xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c (revision 98ce7d32)
1 /*
2  * Copyright 2021 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #include <linux/firmware.h>
25 #include "amdgpu.h"
26 #include "amdgpu_vcn.h"
27 #include "amdgpu_pm.h"
28 #include "amdgpu_cs.h"
29 #include "soc15.h"
30 #include "soc15d.h"
31 #include "soc15_hw_ip.h"
32 #include "vcn_v2_0.h"
33 #include "mmsch_v4_0.h"
34 #include "vcn_v4_0.h"
35 
36 #include "vcn/vcn_4_0_0_offset.h"
37 #include "vcn/vcn_4_0_0_sh_mask.h"
38 #include "ivsrcid/vcn/irqsrcs_vcn_4_0.h"
39 
40 #include <drm/drm_drv.h>
41 
42 #define mmUVD_DPG_LMA_CTL							regUVD_DPG_LMA_CTL
43 #define mmUVD_DPG_LMA_CTL_BASE_IDX						regUVD_DPG_LMA_CTL_BASE_IDX
44 #define mmUVD_DPG_LMA_DATA							regUVD_DPG_LMA_DATA
45 #define mmUVD_DPG_LMA_DATA_BASE_IDX						regUVD_DPG_LMA_DATA_BASE_IDX
46 
47 #define VCN_VID_SOC_ADDRESS_2_0							0x1fb00
48 #define VCN1_VID_SOC_ADDRESS_3_0						0x48300
49 
50 #define VCN_HARVEST_MMSCH								0
51 
52 #define RDECODE_MSG_CREATE							0x00000000
53 #define RDECODE_MESSAGE_CREATE							0x00000001
54 
55 static int amdgpu_ih_clientid_vcns[] = {
56 	SOC15_IH_CLIENTID_VCN,
57 	SOC15_IH_CLIENTID_VCN1
58 };
59 
60 static int vcn_v4_0_start_sriov(struct amdgpu_device *adev);
61 static void vcn_v4_0_set_unified_ring_funcs(struct amdgpu_device *adev);
62 static void vcn_v4_0_set_irq_funcs(struct amdgpu_device *adev);
63 static int vcn_v4_0_set_powergating_state(void *handle,
64         enum amd_powergating_state state);
65 static int vcn_v4_0_pause_dpg_mode(struct amdgpu_device *adev,
66         int inst_idx, struct dpg_pause_state *new_state);
67 static void vcn_v4_0_unified_ring_set_wptr(struct amdgpu_ring *ring);
68 static void vcn_v4_0_set_ras_funcs(struct amdgpu_device *adev);
69 
70 /**
71  * vcn_v4_0_early_init - set function pointers and load microcode
72  *
73  * @handle: amdgpu_device pointer
74  *
75  * Set ring and irq function pointers
76  * Load microcode from filesystem
77  */
78 static int vcn_v4_0_early_init(void *handle)
79 {
80 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
81 	int i;
82 
83 	if (amdgpu_sriov_vf(adev)) {
84 		adev->vcn.harvest_config = VCN_HARVEST_MMSCH;
85 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
86 			if (amdgpu_vcn_is_disabled_vcn(adev, VCN_ENCODE_RING, i)) {
87 				adev->vcn.harvest_config |= 1 << i;
88 				dev_info(adev->dev, "VCN%d is disabled by hypervisor\n", i);
89 			}
90 		}
91 	}
92 
93 	/* re-use enc ring as unified ring */
94 	adev->vcn.num_enc_rings = 1;
95 
96 	vcn_v4_0_set_unified_ring_funcs(adev);
97 	vcn_v4_0_set_irq_funcs(adev);
98 	vcn_v4_0_set_ras_funcs(adev);
99 
100 	return amdgpu_vcn_early_init(adev);
101 }
102 
103 /**
104  * vcn_v4_0_sw_init - sw init for VCN block
105  *
106  * @handle: amdgpu_device pointer
107  *
108  * Load firmware and sw initialization
109  */
110 static int vcn_v4_0_sw_init(void *handle)
111 {
112 	struct amdgpu_ring *ring;
113 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
114 	int i, r;
115 
116 	r = amdgpu_vcn_sw_init(adev);
117 	if (r)
118 		return r;
119 
120 	amdgpu_vcn_setup_ucode(adev);
121 
122 	r = amdgpu_vcn_resume(adev);
123 	if (r)
124 		return r;
125 
126 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
127 		volatile struct amdgpu_vcn4_fw_shared *fw_shared;
128 
129 		if (adev->vcn.harvest_config & (1 << i))
130 			continue;
131 
132 		atomic_set(&adev->vcn.inst[i].sched_score, 0);
133 
134 		/* VCN UNIFIED TRAP */
135 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
136 				VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);
137 		if (r)
138 			return r;
139 
140 		/* VCN POISON TRAP */
141 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
142 				VCN_4_0__SRCID_UVD_POISON, &adev->vcn.inst[i].irq);
143 		if (r)
144 			return r;
145 
146 		ring = &adev->vcn.inst[i].ring_enc[0];
147 		ring->use_doorbell = true;
148 		if (amdgpu_sriov_vf(adev))
149 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + i * (adev->vcn.num_enc_rings + 1) + 1;
150 		else
151 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + 8 * i;
152 
153 		sprintf(ring->name, "vcn_unified_%d", i);
154 
155 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
156 						AMDGPU_RING_PRIO_0, &adev->vcn.inst[i].sched_score);
157 		if (r)
158 			return r;
159 
160 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
161 		fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_FW_SHARED_FLAG_0_UNIFIED_QUEUE);
162 		fw_shared->sq.is_enabled = 1;
163 
164 		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_SMU_DPM_INTERFACE_FLAG);
165 		fw_shared->smu_dpm_interface.smu_interface_type = (adev->flags & AMD_IS_APU) ?
166 			AMDGPU_VCN_SMU_DPM_INTERFACE_APU : AMDGPU_VCN_SMU_DPM_INTERFACE_DGPU;
167 
168 		if (amdgpu_sriov_vf(adev))
169 			fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
170 
171 		if (amdgpu_vcnfw_log)
172 			amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
173 	}
174 
175 	if (amdgpu_sriov_vf(adev)) {
176 		r = amdgpu_virt_alloc_mm_table(adev);
177 		if (r)
178 			return r;
179 	}
180 
181 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
182 		adev->vcn.pause_dpg_mode = vcn_v4_0_pause_dpg_mode;
183 
184 	r = amdgpu_vcn_ras_sw_init(adev);
185 	if (r)
186 		return r;
187 
188 	return 0;
189 }
190 
191 /**
192  * vcn_v4_0_sw_fini - sw fini for VCN block
193  *
194  * @handle: amdgpu_device pointer
195  *
196  * VCN suspend and free up sw allocation
197  */
198 static int vcn_v4_0_sw_fini(void *handle)
199 {
200 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
201 	int i, r, idx;
202 
203 	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
204 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
205 			volatile struct amdgpu_vcn4_fw_shared *fw_shared;
206 
207 			if (adev->vcn.harvest_config & (1 << i))
208 				continue;
209 
210 			fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
211 			fw_shared->present_flag_0 = 0;
212 			fw_shared->sq.is_enabled = 0;
213 		}
214 
215 		drm_dev_exit(idx);
216 	}
217 
218 	if (amdgpu_sriov_vf(adev))
219 		amdgpu_virt_free_mm_table(adev);
220 
221 	r = amdgpu_vcn_suspend(adev);
222 	if (r)
223 		return r;
224 
225 	r = amdgpu_vcn_sw_fini(adev);
226 
227 	return r;
228 }
229 
230 /**
231  * vcn_v4_0_hw_init - start and test VCN block
232  *
233  * @handle: amdgpu_device pointer
234  *
235  * Initialize the hardware, boot up the VCPU and do some testing
236  */
237 static int vcn_v4_0_hw_init(void *handle)
238 {
239 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
240 	struct amdgpu_ring *ring;
241 	int i, r;
242 
243 	if (amdgpu_sriov_vf(adev)) {
244 		r = vcn_v4_0_start_sriov(adev);
245 		if (r)
246 			goto done;
247 
248 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
249 			if (adev->vcn.harvest_config & (1 << i))
250 				continue;
251 
252 			ring = &adev->vcn.inst[i].ring_enc[0];
253 			ring->wptr = 0;
254 			ring->wptr_old = 0;
255 			vcn_v4_0_unified_ring_set_wptr(ring);
256 			ring->sched.ready = true;
257 
258 		}
259 	} else {
260 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
261 			if (adev->vcn.harvest_config & (1 << i))
262 				continue;
263 
264 			ring = &adev->vcn.inst[i].ring_enc[0];
265 
266 			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
267 					((adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i), i);
268 
269 			r = amdgpu_ring_test_helper(ring);
270 			if (r)
271 				goto done;
272 
273 		}
274 	}
275 
276 done:
277 	if (!r)
278 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
279 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
280 
281 	return r;
282 }
283 
284 /**
285  * vcn_v4_0_hw_fini - stop the hardware block
286  *
287  * @handle: amdgpu_device pointer
288  *
289  * Stop the VCN block, mark ring as not ready any more
290  */
291 static int vcn_v4_0_hw_fini(void *handle)
292 {
293 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
294 	int i;
295 
296 	cancel_delayed_work_sync(&adev->vcn.idle_work);
297 
298 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
299 		if (adev->vcn.harvest_config & (1 << i))
300 			continue;
301 		if (!amdgpu_sriov_vf(adev)) {
302 			if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
303                         (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
304                                 RREG32_SOC15(VCN, i, regUVD_STATUS))) {
305                         vcn_v4_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
306 			}
307 		}
308 
309 		amdgpu_irq_put(adev, &adev->vcn.inst[i].irq, 0);
310 	}
311 
312 	return 0;
313 }
314 
315 /**
316  * vcn_v4_0_suspend - suspend VCN block
317  *
318  * @handle: amdgpu_device pointer
319  *
320  * HW fini and suspend VCN block
321  */
322 static int vcn_v4_0_suspend(void *handle)
323 {
324 	int r;
325 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
326 
327 	r = vcn_v4_0_hw_fini(adev);
328 	if (r)
329 		return r;
330 
331 	r = amdgpu_vcn_suspend(adev);
332 
333 	return r;
334 }
335 
336 /**
337  * vcn_v4_0_resume - resume VCN block
338  *
339  * @handle: amdgpu_device pointer
340  *
341  * Resume firmware and hw init VCN block
342  */
343 static int vcn_v4_0_resume(void *handle)
344 {
345 	int r;
346 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
347 
348 	r = amdgpu_vcn_resume(adev);
349 	if (r)
350 		return r;
351 
352 	r = vcn_v4_0_hw_init(adev);
353 
354 	return r;
355 }
356 
357 /**
358  * vcn_v4_0_mc_resume - memory controller programming
