xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c (revision 7272b8bf)
1 /*
2  * Copyright 2021 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 "amdgpu_cs.h"
29 #include "soc15.h"
30 #include "soc15d.h"
31 #include "soc15_hw_ip.h"
32 #include "vcn_v2_0.h"
33 #include "mmsch_v4_0.h"
34 #include "vcn_v4_0.h"
35 
36 #include "vcn/vcn_4_0_0_offset.h"
37 #include "vcn/vcn_4_0_0_sh_mask.h"
38 #include "ivsrcid/vcn/irqsrcs_vcn_4_0.h"
39 
40 #include <drm/drm_drv.h>
41 
42 #define mmUVD_DPG_LMA_CTL							regUVD_DPG_LMA_CTL
43 #define mmUVD_DPG_LMA_CTL_BASE_IDX						regUVD_DPG_LMA_CTL_BASE_IDX
44 #define mmUVD_DPG_LMA_DATA							regUVD_DPG_LMA_DATA
45 #define mmUVD_DPG_LMA_DATA_BASE_IDX						regUVD_DPG_LMA_DATA_BASE_IDX
46 
47 #define VCN_VID_SOC_ADDRESS_2_0							0x1fb00
48 #define VCN1_VID_SOC_ADDRESS_3_0						0x48300
49 
50 #define VCN_HARVEST_MMSCH								0
51 
52 #define RDECODE_MSG_CREATE							0x00000000
53 #define RDECODE_MESSAGE_CREATE							0x00000001
54 
55 static int amdgpu_ih_clientid_vcns[] = {
56 	SOC15_IH_CLIENTID_VCN,
57 	SOC15_IH_CLIENTID_VCN1
58 };
59 
60 static int vcn_v4_0_start_sriov(struct amdgpu_device *adev);
61 static void vcn_v4_0_set_unified_ring_funcs(struct amdgpu_device *adev);
62 static void vcn_v4_0_set_irq_funcs(struct amdgpu_device *adev);
63 static int vcn_v4_0_set_powergating_state(void *handle,
64         enum amd_powergating_state state);
65 static int vcn_v4_0_pause_dpg_mode(struct amdgpu_device *adev,
66         int inst_idx, struct dpg_pause_state *new_state);
67 static void vcn_v4_0_unified_ring_set_wptr(struct amdgpu_ring *ring);
68 static void vcn_v4_0_set_ras_funcs(struct amdgpu_device *adev);
69 
70 /**
71  * vcn_v4_0_early_init - set function pointers and load microcode
72  *
73  * @handle: amdgpu_device pointer
74  *
75  * Set ring and irq function pointers
76  * Load microcode from filesystem
77  */
78 static int vcn_v4_0_early_init(void *handle)
79 {
80 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
81 	int i;
82 
83 	if (amdgpu_sriov_vf(adev)) {
84 		adev->vcn.harvest_config = VCN_HARVEST_MMSCH;
85 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
86 			if (amdgpu_vcn_is_disabled_vcn(adev, VCN_ENCODE_RING, i)) {
87 				adev->vcn.harvest_config |= 1 << i;
88 				dev_info(adev->dev, "VCN%d is disabled by hypervisor\n", i);
89 			}
90 		}
91 	}
92 
93 	/* re-use enc ring as unified ring */
94 	adev->vcn.num_enc_rings = 1;
95 
96 	vcn_v4_0_set_unified_ring_funcs(adev);
97 	vcn_v4_0_set_irq_funcs(adev);
98 	vcn_v4_0_set_ras_funcs(adev);
99 
100 	return amdgpu_vcn_early_init(adev);
101 }
102 
103 /**
104  * vcn_v4_0_sw_init - sw init for VCN block
105  *
106  * @handle: amdgpu_device pointer
107  *
108  * Load firmware and sw initialization
109  */
110 static int vcn_v4_0_sw_init(void *handle)
111 {
112 	struct amdgpu_ring *ring;
113 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
114 	int i, r;
115 
116 	r = amdgpu_vcn_sw_init(adev);
117 	if (r)
118 		return r;
119 
120 	amdgpu_vcn_setup_ucode(adev);
121 
122 	r = amdgpu_vcn_resume(adev);
123 	if (r)
124 		return r;
125 
126 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
127 		volatile struct amdgpu_vcn4_fw_shared *fw_shared;
128 
129 		if (adev->vcn.harvest_config & (1 << i))
130 			continue;
131 
132 		/* Init instance 0 sched_score to 1, so it's scheduled after other instances */
133 		if (i == 0)
134 			atomic_set(&adev->vcn.inst[i].sched_score, 1);
135 		else
136 			atomic_set(&adev->vcn.inst[i].sched_score, 0);
137 
138 		/* VCN UNIFIED TRAP */
139 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
140 				VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);
141 		if (r)
142 			return r;
143 
144 		/* VCN POISON TRAP */
145 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
146 				VCN_4_0__SRCID_UVD_POISON, &adev->vcn.inst[i].ras_poison_irq);
147 		if (r)
148 			return r;
149 
150 		ring = &adev->vcn.inst[i].ring_enc[0];
151 		ring->use_doorbell = true;
152 		if (amdgpu_sriov_vf(adev))
153 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + i * (adev->vcn.num_enc_rings + 1) + 1;
154 		else
155 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + 8 * i;
156 		ring->vm_hub = AMDGPU_MMHUB0(0);
157 		sprintf(ring->name, "vcn_unified_%d", i);
158 
159 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
160 						AMDGPU_RING_PRIO_0, &adev->vcn.inst[i].sched_score);
161 		if (r)
162 			return r;
163 
164 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
165 		fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_FW_SHARED_FLAG_0_UNIFIED_QUEUE);
166 		fw_shared->sq.is_enabled = 1;
167 
168 		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_SMU_DPM_INTERFACE_FLAG);
169 		fw_shared->smu_dpm_interface.smu_interface_type = (adev->flags & AMD_IS_APU) ?
170 			AMDGPU_VCN_SMU_DPM_INTERFACE_APU : AMDGPU_VCN_SMU_DPM_INTERFACE_DGPU;
171 
172 		if (adev->ip_versions[VCN_HWIP][0] == IP_VERSION(4, 0, 2)) {
173 			fw_shared->present_flag_0 |= AMDGPU_FW_SHARED_FLAG_0_DRM_KEY_INJECT;
174 			fw_shared->drm_key_wa.method =
175 				AMDGPU_DRM_KEY_INJECT_WORKAROUND_VCNFW_ASD_HANDSHAKING;
176 		}
177 
178 		if (amdgpu_sriov_vf(adev))
179 			fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
180 
181 		if (amdgpu_vcnfw_log)
182 			amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
183 	}
184 
185 	if (amdgpu_sriov_vf(adev)) {
186 		r = amdgpu_virt_alloc_mm_table(adev);
187 		if (r)
188 			return r;
189 	}
190 
191 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
192 		adev->vcn.pause_dpg_mode = vcn_v4_0_pause_dpg_mode;
193 
194 	r = amdgpu_vcn_ras_sw_init(adev);
195 	if (r)
196 		return r;
197 
198 	return 0;
199 }
200 
201 /**
202  * vcn_v4_0_sw_fini - sw fini for VCN block
203  *
204  * @handle: amdgpu_device pointer
205  *
206  * VCN suspend and free up sw allocation
207  */
208 static int vcn_v4_0_sw_fini(void *handle)
209 {
210 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
211 	int i, r, idx;
212 
213 	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
214 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
215 			volatile struct amdgpu_vcn4_fw_shared *fw_shared;
216 
217 			if (adev->vcn.harvest_config & (1 << i))
218 				continue;
219 
220 			fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
221 			fw_shared->present_flag_0 = 0;
222 			fw_shared->sq.is_enabled = 0;
223 		}
224 
225 		drm_dev_exit(idx);
226 	}
227 
228 	if (amdgpu_sriov_vf(adev))
229 		amdgpu_virt_free_mm_table(adev);
230 
231 	r = amdgpu_vcn_suspend(adev);
232 	if (r)
233 		return r;
234 
235 	r = amdgpu_vcn_sw_fini(adev);
236 
237 	return r;
238 }
239 
240 /**
241  * vcn_v4_0_hw_init - start and test VCN block
242  *
243  * @handle: amdgpu_device pointer
244  *
245  * Initialize the hardware, boot up the VCPU and do some testing
246  */
247 static int vcn_v4_0_hw_init(void *handle)
248 {
249 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
250 	struct amdgpu_ring *ring;
251 	int i, r;
252 
253 	if (amdgpu_sriov_vf(adev)) {
254 		r = vcn_v4_0_start_sriov(adev);
255 		if (r)
256 			goto done;
257 
258 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
259 			if (adev->vcn.harvest_config & (1 << i))
260 				continue;
261 
262 			ring = &adev->vcn.inst[i].ring_enc[0];
263 			ring->wptr = 0;
264 			ring->wptr_old = 0;
265 			vcn_v4_0_unified_ring_set_wptr(ring);
266 			ring->sched.ready = true;
267 
268 		}
269 	} else {
270 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
271 			if (adev->vcn.harvest_config & (1 << i))
272 				continue;
273 
274 			ring = &adev->vcn.inst[i].ring_enc[0];
275 
276 			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
277 					((adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i), i);
278 
279 			r = amdgpu_ring_test_helper(ring);
280 			if (r)
281 				goto done;
282 
283 		}
284 	}
285 
286 done:
287 	if (!r)
288 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
289 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
290 
291 	return r;
292 }
293 
294 /**
295  * vcn_v4_0_hw_fini - stop the hardware block
296  *
297  * @handle: amdgpu_device pointer
298  *
299  * Stop the VCN block, mark ring as not ready any more
300  */
301 static int vcn_v4_0_hw_fini(void *handle)
302 {
303 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
304 	int i;
305 
306 	cancel_delayed_work_sync(&adev->vcn.idle_work);
307 
308 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
309 		if (adev->vcn.harvest_config & (1 << i))
310 			continue;
311 		if (!amdgpu_sriov_vf(adev)) {
312 			if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
313                         (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
314                                 RREG32_SOC15(VCN, i, regUVD_STATUS))) {
315                         vcn_v4_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
316 			}
317 		}
318 		if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN))
319 			amdgpu_irq_put(adev, &adev->vcn.inst[i].ras_poison_irq, 0);
320 	}
321 
322 	return 0;
323 }
324 
325 /**
326  * vcn_v4_0_suspend - suspend VCN block
327  *
328  * @handle: amdgpu_device pointer
329  *
330  * HW fini and suspend VCN block
331  */
332 static int vcn_v4_0_suspend(void *handle)
333 {
334 	int r;
335 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
336 
337 	r = vcn_v4_0_hw_fini(adev);
338 	if (r)
339 		return r;
340 
341 	r = amdgpu_vcn_suspend(adev);
342 
343 	return r;
344 }
345 
346 /**
347  * vcn_v4_0_resume - resume VCN block
348  *
349  * @handle: amdgpu_device pointer
350  *
351  * Resume firmware and hw init VCN block
352  */
353 static int vcn_v4_0_resume(void *handle)
354 {
355 	int r;
