2262 		DRM_ERROR("Unhandled interrupt: %d %d\n",
2263 			  entry->src_id, entry->src_data[0]);
2264 		break;
2265 	}
2266 
2267 	return 0;
2268 }
2269 
2270 static const struct amdgpu_irq_src_funcs vcn_v3_0_irq_funcs = {
2271 	.set = vcn_v3_0_set_interrupt_state,
2272 	.process = vcn_v3_0_process_interrupt,
2273 };
2274 
2275 static void vcn_v3_0_set_irq_funcs(struct amdgpu_device *adev)
2276 {
2277 	int i;
2278 
2279 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2280 		if (adev->vcn.harvest_config & (1 << i))
2281 			continue;
2282 
2283 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
2284 		adev->vcn.inst[i].irq.funcs = &vcn_v3_0_irq_funcs;
2285 	}
2286 }
2287 
2288 static const struct amd_ip_funcs vcn_v3_0_ip_funcs = {
2289 	.name = "vcn_v3_0",
2290 	.early_init = vcn_v3_0_early_init,
2291 	.late_init = NULL,
2292 	.sw_init = vcn_v3_0_sw_init,
2293 	.sw_fini = vcn_v3_0_sw_fini,
2294 	.hw_init = vcn_v3_0_hw_init,
2295 	.hw_fini = vcn_v3_0_hw_fini,
2296 	.suspend = vcn_v3_0_suspend,
2297 	.resume = vcn_v3_0_resume,
2298 	.is_idle = vcn_v3_0_is_idle,
2299 	.wait_for_idle = vcn_v3_0_wait_for_idle,
2300 	.check_soft_reset = NULL,
2301 	.pre_soft_reset = NULL,
2302 	.soft_reset = NULL,
2303 	.post_soft_reset = NULL,
2304 	.set_clockgating_state = vcn_v3_0_set_clockgating_state,
2305 	.set_powergating_state = vcn_v3_0_set_powergating_state,
2306 };
2307 
2308 const struct amdgpu_ip_block_version vcn_v3_0_ip_block =
2309 {
2310 	.type = AMD_IP_BLOCK_TYPE_VCN,
2311 	.major = 3,
2312 	.minor = 0,
2313 	.rev = 0,
2314 	.funcs = &vcn_v3_0_ip_funcs,
2315 };
2316