359  *
360  * @adev: amdgpu_device pointer
361  * @inst: instance number
362  *
363  * Let the VCN memory controller know it's offsets
364  */
365 static void vcn_v4_0_mc_resume(struct amdgpu_device *adev, int inst)
366 {
367 	uint32_t offset, size;
368 	const struct common_firmware_header *hdr;
369 
370 	hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
371 	size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
372 
373 	/* cache window 0: fw */
374 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
375 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
376 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo));
377 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
378 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi));
379 		WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, 0);
380 		offset = 0;
381 	} else {
382 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
383 			lower_32_bits(adev->vcn.inst[inst].gpu_addr));
384 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
385 			upper_32_bits(adev->vcn.inst[inst].gpu_addr));
386 		offset = size;
387                 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
388 	}
389 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE0, size);
390 
391 	/* cache window 1: stack */
392 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
393 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
394 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
395 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
396 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET1, 0);
397 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
398 
399 	/* cache window 2: context */
400 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
401 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
402 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
403 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
404 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET2, 0);
405 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
406 
407 	/* non-cache window */
408 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
409 		lower_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
410 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
411 		upper_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
412 	WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_OFFSET0, 0);
413 	WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_SIZE0,
414 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
415 }
416 
417 /**
418  * vcn_v4_0_mc_resume_dpg_mode - memory controller programming for dpg mode
419  *
420  * @adev: amdgpu_device pointer
421  * @inst_idx: instance number index
422  * @indirect: indirectly write sram
423  *
424  * Let the VCN memory controller know it's offsets with dpg mode
425  */
426 static void vcn_v4_0_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
427 {
428 	uint32_t offset, size;
429 	const struct common_firmware_header *hdr;
430 	hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
431 	size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
432 
433 	/* cache window 0: fw */
434 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
435 		if (!indirect) {
436 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
437 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
438 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
439 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
440 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
441 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
442 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
443 				VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
444 		} else {
445 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
446 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
447 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
448 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
449 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
450 				VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
451 		}
452 		offset = 0;
453 	} else {
454 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
455 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
456 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
457 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
458 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
459 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
460 		offset = size;
461 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
462 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0),
463 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
464 	}
465 
466 	if (!indirect)
467 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
468 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
469 	else
470 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
471 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
472 
473 	/* cache window 1: stack */
474 	if (!indirect) {
475 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
476 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
477 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
478 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
479 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
480 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
481 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
482 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
483 	} else {
484 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
485 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
486 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
487 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
488 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
489 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
490 	}
491 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
492 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
493 
494 	/* cache window 2: context */
495 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
496 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
497 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
498 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
499 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
500 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
501 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
502 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
503 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
504 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
505 
506 	/* non-cache window */
507 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
508 			VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
509 			lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
510 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
511 			VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
512 			upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
513 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
514 			VCN, inst_idx, regUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
515 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
516 			VCN, inst_idx, regUVD_VCPU_NONCACHE_SIZE0),
517 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)), 0, indirect);
518 
519 	/* VCN global tiling registers */
520 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
521 		VCN, 0, regUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
522 }
523 
524 /**
525  * vcn_v4_0_disable_static_power_gating - disable VCN static power gating
526  *
527  * @adev: amdgpu_device pointer
528  * @inst: instance number
529  *
530  * Disable static power gating for VCN block
531  */
532 static void vcn_v4_0_disable_static_power_gating(struct amdgpu_device *adev, int inst)
533 {
534 	uint32_t data = 0;
535 
536 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
537 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
538 			| 1 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
539 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
540 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
541 			| 2 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
542 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
543 			| 2 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
544 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
545 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
546 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
547 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
548 			| 2 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
549 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
550 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
551 
552 		WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
553 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS,
554 			UVD_PGFSM_STATUS__UVDM_UVDU_UVDLM_PWR_ON_3_0, 0x3F3FFFFF);
555 	} else {
556 		uint32_t value;
557 
558 		value = (inst) ? 0x2200800 : 0;
559 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
560 			| 1 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
561 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
562 			| 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
563 			| 1 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
564 			| 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
565 			| 1 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