356 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
357 
358 	r = amdgpu_vcn_resume(adev);
359 	if (r)
360 		return r;
361 
362 	r = vcn_v4_0_hw_init(adev);
363 
364 	return r;
365 }
366 
367 /**
368  * vcn_v4_0_mc_resume - memory controller programming
369  *
370  * @adev: amdgpu_device pointer
371  * @inst: instance number
372  *
373  * Let the VCN memory controller know it's offsets
374  */
375 static void vcn_v4_0_mc_resume(struct amdgpu_device *adev, int inst)
376 {
377 	uint32_t offset, size;
378 	const struct common_firmware_header *hdr;
379 
380 	hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
381 	size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
382 
383 	/* cache window 0: fw */
384 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
385 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
386 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo));
387 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
388 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi));
389 		WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, 0);
390 		offset = 0;
391 	} else {
392 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
393 			lower_32_bits(adev->vcn.inst[inst].gpu_addr));
394 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
395 			upper_32_bits(adev->vcn.inst[inst].gpu_addr));
396 		offset = size;
397                 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
398 	}
399 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE0, size);
400 
401 	/* cache window 1: stack */
402 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
403 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
404 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
405 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
406 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET1, 0);
407 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
408 
409 	/* cache window 2: context */
410 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
411 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
412 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
413 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
414 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET2, 0);
415 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
416 
417 	/* non-cache window */
418 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
419 		lower_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
420 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
421 		upper_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
422 	WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_OFFSET0, 0);
423 	WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_SIZE0,
424 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
425 }
426 
427 /**
428  * vcn_v4_0_mc_resume_dpg_mode - memory controller programming for dpg mode
429  *
430  * @adev: amdgpu_device pointer
431  * @inst_idx: instance number index
432  * @indirect: indirectly write sram
433  *
434  * Let the VCN memory controller know it's offsets with dpg mode
435  */
436 static void vcn_v4_0_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
437 {
438 	uint32_t offset, size;
439 	const struct common_firmware_header *hdr;
440 	hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
441 	size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
442 
443 	/* cache window 0: fw */
444 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
445 		if (!indirect) {
446 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
447 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
448 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
449 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
450 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
451 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
452 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
453 				VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
454 		} else {
455 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
456 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
457 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
458 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
459 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
460 				VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
461 		}
462 		offset = 0;
463 	} else {
464 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
465 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
466 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
467 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
468 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
469 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
470 		offset = size;
471 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
472 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0),
473 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
474 	}
475 
476 	if (!indirect)
477 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
478 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
479 	else
480 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
481 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
482 
483 	/* cache window 1: stack */
484 	if (!indirect) {
485 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
486 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
487 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
488 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
489 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
490 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
491 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
492 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
493 	} else {
494 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
495 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
496 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
497 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
498 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
499 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
500 	}
501 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
502 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
503 
504 	/* cache window 2: context */
505 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
506 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
507 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
508 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
509 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
510 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
511 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
512 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
513 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
514 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
515 
516 	/* non-cache window */
517 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
518 			VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
519 			lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
520 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
521 			VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
522 			upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
523 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
524 			VCN, inst_idx, regUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
525 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
526 			VCN, inst_idx, regUVD_VCPU_NONCACHE_SIZE0),
527 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)), 0, indirect);
528 
529 	/* VCN global tiling registers */
530 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
531 		VCN, 0, regUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
532 }
533 
534 /**
535  * vcn_v4_0_disable_static_power_gating - disable VCN static power gating
536  *
537  * @adev: amdgpu_device pointer
538  * @inst: instance number
539  *
540  * Disable static power gating for VCN block
541  */
542 static void vcn_v4_0_disable_static_power_gating(struct amdgpu_device *adev, int inst)
543 {
544 	uint32_t data = 0;
545 
546 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
547 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
548 			| 1 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
549 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
550 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
551 			| 2 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
552 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
553 			| 2 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
554 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
555 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
556 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
557 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
558 			| 2 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
559 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
560 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
561 
562 		WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
563 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS,
564 			UVD_PGFSM_STATUS__UVDM_UVDU_UVDLM_PWR_ON_3_0, 0x3F3FFFFF);
565 	} else {
566 		uint32_t value;
567 
568 		value = (inst) ? 0x2200800 : 0;
569 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
570 			| 1 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
571 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
572 			| 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
573 			| 1 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
574 			| 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
575 			| 1 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
