xref: /linux/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c (revision 4b660dbd9ee2059850fd30e0df420ca7a38a1856)
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 static int vcn_v4_0_fw_shared_init(struct amdgpu_device *adev, int inst_idx)
104 {
105 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
106 
107 	fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
108 	fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_FW_SHARED_FLAG_0_UNIFIED_QUEUE);
109 	fw_shared->sq.is_enabled = 1;
110 
111 	fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_SMU_DPM_INTERFACE_FLAG);
112 	fw_shared->smu_dpm_interface.smu_interface_type = (adev->flags & AMD_IS_APU) ?
113 		AMDGPU_VCN_SMU_DPM_INTERFACE_APU : AMDGPU_VCN_SMU_DPM_INTERFACE_DGPU;
114 
115 	if (amdgpu_ip_version(adev, VCN_HWIP, 0) ==
116 	    IP_VERSION(4, 0, 2)) {
117 		fw_shared->present_flag_0 |= AMDGPU_FW_SHARED_FLAG_0_DRM_KEY_INJECT;
118 		fw_shared->drm_key_wa.method =
119 			AMDGPU_DRM_KEY_INJECT_WORKAROUND_VCNFW_ASD_HANDSHAKING;
120 	}
121 
122 	if (amdgpu_vcnfw_log)
123 		amdgpu_vcn_fwlog_init(&adev->vcn.inst[inst_idx]);
124 
125 	return 0;
126 }
127 
128 /**
129  * vcn_v4_0_sw_init - sw init for VCN block
130  *
131  * @handle: amdgpu_device pointer
132  *
133  * Load firmware and sw initialization
134  */
135 static int vcn_v4_0_sw_init(void *handle)
136 {
137 	struct amdgpu_ring *ring;
138 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
139 	int i, r;
140 
141 	r = amdgpu_vcn_sw_init(adev);
142 	if (r)
143 		return r;
144 
145 	amdgpu_vcn_setup_ucode(adev);
146 
147 	r = amdgpu_vcn_resume(adev);
148 	if (r)
149 		return r;
150 
151 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
152 		if (adev->vcn.harvest_config & (1 << i))
153 			continue;
154 
155 		/* Init instance 0 sched_score to 1, so it's scheduled after other instances */
156 		if (i == 0)
157 			atomic_set(&adev->vcn.inst[i].sched_score, 1);
158 		else
159 			atomic_set(&adev->vcn.inst[i].sched_score, 0);
160 
161 		/* VCN UNIFIED TRAP */
162 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
163 				VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);
164 		if (r)
165 			return r;
166 
167 		/* VCN POISON TRAP */
168 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
169 				VCN_4_0__SRCID_UVD_POISON, &adev->vcn.inst[i].ras_poison_irq);
170 		if (r)
171 			return r;
172 
173 		ring = &adev->vcn.inst[i].ring_enc[0];
174 		ring->use_doorbell = true;
175 		if (amdgpu_sriov_vf(adev))
176 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + i * (adev->vcn.num_enc_rings + 1) + 1;
177 		else
178 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + 8 * i;
179 		ring->vm_hub = AMDGPU_MMHUB0(0);
180 		sprintf(ring->name, "vcn_unified_%d", i);
181 
182 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
183 						AMDGPU_RING_PRIO_0, &adev->vcn.inst[i].sched_score);
184 		if (r)
185 			return r;
186 
187 		vcn_v4_0_fw_shared_init(adev, i);
188 	}
189 
190 	if (amdgpu_sriov_vf(adev)) {
191 		r = amdgpu_virt_alloc_mm_table(adev);
192 		if (r)
193 			return r;
194 	}
195 
196 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
197 		adev->vcn.pause_dpg_mode = vcn_v4_0_pause_dpg_mode;
198 
199 	r = amdgpu_vcn_ras_sw_init(adev);
200 	if (r)
201 		return r;
202 
203 	return 0;
204 }
205 
206 /**
207  * vcn_v4_0_sw_fini - sw fini for VCN block
208  *
209  * @handle: amdgpu_device pointer
210  *
211  * VCN suspend and free up sw allocation
212  */
213 static int vcn_v4_0_sw_fini(void *handle)
214 {
215 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
216 	int i, r, idx;
217 
218 	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
219 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
220 			volatile struct amdgpu_vcn4_fw_shared *fw_shared;
221 
222 			if (adev->vcn.harvest_config & (1 << i))
223 				continue;
224 
225 			fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
226 			fw_shared->present_flag_0 = 0;
227 			fw_shared->sq.is_enabled = 0;
228 		}
229 
230 		drm_dev_exit(idx);
231 	}
232 
233 	if (amdgpu_sriov_vf(adev))
234 		amdgpu_virt_free_mm_table(adev);
235 
236 	r = amdgpu_vcn_suspend(adev);
237 	if (r)
238 		return r;
239 
240 	r = amdgpu_vcn_sw_fini(adev);
241 
242 	return r;
243 }
244 
245 /**
246  * vcn_v4_0_hw_init - start and test VCN block
247  *
248  * @handle: amdgpu_device pointer
249  *
250  * Initialize the hardware, boot up the VCPU and do some testing
251  */
252 static int vcn_v4_0_hw_init(void *handle)
253 {
254 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
255 	struct amdgpu_ring *ring;
256 	int i, r;
257 
258 	if (amdgpu_sriov_vf(adev)) {
259 		r = vcn_v4_0_start_sriov(adev);
260 		if (r)
261 			goto done;
262 
263 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
264 			if (adev->vcn.harvest_config & (1 << i))
265 				continue;
266 
267 			ring = &adev->vcn.inst[i].ring_enc[0];
268 			ring->wptr = 0;
269 			ring->wptr_old = 0;
270 			vcn_v4_0_unified_ring_set_wptr(ring);
271 			ring->sched.ready = true;
272 
273 		}
274 	} else {
275 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
276 			if (adev->vcn.harvest_config & (1 << i))
277 				continue;
278 
279 			ring = &adev->vcn.inst[i].ring_enc[0];
280 
281 			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
282 					((adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i), i);
283 
284 			r = amdgpu_ring_test_helper(ring);
285 			if (r)
286 				goto done;
287 
288 		}
289 	}
290 
291 done:
292 	if (!r)
293 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
294 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
295 
296 	return r;
297 }
298 
299 /**
300  * vcn_v4_0_hw_fini - stop the hardware block
301  *
302  * @handle: amdgpu_device pointer
303  *
304  * Stop the VCN block, mark ring as not ready any more
305  */
306 static int vcn_v4_0_hw_fini(void *handle)
307 {
308 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
309 	int i;
310 
311 	cancel_delayed_work_sync(&adev->vcn.idle_work);
312 
313 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
314 		if (adev->vcn.harvest_config & (1 << i))
315 			continue;
316 		if (!amdgpu_sriov_vf(adev)) {
317 			if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
318                         (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
319                                 RREG32_SOC15(VCN, i, regUVD_STATUS))) {
320                         vcn_v4_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
321 			}
322 		}
323 		if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN))
324 			amdgpu_irq_put(adev, &adev->vcn.inst[i].ras_poison_irq, 0);
325 	}
326 
327 	return 0;
328 }
329 
330 /**
331  * vcn_v4_0_suspend - suspend VCN block
332  *
333  * @handle: amdgpu_device pointer
334  *
335  * HW fini and suspend VCN block
336  */
337 static int vcn_v4_0_suspend(void *handle)
338 {
339 	int r;
340 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
341 
342 	r = vcn_v4_0_hw_fini(adev);
343 	if (r)
344 		return r;
345 
346 	r = amdgpu_vcn_suspend(adev);
347 
348 	return r;
349 }
350 
351 /**
352  * vcn_v4_0_resume - resume VCN block
353  *
354  * @handle: amdgpu_device pointer
355  *
356  * Resume firmware and hw init VCN block
357  */
358 static int vcn_v4_0_resume(void *handle)
359 {
360 	int r;
361 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
362 
363 	r = amdgpu_vcn_resume(adev);
364 	if (r)
365 		return r;
366 
367 	r = vcn_v4_0_hw_init(adev);
368 
369 	return r;
370 }
371 
372 /**
373  * vcn_v4_0_mc_resume - memory controller programming
374  *
375  * @adev: amdgpu_device pointer
376  * @inst: instance number
377  *
378  * Let the VCN memory controller know it's offsets
379  */
380 static void vcn_v4_0_mc_resume(struct amdgpu_device *adev, int inst)
381 {
382 	uint32_t offset, size;
383 	const struct common_firmware_header *hdr;
384 
385 	hdr = (const struct common_firmware_header *)adev->vcn.fw[inst]->data;
386 	size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
387 
388 	/* cache window 0: fw */
389 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
390 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
391 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo));
392 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
393 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi));
394 		WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, 0);
395 		offset = 0;
396 	} else {
397 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
398 			lower_32_bits(adev->vcn.inst[inst].gpu_addr));
399 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
400 			upper_32_bits(adev->vcn.inst[inst].gpu_addr));
401 		offset = size;
402                 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
403 	}
404 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE0, size);
405 
406 	/* cache window 1: stack */
407 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
408 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
409 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
410 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
411 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET1, 0);
412 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
413 
414 	/* cache window 2: context */
415 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
416 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
417 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
418 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
419 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET2, 0);
420 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