566 			| 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
567 			| 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
568 			| 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
569 			| 1 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
570 			| 1 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
571 			| 1 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
572 			| 1 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
573 
574                 WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
575                 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS, value,  0x3F3FFFFF);
576         }
577 
578         data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
579         data &= ~0x103;
580         if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
581                 data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
582                         UVD_POWER_STATUS__UVD_PG_EN_MASK;
583 
584         WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
585 
586         return;
587 }
588 
589 /**
590  * vcn_v4_0_enable_static_power_gating - enable VCN static power gating
591  *
592  * @adev: amdgpu_device pointer
593  * @inst: instance number
594  *
595  * Enable static power gating for VCN block
596  */
597 static void vcn_v4_0_enable_static_power_gating(struct amdgpu_device *adev, int inst)
598 {
599 	uint32_t data;
600 
601 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
602 		/* Before power off, this indicator has to be turned on */
603 		data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
604 		data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
605 		data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
606 		WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
607 
608 		data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
609 			| 2 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
610 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
611 			| 2 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
612 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
613 			| 2 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
614 			| 2 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
615 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
616 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
617 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
618 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
619 			| 2 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
620 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
621 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
622 		WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
623 
624 		data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
625 			| 2 << UVD_PGFSM_STATUS__UVDS_PWR_STATUS__SHIFT
626 			| 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
627 			| 2 << UVD_PGFSM_STATUS__UVDTC_PWR_STATUS__SHIFT
628 			| 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
629 			| 2 << UVD_PGFSM_STATUS__UVDTA_PWR_STATUS__SHIFT
630 			| 2 << UVD_PGFSM_STATUS__UVDLM_PWR_STATUS__SHIFT
631 			| 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
632 			| 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
633 			| 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT
634 			| 2 << UVD_PGFSM_STATUS__UVDAB_PWR_STATUS__SHIFT
635 			| 2 << UVD_PGFSM_STATUS__UVDTB_PWR_STATUS__SHIFT
636 			| 2 << UVD_PGFSM_STATUS__UVDNA_PWR_STATUS__SHIFT
637 			| 2 << UVD_PGFSM_STATUS__UVDNB_PWR_STATUS__SHIFT);
638 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS, data, 0x3F3FFFFF);
639 	}
640 
641         return;
642 }
643 
644 /**
645  * vcn_v4_0_disable_clock_gating - disable VCN clock gating
646  *
647  * @adev: amdgpu_device pointer
648  * @inst: instance number
649  *
650  * Disable clock gating for VCN block
651  */
652 static void vcn_v4_0_disable_clock_gating(struct amdgpu_device *adev, int inst)
653 {
654 	uint32_t data;
655 
656 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
657 		return;
658 
659 	/* VCN disable CGC */
660 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
661 	data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
662 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
663 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
664 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
665 
666 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_GATE);
667 	data &= ~(UVD_CGC_GATE__SYS_MASK
668 		| UVD_CGC_GATE__UDEC_MASK
669 		| UVD_CGC_GATE__MPEG2_MASK
670 		| UVD_CGC_GATE__REGS_MASK
671 		| UVD_CGC_GATE__RBC_MASK
672 		| UVD_CGC_GATE__LMI_MC_MASK
673 		| UVD_CGC_GATE__LMI_UMC_MASK
674 		| UVD_CGC_GATE__IDCT_MASK
675 		| UVD_CGC_GATE__MPRD_MASK
676 		| UVD_CGC_GATE__MPC_MASK
677 		| UVD_CGC_GATE__LBSI_MASK
678 		| UVD_CGC_GATE__LRBBM_MASK
679 		| UVD_CGC_GATE__UDEC_RE_MASK
680 		| UVD_CGC_GATE__UDEC_CM_MASK
681 		| UVD_CGC_GATE__UDEC_IT_MASK
682 		| UVD_CGC_GATE__UDEC_DB_MASK
683 		| UVD_CGC_GATE__UDEC_MP_MASK
684 		| UVD_CGC_GATE__WCB_MASK
685 		| UVD_CGC_GATE__VCPU_MASK
686 		| UVD_CGC_GATE__MMSCH_MASK);
687 
688 	WREG32_SOC15(VCN, inst, regUVD_CGC_GATE, data);
689 	SOC15_WAIT_ON_RREG(VCN, inst, regUVD_CGC_GATE, 0,  0xFFFFFFFF);
690 
691 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
692 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
693 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
694 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
695 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
696 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
697 		| UVD_CGC_CTRL__SYS_MODE_MASK
698 		| UVD_CGC_CTRL__UDEC_MODE_MASK
699 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
700 		| UVD_CGC_CTRL__REGS_MODE_MASK
701 		| UVD_CGC_CTRL__RBC_MODE_MASK
702 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
703 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
704 		| UVD_CGC_CTRL__IDCT_MODE_MASK
705 		| UVD_CGC_CTRL__MPRD_MODE_MASK
706 		| UVD_CGC_CTRL__MPC_MODE_MASK
707 		| UVD_CGC_CTRL__LBSI_MODE_MASK
708 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
709 		| UVD_CGC_CTRL__WCB_MODE_MASK
710 		| UVD_CGC_CTRL__VCPU_MODE_MASK
711 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
712 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
713 
714 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE);
715 	data |= (UVD_SUVD_CGC_GATE__SRE_MASK
716 		| UVD_SUVD_CGC_GATE__SIT_MASK
717 		| UVD_SUVD_CGC_GATE__SMP_MASK
718 		| UVD_SUVD_CGC_GATE__SCM_MASK
719 		| UVD_SUVD_CGC_GATE__SDB_MASK
720 		| UVD_SUVD_CGC_GATE__SRE_H264_MASK
721 		| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
722 		| UVD_SUVD_CGC_GATE__SIT_H264_MASK
723 		| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
724 		| UVD_SUVD_CGC_GATE__SCM_H264_MASK
725 		| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
726 		| UVD_SUVD_CGC_GATE__SDB_H264_MASK
727 		| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
728 		| UVD_SUVD_CGC_GATE__SCLR_MASK
729 		| UVD_SUVD_CGC_GATE__UVD_SC_MASK
730 		| UVD_SUVD_CGC_GATE__ENT_MASK
731 		| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
732 		| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
733 		| UVD_SUVD_CGC_GATE__SITE_MASK
734 		| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
735 		| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
736 		| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
737 		| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
738 		| UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
739 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE, data);
740 
741 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
742 	data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
743 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
744 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
745 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
746 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
747 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
748 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
749 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
750 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
751 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
752 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
753 }
754 
755 /**
756  * vcn_v4_0_disable_clock_gating_dpg_mode - disable VCN clock gating dpg mode
757  *
758  * @adev: amdgpu_device pointer
759  * @sram_sel: sram select
760  * @inst_idx: instance number index
761  * @indirect: indirectly write sram
762  *
763  * Disable clock gating for VCN block with dpg mode
764  */
765 static void vcn_v4_0_disable_clock_gating_dpg_mode(struct amdgpu_device *adev, uint8_t sram_sel,
766       int inst_idx, uint8_t indirect)
767 {
768 	uint32_t reg_data = 0;
769 
770 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
771 		return;
772 
773 	/* enable sw clock gating control */
774 	reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
775 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
776 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
777 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
778 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
779 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
780 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
781 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
782 		 UVD_CGC_CTRL__SYS_MODE_MASK |
783 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