576 			| 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
577 			| 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
578 			| 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
579 			| 1 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
580 			| 1 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
581 			| 1 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
582 			| 1 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
583 
584                 WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
585                 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS, value,  0x3F3FFFFF);
586         }
587 
588         data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
589         data &= ~0x103;
590         if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
591                 data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
592                         UVD_POWER_STATUS__UVD_PG_EN_MASK;
593 
594         WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
595 
596         return;
597 }
598 
599 /**
600  * vcn_v4_0_enable_static_power_gating - enable VCN static power gating
601  *
602  * @adev: amdgpu_device pointer
603  * @inst: instance number
604  *
605  * Enable static power gating for VCN block
606  */
607 static void vcn_v4_0_enable_static_power_gating(struct amdgpu_device *adev, int inst)
608 {
609 	uint32_t data;
610 
611 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
612 		/* Before power off, this indicator has to be turned on */
613 		data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
614 		data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
615 		data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
616 		WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
617 
618 		data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
619 			| 2 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
620 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
621 			| 2 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
622 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
623 			| 2 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
624 			| 2 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
625 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
626 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
627 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
628 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
629 			| 2 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
630 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
631 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
632 		WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
633 
634 		data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
635 			| 2 << UVD_PGFSM_STATUS__UVDS_PWR_STATUS__SHIFT
636 			| 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
637 			| 2 << UVD_PGFSM_STATUS__UVDTC_PWR_STATUS__SHIFT
638 			| 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
639 			| 2 << UVD_PGFSM_STATUS__UVDTA_PWR_STATUS__SHIFT
640 			| 2 << UVD_PGFSM_STATUS__UVDLM_PWR_STATUS__SHIFT
641 			| 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
642 			| 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
643 			| 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT
644 			| 2 << UVD_PGFSM_STATUS__UVDAB_PWR_STATUS__SHIFT
645 			| 2 << UVD_PGFSM_STATUS__UVDTB_PWR_STATUS__SHIFT
646 			| 2 << UVD_PGFSM_STATUS__UVDNA_PWR_STATUS__SHIFT
647 			| 2 << UVD_PGFSM_STATUS__UVDNB_PWR_STATUS__SHIFT);
648 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS, data, 0x3F3FFFFF);
649 	}
650 
651         return;
652 }
653 
654 /**
655  * vcn_v4_0_disable_clock_gating - disable VCN clock gating
656  *
657  * @adev: amdgpu_device pointer
658  * @inst: instance number
659  *
660  * Disable clock gating for VCN block
661  */
662 static void vcn_v4_0_disable_clock_gating(struct amdgpu_device *adev, int inst)
663 {
664 	uint32_t data;
665 
666 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
667 		return;
668 
669 	/* VCN disable CGC */
670 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
671 	data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
672 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
673 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
674 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
675 
676 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_GATE);
677 	data &= ~(UVD_CGC_GATE__SYS_MASK
678 		| UVD_CGC_GATE__UDEC_MASK
679 		| UVD_CGC_GATE__MPEG2_MASK
680 		| UVD_CGC_GATE__REGS_MASK
681 		| UVD_CGC_GATE__RBC_MASK
682 		| UVD_CGC_GATE__LMI_MC_MASK
683 		| UVD_CGC_GATE__LMI_UMC_MASK
684 		| UVD_CGC_GATE__IDCT_MASK
685 		| UVD_CGC_GATE__MPRD_MASK
686 		| UVD_CGC_GATE__MPC_MASK
687 		| UVD_CGC_GATE__LBSI_MASK
688 		| UVD_CGC_GATE__LRBBM_MASK
689 		| UVD_CGC_GATE__UDEC_RE_MASK
690 		| UVD_CGC_GATE__UDEC_CM_MASK
691 		| UVD_CGC_GATE__UDEC_IT_MASK
692 		| UVD_CGC_GATE__UDEC_DB_MASK
693 		| UVD_CGC_GATE__UDEC_MP_MASK
694 		| UVD_CGC_GATE__WCB_MASK
695 		| UVD_CGC_GATE__VCPU_MASK
696 		| UVD_CGC_GATE__MMSCH_MASK);
697 
698 	WREG32_SOC15(VCN, inst, regUVD_CGC_GATE, data);
699 	SOC15_WAIT_ON_RREG(VCN, inst, regUVD_CGC_GATE, 0,  0xFFFFFFFF);
700 
701 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
702 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
703 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
704 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
705 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
706 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
707 		| UVD_CGC_CTRL__SYS_MODE_MASK
708 		| UVD_CGC_CTRL__UDEC_MODE_MASK
709 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
710 		| UVD_CGC_CTRL__REGS_MODE_MASK
711 		| UVD_CGC_CTRL__RBC_MODE_MASK
712 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
713 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
714 		| UVD_CGC_CTRL__IDCT_MODE_MASK
715 		| UVD_CGC_CTRL__MPRD_MODE_MASK
716 		| UVD_CGC_CTRL__MPC_MODE_MASK
717 		| UVD_CGC_CTRL__LBSI_MODE_MASK
718 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
719 		| UVD_CGC_CTRL__WCB_MODE_MASK
720 		| UVD_CGC_CTRL__VCPU_MODE_MASK
721 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
722 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
723 
724 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE);
725 	data |= (UVD_SUVD_CGC_GATE__SRE_MASK
726 		| UVD_SUVD_CGC_GATE__SIT_MASK
727 		| UVD_SUVD_CGC_GATE__SMP_MASK
728 		| UVD_SUVD_CGC_GATE__SCM_MASK
729 		| UVD_SUVD_CGC_GATE__SDB_MASK
730 		| UVD_SUVD_CGC_GATE__SRE_H264_MASK
731 		| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
732 		| UVD_SUVD_CGC_GATE__SIT_H264_MASK
733 		| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
734 		| UVD_SUVD_CGC_GATE__SCM_H264_MASK
735 		| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
736 		| UVD_SUVD_CGC_GATE__SDB_H264_MASK
737 		| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
738 		| UVD_SUVD_CGC_GATE__SCLR_MASK
739 		| UVD_SUVD_CGC_GATE__UVD_SC_MASK
740 		| UVD_SUVD_CGC_GATE__ENT_MASK
741 		| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
742 		| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
743 		| UVD_SUVD_CGC_GATE__SITE_MASK
744 		| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
745 		| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
746 		| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
747 		| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
748 		| UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
749 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE, data);
750 
751 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
752 	data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
753 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
754 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
755 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
756 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
757 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
758 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
759 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
760 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
761 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
762 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
763 }
764 
765 /**
766  * vcn_v4_0_disable_clock_gating_dpg_mode - disable VCN clock gating dpg mode
767  *
768  * @adev: amdgpu_device pointer
769  * @sram_sel: sram select
770  * @inst_idx: instance number index
771  * @indirect: indirectly write sram
772  *
773  * Disable clock gating for VCN block with dpg mode
774  */
775 static void vcn_v4_0_disable_clock_gating_dpg_mode(struct amdgpu_device *adev, uint8_t sram_sel,
776       int inst_idx, uint8_t indirect)
777 {
778 	uint32_t reg_data = 0;
779 
780 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
781 		return;
782 
783 	/* enable sw clock gating control */
784 	reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
785 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
786 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
787 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
788 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
789 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
790 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
791 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
792 		 UVD_CGC_CTRL__SYS_MODE_MASK |
793 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
794 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
795 		 UVD_CGC_CTRL__REGS_MODE_MASK |
796 		 UVD_CGC_CTRL__RBC_MODE_MASK |