421 
422 	/* non-cache window */
423 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
424 		lower_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
425 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
426 		upper_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
427 	WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_OFFSET0, 0);
428 	WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_SIZE0,
429 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
430 }
431 
432 /**
433  * vcn_v4_0_mc_resume_dpg_mode - memory controller programming for dpg mode
434  *
435  * @adev: amdgpu_device pointer
436  * @inst_idx: instance number index
437  * @indirect: indirectly write sram
438  *
439  * Let the VCN memory controller know it's offsets with dpg mode
440  */
441 static void vcn_v4_0_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
442 {
443 	uint32_t offset, size;
444 	const struct common_firmware_header *hdr;
445 	hdr = (const struct common_firmware_header *)adev->vcn.fw[inst_idx]->data;
446 	size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
447 
448 	/* cache window 0: fw */
449 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
450 		if (!indirect) {
451 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
452 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
453 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
454 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
455 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
456 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
457 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
458 				VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
459 		} else {
460 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
461 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
462 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
463 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
464 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
465 				VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
466 		}
467 		offset = 0;
468 	} else {
469 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
470 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
471 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
472 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
473 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
474 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
475 		offset = size;
476 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
477 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0),
478 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
479 	}
480 
481 	if (!indirect)
482 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
483 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
484 	else
485 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
486 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
487 
488 	/* cache window 1: stack */
489 	if (!indirect) {
490 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
491 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
492 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
493 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
494 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
495 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
496 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
497 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
498 	} else {
499 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
500 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
501 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
502 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
503 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
504 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
505 	}
506 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
507 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
508 
509 	/* cache window 2: context */
510 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
511 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
512 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
513 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
514 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
515 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
516 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
517 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
518 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
519 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
520 
521 	/* non-cache window */
522 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
523 			VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
524 			lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
525 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
526 			VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
527 			upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
528 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
529 			VCN, inst_idx, regUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
530 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
531 			VCN, inst_idx, regUVD_VCPU_NONCACHE_SIZE0),
532 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)), 0, indirect);
533 
534 	/* VCN global tiling registers */
535 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
536 		VCN, 0, regUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
537 }
538 
539 /**
540  * vcn_v4_0_disable_static_power_gating - disable VCN static power gating
541  *
542  * @adev: amdgpu_device pointer
543  * @inst: instance number
544  *
545  * Disable static power gating for VCN block
546  */
547 static void vcn_v4_0_disable_static_power_gating(struct amdgpu_device *adev, int inst)
548 {
549 	uint32_t data = 0;
550 
551 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
552 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
553 			| 1 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
554 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
555 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
556 			| 2 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
557 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
558 			| 2 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
559 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
560 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
561 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
562 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
563 			| 2 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
564 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
565 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
566 
567 		WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
568 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS,
569 			UVD_PGFSM_STATUS__UVDM_UVDU_UVDLM_PWR_ON_3_0, 0x3F3FFFFF);
570 	} else {
571 		uint32_t value;
572 
573 		value = (inst) ? 0x2200800 : 0;
574 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
575 			| 1 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
576 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
577 			| 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
578 			| 1 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
579 			| 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
580 			| 1 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
581 			| 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
582 			| 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
583 			| 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
584 			| 1 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
585 			| 1 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
586 			| 1 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
587 			| 1 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
588 
589                 WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
590                 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS, value,  0x3F3FFFFF);
591         }
592 
593         data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
594         data &= ~0x103;
595         if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
596                 data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
597                         UVD_POWER_STATUS__UVD_PG_EN_MASK;
598 
599         WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
600 
601         return;
602 }
603 
604 /**
605  * vcn_v4_0_enable_static_power_gating - enable VCN static power gating
606  *
607  * @adev: amdgpu_device pointer
608  * @inst: instance number
609  *
610  * Enable static power gating for VCN block
611  */
612 static void vcn_v4_0_enable_static_power_gating(struct amdgpu_device *adev, int inst)
613 {
614 	uint32_t data;
615 
616 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
617 		/* Before power off, this indicator has to be turned on */
618 		data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
619 		data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
620 		data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
621 		WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
622 
623 		data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
624 			| 2 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
625 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
626 			| 2 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
627 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
628 			| 2 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
