xref: /linux/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_3.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 /*
2  * Copyright 2022 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #include <linux/firmware.h>
25 #include <drm/drm_drv.h>
26 
27 #include "amdgpu.h"
28 #include "amdgpu_vcn.h"
29 #include "amdgpu_pm.h"
30 #include "soc15.h"
31 #include "soc15d.h"
32 #include "soc15_hw_ip.h"
33 #include "vcn_v2_0.h"
34 #include "mmsch_v4_0_3.h"
35 
36 #include "vcn/vcn_4_0_3_offset.h"
37 #include "vcn/vcn_4_0_3_sh_mask.h"
38 #include "ivsrcid/vcn/irqsrcs_vcn_4_0.h"
39 
40 #define mmUVD_DPG_LMA_CTL		regUVD_DPG_LMA_CTL
41 #define mmUVD_DPG_LMA_CTL_BASE_IDX	regUVD_DPG_LMA_CTL_BASE_IDX
42 #define mmUVD_DPG_LMA_DATA		regUVD_DPG_LMA_DATA
43 #define mmUVD_DPG_LMA_DATA_BASE_IDX	regUVD_DPG_LMA_DATA_BASE_IDX
44 
45 #define VCN_VID_SOC_ADDRESS_2_0		0x1fb00
46 #define VCN1_VID_SOC_ADDRESS_3_0	0x48300
47 
48 static const struct amdgpu_hwip_reg_entry vcn_reg_list_4_0_3[] = {
49 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_POWER_STATUS),
50 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_STATUS),
51 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_CONTEXT_ID),
52 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_CONTEXT_ID2),
53 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_GPCOM_VCPU_DATA0),
54 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_GPCOM_VCPU_DATA1),
55 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_GPCOM_VCPU_CMD),
56 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_HI),
57 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_LO),
58 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_HI2),
59 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_LO2),
60 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_HI3),
61 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_LO3),
62 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_HI4),
63 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_LO4),
64 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_RPTR),
65 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_WPTR),
66 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_RPTR2),
67 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_WPTR2),
68 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_RPTR3),
69 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_WPTR3),
70 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_RPTR4),
71 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_WPTR4),
72 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_SIZE),
73 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_SIZE2),
74 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_SIZE3),
75 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_SIZE4),
76 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_PGFSM_CONFIG),
77 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_PGFSM_STATUS),
78 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_DPG_LMA_CTL),
79 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_DPG_LMA_DATA),
80 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_DPG_LMA_MASK),
81 	SOC15_REG_ENTRY_STR(VCN, 0, regUVD_DPG_PAUSE)
82 };
83 
84 #define NORMALIZE_VCN_REG_OFFSET(offset) \
85 		(offset & 0x1FFFF)
86 
87 static int vcn_v4_0_3_start_sriov(struct amdgpu_device *adev);
88 static void vcn_v4_0_3_set_unified_ring_funcs(struct amdgpu_device *adev);
89 static void vcn_v4_0_3_set_irq_funcs(struct amdgpu_device *adev);
90 static int vcn_v4_0_3_set_powergating_state(void *handle,
91 		enum amd_powergating_state state);
92 static int vcn_v4_0_3_pause_dpg_mode(struct amdgpu_device *adev,
93 		int inst_idx, struct dpg_pause_state *new_state);
94 static void vcn_v4_0_3_unified_ring_set_wptr(struct amdgpu_ring *ring);
95 static void vcn_v4_0_3_set_ras_funcs(struct amdgpu_device *adev);
96 static void vcn_v4_0_3_enable_ras(struct amdgpu_device *adev,
97 				  int inst_idx, bool indirect);
98 /**
99  * vcn_v4_0_3_early_init - set function pointers
100  *
101  * @handle: amdgpu_device pointer
102  *
103  * Set ring and irq function pointers
104  */
105 static int vcn_v4_0_3_early_init(void *handle)
106 {
107 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
108 
109 	/* re-use enc ring as unified ring */
110 	adev->vcn.num_enc_rings = 1;
111 
112 	vcn_v4_0_3_set_unified_ring_funcs(adev);
113 	vcn_v4_0_3_set_irq_funcs(adev);
114 	vcn_v4_0_3_set_ras_funcs(adev);
115 
116 	return amdgpu_vcn_early_init(adev);
117 }
118 
119 /**
120  * vcn_v4_0_3_sw_init - sw init for VCN block
121  *
122  * @handle: amdgpu_device pointer
123  *
124  * Load firmware and sw initialization
125  */
126 static int vcn_v4_0_3_sw_init(void *handle)
127 {
128 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
129 	struct amdgpu_ring *ring;
130 	int i, r, vcn_inst;
131 	uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_4_0_3);
132 	uint32_t *ptr;
133 
134 	r = amdgpu_vcn_sw_init(adev);
135 	if (r)
136 		return r;
137 
138 	amdgpu_vcn_setup_ucode(adev);
139 
140 	r = amdgpu_vcn_resume(adev);
141 	if (r)
142 		return r;
143 
144 	/* VCN DEC TRAP */
145 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VCN,
146 		VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst->irq);
147 	if (r)
148 		return r;
149 
150 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
151 		volatile struct amdgpu_vcn4_fw_shared *fw_shared;
152 
153 		vcn_inst = GET_INST(VCN, i);
154 
155 		ring = &adev->vcn.inst[i].ring_enc[0];
156 		ring->use_doorbell = true;
157 
158 		if (!amdgpu_sriov_vf(adev))
159 			ring->doorbell_index =
160 				(adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
161 				9 * vcn_inst;
162 		else
163 			ring->doorbell_index =
164 				(adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
165 				32 * vcn_inst;
166 
167 		ring->vm_hub = AMDGPU_MMHUB0(adev->vcn.inst[i].aid_id);
168 		sprintf(ring->name, "vcn_unified_%d", adev->vcn.inst[i].aid_id);
169 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0,
170 				     AMDGPU_RING_PRIO_DEFAULT,
171 				     &adev->vcn.inst[i].sched_score);
172 		if (r)
173 			return r;
174 
175 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
176 		fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_FW_SHARED_FLAG_0_UNIFIED_QUEUE);
177 		fw_shared->sq.is_enabled = true;
178 
179 		if (amdgpu_vcnfw_log)
180 			amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
181 	}
182 
183 	if (amdgpu_sriov_vf(adev)) {
184 		r = amdgpu_virt_alloc_mm_table(adev);
185 		if (r)
186 			return r;
187 	}
188 
189 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
190 		adev->vcn.pause_dpg_mode = vcn_v4_0_3_pause_dpg_mode;
191 
192 	if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN)) {
193 		r = amdgpu_vcn_ras_sw_init(adev);
194 		if (r) {
195 			dev_err(adev->dev, "Failed to initialize vcn ras block!\n");
196 			return r;
197 		}
198 	}
199 
200 	/* Allocate memory for VCN IP Dump buffer */
201 	ptr = kcalloc(adev->vcn.num_vcn_inst * reg_count, sizeof(uint32_t), GFP_KERNEL);
202 	if (!ptr) {
203 		DRM_ERROR("Failed to allocate memory for VCN IP Dump\n");
204 		adev->vcn.ip_dump = NULL;
205 	} else {
206 		adev->vcn.ip_dump = ptr;
207 	}
208 
209 	return 0;
210 }
211 
212 /**
213  * vcn_v4_0_3_sw_fini - sw fini for VCN block
214  *
215  * @handle: amdgpu_device pointer
216  *
217  * VCN suspend and free up sw allocation
218  */
219 static int vcn_v4_0_3_sw_fini(void *handle)
220 {
221 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
222 	int i, r, idx;
223 
224 	if (drm_dev_enter(&adev->ddev, &idx)) {
225 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
226 			volatile struct amdgpu_vcn4_fw_shared *fw_shared;
227 
228 			fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
229 			fw_shared->present_flag_0 = 0;
230 			fw_shared->sq.is_enabled = cpu_to_le32(false);
231 		}
232 		drm_dev_exit(idx);
233 	}
234 
235 	if (amdgpu_sriov_vf(adev))
236 		amdgpu_virt_free_mm_table(adev);
237 
238 	r = amdgpu_vcn_suspend(adev);
239 	if (r)
240 		return r;
241 
242 	r = amdgpu_vcn_sw_fini(adev);
243 
244 	kfree(adev->vcn.ip_dump);
245 
246 	return r;
247 }
248 
249 /**
250  * vcn_v4_0_3_hw_init - start and test VCN block
251  *
252  * @handle: amdgpu_device pointer
253  *
254  * Initialize the hardware, boot up the VCPU and do some testing
255  */
256 static int vcn_v4_0_3_hw_init(void *handle)
257 {
258 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
259 	struct amdgpu_ring *ring;
260 	int i, r, vcn_inst;
261 
262 	if (amdgpu_sriov_vf(adev)) {
263 		r = vcn_v4_0_3_start_sriov(adev);
264 		if (r)
265 			return r;
266 
267 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
268 			ring = &adev->vcn.inst[i].ring_enc[0];
269 			ring->wptr = 0;
270 			ring->wptr_old = 0;
271 			vcn_v4_0_3_unified_ring_set_wptr(ring);
272 			ring->sched.ready = true;
273 		}
274 	} else {
275 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
276 			vcn_inst = GET_INST(VCN, i);
277 			ring = &adev->vcn.inst[i].ring_enc[0];
278 
279 			if (ring->use_doorbell) {
280 				adev->nbio.funcs->vcn_doorbell_range(
281 					adev, ring->use_doorbell,
282 					(adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
283 						9 * vcn_inst,
