xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c (revision 078b39c9)
1 /*
2  * Copyright 2021 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #include <linux/firmware.h>
25 #include "amdgpu.h"
26 #include "amdgpu_vcn.h"
27 #include "amdgpu_pm.h"
28 #include "amdgpu_cs.h"
29 #include "soc15.h"
30 #include "soc15d.h"
31 #include "soc15_hw_ip.h"
32 #include "vcn_v2_0.h"
33 #include "mmsch_v4_0.h"
34 #include "vcn_v4_0.h"
35 
36 #include "vcn/vcn_4_0_0_offset.h"
37 #include "vcn/vcn_4_0_0_sh_mask.h"
38 #include "ivsrcid/vcn/irqsrcs_vcn_4_0.h"
39 
40 #include <drm/drm_drv.h>
41 
42 #define mmUVD_DPG_LMA_CTL							regUVD_DPG_LMA_CTL
43 #define mmUVD_DPG_LMA_CTL_BASE_IDX						regUVD_DPG_LMA_CTL_BASE_IDX
44 #define mmUVD_DPG_LMA_DATA							regUVD_DPG_LMA_DATA
45 #define mmUVD_DPG_LMA_DATA_BASE_IDX						regUVD_DPG_LMA_DATA_BASE_IDX
46 
47 #define VCN_VID_SOC_ADDRESS_2_0							0x1fb00
48 #define VCN1_VID_SOC_ADDRESS_3_0						0x48300
49 
50 #define VCN_HARVEST_MMSCH								0
51 
52 #define RDECODE_MSG_CREATE							0x00000000
53 #define RDECODE_MESSAGE_CREATE							0x00000001
54 
55 static int amdgpu_ih_clientid_vcns[] = {
56 	SOC15_IH_CLIENTID_VCN,
57 	SOC15_IH_CLIENTID_VCN1
58 };
59 
60 static int vcn_v4_0_start_sriov(struct amdgpu_device *adev);
61 static void vcn_v4_0_set_unified_ring_funcs(struct amdgpu_device *adev);
62 static void vcn_v4_0_set_irq_funcs(struct amdgpu_device *adev);
63 static int vcn_v4_0_set_powergating_state(void *handle,
64         enum amd_powergating_state state);
65 static int vcn_v4_0_pause_dpg_mode(struct amdgpu_device *adev,
66         int inst_idx, struct dpg_pause_state *new_state);
67 static void vcn_v4_0_unified_ring_set_wptr(struct amdgpu_ring *ring);
68 static void vcn_v4_0_set_ras_funcs(struct amdgpu_device *adev);
69 
70 /**
71  * vcn_v4_0_early_init - set function pointers and load microcode
72  *
73  * @handle: amdgpu_device pointer
74  *
75  * Set ring and irq function pointers
76  * Load microcode from filesystem
77  */
78 static int vcn_v4_0_early_init(void *handle)
79 {
80 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
81 	int i;
82 
83 	if (amdgpu_sriov_vf(adev)) {
84 		adev->vcn.harvest_config = VCN_HARVEST_MMSCH;
85 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
86 			if (amdgpu_vcn_is_disabled_vcn(adev, VCN_ENCODE_RING, i)) {
87 				adev->vcn.harvest_config |= 1 << i;
88 				dev_info(adev->dev, "VCN%d is disabled by hypervisor\n", i);
89 			}
90 		}
91 	}
92 
93 	/* re-use enc ring as unified ring */
94 	adev->vcn.num_enc_rings = 1;
95 
96 	vcn_v4_0_set_unified_ring_funcs(adev);
97 	vcn_v4_0_set_irq_funcs(adev);
98 	vcn_v4_0_set_ras_funcs(adev);
99 
100 	return amdgpu_vcn_early_init(adev);
101 }
102 
103 /**
104  * vcn_v4_0_sw_init - sw init for VCN block
105  *
106  * @handle: amdgpu_device pointer
107  *
108  * Load firmware and sw initialization
109  */
110 static int vcn_v4_0_sw_init(void *handle)
111 {
112 	struct amdgpu_ring *ring;
113 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
114 	int i, r;
115 
116 	r = amdgpu_vcn_sw_init(adev);
117 	if (r)
118 		return r;
119 
120 	amdgpu_vcn_setup_ucode(adev);
121 
122 	r = amdgpu_vcn_resume(adev);
123 	if (r)
124 		return r;
125 
126 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
127 		volatile struct amdgpu_vcn4_fw_shared *fw_shared;
128 
129 		if (adev->vcn.harvest_config & (1 << i))
130 			continue;
131 
132 		/* Init instance 0 sched_score to 1, so it's scheduled after other instances */
133 		if (i == 0)
134 			atomic_set(&adev->vcn.inst[i].sched_score, 1);
135 		else
136 			atomic_set(&adev->vcn.inst[i].sched_score, 0);
137 
138 		/* VCN UNIFIED TRAP */
139 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
140 				VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);
141 		if (r)
142 			return r;
143 
144 		/* VCN POISON TRAP */
145 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
146 				VCN_4_0__SRCID_UVD_POISON, &adev->vcn.inst[i].ras_poison_irq);
147 		if (r)
148 			return r;
149 
150 		ring = &adev->vcn.inst[i].ring_enc[0];
151 		ring->use_doorbell = true;
152 		if (amdgpu_sriov_vf(adev))
153 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + i * (adev->vcn.num_enc_rings + 1) + 1;
154 		else
155 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + 8 * i;
156 		ring->vm_hub = AMDGPU_MMHUB0(0);
157 		sprintf(ring->name, "vcn_unified_%d", i);
158 
159 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
160 						AMDGPU_RING_PRIO_0, &adev->vcn.inst[i].sched_score);
161 		if (r)
162 			return r;
163 
164 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
165 		fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_FW_SHARED_FLAG_0_UNIFIED_QUEUE);
166 		fw_shared->sq.is_enabled = 1;
167 
168 		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_SMU_DPM_INTERFACE_FLAG);
169 		fw_shared->smu_dpm_interface.smu_interface_type = (adev->flags & AMD_IS_APU) ?
170 			AMDGPU_VCN_SMU_DPM_INTERFACE_APU : AMDGPU_VCN_SMU_DPM_INTERFACE_DGPU;
171 
172 		if (amdgpu_sriov_vf(adev))
173 			fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
