xref: /linux/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c (revision 7fc7f25419f5a6b09199ba4b5026b94ef184fa79)
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 
35 #include "vcn/vcn_4_0_0_offset.h"
36 #include "vcn/vcn_4_0_0_sh_mask.h"
37 #include "ivsrcid/vcn/irqsrcs_vcn_4_0.h"
38 
39 #include <drm/drm_drv.h>
40 
41 #define mmUVD_DPG_LMA_CTL							regUVD_DPG_LMA_CTL
42 #define mmUVD_DPG_LMA_CTL_BASE_IDX						regUVD_DPG_LMA_CTL_BASE_IDX
43 #define mmUVD_DPG_LMA_DATA							regUVD_DPG_LMA_DATA
44 #define mmUVD_DPG_LMA_DATA_BASE_IDX						regUVD_DPG_LMA_DATA_BASE_IDX
45 
46 #define VCN_VID_SOC_ADDRESS_2_0							0x1fb00
47 #define VCN1_VID_SOC_ADDRESS_3_0						0x48300
48 
49 #define VCN_HARVEST_MMSCH								0
50 
51 #define RDECODE_MSG_CREATE							0x00000000
52 #define RDECODE_MESSAGE_CREATE							0x00000001
53 
54 static int amdgpu_ih_clientid_vcns[] = {
55 	SOC15_IH_CLIENTID_VCN,
56 	SOC15_IH_CLIENTID_VCN1
57 };
58 
59 static int vcn_v4_0_start_sriov(struct amdgpu_device *adev);
60 static void vcn_v4_0_set_unified_ring_funcs(struct amdgpu_device *adev);
61 static void vcn_v4_0_set_irq_funcs(struct amdgpu_device *adev);
62 static int vcn_v4_0_set_powergating_state(void *handle,
63         enum amd_powergating_state state);
64 static int vcn_v4_0_pause_dpg_mode(struct amdgpu_device *adev,
65         int inst_idx, struct dpg_pause_state *new_state);
66 static void vcn_v4_0_unified_ring_set_wptr(struct amdgpu_ring *ring);
67 
68 /**
69  * vcn_v4_0_early_init - set function pointers
70  *
71  * @handle: amdgpu_device pointer
72  *
73  * Set ring and irq function pointers
74  */
75 static int vcn_v4_0_early_init(void *handle)
76 {
77 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
78 
79 	if (amdgpu_sriov_vf(adev))
80 		adev->vcn.harvest_config = VCN_HARVEST_MMSCH;
81 
82 	/* re-use enc ring as unified ring */
83 	adev->vcn.num_enc_rings = 1;
84 
85 	vcn_v4_0_set_unified_ring_funcs(adev);
86 	vcn_v4_0_set_irq_funcs(adev);
87 
88 	return 0;
89 }
90 
91 /**
92  * vcn_v4_0_sw_init - sw init for VCN block
93  *
94  * @handle: amdgpu_device pointer
95  *
96  * Load firmware and sw initialization
97  */
98 static int vcn_v4_0_sw_init(void *handle)
99 {
100 	struct amdgpu_ring *ring;
101 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
102 	int i, r;
103 	int vcn_doorbell_index = 0;
104 
105 	r = amdgpu_vcn_sw_init(adev);
106 	if (r)
107 		return r;
108 
109 	amdgpu_vcn_setup_ucode(adev);
110 
111 	r = amdgpu_vcn_resume(adev);
112 	if (r)
113 		return r;
114 
115 	if (amdgpu_sriov_vf(adev)) {
116 		vcn_doorbell_index = adev->doorbell_index.vcn.vcn_ring0_1 - MMSCH_DOORBELL_OFFSET;
117 		/* get DWORD offset */
118 		vcn_doorbell_index = vcn_doorbell_index << 1;
119 	}
120 
121 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
122 		volatile struct amdgpu_vcn4_fw_shared *fw_shared;
123 
124 		if (adev->vcn.harvest_config & (1 << i))
125 			continue;
126 
127 		atomic_set(&adev->vcn.inst[i].sched_score, 0);
128 
129 		/* VCN UNIFIED TRAP */
130 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
131 				VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);
132 		if (r)
133 			return r;
134 
135 		ring = &adev->vcn.inst[i].ring_enc[0];
136 		ring->use_doorbell = true;
137 		if (amdgpu_sriov_vf(adev))
138 			ring->doorbell_index = vcn_doorbell_index + i * (adev->vcn.num_enc_rings + 1) + 1;
139 		else
140 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + 8 * i;
141 
142 		sprintf(ring->name, "vcn_unified_%d", i);
143 
144 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
145 						AMDGPU_RING_PRIO_0, &adev->vcn.inst[i].sched_score);
146 		if (r)
147 			return r;
148 
149 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
150 		fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_FW_SHARED_FLAG_0_UNIFIED_QUEUE);
151 		fw_shared->sq.is_enabled = 1;
152 
153 		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_SMU_DPM_INTERFACE_FLAG);
154 		fw_shared->smu_dpm_interface.smu_interface_type = (adev->flags & AMD_IS_APU) ?
155 			AMDGPU_VCN_SMU_DPM_INTERFACE_APU : AMDGPU_VCN_SMU_DPM_INTERFACE_DGPU;
156 
157 		if (amdgpu_sriov_vf(adev))
158 			fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
159 
160 		if (amdgpu_vcnfw_log)
161 			amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
162 	}
163 
164 	if (amdgpu_sriov_vf(adev)) {
165 		r = amdgpu_virt_alloc_mm_table(adev);
166 		if (r)
167 			return r;
168 	}
169 
170 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
171 		adev->vcn.pause_dpg_mode = vcn_v4_0_pause_dpg_mode;
172 
173 	return 0;
174 }
175 
176 /**
177  * vcn_v4_0_sw_fini - sw fini for VCN block
178  *
179  * @handle: amdgpu_device pointer
180  *
181  * VCN suspend and free up sw allocation
182  */
183 static int vcn_v4_0_sw_fini(void *handle)
184 {
185 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
186 	int i, r, idx;
187 
188 	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
189 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
190 			volatile struct amdgpu_vcn4_fw_shared *fw_shared;
191 
192 			if (adev->vcn.harvest_config & (1 << i))
193 				continue;
194 
195 			fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
196 			fw_shared->present_flag_0 = 0;
197 			fw_shared->sq.is_enabled = 0;
198 		}
199 
200 		drm_dev_exit(idx);
201 	}
202 
203 	if (amdgpu_sriov_vf(adev))
204 		amdgpu_virt_free_mm_table(adev);
205 
206 	r = amdgpu_vcn_suspend(adev);
207 	if (r)
208 		return r;
209 
210 	r = amdgpu_vcn_sw_fini(adev);
211 
212 	return r;
213 }
214 
215 /**
216  * vcn_v4_0_hw_init - start and test VCN block
217  *
218  * @handle: amdgpu_device pointer
219  *
220  * Initialize the hardware, boot up the VCPU and do some testing
221  */
222 static int vcn_v4_0_hw_init(void *handle)
223 {
224 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
225 	struct amdgpu_ring *ring;
226 	int i, r;
227 
228 	if (amdgpu_sriov_vf(adev)) {
229 		r = vcn_v4_0_start_sriov(adev);
230 		if (r)
231 			goto done;
232 
233 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
234 			if (adev->vcn.harvest_config & (1 << i))
235 				continue;
236 
237 			ring = &adev->vcn.inst[i].ring_enc[0];
238 			if (amdgpu_vcn_is_disabled_vcn(adev, VCN_ENCODE_RING, i)) {
239 				ring->sched.ready = false;
240 				ring->no_scheduler = true;
241 				dev_info(adev->dev, "ring %s is disabled by hypervisor\n", ring->name);
242 			} else {
243 				ring->wptr = 0;
244 				ring->wptr_old = 0;
245 				vcn_v4_0_unified_ring_set_wptr(ring);
246 				ring->sched.ready = true;
247 			}
248 		}
249 	} else {
250 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
251 			if (adev->vcn.harvest_config & (1 << i))
252 				continue;
253 
254 			ring = &adev->vcn.inst[i].ring_enc[0];
255 
256 			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
257 					((adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i), i);
258 
259 			r = amdgpu_ring_test_helper(ring);
260 			if (r)
261 				goto done;
262 
263 		}
264 	}
265 
266 done:
267 	if (!r)
268 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
269 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
270 
271 	return r;
272 }
273 
274 /**
275  * vcn_v4_0_hw_fini - stop the hardware block
276  *
277  * @handle: amdgpu_device pointer
278  *
279  * Stop the VCN block, mark ring as not ready any more
280  */
281 static int vcn_v4_0_hw_fini(void *handle)
282 {
283 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
284 	int i;
285 
286 	cancel_delayed_work_sync(&adev->vcn.idle_work);
287 
288 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
289 		if (adev->vcn.harvest_config & (1 << i))
290 			continue;
291 		if (!amdgpu_sriov_vf(adev)) {
292 			if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
293                         (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
294                                 RREG32_SOC15(VCN, i, regUVD_STATUS))) {
295                         vcn_v4_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
296 			}
297 		}
298 
299 	}
300 
301 	return 0;
302 }
303 
304 /**
305  * vcn_v4_0_suspend - suspend VCN block
306  *
307  * @handle: amdgpu_device pointer
308  *
309  * HW fini and suspend VCN block
310  */
311 static int vcn_v4_0_suspend(void *handle)
312 {
313 	int r;
314 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
315 
316 	r = vcn_v4_0_hw_fini(adev);
317 	if (r)
318 		return r;
319 
320 	r = amdgpu_vcn_suspend(adev);
321 
322 	return r;
323 }
324 
325 /**
326  * vcn_v4_0_resume - resume VCN block
327  *
328  * @handle: amdgpu_device pointer
329  *
330  * Resume firmware and hw init VCN block
331  */
332 static int vcn_v4_0_resume(void *handle)
333 {
334 	int r;
335 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
336 
337 	r = amdgpu_vcn_resume(adev);
338 	if (r)
339 		return r;
340 
341 	r = vcn_v4_0_hw_init(adev);
342 
343 	return r;
344 }
345 
346 /**
347  * vcn_v4_0_mc_resume - memory controller programming
348  *