284 					adev->vcn.inst[i].aid_id);
285 
286 				WREG32_SOC15(
287 					VCN, GET_INST(VCN, ring->me),
288 					regVCN_RB1_DB_CTRL,
289 					ring->doorbell_index
290 							<< VCN_RB1_DB_CTRL__OFFSET__SHIFT |
291 						VCN_RB1_DB_CTRL__EN_MASK);
292 
293 				/* Read DB_CTRL to flush the write DB_CTRL command. */
294 				RREG32_SOC15(
295 					VCN, GET_INST(VCN, ring->me),
296 					regVCN_RB1_DB_CTRL);
297 			}
298 
299 			r = amdgpu_ring_test_helper(ring);
300 			if (r)
301 				return r;
302 		}
303 	}
304 
305 	return r;
306 }
307 
308 /**
309  * vcn_v4_0_3_hw_fini - stop the hardware block
310  *
311  * @handle: amdgpu_device pointer
312  *
313  * Stop the VCN block, mark ring as not ready any more
314  */
315 static int vcn_v4_0_3_hw_fini(void *handle)
316 {
317 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
318 
319 	cancel_delayed_work_sync(&adev->vcn.idle_work);
320 
321 	if (adev->vcn.cur_state != AMD_PG_STATE_GATE)
322 		vcn_v4_0_3_set_powergating_state(adev, AMD_PG_STATE_GATE);
323 
324 	return 0;
325 }
326 
327 /**
328  * vcn_v4_0_3_suspend - suspend VCN block
329  *
330  * @handle: amdgpu_device pointer
331  *
332  * HW fini and suspend VCN block
333  */
334 static int vcn_v4_0_3_suspend(void *handle)
335 {
336 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
337 	int r;
338 
339 	r = vcn_v4_0_3_hw_fini(adev);
340 	if (r)
341 		return r;
342 
343 	r = amdgpu_vcn_suspend(adev);
344 
345 	return r;
346 }
347 
348 /**
349  * vcn_v4_0_3_resume - resume VCN block
350  *
351  * @handle: amdgpu_device pointer
352  *
353  * Resume firmware and hw init VCN block
354  */
355 static int vcn_v4_0_3_resume(void *handle)
356 {
357 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
358 	int r;
359 
360 	r = amdgpu_vcn_resume(adev);
361 	if (r)
362 		return r;
363 
364 	r = vcn_v4_0_3_hw_init(adev);
365 
366 	return r;
367 }
368 
369 /**
370  * vcn_v4_0_3_mc_resume - memory controller programming
371  *
372  * @adev: amdgpu_device pointer
373  * @inst_idx: instance number
374  *
375  * Let the VCN memory controller know it's offsets
376  */
377 static void vcn_v4_0_3_mc_resume(struct amdgpu_device *adev, int inst_idx)
378 {
379 	uint32_t offset, size, vcn_inst;
380 	const struct common_firmware_header *hdr;
381 
382 	hdr = (const struct common_firmware_header *)adev->vcn.fw[inst_idx]->data;
383 	size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
384 
385 	vcn_inst = GET_INST(VCN, inst_idx);
386 	/* cache window 0: fw */
387 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
388 		WREG32_SOC15(
389 			VCN, vcn_inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
390 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx]
391 				 .tmr_mc_addr_lo));
392 		WREG32_SOC15(
393 			VCN, vcn_inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
394 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx]
395 				 .tmr_mc_addr_hi));
396 		WREG32_SOC15(VCN, vcn_inst, regUVD_VCPU_CACHE_OFFSET0, 0);
397 		offset = 0;
398 	} else {
399 		WREG32_SOC15(VCN, vcn_inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
400 			     lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr));
401 		WREG32_SOC15(VCN, vcn_inst,
402 			     regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
403 			     upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr));
404 		offset = size;
405 		WREG32_SOC15(VCN, vcn_inst, regUVD_VCPU_CACHE_OFFSET0,
406 			     AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
407 	}
408 	WREG32_SOC15(VCN, vcn_inst, regUVD_VCPU_CACHE_SIZE0, size);
409 
410 	/* cache window 1: stack */
411 	WREG32_SOC15(VCN, vcn_inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
412 		     lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset));
413 	WREG32_SOC15(VCN, vcn_inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
414 		     upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset));
415 	WREG32_SOC15(VCN, vcn_inst, regUVD_VCPU_CACHE_OFFSET1, 0);
416 	WREG32_SOC15(VCN, vcn_inst, regUVD_VCPU_CACHE_SIZE1,
417 		     AMDGPU_VCN_STACK_SIZE);
418 
419 	/* cache window 2: context */
420 	WREG32_SOC15(VCN, vcn_inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
421 		     lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset +
422 				   AMDGPU_VCN_STACK_SIZE));
423 	WREG32_SOC15(VCN, vcn_inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
424 		     upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset +
425 				   AMDGPU_VCN_STACK_SIZE));
426 	WREG32_SOC15(VCN, vcn_inst, regUVD_VCPU_CACHE_OFFSET2, 0);
427 	WREG32_SOC15(VCN, vcn_inst, regUVD_VCPU_CACHE_SIZE2,
428 		     AMDGPU_VCN_CONTEXT_SIZE);
429 
430 	/* non-cache window */
431 	WREG32_SOC15(
432 		VCN, vcn_inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
433 		lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr));
434 	WREG32_SOC15(
435 		VCN, vcn_inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
436 		upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr));
437 	WREG32_SOC15(VCN, vcn_inst, regUVD_VCPU_NONCACHE_OFFSET0, 0);
438 	WREG32_SOC15(
439 		VCN, vcn_inst, regUVD_VCPU_NONCACHE_SIZE0,
440 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
441 }
442 
443 /**
444  * vcn_v4_0_3_mc_resume_dpg_mode - memory controller programming for dpg mode
445  *
446  * @adev: amdgpu_device pointer
447  * @inst_idx: instance number index
448  * @indirect: indirectly write sram
449  *
450  * Let the VCN memory controller know it's offsets with dpg mode
451  */
452 static void vcn_v4_0_3_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
453 {
454 	uint32_t offset, size;
455 	const struct common_firmware_header *hdr;
456 
457 	hdr = (const struct common_firmware_header *)adev->vcn.fw[inst_idx]->data;
458 	size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
459 
460 	/* cache window 0: fw */
461 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
462 		if (!indirect) {
463 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
464 				VCN, 0, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
465 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN +
466 					inst_idx].tmr_mc_addr_lo), 0, indirect);
467 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
468 				VCN, 0, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
469 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN +
470 					inst_idx].tmr_mc_addr_hi), 0, indirect);
471 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
472 				VCN, 0, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
473 		} else {
474 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
475 				VCN, 0, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
476 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
477 				VCN, 0, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
478 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
479 				VCN, 0, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
480 		}
481 		offset = 0;
482 	} else {
483 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
484 			VCN, 0, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
485 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
486 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
487 			VCN, 0, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
488 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
489 		offset = size;
490 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
491 			VCN, 0, regUVD_VCPU_CACHE_OFFSET0),
492 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
493 	}
494 
495 	if (!indirect)
496 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
497 			VCN, 0, regUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
498 	else
499 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
500 			VCN, 0, regUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
501 
502 	/* cache window 1: stack */
503 	if (!indirect) {
504 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
505 			VCN, 0, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
506 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
507 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
508 			VCN, 0, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
509 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
510 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
511 			VCN, 0, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
512 	} else {
513 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
514 			VCN, 0, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
515 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
516 			VCN, 0, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
517 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
518 			VCN, 0, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
519 	}
520 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
521 			VCN, 0, regUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
522 
523 	/* cache window 2: context */
524 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
525 			VCN, 0, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
526 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset +
527 				AMDGPU_VCN_STACK_SIZE), 0, indirect);
528 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