784 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
785 		 UVD_CGC_CTRL__REGS_MODE_MASK |
786 		 UVD_CGC_CTRL__RBC_MODE_MASK |
787 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
788 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
789 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
790 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
791 		 UVD_CGC_CTRL__MPC_MODE_MASK |
792 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
793 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
794 		 UVD_CGC_CTRL__WCB_MODE_MASK |
795 		 UVD_CGC_CTRL__VCPU_MODE_MASK);
796 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
797 		VCN, inst_idx, regUVD_CGC_CTRL), reg_data, sram_sel, indirect);
798 
799 	/* turn off clock gating */
800 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
801 		VCN, inst_idx, regUVD_CGC_GATE), 0, sram_sel, indirect);
802 
803 	/* turn on SUVD clock gating */
804 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
805 		VCN, inst_idx, regUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
806 
807 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
808 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
809 		VCN, inst_idx, regUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
810 }
811 
812 /**
813  * vcn_v4_0_enable_clock_gating - enable VCN clock gating
814  *
815  * @adev: amdgpu_device pointer
816  * @inst: instance number
817  *
818  * Enable clock gating for VCN block
819  */
820 static void vcn_v4_0_enable_clock_gating(struct amdgpu_device *adev, int inst)
821 {
822 	uint32_t data;
823 
824 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
825 		return;
826 
827 	/* enable VCN CGC */
828 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
829 	data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
830 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
831 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
832 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
833 
834 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
835 	data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
836 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
837 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
838 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
839 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
840 		| UVD_CGC_CTRL__SYS_MODE_MASK
841 		| UVD_CGC_CTRL__UDEC_MODE_MASK
842 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
843 		| UVD_CGC_CTRL__REGS_MODE_MASK
844 		| UVD_CGC_CTRL__RBC_MODE_MASK
845 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
846 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
847 		| UVD_CGC_CTRL__IDCT_MODE_MASK
848 		| UVD_CGC_CTRL__MPRD_MODE_MASK
849 		| UVD_CGC_CTRL__MPC_MODE_MASK
850 		| UVD_CGC_CTRL__LBSI_MODE_MASK
851 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
852 		| UVD_CGC_CTRL__WCB_MODE_MASK
853 		| UVD_CGC_CTRL__VCPU_MODE_MASK
854 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
855 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
856 
857 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
858 	data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
859 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
860 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
861 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
862 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
863 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
864 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
865 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
866 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
867 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
868 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
869 
870 	return;
871 }
872 
873 static void vcn_v4_0_enable_ras(struct amdgpu_device *adev, int inst_idx,
874 				bool indirect)
875 {
876 	uint32_t tmp;
877 
878 	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN))
879 		return;
880 
881 	tmp = VCN_RAS_CNTL__VCPU_VCODEC_REARM_MASK |
882 	      VCN_RAS_CNTL__VCPU_VCODEC_IH_EN_MASK |
883 	      VCN_RAS_CNTL__VCPU_VCODEC_PMI_EN_MASK |
884 	      VCN_RAS_CNTL__VCPU_VCODEC_STALL_EN_MASK;
885 	WREG32_SOC15_DPG_MODE(inst_idx,
886 			      SOC15_DPG_MODE_OFFSET(VCN, 0, regVCN_RAS_CNTL),
887 			      tmp, 0, indirect);
888 
889 	tmp = UVD_SYS_INT_EN__RASCNTL_VCPU_VCODEC_EN_MASK;
890 	WREG32_SOC15_DPG_MODE(inst_idx,
891 			      SOC15_DPG_MODE_OFFSET(VCN, 0, regUVD_SYS_INT_EN),
892 			      tmp, 0, indirect);
893 }
894 
895 /**
896  * vcn_v4_0_start_dpg_mode - VCN start with dpg mode
897  *
898  * @adev: amdgpu_device pointer
899  * @inst_idx: instance number index
900  * @indirect: indirectly write sram
901  *
902  * Start VCN block with dpg mode
903  */
904 static int vcn_v4_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
905 {
906 	volatile struct amdgpu_vcn4_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
907 	struct amdgpu_ring *ring;
908 	uint32_t tmp;
909 
910 	/* disable register anti-hang mechanism */
911 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 1,
912 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
913 	/* enable dynamic power gating mode */
914 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS);
915 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
916 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
917 	WREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS, tmp);
918 
919 	if (indirect)
920 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
921 
922 	/* enable clock gating */
923 	vcn_v4_0_disable_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
924 
925 	/* enable VCPU clock */
926 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
927 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK | UVD_VCPU_CNTL__BLK_RST_MASK;
928 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
929 		VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
930 
931 	/* disable master interupt */
932 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
933 		VCN, inst_idx, regUVD_MASTINT_EN), 0, 0, indirect);
934 
935 	/* setup regUVD_LMI_CTRL */
936 	tmp = (UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
937 		UVD_LMI_CTRL__REQ_MODE_MASK |
938 		UVD_LMI_CTRL__CRC_RESET_MASK |
939 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
940 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
941 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
942 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
943 		0x00100000L);
944 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
945 		VCN, inst_idx, regUVD_LMI_CTRL), tmp, 0, indirect);
946 
947 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
948 		VCN, inst_idx, regUVD_MPC_CNTL),
949 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
950 
951 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
952 		VCN, inst_idx, regUVD_MPC_SET_MUXA0),
953 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
954 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
955 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
956 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
957 
958 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
959 		VCN, inst_idx, regUVD_MPC_SET_MUXB0),
960 		 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
961 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
962 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
963 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
964 
965 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
966 		VCN, inst_idx, regUVD_MPC_SET_MUX),
967 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
968 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
969 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
970 
971 	vcn_v4_0_mc_resume_dpg_mode(adev, inst_idx, indirect);
972 
973 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
974 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
975 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
976 		VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
977 
978 	/* enable LMI MC and UMC channels */
979 	tmp = 0x1f << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT;
980 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
981 		VCN, inst_idx, regUVD_LMI_CTRL2), tmp, 0, indirect);
982 
983 	vcn_v4_0_enable_ras(adev, inst_idx, indirect);
984 
985 	/* enable master interrupt */
986 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
987 		VCN, inst_idx, regUVD_MASTINT_EN),
988 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
989 
990 
991 	if (indirect)
992 		psp_update_vcn_sram(adev, inst_idx, adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
993 			(uint32_t)((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
994 				(uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr));
995 
996 	ring = &adev->vcn.inst[inst_idx].ring_enc[0];
997 
998 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_LO, ring->gpu_addr);
999 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1000 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_SIZE, ring->ring_size / 4);
1001 
1002 	tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
1003 	tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
1004 	WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
1005 	fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
1006 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR, 0);