797 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
798 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
799 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
800 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
801 		 UVD_CGC_CTRL__MPC_MODE_MASK |
802 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
803 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
804 		 UVD_CGC_CTRL__WCB_MODE_MASK |
805 		 UVD_CGC_CTRL__VCPU_MODE_MASK);
806 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
807 		VCN, inst_idx, regUVD_CGC_CTRL), reg_data, sram_sel, indirect);
808 
809 	/* turn off clock gating */
810 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
811 		VCN, inst_idx, regUVD_CGC_GATE), 0, sram_sel, indirect);
812 
813 	/* turn on SUVD clock gating */
814 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
815 		VCN, inst_idx, regUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
816 
817 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
818 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
819 		VCN, inst_idx, regUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
820 }
821 
822 /**
823  * vcn_v4_0_enable_clock_gating - enable VCN clock gating
824  *
825  * @adev: amdgpu_device pointer
826  * @inst: instance number
827  *
828  * Enable clock gating for VCN block
829  */
830 static void vcn_v4_0_enable_clock_gating(struct amdgpu_device *adev, int inst)
831 {
832 	uint32_t data;
833 
834 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
835 		return;
836 
837 	/* enable VCN CGC */
838 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
839 	data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
840 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
841 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
842 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
843 
844 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
845 	data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
846 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
847 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
848 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
849 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
850 		| UVD_CGC_CTRL__SYS_MODE_MASK
851 		| UVD_CGC_CTRL__UDEC_MODE_MASK
852 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
853 		| UVD_CGC_CTRL__REGS_MODE_MASK
854 		| UVD_CGC_CTRL__RBC_MODE_MASK
855 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
856 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
857 		| UVD_CGC_CTRL__IDCT_MODE_MASK
858 		| UVD_CGC_CTRL__MPRD_MODE_MASK
859 		| UVD_CGC_CTRL__MPC_MODE_MASK
860 		| UVD_CGC_CTRL__LBSI_MODE_MASK
861 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
862 		| UVD_CGC_CTRL__WCB_MODE_MASK
863 		| UVD_CGC_CTRL__VCPU_MODE_MASK
864 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
865 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
866 
867 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
868 	data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
869 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
870 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
871 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
872 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
873 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
874 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
875 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
876 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
877 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
878 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
879 
880 	return;
881 }
882 
883 static void vcn_v4_0_enable_ras(struct amdgpu_device *adev, int inst_idx,
884 				bool indirect)
885 {
886 	uint32_t tmp;
887 
888 	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN))
889 		return;
890 
891 	tmp = VCN_RAS_CNTL__VCPU_VCODEC_REARM_MASK |
892 	      VCN_RAS_CNTL__VCPU_VCODEC_IH_EN_MASK |
893 	      VCN_RAS_CNTL__VCPU_VCODEC_PMI_EN_MASK |
894 	      VCN_RAS_CNTL__VCPU_VCODEC_STALL_EN_MASK;
895 	WREG32_SOC15_DPG_MODE(inst_idx,
896 			      SOC15_DPG_MODE_OFFSET(VCN, 0, regVCN_RAS_CNTL),
897 			      tmp, 0, indirect);
898 
899 	tmp = UVD_SYS_INT_EN__RASCNTL_VCPU_VCODEC_EN_MASK;
900 	WREG32_SOC15_DPG_MODE(inst_idx,
901 			      SOC15_DPG_MODE_OFFSET(VCN, 0, regUVD_SYS_INT_EN),
902 			      tmp, 0, indirect);
903 }
904 
905 /**
906  * vcn_v4_0_start_dpg_mode - VCN start with dpg mode
907  *
908  * @adev: amdgpu_device pointer
909  * @inst_idx: instance number index
910  * @indirect: indirectly write sram
911  *
912  * Start VCN block with dpg mode
913  */
914 static int vcn_v4_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
915 {
916 	volatile struct amdgpu_vcn4_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
917 	struct amdgpu_ring *ring;
918 	uint32_t tmp;
919 
920 	/* disable register anti-hang mechanism */
921 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 1,
922 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
923 	/* enable dynamic power gating mode */
924 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS);
925 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
926 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
927 	WREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS, tmp);
928 
929 	if (indirect)
930 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
931 
932 	/* enable clock gating */
933 	vcn_v4_0_disable_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
934 
935 	/* enable VCPU clock */
936 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
937 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK | UVD_VCPU_CNTL__BLK_RST_MASK;
938 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
939 		VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
940 
941 	/* disable master interupt */
942 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
943 		VCN, inst_idx, regUVD_MASTINT_EN), 0, 0, indirect);
944 
945 	/* setup regUVD_LMI_CTRL */
946 	tmp = (UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
947 		UVD_LMI_CTRL__REQ_MODE_MASK |
948 		UVD_LMI_CTRL__CRC_RESET_MASK |
949 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
950 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
951 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
952 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
953 		0x00100000L);
954 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
955 		VCN, inst_idx, regUVD_LMI_CTRL), tmp, 0, indirect);
956 
957 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
958 		VCN, inst_idx, regUVD_MPC_CNTL),
959 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
960 
961 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
962 		VCN, inst_idx, regUVD_MPC_SET_MUXA0),
963 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
964 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
965 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
966 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
967 
968 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
969 		VCN, inst_idx, regUVD_MPC_SET_MUXB0),
970 		 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
971 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
972 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
973 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
974 
975 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
976 		VCN, inst_idx, regUVD_MPC_SET_MUX),
977 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
978 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
979 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
980 
981 	vcn_v4_0_mc_resume_dpg_mode(adev, inst_idx, indirect);
982 
983 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
984 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
985 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
986 		VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
987 
988 	/* enable LMI MC and UMC channels */
989 	tmp = 0x1f << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT;
990 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
991 		VCN, inst_idx, regUVD_LMI_CTRL2), tmp, 0, indirect);
992 
993 	vcn_v4_0_enable_ras(adev, inst_idx, indirect);
994 
995 	/* enable master interrupt */
996 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
997 		VCN, inst_idx, regUVD_MASTINT_EN),
998 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
999 
1000 
1001 	if (indirect)
1002 		amdgpu_vcn_psp_update_sram(adev, inst_idx, 0);
1003 
1004 	ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1005 
1006 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_LO, ring->gpu_addr);
1007 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1008 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_SIZE, ring->ring_size / 4);
1009 
1010 	tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
1011 	tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
1012 	WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
1013 	fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
1014 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR, 0);
1015 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, 0);
1016 
1017 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR);
1018 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, tmp);
1019 	ring->wptr = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
1020 
1021 	tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
1022 	tmp |= VCN_RB_ENABLE__RB1_EN_MASK;
1023 	WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
1024 	fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
1025 
1026 	WREG32_SOC15(VCN, inst_idx, regVCN_RB1_DB_CTRL,