529 			VCN, 0, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
530 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset +
531 				AMDGPU_VCN_STACK_SIZE), 0, indirect);
532 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
533 			VCN, 0, regUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
534 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
535 			VCN, 0, regUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
536 
537 	/* non-cache window */
538 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
539 			VCN, 0, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
540 			lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
541 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
542 			VCN, 0, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
543 			upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
544 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
545 			VCN, 0, regUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
546 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
547 			VCN, 0, regUVD_VCPU_NONCACHE_SIZE0),
548 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)), 0, indirect);
549 
550 	/* VCN global tiling registers */
551 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
552 		VCN, 0, regUVD_GFX8_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
553 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
554 		VCN, 0, regUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
555 }
556 
557 /**
558  * vcn_v4_0_3_disable_clock_gating - disable VCN clock gating
559  *
560  * @adev: amdgpu_device pointer
561  * @inst_idx: instance number
562  *
563  * Disable clock gating for VCN block
564  */
565 static void vcn_v4_0_3_disable_clock_gating(struct amdgpu_device *adev, int inst_idx)
566 {
567 	uint32_t data;
568 	int vcn_inst;
569 
570 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
571 		return;
572 
573 	vcn_inst = GET_INST(VCN, inst_idx);
574 
575 	/* VCN disable CGC */
576 	data = RREG32_SOC15(VCN, vcn_inst, regUVD_CGC_CTRL);
577 	data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
578 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
579 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
580 	WREG32_SOC15(VCN, vcn_inst, regUVD_CGC_CTRL, data);
581 
582 	data = RREG32_SOC15(VCN, vcn_inst, regUVD_CGC_GATE);
583 	data &= ~(UVD_CGC_GATE__SYS_MASK
584 		| UVD_CGC_GATE__MPEG2_MASK
585 		| UVD_CGC_GATE__REGS_MASK
586 		| UVD_CGC_GATE__RBC_MASK
587 		| UVD_CGC_GATE__LMI_MC_MASK
588 		| UVD_CGC_GATE__LMI_UMC_MASK
589 		| UVD_CGC_GATE__MPC_MASK
590 		| UVD_CGC_GATE__LBSI_MASK
591 		| UVD_CGC_GATE__LRBBM_MASK
592 		| UVD_CGC_GATE__WCB_MASK
593 		| UVD_CGC_GATE__VCPU_MASK
594 		| UVD_CGC_GATE__MMSCH_MASK);
595 
596 	WREG32_SOC15(VCN, vcn_inst, regUVD_CGC_GATE, data);
597 	SOC15_WAIT_ON_RREG(VCN, vcn_inst, regUVD_CGC_GATE, 0, 0xFFFFFFFF);
598 
599 	data = RREG32_SOC15(VCN, vcn_inst, regUVD_CGC_CTRL);
600 	data &= ~(UVD_CGC_CTRL__SYS_MODE_MASK
601 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
602 		| UVD_CGC_CTRL__REGS_MODE_MASK
603 		| UVD_CGC_CTRL__RBC_MODE_MASK
604 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
605 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
606 		| UVD_CGC_CTRL__MPC_MODE_MASK
607 		| UVD_CGC_CTRL__LBSI_MODE_MASK
608 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
609 		| UVD_CGC_CTRL__WCB_MODE_MASK
610 		| UVD_CGC_CTRL__VCPU_MODE_MASK
611 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
612 	WREG32_SOC15(VCN, vcn_inst, regUVD_CGC_CTRL, data);
613 
614 	data = RREG32_SOC15(VCN, vcn_inst, regUVD_SUVD_CGC_GATE);
615 	data |= (UVD_SUVD_CGC_GATE__SRE_MASK
616 		| UVD_SUVD_CGC_GATE__SIT_MASK
617 		| UVD_SUVD_CGC_GATE__SMP_MASK
618 		| UVD_SUVD_CGC_GATE__SCM_MASK
619 		| UVD_SUVD_CGC_GATE__SDB_MASK
620 		| UVD_SUVD_CGC_GATE__SRE_H264_MASK
621 		| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
622 		| UVD_SUVD_CGC_GATE__SIT_H264_MASK
623 		| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
624 		| UVD_SUVD_CGC_GATE__SCM_H264_MASK
625 		| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
626 		| UVD_SUVD_CGC_GATE__SDB_H264_MASK
627 		| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
628 		| UVD_SUVD_CGC_GATE__ENT_MASK
629 		| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
630 		| UVD_SUVD_CGC_GATE__SITE_MASK
631 		| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
632 		| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
633 		| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
634 		| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
635 		| UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
636 	WREG32_SOC15(VCN, vcn_inst, regUVD_SUVD_CGC_GATE, data);
637 
638 	data = RREG32_SOC15(VCN, vcn_inst, regUVD_SUVD_CGC_CTRL);
639 	data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
640 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
641 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
642 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
643 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
644 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
645 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
646 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
647 	WREG32_SOC15(VCN, vcn_inst, regUVD_SUVD_CGC_CTRL, data);
648 }
649 
650 /**
651  * vcn_v4_0_3_disable_clock_gating_dpg_mode - disable VCN clock gating dpg mode
652  *
653  * @adev: amdgpu_device pointer
654  * @sram_sel: sram select
655  * @inst_idx: instance number index
656  * @indirect: indirectly write sram
657  *
658  * Disable clock gating for VCN block with dpg mode
659  */
660 static void vcn_v4_0_3_disable_clock_gating_dpg_mode(struct amdgpu_device *adev, uint8_t sram_sel,
661 				int inst_idx, uint8_t indirect)
662 {
663 	uint32_t reg_data = 0;
664 
665 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
666 		return;
667 
668 	/* enable sw clock gating control */
669 	reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
670 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
671 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
672 	reg_data &= ~(UVD_CGC_CTRL__SYS_MODE_MASK |
673 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
674 		 UVD_CGC_CTRL__REGS_MODE_MASK |
675 		 UVD_CGC_CTRL__RBC_MODE_MASK |
676 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
677 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
678 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
679 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
680 		 UVD_CGC_CTRL__MPC_MODE_MASK |
681 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
682 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
683 		 UVD_CGC_CTRL__WCB_MODE_MASK |
684 		 UVD_CGC_CTRL__VCPU_MODE_MASK);
685 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
686 		VCN, 0, regUVD_CGC_CTRL), reg_data, sram_sel, indirect);
687 
688 	/* turn off clock gating */
689 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
690 		VCN, 0, regUVD_CGC_GATE), 0, sram_sel, indirect);
691 
692 	/* turn on SUVD clock gating */
693 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
694 		VCN, 0, regUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
695 
696 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
697 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
698 		VCN, 0, regUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
699 }
700 
701 /**
702  * vcn_v4_0_3_enable_clock_gating - enable VCN clock gating
703  *
704  * @adev: amdgpu_device pointer
705  * @inst_idx: instance number
706  *
707  * Enable clock gating for VCN block
708  */
709 static void vcn_v4_0_3_enable_clock_gating(struct amdgpu_device *adev, int inst_idx)
710 {
711 	uint32_t data;
712 	int vcn_inst;
713 
714 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
715 		return;
716 
717 	vcn_inst = GET_INST(VCN, inst_idx);
718 
719 	/* enable VCN CGC */
720 	data = RREG32_SOC15(VCN, vcn_inst, regUVD_CGC_CTRL);
721 	data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
722 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
723 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
724 	WREG32_SOC15(VCN, vcn_inst, regUVD_CGC_CTRL, data);
725 
726 	data = RREG32_SOC15(VCN, vcn_inst, regUVD_CGC_CTRL);
727 	data |= (UVD_CGC_CTRL__SYS_MODE_MASK
728 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
729 		| UVD_CGC_CTRL__REGS_MODE_MASK
730 		| UVD_CGC_CTRL__RBC_MODE_MASK
731 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
732 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
733 		| UVD_CGC_CTRL__MPC_MODE_MASK
734 		| UVD_CGC_CTRL__LBSI_MODE_MASK
735 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
736 		| UVD_CGC_CTRL__WCB_MODE_MASK
737 		| UVD_CGC_CTRL__VCPU_MODE_MASK);
738 	WREG32_SOC15(VCN, vcn_inst, regUVD_CGC_CTRL, data);
739 
740 	data = RREG32_SOC15(VCN, vcn_inst, regUVD_SUVD_CGC_CTRL);
741 	data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
742 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
743 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
744 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
745 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