1007 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, 0);
1008 
1009 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR);
1010 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, tmp);
1011 	ring->wptr = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
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 | FW_QUEUE_DPG_HOLD_OFF);
1017 
1018 	WREG32_SOC15(VCN, inst_idx, regVCN_RB1_DB_CTRL,
1019 			ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
1020 			VCN_RB1_DB_CTRL__EN_MASK);
1021 
1022 	return 0;
1023 }
1024 
1025 
1026 /**
1027  * vcn_v4_0_start - VCN start
1028  *
1029  * @adev: amdgpu_device pointer
1030  *
1031  * Start VCN block
1032  */
1033 static int vcn_v4_0_start(struct amdgpu_device *adev)
1034 {
1035 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1036 	struct amdgpu_ring *ring;
1037 	uint32_t tmp;
1038 	int i, j, k, r;
1039 
1040 	if (adev->pm.dpm_enabled)
1041 		amdgpu_dpm_enable_uvd(adev, true);
1042 
1043 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1044 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1045 
1046 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1047 			r = vcn_v4_0_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
1048 			continue;
1049 		}
1050 
1051 		/* disable VCN power gating */
1052 		vcn_v4_0_disable_static_power_gating(adev, i);
1053 
1054 		/* set VCN status busy */
1055 		tmp = RREG32_SOC15(VCN, i, regUVD_STATUS) | UVD_STATUS__UVD_BUSY;
1056 		WREG32_SOC15(VCN, i, regUVD_STATUS, tmp);
1057 
1058 		/*SW clock gating */
1059 		vcn_v4_0_disable_clock_gating(adev, i);
1060 
1061 		/* enable VCPU clock */
1062 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1063 				UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1064 
1065 		/* disable master interrupt */
1066 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN), 0,
1067 				~UVD_MASTINT_EN__VCPU_EN_MASK);
1068 
1069 		/* enable LMI MC and UMC channels */
1070 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_LMI_CTRL2), 0,
1071 				~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1072 
1073 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1074 		tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1075 		tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1076 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1077 
1078 		/* setup regUVD_LMI_CTRL */
1079 		tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL);
1080 		WREG32_SOC15(VCN, i, regUVD_LMI_CTRL, tmp |
1081 				UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
1082 				UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1083 				UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1084 				UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1085 
1086 		/* setup regUVD_MPC_CNTL */
1087 		tmp = RREG32_SOC15(VCN, i, regUVD_MPC_CNTL);
1088 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1089 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1090 		WREG32_SOC15(VCN, i, regUVD_MPC_CNTL, tmp);
1091 
1092 		/* setup UVD_MPC_SET_MUXA0 */
1093 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXA0,
1094 				((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1095 				 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1096 				 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1097 				 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1098 
1099 		/* setup UVD_MPC_SET_MUXB0 */
1100 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXB0,
1101 				((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1102 				 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1103 				 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1104 				 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1105 
1106 		/* setup UVD_MPC_SET_MUX */
1107 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUX,
1108 				((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1109 				 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1110 				 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1111 
1112 		vcn_v4_0_mc_resume(adev, i);
1113 
1114 		/* VCN global tiling registers */
1115 		WREG32_SOC15(VCN, i, regUVD_GFX10_ADDR_CONFIG,
1116 				adev->gfx.config.gb_addr_config);
1117 
1118 		/* unblock VCPU register access */
1119 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL), 0,
1120 				~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1121 
1122 		/* release VCPU reset to boot */
1123 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1124 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1125 
1126 		for (j = 0; j < 10; ++j) {
1127 			uint32_t status;
1128 
1129 			for (k = 0; k < 100; ++k) {
1130 				status = RREG32_SOC15(VCN, i, regUVD_STATUS);
1131 				if (status & 2)
1132 					break;
1133 				mdelay(10);
1134 				if (amdgpu_emu_mode==1)
1135 					msleep(1);
1136 			}
1137 
1138 			if (amdgpu_emu_mode==1) {
1139 				r = -1;
1140 				if (status & 2) {
1141 					r = 0;
1142 					break;
1143 				}
1144 			} else {
1145 				r = 0;
1146 				if (status & 2)
1147 					break;
1148 
1149 				dev_err(adev->dev, "VCN[%d] is not responding, trying to reset the VCPU!!!\n", i);
1150 				WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1151 							UVD_VCPU_CNTL__BLK_RST_MASK,
1152 							~UVD_VCPU_CNTL__BLK_RST_MASK);
1153 				mdelay(10);
1154 				WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1155 						~UVD_VCPU_CNTL__BLK_RST_MASK);
1156 
1157 				mdelay(10);
1158 				r = -1;
1159 			}
1160 		}
1161 
1162 		if (r) {
1163 			dev_err(adev->dev, "VCN[%d] is not responding, giving up!!!\n", i);
1164 			return r;
1165 		}
1166 
1167 		/* enable master interrupt */
1168 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN),
1169 				UVD_MASTINT_EN__VCPU_EN_MASK,
1170 				~UVD_MASTINT_EN__VCPU_EN_MASK);
1171 
1172 		/* clear the busy bit of VCN_STATUS */
1173 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_STATUS), 0,
1174 				~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1175 
1176 		ring = &adev->vcn.inst[i].ring_enc[0];
1177 		WREG32_SOC15(VCN, i, regVCN_RB1_DB_CTRL,
1178 				ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
1179 				VCN_RB1_DB_CTRL__EN_MASK);
1180 
1181 		WREG32_SOC15(VCN, i, regUVD_RB_BASE_LO, ring->gpu_addr);
1182 		WREG32_SOC15(VCN, i, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1183 		WREG32_SOC15(VCN, i, regUVD_RB_SIZE, ring->ring_size / 4);
1184 
1185 		tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1186 		tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
1187 		WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1188 		fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
1189 		WREG32_SOC15(VCN, i, regUVD_RB_RPTR, 0);
1190 		WREG32_SOC15(VCN, i, regUVD_RB_WPTR, 0);
1191 
1192 		tmp = RREG32_SOC15(VCN, i, regUVD_RB_RPTR);
1193 		WREG32_SOC15(VCN, i, regUVD_RB_WPTR, tmp);
1194 		ring->wptr = RREG32_SOC15(VCN, i, regUVD_RB_WPTR);
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 | FW_QUEUE_DPG_HOLD_OFF);
1200 	}
1201 
1202 	return 0;
1203 }
1204 
1205 static int vcn_v4_0_start_sriov(struct amdgpu_device *adev)
1206 {
1207 	int i;
1208 	struct amdgpu_ring *ring_enc;
1209 	uint64_t cache_addr;
1210 	uint64_t rb_enc_addr;
1211 	uint64_t ctx_addr;
1212 	uint32_t param, resp, expected;
1213 	uint32_t offset, cache_size;
1214 	uint32_t tmp, timeout;
1215 
1216 	struct amdgpu_mm_table *table = &adev->virt.mm_table;
1217 	uint32_t *table_loc;
1218 	uint32_t table_size;
1219 	uint32_t size, size_dw;
1220 	uint32_t init_status;
1221 	uint32_t enabled_vcn;
1222 
1223 	struct mmsch_v4_0_cmd_direct_write
1224 		direct_wt = { {0} };
1225 	struct mmsch_v4_0_cmd_direct_read_modify_write
1226 		direct_rd_mod_wt = { {0} };
1227 	struct mmsch_v4_0_cmd_end end = { {0} };
1228 	struct mmsch_v4_0_init_header header;
1229 
1230 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1231 	volatile struct amdgpu_fw_shared_rb_setup *rb_setup;
1232 
1233 	direct_wt.cmd_header.command_type =
1234 		MMSCH_COMMAND__DIRECT_REG_WRITE;
1235 	direct_rd_mod_wt.cmd_header.command_type =
1236 		MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1237 	end.cmd_header.command_type =
1238 		MMSCH_COMMAND__END;
1239 
1240 	header.version = MMSCH_VERSION;
1241 	header.total_size = sizeof(struct mmsch_v4_0_init_header) >> 2;
1242 	for (i = 0; i < AMDGPU_MAX_VCN_INSTANCES; i++) {
1243 		header.inst[i].init_status = 0;
1244 		header.inst[i].table_offset = 0;
1245 		header.inst[i].table_size = 0;
1246 	}
1247 
1248 	table_loc = (uint32_t *)table->cpu_addr;
1249 	table_loc += header.total_size;
1250 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1251 		if (adev->vcn.harvest_config & (1 << i))
1252 			continue;
1253 
1254 		table_size = 0;
1255 
1256 		MMSCH_V4_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(VCN, i,
1257 			regUVD_STATUS),
1258 			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1259 
1260 		cache_size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1261 
1262 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1263 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1264 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1265 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1266 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1267 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1268 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1269 			offset = 0;
1270 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1271 				regUVD_VCPU_CACHE_OFFSET0),