349  * @adev: amdgpu_device pointer
350  * @inst: instance number
351  *
352  * Let the VCN memory controller know it's offsets
353  */
354 static void vcn_v4_0_mc_resume(struct amdgpu_device *adev, int inst)
355 {
356 	uint32_t offset, size;
357 	const struct common_firmware_header *hdr;
358 
359 	hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
360 	size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
361 
362 	/* cache window 0: fw */
363 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
364 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
365 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo));
366 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
367 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi));
368 		WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, 0);
369 		offset = 0;
370 	} else {
371 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
372 			lower_32_bits(adev->vcn.inst[inst].gpu_addr));
373 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
374 			upper_32_bits(adev->vcn.inst[inst].gpu_addr));
375 		offset = size;
376                 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
377 	}
378 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE0, size);
379 
380 	/* cache window 1: stack */
381 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
382 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
383 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
384 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
385 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET1, 0);
386 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
387 
388 	/* cache window 2: context */
389 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
390 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
391 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
392 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
393 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET2, 0);
394 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
395 
396 	/* non-cache window */
397 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
398 		lower_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
399 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
400 		upper_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
401 	WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_OFFSET0, 0);
402 	WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_SIZE0,
403 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
404 }
405 
406 /**
407  * vcn_v4_0_mc_resume_dpg_mode - memory controller programming for dpg mode
408  *
409  * @adev: amdgpu_device pointer
410  * @inst_idx: instance number index
411  * @indirect: indirectly write sram
412  *
413  * Let the VCN memory controller know it's offsets with dpg mode
414  */
415 static void vcn_v4_0_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
416 {
417 	uint32_t offset, size;
418 	const struct common_firmware_header *hdr;
419 	hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
420 	size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
421 
422 	/* cache window 0: fw */
423 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
424 		if (!indirect) {
425 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
426 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
427 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
428 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
429 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
430 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
431 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
432 				VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
433 		} else {
434 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
435 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
436 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
437 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
438 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
439 				VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
440 		}
441 		offset = 0;
442 	} else {
443 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
444 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
445 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
446 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
447 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
448 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
449 		offset = size;
450 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
451 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0),
452 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
453 	}
454 
455 	if (!indirect)
456 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
457 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
458 	else
459 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
460 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
461 
462 	/* cache window 1: stack */
463 	if (!indirect) {
464 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
465 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
466 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
467 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
468 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
469 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
470 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
471 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
472 	} else {
473 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
474 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
475 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
476 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
477 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
478 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
479 	}
480 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
481 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
482 
483 	/* cache window 2: context */
484 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
485 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
486 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
487 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
488 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
489 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
490 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
491 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
492 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
493 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
494 
495 	/* non-cache window */
496 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
497 			VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
498 			lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
499 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
500 			VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
501 			upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
502 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
503 			VCN, inst_idx, regUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
504 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
505 			VCN, inst_idx, regUVD_VCPU_NONCACHE_SIZE0),
506 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)), 0, indirect);
507 
508 	/* VCN global tiling registers */
509 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
510 		VCN, 0, regUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
511 }
512 
513 /**
514  * vcn_v4_0_disable_static_power_gating - disable VCN static power gating
515  *
516  * @adev: amdgpu_device pointer
517  * @inst: instance number
518  *
519  * Disable static power gating for VCN block
520  */
521 static void vcn_v4_0_disable_static_power_gating(struct amdgpu_device *adev, int inst)
522 {
523 	uint32_t data = 0;
524 
525 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
526 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
527 			| 1 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
528 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
529 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
530 			| 2 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
531 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
532 			| 2 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
533 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
534 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
535 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
536 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
537 			| 2 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
538 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
539 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
540 
541 		WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
542 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS,
543 			UVD_PGFSM_STATUS__UVDM_UVDU_UVDLM_PWR_ON_3_0, 0x3F3FFFFF);
544 	} else {
545 		uint32_t value;
546 
547 		value = (inst) ? 0x2200800 : 0;