746 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
747 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
748 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
749 	WREG32_SOC15(VCN, vcn_inst, regUVD_SUVD_CGC_CTRL, data);
750 }
751 
752 /**
753  * vcn_v4_0_3_start_dpg_mode - VCN start with dpg mode
754  *
755  * @adev: amdgpu_device pointer
756  * @inst_idx: instance number index
757  * @indirect: indirectly write sram
758  *
759  * Start VCN block with dpg mode
760  */
761 static int vcn_v4_0_3_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
762 {
763 	volatile struct amdgpu_vcn4_fw_shared *fw_shared =
764 						adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
765 	struct amdgpu_ring *ring;
766 	int vcn_inst;
767 	uint32_t tmp;
768 
769 	vcn_inst = GET_INST(VCN, inst_idx);
770 	/* disable register anti-hang mechanism */
771 	WREG32_P(SOC15_REG_OFFSET(VCN, vcn_inst, regUVD_POWER_STATUS), 1,
772 		 ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
773 	/* enable dynamic power gating mode */
774 	tmp = RREG32_SOC15(VCN, vcn_inst, regUVD_POWER_STATUS);
775 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
776 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
777 	WREG32_SOC15(VCN, vcn_inst, regUVD_POWER_STATUS, tmp);
778 
779 	if (indirect) {
780 		DRM_DEV_DEBUG(adev->dev, "VCN %d start: on AID %d",
781 			inst_idx, adev->vcn.inst[inst_idx].aid_id);
782 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr =
783 				(uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
784 		/* Use dummy register 0xDEADBEEF passing AID selection to PSP FW */
785 		WREG32_SOC15_DPG_MODE(inst_idx, 0xDEADBEEF,
786 			adev->vcn.inst[inst_idx].aid_id, 0, true);
787 	}
788 
789 	/* enable clock gating */
790 	vcn_v4_0_3_disable_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
791 
792 	/* enable VCPU clock */
793 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
794 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
795 	tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
796 
797 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
798 		VCN, 0, regUVD_VCPU_CNTL), tmp, 0, indirect);
799 
800 	/* disable master interrupt */
801 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
802 		VCN, 0, regUVD_MASTINT_EN), 0, 0, indirect);
803 
804 	/* setup regUVD_LMI_CTRL */
805 	tmp = (UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
806 		UVD_LMI_CTRL__REQ_MODE_MASK |
807 		UVD_LMI_CTRL__CRC_RESET_MASK |
808 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
809 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
810 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
811 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
812 		0x00100000L);
813 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
814 		VCN, 0, regUVD_LMI_CTRL), tmp, 0, indirect);
815 
816 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
817 		VCN, 0, regUVD_MPC_CNTL),
818 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
819 
820 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
821 		VCN, 0, regUVD_MPC_SET_MUXA0),
822 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
823 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
824 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
825 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
826 
827 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
828 		VCN, 0, regUVD_MPC_SET_MUXB0),
829 		 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
830 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
831 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
832 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
833 
834 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
835 		VCN, 0, regUVD_MPC_SET_MUX),
836 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
837 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
838 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
839 
840 	vcn_v4_0_3_mc_resume_dpg_mode(adev, inst_idx, indirect);
841 
842 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
843 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
844 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
845 		VCN, 0, regUVD_VCPU_CNTL), tmp, 0, indirect);
846 
847 	/* enable LMI MC and UMC channels */
848 	tmp = 0x1f << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT;
849 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
850 		VCN, 0, regUVD_LMI_CTRL2), tmp, 0, indirect);
851 
852 	vcn_v4_0_3_enable_ras(adev, inst_idx, indirect);
853 
854 	/* enable master interrupt */
855 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
856 		VCN, 0, regUVD_MASTINT_EN),
857 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
858 
859 	if (indirect)
860 		amdgpu_vcn_psp_update_sram(adev, inst_idx, AMDGPU_UCODE_ID_VCN0_RAM);
861 
862 	ring = &adev->vcn.inst[inst_idx].ring_enc[0];
863 
864 	/* program the RB_BASE for ring buffer */
865 	WREG32_SOC15(VCN, vcn_inst, regUVD_RB_BASE_LO,
866 		     lower_32_bits(ring->gpu_addr));
867 	WREG32_SOC15(VCN, vcn_inst, regUVD_RB_BASE_HI,
868 		     upper_32_bits(ring->gpu_addr));
869 
870 	WREG32_SOC15(VCN, vcn_inst, regUVD_RB_SIZE,
871 		     ring->ring_size / sizeof(uint32_t));
872 
873 	/* resetting ring, fw should not check RB ring */
874 	tmp = RREG32_SOC15(VCN, vcn_inst, regVCN_RB_ENABLE);
875 	tmp &= ~(VCN_RB_ENABLE__RB_EN_MASK);
876 	WREG32_SOC15(VCN, vcn_inst, regVCN_RB_ENABLE, tmp);
877 	fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
878 
879 	/* Initialize the ring buffer's read and write pointers */
880 	WREG32_SOC15(VCN, vcn_inst, regUVD_RB_RPTR, 0);
881 	WREG32_SOC15(VCN, vcn_inst, regUVD_RB_WPTR, 0);
882 	ring->wptr = RREG32_SOC15(VCN, vcn_inst, regUVD_RB_WPTR);
883 
884 	tmp = RREG32_SOC15(VCN, vcn_inst, regVCN_RB_ENABLE);
885 	tmp |= VCN_RB_ENABLE__RB_EN_MASK;
886 	WREG32_SOC15(VCN, vcn_inst, regVCN_RB_ENABLE, tmp);
887 	fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
888 
889 	/*resetting done, fw can check RB ring */
890 	fw_shared->sq.queue_mode &= cpu_to_le32(~FW_QUEUE_RING_RESET);
891 
892 	return 0;
893 }
894 
895 static int vcn_v4_0_3_start_sriov(struct amdgpu_device *adev)
896 {
897 	int i, vcn_inst;
898 	struct amdgpu_ring *ring_enc;
899 	uint64_t cache_addr;
900 	uint64_t rb_enc_addr;
901 	uint64_t ctx_addr;
902 	uint32_t param, resp, expected;
903 	uint32_t offset, cache_size;
904 	uint32_t tmp, timeout;
905 
906 	struct amdgpu_mm_table *table = &adev->virt.mm_table;
907 	uint32_t *table_loc;
908 	uint32_t table_size;
909 	uint32_t size, size_dw;
910 	uint32_t init_status;
911 	uint32_t enabled_vcn;
912 
913 	struct mmsch_v4_0_cmd_direct_write
914 		direct_wt = { {0} };
915 	struct mmsch_v4_0_cmd_direct_read_modify_write
916 		direct_rd_mod_wt = { {0} };
917 	struct mmsch_v4_0_cmd_end end = { {0} };
918 	struct mmsch_v4_0_3_init_header header;
919 
920 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
921 	volatile struct amdgpu_fw_shared_rb_setup *rb_setup;
922 
923 	direct_wt.cmd_header.command_type =
924 		MMSCH_COMMAND__DIRECT_REG_WRITE;
925 	direct_rd_mod_wt.cmd_header.command_type =
926 		MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
927 	end.cmd_header.command_type = MMSCH_COMMAND__END;
928 
929 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
930 		vcn_inst = GET_INST(VCN, i);
931 
932 		memset(&header, 0, sizeof(struct mmsch_v4_0_3_init_header));
933 		header.version = MMSCH_VERSION;
934 		header.total_size = sizeof(struct mmsch_v4_0_3_init_header) >> 2;
935 
936 		table_loc = (uint32_t *)table->cpu_addr;
937 		table_loc += header.total_size;
938 
939 		table_size = 0;
940 
941 		MMSCH_V4_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(VCN, 0, regUVD_STATUS),
942 			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
943 
944 		cache_size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw[i]->size + 4);
945 
946 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
947 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
948 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
949 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
950 
951 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
952 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
953 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
954 
955 			offset = 0;
956 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
957 				regUVD_VCPU_CACHE_OFFSET0), 0);
958 		} else {
959 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
960 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
961 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
962 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
963 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
964 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
965 			offset = cache_size;
966 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
967 				regUVD_VCPU_CACHE_OFFSET0),
968 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
969 		}
970 
971 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
972 			regUVD_VCPU_CACHE_SIZE0),
973 			cache_size);
974 
975 		cache_addr = adev->vcn.inst[vcn_inst].gpu_addr + offset;
976 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