629 			| 2 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
630 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
631 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
632 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
633 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
634 			| 2 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
635 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
636 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
637 		WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
638 
639 		data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
640 			| 2 << UVD_PGFSM_STATUS__UVDS_PWR_STATUS__SHIFT
641 			| 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
642 			| 2 << UVD_PGFSM_STATUS__UVDTC_PWR_STATUS__SHIFT
643 			| 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
644 			| 2 << UVD_PGFSM_STATUS__UVDTA_PWR_STATUS__SHIFT
645 			| 2 << UVD_PGFSM_STATUS__UVDLM_PWR_STATUS__SHIFT
646 			| 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
647 			| 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
648 			| 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT
649 			| 2 << UVD_PGFSM_STATUS__UVDAB_PWR_STATUS__SHIFT
650 			| 2 << UVD_PGFSM_STATUS__UVDTB_PWR_STATUS__SHIFT
651 			| 2 << UVD_PGFSM_STATUS__UVDNA_PWR_STATUS__SHIFT
652 			| 2 << UVD_PGFSM_STATUS__UVDNB_PWR_STATUS__SHIFT);
653 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS, data, 0x3F3FFFFF);
654 	}
655 
656         return;
657 }
658 
659 /**
660  * vcn_v4_0_disable_clock_gating - disable VCN clock gating
661  *
662  * @adev: amdgpu_device pointer
663  * @inst: instance number
664  *
665  * Disable clock gating for VCN block
666  */
667 static void vcn_v4_0_disable_clock_gating(struct amdgpu_device *adev, int inst)
668 {
669 	uint32_t data;
670 
671 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
672 		return;
673 
674 	/* VCN disable CGC */
675 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
676 	data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
677 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
678 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
679 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
680 
681 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_GATE);
682 	data &= ~(UVD_CGC_GATE__SYS_MASK
683 		| UVD_CGC_GATE__UDEC_MASK
684 		| UVD_CGC_GATE__MPEG2_MASK
685 		| UVD_CGC_GATE__REGS_MASK
686 		| UVD_CGC_GATE__RBC_MASK
687 		| UVD_CGC_GATE__LMI_MC_MASK
688 		| UVD_CGC_GATE__LMI_UMC_MASK
689 		| UVD_CGC_GATE__IDCT_MASK
690 		| UVD_CGC_GATE__MPRD_MASK
691 		| UVD_CGC_GATE__MPC_MASK
692 		| UVD_CGC_GATE__LBSI_MASK
693 		| UVD_CGC_GATE__LRBBM_MASK
694 		| UVD_CGC_GATE__UDEC_RE_MASK
695 		| UVD_CGC_GATE__UDEC_CM_MASK
696 		| UVD_CGC_GATE__UDEC_IT_MASK
697 		| UVD_CGC_GATE__UDEC_DB_MASK
698 		| UVD_CGC_GATE__UDEC_MP_MASK
699 		| UVD_CGC_GATE__WCB_MASK
700 		| UVD_CGC_GATE__VCPU_MASK
701 		| UVD_CGC_GATE__MMSCH_MASK);
702 
703 	WREG32_SOC15(VCN, inst, regUVD_CGC_GATE, data);
704 	SOC15_WAIT_ON_RREG(VCN, inst, regUVD_CGC_GATE, 0,  0xFFFFFFFF);
705 
706 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
707 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
708 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
709 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
710 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
711 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
712 		| UVD_CGC_CTRL__SYS_MODE_MASK
713 		| UVD_CGC_CTRL__UDEC_MODE_MASK
714 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
715 		| UVD_CGC_CTRL__REGS_MODE_MASK
716 		| UVD_CGC_CTRL__RBC_MODE_MASK
717 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
718 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
719 		| UVD_CGC_CTRL__IDCT_MODE_MASK
720 		| UVD_CGC_CTRL__MPRD_MODE_MASK
721 		| UVD_CGC_CTRL__MPC_MODE_MASK
722 		| UVD_CGC_CTRL__LBSI_MODE_MASK
723 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
724 		| UVD_CGC_CTRL__WCB_MODE_MASK
725 		| UVD_CGC_CTRL__VCPU_MODE_MASK
726 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
727 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
728 
729 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE);
730 	data |= (UVD_SUVD_CGC_GATE__SRE_MASK
731 		| UVD_SUVD_CGC_GATE__SIT_MASK
732 		| UVD_SUVD_CGC_GATE__SMP_MASK
733 		| UVD_SUVD_CGC_GATE__SCM_MASK
734 		| UVD_SUVD_CGC_GATE__SDB_MASK
735 		| UVD_SUVD_CGC_GATE__SRE_H264_MASK
736 		| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
737 		| UVD_SUVD_CGC_GATE__SIT_H264_MASK
738 		| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
739 		| UVD_SUVD_CGC_GATE__SCM_H264_MASK
740 		| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
741 		| UVD_SUVD_CGC_GATE__SDB_H264_MASK
742 		| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
743 		| UVD_SUVD_CGC_GATE__SCLR_MASK
744 		| UVD_SUVD_CGC_GATE__UVD_SC_MASK
745 		| UVD_SUVD_CGC_GATE__ENT_MASK
746 		| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
747 		| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
748 		| UVD_SUVD_CGC_GATE__SITE_MASK
749 		| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
750 		| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
751 		| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
752 		| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
753 		| UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
754 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE, data);
755 
756 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
757 	data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
758 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
759 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
760 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
761 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
762 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
763 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
764 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
765 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
766 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
767 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
768 }
769 
770 /**
771  * vcn_v4_0_disable_clock_gating_dpg_mode - disable VCN clock gating dpg mode
772  *
773  * @adev: amdgpu_device pointer
774  * @sram_sel: sram select
775  * @inst_idx: instance number index
776  * @indirect: indirectly write sram
777  *
778  * Disable clock gating for VCN block with dpg mode
779  */
780 static void vcn_v4_0_disable_clock_gating_dpg_mode(struct amdgpu_device *adev, uint8_t sram_sel,
781       int inst_idx, uint8_t indirect)
782 {
783 	uint32_t reg_data = 0;
784 
785 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
786 		return;
787 
788 	/* enable sw clock gating control */
789 	reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
790 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
791 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
792 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
793 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
794 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
795 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
796 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
797 		 UVD_CGC_CTRL__SYS_MODE_MASK |
798 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
799 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
800 		 UVD_CGC_CTRL__REGS_MODE_MASK |
801 		 UVD_CGC_CTRL__RBC_MODE_MASK |
802 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
803 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
804 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
805 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
806 		 UVD_CGC_CTRL__MPC_MODE_MASK |
807 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
808 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
809 		 UVD_CGC_CTRL__WCB_MODE_MASK |
810 		 UVD_CGC_CTRL__VCPU_MODE_MASK);
811 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
812 		VCN, inst_idx, regUVD_CGC_CTRL), reg_data, sram_sel, indirect);
813 
814 	/* turn off clock gating */
815 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
816 		VCN, inst_idx, regUVD_CGC_GATE), 0, sram_sel, indirect);
817 
818 	/* turn on SUVD clock gating */
819 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
820 		VCN, inst_idx, regUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
821 
822 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
823 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
824 		VCN, inst_idx, regUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
825 }
826 
827 /**
828  * vcn_v4_0_enable_clock_gating - enable VCN clock gating
829  *
830  * @adev: amdgpu_device pointer
831  * @inst: instance number
832  *
833  * Enable clock gating for VCN block
834  */
835 static void vcn_v4_0_enable_clock_gating(struct amdgpu_device *adev, int inst)
836 {
837 	uint32_t data;
838 
839 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
840 		return;
841 
842 	/* enable VCN CGC */
843 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
844 	data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
845 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
846 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
847 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
848 
849 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
850 	data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
851 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
852 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