174 
175 		if (amdgpu_vcnfw_log)
176 			amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
177 	}
178 
179 	if (amdgpu_sriov_vf(adev)) {
180 		r = amdgpu_virt_alloc_mm_table(adev);
181 		if (r)
182 			return r;
183 	}
184 
185 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
186 		adev->vcn.pause_dpg_mode = vcn_v4_0_pause_dpg_mode;
187 
188 	r = amdgpu_vcn_ras_sw_init(adev);
189 	if (r)
190 		return r;
191 
192 	return 0;
193 }
194 
195 /**
196  * vcn_v4_0_sw_fini - sw fini for VCN block
197  *
198  * @handle: amdgpu_device pointer
199  *
200  * VCN suspend and free up sw allocation
201  */
202 static int vcn_v4_0_sw_fini(void *handle)
203 {
204 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
205 	int i, r, idx;
206 
207 	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
208 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
209 			volatile struct amdgpu_vcn4_fw_shared *fw_shared;
210 
211 			if (adev->vcn.harvest_config & (1 << i))
212 				continue;
213 
214 			fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
215 			fw_shared->present_flag_0 = 0;
216 			fw_shared->sq.is_enabled = 0;
217 		}
218 
219 		drm_dev_exit(idx);
220 	}
221 
222 	if (amdgpu_sriov_vf(adev))
223 		amdgpu_virt_free_mm_table(adev);
224 
225 	r = amdgpu_vcn_suspend(adev);
226 	if (r)
227 		return r;
228 
229 	r = amdgpu_vcn_sw_fini(adev);
230 
231 	return r;
232 }
233 
234 /**
235  * vcn_v4_0_hw_init - start and test VCN block
236  *
237  * @handle: amdgpu_device pointer
238  *
239  * Initialize the hardware, boot up the VCPU and do some testing
240  */
241 static int vcn_v4_0_hw_init(void *handle)
242 {
243 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
244 	struct amdgpu_ring *ring;
245 	int i, r;
246 
247 	if (amdgpu_sriov_vf(adev)) {
248 		r = vcn_v4_0_start_sriov(adev);
249 		if (r)
250 			goto done;
251 
252 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
253 			if (adev->vcn.harvest_config & (1 << i))
254 				continue;
255 
256 			ring = &adev->vcn.inst[i].ring_enc[0];
257 			ring->wptr = 0;
258 			ring->wptr_old = 0;
259 			vcn_v4_0_unified_ring_set_wptr(ring);
260 			ring->sched.ready = true;
261 
262 		}
263 	} else {
264 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
265 			if (adev->vcn.harvest_config & (1 << i))
266 				continue;
267 
268 			ring = &adev->vcn.inst[i].ring_enc[0];
269 
270 			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
271 					((adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i), i);
272 
273 			r = amdgpu_ring_test_helper(ring);
274 			if (r)
275 				goto done;
276 
277 		}
278 	}
279 
280 done:
281 	if (!r)
282 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
283 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
284 
285 	return r;
286 }
287 
288 /**
289  * vcn_v4_0_hw_fini - stop the hardware block
290  *
291  * @handle: amdgpu_device pointer
292  *
293  * Stop the VCN block, mark ring as not ready any more
294  */
295 static int vcn_v4_0_hw_fini(void *handle)
296 {
297 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
298 	int i;
299 
300 	cancel_delayed_work_sync(&adev->vcn.idle_work);
301 
302 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
303 		if (adev->vcn.harvest_config & (1 << i))
304 			continue;
305 		if (!amdgpu_sriov_vf(adev)) {
306 			if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
307                         (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
308                                 RREG32_SOC15(VCN, i, regUVD_STATUS))) {
309                         vcn_v4_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
310 			}
311 		}
312 		if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN))
313 			amdgpu_irq_put(adev, &adev->vcn.inst[i].ras_poison_irq, 0);
314 	}
315 
316 	return 0;
317 }
318 
319 /**
320  * vcn_v4_0_suspend - suspend VCN block
321  *
322  * @handle: amdgpu_device pointer
323  *
324  * HW fini and suspend VCN block
325  */
326 static int vcn_v4_0_suspend(void *handle)
327 {
328 	int r;
329 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
330 
331 	r = vcn_v4_0_hw_fini(adev);
332 	if (r)
333 		return r;
334 
335 	r = amdgpu_vcn_suspend(adev);
336 
337 	return r;
338 }
339 
340 /**
341  * vcn_v4_0_resume - resume VCN block
342  *
343  * @handle: amdgpu_device pointer
344  *
345  * Resume firmware and hw init VCN block
346  */
347 static int vcn_v4_0_resume(void *handle)
348 {
349 	int r;
350 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
351 
352 	r = amdgpu_vcn_resume(adev);
353 	if (r)
354 		return r;
355 
356 	r = vcn_v4_0_hw_init(adev);
357 
358 	return r;
359 }
360 
361 /**
362  * vcn_v4_0_mc_resume - memory controller programming
363  *
364  * @adev: amdgpu_device pointer
365  * @inst: instance number
366  *
367  * Let the VCN memory controller know it's offsets
368  */
369 static void vcn_v4_0_mc_resume(struct amdgpu_device *adev, int inst)
370 {
371 	uint32_t offset, size;
372 	const struct common_firmware_header *hdr;
373 
374 	hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