548 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
549 			| 1 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
550 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
551 			| 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
552 			| 1 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
553 			| 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
554 			| 1 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
555 			| 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
556 			| 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
557 			| 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
558 			| 1 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
559 			| 1 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
560 			| 1 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
561 			| 1 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
562 
563                 WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
564                 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS, value,  0x3F3FFFFF);
565         }
566 
567         data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
568         data &= ~0x103;
569         if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
570                 data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
571                         UVD_POWER_STATUS__UVD_PG_EN_MASK;
572 
573         WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
574 
575         return;
576 }
577 
578 /**
579  * vcn_v4_0_enable_static_power_gating - enable VCN static power gating
580  *
581  * @adev: amdgpu_device pointer
582  * @inst: instance number
583  *
584  * Enable static power gating for VCN block
585  */
586 static void vcn_v4_0_enable_static_power_gating(struct amdgpu_device *adev, int inst)
587 {
588 	uint32_t data;
589 
590 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
591 		/* Before power off, this indicator has to be turned on */
592 		data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
593 		data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
594 		data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
595 		WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
596 
597 		data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
598 			| 2 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
599 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
600 			| 2 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
601 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
602 			| 2 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
603 			| 2 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
604 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
605 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
606 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
607 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
608 			| 2 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
609 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
610 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
611 		WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
612 
613 		data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
614 			| 2 << UVD_PGFSM_STATUS__UVDS_PWR_STATUS__SHIFT
615 			| 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
616 			| 2 << UVD_PGFSM_STATUS__UVDTC_PWR_STATUS__SHIFT
617 			| 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
618 			| 2 << UVD_PGFSM_STATUS__UVDTA_PWR_STATUS__SHIFT
619 			| 2 << UVD_PGFSM_STATUS__UVDLM_PWR_STATUS__SHIFT
620 			| 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
621 			| 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
622 			| 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT
623 			| 2 << UVD_PGFSM_STATUS__UVDAB_PWR_STATUS__SHIFT
624 			| 2 << UVD_PGFSM_STATUS__UVDTB_PWR_STATUS__SHIFT
625 			| 2 << UVD_PGFSM_STATUS__UVDNA_PWR_STATUS__SHIFT
626 			| 2 << UVD_PGFSM_STATUS__UVDNB_PWR_STATUS__SHIFT);
627 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS, data, 0x3F3FFFFF);
628 	}
629 
630         return;
631 }
632 
633 /**
634  * vcn_v4_0_disable_clock_gating - disable VCN clock gating
635  *
636  * @adev: amdgpu_device pointer
637  * @inst: instance number
638  *
639  * Disable clock gating for VCN block
640  */
641 static void vcn_v4_0_disable_clock_gating(struct amdgpu_device *adev, int inst)
642 {
643 	uint32_t data;
644 
645 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
646 		return;
647 
648 	/* VCN disable CGC */
649 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
650 	data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
651 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
652 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
653 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
654 
655 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_GATE);
656 	data &= ~(UVD_CGC_GATE__SYS_MASK
657 		| UVD_CGC_GATE__UDEC_MASK
658 		| UVD_CGC_GATE__MPEG2_MASK
659 		| UVD_CGC_GATE__REGS_MASK
660 		| UVD_CGC_GATE__RBC_MASK
661 		| UVD_CGC_GATE__LMI_MC_MASK
662 		| UVD_CGC_GATE__LMI_UMC_MASK
663 		| UVD_CGC_GATE__IDCT_MASK
664 		| UVD_CGC_GATE__MPRD_MASK
665 		| UVD_CGC_GATE__MPC_MASK
666 		| UVD_CGC_GATE__LBSI_MASK
667 		| UVD_CGC_GATE__LRBBM_MASK
668 		| UVD_CGC_GATE__UDEC_RE_MASK
669 		| UVD_CGC_GATE__UDEC_CM_MASK
670 		| UVD_CGC_GATE__UDEC_IT_MASK
671 		| UVD_CGC_GATE__UDEC_DB_MASK
672 		| UVD_CGC_GATE__UDEC_MP_MASK
673 		| UVD_CGC_GATE__WCB_MASK
674 		| UVD_CGC_GATE__VCPU_MASK
675 		| UVD_CGC_GATE__MMSCH_MASK);
676 
677 	WREG32_SOC15(VCN, inst, regUVD_CGC_GATE, data);
678 	SOC15_WAIT_ON_RREG(VCN, inst, regUVD_CGC_GATE, 0,  0xFFFFFFFF);
679 
680 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
681 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
682 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
683 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
684 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
685 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
686 		| UVD_CGC_CTRL__SYS_MODE_MASK
687 		| UVD_CGC_CTRL__UDEC_MODE_MASK
688 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
689 		| UVD_CGC_CTRL__REGS_MODE_MASK
690 		| UVD_CGC_CTRL__RBC_MODE_MASK
691 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
692 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
693 		| UVD_CGC_CTRL__IDCT_MODE_MASK
694 		| UVD_CGC_CTRL__MPRD_MODE_MASK
695 		| UVD_CGC_CTRL__MPC_MODE_MASK
696 		| UVD_CGC_CTRL__LBSI_MODE_MASK
697 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
698 		| UVD_CGC_CTRL__WCB_MODE_MASK
699 		| UVD_CGC_CTRL__VCPU_MODE_MASK
700 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
701 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
702 
703 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE);
704 	data |= (UVD_SUVD_CGC_GATE__SRE_MASK
705 		| UVD_SUVD_CGC_GATE__SIT_MASK
706 		| UVD_SUVD_CGC_GATE__SMP_MASK
707 		| UVD_SUVD_CGC_GATE__SCM_MASK
708 		| UVD_SUVD_CGC_GATE__SDB_MASK
709 		| UVD_SUVD_CGC_GATE__SRE_H264_MASK
710 		| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
711 		| UVD_SUVD_CGC_GATE__SIT_H264_MASK
712 		| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
713 		| UVD_SUVD_CGC_GATE__SCM_H264_MASK
714 		| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
715 		| UVD_SUVD_CGC_GATE__SDB_H264_MASK
716 		| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
717 		| UVD_SUVD_CGC_GATE__SCLR_MASK
718 		| UVD_SUVD_CGC_GATE__UVD_SC_MASK
719 		| UVD_SUVD_CGC_GATE__ENT_MASK
720 		| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
721 		| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
722 		| UVD_SUVD_CGC_GATE__SITE_MASK
723 		| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
724 		| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
725 		| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
726 		| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
727 		| UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
728 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE, data);
729 
730 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
731 	data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
732 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
733 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
734 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
735 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
736 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
737 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
738 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
739 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
740 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
741 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
742 }
743 
744 /**
745  * vcn_v4_0_disable_clock_gating_dpg_mode - disable VCN clock gating dpg mode
746  *
747  * @adev: amdgpu_device pointer
748  * @sram_sel: sram select