1272 				0);
1273 		} else {
1274 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1275 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1276 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1277 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1278 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1279 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1280 			offset = cache_size;
1281 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1282 				regUVD_VCPU_CACHE_OFFSET0),
1283 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1284 		}
1285 
1286 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1287 			regUVD_VCPU_CACHE_SIZE0),
1288 			cache_size);
1289 
1290 		cache_addr = adev->vcn.inst[i].gpu_addr + offset;
1291 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1292 			regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1293 			lower_32_bits(cache_addr));
1294 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1295 			regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1296 			upper_32_bits(cache_addr));
1297 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1298 			regUVD_VCPU_CACHE_OFFSET1),
1299 			0);
1300 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1301 			regUVD_VCPU_CACHE_SIZE1),
1302 			AMDGPU_VCN_STACK_SIZE);
1303 
1304 		cache_addr = adev->vcn.inst[i].gpu_addr + offset +
1305 			AMDGPU_VCN_STACK_SIZE;
1306 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1307 			regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1308 			lower_32_bits(cache_addr));
1309 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1310 			regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1311 			upper_32_bits(cache_addr));
1312 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1313 			regUVD_VCPU_CACHE_OFFSET2),
1314 			0);
1315 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1316 			regUVD_VCPU_CACHE_SIZE2),
1317 			AMDGPU_VCN_CONTEXT_SIZE);
1318 
1319 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1320 		rb_setup = &fw_shared->rb_setup;
1321 
1322 		ring_enc = &adev->vcn.inst[i].ring_enc[0];
1323 		ring_enc->wptr = 0;
1324 		rb_enc_addr = ring_enc->gpu_addr;
1325 
1326 		rb_setup->is_rb_enabled_flags |= RB_ENABLED;
1327 		rb_setup->rb_addr_lo = lower_32_bits(rb_enc_addr);
1328 		rb_setup->rb_addr_hi = upper_32_bits(rb_enc_addr);
1329 		rb_setup->rb_size = ring_enc->ring_size / 4;
1330 		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
1331 
1332 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1333 			regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
1334 			lower_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
1335 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1336 			regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
1337 			upper_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
1338 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1339 			regUVD_VCPU_NONCACHE_SIZE0),
1340 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
1341 
1342 		/* add end packet */
1343 		MMSCH_V4_0_INSERT_END();
1344 
1345 		/* refine header */
1346 		header.inst[i].init_status = 0;
1347 		header.inst[i].table_offset = header.total_size;
1348 		header.inst[i].table_size = table_size;
1349 		header.total_size += table_size;
1350 	}
1351 
1352 	/* Update init table header in memory */
1353 	size = sizeof(struct mmsch_v4_0_init_header);
1354 	table_loc = (uint32_t *)table->cpu_addr;
1355 	memcpy((void *)table_loc, &header, size);
1356 
1357 	/* message MMSCH (in VCN[0]) to initialize this client
1358 	 * 1, write to mmsch_vf_ctx_addr_lo/hi register with GPU mc addr
1359 	 * of memory descriptor location
1360 	 */
1361 	ctx_addr = table->gpu_addr;
1362 	WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_ADDR_LO, lower_32_bits(ctx_addr));
1363 	WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_ADDR_HI, upper_32_bits(ctx_addr));
1364 
1365 	/* 2, update vmid of descriptor */
1366 	tmp = RREG32_SOC15(VCN, 0, regMMSCH_VF_VMID);
1367 	tmp &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1368 	/* use domain0 for MM scheduler */
1369 	tmp |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1370 	WREG32_SOC15(VCN, 0, regMMSCH_VF_VMID, tmp);
1371 
1372 	/* 3, notify mmsch about the size of this descriptor */
1373 	size = header.total_size;
1374 	WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_SIZE, size);
1375 
1376 	/* 4, set resp to zero */
1377 	WREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_RESP, 0);
1378 
1379 	/* 5, kick off the initialization and wait until
1380 	 * MMSCH_VF_MAILBOX_RESP becomes non-zero
1381 	 */
1382 	param = 0x00000001;
1383 	WREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_HOST, param);
1384 	tmp = 0;
1385 	timeout = 1000;
1386 	resp = 0;
1387 	expected = MMSCH_VF_MAILBOX_RESP__OK;
1388 	while (resp != expected) {
1389 		resp = RREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_RESP);
1390 		if (resp != 0)
1391 			break;
1392 
1393 		udelay(10);
1394 		tmp = tmp + 10;
1395 		if (tmp >= timeout) {
1396 			DRM_ERROR("failed to init MMSCH. TIME-OUT after %d usec"\
1397 				" waiting for regMMSCH_VF_MAILBOX_RESP "\
1398 				"(expected=0x%08x, readback=0x%08x)\n",
1399 				tmp, expected, resp);
1400 			return -EBUSY;
1401 		}
1402 	}
1403 	enabled_vcn = amdgpu_vcn_is_disabled_vcn(adev, VCN_DECODE_RING, 0) ? 1 : 0;
1404 	init_status = ((struct mmsch_v4_0_init_header *)(table_loc))->inst[enabled_vcn].init_status;
1405 	if (resp != expected && resp != MMSCH_VF_MAILBOX_RESP__INCOMPLETE
1406 	&& init_status != MMSCH_VF_ENGINE_STATUS__PASS)
1407 		DRM_ERROR("MMSCH init status is incorrect! readback=0x%08x, header init "\
1408 			"status for VCN%x: 0x%x\n", resp, enabled_vcn, init_status);
1409 
1410 	return 0;
1411 }
1412 
1413 /**
1414  * vcn_v4_0_stop_dpg_mode - VCN stop with dpg mode
1415  *
1416  * @adev: amdgpu_device pointer
1417  * @inst_idx: instance number index
1418  *
1419  * Stop VCN block with dpg mode
1420  */
1421 static void vcn_v4_0_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1422 {
1423 	uint32_t tmp;
1424 
1425 	/* Wait for power status to be 1 */
1426 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1427 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1428 
1429 	/* wait for read ptr to be equal to write ptr */
1430 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
1431 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1432 
1433 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1434 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1435 
1436 	/* disable dynamic power gating mode */
1437 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 0,
1438 		~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1439 }
1440 
1441 /**
1442  * vcn_v4_0_stop - VCN stop
1443  *
1444  * @adev: amdgpu_device pointer
1445  *
1446  * Stop VCN block
1447  */
1448 static int vcn_v4_0_stop(struct amdgpu_device *adev)
1449 {
1450 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1451 	uint32_t tmp;
1452 	int i, r = 0;
1453 
1454 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1455 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1456 		fw_shared->sq.queue_mode |= FW_QUEUE_DPG_HOLD_OFF;
1457 
1458 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1459 			vcn_v4_0_stop_dpg_mode(adev, i);
1460 			continue;
1461 		}
1462 
1463 		/* wait for vcn idle */
1464 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1465 		if (r)
1466 			return r;
1467 
1468 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1469 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1470 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1471 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1472 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1473 		if (r)
1474 			return r;
1475 
1476 		/* disable LMI UMC channel */
1477 		tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL2);
1478 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1479 		WREG32_SOC15(VCN, i, regUVD_LMI_CTRL2, tmp);
1480 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK |
1481 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1482 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1483 		if (r)
1484 			return r;
1485 
1486 		/* block VCPU register access */
1487 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL),
1488 				UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1489 				~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1490 
1491 		/* reset VCPU */
1492 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1493 				UVD_VCPU_CNTL__BLK_RST_MASK,
1494 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1495 
1496 		/* disable VCPU clock */
1497 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1498 				~(UVD_VCPU_CNTL__CLK_EN_MASK));
1499 
1500 		/* apply soft reset */
1501 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1502 		tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1503 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1504 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1505 		tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1506 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1507 
1508 		/* clear status */
1509 		WREG32_SOC15(VCN, i, regUVD_STATUS, 0);
1510 
1511 		/* apply HW clock gating */
1512 		vcn_v4_0_enable_clock_gating(adev, i);
1513 
1514 		/* enable VCN power gating */
1515 		vcn_v4_0_enable_static_power_gating(adev, i);