375 	size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
376 
377 	/* cache window 0: fw */
378 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
379 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
380 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo));
381 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
382 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi));
383 		WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, 0);
384 		offset = 0;
385 	} else {
386 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
387 			lower_32_bits(adev->vcn.inst[inst].gpu_addr));
388 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
389 			upper_32_bits(adev->vcn.inst[inst].gpu_addr));
390 		offset = size;
391                 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
392 	}
393 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE0, size);
394 
395 	/* cache window 1: stack */
396 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
397 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
398 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
399 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
400 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET1, 0);
401 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
402 
403 	/* cache window 2: context */
404 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
405 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
406 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
407 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
408 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET2, 0);
409 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
410 
411 	/* non-cache window */
412 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
413 		lower_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
414 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
415 		upper_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
416 	WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_OFFSET0, 0);
417 	WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_SIZE0,
418 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
419 }
420 
421 /**
422  * vcn_v4_0_mc_resume_dpg_mode - memory controller programming for dpg mode
423  *
424  * @adev: amdgpu_device pointer
425  * @inst_idx: instance number index
426  * @indirect: indirectly write sram
427  *
428  * Let the VCN memory controller know it's offsets with dpg mode
429  */
430 static void vcn_v4_0_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
431 {
432 	uint32_t offset, size;
433 	const struct common_firmware_header *hdr;
434 	hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
435 	size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
436 
437 	/* cache window 0: fw */
438 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
439 		if (!indirect) {
440 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
441 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
442 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
443 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
444 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
445 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
446 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
447 				VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
448 		} else {
449 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
450 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
451 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
452 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
453 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
454 				VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
455 		}
456 		offset = 0;
457 	} else {
458 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
459 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
460 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
461 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
462 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
463 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
464 		offset = size;
465 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
466 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0),
467 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
468 	}
469 
470 	if (!indirect)
471 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
472 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
473 	else
474 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
475 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
476 
477 	/* cache window 1: stack */
478 	if (!indirect) {
479 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
480 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
481 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
482 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
483 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
484 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
485 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
486 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