749  * @inst_idx: instance number index
750  * @indirect: indirectly write sram
751  *
752  * Disable clock gating for VCN block with dpg mode
753  */
754 static void vcn_v4_0_disable_clock_gating_dpg_mode(struct amdgpu_device *adev, uint8_t sram_sel,
755       int inst_idx, uint8_t indirect)
756 {
757 	uint32_t reg_data = 0;
758 
759 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
760 		return;
761 
762 	/* enable sw clock gating control */
763 	reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
764 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
765 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
766 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
767 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
768 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
769 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
770 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
771 		 UVD_CGC_CTRL__SYS_MODE_MASK |
772 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
773 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
774 		 UVD_CGC_CTRL__REGS_MODE_MASK |
775 		 UVD_CGC_CTRL__RBC_MODE_MASK |
776 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
777 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
778 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
779 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
780 		 UVD_CGC_CTRL__MPC_MODE_MASK |
781 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
782 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
783 		 UVD_CGC_CTRL__WCB_MODE_MASK |
784 		 UVD_CGC_CTRL__VCPU_MODE_MASK);
785 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
786 		VCN, inst_idx, regUVD_CGC_CTRL), reg_data, sram_sel, indirect);
787 
788 	/* turn off clock gating */
789 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
790 		VCN, inst_idx, regUVD_CGC_GATE), 0, sram_sel, indirect);
791 
792 	/* turn on SUVD clock gating */
793 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
794 		VCN, inst_idx, regUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
795 
796 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
797 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
798 		VCN, inst_idx, regUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
799 }
800 
801 /**
802  * vcn_v4_0_enable_clock_gating - enable VCN clock gating
803  *
804  * @adev: amdgpu_device pointer
805  * @inst: instance number
806  *
807  * Enable clock gating for VCN block
808  */
809 static void vcn_v4_0_enable_clock_gating(struct amdgpu_device *adev, int inst)
810 {
811 	uint32_t data;
812 
813 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
814 		return;
815 
816 	/* enable VCN CGC */
817 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
818 	data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
819 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
820 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
821 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
822 
823 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
824 	data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
825 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
826 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
827 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
828 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
829 		| UVD_CGC_CTRL__SYS_MODE_MASK
830 		| UVD_CGC_CTRL__UDEC_MODE_MASK
831 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
832 		| UVD_CGC_CTRL__REGS_MODE_MASK
833 		| UVD_CGC_CTRL__RBC_MODE_MASK
834 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
835 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
836 		| UVD_CGC_CTRL__IDCT_MODE_MASK
837 		| UVD_CGC_CTRL__MPRD_MODE_MASK
838 		| UVD_CGC_CTRL__MPC_MODE_MASK
839 		| UVD_CGC_CTRL__LBSI_MODE_MASK
840 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
841 		| UVD_CGC_CTRL__WCB_MODE_MASK
842 		| UVD_CGC_CTRL__VCPU_MODE_MASK
843 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
844 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
845 
846 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
847 	data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
848 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
849 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
850 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
851 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
852 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
853 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
854 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
855 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
856 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
857 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
858 
859 	return;
860 }
861 
862 /**
863  * vcn_v4_0_start_dpg_mode - VCN start with dpg mode
864  *
865  * @adev: amdgpu_device pointer
866  * @inst_idx: instance number index
867  * @indirect: indirectly write sram
868  *
869  * Start VCN block with dpg mode
870  */
871 static int vcn_v4_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
872 {
873 	volatile struct amdgpu_vcn4_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
874 	struct amdgpu_ring *ring;
875 	uint32_t tmp;
876 
877 	/* disable register anti-hang mechanism */
878 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 1,
879 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
880 	/* enable dynamic power gating mode */
881 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS);
882 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
883 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
884 	WREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS, tmp);
885 
886 	if (indirect)
887 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
888 
889 	/* enable clock gating */
890 	vcn_v4_0_disable_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
891 
892 	/* enable VCPU clock */
893 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
894 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK | UVD_VCPU_CNTL__BLK_RST_MASK;
895 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
896 		VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
897 
898 	/* disable master interupt */
899 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
900 		VCN, inst_idx, regUVD_MASTINT_EN), 0, 0, indirect);
901 
902 	/* setup regUVD_LMI_CTRL */
903 	tmp = (UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
904 		UVD_LMI_CTRL__REQ_MODE_MASK |
905 		UVD_LMI_CTRL__CRC_RESET_MASK |
906 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
907 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
908 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
909 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
910 		0x00100000L);
911 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
912 		VCN, inst_idx, regUVD_LMI_CTRL), tmp, 0, indirect);
913 
914 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
915 		VCN, inst_idx, regUVD_MPC_CNTL),
916 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
917 
918 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
919 		VCN, inst_idx, regUVD_MPC_SET_MUXA0),
920 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
921 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
922 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
923 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
924 
925 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
926 		VCN, inst_idx, regUVD_MPC_SET_MUXB0),
927 		 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
928 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
929 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
930 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
931 
932 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
933 		VCN, inst_idx, regUVD_MPC_SET_MUX),
934 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
935 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
936 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
937 
938 	vcn_v4_0_mc_resume_dpg_mode(adev, inst_idx, indirect);
939 
940 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
941 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
942 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
943 		VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
944 
945 	/* enable LMI MC and UMC channels */
946 	tmp = 0x1f << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT;
947 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
948 		VCN, inst_idx, regUVD_LMI_CTRL2), tmp, 0, indirect);
949 
950 	/* enable master interrupt */
951 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
952 		VCN, inst_idx, regUVD_MASTINT_EN),
953 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
954 
955 
956 	if (indirect)
957 		psp_update_vcn_sram(adev, inst_idx, adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