977 			regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), lower_32_bits(cache_addr));
978 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
979 			regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), upper_32_bits(cache_addr));
980 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
981 			regUVD_VCPU_CACHE_OFFSET1), 0);
982 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
983 			regUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE);
984 
985 		cache_addr = adev->vcn.inst[vcn_inst].gpu_addr + offset +
986 			AMDGPU_VCN_STACK_SIZE;
987 
988 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
989 			regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW), lower_32_bits(cache_addr));
990 
991 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
992 			regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH), upper_32_bits(cache_addr));
993 
994 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
995 			regUVD_VCPU_CACHE_OFFSET2), 0);
996 
997 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
998 			regUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE);
999 
1000 		fw_shared = adev->vcn.inst[vcn_inst].fw_shared.cpu_addr;
1001 		rb_setup = &fw_shared->rb_setup;
1002 
1003 		ring_enc = &adev->vcn.inst[vcn_inst].ring_enc[0];
1004 		ring_enc->wptr = 0;
1005 		rb_enc_addr = ring_enc->gpu_addr;
1006 
1007 		rb_setup->is_rb_enabled_flags |= RB_ENABLED;
1008 		rb_setup->rb_addr_lo = lower_32_bits(rb_enc_addr);
1009 		rb_setup->rb_addr_hi = upper_32_bits(rb_enc_addr);
1010 		rb_setup->rb_size = ring_enc->ring_size / 4;
1011 		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
1012 
1013 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
1014 			regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
1015 			lower_32_bits(adev->vcn.inst[vcn_inst].fw_shared.gpu_addr));
1016 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
1017 			regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
1018 			upper_32_bits(adev->vcn.inst[vcn_inst].fw_shared.gpu_addr));
1019 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, 0,
1020 			regUVD_VCPU_NONCACHE_SIZE0),
1021 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
1022 		MMSCH_V4_0_INSERT_END();
1023 
1024 		header.vcn0.init_status = 0;
1025 		header.vcn0.table_offset = header.total_size;
1026 		header.vcn0.table_size = table_size;
1027 		header.total_size += table_size;
1028 
1029 		/* Send init table to mmsch */
1030 		size = sizeof(struct mmsch_v4_0_3_init_header);
1031 		table_loc = (uint32_t *)table->cpu_addr;
1032 		memcpy((void *)table_loc, &header, size);
1033 
1034 		ctx_addr = table->gpu_addr;
1035 		WREG32_SOC15(VCN, vcn_inst, regMMSCH_VF_CTX_ADDR_LO, lower_32_bits(ctx_addr));
1036 		WREG32_SOC15(VCN, vcn_inst, regMMSCH_VF_CTX_ADDR_HI, upper_32_bits(ctx_addr));
1037 
1038 		tmp = RREG32_SOC15(VCN, vcn_inst, regMMSCH_VF_VMID);
1039 		tmp &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1040 		tmp |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1041 		WREG32_SOC15(VCN, vcn_inst, regMMSCH_VF_VMID, tmp);
1042 
1043 		size = header.total_size;
1044 		WREG32_SOC15(VCN, vcn_inst, regMMSCH_VF_CTX_SIZE, size);
1045 
1046 		WREG32_SOC15(VCN, vcn_inst, regMMSCH_VF_MAILBOX_RESP, 0);
1047 
1048 		param = 0x00000001;
1049 		WREG32_SOC15(VCN, vcn_inst, regMMSCH_VF_MAILBOX_HOST, param);
1050 		tmp = 0;
1051 		timeout = 1000;
1052 		resp = 0;
1053 		expected = MMSCH_VF_MAILBOX_RESP__OK;
1054 		while (resp != expected) {
1055 			resp = RREG32_SOC15(VCN, vcn_inst, regMMSCH_VF_MAILBOX_RESP);
1056 			if (resp != 0)
1057 				break;
1058 
1059 			udelay(10);
1060 			tmp = tmp + 10;
1061 			if (tmp >= timeout) {
1062 				DRM_ERROR("failed to init MMSCH. TIME-OUT after %d usec"\
1063 					" waiting for regMMSCH_VF_MAILBOX_RESP "\
1064 					"(expected=0x%08x, readback=0x%08x)\n",
1065 					tmp, expected, resp);
1066 				return -EBUSY;
1067 			}
1068 		}
1069 
1070 		enabled_vcn = amdgpu_vcn_is_disabled_vcn(adev, VCN_DECODE_RING, 0) ? 1 : 0;
1071 		init_status = ((struct mmsch_v4_0_3_init_header *)(table_loc))->vcn0.init_status;
1072 		if (resp != expected && resp != MMSCH_VF_MAILBOX_RESP__INCOMPLETE
1073 					&& init_status != MMSCH_VF_ENGINE_STATUS__PASS) {
1074 			DRM_ERROR("MMSCH init status is incorrect! readback=0x%08x, header init "\
1075 				"status for VCN%x: 0x%x\n", resp, enabled_vcn, init_status);
1076 		}
1077 	}
1078 
1079 	return 0;
1080 }
1081 
1082 /**
1083  * vcn_v4_0_3_start - VCN start
1084  *
1085  * @adev: amdgpu_device pointer
1086  *
1087  * Start VCN block
1088  */
1089 static int vcn_v4_0_3_start(struct amdgpu_device *adev)
1090 {
1091 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1092 	struct amdgpu_ring *ring;
1093 	int i, j, k, r, vcn_inst;
1094 	uint32_t tmp;
1095 
1096 	if (adev->pm.dpm_enabled)
1097 		amdgpu_dpm_enable_uvd(adev, true);
1098 
1099 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1100 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1101 			r = vcn_v4_0_3_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
1102 			continue;
1103 		}
1104 
1105 		vcn_inst = GET_INST(VCN, i);
1106 		/* set VCN status busy */
1107 		tmp = RREG32_SOC15(VCN, vcn_inst, regUVD_STATUS) |
1108 		      UVD_STATUS__UVD_BUSY;
1109 		WREG32_SOC15(VCN, vcn_inst, regUVD_STATUS, tmp);
1110 
1111 		/*SW clock gating */
1112 		vcn_v4_0_3_disable_clock_gating(adev, i);
1113 
1114 		/* enable VCPU clock */
1115 		WREG32_P(SOC15_REG_OFFSET(VCN, vcn_inst, regUVD_VCPU_CNTL),
1116 			 UVD_VCPU_CNTL__CLK_EN_MASK,
1117 			 ~UVD_VCPU_CNTL__CLK_EN_MASK);
1118 
1119 		/* disable master interrupt */
1120 		WREG32_P(SOC15_REG_OFFSET(VCN, vcn_inst, regUVD_MASTINT_EN), 0,
1121 			 ~UVD_MASTINT_EN__VCPU_EN_MASK);
1122 
1123 		/* enable LMI MC and UMC channels */
1124 		WREG32_P(SOC15_REG_OFFSET(VCN, vcn_inst, regUVD_LMI_CTRL2), 0,
1125 			 ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1126 
1127 		tmp = RREG32_SOC15(VCN, vcn_inst, regUVD_SOFT_RESET);
1128 		tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1129 		tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1130 		WREG32_SOC15(VCN, vcn_inst, regUVD_SOFT_RESET, tmp);
1131 
1132 		/* setup regUVD_LMI_CTRL */
1133 		tmp = RREG32_SOC15(VCN, vcn_inst, regUVD_LMI_CTRL);
1134 		WREG32_SOC15(VCN, vcn_inst, regUVD_LMI_CTRL,
1135 			     tmp | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
1136 				     UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1137 				     UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1138 				     UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1139 
1140 		/* setup regUVD_MPC_CNTL */
1141 		tmp = RREG32_SOC15(VCN, vcn_inst, regUVD_MPC_CNTL);
1142 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1143 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1144 		WREG32_SOC15(VCN, vcn_inst, regUVD_MPC_CNTL, tmp);
1145 
1146 		/* setup UVD_MPC_SET_MUXA0 */
1147 		WREG32_SOC15(VCN, vcn_inst, regUVD_MPC_SET_MUXA0,
1148 			     ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1149 			      (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1150 			      (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1151 			      (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1152 
1153 		/* setup UVD_MPC_SET_MUXB0 */
1154 		WREG32_SOC15(VCN, vcn_inst, regUVD_MPC_SET_MUXB0,
1155 			     ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1156 			      (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1157 			      (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1158 			      (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1159 
1160 		/* setup UVD_MPC_SET_MUX */
1161 		WREG32_SOC15(VCN, vcn_inst, regUVD_MPC_SET_MUX,
1162 			     ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1163 			      (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1164 			      (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1165 
1166 		vcn_v4_0_3_mc_resume(adev, i);
1167 
1168 		/* VCN global tiling registers */
1169 		WREG32_SOC15(VCN, vcn_inst, regUVD_GFX8_ADDR_CONFIG,
1170 			     adev->gfx.config.gb_addr_config);
1171 		WREG32_SOC15(VCN, vcn_inst, regUVD_GFX10_ADDR_CONFIG,
1172 			     adev->gfx.config.gb_addr_config);
1173 
1174 		/* unblock VCPU register access */
1175 		WREG32_P(SOC15_REG_OFFSET(VCN, vcn_inst, regUVD_RB_ARB_CTRL), 0,
1176 			 ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1177 
1178 		/* release VCPU reset to boot */
1179 		WREG32_P(SOC15_REG_OFFSET(VCN, vcn_inst, regUVD_VCPU_CNTL), 0,
1180 			 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1181 
1182 		for (j = 0; j < 10; ++j) {
1183 			uint32_t status;
1184 
1185 			for (k = 0; k < 100; ++k) {
1186 				status = RREG32_SOC15(VCN, vcn_inst,
1187 						      regUVD_STATUS);
1188 				if (status & 2)
1189 					break;
1190 				mdelay(10);
1191 			}
1192 			r = 0;
1193 			if (status & 2)
1194 				break;
1195 
1196 			DRM_DEV_ERROR(adev->dev,
1197 				"VCN decode not responding, trying to reset the VCPU!!!\n");
1198 			WREG32_P(SOC15_REG_OFFSET(VCN, vcn_inst,
1199 						  regUVD_VCPU_CNTL),
1200 				 UVD_VCPU_CNTL__BLK_RST_MASK,
1201 				 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1202 			mdelay(10);