487 	} else {
488 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
489 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
490 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
491 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
492 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
493 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
494 	}
495 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
496 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
497 
498 	/* cache window 2: context */
499 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
500 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
501 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
502 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
503 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
504 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
505 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
506 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
507 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
508 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
509 
510 	/* non-cache window */
511 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
512 			VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
513 			lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
514 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
515 			VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
516 			upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
517 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
518 			VCN, inst_idx, regUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
519 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
520 			VCN, inst_idx, regUVD_VCPU_NONCACHE_SIZE0),
521 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)), 0, indirect);
522 
523 	/* VCN global tiling registers */
524 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
525 		VCN, 0, regUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
526 }
527 
528 /**
529  * vcn_v4_0_disable_static_power_gating - disable VCN static power gating
530  *
531  * @adev: amdgpu_device pointer
532  * @inst: instance number
533  *
534  * Disable static power gating for VCN block
535  */
536 static void vcn_v4_0_disable_static_power_gating(struct amdgpu_device *adev, int inst)
537 {
538 	uint32_t data = 0;
539 
540 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
541 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
542 			| 1 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
543 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
544 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
545 			| 2 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
546 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
547 			| 2 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
548 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
549 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
550 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
551 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
552 			| 2 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
553 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
554 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
555 
556 		WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
557 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS,
558 			UVD_PGFSM_STATUS__UVDM_UVDU_UVDLM_PWR_ON_3_0, 0x3F3FFFFF);
559 	} else {
560 		uint32_t value;
561 
562 		value = (inst) ? 0x2200800 : 0;
563 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
564 			| 1 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
565 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
566 			| 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
567 			| 1 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
568 			| 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
569 			| 1 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
570 			| 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
571 			| 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
572 			| 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
573 			| 1 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
574 			| 1 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
575 			| 1 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
576 			| 1 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
577 
578                 WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
579                 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS, value,  0x3F3FFFFF);
580         }
581 
582         data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
583         data &= ~0x103;
584         if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
585                 data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
586                         UVD_POWER_STATUS__UVD_PG_EN_MASK;
587 
588         WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
589 
590         return;
591 }
592 
593 /**