958 			(uint32_t)((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
959 				(uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr));
960 
961 	ring = &adev->vcn.inst[inst_idx].ring_enc[0];
962 
963 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_LO, ring->gpu_addr);
964 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
965 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_SIZE, ring->ring_size / 4);
966 
967 	tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
968 	tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
969 	WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
970 	fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
971 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR, 0);
972 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, 0);
973 
974 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR);
975 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, tmp);
976 	ring->wptr = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
977 
978 	tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
979 	tmp |= VCN_RB_ENABLE__RB1_EN_MASK;
980 	WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
981 	fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
982 
983 	WREG32_SOC15(VCN, inst_idx, regVCN_RB1_DB_CTRL,
984 			ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
985 			VCN_RB1_DB_CTRL__EN_MASK);
986 
987 	return 0;
988 }
989 
990 
991 /**
992  * vcn_v4_0_start - VCN start
993  *
994  * @adev: amdgpu_device pointer
995  *
996  * Start VCN block
997  */
998 static int vcn_v4_0_start(struct amdgpu_device *adev)
999 {
1000 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1001 	struct amdgpu_ring *ring;
1002 	uint32_t tmp;
1003 	int i, j, k, r;
1004 
1005 	if (adev->pm.dpm_enabled)
1006 		amdgpu_dpm_enable_uvd(adev, true);
1007 
1008 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1009 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1010 
1011 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1012 			r = vcn_v4_0_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
1013 			continue;
1014 		}
1015 
1016 		/* disable VCN power gating */
1017 		vcn_v4_0_disable_static_power_gating(adev, i);
1018 
1019 		/* set VCN status busy */
1020 		tmp = RREG32_SOC15(VCN, i, regUVD_STATUS) | UVD_STATUS__UVD_BUSY;
1021 		WREG32_SOC15(VCN, i, regUVD_STATUS, tmp);
1022 
1023 		/*SW clock gating */
1024 		vcn_v4_0_disable_clock_gating(adev, i);
1025 
1026 		/* enable VCPU clock */
1027 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1028 				UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1029 
1030 		/* disable master interrupt */
1031 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN), 0,
1032 				~UVD_MASTINT_EN__VCPU_EN_MASK);
1033 
1034 		/* enable LMI MC and UMC channels */
1035 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_LMI_CTRL2), 0,
1036 				~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1037 
1038 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1039 		tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1040 		tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1041 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1042 
1043 		/* setup regUVD_LMI_CTRL */
1044 		tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL);
1045 		WREG32_SOC15(VCN, i, regUVD_LMI_CTRL, tmp |
1046 				UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
1047 				UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1048 				UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1049 				UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1050 
1051 		/* setup regUVD_MPC_CNTL */
1052 		tmp = RREG32_SOC15(VCN, i, regUVD_MPC_CNTL);
1053 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1054 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1055 		WREG32_SOC15(VCN, i, regUVD_MPC_CNTL, tmp);
1056 
1057 		/* setup UVD_MPC_SET_MUXA0 */
1058 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXA0,
1059 				((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1060 				 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1061 				 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1062 				 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1063 
1064 		/* setup UVD_MPC_SET_MUXB0 */
1065 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXB0,
1066 				((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1067 				 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1068 				 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1069 				 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1070 
1071 		/* setup UVD_MPC_SET_MUX */
1072 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUX,
1073 				((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1074 				 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1075 				 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1076 
1077 		vcn_v4_0_mc_resume(adev, i);
1078 
1079 		/* VCN global tiling registers */
1080 		WREG32_SOC15(VCN, i, regUVD_GFX10_ADDR_CONFIG,
1081 				adev->gfx.config.gb_addr_config);
1082 
1083 		/* unblock VCPU register access */
1084 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL), 0,
1085 				~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1086 
1087 		/* release VCPU reset to boot */
1088 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1089 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1090 
1091 		for (j = 0; j < 10; ++j) {
1092 			uint32_t status;
1093 
1094 			for (k = 0; k < 100; ++k) {
1095 				status = RREG32_SOC15(VCN, i, regUVD_STATUS);
1096 				if (status & 2)
1097 					break;
1098 				mdelay(10);
1099 				if (amdgpu_emu_mode==1)
1100 					msleep(1);
1101 			}
1102 
1103 			if (amdgpu_emu_mode==1) {
1104 				r = -1;
1105 				if (status & 2) {
1106 					r = 0;
1107 					break;
1108 				}
1109 			} else {
1110 				r = 0;
1111 				if (status & 2)
1112 					break;
1113 
1114 				dev_err(adev->dev, "VCN[%d] is not responding, trying to reset the VCPU!!!\n", i);
1115 				WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1116 							UVD_VCPU_CNTL__BLK_RST_MASK,
1117 							~UVD_VCPU_CNTL__BLK_RST_MASK);
1118 				mdelay(10);
1119 				WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1120 						~UVD_VCPU_CNTL__BLK_RST_MASK);
1121 
1122 				mdelay(10);
1123 				r = -1;
1124 			}
1125 		}
1126 
1127 		if (r) {
1128 			dev_err(adev->dev, "VCN[%d] is not responding, giving up!!!\n", i);
1129 			return r;
1130 		}
1131 
1132 		/* enable master interrupt */
1133 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN),
1134 				UVD_MASTINT_EN__VCPU_EN_MASK,
1135 				~UVD_MASTINT_EN__VCPU_EN_MASK);
1136 
1137 		/* clear the busy bit of VCN_STATUS */
1138 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_STATUS), 0,
1139 				~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1140 
1141 		ring = &adev->vcn.inst[i].ring_enc[0];
1142 		WREG32_SOC15(VCN, i, regVCN_RB1_DB_CTRL,
1143 				ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
1144 				VCN_RB1_DB_CTRL__EN_MASK);
1145 
1146 		WREG32_SOC15(VCN, i, regUVD_RB_BASE_LO, ring->gpu_addr);
1147 		WREG32_SOC15(VCN, i, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1148 		WREG32_SOC15(VCN, i, regUVD_RB_SIZE, ring->ring_size / 4);
1149 
1150 		tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1151 		tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
1152 		WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1153 		fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
1154 		WREG32_SOC15(VCN, i, regUVD_RB_RPTR, 0);
1155 		WREG32_SOC15(VCN, i, regUVD_RB_WPTR, 0);
1156 
1157 		tmp = RREG32_SOC15(VCN, i, regUVD_RB_RPTR);
1158 		WREG32_SOC15(VCN, i, regUVD_RB_WPTR, tmp);
1159 		ring->wptr = RREG32_SOC15(VCN, i, regUVD_RB_WPTR);
1160 
1161 		tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1162 		tmp |= VCN_RB_ENABLE__RB1_EN_MASK;
1163 		WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1164 		fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
1165 	}
1166 
1167 	return 0;
1168 }
1169 
1170 static int vcn_v4_0_start_sriov(struct amdgpu_device *adev)
1171 {
1172 	int i;
1173 	struct amdgpu_ring *ring_enc;
1174 	uint64_t cache_addr;
1175 	uint64_t rb_enc_addr;
1176 	uint64_t ctx_addr;
1177 	uint32_t param, resp, expected;
1178 	uint32_t offset, cache_size;
1179 	uint32_t tmp, timeout;
1180 
1181 	struct amdgpu_mm_table *table = &adev->virt.mm_table;
1182 	uint32_t *table_loc;
1183 	uint32_t table_size;
1184 	uint32_t size, size_dw;
1185 	uint32_t init_status;
1186 	uint32_t enabled_vcn;
1187 
1188 	struct mmsch_v4_0_cmd_direct_write
1189 		direct_wt = { {0} };
1190 	struct mmsch_v4_0_cmd_direct_read_modify_write
1191 		direct_rd_mod_wt = { {0} };
1192 	struct mmsch_v4_0_cmd_end end = { {0} };
1193 	struct mmsch_v4_0_init_header header;
1194 
1195 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1196 	volatile struct amdgpu_fw_shared_rb_setup *rb_setup;
1197 
1198 	direct_wt.cmd_header.command_type =