1203 			WREG32_P(SOC15_REG_OFFSET(VCN, vcn_inst,
1204 						  regUVD_VCPU_CNTL),
1205 				 0, ~UVD_VCPU_CNTL__BLK_RST_MASK);
1206 
1207 			mdelay(10);
1208 			r = -1;
1209 		}
1210 
1211 		if (r) {
1212 			DRM_DEV_ERROR(adev->dev, "VCN decode not responding, giving up!!!\n");
1213 			return r;
1214 		}
1215 
1216 		/* enable master interrupt */
1217 		WREG32_P(SOC15_REG_OFFSET(VCN, vcn_inst, regUVD_MASTINT_EN),
1218 			 UVD_MASTINT_EN__VCPU_EN_MASK,
1219 			 ~UVD_MASTINT_EN__VCPU_EN_MASK);
1220 
1221 		/* clear the busy bit of VCN_STATUS */
1222 		WREG32_P(SOC15_REG_OFFSET(VCN, vcn_inst, regUVD_STATUS), 0,
1223 			 ~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1224 
1225 		ring = &adev->vcn.inst[i].ring_enc[0];
1226 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1227 
1228 		/* program the RB_BASE for ring buffer */
1229 		WREG32_SOC15(VCN, vcn_inst, regUVD_RB_BASE_LO,
1230 			     lower_32_bits(ring->gpu_addr));
1231 		WREG32_SOC15(VCN, vcn_inst, regUVD_RB_BASE_HI,
1232 			     upper_32_bits(ring->gpu_addr));
1233 
1234 		WREG32_SOC15(VCN, vcn_inst, regUVD_RB_SIZE,
1235 			     ring->ring_size / sizeof(uint32_t));
1236 
1237 		/* resetting ring, fw should not check RB ring */
1238 		tmp = RREG32_SOC15(VCN, vcn_inst, regVCN_RB_ENABLE);
1239 		tmp &= ~(VCN_RB_ENABLE__RB_EN_MASK);
1240 		WREG32_SOC15(VCN, vcn_inst, regVCN_RB_ENABLE, tmp);
1241 
1242 		/* Initialize the ring buffer's read and write pointers */
1243 		WREG32_SOC15(VCN, vcn_inst, regUVD_RB_RPTR, 0);
1244 		WREG32_SOC15(VCN, vcn_inst, regUVD_RB_WPTR, 0);
1245 
1246 		tmp = RREG32_SOC15(VCN, vcn_inst, regVCN_RB_ENABLE);
1247 		tmp |= VCN_RB_ENABLE__RB_EN_MASK;
1248 		WREG32_SOC15(VCN, vcn_inst, regVCN_RB_ENABLE, tmp);
1249 
1250 		ring->wptr = RREG32_SOC15(VCN, vcn_inst, regUVD_RB_WPTR);
1251 		fw_shared->sq.queue_mode &=
1252 			cpu_to_le32(~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF));
1253 
1254 	}
1255 	return 0;
1256 }
1257 
1258 /**
1259  * vcn_v4_0_3_stop_dpg_mode - VCN stop with dpg mode
1260  *
1261  * @adev: amdgpu_device pointer
1262  * @inst_idx: instance number index
1263  *
1264  * Stop VCN block with dpg mode
1265  */
1266 static int vcn_v4_0_3_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1267 {
1268 	uint32_t tmp;
1269 	int vcn_inst;
1270 
1271 	vcn_inst = GET_INST(VCN, inst_idx);
1272 
1273 	/* Wait for power status to be 1 */
1274 	SOC15_WAIT_ON_RREG(VCN, vcn_inst, regUVD_POWER_STATUS, 1,
1275 			   UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1276 
1277 	/* wait for read ptr to be equal to write ptr */
1278 	tmp = RREG32_SOC15(VCN, vcn_inst, regUVD_RB_WPTR);
1279 	SOC15_WAIT_ON_RREG(VCN, vcn_inst, regUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1280 
1281 	SOC15_WAIT_ON_RREG(VCN, vcn_inst, regUVD_POWER_STATUS, 1,
1282 			   UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1283 
1284 	/* disable dynamic power gating mode */
1285 	WREG32_P(SOC15_REG_OFFSET(VCN, vcn_inst, regUVD_POWER_STATUS), 0,
1286 		 ~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1287 	return 0;
1288 }
1289 
1290 /**
1291  * vcn_v4_0_3_stop - VCN stop
1292  *
1293  * @adev: amdgpu_device pointer
1294  *
1295  * Stop VCN block
1296  */
1297 static int vcn_v4_0_3_stop(struct amdgpu_device *adev)
1298 {
1299 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1300 	int i, r = 0, vcn_inst;
1301 	uint32_t tmp;
1302 
1303 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1304 		vcn_inst = GET_INST(VCN, i);
1305 
1306 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1307 		fw_shared->sq.queue_mode |= FW_QUEUE_DPG_HOLD_OFF;
1308 
1309 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1310 			vcn_v4_0_3_stop_dpg_mode(adev, i);
1311 			continue;
1312 		}
1313 
1314 		/* wait for vcn idle */
1315 		r = SOC15_WAIT_ON_RREG(VCN, vcn_inst, regUVD_STATUS,
1316 				       UVD_STATUS__IDLE, 0x7);
1317 		if (r)
1318 			goto Done;
1319 
1320 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1321 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1322 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1323 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1324 		r = SOC15_WAIT_ON_RREG(VCN, vcn_inst, regUVD_LMI_STATUS, tmp,
1325 				       tmp);
1326 		if (r)
1327 			goto Done;
1328 
1329 		/* stall UMC channel */
1330 		tmp = RREG32_SOC15(VCN, vcn_inst, regUVD_LMI_CTRL2);
1331 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1332 		WREG32_SOC15(VCN, vcn_inst, regUVD_LMI_CTRL2, tmp);
1333 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK |
1334 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1335 		r = SOC15_WAIT_ON_RREG(VCN, vcn_inst, regUVD_LMI_STATUS, tmp,
1336 				       tmp);
1337 		if (r)
1338 			goto Done;
1339 
1340 		/* Unblock VCPU Register access */
1341 		WREG32_P(SOC15_REG_OFFSET(VCN, vcn_inst, regUVD_RB_ARB_CTRL),
1342 			 UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1343 			 ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1344 
1345 		/* release VCPU reset to boot */
1346 		WREG32_P(SOC15_REG_OFFSET(VCN, vcn_inst, regUVD_VCPU_CNTL),
1347 			 UVD_VCPU_CNTL__BLK_RST_MASK,
1348 			 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1349 
1350 		/* disable VCPU clock */
1351 		WREG32_P(SOC15_REG_OFFSET(VCN, vcn_inst, regUVD_VCPU_CNTL), 0,
1352 			 ~(UVD_VCPU_CNTL__CLK_EN_MASK));
1353 
1354 		/* reset LMI UMC/LMI/VCPU */
1355 		tmp = RREG32_SOC15(VCN, vcn_inst, regUVD_SOFT_RESET);
1356 		tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1357 		WREG32_SOC15(VCN, vcn_inst, regUVD_SOFT_RESET, tmp);
1358 
1359 		tmp = RREG32_SOC15(VCN, vcn_inst, regUVD_SOFT_RESET);
1360 		tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1361 		WREG32_SOC15(VCN, vcn_inst, regUVD_SOFT_RESET, tmp);
1362 
1363 		/* clear VCN status */
1364 		WREG32_SOC15(VCN, vcn_inst, regUVD_STATUS, 0);
1365 
1366 		/* apply HW clock gating */
1367 		vcn_v4_0_3_enable_clock_gating(adev, i);
1368 	}
1369 Done:
1370 	if (adev->pm.dpm_enabled)
1371 		amdgpu_dpm_enable_uvd(adev, false);
1372 
1373 	return 0;
1374 }
1375 
1376 /**
1377  * vcn_v4_0_3_pause_dpg_mode - VCN pause with dpg mode
1378  *
1379  * @adev: amdgpu_device pointer
1380  * @inst_idx: instance number index
1381  * @new_state: pause state
1382  *
1383  * Pause dpg mode for VCN block
1384  */
1385 static int vcn_v4_0_3_pause_dpg_mode(struct amdgpu_device *adev, int inst_idx,
1386 				struct dpg_pause_state *new_state)
1387 {
1388 
1389 	return 0;
1390 }
1391 
1392 /**
1393  * vcn_v4_0_3_unified_ring_get_rptr - get unified read pointer
1394  *
1395  * @ring: amdgpu_ring pointer
1396  *
1397  * Returns the current hardware unified read pointer
1398  */
1399 static uint64_t vcn_v4_0_3_unified_ring_get_rptr(struct amdgpu_ring *ring)
1400 {
1401 	struct amdgpu_device *adev = ring->adev;
1402 
1403 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1404 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1405 
1406 	return RREG32_SOC15(VCN, GET_INST(VCN, ring->me), regUVD_RB_RPTR);
1407 }
1408 
1409 /**
1410  * vcn_v4_0_3_unified_ring_get_wptr - get unified write pointer
1411  *
1412  * @ring: amdgpu_ring pointer
1413  *
1414  * Returns the current hardware unified write pointer
1415  */
1416 static uint64_t vcn_v4_0_3_unified_ring_get_wptr(struct amdgpu_ring *ring)
1417 {
1418 	struct amdgpu_device *adev = ring->adev;
1419 
1420 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1421 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1422 
1423 	if (ring->use_doorbell)
1424 		return *ring->wptr_cpu_addr;
1425 	else
1426 		return RREG32_SOC15(VCN, GET_INST(VCN, ring->me),
1427 				    regUVD_RB_WPTR);
1428 }
1429 
1430 static void vcn_v4_0_3_enc_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
1431 				uint32_t val, uint32_t mask)
1432 {
1433 	/* For VF, only local offsets should be used */
1434 	if (amdgpu_sriov_vf(ring->adev))
1435 		reg = NORMALIZE_VCN_REG_OFFSET(reg);
1436 
1437 	amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WAIT);
1438 	amdgpu_ring_write(ring, reg << 2);
1439 	amdgpu_ring_write(ring, mask);
1440 	amdgpu_ring_write(ring, val);
1441 }
1442 
1443 static void vcn_v4_0_3_enc_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg, uint32_t val)
1444 {
1445 	/* For VF, only local offsets should be used */
1446 	if (amdgpu_sriov_vf(ring->adev))
1447 		reg = NORMALIZE_VCN_REG_OFFSET(reg);
1448 
1449 	amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WRITE);
1450 	amdgpu_ring_write(ring,	reg << 2);
1451 	amdgpu_ring_write(ring, val);
1452 }
1453 
1454 static void vcn_v4_0_3_enc_ring_emit_vm_flush(struct amdgpu_ring *ring,
1455 				unsigned int vmid, uint64_t pd_addr)
1456 {
1457 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->vm_hub];
1458 
1459 	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1460 
1461 	/* wait for reg writes */
1462 	vcn_v4_0_3_enc_ring_emit_reg_wait(ring, hub->ctx0_ptb_addr_lo32 +
1463 					vmid * hub->ctx_addr_distance,
1464 					lower_32_bits(pd_addr), 0xffffffff);
1465 }
1466 
1467 static void vcn_v4_0_3_ring_emit_hdp_flush(struct amdgpu_ring *ring)
1468 {
1469 	/* VCN engine access for HDP flush doesn't work when RRMT is enabled.
1470 	 * This is a workaround to avoid any HDP flush through VCN ring.