853 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
854 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
855 		| UVD_CGC_CTRL__SYS_MODE_MASK
856 		| UVD_CGC_CTRL__UDEC_MODE_MASK
857 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
858 		| UVD_CGC_CTRL__REGS_MODE_MASK
859 		| UVD_CGC_CTRL__RBC_MODE_MASK
860 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
861 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
862 		| UVD_CGC_CTRL__IDCT_MODE_MASK
863 		| UVD_CGC_CTRL__MPRD_MODE_MASK
864 		| UVD_CGC_CTRL__MPC_MODE_MASK
865 		| UVD_CGC_CTRL__LBSI_MODE_MASK
866 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
867 		| UVD_CGC_CTRL__WCB_MODE_MASK
868 		| UVD_CGC_CTRL__VCPU_MODE_MASK
869 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
870 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
871 
872 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
873 	data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
874 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
875 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
876 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
877 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
878 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
879 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
880 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
881 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
882 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
883 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
884 }
885 
886 static void vcn_v4_0_enable_ras(struct amdgpu_device *adev, int inst_idx,
887 				bool indirect)
888 {
889 	uint32_t tmp;
890 
891 	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN))
892 		return;
893 
894 	tmp = VCN_RAS_CNTL__VCPU_VCODEC_REARM_MASK |
895 	      VCN_RAS_CNTL__VCPU_VCODEC_IH_EN_MASK |
896 	      VCN_RAS_CNTL__VCPU_VCODEC_PMI_EN_MASK |
897 	      VCN_RAS_CNTL__VCPU_VCODEC_STALL_EN_MASK;
898 	WREG32_SOC15_DPG_MODE(inst_idx,
899 			      SOC15_DPG_MODE_OFFSET(VCN, 0, regVCN_RAS_CNTL),
900 			      tmp, 0, indirect);
901 
902 	tmp = UVD_SYS_INT_EN__RASCNTL_VCPU_VCODEC_EN_MASK;
903 	WREG32_SOC15_DPG_MODE(inst_idx,
904 			      SOC15_DPG_MODE_OFFSET(VCN, 0, regUVD_SYS_INT_EN),
905 			      tmp, 0, indirect);
906 }
907 
908 /**
909  * vcn_v4_0_start_dpg_mode - VCN start with dpg mode
910  *
911  * @adev: amdgpu_device pointer
912  * @inst_idx: instance number index
913  * @indirect: indirectly write sram
914  *
915  * Start VCN block with dpg mode
916  */
917 static int vcn_v4_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
918 {
919 	volatile struct amdgpu_vcn4_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
920 	struct amdgpu_ring *ring;
921 	uint32_t tmp;
922 
923 	/* disable register anti-hang mechanism */
924 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 1,
925 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
926 	/* enable dynamic power gating mode */
927 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS);
928 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
929 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
930 	WREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS, tmp);
931 
932 	if (indirect)
933 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
934 
935 	/* enable clock gating */
936 	vcn_v4_0_disable_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
937 
938 	/* enable VCPU clock */
939 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
940 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK | UVD_VCPU_CNTL__BLK_RST_MASK;
941 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
942 		VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
943 
944 	/* disable master interupt */
945 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
946 		VCN, inst_idx, regUVD_MASTINT_EN), 0, 0, indirect);
947 
948 	/* setup regUVD_LMI_CTRL */
949 	tmp = (UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
950 		UVD_LMI_CTRL__REQ_MODE_MASK |
951 		UVD_LMI_CTRL__CRC_RESET_MASK |
952 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
953 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
954 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
955 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
956 		0x00100000L);
957 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
958 		VCN, inst_idx, regUVD_LMI_CTRL), tmp, 0, indirect);
959 
960 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
961 		VCN, inst_idx, regUVD_MPC_CNTL),
962 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
963 
964 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
965 		VCN, inst_idx, regUVD_MPC_SET_MUXA0),
966 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
967 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
968 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
969 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
970 
971 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
972 		VCN, inst_idx, regUVD_MPC_SET_MUXB0),
973 		 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
974 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
975 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
976 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
977 
978 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
979 		VCN, inst_idx, regUVD_MPC_SET_MUX),
980 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
981 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
982 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
983 
984 	vcn_v4_0_mc_resume_dpg_mode(adev, inst_idx, indirect);
985 
986 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
987 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
988 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
989 		VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
990 
991 	/* enable LMI MC and UMC channels */
992 	tmp = 0x1f << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT;
993 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
994 		VCN, inst_idx, regUVD_LMI_CTRL2), tmp, 0, indirect);
995 
996 	vcn_v4_0_enable_ras(adev, inst_idx, indirect);
997 
998 	/* enable master interrupt */
999 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1000 		VCN, inst_idx, regUVD_MASTINT_EN),
1001 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
1002 
1003 
1004 	if (indirect)
1005 		amdgpu_vcn_psp_update_sram(adev, inst_idx, 0);
1006 
1007 	ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1008 
1009 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_LO, ring->gpu_addr);
1010 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1011 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_SIZE, ring->ring_size / 4);
1012 
1013 	tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
1014 	tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
1015 	WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
1016 	fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
1017 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR, 0);
1018 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, 0);
1019 
1020 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR);
1021 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, tmp);
1022 	ring->wptr = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
1023 
1024 	tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
1025 	tmp |= VCN_RB_ENABLE__RB1_EN_MASK;
1026 	WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
1027 	fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
1028 
1029 	WREG32_SOC15(VCN, inst_idx, regVCN_RB1_DB_CTRL,
1030 			ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
1031 			VCN_RB1_DB_CTRL__EN_MASK);
1032 
1033 	return 0;
1034 }
1035 
1036 
1037 /**
1038  * vcn_v4_0_start - VCN start
1039  *
1040  * @adev: amdgpu_device pointer
1041  *
1042  * Start VCN block
1043  */
1044 static int vcn_v4_0_start(struct amdgpu_device *adev)
1045 {
1046 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1047 	struct amdgpu_ring *ring;
1048 	uint32_t tmp;
1049 	int i, j, k, r;
1050 
1051 	if (adev->pm.dpm_enabled)
1052 		amdgpu_dpm_enable_uvd(adev, true);
1053 
1054 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1055 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1056 
1057 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1058 			r = vcn_v4_0_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
1059 			continue;
1060 		}
1061 
1062 		/* disable VCN power gating */
1063 		vcn_v4_0_disable_static_power_gating(adev, i);
1064 
1065 		/* set VCN status busy */
1066 		tmp = RREG32_SOC15(VCN, i, regUVD_STATUS) | UVD_STATUS__UVD_BUSY;
1067 		WREG32_SOC15(VCN, i, regUVD_STATUS, tmp);
1068 
1069 		/*SW clock gating */
1070 		vcn_v4_0_disable_clock_gating(adev, i);
1071 
1072 		/* enable VCPU clock */
1073 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1074 				UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1075 
1076 		/* disable master interrupt */
1077 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN), 0,
1078 				~UVD_MASTINT_EN__VCPU_EN_MASK);
1079 
1080 		/* enable LMI MC and UMC channels */
1081 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_LMI_CTRL2), 0,
1082 				~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1083 
1084 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1085 		tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1086 		tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1087 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1088 
1089 		/* setup regUVD_LMI_CTRL */