594  * vcn_v4_0_enable_static_power_gating - enable VCN static power gating
595  *
596  * @adev: amdgpu_device pointer
597  * @inst: instance number
598  *
599  * Enable static power gating for VCN block
600  */
601 static void vcn_v4_0_enable_static_power_gating(struct amdgpu_device *adev, int inst)
602 {
603 	uint32_t data;
604 
605 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
606 		/* Before power off, this indicator has to be turned on */
607 		data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
608 		data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
609 		data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
610 		WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
611 
612 		data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
613 			| 2 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
614 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
615 			| 2 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
616 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
617 			| 2 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
618 			| 2 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
619 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
620 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
621 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
622 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
623 			| 2 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
624 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
625 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
626 		WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
627 
628 		data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
629 			| 2 << UVD_PGFSM_STATUS__UVDS_PWR_STATUS__SHIFT
630 			| 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
631 			| 2 << UVD_PGFSM_STATUS__UVDTC_PWR_STATUS__SHIFT
632 			| 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
633 			| 2 << UVD_PGFSM_STATUS__UVDTA_PWR_STATUS__SHIFT
634 			| 2 << UVD_PGFSM_STATUS__UVDLM_PWR_STATUS__SHIFT
635 			| 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
636 			| 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
637 			| 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT
638 			| 2 << UVD_PGFSM_STATUS__UVDAB_PWR_STATUS__SHIFT
639 			| 2 << UVD_PGFSM_STATUS__UVDTB_PWR_STATUS__SHIFT
640 			| 2 << UVD_PGFSM_STATUS__UVDNA_PWR_STATUS__SHIFT
641 			| 2 << UVD_PGFSM_STATUS__UVDNB_PWR_STATUS__SHIFT);
642 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS, data, 0x3F3FFFFF);
643 	}
644 
645         return;
646 }
647 
648 /**
649  * vcn_v4_0_disable_clock_gating - disable VCN clock gating
650  *
651  * @adev: amdgpu_device pointer
652  * @inst: instance number
653  *
654  * Disable clock gating for VCN block
655  */
656 static void vcn_v4_0_disable_clock_gating(struct amdgpu_device *adev, int inst)
657 {
658 	uint32_t data;
659 
660 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
661 		return;
662 
663 	/* VCN disable CGC */
664 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
665 	data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
666 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
667 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
668 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
669 
670 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_GATE);
671 	data &= ~(UVD_CGC_GATE__SYS_MASK
672 		| UVD_CGC_GATE__UDEC_MASK
673 		| UVD_CGC_GATE__MPEG2_MASK
674 		| UVD_CGC_GATE__REGS_MASK
675 		| UVD_CGC_GATE__RBC_MASK
676 		| UVD_CGC_GATE__LMI_MC_MASK
677 		| UVD_CGC_GATE__LMI_UMC_MASK
678 		| UVD_CGC_GATE__IDCT_MASK
679 		| UVD_CGC_GATE__MPRD_MASK
680 		| UVD_CGC_GATE__MPC_MASK
681 		| UVD_CGC_GATE__LBSI_MASK
682 		| UVD_CGC_GATE__LRBBM_MASK
683 		| UVD_CGC_GATE__UDEC_RE_MASK
684 		| UVD_CGC_GATE__UDEC_CM_MASK
685 		| UVD_CGC_GATE__UDEC_IT_MASK
686 		| UVD_CGC_GATE__UDEC_DB_MASK
687 		| UVD_CGC_GATE__UDEC_MP_MASK
688 		| UVD_CGC_GATE__WCB_MASK
689 		| UVD_CGC_GATE__VCPU_MASK
690 		| UVD_CGC_GATE__MMSCH_MASK);
691 
692 	WREG32_SOC15(VCN, inst, regUVD_CGC_GATE, data);
693 	SOC15_WAIT_ON_RREG(VCN, inst, regUVD_CGC_GATE, 0,  0xFFFFFFFF);
694 
695 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
696 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
697 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
698 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
699 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
700 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
701 		| UVD_CGC_CTRL__SYS_MODE_MASK
702 		| UVD_CGC_CTRL__UDEC_MODE_MASK
703 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
704 		| UVD_CGC_CTRL__REGS_MODE_MASK
705 		| UVD_CGC_CTRL__RBC_MODE_MASK
706 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
707 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
708 		| UVD_CGC_CTRL__IDCT_MODE_MASK
709 		| UVD_CGC_CTRL__MPRD_MODE_MASK
710 		| UVD_CGC_CTRL__MPC_MODE_MASK
711 		| UVD_CGC_CTRL__LBSI_MODE_MASK
712 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