1471 	 */
1472 }
1473 
1474 /**
1475  * vcn_v4_0_3_unified_ring_set_wptr - set enc write pointer
1476  *
1477  * @ring: amdgpu_ring pointer
1478  *
1479  * Commits the enc write pointer to the hardware
1480  */
1481 static void vcn_v4_0_3_unified_ring_set_wptr(struct amdgpu_ring *ring)
1482 {
1483 	struct amdgpu_device *adev = ring->adev;
1484 
1485 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1486 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1487 
1488 	if (ring->use_doorbell) {
1489 		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1490 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1491 	} else {
1492 		WREG32_SOC15(VCN, GET_INST(VCN, ring->me), regUVD_RB_WPTR,
1493 			     lower_32_bits(ring->wptr));
1494 	}
1495 }
1496 
1497 static const struct amdgpu_ring_funcs vcn_v4_0_3_unified_ring_vm_funcs = {
1498 	.type = AMDGPU_RING_TYPE_VCN_ENC,
1499 	.align_mask = 0x3f,
1500 	.nop = VCN_ENC_CMD_NO_OP,
1501 	.get_rptr = vcn_v4_0_3_unified_ring_get_rptr,
1502 	.get_wptr = vcn_v4_0_3_unified_ring_get_wptr,
1503 	.set_wptr = vcn_v4_0_3_unified_ring_set_wptr,
1504 	.emit_frame_size =
1505 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1506 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1507 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1508 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1509 		1, /* vcn_v2_0_enc_ring_insert_end */
1510 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1511 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
1512 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
1513 	.emit_vm_flush = vcn_v4_0_3_enc_ring_emit_vm_flush,
1514 	.emit_hdp_flush = vcn_v4_0_3_ring_emit_hdp_flush,
1515 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
1516 	.test_ib = amdgpu_vcn_unified_ring_test_ib,
1517 	.insert_nop = amdgpu_ring_insert_nop,
1518 	.insert_end = vcn_v2_0_enc_ring_insert_end,
1519 	.pad_ib = amdgpu_ring_generic_pad_ib,
1520 	.begin_use = amdgpu_vcn_ring_begin_use,
1521 	.end_use = amdgpu_vcn_ring_end_use,
1522 	.emit_wreg = vcn_v4_0_3_enc_ring_emit_wreg,
1523 	.emit_reg_wait = vcn_v4_0_3_enc_ring_emit_reg_wait,
1524 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1525 };
1526 
1527 /**
1528  * vcn_v4_0_3_set_unified_ring_funcs - set unified ring functions
1529  *
1530  * @adev: amdgpu_device pointer
1531  *
1532  * Set unified ring functions
1533  */
1534 static void vcn_v4_0_3_set_unified_ring_funcs(struct amdgpu_device *adev)
1535 {
1536 	int i, vcn_inst;
1537 
1538 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1539 		adev->vcn.inst[i].ring_enc[0].funcs = &vcn_v4_0_3_unified_ring_vm_funcs;
1540 		adev->vcn.inst[i].ring_enc[0].me = i;
1541 		vcn_inst = GET_INST(VCN, i);
1542 		adev->vcn.inst[i].aid_id =
1543 			vcn_inst / adev->vcn.num_inst_per_aid;
1544 	}
1545 }
1546 
1547 /**
1548  * vcn_v4_0_3_is_idle - check VCN block is idle
1549  *
1550  * @handle: amdgpu_device pointer
1551  *
1552  * Check whether VCN block is idle
1553  */
1554 static bool vcn_v4_0_3_is_idle(void *handle)
1555 {
1556 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1557 	int i, ret = 1;
1558 
1559 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1560 		ret &= (RREG32_SOC15(VCN, GET_INST(VCN, i), regUVD_STATUS) ==
1561 			UVD_STATUS__IDLE);
1562 	}
1563 
1564 	return ret;
1565 }
1566 
1567 /**
1568  * vcn_v4_0_3_wait_for_idle - wait for VCN block idle
1569  *
1570  * @handle: amdgpu_device pointer
1571  *
1572  * Wait for VCN block idle
1573  */
1574 static int vcn_v4_0_3_wait_for_idle(void *handle)
1575 {
1576 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1577 	int i, ret = 0;
1578 
1579 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1580 		ret = SOC15_WAIT_ON_RREG(VCN, GET_INST(VCN, i), regUVD_STATUS,
1581 					 UVD_STATUS__IDLE, UVD_STATUS__IDLE);
1582 		if (ret)
1583 			return ret;
1584 	}
1585 
1586 	return ret;
1587 }
1588 
1589 /* vcn_v4_0_3_set_clockgating_state - set VCN block clockgating state
1590  *
1591  * @handle: amdgpu_device pointer
1592  * @state: clock gating state
1593  *
1594  * Set VCN block clockgating state
1595  */
1596 static int vcn_v4_0_3_set_clockgating_state(void *handle,
1597 					  enum amd_clockgating_state state)
1598 {
1599 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1600 	bool enable = state == AMD_CG_STATE_GATE;
1601 	int i;
1602 
1603 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1604 		if (enable) {
1605 			if (RREG32_SOC15(VCN, GET_INST(VCN, i),
1606 					 regUVD_STATUS) != UVD_STATUS__IDLE)
1607 				return -EBUSY;
1608 			vcn_v4_0_3_enable_clock_gating(adev, i);
1609 		} else {
1610 			vcn_v4_0_3_disable_clock_gating(adev, i);
1611 		}
1612 	}
1613 	return 0;
1614 }
1615 
1616 /**
1617  * vcn_v4_0_3_set_powergating_state - set VCN block powergating state
1618  *
1619  * @handle: amdgpu_device pointer
1620  * @state: power gating state
1621  *
1622  * Set VCN block powergating state
1623  */
1624 static int vcn_v4_0_3_set_powergating_state(void *handle,
1625 					  enum amd_powergating_state state)
1626 {
1627 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1628 	int ret;
1629 
1630 	/* for SRIOV, guest should not control VCN Power-gating
1631 	 * MMSCH FW should control Power-gating and clock-gating
1632 	 * guest should avoid touching CGC and PG
1633 	 */
1634 	if (amdgpu_sriov_vf(adev)) {
1635 		adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
1636 		return 0;
1637 	}
1638 
1639 	if (state == adev->vcn.cur_state)
1640 		return 0;
1641 
1642 	if (state == AMD_PG_STATE_GATE)
1643 		ret = vcn_v4_0_3_stop(adev);
1644 	else
1645 		ret = vcn_v4_0_3_start(adev);
1646 
1647 	if (!ret)
1648 		adev->vcn.cur_state = state;
1649 
1650 	return ret;
1651 }
1652 
1653 /**
1654  * vcn_v4_0_3_set_interrupt_state - set VCN block interrupt state
1655  *
1656  * @adev: amdgpu_device pointer
1657  * @source: interrupt sources
1658  * @type: interrupt types
1659  * @state: interrupt states
1660  *
1661  * Set VCN block interrupt state
1662  */
1663 static int vcn_v4_0_3_set_interrupt_state(struct amdgpu_device *adev,
1664 					struct amdgpu_irq_src *source,
1665 					unsigned int type,
1666 					enum amdgpu_interrupt_state state)
1667 {
1668 	return 0;
1669 }
1670 
1671 /**
1672  * vcn_v4_0_3_process_interrupt - process VCN block interrupt
1673  *
1674  * @adev: amdgpu_device pointer
1675  * @source: interrupt sources
1676  * @entry: interrupt entry from clients and sources
1677  *
1678  * Process VCN block interrupt
1679  */
1680 static int vcn_v4_0_3_process_interrupt(struct amdgpu_device *adev,
1681 				      struct amdgpu_irq_src *source,
1682 				      struct amdgpu_iv_entry *entry)
1683 {
1684 	uint32_t i, inst;
1685 
1686 	i = node_id_to_phys_map[entry->node_id];
1687 
1688 	DRM_DEV_DEBUG(adev->dev, "IH: VCN TRAP\n");
1689 
1690 	for (inst = 0; inst < adev->vcn.num_vcn_inst; ++inst)
1691 		if (adev->vcn.inst[inst].aid_id == i)
1692 			break;
1693 
1694 	if (inst >= adev->vcn.num_vcn_inst) {
1695 		dev_WARN_ONCE(adev->dev, 1,
1696 			      "Interrupt received for unknown VCN instance %d",
1697 			      entry->node_id);
1698 		return 0;
1699 	}
1700 
1701 	switch (entry->src_id) {
1702 	case VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
1703 		amdgpu_fence_process(&adev->vcn.inst[inst].ring_enc[0]);
1704 		break;
1705 	default:
1706 		DRM_DEV_ERROR(adev->dev, "Unhandled interrupt: %d %d\n",
1707 			  entry->src_id, entry->src_data[0]);
1708 		break;