1090 		tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL);
1091 		WREG32_SOC15(VCN, i, regUVD_LMI_CTRL, tmp |
1092 				UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
1093 				UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1094 				UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1095 				UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1096 
1097 		/* setup regUVD_MPC_CNTL */
1098 		tmp = RREG32_SOC15(VCN, i, regUVD_MPC_CNTL);
1099 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1100 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1101 		WREG32_SOC15(VCN, i, regUVD_MPC_CNTL, tmp);
1102 
1103 		/* setup UVD_MPC_SET_MUXA0 */
1104 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXA0,
1105 				((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1106 				 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1107 				 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1108 				 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1109 
1110 		/* setup UVD_MPC_SET_MUXB0 */
1111 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXB0,
1112 				((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1113 				 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1114 				 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1115 				 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1116 
1117 		/* setup UVD_MPC_SET_MUX */
1118 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUX,
1119 				((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1120 				 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1121 				 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1122 
1123 		vcn_v4_0_mc_resume(adev, i);
1124 
1125 		/* VCN global tiling registers */
1126 		WREG32_SOC15(VCN, i, regUVD_GFX10_ADDR_CONFIG,
1127 				adev->gfx.config.gb_addr_config);
1128 
1129 		/* unblock VCPU register access */
1130 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL), 0,
1131 				~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1132 
1133 		/* release VCPU reset to boot */
1134 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1135 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1136 
1137 		for (j = 0; j < 10; ++j) {
1138 			uint32_t status;
1139 
1140 			for (k = 0; k < 100; ++k) {
1141 				status = RREG32_SOC15(VCN, i, regUVD_STATUS);
1142 				if (status & 2)
1143 					break;
1144 				mdelay(10);
1145 				if (amdgpu_emu_mode == 1)
1146 					msleep(1);
1147 			}
1148 
1149 			if (amdgpu_emu_mode == 1) {
1150 				r = -1;
1151 				if (status & 2) {
1152 					r = 0;
1153 					break;
1154 				}
1155 			} else {
1156 				r = 0;
1157 				if (status & 2)
1158 					break;
1159 
1160 				dev_err(adev->dev, "VCN[%d] is not responding, trying to reset the VCPU!!!\n", i);
1161 				WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1162 							UVD_VCPU_CNTL__BLK_RST_MASK,
1163 							~UVD_VCPU_CNTL__BLK_RST_MASK);
1164 				mdelay(10);
1165 				WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1166 						~UVD_VCPU_CNTL__BLK_RST_MASK);
1167 
1168 				mdelay(10);
1169 				r = -1;
1170 			}
1171 		}
1172 
1173 		if (r) {
1174 			dev_err(adev->dev, "VCN[%d] is not responding, giving up!!!\n", i);
1175 			return r;
1176 		}
1177 
1178 		/* enable master interrupt */
1179 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN),
1180 				UVD_MASTINT_EN__VCPU_EN_MASK,
1181 				~UVD_MASTINT_EN__VCPU_EN_MASK);
1182 
1183 		/* clear the busy bit of VCN_STATUS */
1184 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_STATUS), 0,
1185 				~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1186 
1187 		ring = &adev->vcn.inst[i].ring_enc[0];
1188 		WREG32_SOC15(VCN, i, regVCN_RB1_DB_CTRL,
1189 				ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
1190 				VCN_RB1_DB_CTRL__EN_MASK);
1191 
1192 		WREG32_SOC15(VCN, i, regUVD_RB_BASE_LO, ring->gpu_addr);
1193 		WREG32_SOC15(VCN, i, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1194 		WREG32_SOC15(VCN, i, regUVD_RB_SIZE, ring->ring_size / 4);
1195 
1196 		tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1197 		tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
1198 		WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1199 		fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
1200 		WREG32_SOC15(VCN, i, regUVD_RB_RPTR, 0);
1201 		WREG32_SOC15(VCN, i, regUVD_RB_WPTR, 0);
1202 
1203 		tmp = RREG32_SOC15(VCN, i, regUVD_RB_RPTR);
1204 		WREG32_SOC15(VCN, i, regUVD_RB_WPTR, tmp);
1205 		ring->wptr = RREG32_SOC15(VCN, i, regUVD_RB_WPTR);
1206 
1207 		tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1208 		tmp |= VCN_RB_ENABLE__RB1_EN_MASK;
1209 		WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1210 		fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
1211 	}
1212 
1213 	return 0;
1214 }
1215 
1216 static int vcn_v4_0_init_ring_metadata(struct amdgpu_device *adev, uint32_t vcn_inst, struct amdgpu_ring *ring_enc)
1217 {
1218 	struct amdgpu_vcn_rb_metadata *rb_metadata = NULL;
1219 	uint8_t *rb_ptr = (uint8_t *)ring_enc->ring;
1220 
1221 	rb_ptr += ring_enc->ring_size;
1222 	rb_metadata = (struct amdgpu_vcn_rb_metadata *)rb_ptr;
1223 
1224 	memset(rb_metadata, 0, sizeof(struct amdgpu_vcn_rb_metadata));
1225 	rb_metadata->size = sizeof(struct amdgpu_vcn_rb_metadata);
1226 	rb_metadata->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
1227 	rb_metadata->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_DECOUPLE_FLAG);
1228 	rb_metadata->version = 1;
1229 	rb_metadata->ring_id = vcn_inst & 0xFF;
1230 
1231 	return 0;
1232 }
1233 
1234 static int vcn_v4_0_start_sriov(struct amdgpu_device *adev)
1235 {
1236 	int i;
1237 	struct amdgpu_ring *ring_enc;
1238 	uint64_t cache_addr;
1239 	uint64_t rb_enc_addr;
1240 	uint64_t ctx_addr;
1241 	uint32_t param, resp, expected;
1242 	uint32_t offset, cache_size;
1243 	uint32_t tmp, timeout;
1244 
1245 	struct amdgpu_mm_table *table = &adev->virt.mm_table;
1246 	uint32_t *table_loc;
1247 	uint32_t table_size;
1248 	uint32_t size, size_dw;
1249 	uint32_t init_status;
1250 	uint32_t enabled_vcn;
1251 
1252 	struct mmsch_v4_0_cmd_direct_write
1253 		direct_wt = { {0} };
1254 	struct mmsch_v4_0_cmd_direct_read_modify_write
1255 		direct_rd_mod_wt = { {0} };
1256 	struct mmsch_v4_0_cmd_end end = { {0} };
1257 	struct mmsch_v4_0_init_header header;
1258 
1259 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1260 	volatile struct amdgpu_fw_shared_rb_setup *rb_setup;
1261 
1262 	direct_wt.cmd_header.command_type =
1263 		MMSCH_COMMAND__DIRECT_REG_WRITE;
1264 	direct_rd_mod_wt.cmd_header.command_type =
1265 		MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1266 	end.cmd_header.command_type =
1267 		MMSCH_COMMAND__END;
1268 
1269 	header.version = MMSCH_VERSION;
1270 	header.total_size = sizeof(struct mmsch_v4_0_init_header) >> 2;
1271 	for (i = 0; i < MMSCH_V4_0_VCN_INSTANCES; i++) {
1272 		header.inst[i].init_status = 0;
1273 		header.inst[i].table_offset = 0;
1274 		header.inst[i].table_size = 0;
1275 	}
1276 
1277 	table_loc = (uint32_t *)table->cpu_addr;
1278 	table_loc += header.total_size;
1279 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1280 		if (adev->vcn.harvest_config & (1 << i))
1281 			continue;
1282 
1283 		// Must re/init fw_shared at beginning
1284 		vcn_v4_0_fw_shared_init(adev, i);
1285 
1286 		table_size = 0;
1287 
1288 		MMSCH_V4_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(VCN, i,
1289 			regUVD_STATUS),
1290 			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1291 
1292 		cache_size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw[i]->size + 4);
1293 
1294 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1295 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1296 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1297 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1298 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1299 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1300 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1301 			offset = 0;
1302 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1303 				regUVD_VCPU_CACHE_OFFSET0),
1304 				0);
1305 		} else {
1306 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1307 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1308 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1309 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1310 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1311 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1312 			offset = cache_size;
1313 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1314 				regUVD_VCPU_CACHE_OFFSET0),
1315 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1316 		}
1317 
1318 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1319 			regUVD_VCPU_CACHE_SIZE0),
1320 			cache_size);
1321 
1322 		cache_addr = adev->vcn.inst[i].gpu_addr + offset;
1323 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1324 			regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1325 			lower_32_bits(cache_addr));
1326 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1327 			regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1328 			upper_32_bits(cache_addr));
1329 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1330 			regUVD_VCPU_CACHE_OFFSET1),
1331 			0);
1332 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1333 			regUVD_VCPU_CACHE_SIZE1),
1334 			AMDGPU_VCN_STACK_SIZE);
1335 
1336 		cache_addr = adev->vcn.inst[i].gpu_addr + offset +
1337 			AMDGPU_VCN_STACK_SIZE;