1199 		MMSCH_COMMAND__DIRECT_REG_WRITE;
1200 	direct_rd_mod_wt.cmd_header.command_type =
1201 		MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1202 	end.cmd_header.command_type =
1203 		MMSCH_COMMAND__END;
1204 
1205 	header.version = MMSCH_VERSION;
1206 	header.total_size = sizeof(struct mmsch_v4_0_init_header) >> 2;
1207 	for (i = 0; i < AMDGPU_MAX_VCN_INSTANCES; i++) {
1208 		header.inst[i].init_status = 0;
1209 		header.inst[i].table_offset = 0;
1210 		header.inst[i].table_size = 0;
1211 	}
1212 
1213 	table_loc = (uint32_t *)table->cpu_addr;
1214 	table_loc += header.total_size;
1215 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1216 		if (adev->vcn.harvest_config & (1 << i))
1217 			continue;
1218 
1219 		table_size = 0;
1220 
1221 		MMSCH_V4_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(VCN, i,
1222 			regUVD_STATUS),
1223 			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1224 
1225 		cache_size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1226 
1227 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1228 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1229 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1230 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1231 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1232 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1233 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1234 			offset = 0;
1235 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1236 				regUVD_VCPU_CACHE_OFFSET0),
1237 				0);
1238 		} else {
1239 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1240 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1241 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1242 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1243 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1244 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1245 			offset = cache_size;
1246 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1247 				regUVD_VCPU_CACHE_OFFSET0),
1248 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1249 		}
1250 
1251 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1252 			regUVD_VCPU_CACHE_SIZE0),
1253 			cache_size);
1254 
1255 		cache_addr = adev->vcn.inst[i].gpu_addr + offset;
1256 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1257 			regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1258 			lower_32_bits(cache_addr));
1259 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1260 			regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1261 			upper_32_bits(cache_addr));
1262 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1263 			regUVD_VCPU_CACHE_OFFSET1),
1264 			0);
1265 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1266 			regUVD_VCPU_CACHE_SIZE1),
1267 			AMDGPU_VCN_STACK_SIZE);
1268 
1269 		cache_addr = adev->vcn.inst[i].gpu_addr + offset +
1270 			AMDGPU_VCN_STACK_SIZE;
1271 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1272 			regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1273 			lower_32_bits(cache_addr));
1274 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1275 			regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1276 			upper_32_bits(cache_addr));
1277 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1278 			regUVD_VCPU_CACHE_OFFSET2),
1279 			0);
1280 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1281 			regUVD_VCPU_CACHE_SIZE2),
1282 			AMDGPU_VCN_CONTEXT_SIZE);
1283 
1284 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1285 		rb_setup = &fw_shared->rb_setup;
1286 
1287 		ring_enc = &adev->vcn.inst[i].ring_enc[0];
1288 		ring_enc->wptr = 0;
1289 		rb_enc_addr = ring_enc->gpu_addr;
1290 
1291 		rb_setup->is_rb_enabled_flags |= RB_ENABLED;
1292 		rb_setup->rb_addr_lo = lower_32_bits(rb_enc_addr);
1293 		rb_setup->rb_addr_hi = upper_32_bits(rb_enc_addr);
1294 		rb_setup->rb_size = ring_enc->ring_size / 4;
1295 		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
1296 
1297 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1298 			regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
1299 			lower_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
1300 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1301 			regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
1302 			upper_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
1303 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1304 			regUVD_VCPU_NONCACHE_SIZE0),
1305 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
1306 
1307 		/* add end packet */
1308 		MMSCH_V4_0_INSERT_END();
1309 
1310 		/* refine header */
1311 		header.inst[i].init_status = 0;
1312 		header.inst[i].table_offset = header.total_size;
1313 		header.inst[i].table_size = table_size;
1314 		header.total_size += table_size;
1315 	}
1316 
1317 	/* Update init table header in memory */
1318 	size = sizeof(struct mmsch_v4_0_init_header);
1319 	table_loc = (uint32_t *)table->cpu_addr;
1320 	memcpy((void *)table_loc, &header, size);
1321 
1322 	/* message MMSCH (in VCN[0]) to initialize this client
1323 	 * 1, write to mmsch_vf_ctx_addr_lo/hi register with GPU mc addr
1324 	 * of memory descriptor location
1325 	 */
1326 	ctx_addr = table->gpu_addr;
1327 	WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_ADDR_LO, lower_32_bits(ctx_addr));
1328 	WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_ADDR_HI, upper_32_bits(ctx_addr));
1329 
1330 	/* 2, update vmid of descriptor */
1331 	tmp = RREG32_SOC15(VCN, 0, regMMSCH_VF_VMID);
1332 	tmp &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1333 	/* use domain0 for MM scheduler */
1334 	tmp |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1335 	WREG32_SOC15(VCN, 0, regMMSCH_VF_VMID, tmp);
1336 
1337 	/* 3, notify mmsch about the size of this descriptor */
1338 	size = header.total_size;
1339 	WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_SIZE, size);
1340 
1341 	/* 4, set resp to zero */
1342 	WREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_RESP, 0);
1343 
1344 	/* 5, kick off the initialization and wait until
1345 	 * MMSCH_VF_MAILBOX_RESP becomes non-zero
1346 	 */
1347 	param = 0x00000001;
1348 	WREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_HOST, param);
1349 	tmp = 0;
1350 	timeout = 1000;
1351 	resp = 0;
1352 	expected = MMSCH_VF_MAILBOX_RESP__OK;
1353 	while (resp != expected) {
1354 		resp = RREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_RESP);
1355 		if (resp != 0)
1356 			break;
1357 
1358 		udelay(10);
1359 		tmp = tmp + 10;
1360 		if (tmp >= timeout) {
1361 			DRM_ERROR("failed to init MMSCH. TIME-OUT after %d usec"\
1362 				" waiting for regMMSCH_VF_MAILBOX_RESP "\
1363 				"(expected=0x%08x, readback=0x%08x)\n",
1364 				tmp, expected, resp);
1365 			return -EBUSY;
1366 		}
1367 	}
1368 	enabled_vcn = amdgpu_vcn_is_disabled_vcn(adev, VCN_DECODE_RING, 0) ? 1 : 0;
1369 	init_status = ((struct mmsch_v4_0_init_header *)(table_loc))->inst[enabled_vcn].init_status;
1370 	if (resp != expected && resp != MMSCH_VF_MAILBOX_RESP__INCOMPLETE
1371 	&& init_status != MMSCH_VF_ENGINE_STATUS__PASS)
1372 		DRM_ERROR("MMSCH init status is incorrect! readback=0x%08x, header init "\
1373 			"status for VCN%x: 0x%x\n", resp, enabled_vcn, init_status);
1374 
1375 	return 0;
1376 }
1377 
1378 /**
1379  * vcn_v4_0_stop_dpg_mode - VCN stop with dpg mode
1380  *
1381  * @adev: amdgpu_device pointer
1382  * @inst_idx: instance number index
1383  *
1384  * Stop VCN block with dpg mode
1385  */
1386 static void vcn_v4_0_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1387 {
1388 	uint32_t tmp;
1389 
1390 	/* Wait for power status to be 1 */
1391 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1392 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1393 
1394 	/* wait for read ptr to be equal to write ptr */
1395 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
1396 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1397 
1398 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1399 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1400 
1401 	/* disable dynamic power gating mode */
1402 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 0,
1403 		~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1404 }
1405 
1406 /**
1407  * vcn_v4_0_stop - VCN stop
1408  *
1409  * @adev: amdgpu_device pointer
1410  *
1411  * Stop VCN block
1412  */
1413 static int vcn_v4_0_stop(struct amdgpu_device *adev)
1414 {
1415 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1416 	uint32_t tmp;
1417 	int i, r = 0;
1418 
1419 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1420 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1421 		fw_shared->sq.queue_mode |= FW_QUEUE_DPG_HOLD_OFF;
1422 
1423 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1424 			vcn_v4_0_stop_dpg_mode(adev, i);
1425 			continue;
1426 		}
1427 
1428 		/* wait for vcn idle */