1027 			ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
1028 			VCN_RB1_DB_CTRL__EN_MASK);
1029 
1030 	return 0;
1031 }
1032 
1033 
1034 /**
1035  * vcn_v4_0_start - VCN start
1036  *
1037  * @adev: amdgpu_device pointer
1038  *
1039  * Start VCN block
1040  */
1041 static int vcn_v4_0_start(struct amdgpu_device *adev)
1042 {
1043 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1044 	struct amdgpu_ring *ring;
1045 	uint32_t tmp;
1046 	int i, j, k, r;
1047 
1048 	if (adev->pm.dpm_enabled)
1049 		amdgpu_dpm_enable_uvd(adev, true);
1050 
1051 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1052 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1053 
1054 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1055 			r = vcn_v4_0_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
1056 			continue;
1057 		}
1058 
1059 		/* disable VCN power gating */
1060 		vcn_v4_0_disable_static_power_gating(adev, i);
1061 
1062 		/* set VCN status busy */
1063 		tmp = RREG32_SOC15(VCN, i, regUVD_STATUS) | UVD_STATUS__UVD_BUSY;
1064 		WREG32_SOC15(VCN, i, regUVD_STATUS, tmp);
1065 
1066 		/*SW clock gating */
1067 		vcn_v4_0_disable_clock_gating(adev, i);
1068 
1069 		/* enable VCPU clock */
1070 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1071 				UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1072 
1073 		/* disable master interrupt */
1074 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN), 0,
1075 				~UVD_MASTINT_EN__VCPU_EN_MASK);
1076 
1077 		/* enable LMI MC and UMC channels */
1078 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_LMI_CTRL2), 0,
1079 				~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1080 
1081 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1082 		tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1083 		tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1084 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1085 
1086 		/* setup regUVD_LMI_CTRL */
1087 		tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL);
1088 		WREG32_SOC15(VCN, i, regUVD_LMI_CTRL, tmp |
1089 				UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
1090 				UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1091 				UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1092 				UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1093 
1094 		/* setup regUVD_MPC_CNTL */
1095 		tmp = RREG32_SOC15(VCN, i, regUVD_MPC_CNTL);
1096 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1097 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1098 		WREG32_SOC15(VCN, i, regUVD_MPC_CNTL, tmp);
1099 
1100 		/* setup UVD_MPC_SET_MUXA0 */
1101 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXA0,
1102 				((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1103 				 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1104 				 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1105 				 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1106 
1107 		/* setup UVD_MPC_SET_MUXB0 */
1108 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXB0,
1109 				((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1110 				 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1111 				 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1112 				 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1113 
1114 		/* setup UVD_MPC_SET_MUX */
1115 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUX,
1116 				((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1117 				 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1118 				 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1119 
1120 		vcn_v4_0_mc_resume(adev, i);
1121 
1122 		/* VCN global tiling registers */
1123 		WREG32_SOC15(VCN, i, regUVD_GFX10_ADDR_CONFIG,
1124 				adev->gfx.config.gb_addr_config);
1125 
1126 		/* unblock VCPU register access */
1127 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL), 0,
1128 				~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1129 
1130 		/* release VCPU reset to boot */
1131 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1132 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1133 
1134 		for (j = 0; j < 10; ++j) {
1135 			uint32_t status;
1136 
1137 			for (k = 0; k < 100; ++k) {
1138 				status = RREG32_SOC15(VCN, i, regUVD_STATUS);
1139 				if (status & 2)
1140 					break;
1141 				mdelay(10);
1142 				if (amdgpu_emu_mode == 1)
1143 					msleep(1);
1144 			}
1145 
1146 			if (amdgpu_emu_mode == 1) {
1147 				r = -1;
1148 				if (status & 2) {
1149 					r = 0;
1150 					break;
1151 				}
1152 			} else {
1153 				r = 0;
1154 				if (status & 2)
1155 					break;
1156 
1157 				dev_err(adev->dev, "VCN[%d] is not responding, trying to reset the VCPU!!!\n", i);
1158 				WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1159 							UVD_VCPU_CNTL__BLK_RST_MASK,
1160 							~UVD_VCPU_CNTL__BLK_RST_MASK);
1161 				mdelay(10);
1162 				WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1163 						~UVD_VCPU_CNTL__BLK_RST_MASK);
1164 
1165 				mdelay(10);
1166 				r = -1;
1167 			}
1168 		}
1169 
1170 		if (r) {
1171 			dev_err(adev->dev, "VCN[%d] is not responding, giving up!!!\n", i);
1172 			return r;
1173 		}
1174 
1175 		/* enable master interrupt */
1176 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN),
1177 				UVD_MASTINT_EN__VCPU_EN_MASK,
1178 				~UVD_MASTINT_EN__VCPU_EN_MASK);
1179 
1180 		/* clear the busy bit of VCN_STATUS */
1181 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_STATUS), 0,
1182 				~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1183 
1184 		ring = &adev->vcn.inst[i].ring_enc[0];
1185 		WREG32_SOC15(VCN, i, regVCN_RB1_DB_CTRL,
1186 				ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
1187 				VCN_RB1_DB_CTRL__EN_MASK);
1188 
1189 		WREG32_SOC15(VCN, i, regUVD_RB_BASE_LO, ring->gpu_addr);
1190 		WREG32_SOC15(VCN, i, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1191 		WREG32_SOC15(VCN, i, regUVD_RB_SIZE, ring->ring_size / 4);
1192 
1193 		tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1194 		tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
1195 		WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1196 		fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
1197 		WREG32_SOC15(VCN, i, regUVD_RB_RPTR, 0);
1198 		WREG32_SOC15(VCN, i, regUVD_RB_WPTR, 0);
1199 
1200 		tmp = RREG32_SOC15(VCN, i, regUVD_RB_RPTR);
1201 		WREG32_SOC15(VCN, i, regUVD_RB_WPTR, tmp);
1202 		ring->wptr = RREG32_SOC15(VCN, i, regUVD_RB_WPTR);
1203 
1204 		tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1205 		tmp |= VCN_RB_ENABLE__RB1_EN_MASK;
1206 		WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1207 		fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
1208 	}
1209 
1210 	return 0;
1211 }
1212 
1213 static int vcn_v4_0_start_sriov(struct amdgpu_device *adev)
1214 {
1215 	int i;
1216 	struct amdgpu_ring *ring_enc;
1217 	uint64_t cache_addr;
1218 	uint64_t rb_enc_addr;
1219 	uint64_t ctx_addr;
1220 	uint32_t param, resp, expected;
1221 	uint32_t offset, cache_size;
1222 	uint32_t tmp, timeout;
1223 
1224 	struct amdgpu_mm_table *table = &adev->virt.mm_table;
1225 	uint32_t *table_loc;
1226 	uint32_t table_size;
1227 	uint32_t size, size_dw;
1228 	uint32_t init_status;
1229 	uint32_t enabled_vcn;
1230 
1231 	struct mmsch_v4_0_cmd_direct_write
1232 		direct_wt = { {0} };
1233 	struct mmsch_v4_0_cmd_direct_read_modify_write
1234 		direct_rd_mod_wt = { {0} };
1235 	struct mmsch_v4_0_cmd_end end = { {0} };
1236 	struct mmsch_v4_0_init_header header;
1237 
1238 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1239 	volatile struct amdgpu_fw_shared_rb_setup *rb_setup;
1240 
1241 	direct_wt.cmd_header.command_type =
1242 		MMSCH_COMMAND__DIRECT_REG_WRITE;
1243 	direct_rd_mod_wt.cmd_header.command_type =
1244 		MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1245 	end.cmd_header.command_type =
1246 		MMSCH_COMMAND__END;
1247 
1248 	header.version = MMSCH_VERSION;
1249 	header.total_size = sizeof(struct mmsch_v4_0_init_header) >> 2;
1250 	for (i = 0; i < MMSCH_V4_0_VCN_INSTANCES; i++) {
1251 		header.inst[i].init_status = 0;
1252 		header.inst[i].table_offset = 0;
1253 		header.inst[i].table_size = 0;
1254 	}
1255 
1256 	table_loc = (uint32_t *)table->cpu_addr;
1257 	table_loc += header.total_size;
1258 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1259 		if (adev->vcn.harvest_config & (1 << i))
1260 			continue;
1261 
1262 		table_size = 0;
1263 
1264 		MMSCH_V4_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(VCN, i,
1265 			regUVD_STATUS),
1266 			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1267 
1268 		cache_size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1269 
1270 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1271 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1272 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1273 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1274 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1275 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1276 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1277 			offset = 0;
1278 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1279 				regUVD_VCPU_CACHE_OFFSET0),
1280 				0);
1281 		} else {
1282 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1283 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1284 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1285 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1286 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1287 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1288 			offset = cache_size;