1338 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1339 			regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1340 			lower_32_bits(cache_addr));
1341 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1342 			regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1343 			upper_32_bits(cache_addr));
1344 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1345 			regUVD_VCPU_CACHE_OFFSET2),
1346 			0);
1347 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1348 			regUVD_VCPU_CACHE_SIZE2),
1349 			AMDGPU_VCN_CONTEXT_SIZE);
1350 
1351 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1352 		rb_setup = &fw_shared->rb_setup;
1353 
1354 		ring_enc = &adev->vcn.inst[i].ring_enc[0];
1355 		ring_enc->wptr = 0;
1356 		rb_enc_addr = ring_enc->gpu_addr;
1357 
1358 		rb_setup->is_rb_enabled_flags |= RB_ENABLED;
1359 		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
1360 
1361 		if (amdgpu_sriov_is_vcn_rb_decouple(adev)) {
1362 			vcn_v4_0_init_ring_metadata(adev, i, ring_enc);
1363 
1364 			memset((void *)&rb_setup->rb_info, 0, sizeof(struct amdgpu_vcn_rb_setup_info) * MAX_NUM_VCN_RB_SETUP);
1365 			if (!(adev->vcn.harvest_config & (1 << 0))) {
1366 				rb_setup->rb_info[0].rb_addr_lo = lower_32_bits(adev->vcn.inst[0].ring_enc[0].gpu_addr);
1367 				rb_setup->rb_info[0].rb_addr_hi = upper_32_bits(adev->vcn.inst[0].ring_enc[0].gpu_addr);
1368 				rb_setup->rb_info[0].rb_size = adev->vcn.inst[0].ring_enc[0].ring_size / 4;
1369 			}
1370 			if (!(adev->vcn.harvest_config & (1 << 1))) {
1371 				rb_setup->rb_info[2].rb_addr_lo = lower_32_bits(adev->vcn.inst[1].ring_enc[0].gpu_addr);
1372 				rb_setup->rb_info[2].rb_addr_hi = upper_32_bits(adev->vcn.inst[1].ring_enc[0].gpu_addr);
1373 				rb_setup->rb_info[2].rb_size = adev->vcn.inst[1].ring_enc[0].ring_size / 4;
1374 			}
1375 			fw_shared->decouple.is_enabled = 1;
1376 			fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_DECOUPLE_FLAG);
1377 		} else {
1378 			rb_setup->rb_addr_lo = lower_32_bits(rb_enc_addr);
1379 			rb_setup->rb_addr_hi = upper_32_bits(rb_enc_addr);
1380 			rb_setup->rb_size = ring_enc->ring_size / 4;
1381 		}
1382 
1383 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1384 			regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
1385 			lower_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
1386 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1387 			regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
1388 			upper_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
1389 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1390 			regUVD_VCPU_NONCACHE_SIZE0),
1391 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
1392 
1393 		/* add end packet */
1394 		MMSCH_V4_0_INSERT_END();
1395 
1396 		/* refine header */
1397 		header.inst[i].init_status = 0;
1398 		header.inst[i].table_offset = header.total_size;
1399 		header.inst[i].table_size = table_size;
1400 		header.total_size += table_size;
1401 	}
1402 
1403 	/* Update init table header in memory */
1404 	size = sizeof(struct mmsch_v4_0_init_header);
1405 	table_loc = (uint32_t *)table->cpu_addr;
1406 	memcpy((void *)table_loc, &header, size);
1407 
1408 	/* message MMSCH (in VCN[0]) to initialize this client
1409 	 * 1, write to mmsch_vf_ctx_addr_lo/hi register with GPU mc addr
1410 	 * of memory descriptor location
1411 	 */
1412 	ctx_addr = table->gpu_addr;
1413 	WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_ADDR_LO, lower_32_bits(ctx_addr));
1414 	WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_ADDR_HI, upper_32_bits(ctx_addr));
1415 
1416 	/* 2, update vmid of descriptor */
1417 	tmp = RREG32_SOC15(VCN, 0, regMMSCH_VF_VMID);
1418 	tmp &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1419 	/* use domain0 for MM scheduler */
1420 	tmp |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1421 	WREG32_SOC15(VCN, 0, regMMSCH_VF_VMID, tmp);
1422 
1423 	/* 3, notify mmsch about the size of this descriptor */
1424 	size = header.total_size;
1425 	WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_SIZE, size);
1426 
1427 	/* 4, set resp to zero */
1428 	WREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_RESP, 0);
1429 
1430 	/* 5, kick off the initialization and wait until
1431 	 * MMSCH_VF_MAILBOX_RESP becomes non-zero
1432 	 */
1433 	param = 0x00000001;
1434 	WREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_HOST, param);
1435 	tmp = 0;
1436 	timeout = 1000;
1437 	resp = 0;
1438 	expected = MMSCH_VF_MAILBOX_RESP__OK;
1439 	while (resp != expected) {
1440 		resp = RREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_RESP);
1441 		if (resp != 0)
1442 			break;
1443 
1444 		udelay(10);
1445 		tmp = tmp + 10;
1446 		if (tmp >= timeout) {
1447 			DRM_ERROR("failed to init MMSCH. TIME-OUT after %d usec"\
1448 				" waiting for regMMSCH_VF_MAILBOX_RESP "\
1449 				"(expected=0x%08x, readback=0x%08x)\n",
1450 				tmp, expected, resp);
1451 			return -EBUSY;
1452 		}
1453 	}
1454 	enabled_vcn = amdgpu_vcn_is_disabled_vcn(adev, VCN_DECODE_RING, 0) ? 1 : 0;
1455 	init_status = ((struct mmsch_v4_0_init_header *)(table_loc))->inst[enabled_vcn].init_status;
1456 	if (resp != expected && resp != MMSCH_VF_MAILBOX_RESP__INCOMPLETE
1457 	&& init_status != MMSCH_VF_ENGINE_STATUS__PASS)
1458 		DRM_ERROR("MMSCH init status is incorrect! readback=0x%08x, header init "\
1459 			"status for VCN%x: 0x%x\n", resp, enabled_vcn, init_status);
1460 
1461 	return 0;
1462 }
1463 
1464 /**
1465  * vcn_v4_0_stop_dpg_mode - VCN stop with dpg mode
1466  *
1467  * @adev: amdgpu_device pointer
1468  * @inst_idx: instance number index
1469  *
1470  * Stop VCN block with dpg mode
1471  */
1472 static void vcn_v4_0_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1473 {
1474 	struct dpg_pause_state state = {.fw_based = VCN_DPG_STATE__UNPAUSE};
1475 	uint32_t tmp;
1476 
1477 	vcn_v4_0_pause_dpg_mode(adev, inst_idx, &state);
1478 	/* Wait for power status to be 1 */
1479 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1480 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1481 
1482 	/* wait for read ptr to be equal to write ptr */
1483 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
1484 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1485 
1486 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1487 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1488 
1489 	/* disable dynamic power gating mode */
1490 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 0,
1491 		~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1492 }
1493 
1494 /**
1495  * vcn_v4_0_stop - VCN stop
1496  *
1497  * @adev: amdgpu_device pointer
1498  *
1499  * Stop VCN block
1500  */
1501 static int vcn_v4_0_stop(struct amdgpu_device *adev)
1502 {
1503 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1504 	uint32_t tmp;
1505 	int i, r = 0;
1506 
1507 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1508 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1509 		fw_shared->sq.queue_mode |= FW_QUEUE_DPG_HOLD_OFF;
1510 
1511 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1512 			vcn_v4_0_stop_dpg_mode(adev, i);
1513 			continue;
1514 		}
1515 
1516 		/* wait for vcn idle */
1517 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1518 		if (r)
1519 			return r;
1520 
1521 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1522 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1523 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1524 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1525 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1526 		if (r)
1527 			return r;
1528 
1529 		/* disable LMI UMC channel */
1530 		tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL2);
1531 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1532 		WREG32_SOC15(VCN, i, regUVD_LMI_CTRL2, tmp);
1533 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK |
1534 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1535 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1536 		if (r)
1537 			return r;
1538 
1539 		/* block VCPU register access */
1540 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL),
1541 				UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1542 				~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1543 
1544 		/* reset VCPU */
1545 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1546 				UVD_VCPU_CNTL__BLK_RST_MASK,
1547 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1548 
1549 		/* disable VCPU clock */
1550 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1551 				~(UVD_VCPU_CNTL__CLK_EN_MASK));
1552 
1553 		/* apply soft reset */
1554 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1555 		tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1556 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1557 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1558 		tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1559 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1560 
1561 		/* clear status */
1562 		WREG32_SOC15(VCN, i, regUVD_STATUS, 0);
1563 
1564 		/* apply HW clock gating */
1565 		vcn_v4_0_enable_clock_gating(adev, i);
1566 
1567 		/* enable VCN power gating */
1568 		vcn_v4_0_enable_static_power_gating(adev, i);
1569 	}
1570 
1571 	if (adev->pm.dpm_enabled)
1572 		amdgpu_dpm_enable_uvd(adev, false);
1573 
1574 	return 0;
1575 }
1576 
1577 /**
1578  * vcn_v4_0_pause_dpg_mode - VCN pause with dpg mode
1579  *
1580  * @adev: amdgpu_device pointer
1581  * @inst_idx: instance number index
1582  * @new_state: pause state
1583  *
1584  * Pause dpg mode for VCN block
1585  */