713 		| UVD_CGC_CTRL__WCB_MODE_MASK
714 		| UVD_CGC_CTRL__VCPU_MODE_MASK
715 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
716 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
717 
718 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE);
719 	data |= (UVD_SUVD_CGC_GATE__SRE_MASK
720 		| UVD_SUVD_CGC_GATE__SIT_MASK
721 		| UVD_SUVD_CGC_GATE__SMP_MASK
722 		| UVD_SUVD_CGC_GATE__SCM_MASK
723 		| UVD_SUVD_CGC_GATE__SDB_MASK
724 		| UVD_SUVD_CGC_GATE__SRE_H264_MASK
725 		| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
726 		| UVD_SUVD_CGC_GATE__SIT_H264_MASK
727 		| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
728 		| UVD_SUVD_CGC_GATE__SCM_H264_MASK
729 		| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
730 		| UVD_SUVD_CGC_GATE__SDB_H264_MASK
731 		| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
732 		| UVD_SUVD_CGC_GATE__SCLR_MASK
733 		| UVD_SUVD_CGC_GATE__UVD_SC_MASK
734 		| UVD_SUVD_CGC_GATE__ENT_MASK
735 		| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
736 		| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
737 		| UVD_SUVD_CGC_GATE__SITE_MASK
738 		| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
739 		| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
740 		| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
741 		| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
742 		| UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
743 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE, data);
744 
745 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
746 	data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
747 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
748 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
749 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
750 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
751 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
752 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
753 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
754 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
755 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
756 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
757 }
758 
759 /**
760  * vcn_v4_0_disable_clock_gating_dpg_mode - disable VCN clock gating dpg mode
761  *
762  * @adev: amdgpu_device pointer
763  * @sram_sel: sram select
764  * @inst_idx: instance number index
765  * @indirect: indirectly write sram
766  *
767  * Disable clock gating for VCN block with dpg mode
768  */
769 static void vcn_v4_0_disable_clock_gating_dpg_mode(struct amdgpu_device *adev, uint8_t sram_sel,
770       int inst_idx, uint8_t indirect)
771 {
772 	uint32_t reg_data = 0;
773 
774 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
775 		return;
776 
777 	/* enable sw clock gating control */
778 	reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
779 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
780 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
781 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
782 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
783 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
784 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
785 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
786 		 UVD_CGC_CTRL__SYS_MODE_MASK |
787 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
788 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
789 		 UVD_CGC_CTRL__REGS_MODE_MASK |
790 		 UVD_CGC_CTRL__RBC_MODE_MASK |
791 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
792 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
793 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
794 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
795 		 UVD_CGC_CTRL__MPC_MODE_MASK |
796 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
797 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
798 		 UVD_CGC_CTRL__WCB_MODE_MASK |
799 		 UVD_CGC_CTRL__VCPU_MODE_MASK);
800 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
801 		VCN, inst_idx, regUVD_CGC_CTRL), reg_data, sram_sel, indirect);
802 
803 	/* turn off clock gating */
804 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
805 		VCN, inst_idx, regUVD_CGC_GATE), 0, sram_sel, indirect);
806 
807 	/* turn on SUVD clock gating */
808 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
809 		VCN, inst_idx, regUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
810 
811 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
812 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
813 		VCN, inst_idx, regUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
814 }
815 
816 /**
817  * vcn_v4_0_enable_clock_gating - enable VCN clock gating
818  *
819  * @adev: amdgpu_device pointer
820  * @inst: instance number
821  *
822  * Enable clock gating for VCN block
823  */
824 static void vcn_v4_0_enable_clock_gating(struct amdgpu_device *adev, int inst)
825 {
826 	uint32_t data;
827 
828 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
829 		return;
830 
831 	/* enable VCN CGC */