1289 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1290 				regUVD_VCPU_CACHE_OFFSET0),
1291 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1292 		}
1293 
1294 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1295 			regUVD_VCPU_CACHE_SIZE0),
1296 			cache_size);
1297 
1298 		cache_addr = adev->vcn.inst[i].gpu_addr + offset;
1299 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1300 			regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1301 			lower_32_bits(cache_addr));
1302 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1303 			regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1304 			upper_32_bits(cache_addr));
1305 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1306 			regUVD_VCPU_CACHE_OFFSET1),
1307 			0);
1308 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1309 			regUVD_VCPU_CACHE_SIZE1),
1310 			AMDGPU_VCN_STACK_SIZE);
1311 
1312 		cache_addr = adev->vcn.inst[i].gpu_addr + offset +
1313 			AMDGPU_VCN_STACK_SIZE;
1314 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1315 			regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1316 			lower_32_bits(cache_addr));
1317 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1318 			regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1319 			upper_32_bits(cache_addr));
1320 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1321 			regUVD_VCPU_CACHE_OFFSET2),
1322 			0);
1323 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1324 			regUVD_VCPU_CACHE_SIZE2),
1325 			AMDGPU_VCN_CONTEXT_SIZE);
1326 
1327 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1328 		rb_setup = &fw_shared->rb_setup;
1329 
1330 		ring_enc = &adev->vcn.inst[i].ring_enc[0];
1331 		ring_enc->wptr = 0;
1332 		rb_enc_addr = ring_enc->gpu_addr;
1333 
1334 		rb_setup->is_rb_enabled_flags |= RB_ENABLED;
1335 		rb_setup->rb_addr_lo = lower_32_bits(rb_enc_addr);
1336 		rb_setup->rb_addr_hi = upper_32_bits(rb_enc_addr);
1337 		rb_setup->rb_size = ring_enc->ring_size / 4;
1338 		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
1339 
1340 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1341 			regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
1342 			lower_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
1343 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1344 			regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
1345 			upper_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
1346 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1347 			regUVD_VCPU_NONCACHE_SIZE0),
1348 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
1349 
1350 		/* add end packet */
1351 		MMSCH_V4_0_INSERT_END();
1352 
1353 		/* refine header */
1354 		header.inst[i].init_status = 0;
1355 		header.inst[i].table_offset = header.total_size;
1356 		header.inst[i].table_size = table_size;
1357 		header.total_size += table_size;
1358 	}
1359 
1360 	/* Update init table header in memory */
1361 	size = sizeof(struct mmsch_v4_0_init_header);
1362 	table_loc = (uint32_t *)table->cpu_addr;
1363 	memcpy((void *)table_loc, &header, size);
1364 
1365 	/* message MMSCH (in VCN[0]) to initialize this client
1366 	 * 1, write to mmsch_vf_ctx_addr_lo/hi register with GPU mc addr
1367 	 * of memory descriptor location
1368 	 */
1369 	ctx_addr = table->gpu_addr;
1370 	WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_ADDR_LO, lower_32_bits(ctx_addr));
1371 	WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_ADDR_HI, upper_32_bits(ctx_addr));
1372 
1373 	/* 2, update vmid of descriptor */
1374 	tmp = RREG32_SOC15(VCN, 0, regMMSCH_VF_VMID);
1375 	tmp &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1376 	/* use domain0 for MM scheduler */
1377 	tmp |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1378 	WREG32_SOC15(VCN, 0, regMMSCH_VF_VMID, tmp);
1379 
1380 	/* 3, notify mmsch about the size of this descriptor */
1381 	size = header.total_size;
1382 	WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_SIZE, size);
1383 
1384 	/* 4, set resp to zero */
1385 	WREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_RESP, 0);
1386 
1387 	/* 5, kick off the initialization and wait until
1388 	 * MMSCH_VF_MAILBOX_RESP becomes non-zero
1389 	 */
1390 	param = 0x00000001;
1391 	WREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_HOST, param);
1392 	tmp = 0;
1393 	timeout = 1000;
1394 	resp = 0;
1395 	expected = MMSCH_VF_MAILBOX_RESP__OK;
1396 	while (resp != expected) {
1397 		resp = RREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_RESP);
1398 		if (resp != 0)
1399 			break;
1400 
1401 		udelay(10);
1402 		tmp = tmp + 10;
1403 		if (tmp >= timeout) {
1404 			DRM_ERROR("failed to init MMSCH. TIME-OUT after %d usec"\
1405 				" waiting for regMMSCH_VF_MAILBOX_RESP "\
1406 				"(expected=0x%08x, readback=0x%08x)\n",
1407 				tmp, expected, resp);
1408 			return -EBUSY;
1409 		}
1410 	}
1411 	enabled_vcn = amdgpu_vcn_is_disabled_vcn(adev, VCN_DECODE_RING, 0) ? 1 : 0;
1412 	init_status = ((struct mmsch_v4_0_init_header *)(table_loc))->inst[enabled_vcn].init_status;
1413 	if (resp != expected && resp != MMSCH_VF_MAILBOX_RESP__INCOMPLETE
1414 	&& init_status != MMSCH_VF_ENGINE_STATUS__PASS)
1415 		DRM_ERROR("MMSCH init status is incorrect! readback=0x%08x, header init "\
1416 			"status for VCN%x: 0x%x\n", resp, enabled_vcn, init_status);
1417 
1418 	return 0;
1419 }
1420 
1421 /**
1422  * vcn_v4_0_stop_dpg_mode - VCN stop with dpg mode
1423  *
1424  * @adev: amdgpu_device pointer
1425  * @inst_idx: instance number index
1426  *
1427  * Stop VCN block with dpg mode
1428  */
1429 static void vcn_v4_0_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1430 {
1431 	struct dpg_pause_state state = {.fw_based = VCN_DPG_STATE__UNPAUSE};
1432 	uint32_t tmp;
1433 
1434 	vcn_v4_0_pause_dpg_mode(adev, inst_idx, &state);
1435 	/* Wait for power status to be 1 */
1436 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1437 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1438 
1439 	/* wait for read ptr to be equal to write ptr */
1440 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
1441 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1442 
1443 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1444 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1445 
1446 	/* disable dynamic power gating mode */
1447 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 0,
1448 		~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1449 }
1450 
1451 /**
1452  * vcn_v4_0_stop - VCN stop
1453  *
1454  * @adev: amdgpu_device pointer
1455  *
1456  * Stop VCN block
1457  */
1458 static int vcn_v4_0_stop(struct amdgpu_device *adev)
1459 {
1460 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1461 	uint32_t tmp;
1462 	int i, r = 0;
1463 
1464 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1465 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1466 		fw_shared->sq.queue_mode |= FW_QUEUE_DPG_HOLD_OFF;
1467 
1468 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1469 			vcn_v4_0_stop_dpg_mode(adev, i);
1470 			continue;
1471 		}
1472 
1473 		/* wait for vcn idle */
1474 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1475 		if (r)
1476 			return r;
1477 
1478 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1479 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1480 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1481 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1482 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1483 		if (r)
1484 			return r;
1485 
1486 		/* disable LMI UMC channel */
1487 		tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL2);
1488 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1489 		WREG32_SOC15(VCN, i, regUVD_LMI_CTRL2, tmp);
1490 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK |
1491 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1492 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1493 		if (r)
1494 			return r;
1495 
1496 		/* block VCPU register access */
1497 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL),
1498 				UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1499 				~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1500 
1501 		/* reset VCPU */
1502 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1503 				UVD_VCPU_CNTL__BLK_RST_MASK,
1504 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1505 
1506 		/* disable VCPU clock */
1507 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1508 				~(UVD_VCPU_CNTL__CLK_EN_MASK));
1509 
1510 		/* apply soft reset */
1511 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1512 		tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1513 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1514 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1515 		tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1516 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1517 
1518 		/* clear status */
1519 		WREG32_SOC15(VCN, i, regUVD_STATUS, 0);
1520 
1521 		/* apply HW clock gating */
1522 		vcn_v4_0_enable_clock_gating(adev, i);
1523 
1524 		/* enable VCN power gating */
1525 		vcn_v4_0_enable_static_power_gating(adev, i);
1526 	}
1527 
1528 	if (adev->pm.dpm_enabled)
1529 		amdgpu_dpm_enable_uvd(adev, false);
1530 
1531 	return 0;
1532 }
1533 
1534 /**
1535  * vcn_v4_0_pause_dpg_mode - VCN pause with dpg mode
1536  *
1537  * @adev: amdgpu_device pointer
1538  * @inst_idx: instance number index
1539  * @new_state: pause state
1540  *
1541  * Pause dpg mode for VCN block
1542  */
1543 static int vcn_v4_0_pause_dpg_mode(struct amdgpu_device *adev, int inst_idx,
1544       struct dpg_pause_state *new_state)
1545 {
1546 	uint32_t reg_data = 0;