1586 static int vcn_v4_0_pause_dpg_mode(struct amdgpu_device *adev, int inst_idx,
1587       struct dpg_pause_state *new_state)
1588 {
1589 	uint32_t reg_data = 0;
1590 	int ret_code;
1591 
1592 	/* pause/unpause if state is changed */
1593 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1594 		DRM_DEV_DEBUG(adev->dev, "dpg pause state changed %d -> %d",
1595 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1596 		reg_data = RREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE) &
1597 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1598 
1599 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1600 			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 0x1,
1601 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1602 
1603 			if (!ret_code) {
1604 				/* pause DPG */
1605 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1606 				WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1607 
1608 				/* wait for ACK */
1609 				SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_DPG_PAUSE,
1610 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1611 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1612 
1613 				SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS,
1614 					UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1615 			}
1616 		} else {
1617 			/* unpause dpg, no need to wait */
1618 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1619 			WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1620 		}
1621 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1622 	}
1623 
1624 	return 0;
1625 }
1626 
1627 /**
1628  * vcn_v4_0_unified_ring_get_rptr - get unified read pointer
1629  *
1630  * @ring: amdgpu_ring pointer
1631  *
1632  * Returns the current hardware unified read pointer
1633  */
1634 static uint64_t vcn_v4_0_unified_ring_get_rptr(struct amdgpu_ring *ring)
1635 {
1636 	struct amdgpu_device *adev = ring->adev;
1637 
1638 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1639 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1640 
1641 	return RREG32_SOC15(VCN, ring->me, regUVD_RB_RPTR);
1642 }
1643 
1644 /**
1645  * vcn_v4_0_unified_ring_get_wptr - get unified write pointer
1646  *
1647  * @ring: amdgpu_ring pointer
1648  *
1649  * Returns the current hardware unified write pointer
1650  */
1651 static uint64_t vcn_v4_0_unified_ring_get_wptr(struct amdgpu_ring *ring)
1652 {
1653 	struct amdgpu_device *adev = ring->adev;
1654 
1655 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1656 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1657 
1658 	if (ring->use_doorbell)
1659 		return *ring->wptr_cpu_addr;
1660 	else
1661 		return RREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR);
1662 }
1663 
1664 /**
1665  * vcn_v4_0_unified_ring_set_wptr - set enc write pointer
1666  *
1667  * @ring: amdgpu_ring pointer
1668  *
1669  * Commits the enc write pointer to the hardware
1670  */
1671 static void vcn_v4_0_unified_ring_set_wptr(struct amdgpu_ring *ring)
1672 {
1673 	struct amdgpu_device *adev = ring->adev;
1674 
1675 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1676 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1677 
1678 	if (ring->use_doorbell) {
1679 		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1680 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1681 	} else {
1682 		WREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR, lower_32_bits(ring->wptr));
1683 	}
1684 }
1685 
1686 static int vcn_v4_0_limit_sched(struct amdgpu_cs_parser *p,
1687 				struct amdgpu_job *job)
1688 {
1689 	struct drm_gpu_scheduler **scheds;
1690 
1691 	/* The create msg must be in the first IB submitted */
1692 	if (atomic_read(&job->base.entity->fence_seq))
1693 		return -EINVAL;
1694 
1695 	/* if VCN0 is harvested, we can't support AV1 */
1696 	if (p->adev->vcn.harvest_config & AMDGPU_VCN_HARVEST_VCN0)
1697 		return -EINVAL;
1698 
1699 	scheds = p->adev->gpu_sched[AMDGPU_HW_IP_VCN_ENC]
1700 		[AMDGPU_RING_PRIO_0].sched;
1701 	drm_sched_entity_modify_sched(job->base.entity, scheds, 1);
1702 	return 0;
1703 }
1704 
1705 static int vcn_v4_0_dec_msg(struct amdgpu_cs_parser *p, struct amdgpu_job *job,
1706 			    uint64_t addr)
1707 {
1708 	struct ttm_operation_ctx ctx = { false, false };
1709 	struct amdgpu_bo_va_mapping *map;
1710 	uint32_t *msg, num_buffers;
1711 	struct amdgpu_bo *bo;
1712 	uint64_t start, end;
1713 	unsigned int i;
1714 	void *ptr;
1715 	int r;
1716 
1717 	addr &= AMDGPU_GMC_HOLE_MASK;
1718 	r = amdgpu_cs_find_mapping(p, addr, &bo, &map);
1719 	if (r) {
1720 		DRM_ERROR("Can't find BO for addr 0x%08llx\n", addr);
1721 		return r;
1722 	}
1723 
1724 	start = map->start * AMDGPU_GPU_PAGE_SIZE;
1725 	end = (map->last + 1) * AMDGPU_GPU_PAGE_SIZE;
1726 	if (addr & 0x7) {
1727 		DRM_ERROR("VCN messages must be 8 byte aligned!\n");
1728 		return -EINVAL;
1729 	}
1730 
1731 	bo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
1732 	amdgpu_bo_placement_from_domain(bo, bo->allowed_domains);
1733 	r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
1734 	if (r) {
1735 		DRM_ERROR("Failed validating the VCN message BO (%d)!\n", r);
1736 		return r;
1737 	}
1738 
1739 	r = amdgpu_bo_kmap(bo, &ptr);
1740 	if (r) {
1741 		DRM_ERROR("Failed mapping the VCN message (%d)!\n", r);
1742 		return r;
1743 	}
1744 
1745 	msg = ptr + addr - start;
1746 
1747 	/* Check length */
1748 	if (msg[1] > end - addr) {
1749 		r = -EINVAL;
1750 		goto out;
1751 	}
1752 
1753 	if (msg[3] != RDECODE_MSG_CREATE)
1754 		goto out;
1755 
1756 	num_buffers = msg[2];
1757 	for (i = 0, msg = &msg[6]; i < num_buffers; ++i, msg += 4) {
1758 		uint32_t offset, size, *create;
1759 
1760 		if (msg[0] != RDECODE_MESSAGE_CREATE)
1761 			continue;
1762 
1763 		offset = msg[1];
1764 		size = msg[2];
1765 
1766 		if (offset + size > end) {
1767 			r = -EINVAL;
1768 			goto out;
1769 		}
1770 
1771 		create = ptr + addr + offset - start;
1772 
1773 		/* H264, HEVC and VP9 can run on any instance */
1774 		if (create[0] == 0x7 || create[0] == 0x10 || create[0] == 0x11)
1775 			continue;
1776 
1777 		r = vcn_v4_0_limit_sched(p, job);
1778 		if (r)
1779 			goto out;
1780 	}
1781 
1782 out:
1783 	amdgpu_bo_kunmap(bo);
1784 	return r;
1785 }
1786 
1787 #define RADEON_VCN_ENGINE_TYPE_ENCODE			(0x00000002)
1788 #define RADEON_VCN_ENGINE_TYPE_DECODE			(0x00000003)
1789 
1790 #define RADEON_VCN_ENGINE_INFO				(0x30000001)
1791 #define RADEON_VCN_ENGINE_INFO_MAX_OFFSET		16
1792 
1793 #define RENCODE_ENCODE_STANDARD_AV1			2
1794 #define RENCODE_IB_PARAM_SESSION_INIT			0x00000003
1795 #define RENCODE_IB_PARAM_SESSION_INIT_MAX_OFFSET	64
1796 
1797 /* return the offset in ib if id is found, -1 otherwise
1798  * to speed up the searching we only search upto max_offset
1799  */
1800 static int vcn_v4_0_enc_find_ib_param(struct amdgpu_ib *ib, uint32_t id, int max_offset)
1801 {
1802 	int i;
1803 
1804 	for (i = 0; i < ib->length_dw && i < max_offset && ib->ptr[i] >= 8; i += ib->ptr[i]/4) {
1805 		if (ib->ptr[i + 1] == id)
1806 			return i;
1807 	}
1808 	return -1;
1809 }
1810 
1811 static int vcn_v4_0_ring_patch_cs_in_place(struct amdgpu_cs_parser *p,
1812 					   struct amdgpu_job *job,
1813 					   struct amdgpu_ib *ib)
1814 {
1815 	struct amdgpu_ring *ring = amdgpu_job_ring(job);
1816 	struct amdgpu_vcn_decode_buffer *decode_buffer;
1817 	uint64_t addr;
1818 	uint32_t val;
1819 	int idx;
1820 
1821 	/* The first instance can decode anything */
1822 	if (!ring->me)
1823 		return 0;
1824 
1825 	/* RADEON_VCN_ENGINE_INFO is at the top of ib block */
1826 	idx = vcn_v4_0_enc_find_ib_param(ib, RADEON_VCN_ENGINE_INFO,
1827 			RADEON_VCN_ENGINE_INFO_MAX_OFFSET);
1828 	if (idx < 0) /* engine info is missing */
1829 		return 0;
1830 
1831 	val = amdgpu_ib_get_value(ib, idx + 2); /* RADEON_VCN_ENGINE_TYPE */
1832 	if (val == RADEON_VCN_ENGINE_TYPE_DECODE) {
1833 		decode_buffer = (struct amdgpu_vcn_decode_buffer *)&ib->ptr[idx + 6];
1834 
1835 		if (!(decode_buffer->valid_buf_flag  & 0x1))
1836 			return 0;
1837 
1838 		addr = ((u64)decode_buffer->msg_buffer_address_hi) << 32 |
1839 			decode_buffer->msg_buffer_address_lo;
1840 		return vcn_v4_0_dec_msg(p, job, addr);
1841 	} else if (val == RADEON_VCN_ENGINE_TYPE_ENCODE) {
1842 		idx = vcn_v4_0_enc_find_ib_param(ib, RENCODE_IB_PARAM_SESSION_INIT,
1843 			RENCODE_IB_PARAM_SESSION_INIT_MAX_OFFSET);
1844 		if (idx >= 0 && ib->ptr[idx + 2] == RENCODE_ENCODE_STANDARD_AV1)
1845 			return vcn_v4_0_limit_sched(p, job);
1846 	}
1847 	return 0;
1848 }
1849 
1850 static struct amdgpu_ring_funcs vcn_v4_0_unified_ring_vm_funcs = {
1851 	.type = AMDGPU_RING_TYPE_VCN_ENC,
1852 	.align_mask = 0x3f,
1853 	.nop = VCN_ENC_CMD_NO_OP,
1854 	.extra_dw = sizeof(struct amdgpu_vcn_rb_metadata),
1855 	.get_rptr = vcn_v4_0_unified_ring_get_rptr,
1856 	.get_wptr = vcn_v4_0_unified_ring_get_wptr,
1857 	.set_wptr = vcn_v4_0_unified_ring_set_wptr,
1858 	.patch_cs_in_place = vcn_v4_0_ring_patch_cs_in_place,
1859 	.emit_frame_size =
1860 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1861 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1862 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1863 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1864 		1, /* vcn_v2_0_enc_ring_insert_end */
1865 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1866 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
1867 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
1868 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1869 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
1870 	.test_ib = amdgpu_vcn_unified_ring_test_ib,