1429 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1430 		if (r)
1431 			return r;
1432 
1433 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1434 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1435 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1436 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1437 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1438 		if (r)
1439 			return r;
1440 
1441 		/* disable LMI UMC channel */
1442 		tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL2);
1443 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1444 		WREG32_SOC15(VCN, i, regUVD_LMI_CTRL2, tmp);
1445 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK |
1446 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1447 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1448 		if (r)
1449 			return r;
1450 
1451 		/* block VCPU register access */
1452 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL),
1453 				UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1454 				~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1455 
1456 		/* reset VCPU */
1457 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1458 				UVD_VCPU_CNTL__BLK_RST_MASK,
1459 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1460 
1461 		/* disable VCPU clock */
1462 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1463 				~(UVD_VCPU_CNTL__CLK_EN_MASK));
1464 
1465 		/* apply soft reset */
1466 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1467 		tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1468 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1469 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1470 		tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1471 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1472 
1473 		/* clear status */
1474 		WREG32_SOC15(VCN, i, regUVD_STATUS, 0);
1475 
1476 		/* apply HW clock gating */
1477 		vcn_v4_0_enable_clock_gating(adev, i);
1478 
1479 		/* enable VCN power gating */
1480 		vcn_v4_0_enable_static_power_gating(adev, i);
1481 	}
1482 
1483 	if (adev->pm.dpm_enabled)
1484 		amdgpu_dpm_enable_uvd(adev, false);
1485 
1486 	return 0;
1487 }
1488 
1489 /**
1490  * vcn_v4_0_pause_dpg_mode - VCN pause with dpg mode
1491  *
1492  * @adev: amdgpu_device pointer
1493  * @inst_idx: instance number index
1494  * @new_state: pause state
1495  *
1496  * Pause dpg mode for VCN block
1497  */
1498 static int vcn_v4_0_pause_dpg_mode(struct amdgpu_device *adev, int inst_idx,
1499       struct dpg_pause_state *new_state)
1500 {
1501 	uint32_t reg_data = 0;
1502 	int ret_code;
1503 
1504 	/* pause/unpause if state is changed */
1505 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1506 		DRM_DEV_DEBUG(adev->dev, "dpg pause state changed %d -> %d",
1507 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1508 		reg_data = RREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE) &
1509 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1510 
1511 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1512 			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 0x1,
1513 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1514 
1515 			if (!ret_code) {
1516 				/* pause DPG */
1517 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1518 				WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1519 
1520 				/* wait for ACK */
1521 				SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_DPG_PAUSE,
1522 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1523 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1524 
1525 				SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS,
1526 					UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1527 			}
1528 		} else {
1529 			/* unpause dpg, no need to wait */
1530 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1531 			WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1532 		}
1533 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1534 	}
1535 
1536 	return 0;
1537 }
1538 
1539 /**
1540  * vcn_v4_0_unified_ring_get_rptr - get unified read pointer
1541  *
1542  * @ring: amdgpu_ring pointer
1543  *
1544  * Returns the current hardware unified read pointer
1545  */
1546 static uint64_t vcn_v4_0_unified_ring_get_rptr(struct amdgpu_ring *ring)
1547 {
1548 	struct amdgpu_device *adev = ring->adev;
1549 
1550 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1551 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1552 
1553 	return RREG32_SOC15(VCN, ring->me, regUVD_RB_RPTR);
1554 }
1555 
1556 /**
1557  * vcn_v4_0_unified_ring_get_wptr - get unified write pointer
1558  *
1559  * @ring: amdgpu_ring pointer
1560  *
1561  * Returns the current hardware unified write pointer
1562  */
1563 static uint64_t vcn_v4_0_unified_ring_get_wptr(struct amdgpu_ring *ring)
1564 {
1565 	struct amdgpu_device *adev = ring->adev;
1566 
1567 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1568 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1569 
1570 	if (ring->use_doorbell)
1571 		return *ring->wptr_cpu_addr;
1572 	else
1573 		return RREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR);
1574 }
1575 
1576 /**
1577  * vcn_v4_0_unified_ring_set_wptr - set enc write pointer
1578  *
1579  * @ring: amdgpu_ring pointer
1580  *
1581  * Commits the enc write pointer to the hardware
1582  */
1583 static void vcn_v4_0_unified_ring_set_wptr(struct amdgpu_ring *ring)
1584 {
1585 	struct amdgpu_device *adev = ring->adev;
1586 
1587 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1588 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1589 
1590 	if (ring->use_doorbell) {
1591 		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1592 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1593 	} else {
1594 		WREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR, lower_32_bits(ring->wptr));
1595 	}
1596 }
1597 
1598 static int vcn_v4_0_limit_sched(struct amdgpu_cs_parser *p,
1599 				struct amdgpu_job *job)
1600 {
1601 	struct drm_gpu_scheduler **scheds;
1602 
1603 	/* The create msg must be in the first IB submitted */
1604 	if (atomic_read(&job->base.entity->fence_seq))
1605 		return -EINVAL;
1606 
1607 	scheds = p->adev->gpu_sched[AMDGPU_HW_IP_VCN_ENC]
1608 		[AMDGPU_RING_PRIO_0].sched;
1609 	drm_sched_entity_modify_sched(job->base.entity, scheds, 1);
1610 	return 0;
1611 }
1612 
1613 static int vcn_v4_0_dec_msg(struct amdgpu_cs_parser *p, struct amdgpu_job *job,
1614 			    uint64_t addr)
1615 {
1616 	struct ttm_operation_ctx ctx = { false, false };
1617 	struct amdgpu_bo_va_mapping *map;
1618 	uint32_t *msg, num_buffers;
1619 	struct amdgpu_bo *bo;
1620 	uint64_t start, end;
1621 	unsigned int i;
1622 	void *ptr;
1623 	int r;
1624 
1625 	addr &= AMDGPU_GMC_HOLE_MASK;
1626 	r = amdgpu_cs_find_mapping(p, addr, &bo, &map);
1627 	if (r) {
1628 		DRM_ERROR("Can't find BO for addr 0x%08llx\n", addr);
1629 		return r;
1630 	}
1631 
1632 	start = map->start * AMDGPU_GPU_PAGE_SIZE;
1633 	end = (map->last + 1) * AMDGPU_GPU_PAGE_SIZE;
1634 	if (addr & 0x7) {
1635 		DRM_ERROR("VCN messages must be 8 byte aligned!\n");
1636 		return -EINVAL;
1637 	}
1638 
1639 	bo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
1640 	amdgpu_bo_placement_from_domain(bo, bo->allowed_domains);
1641 	r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
1642 	if (r) {
1643 		DRM_ERROR("Failed validating the VCN message BO (%d)!\n", r);
1644 		return r;
1645 	}
1646 
1647 	r = amdgpu_bo_kmap(bo, &ptr);
1648 	if (r) {
1649 		DRM_ERROR("Failed mapping the VCN message (%d)!\n", r);
1650 		return r;
1651 	}
1652 
1653 	msg = ptr + addr - start;
1654 
1655 	/* Check length */
1656 	if (msg[1] > end - addr) {
1657 		r = -EINVAL;
1658 		goto out;
1659 	}
1660 
1661 	if (msg[3] != RDECODE_MSG_CREATE)
1662 		goto out;
1663 
1664 	num_buffers = msg[2];
1665 	for (i = 0, msg = &msg[6]; i < num_buffers; ++i, msg += 4) {
1666 		uint32_t offset, size, *create;
1667 
1668 		if (msg[0] != RDECODE_MESSAGE_CREATE)
1669 			continue;
1670 
1671 		offset = msg[1];
1672 		size = msg[2];
1673 
1674 		if (offset + size > end) {
1675 			r = -EINVAL;
1676 			goto out;
1677 		}
1678 
1679 		create = ptr + addr + offset - start;
1680 
1681 		/* H246, HEVC and VP9 can run on any instance */
1682 		if (create[0] == 0x7 || create[0] == 0x10 || create[0] == 0x11)
1683 			continue;
1684 
1685 		r = vcn_v4_0_limit_sched(p, job);
1686 		if (r)
1687 			goto out;
1688 	}
1689 
1690 out:
1691 	amdgpu_bo_kunmap(bo);
1692 	return r;
1693 }
1694 
1695 #define RADEON_VCN_ENGINE_TYPE_DECODE                                 (0x00000003)
1696 
1697 static int vcn_v4_0_ring_patch_cs_in_place(struct amdgpu_cs_parser *p,
1698 					   struct amdgpu_job *job,
1699 					   struct amdgpu_ib *ib)
1700 {
1701 	struct amdgpu_ring *ring = amdgpu_job_ring(job);
1702 	struct amdgpu_vcn_decode_buffer *decode_buffer;
1703 	uint64_t addr;
1704 	uint32_t val;
1705 
1706 	/* The first instance can decode anything */
1707 	if (!ring->me)