1547 	int ret_code;
1548 
1549 	/* pause/unpause if state is changed */
1550 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1551 		DRM_DEV_DEBUG(adev->dev, "dpg pause state changed %d -> %d",
1552 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1553 		reg_data = RREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE) &
1554 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1555 
1556 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1557 			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 0x1,
1558 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1559 
1560 			if (!ret_code) {
1561 				/* pause DPG */
1562 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1563 				WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1564 
1565 				/* wait for ACK */
1566 				SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_DPG_PAUSE,
1567 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1568 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1569 
1570 				SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS,
1571 					UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1572 			}
1573 		} else {
1574 			/* unpause dpg, no need to wait */
1575 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1576 			WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1577 		}
1578 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1579 	}
1580 
1581 	return 0;
1582 }
1583 
1584 /**
1585  * vcn_v4_0_unified_ring_get_rptr - get unified read pointer
1586  *
1587  * @ring: amdgpu_ring pointer
1588  *
1589  * Returns the current hardware unified read pointer
1590  */
1591 static uint64_t vcn_v4_0_unified_ring_get_rptr(struct amdgpu_ring *ring)
1592 {
1593 	struct amdgpu_device *adev = ring->adev;
1594 
1595 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1596 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1597 
1598 	return RREG32_SOC15(VCN, ring->me, regUVD_RB_RPTR);
1599 }
1600 
1601 /**
1602  * vcn_v4_0_unified_ring_get_wptr - get unified write pointer
1603  *
1604  * @ring: amdgpu_ring pointer
1605  *
1606  * Returns the current hardware unified write pointer
1607  */
1608 static uint64_t vcn_v4_0_unified_ring_get_wptr(struct amdgpu_ring *ring)
1609 {
1610 	struct amdgpu_device *adev = ring->adev;
1611 
1612 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1613 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1614 
1615 	if (ring->use_doorbell)
1616 		return *ring->wptr_cpu_addr;
1617 	else
1618 		return RREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR);
1619 }
1620 
1621 /**
1622  * vcn_v4_0_unified_ring_set_wptr - set enc write pointer
1623  *
1624  * @ring: amdgpu_ring pointer
1625  *
1626  * Commits the enc write pointer to the hardware
1627  */
1628 static void vcn_v4_0_unified_ring_set_wptr(struct amdgpu_ring *ring)
1629 {
1630 	struct amdgpu_device *adev = ring->adev;
1631 
1632 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1633 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1634 
1635 	if (ring->use_doorbell) {
1636 		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1637 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1638 	} else {
1639 		WREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR, lower_32_bits(ring->wptr));
1640 	}
1641 }
1642 
1643 static int vcn_v4_0_limit_sched(struct amdgpu_cs_parser *p,
1644 				struct amdgpu_job *job)
1645 {
1646 	struct drm_gpu_scheduler **scheds;
1647 
1648 	/* The create msg must be in the first IB submitted */
1649 	if (atomic_read(&job->base.entity->fence_seq))
1650 		return -EINVAL;
1651 
1652 	/* if VCN0 is harvested, we can't support AV1 */
1653 	if (p->adev->vcn.harvest_config & AMDGPU_VCN_HARVEST_VCN0)
1654 		return -EINVAL;
1655 
1656 	scheds = p->adev->gpu_sched[AMDGPU_HW_IP_VCN_ENC]
1657 		[AMDGPU_RING_PRIO_0].sched;
1658 	drm_sched_entity_modify_sched(job->base.entity, scheds, 1);
1659 	return 0;
1660 }
1661 
1662 static int vcn_v4_0_dec_msg(struct amdgpu_cs_parser *p, struct amdgpu_job *job,
1663 			    uint64_t addr)
1664 {
1665 	struct ttm_operation_ctx ctx = { false, false };
1666 	struct amdgpu_bo_va_mapping *map;
1667 	uint32_t *msg, num_buffers;
1668 	struct amdgpu_bo *bo;
1669 	uint64_t start, end;
1670 	unsigned int i;
1671 	void *ptr;
1672 	int r;
1673 
1674 	addr &= AMDGPU_GMC_HOLE_MASK;
1675 	r = amdgpu_cs_find_mapping(p, addr, &bo, &map);
1676 	if (r) {
1677 		DRM_ERROR("Can't find BO for addr 0x%08llx\n", addr);
1678 		return r;
1679 	}
1680 
1681 	start = map->start * AMDGPU_GPU_PAGE_SIZE;
1682 	end = (map->last + 1) * AMDGPU_GPU_PAGE_SIZE;
1683 	if (addr & 0x7) {
1684 		DRM_ERROR("VCN messages must be 8 byte aligned!\n");
1685 		return -EINVAL;
1686 	}
1687 
1688 	bo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
1689 	amdgpu_bo_placement_from_domain(bo, bo->allowed_domains);
1690 	r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
1691 	if (r) {
1692 		DRM_ERROR("Failed validating the VCN message BO (%d)!\n", r);
1693 		return r;
1694 	}
1695 
1696 	r = amdgpu_bo_kmap(bo, &ptr);
1697 	if (r) {
1698 		DRM_ERROR("Failed mapping the VCN message (%d)!\n", r);
1699 		return r;
1700 	}
1701 
1702 	msg = ptr + addr - start;
1703 
1704 	/* Check length */
1705 	if (msg[1] > end - addr) {
1706 		r = -EINVAL;
1707 		goto out;
1708 	}
1709 
1710 	if (msg[3] != RDECODE_MSG_CREATE)
1711 		goto out;
1712 
1713 	num_buffers = msg[2];
1714 	for (i = 0, msg = &msg[6]; i < num_buffers; ++i, msg += 4) {
1715 		uint32_t offset, size, *create;
1716 
1717 		if (msg[0] != RDECODE_MESSAGE_CREATE)
1718 			continue;
1719 
1720 		offset = msg[1];
1721 		size = msg[2];
1722 
1723 		if (offset + size > end) {
1724 			r = -EINVAL;
1725 			goto out;
1726 		}
1727 
1728 		create = ptr + addr + offset - start;
1729 
1730 		/* H264, HEVC and VP9 can run on any instance */
1731 		if (create[0] == 0x7 || create[0] == 0x10 || create[0] == 0x11)
1732 			continue;
1733 
1734 		r = vcn_v4_0_limit_sched(p, job);
1735 		if (r)
1736 			goto out;
1737 	}
1738 
1739 out:
1740 	amdgpu_bo_kunmap(bo);
1741 	return r;
1742 }
1743 
1744 #define RADEON_VCN_ENGINE_TYPE_ENCODE			(0x00000002)
1745 #define RADEON_VCN_ENGINE_TYPE_DECODE			(0x00000003)
1746 
1747 #define RADEON_VCN_ENGINE_INFO				(0x30000001)
1748 #define RADEON_VCN_ENGINE_INFO_MAX_OFFSET		16
1749 
1750 #define RENCODE_ENCODE_STANDARD_AV1			2
1751 #define RENCODE_IB_PARAM_SESSION_INIT			0x00000003
1752 #define RENCODE_IB_PARAM_SESSION_INIT_MAX_OFFSET	64
1753 
1754 /* return the offset in ib if id is found, -1 otherwise
1755  * to speed up the searching we only search upto max_offset
1756  */
1757 static int vcn_v4_0_enc_find_ib_param(struct amdgpu_ib *ib, uint32_t id, int max_offset)
1758 {
1759 	int i;
1760 
1761 	for (i = 0; i < ib->length_dw && i < max_offset && ib->ptr[i] >= 8; i += ib->ptr[i]/4) {
1762 		if (ib->ptr[i + 1] == id)
1763 			return i;
1764 	}
1765 	return -1;
1766 }
1767 
1768 static int vcn_v4_0_ring_patch_cs_in_place(struct amdgpu_cs_parser *p,
1769 					   struct amdgpu_job *job,
1770 					   struct amdgpu_ib *ib)
1771 {
1772 	struct amdgpu_ring *ring = amdgpu_job_ring(job);
1773 	struct amdgpu_vcn_decode_buffer *decode_buffer;
1774 	uint64_t addr;
1775 	uint32_t val;
1776 	int idx;
1777 
1778 	/* The first instance can decode anything */
1779 	if (!ring->me)
1780 		return 0;
1781 
1782 	/* RADEON_VCN_ENGINE_INFO is at the top of ib block */
1783 	idx = vcn_v4_0_enc_find_ib_param(ib, RADEON_VCN_ENGINE_INFO,
1784 			RADEON_VCN_ENGINE_INFO_MAX_OFFSET);
1785 	if (idx < 0) /* engine info is missing */
1786 		return 0;
1787 
1788 	val = amdgpu_ib_get_value(ib, idx + 2); /* RADEON_VCN_ENGINE_TYPE */
1789 	if (val == RADEON_VCN_ENGINE_TYPE_DECODE) {
1790 		decode_buffer = (struct amdgpu_vcn_decode_buffer *)&ib->ptr[idx + 6];
1791 
1792 		if (!(decode_buffer->valid_buf_flag  & 0x1))
1793 			return 0;
1794 
1795 		addr = ((u64)decode_buffer->msg_buffer_address_hi) << 32 |
1796 			decode_buffer->msg_buffer_address_lo;
1797 		return vcn_v4_0_dec_msg(p, job, addr);
1798 	} else if (val == RADEON_VCN_ENGINE_TYPE_ENCODE) {
1799 		idx = vcn_v4_0_enc_find_ib_param(ib, RENCODE_IB_PARAM_SESSION_INIT,
1800 			RENCODE_IB_PARAM_SESSION_INIT_MAX_OFFSET);
1801 		if (idx >= 0 && ib->ptr[idx + 2] == RENCODE_ENCODE_STANDARD_AV1)
1802 			return vcn_v4_0_limit_sched(p, job);
1803 	}
1804 	return 0;
1805 }
1806 
1807 static struct amdgpu_ring_funcs vcn_v4_0_unified_ring_vm_funcs = {
1808 	.type = AMDGPU_RING_TYPE_VCN_ENC,
1809 	.align_mask = 0x3f,
1810 	.nop = VCN_ENC_CMD_NO_OP,
1811 	.get_rptr = vcn_v4_0_unified_ring_get_rptr,
1812 	.get_wptr = vcn_v4_0_unified_ring_get_wptr,
1813 	.set_wptr = vcn_v4_0_unified_ring_set_wptr,
1814 	.patch_cs_in_place = vcn_v4_0_ring_patch_cs_in_place,
1815 	.emit_frame_size =
1816 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1817 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1818 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1819 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1820 		1, /* vcn_v2_0_enc_ring_insert_end */
1821 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1822 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
1823 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
1824 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1825 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
1826 	.test_ib = amdgpu_vcn_unified_ring_test_ib,
1827 	.insert_nop = amdgpu_ring_insert_nop,
1828 	.insert_end = vcn_v2_0_enc_ring_insert_end,
1829 	.pad_ib = amdgpu_ring_generic_pad_ib,
1830 	.begin_use = amdgpu_vcn_ring_begin_use,
1831 	.end_use = amdgpu_vcn_ring_end_use,