1871 	.insert_nop = amdgpu_ring_insert_nop,
1872 	.insert_end = vcn_v2_0_enc_ring_insert_end,
1873 	.pad_ib = amdgpu_ring_generic_pad_ib,
1874 	.begin_use = amdgpu_vcn_ring_begin_use,
1875 	.end_use = amdgpu_vcn_ring_end_use,
1876 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1877 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1878 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1879 };
1880 
1881 /**
1882  * vcn_v4_0_set_unified_ring_funcs - set unified ring functions
1883  *
1884  * @adev: amdgpu_device pointer
1885  *
1886  * Set unified ring functions
1887  */
1888 static void vcn_v4_0_set_unified_ring_funcs(struct amdgpu_device *adev)
1889 {
1890 	int i;
1891 
1892 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1893 		if (adev->vcn.harvest_config & (1 << i))
1894 			continue;
1895 
1896 		if (amdgpu_ip_version(adev, VCN_HWIP, 0) == IP_VERSION(4, 0, 2))
1897 			vcn_v4_0_unified_ring_vm_funcs.secure_submission_supported = true;
1898 
1899 		adev->vcn.inst[i].ring_enc[0].funcs =
1900 		       (const struct amdgpu_ring_funcs *)&vcn_v4_0_unified_ring_vm_funcs;
1901 		adev->vcn.inst[i].ring_enc[0].me = i;
1902 
1903 		DRM_INFO("VCN(%d) encode/decode are enabled in VM mode\n", i);
1904 	}
1905 }
1906 
1907 /**
1908  * vcn_v4_0_is_idle - check VCN block is idle
1909  *
1910  * @handle: amdgpu_device pointer
1911  *
1912  * Check whether VCN block is idle
1913  */
1914 static bool vcn_v4_0_is_idle(void *handle)
1915 {
1916 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1917 	int i, ret = 1;
1918 
1919 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1920 		if (adev->vcn.harvest_config & (1 << i))
1921 			continue;
1922 
1923 		ret &= (RREG32_SOC15(VCN, i, regUVD_STATUS) == UVD_STATUS__IDLE);
1924 	}
1925 
1926 	return ret;
1927 }
1928 
1929 /**
1930  * vcn_v4_0_wait_for_idle - wait for VCN block idle
1931  *
1932  * @handle: amdgpu_device pointer
1933  *
1934  * Wait for VCN block idle
1935  */
1936 static int vcn_v4_0_wait_for_idle(void *handle)
1937 {
1938 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1939 	int i, ret = 0;
1940 
1941 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1942 		if (adev->vcn.harvest_config & (1 << i))
1943 			continue;
1944 
1945 		ret = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE,
1946 			UVD_STATUS__IDLE);
1947 		if (ret)
1948 			return ret;
1949 	}
1950 
1951 	return ret;
1952 }
1953 
1954 /**
1955  * vcn_v4_0_set_clockgating_state - set VCN block clockgating state
1956  *
1957  * @handle: amdgpu_device pointer
1958  * @state: clock gating state
1959  *
1960  * Set VCN block clockgating state
1961  */
1962 static int vcn_v4_0_set_clockgating_state(void *handle, enum amd_clockgating_state state)
1963 {
1964 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1965 	bool enable = state == AMD_CG_STATE_GATE;
1966 	int i;
1967 
1968 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1969 		if (adev->vcn.harvest_config & (1 << i))
1970 			continue;
1971 
1972 		if (enable) {
1973 			if (RREG32_SOC15(VCN, i, regUVD_STATUS) != UVD_STATUS__IDLE)
1974 				return -EBUSY;
1975 			vcn_v4_0_enable_clock_gating(adev, i);
1976 		} else {
1977 			vcn_v4_0_disable_clock_gating(adev, i);
1978 		}
1979 	}
1980 
1981 	return 0;
1982 }
1983 
1984 /**
1985  * vcn_v4_0_set_powergating_state - set VCN block powergating state
1986  *
1987  * @handle: amdgpu_device pointer
1988  * @state: power gating state
1989  *
1990  * Set VCN block powergating state
1991  */
1992 static int vcn_v4_0_set_powergating_state(void *handle, enum amd_powergating_state state)
1993 {
1994 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1995 	int ret;
1996 
1997 	/* for SRIOV, guest should not control VCN Power-gating
1998 	 * MMSCH FW should control Power-gating and clock-gating
1999 	 * guest should avoid touching CGC and PG
2000 	 */
2001 	if (amdgpu_sriov_vf(adev)) {
2002 		adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
2003 		return 0;
2004 	}
2005 
2006 	if (state == adev->vcn.cur_state)
2007 		return 0;
2008 
2009 	if (state == AMD_PG_STATE_GATE)
2010 		ret = vcn_v4_0_stop(adev);
2011 	else
2012 		ret = vcn_v4_0_start(adev);
2013 
2014 	if (!ret)
2015 		adev->vcn.cur_state = state;
2016 
2017 	return ret;
2018 }
2019 
2020 /**
2021  * vcn_v4_0_set_ras_interrupt_state - set VCN block RAS interrupt state
2022  *
2023  * @adev: amdgpu_device pointer
2024  * @source: interrupt sources
2025  * @type: interrupt types
2026  * @state: interrupt states
2027  *
2028  * Set VCN block RAS interrupt state
2029  */
2030 static int vcn_v4_0_set_ras_interrupt_state(struct amdgpu_device *adev,
2031 	struct amdgpu_irq_src *source,
2032 	unsigned int type,
2033 	enum amdgpu_interrupt_state state)
2034 {
2035 	return 0;
2036 }
2037 
2038 /**
2039  * vcn_v4_0_process_interrupt - process VCN block interrupt
2040  *
2041  * @adev: amdgpu_device pointer
2042  * @source: interrupt sources
2043  * @entry: interrupt entry from clients and sources
2044  *
2045  * Process VCN block interrupt
2046  */
2047 static int vcn_v4_0_process_interrupt(struct amdgpu_device *adev, struct amdgpu_irq_src *source,
2048       struct amdgpu_iv_entry *entry)
2049 {
2050 	uint32_t ip_instance;
2051 
2052 	if (amdgpu_sriov_is_vcn_rb_decouple(adev)) {
2053 		ip_instance = entry->ring_id;
2054 	} else {
2055 		switch (entry->client_id) {
2056 		case SOC15_IH_CLIENTID_VCN:
2057 			ip_instance = 0;
2058 			break;
2059 		case SOC15_IH_CLIENTID_VCN1:
2060 			ip_instance = 1;
2061 			break;
2062 		default:
2063 			DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
2064 			return 0;
2065 		}
2066 	}
2067 
2068 	DRM_DEBUG("IH: VCN TRAP\n");
2069 
2070 	switch (entry->src_id) {
2071 	case VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
2072 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
2073 		break;
2074 	default:
2075 		DRM_ERROR("Unhandled interrupt: %d %d\n",
2076 			  entry->src_id, entry->src_data[0]);
2077 		break;
2078 	}
2079 
2080 	return 0;
2081 }
2082 
2083 static const struct amdgpu_irq_src_funcs vcn_v4_0_irq_funcs = {
2084 	.process = vcn_v4_0_process_interrupt,
2085 };
2086 
2087 static const struct amdgpu_irq_src_funcs vcn_v4_0_ras_irq_funcs = {
2088 	.set = vcn_v4_0_set_ras_interrupt_state,
2089 	.process = amdgpu_vcn_process_poison_irq,
2090 };
2091 
2092 /**
2093  * vcn_v4_0_set_irq_funcs - set VCN block interrupt irq functions
2094  *
2095  * @adev: amdgpu_device pointer
2096  *
2097  * Set VCN block interrupt irq functions
2098  */
2099 static void vcn_v4_0_set_irq_funcs(struct amdgpu_device *adev)
2100 {
2101 	int i;
2102 
2103 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2104 		if (adev->vcn.harvest_config & (1 << i))
2105 			continue;
2106 
2107 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
2108 		adev->vcn.inst[i].irq.funcs = &vcn_v4_0_irq_funcs;
2109 
2110 		adev->vcn.inst[i].ras_poison_irq.num_types = adev->vcn.num_enc_rings + 1;
2111 		adev->vcn.inst[i].ras_poison_irq.funcs = &vcn_v4_0_ras_irq_funcs;
2112 	}
2113 }
2114 
2115 static const struct amd_ip_funcs vcn_v4_0_ip_funcs = {
2116 	.name = "vcn_v4_0",
2117 	.early_init = vcn_v4_0_early_init,
2118 	.late_init = NULL,
2119 	.sw_init = vcn_v4_0_sw_init,
2120 	.sw_fini = vcn_v4_0_sw_fini,
2121 	.hw_init = vcn_v4_0_hw_init,
2122 	.hw_fini = vcn_v4_0_hw_fini,
2123 	.suspend = vcn_v4_0_suspend,
2124 	.resume = vcn_v4_0_resume,
2125 	.is_idle = vcn_v4_0_is_idle,
2126 	.wait_for_idle = vcn_v4_0_wait_for_idle,
2127 	.check_soft_reset = NULL,
2128 	.pre_soft_reset = NULL,
2129 	.soft_reset = NULL,
2130 	.post_soft_reset = NULL,
2131 	.set_clockgating_state = vcn_v4_0_set_clockgating_state,
2132 	.set_powergating_state = vcn_v4_0_set_powergating_state,
2133 };
2134 
2135 const struct amdgpu_ip_block_version vcn_v4_0_ip_block = {
2136 	.type = AMD_IP_BLOCK_TYPE_VCN,
2137 	.major = 4,
2138 	.minor = 0,
2139 	.rev = 0,
2140 	.funcs = &vcn_v4_0_ip_funcs,
2141 };
2142 
2143 static uint32_t vcn_v4_0_query_poison_by_instance(struct amdgpu_device *adev,
2144 			uint32_t instance, uint32_t sub_block)
2145 {
2146 	uint32_t poison_stat = 0, reg_value = 0;
2147 
2148 	switch (sub_block) {
2149 	case AMDGPU_VCN_V4_0_VCPU_VCODEC:
2150 		reg_value = RREG32_SOC15(VCN, instance, regUVD_RAS_VCPU_VCODEC_STATUS);
2151 		poison_stat = REG_GET_FIELD(reg_value, UVD_RAS_VCPU_VCODEC_STATUS, POISONED_PF);
2152 		break;
2153 	default:
2154 		break;
2155 	}
2156 
2157 	if (poison_stat)
2158 		dev_info(adev->dev, "Poison detected in VCN%d, sub_block%d\n",
2159 			instance, sub_block);
2160 
2161 	return poison_stat;
2162 }
2163 
2164 static bool vcn_v4_0_query_ras_poison_status(struct amdgpu_device *adev)
2165 {
2166 	uint32_t inst, sub;
2167 	uint32_t poison_stat = 0;
2168 
2169 	for (inst = 0; inst < adev->vcn.num_vcn_inst; inst++)
2170 		for (sub = 0; sub < AMDGPU_VCN_V4_0_MAX_SUB_BLOCK; sub++)
2171 			poison_stat +=
2172 				vcn_v4_0_query_poison_by_instance(adev, inst, sub);
2173 
2174 	return !!poison_stat;
2175 }
2176 
2177 const struct amdgpu_ras_block_hw_ops vcn_v4_0_ras_hw_ops = {
2178 	.query_poison_status = vcn_v4_0_query_ras_poison_status,
2179 };
2180 
2181 static struct amdgpu_vcn_ras vcn_v4_0_ras = {
2182 	.ras_block = {
2183 		.hw_ops = &vcn_v4_0_ras_hw_ops,
2184 		.ras_late_init = amdgpu_vcn_ras_late_init,
2185 	},
2186 };
2187 
2188 static void vcn_v4_0_set_ras_funcs(struct amdgpu_device *adev)
2189 {
2190 	switch (amdgpu_ip_version(adev, VCN_HWIP, 0)) {
2191 	case IP_VERSION(4, 0, 0):
2192 		adev->vcn.ras = &vcn_v4_0_ras;
2193 		break;
2194 	default:
2195 		break;
2196 	}
2197 }
2198