1516 	}
1517 
1518 	if (adev->pm.dpm_enabled)
1519 		amdgpu_dpm_enable_uvd(adev, false);
1520 
1521 	return 0;
1522 }
1523 
1524 /**
1525  * vcn_v4_0_pause_dpg_mode - VCN pause with dpg mode
1526  *
1527  * @adev: amdgpu_device pointer
1528  * @inst_idx: instance number index
1529  * @new_state: pause state
1530  *
1531  * Pause dpg mode for VCN block
1532  */
1533 static int vcn_v4_0_pause_dpg_mode(struct amdgpu_device *adev, int inst_idx,
1534       struct dpg_pause_state *new_state)
1535 {
1536 	uint32_t reg_data = 0;
1537 	int ret_code;
1538 
1539 	/* pause/unpause if state is changed */
1540 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1541 		DRM_DEV_DEBUG(adev->dev, "dpg pause state changed %d -> %d",
1542 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1543 		reg_data = RREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE) &
1544 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1545 
1546 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1547 			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 0x1,
1548 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1549 
1550 			if (!ret_code) {
1551 				/* pause DPG */
1552 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1553 				WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1554 
1555 				/* wait for ACK */
1556 				SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_DPG_PAUSE,
1557 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1558 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1559 
1560 				SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS,
1561 					UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1562 			}
1563 		} else {
1564 			/* unpause dpg, no need to wait */
1565 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1566 			WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1567 		}
1568 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1569 	}
1570 
1571 	return 0;
1572 }
1573 
1574 /**
1575  * vcn_v4_0_unified_ring_get_rptr - get unified read pointer
1576  *
1577  * @ring: amdgpu_ring pointer
1578  *
1579  * Returns the current hardware unified read pointer
1580  */
1581 static uint64_t vcn_v4_0_unified_ring_get_rptr(struct amdgpu_ring *ring)
1582 {
1583 	struct amdgpu_device *adev = ring->adev;
1584 
1585 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1586 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1587 
1588 	return RREG32_SOC15(VCN, ring->me, regUVD_RB_RPTR);
1589 }
1590 
1591 /**
1592  * vcn_v4_0_unified_ring_get_wptr - get unified write pointer
1593  *
1594  * @ring: amdgpu_ring pointer
1595  *
1596  * Returns the current hardware unified write pointer
1597  */
1598 static uint64_t vcn_v4_0_unified_ring_get_wptr(struct amdgpu_ring *ring)
1599 {
1600 	struct amdgpu_device *adev = ring->adev;
1601 
1602 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1603 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1604 
1605 	if (ring->use_doorbell)
1606 		return *ring->wptr_cpu_addr;
1607 	else
1608 		return RREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR);
1609 }
1610 
1611 /**
1612  * vcn_v4_0_unified_ring_set_wptr - set enc write pointer
1613  *
1614  * @ring: amdgpu_ring pointer
1615  *
1616  * Commits the enc write pointer to the hardware
1617  */
1618 static void vcn_v4_0_unified_ring_set_wptr(struct amdgpu_ring *ring)
1619 {
1620 	struct amdgpu_device *adev = ring->adev;
1621 
1622 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1623 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1624 
1625 	if (ring->use_doorbell) {
1626 		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1627 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1628 	} else {
1629 		WREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR, lower_32_bits(ring->wptr));
1630 	}
1631 }
1632 
1633 static int vcn_v4_0_limit_sched(struct amdgpu_cs_parser *p,
1634 				struct amdgpu_job *job)
1635 {
1636 	struct drm_gpu_scheduler **scheds;
1637 
1638 	/* The create msg must be in the first IB submitted */
1639 	if (atomic_read(&job->base.entity->fence_seq))
1640 		return -EINVAL;
1641 
1642 	/* if VCN0 is harvested, we can't support AV1 */
1643 	if (p->adev->vcn.harvest_config & AMDGPU_VCN_HARVEST_VCN0)
1644 		return -EINVAL;
1645 
1646 	scheds = p->adev->gpu_sched[AMDGPU_HW_IP_VCN_ENC]
1647 		[AMDGPU_RING_PRIO_0].sched;
1648 	drm_sched_entity_modify_sched(job->base.entity, scheds, 1);
1649 	return 0;
1650 }
1651 
1652 static int vcn_v4_0_dec_msg(struct amdgpu_cs_parser *p, struct amdgpu_job *job,
1653 			    uint64_t addr)
1654 {
1655 	struct ttm_operation_ctx ctx = { false, false };
1656 	struct amdgpu_bo_va_mapping *map;
1657 	uint32_t *msg, num_buffers;
1658 	struct amdgpu_bo *bo;
1659 	uint64_t start, end;
1660 	unsigned int i;
1661 	void *ptr;
1662 	int r;
1663 
1664 	addr &= AMDGPU_GMC_HOLE_MASK;
1665 	r = amdgpu_cs_find_mapping(p, addr, &bo, &map);
1666 	if (r) {
1667 		DRM_ERROR("Can't find BO for addr 0x%08llx\n", addr);
1668 		return r;
1669 	}
1670 
1671 	start = map->start * AMDGPU_GPU_PAGE_SIZE;
1672 	end = (map->last + 1) * AMDGPU_GPU_PAGE_SIZE;
1673 	if (addr & 0x7) {
1674 		DRM_ERROR("VCN messages must be 8 byte aligned!\n");
1675 		return -EINVAL;
1676 	}
1677 
1678 	bo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
1679 	amdgpu_bo_placement_from_domain(bo, bo->allowed_domains);
1680 	r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
1681 	if (r) {
1682 		DRM_ERROR("Failed validating the VCN message BO (%d)!\n", r);
1683 		return r;
1684 	}
1685 
1686 	r = amdgpu_bo_kmap(bo, &ptr);
1687 	if (r) {
1688 		DRM_ERROR("Failed mapping the VCN message (%d)!\n", r);
1689 		return r;
1690 	}
1691 
1692 	msg = ptr + addr - start;
1693 
1694 	/* Check length */
1695 	if (msg[1] > end - addr) {
1696 		r = -EINVAL;
1697 		goto out;
1698 	}
1699 
1700 	if (msg[3] != RDECODE_MSG_CREATE)
1701 		goto out;
1702 
1703 	num_buffers = msg[2];
1704 	for (i = 0, msg = &msg[6]; i < num_buffers; ++i, msg += 4) {
1705 		uint32_t offset, size, *create;
1706 
1707 		if (msg[0] != RDECODE_MESSAGE_CREATE)
1708 			continue;
1709 
1710 		offset = msg[1];
1711 		size = msg[2];
1712 
1713 		if (offset + size > end) {
1714 			r = -EINVAL;
1715 			goto out;
1716 		}
1717 
1718 		create = ptr + addr + offset - start;
1719 
1720 		/* H264, HEVC and VP9 can run on any instance */
1721 		if (create[0] == 0x7 || create[0] == 0x10 || create[0] == 0x11)
1722 			continue;
1723 
1724 		r = vcn_v4_0_limit_sched(p, job);
1725 		if (r)
1726 			goto out;
1727 	}
1728 
1729 out:
1730 	amdgpu_bo_kunmap(bo);
1731 	return r;
1732 }
1733 
1734 #define RADEON_VCN_ENGINE_TYPE_ENCODE			(0x00000002)
1735 #define RADEON_VCN_ENGINE_TYPE_DECODE			(0x00000003)
1736 
1737 #define RADEON_VCN_ENGINE_INFO				(0x30000001)
1738 #define RADEON_VCN_ENGINE_INFO_MAX_OFFSET		16
1739 
1740 #define RENCODE_ENCODE_STANDARD_AV1			2
1741 #define RENCODE_IB_PARAM_SESSION_INIT			0x00000003
1742 #define RENCODE_IB_PARAM_SESSION_INIT_MAX_OFFSET	64
1743 
1744 /* return the offset in ib if id is found, -1 otherwise
1745  * to speed up the searching we only search upto max_offset
1746  */
1747 static int vcn_v4_0_enc_find_ib_param(struct amdgpu_ib *ib, uint32_t id, int max_offset)
1748 {
1749 	int i;
1750 
1751 	for (i = 0; i < ib->length_dw && i < max_offset && ib->ptr[i] >= 8; i += ib->ptr[i]/4) {
1752 		if (ib->ptr[i + 1] == id)
1753 			return i;
1754 	}
1755 	return -1;
1756 }
1757 
1758 static int vcn_v4_0_ring_patch_cs_in_place(struct amdgpu_cs_parser *p,
1759 					   struct amdgpu_job *job,
1760 					   struct amdgpu_ib *ib)
1761 {
1762 	struct amdgpu_ring *ring = amdgpu_job_ring(job);
1763 	struct amdgpu_vcn_decode_buffer *decode_buffer;
1764 	uint64_t addr;
1765 	uint32_t val;
1766 	int idx;
1767 
1768 	/* The first instance can decode anything */
1769 	if (!ring->me)
1770 		return 0;
1771 
1772 	/* RADEON_VCN_ENGINE_INFO is at the top of ib block */
1773 	idx = vcn_v4_0_enc_find_ib_param(ib, RADEON_VCN_ENGINE_INFO,
1774 			RADEON_VCN_ENGINE_INFO_MAX_OFFSET);
1775 	if (idx < 0) /* engine info is missing */
1776 		return 0;
1777 
1778 	val = amdgpu_ib_get_value(ib, idx + 2); /* RADEON_VCN_ENGINE_TYPE */
1779 	if (val == RADEON_VCN_ENGINE_TYPE_DECODE) {
1780 		decode_buffer = (struct amdgpu_vcn_decode_buffer *)&ib->ptr[idx + 6];
1781 
1782 		if (!(decode_buffer->valid_buf_flag  & 0x1))
1783 			return 0;
1784 
1785 		addr = ((u64)decode_buffer->msg_buffer_address_hi) << 32 |
1786 			decode_buffer->msg_buffer_address_lo;
1787 		return vcn_v4_0_dec_msg(p, job, addr);
1788 	} else if (val == RADEON_VCN_ENGINE_TYPE_ENCODE) {
1789 		idx = vcn_v4_0_enc_find_ib_param(ib, RENCODE_IB_PARAM_SESSION_INIT,
1790 			RENCODE_IB_PARAM_SESSION_INIT_MAX_OFFSET);
1791 		if (idx >= 0 && ib->ptr[idx + 2] == RENCODE_ENCODE_STANDARD_AV1)
1792 			return vcn_v4_0_limit_sched(p, job);
1793 	}
1794 	return 0;
1795 }
1796 
1797 static const struct amdgpu_ring_funcs vcn_v4_0_unified_ring_vm_funcs = {
1798 	.type = AMDGPU_RING_TYPE_VCN_ENC,
1799 	.align_mask = 0x3f,
1800 	.nop = VCN_ENC_CMD_NO_OP,
1801 	.vmhub = AMDGPU_MMHUB_0,
1802 	.get_rptr = vcn_v4_0_unified_ring_get_rptr,
1803 	.get_wptr = vcn_v4_0_unified_ring_get_wptr,
1804 	.set_wptr = vcn_v4_0_unified_ring_set_wptr,
1805 	.patch_cs_in_place = vcn_v4_0_ring_patch_cs_in_place,
1806 	.emit_frame_size =
1807 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1808 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1809 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1810 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1811 		1, /* vcn_v2_0_enc_ring_insert_end */