1832 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1833 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1834 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1835 };
1836 
1837 /**
1838  * vcn_v4_0_set_unified_ring_funcs - set unified ring functions
1839  *
1840  * @adev: amdgpu_device pointer
1841  *
1842  * Set unified ring functions
1843  */
1844 static void vcn_v4_0_set_unified_ring_funcs(struct amdgpu_device *adev)
1845 {
1846 	int i;
1847 
1848 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1849 		if (adev->vcn.harvest_config & (1 << i))
1850 			continue;
1851 
1852 		if (adev->ip_versions[VCN_HWIP][0] == IP_VERSION(4, 0, 2))
1853 			vcn_v4_0_unified_ring_vm_funcs.secure_submission_supported = true;
1854 
1855 		adev->vcn.inst[i].ring_enc[0].funcs =
1856 		       (const struct amdgpu_ring_funcs *)&vcn_v4_0_unified_ring_vm_funcs;
1857 		adev->vcn.inst[i].ring_enc[0].me = i;
1858 
1859 		DRM_INFO("VCN(%d) encode/decode are enabled in VM mode\n", i);
1860 	}
1861 }
1862 
1863 /**
1864  * vcn_v4_0_is_idle - check VCN block is idle
1865  *
1866  * @handle: amdgpu_device pointer
1867  *
1868  * Check whether VCN block is idle
1869  */
1870 static bool vcn_v4_0_is_idle(void *handle)
1871 {
1872 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1873 	int i, ret = 1;
1874 
1875 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1876 		if (adev->vcn.harvest_config & (1 << i))
1877 			continue;
1878 
1879 		ret &= (RREG32_SOC15(VCN, i, regUVD_STATUS) == UVD_STATUS__IDLE);
1880 	}
1881 
1882 	return ret;
1883 }
1884 
1885 /**
1886  * vcn_v4_0_wait_for_idle - wait for VCN block idle
1887  *
1888  * @handle: amdgpu_device pointer
1889  *
1890  * Wait for VCN block idle
1891  */
1892 static int vcn_v4_0_wait_for_idle(void *handle)
1893 {
1894 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1895 	int i, ret = 0;
1896 
1897 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1898 		if (adev->vcn.harvest_config & (1 << i))
1899 			continue;
1900 
1901 		ret = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE,
1902 			UVD_STATUS__IDLE);
1903 		if (ret)
1904 			return ret;
1905 	}
1906 
1907 	return ret;
1908 }
1909 
1910 /**
1911  * vcn_v4_0_set_clockgating_state - set VCN block clockgating state
1912  *
1913  * @handle: amdgpu_device pointer
1914  * @state: clock gating state
1915  *
1916  * Set VCN block clockgating state
1917  */
1918 static int vcn_v4_0_set_clockgating_state(void *handle, enum amd_clockgating_state state)
1919 {
1920 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1921 	bool enable = state == AMD_CG_STATE_GATE;
1922 	int i;
1923 
1924 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1925 		if (adev->vcn.harvest_config & (1 << i))
1926 			continue;
1927 
1928 		if (enable) {
1929 			if (RREG32_SOC15(VCN, i, regUVD_STATUS) != UVD_STATUS__IDLE)
1930 				return -EBUSY;
1931 			vcn_v4_0_enable_clock_gating(adev, i);
1932 		} else {
1933 			vcn_v4_0_disable_clock_gating(adev, i);
1934 		}
1935 	}
1936 
1937 	return 0;
1938 }
1939 
1940 /**
1941  * vcn_v4_0_set_powergating_state - set VCN block powergating state
1942  *
1943  * @handle: amdgpu_device pointer
1944  * @state: power gating state
1945  *
1946  * Set VCN block powergating state
1947  */
1948 static int vcn_v4_0_set_powergating_state(void *handle, enum amd_powergating_state state)
1949 {
1950 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1951 	int ret;
1952 
1953 	/* for SRIOV, guest should not control VCN Power-gating
1954 	 * MMSCH FW should control Power-gating and clock-gating
1955 	 * guest should avoid touching CGC and PG
1956 	 */
1957 	if (amdgpu_sriov_vf(adev)) {
1958 		adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
1959 		return 0;
1960 	}
1961 
1962 	if (state == adev->vcn.cur_state)
1963 		return 0;
1964 
1965 	if (state == AMD_PG_STATE_GATE)
1966 		ret = vcn_v4_0_stop(adev);
1967 	else
1968 		ret = vcn_v4_0_start(adev);
1969 
1970 	if (!ret)
1971 		adev->vcn.cur_state = state;
1972 
1973 	return ret;
1974 }
1975 
1976 /**
1977  * vcn_v4_0_set_interrupt_state - set VCN block interrupt state
1978  *
1979  * @adev: amdgpu_device pointer
1980  * @source: interrupt sources
1981  * @type: interrupt types
1982  * @state: interrupt states
1983  *
1984  * Set VCN block interrupt state
1985  */
1986 static int vcn_v4_0_set_interrupt_state(struct amdgpu_device *adev, struct amdgpu_irq_src *source,
1987       unsigned type, enum amdgpu_interrupt_state state)
1988 {
1989 	return 0;
1990 }
1991 
1992 /**
1993  * vcn_v4_0_set_ras_interrupt_state - set VCN block RAS interrupt state
1994  *
1995  * @adev: amdgpu_device pointer
1996  * @source: interrupt sources
1997  * @type: interrupt types
1998  * @state: interrupt states
1999  *
2000  * Set VCN block RAS interrupt state
2001  */
2002 static int vcn_v4_0_set_ras_interrupt_state(struct amdgpu_device *adev,
2003 	struct amdgpu_irq_src *source,
2004 	unsigned int type,
2005 	enum amdgpu_interrupt_state state)
2006 {
2007 	return 0;
2008 }
2009 
2010 /**
2011  * vcn_v4_0_process_interrupt - process VCN block interrupt
2012  *
2013  * @adev: amdgpu_device pointer
2014  * @source: interrupt sources
2015  * @entry: interrupt entry from clients and sources
2016  *
2017  * Process VCN block interrupt
2018  */
2019 static int vcn_v4_0_process_interrupt(struct amdgpu_device *adev, struct amdgpu_irq_src *source,
2020       struct amdgpu_iv_entry *entry)
2021 {
2022 	uint32_t ip_instance;
2023 
2024 	switch (entry->client_id) {
2025 	case SOC15_IH_CLIENTID_VCN:
2026 		ip_instance = 0;
2027 		break;
2028 	case SOC15_IH_CLIENTID_VCN1:
2029 		ip_instance = 1;
2030 		break;
2031 	default:
2032 		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
2033 		return 0;
2034 	}
2035 
2036 	DRM_DEBUG("IH: VCN TRAP\n");
2037 
2038 	switch (entry->src_id) {
2039 	case VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
2040 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
2041 		break;
2042 	default:
2043 		DRM_ERROR("Unhandled interrupt: %d %d\n",
2044 			  entry->src_id, entry->src_data[0]);
2045 		break;
2046 	}
2047 
2048 	return 0;
2049 }
2050 
2051 static const struct amdgpu_irq_src_funcs vcn_v4_0_irq_funcs = {
2052 	.set = vcn_v4_0_set_interrupt_state,
2053 	.process = vcn_v4_0_process_interrupt,
2054 };
2055 
2056 static const struct amdgpu_irq_src_funcs vcn_v4_0_ras_irq_funcs = {
2057 	.set = vcn_v4_0_set_ras_interrupt_state,
2058 	.process = amdgpu_vcn_process_poison_irq,
2059 };
2060 
2061 /**
2062  * vcn_v4_0_set_irq_funcs - set VCN block interrupt irq functions
2063  *
2064  * @adev: amdgpu_device pointer
2065  *
2066  * Set VCN block interrupt irq functions
2067  */
2068 static void vcn_v4_0_set_irq_funcs(struct amdgpu_device *adev)
2069 {
2070 	int i;
2071 
2072 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2073 		if (adev->vcn.harvest_config & (1 << i))
2074 			continue;
2075 
2076 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
2077 		adev->vcn.inst[i].irq.funcs = &vcn_v4_0_irq_funcs;
2078 
2079 		adev->vcn.inst[i].ras_poison_irq.num_types = adev->vcn.num_enc_rings + 1;
2080 		adev->vcn.inst[i].ras_poison_irq.funcs = &vcn_v4_0_ras_irq_funcs;
2081 	}
2082 }
2083 
2084 static const struct amd_ip_funcs vcn_v4_0_ip_funcs = {
2085 	.name = "vcn_v4_0",
2086 	.early_init = vcn_v4_0_early_init,
2087 	.late_init = NULL,
2088 	.sw_init = vcn_v4_0_sw_init,
2089 	.sw_fini = vcn_v4_0_sw_fini,
2090 	.hw_init = vcn_v4_0_hw_init,
2091 	.hw_fini = vcn_v4_0_hw_fini,
2092 	.suspend = vcn_v4_0_suspend,
2093 	.resume = vcn_v4_0_resume,
2094 	.is_idle = vcn_v4_0_is_idle,
2095 	.wait_for_idle = vcn_v4_0_wait_for_idle,
2096 	.check_soft_reset = NULL,
2097 	.pre_soft_reset = NULL,
2098 	.soft_reset = NULL,
2099 	.post_soft_reset = NULL,
2100 	.set_clockgating_state = vcn_v4_0_set_clockgating_state,
2101 	.set_powergating_state = vcn_v4_0_set_powergating_state,
2102 };
2103 
2104 const struct amdgpu_ip_block_version vcn_v4_0_ip_block = {
2105 	.type = AMD_IP_BLOCK_TYPE_VCN,
2106 	.major = 4,
2107 	.minor = 0,
2108 	.rev = 0,
2109 	.funcs = &vcn_v4_0_ip_funcs,
2110 };
2111 
2112 static uint32_t vcn_v4_0_query_poison_by_instance(struct amdgpu_device *adev,
2113 			uint32_t instance, uint32_t sub_block)
2114 {
2115 	uint32_t poison_stat = 0, reg_value = 0;
2116 
2117 	switch (sub_block) {
2118 	case AMDGPU_VCN_V4_0_VCPU_VCODEC:
2119 		reg_value = RREG32_SOC15(VCN, instance, regUVD_RAS_VCPU_VCODEC_STATUS);
2120 		poison_stat = REG_GET_FIELD(reg_value, UVD_RAS_VCPU_VCODEC_STATUS, POISONED_PF);
2121 		break;
2122 	default:
2123 		break;
2124 	}
2125 
2126 	if (poison_stat)
2127 		dev_info(adev->dev, "Poison detected in VCN%d, sub_block%d\n",
2128 			instance, sub_block);
2129 
2130 	return poison_stat;
2131 }
2132 
2133 static bool vcn_v4_0_query_ras_poison_status(struct amdgpu_device *adev)
2134 {
2135 	uint32_t inst, sub;
2136 	uint32_t poison_stat = 0;
2137 
2138 	for (inst = 0; inst < adev->vcn.num_vcn_inst; inst++)
2139 		for (sub = 0; sub < AMDGPU_VCN_V4_0_MAX_SUB_BLOCK; sub++)
2140 			poison_stat +=
2141 				vcn_v4_0_query_poison_by_instance(adev, inst, sub);
2142 
2143 	return !!poison_stat;
2144 }
2145 
2146 const struct amdgpu_ras_block_hw_ops vcn_v4_0_ras_hw_ops = {
2147 	.query_poison_status = vcn_v4_0_query_ras_poison_status,
2148 };
2149 
2150 static struct amdgpu_vcn_ras vcn_v4_0_ras = {
2151 	.ras_block = {
2152 		.hw_ops = &vcn_v4_0_ras_hw_ops,
2153 		.ras_late_init = amdgpu_vcn_ras_late_init,
2154 	},
2155 };
2156 
2157 static void vcn_v4_0_set_ras_funcs(struct amdgpu_device *adev)
2158 {
2159 	switch (adev->ip_versions[VCN_HWIP][0]) {
2160 	case IP_VERSION(4, 0, 0):
2161 		adev->vcn.ras = &vcn_v4_0_ras;
2162 		break;
2163 	default:
2164 		break;
2165 	}
2166 }
2167