1709 	}
1710 
1711 	return 0;
1712 }
1713 
1714 static const struct amdgpu_irq_src_funcs vcn_v4_0_3_irq_funcs = {
1715 	.set = vcn_v4_0_3_set_interrupt_state,
1716 	.process = vcn_v4_0_3_process_interrupt,
1717 };
1718 
1719 /**
1720  * vcn_v4_0_3_set_irq_funcs - set VCN block interrupt irq functions
1721  *
1722  * @adev: amdgpu_device pointer
1723  *
1724  * Set VCN block interrupt irq functions
1725  */
1726 static void vcn_v4_0_3_set_irq_funcs(struct amdgpu_device *adev)
1727 {
1728 	int i;
1729 
1730 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1731 		adev->vcn.inst->irq.num_types++;
1732 	}
1733 	adev->vcn.inst->irq.funcs = &vcn_v4_0_3_irq_funcs;
1734 }
1735 
1736 static void vcn_v4_0_3_print_ip_state(void *handle, struct drm_printer *p)
1737 {
1738 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1739 	int i, j;
1740 	uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_4_0_3);
1741 	uint32_t inst_off, is_powered;
1742 
1743 	if (!adev->vcn.ip_dump)
1744 		return;
1745 
1746 	drm_printf(p, "num_instances:%d\n", adev->vcn.num_vcn_inst);
1747 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1748 		if (adev->vcn.harvest_config & (1 << i)) {
1749 			drm_printf(p, "\nHarvested Instance:VCN%d Skipping dump\n", i);
1750 			continue;
1751 		}
1752 
1753 		inst_off = i * reg_count;
1754 		is_powered = (adev->vcn.ip_dump[inst_off] &
1755 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK) != 1;
1756 
1757 		if (is_powered) {
1758 			drm_printf(p, "\nActive Instance:VCN%d\n", i);
1759 			for (j = 0; j < reg_count; j++)
1760 				drm_printf(p, "%-50s \t 0x%08x\n", vcn_reg_list_4_0_3[j].reg_name,
1761 					   adev->vcn.ip_dump[inst_off + j]);
1762 		} else {
1763 			drm_printf(p, "\nInactive Instance:VCN%d\n", i);
1764 		}
1765 	}
1766 }
1767 
1768 static void vcn_v4_0_3_dump_ip_state(void *handle)
1769 {
1770 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1771 	int i, j;
1772 	bool is_powered;
1773 	uint32_t inst_off, inst_id;
1774 	uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_4_0_3);
1775 
1776 	if (!adev->vcn.ip_dump)
1777 		return;
1778 
1779 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1780 		if (adev->vcn.harvest_config & (1 << i))
1781 			continue;
1782 
1783 		inst_id = GET_INST(VCN, i);
1784 		inst_off = i * reg_count;
1785 		/* mmUVD_POWER_STATUS is always readable and is first element of the array */
1786 		adev->vcn.ip_dump[inst_off] = RREG32_SOC15(VCN, inst_id, regUVD_POWER_STATUS);
1787 		is_powered = (adev->vcn.ip_dump[inst_off] &
1788 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK) != 1;
1789 
1790 		if (is_powered)
1791 			for (j = 1; j < reg_count; j++)
1792 				adev->vcn.ip_dump[inst_off + j] =
1793 					RREG32(SOC15_REG_ENTRY_OFFSET_INST(vcn_reg_list_4_0_3[j],
1794 									   inst_id));
1795 	}
1796 }
1797 
1798 static const struct amd_ip_funcs vcn_v4_0_3_ip_funcs = {
1799 	.name = "vcn_v4_0_3",
1800 	.early_init = vcn_v4_0_3_early_init,
1801 	.late_init = NULL,
1802 	.sw_init = vcn_v4_0_3_sw_init,
1803 	.sw_fini = vcn_v4_0_3_sw_fini,
1804 	.hw_init = vcn_v4_0_3_hw_init,
1805 	.hw_fini = vcn_v4_0_3_hw_fini,
1806 	.suspend = vcn_v4_0_3_suspend,
1807 	.resume = vcn_v4_0_3_resume,
1808 	.is_idle = vcn_v4_0_3_is_idle,
1809 	.wait_for_idle = vcn_v4_0_3_wait_for_idle,
1810 	.check_soft_reset = NULL,
1811 	.pre_soft_reset = NULL,
1812 	.soft_reset = NULL,
1813 	.post_soft_reset = NULL,
1814 	.set_clockgating_state = vcn_v4_0_3_set_clockgating_state,
1815 	.set_powergating_state = vcn_v4_0_3_set_powergating_state,
1816 	.dump_ip_state = vcn_v4_0_3_dump_ip_state,
1817 	.print_ip_state = vcn_v4_0_3_print_ip_state,
1818 };
1819 
1820 const struct amdgpu_ip_block_version vcn_v4_0_3_ip_block = {
1821 	.type = AMD_IP_BLOCK_TYPE_VCN,
1822 	.major = 4,
1823 	.minor = 0,
1824 	.rev = 3,
1825 	.funcs = &vcn_v4_0_3_ip_funcs,
1826 };
1827 
1828 static const struct amdgpu_ras_err_status_reg_entry vcn_v4_0_3_ue_reg_list[] = {
1829 	{AMDGPU_RAS_REG_ENTRY(VCN, 0, regVCN_UE_ERR_STATUS_LO_VIDD, regVCN_UE_ERR_STATUS_HI_VIDD),
1830 	1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "VIDD"},
1831 	{AMDGPU_RAS_REG_ENTRY(VCN, 0, regVCN_UE_ERR_STATUS_LO_VIDV, regVCN_UE_ERR_STATUS_HI_VIDV),
1832 	1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "VIDV"},
1833 };
1834 
1835 static void vcn_v4_0_3_inst_query_ras_error_count(struct amdgpu_device *adev,
1836 						  uint32_t vcn_inst,
1837 						  void *ras_err_status)
1838 {
1839 	struct ras_err_data *err_data = (struct ras_err_data *)ras_err_status;
1840 
1841 	/* vcn v4_0_3 only support query uncorrectable errors */
1842 	amdgpu_ras_inst_query_ras_error_count(adev,
1843 			vcn_v4_0_3_ue_reg_list,
1844 			ARRAY_SIZE(vcn_v4_0_3_ue_reg_list),
1845 			NULL, 0, GET_INST(VCN, vcn_inst),
1846 			AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE,
1847 			&err_data->ue_count);
1848 }
1849 
1850 static void vcn_v4_0_3_query_ras_error_count(struct amdgpu_device *adev,
1851 					     void *ras_err_status)
1852 {
1853 	uint32_t i;
1854 
1855 	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN)) {
1856 		dev_warn(adev->dev, "VCN RAS is not supported\n");
1857 		return;
1858 	}
1859 
1860 	for (i = 0; i < adev->vcn.num_vcn_inst; i++)
1861 		vcn_v4_0_3_inst_query_ras_error_count(adev, i, ras_err_status);
1862 }
1863 
1864 static void vcn_v4_0_3_inst_reset_ras_error_count(struct amdgpu_device *adev,
1865 						  uint32_t vcn_inst)
1866 {
1867 	amdgpu_ras_inst_reset_ras_error_count(adev,
1868 					vcn_v4_0_3_ue_reg_list,
1869 					ARRAY_SIZE(vcn_v4_0_3_ue_reg_list),
1870 					GET_INST(VCN, vcn_inst));
1871 }
1872 
1873 static void vcn_v4_0_3_reset_ras_error_count(struct amdgpu_device *adev)
1874 {
1875 	uint32_t i;
1876 
1877 	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN)) {
1878 		dev_warn(adev->dev, "VCN RAS is not supported\n");
1879 		return;
1880 	}
1881 
1882 	for (i = 0; i < adev->vcn.num_vcn_inst; i++)
1883 		vcn_v4_0_3_inst_reset_ras_error_count(adev, i);
1884 }
1885 
1886 static const struct amdgpu_ras_block_hw_ops vcn_v4_0_3_ras_hw_ops = {
1887 	.query_ras_error_count = vcn_v4_0_3_query_ras_error_count,
1888 	.reset_ras_error_count = vcn_v4_0_3_reset_ras_error_count,
1889 };
1890 
1891 static struct amdgpu_vcn_ras vcn_v4_0_3_ras = {
1892 	.ras_block = {
1893 		.hw_ops = &vcn_v4_0_3_ras_hw_ops,
1894 	},
1895 };
1896 
1897 static void vcn_v4_0_3_set_ras_funcs(struct amdgpu_device *adev)
1898 {
1899 	adev->vcn.ras = &vcn_v4_0_3_ras;
1900 }
1901 
1902 static void vcn_v4_0_3_enable_ras(struct amdgpu_device *adev,
1903 				  int inst_idx, bool indirect)
1904 {
1905 	uint32_t tmp;
1906 
1907 	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN))
1908 		return;
1909 
1910 	tmp = VCN_RAS_CNTL__VCPU_VCODEC_REARM_MASK |
1911 	      VCN_RAS_CNTL__VCPU_VCODEC_IH_EN_MASK |
1912 	      VCN_RAS_CNTL__VCPU_VCODEC_PMI_EN_MASK |
1913 	      VCN_RAS_CNTL__VCPU_VCODEC_STALL_EN_MASK;
1914 	WREG32_SOC15_DPG_MODE(inst_idx,
1915 			      SOC15_DPG_MODE_OFFSET(VCN, 0, regVCN_RAS_CNTL),
1916 			      tmp, 0, indirect);
1917 
1918 	tmp = UVD_VCPU_INT_EN2__RASCNTL_VCPU_VCODEC_EN_MASK;
1919 	WREG32_SOC15_DPG_MODE(inst_idx,
1920 			      SOC15_DPG_MODE_OFFSET(VCN, 0, regUVD_VCPU_INT_EN2),
1921 			      tmp, 0, indirect);
1922 
1923 	tmp = UVD_SYS_INT_EN__RASCNTL_VCPU_VCODEC_EN_MASK;
1924 	WREG32_SOC15_DPG_MODE(inst_idx,
1925 			      SOC15_DPG_MODE_OFFSET(VCN, 0, regUVD_SYS_INT_EN),
1926 			      tmp, 0, indirect);
1927 }
1928