1812 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1813 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
1814 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
1815 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1816 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
1817 	.test_ib = amdgpu_vcn_unified_ring_test_ib,
1818 	.insert_nop = amdgpu_ring_insert_nop,
1819 	.insert_end = vcn_v2_0_enc_ring_insert_end,
1820 	.pad_ib = amdgpu_ring_generic_pad_ib,
1821 	.begin_use = amdgpu_vcn_ring_begin_use,
1822 	.end_use = amdgpu_vcn_ring_end_use,
1823 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1824 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1825 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1826 };
1827 
1828 /**
1829  * vcn_v4_0_set_unified_ring_funcs - set unified ring functions
1830  *
1831  * @adev: amdgpu_device pointer
1832  *
1833  * Set unified ring functions
1834  */
1835 static void vcn_v4_0_set_unified_ring_funcs(struct amdgpu_device *adev)
1836 {
1837 	int i;
1838 
1839 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1840 		if (adev->vcn.harvest_config & (1 << i))
1841 			continue;
1842 
1843 		adev->vcn.inst[i].ring_enc[0].funcs = &vcn_v4_0_unified_ring_vm_funcs;
1844 		adev->vcn.inst[i].ring_enc[0].me = i;
1845 
1846 		DRM_INFO("VCN(%d) encode/decode are enabled in VM mode\n", i);
1847 	}
1848 }
1849 
1850 /**
1851  * vcn_v4_0_is_idle - check VCN block is idle
1852  *
1853  * @handle: amdgpu_device pointer
1854  *
1855  * Check whether VCN block is idle
1856  */
1857 static bool vcn_v4_0_is_idle(void *handle)
1858 {
1859 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1860 	int i, ret = 1;
1861 
1862 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1863 		if (adev->vcn.harvest_config & (1 << i))
1864 			continue;
1865 
1866 		ret &= (RREG32_SOC15(VCN, i, regUVD_STATUS) == UVD_STATUS__IDLE);
1867 	}
1868 
1869 	return ret;
1870 }
1871 
1872 /**
1873  * vcn_v4_0_wait_for_idle - wait for VCN block idle
1874  *
1875  * @handle: amdgpu_device pointer
1876  *
1877  * Wait for VCN block idle
1878  */
1879 static int vcn_v4_0_wait_for_idle(void *handle)
1880 {
1881 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1882 	int i, ret = 0;
1883 
1884 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1885 		if (adev->vcn.harvest_config & (1 << i))
1886 			continue;
1887 
1888 		ret = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE,
1889 			UVD_STATUS__IDLE);
1890 		if (ret)
1891 			return ret;
1892 	}
1893 
1894 	return ret;
1895 }
1896 
1897 /**
1898  * vcn_v4_0_set_clockgating_state - set VCN block clockgating state
1899  *
1900  * @handle: amdgpu_device pointer
1901  * @state: clock gating state
1902  *
1903  * Set VCN block clockgating state
1904  */
1905 static int vcn_v4_0_set_clockgating_state(void *handle, enum amd_clockgating_state state)
1906 {
1907 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1908 	bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
1909 	int i;
1910 
1911 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1912 		if (adev->vcn.harvest_config & (1 << i))
1913 			continue;
1914 
1915 		if (enable) {
1916 			if (RREG32_SOC15(VCN, i, regUVD_STATUS) != UVD_STATUS__IDLE)
1917 				return -EBUSY;
1918 			vcn_v4_0_enable_clock_gating(adev, i);
1919 		} else {
1920 			vcn_v4_0_disable_clock_gating(adev, i);
1921 		}
1922 	}
1923 
1924 	return 0;
1925 }
1926 
1927 /**
1928  * vcn_v4_0_set_powergating_state - set VCN block powergating state
1929  *
1930  * @handle: amdgpu_device pointer
1931  * @state: power gating state
1932  *
1933  * Set VCN block powergating state
1934  */
1935 static int vcn_v4_0_set_powergating_state(void *handle, enum amd_powergating_state state)
1936 {
1937 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1938 	int ret;
1939 
1940 	/* for SRIOV, guest should not control VCN Power-gating
1941 	 * MMSCH FW should control Power-gating and clock-gating
1942 	 * guest should avoid touching CGC and PG
1943 	 */
1944 	if (amdgpu_sriov_vf(adev)) {
1945 		adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
1946 		return 0;
1947 	}
1948 
1949 	if(state == adev->vcn.cur_state)
1950 		return 0;
1951 
1952 	if (state == AMD_PG_STATE_GATE)
1953 		ret = vcn_v4_0_stop(adev);
1954 	else
1955 		ret = vcn_v4_0_start(adev);
1956 
1957 	if(!ret)
1958 		adev->vcn.cur_state = state;
1959 
1960 	return ret;
1961 }
1962 
1963 /**
1964  * vcn_v4_0_set_interrupt_state - set VCN block interrupt state
1965  *
1966  * @adev: amdgpu_device pointer
1967  * @source: interrupt sources
1968  * @type: interrupt types
1969  * @state: interrupt states
1970  *
1971  * Set VCN block interrupt state
1972  */
1973 static int vcn_v4_0_set_interrupt_state(struct amdgpu_device *adev, struct amdgpu_irq_src *source,
1974       unsigned type, enum amdgpu_interrupt_state state)
1975 {
1976 	return 0;
1977 }
1978 
1979 /**
1980  * vcn_v4_0_process_interrupt - process VCN block interrupt
1981  *
1982  * @adev: amdgpu_device pointer
1983  * @source: interrupt sources
1984  * @entry: interrupt entry from clients and sources
1985  *
1986  * Process VCN block interrupt
1987  */
1988 static int vcn_v4_0_process_interrupt(struct amdgpu_device *adev, struct amdgpu_irq_src *source,
1989       struct amdgpu_iv_entry *entry)
1990 {
1991 	uint32_t ip_instance;
1992 
1993 	switch (entry->client_id) {
1994 	case SOC15_IH_CLIENTID_VCN:
1995 		ip_instance = 0;
1996 		break;
1997 	case SOC15_IH_CLIENTID_VCN1:
1998 		ip_instance = 1;
1999 		break;
2000 	default:
2001 		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
2002 		return 0;
2003 	}
2004 
2005 	DRM_DEBUG("IH: VCN TRAP\n");
2006 
2007 	switch (entry->src_id) {
2008 	case VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
2009 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
2010 		break;
2011 	case VCN_4_0__SRCID_UVD_POISON:
2012 		amdgpu_vcn_process_poison_irq(adev, source, entry);
2013 		break;
2014 	default:
2015 		DRM_ERROR("Unhandled interrupt: %d %d\n",
2016 			  entry->src_id, entry->src_data[0]);
2017 		break;
2018 	}
2019 
2020 	return 0;
2021 }
2022 
2023 static const struct amdgpu_irq_src_funcs vcn_v4_0_irq_funcs = {
2024 	.set = vcn_v4_0_set_interrupt_state,
2025 	.process = vcn_v4_0_process_interrupt,
2026 };
2027 
2028 /**
2029  * vcn_v4_0_set_irq_funcs - set VCN block interrupt irq functions
2030  *
2031  * @adev: amdgpu_device pointer
2032  *
2033  * Set VCN block interrupt irq functions
2034  */
2035 static void vcn_v4_0_set_irq_funcs(struct amdgpu_device *adev)
2036 {
2037 	int i;
2038 
2039 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2040 		if (adev->vcn.harvest_config & (1 << i))
2041 			continue;
2042 
2043 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
2044 		adev->vcn.inst[i].irq.funcs = &vcn_v4_0_irq_funcs;
2045 	}
2046 }
2047 
2048 static const struct amd_ip_funcs vcn_v4_0_ip_funcs = {
2049 	.name = "vcn_v4_0",
2050 	.early_init = vcn_v4_0_early_init,
2051 	.late_init = NULL,
2052 	.sw_init = vcn_v4_0_sw_init,
2053 	.sw_fini = vcn_v4_0_sw_fini,
2054 	.hw_init = vcn_v4_0_hw_init,
2055 	.hw_fini = vcn_v4_0_hw_fini,
2056 	.suspend = vcn_v4_0_suspend,
2057 	.resume = vcn_v4_0_resume,
2058 	.is_idle = vcn_v4_0_is_idle,
2059 	.wait_for_idle = vcn_v4_0_wait_for_idle,
2060 	.check_soft_reset = NULL,
2061 	.pre_soft_reset = NULL,
2062 	.soft_reset = NULL,
2063 	.post_soft_reset = NULL,
2064 	.set_clockgating_state = vcn_v4_0_set_clockgating_state,
2065 	.set_powergating_state = vcn_v4_0_set_powergating_state,
2066 };
2067 
2068 const struct amdgpu_ip_block_version vcn_v4_0_ip_block =
2069 {
2070 	.type = AMD_IP_BLOCK_TYPE_VCN,
2071 	.major = 4,
2072 	.minor = 0,
2073 	.rev = 0,
2074 	.funcs = &vcn_v4_0_ip_funcs,
2075 };
2076 
2077 static uint32_t vcn_v4_0_query_poison_by_instance(struct amdgpu_device *adev,
2078 			uint32_t instance, uint32_t sub_block)
2079 {
2080 	uint32_t poison_stat = 0, reg_value = 0;
2081 
2082 	switch (sub_block) {
2083 	case AMDGPU_VCN_V4_0_VCPU_VCODEC:
2084 		reg_value = RREG32_SOC15(VCN, instance, regUVD_RAS_VCPU_VCODEC_STATUS);
2085 		poison_stat = REG_GET_FIELD(reg_value, UVD_RAS_VCPU_VCODEC_STATUS, POISONED_PF);
2086 		break;
2087 	default:
2088 		break;
2089 	}
2090 
2091 	if (poison_stat)
2092 		dev_info(adev->dev, "Poison detected in VCN%d, sub_block%d\n",
2093 			instance, sub_block);
2094 
2095 	return poison_stat;
2096 }
2097 
2098 static bool vcn_v4_0_query_ras_poison_status(struct amdgpu_device *adev)
2099 {
2100 	uint32_t inst, sub;
2101 	uint32_t poison_stat = 0;
2102 
2103 	for (inst = 0; inst < adev->vcn.num_vcn_inst; inst++)
2104 		for (sub = 0; sub < AMDGPU_VCN_V4_0_MAX_SUB_BLOCK; sub++)
2105 			poison_stat +=
2106 				vcn_v4_0_query_poison_by_instance(adev, inst, sub);
2107 
2108 	return !!poison_stat;
2109 }
2110 
2111 const struct amdgpu_ras_block_hw_ops vcn_v4_0_ras_hw_ops = {
2112 	.query_poison_status = vcn_v4_0_query_ras_poison_status,
2113 };
2114 
2115 static struct amdgpu_vcn_ras vcn_v4_0_ras = {
2116 	.ras_block = {
2117 		.hw_ops = &vcn_v4_0_ras_hw_ops,
2118 	},
2119 };
2120 
2121 static void vcn_v4_0_set_ras_funcs(struct amdgpu_device *adev)
2122 {
2123 	switch (adev->ip_versions[VCN_HWIP][0]) {
2124 	case IP_VERSION(4, 0, 0):
2125 		adev->vcn.ras = &vcn_v4_0_ras;
2126 		break;
2127 	default:
2128 		break;
2129 	}
2130 }
2131