1708 		return 0;
1709 
1710 	/* unified queue ib header has 8 double words. */
1711 	if (ib->length_dw < 8)
1712 		return 0;
1713 
1714 	val = amdgpu_ib_get_value(ib, 6); //RADEON_VCN_ENGINE_TYPE
1715 	if (val != RADEON_VCN_ENGINE_TYPE_DECODE)
1716 		return 0;
1717 
1718 	decode_buffer = (struct amdgpu_vcn_decode_buffer *)&ib->ptr[10];
1719 
1720 	if (!(decode_buffer->valid_buf_flag  & 0x1))
1721 		return 0;
1722 
1723 	addr = ((u64)decode_buffer->msg_buffer_address_hi) << 32 |
1724 		decode_buffer->msg_buffer_address_lo;
1725 	return vcn_v4_0_dec_msg(p, job, addr);
1726 }
1727 
1728 static const struct amdgpu_ring_funcs vcn_v4_0_unified_ring_vm_funcs = {
1729 	.type = AMDGPU_RING_TYPE_VCN_ENC,
1730 	.align_mask = 0x3f,
1731 	.nop = VCN_ENC_CMD_NO_OP,
1732 	.vmhub = AMDGPU_MMHUB_0,
1733 	.get_rptr = vcn_v4_0_unified_ring_get_rptr,
1734 	.get_wptr = vcn_v4_0_unified_ring_get_wptr,
1735 	.set_wptr = vcn_v4_0_unified_ring_set_wptr,
1736 	.patch_cs_in_place = vcn_v4_0_ring_patch_cs_in_place,
1737 	.emit_frame_size =
1738 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1739 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1740 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1741 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1742 		1, /* vcn_v2_0_enc_ring_insert_end */
1743 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1744 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
1745 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
1746 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1747 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
1748 	.test_ib = amdgpu_vcn_unified_ring_test_ib,
1749 	.insert_nop = amdgpu_ring_insert_nop,
1750 	.insert_end = vcn_v2_0_enc_ring_insert_end,
1751 	.pad_ib = amdgpu_ring_generic_pad_ib,
1752 	.begin_use = amdgpu_vcn_ring_begin_use,
1753 	.end_use = amdgpu_vcn_ring_end_use,
1754 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1755 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1756 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1757 };
1758 
1759 /**
1760  * vcn_v4_0_set_unified_ring_funcs - set unified ring functions
1761  *
1762  * @adev: amdgpu_device pointer
1763  *
1764  * Set unified ring functions
1765  */
1766 static void vcn_v4_0_set_unified_ring_funcs(struct amdgpu_device *adev)
1767 {
1768 	int i;
1769 
1770 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1771 		if (adev->vcn.harvest_config & (1 << i))
1772 			continue;
1773 
1774 		adev->vcn.inst[i].ring_enc[0].funcs = &vcn_v4_0_unified_ring_vm_funcs;
1775 		adev->vcn.inst[i].ring_enc[0].me = i;
1776 
1777 		DRM_INFO("VCN(%d) encode/decode are enabled in VM mode\n", i);
1778 	}
1779 }
1780 
1781 /**
1782  * vcn_v4_0_is_idle - check VCN block is idle
1783  *
1784  * @handle: amdgpu_device pointer
1785  *
1786  * Check whether VCN block is idle
1787  */
1788 static bool vcn_v4_0_is_idle(void *handle)
1789 {
1790 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1791 	int i, ret = 1;
1792 
1793 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1794 		if (adev->vcn.harvest_config & (1 << i))
1795 			continue;
1796 
1797 		ret &= (RREG32_SOC15(VCN, i, regUVD_STATUS) == UVD_STATUS__IDLE);
1798 	}
1799 
1800 	return ret;
1801 }
1802 
1803 /**
1804  * vcn_v4_0_wait_for_idle - wait for VCN block idle
1805  *
1806  * @handle: amdgpu_device pointer
1807  *
1808  * Wait for VCN block idle
1809  */
1810 static int vcn_v4_0_wait_for_idle(void *handle)
1811 {
1812 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1813 	int i, ret = 0;
1814 
1815 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1816 		if (adev->vcn.harvest_config & (1 << i))
1817 			continue;
1818 
1819 		ret = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE,
1820 			UVD_STATUS__IDLE);
1821 		if (ret)
1822 			return ret;
1823 	}
1824 
1825 	return ret;
1826 }
1827 
1828 /**
1829  * vcn_v4_0_set_clockgating_state - set VCN block clockgating state
1830  *
1831  * @handle: amdgpu_device pointer
1832  * @state: clock gating state
1833  *
1834  * Set VCN block clockgating state
1835  */
1836 static int vcn_v4_0_set_clockgating_state(void *handle, enum amd_clockgating_state state)
1837 {
1838 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1839 	bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
1840 	int i;
1841 
1842 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1843 		if (adev->vcn.harvest_config & (1 << i))
1844 			continue;
1845 
1846 		if (enable) {
1847 			if (RREG32_SOC15(VCN, i, regUVD_STATUS) != UVD_STATUS__IDLE)
1848 				return -EBUSY;
1849 			vcn_v4_0_enable_clock_gating(adev, i);
1850 		} else {
1851 			vcn_v4_0_disable_clock_gating(adev, i);
1852 		}
1853 	}
1854 
1855 	return 0;
1856 }
1857 
1858 /**
1859  * vcn_v4_0_set_powergating_state - set VCN block powergating state
1860  *
1861  * @handle: amdgpu_device pointer
1862  * @state: power gating state
1863  *
1864  * Set VCN block powergating state
1865  */
1866 static int vcn_v4_0_set_powergating_state(void *handle, enum amd_powergating_state state)
1867 {
1868 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1869 	int ret;
1870 
1871 	/* for SRIOV, guest should not control VCN Power-gating
1872 	 * MMSCH FW should control Power-gating and clock-gating
1873 	 * guest should avoid touching CGC and PG
1874 	 */
1875 	if (amdgpu_sriov_vf(adev)) {
1876 		adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
1877 		return 0;
1878 	}
1879 
1880 	if(state == adev->vcn.cur_state)
1881 		return 0;
1882 
1883 	if (state == AMD_PG_STATE_GATE)
1884 		ret = vcn_v4_0_stop(adev);
1885 	else
1886 		ret = vcn_v4_0_start(adev);
1887 
1888 	if(!ret)
1889 		adev->vcn.cur_state = state;
1890 
1891 	return ret;
1892 }
1893 
1894 /**
1895  * vcn_v4_0_set_interrupt_state - set VCN block interrupt state
1896  *
1897  * @adev: amdgpu_device pointer
1898  * @source: interrupt sources
1899  * @type: interrupt types
1900  * @state: interrupt states
1901  *
1902  * Set VCN block interrupt state
1903  */
1904 static int vcn_v4_0_set_interrupt_state(struct amdgpu_device *adev, struct amdgpu_irq_src *source,
1905       unsigned type, enum amdgpu_interrupt_state state)
1906 {
1907 	return 0;
1908 }
1909 
1910 /**
1911  * vcn_v4_0_process_interrupt - process VCN block interrupt
1912  *
1913  * @adev: amdgpu_device pointer
1914  * @source: interrupt sources
1915  * @entry: interrupt entry from clients and sources
1916  *
1917  * Process VCN block interrupt
1918  */
1919 static int vcn_v4_0_process_interrupt(struct amdgpu_device *adev, struct amdgpu_irq_src *source,
1920       struct amdgpu_iv_entry *entry)
1921 {
1922 	uint32_t ip_instance;
1923 
1924 	switch (entry->client_id) {
1925 	case SOC15_IH_CLIENTID_VCN:
1926 		ip_instance = 0;
1927 		break;
1928 	case SOC15_IH_CLIENTID_VCN1:
1929 		ip_instance = 1;
1930 		break;
1931 	default:
1932 		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
1933 		return 0;
1934 	}
1935 
1936 	DRM_DEBUG("IH: VCN TRAP\n");
1937 
1938 	switch (entry->src_id) {
1939 	case VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
1940 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
1941 		break;
1942 	default:
1943 		DRM_ERROR("Unhandled interrupt: %d %d\n",
1944 			  entry->src_id, entry->src_data[0]);
1945 		break;
1946 	}
1947 
1948 	return 0;
1949 }
1950 
1951 static const struct amdgpu_irq_src_funcs vcn_v4_0_irq_funcs = {
1952 	.set = vcn_v4_0_set_interrupt_state,
1953 	.process = vcn_v4_0_process_interrupt,
1954 };
1955 
1956 /**
1957  * vcn_v4_0_set_irq_funcs - set VCN block interrupt irq functions
1958  *
1959  * @adev: amdgpu_device pointer
1960  *
1961  * Set VCN block interrupt irq functions
1962  */
1963 static void vcn_v4_0_set_irq_funcs(struct amdgpu_device *adev)
1964 {
1965 	int i;
1966 
1967 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1968 		if (adev->vcn.harvest_config & (1 << i))
1969 			continue;
1970 
1971 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
1972 		adev->vcn.inst[i].irq.funcs = &vcn_v4_0_irq_funcs;
1973 	}
1974 }
1975 
1976 static const struct amd_ip_funcs vcn_v4_0_ip_funcs = {
1977 	.name = "vcn_v4_0",
1978 	.early_init = vcn_v4_0_early_init,
1979 	.late_init = NULL,
1980 	.sw_init = vcn_v4_0_sw_init,
1981 	.sw_fini = vcn_v4_0_sw_fini,
1982 	.hw_init = vcn_v4_0_hw_init,
1983 	.hw_fini = vcn_v4_0_hw_fini,
1984 	.suspend = vcn_v4_0_suspend,
1985 	.resume = vcn_v4_0_resume,
1986 	.is_idle = vcn_v4_0_is_idle,
1987 	.wait_for_idle = vcn_v4_0_wait_for_idle,
1988 	.check_soft_reset = NULL,
1989 	.pre_soft_reset = NULL,
1990 	.soft_reset = NULL,
1991 	.post_soft_reset = NULL,
1992 	.set_clockgating_state = vcn_v4_0_set_clockgating_state,
1993 	.set_powergating_state = vcn_v4_0_set_powergating_state,
1994 };
1995 
1996 const struct amdgpu_ip_block_version vcn_v4_0_ip_block =
1997 {
1998 	.type = AMD_IP_BLOCK_TYPE_VCN,
1999 	.major = 4,
2000 	.minor = 0,
2001 	.rev = 0,
2002 	.funcs = &vcn_v4_0_ip_funcs,
2003 };
2004