832 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
833 	data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
834 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
835 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
836 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
837 
838 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
839 	data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
840 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
841 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
842 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
843 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
844 		| UVD_CGC_CTRL__SYS_MODE_MASK
845 		| UVD_CGC_CTRL__UDEC_MODE_MASK
846 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
847 		| UVD_CGC_CTRL__REGS_MODE_MASK
848 		| UVD_CGC_CTRL__RBC_MODE_MASK
849 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
850 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
851 		| UVD_CGC_CTRL__IDCT_MODE_MASK
852 		| UVD_CGC_CTRL__MPRD_MODE_MASK
853 		| UVD_CGC_CTRL__MPC_MODE_MASK
854 		| UVD_CGC_CTRL__LBSI_MODE_MASK
855 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
856 		| UVD_CGC_CTRL__WCB_MODE_MASK
857 		| UVD_CGC_CTRL__VCPU_MODE_MASK
858 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
859 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
860 
861 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
862 	data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
863 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
864 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
865 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
866 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
867 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
868 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
869 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
870 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
871 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
872 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
873 
874 	return;
875 }
876 
877 static void vcn_v4_0_enable_ras(struct amdgpu_device *adev, int inst_idx,
878 				bool indirect)
879 {
880 	uint32_t tmp;
881 
882 	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN))
883 		return;
884 
885 	tmp = VCN_RAS_CNTL__VCPU_VCODEC_REARM_MASK |
886 	      VCN_RAS_CNTL__VCPU_VCODEC_IH_EN_MASK |
887 	      VCN_RAS_CNTL__VCPU_VCODEC_PMI_EN_MASK |
888 	      VCN_RAS_CNTL__VCPU_VCODEC_STALL_EN_MASK;
889 	WREG32_SOC15_DPG_MODE(inst_idx,
890 			      SOC15_DPG_MODE_OFFSET(VCN, 0, regVCN_RAS_CNTL),
891 			      tmp, 0, indirect);
892 
893 	tmp = UVD_SYS_INT_EN__RASCNTL_VCPU_VCODEC_EN_MASK;
894 	WREG32_SOC15_DPG_MODE(inst_idx,
895 			      SOC15_DPG_MODE_OFFSET(VCN, 0, regUVD_SYS_INT_EN),
896 			      tmp, 0, indirect);
897 }
898 
899 /**
900  * vcn_v4_0_start_dpg_mode - VCN start with dpg mode
901  *
902  * @adev: amdgpu_device pointer
903  * @inst_idx: instance number index
904  * @indirect: indirectly write sram
905  *
906  * Start VCN block with dpg mode
907  */
908 static int vcn_v4_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
909 {
910 	volatile struct amdgpu_vcn4_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
911 	struct amdgpu_ring *ring;
912 	uint32_t tmp;
913 
914 	/* disable register anti-hang mechanism */
915 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 1,
916 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
917 	/* enable dynamic power gating mode */
918 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS);
919 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
920 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
921 	WREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS, tmp);
922 
923 	if (indirect)
924 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
925 
926 	/* enable clock gating */
927 	vcn_v4_0_disable_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
928 
929 	/* enable VCPU clock */
930 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
931 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK | UVD_VCPU_CNTL__BLK_RST_MASK;
932 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
933 		VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
934 
935 	/* disable master interupt */
936 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
937 		VCN, inst_idx, regUVD_MASTINT_EN), 0, 0, indirect);
938 
939 	/* setup regUVD_LMI_CTRL */
940 	tmp = (UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
941 		UVD_LMI_CTRL__REQ_MODE_MASK |
942 		UVD_LMI_CTRL__CRC_RESET_MASK |
943 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
944 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
945 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
946 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
947 		0x00100000L);
948 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
949 		VCN, inst_idx, regUVD_LMI_CTRL), tmp, 0, indirect);
950 
951 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
952 		VCN, inst_idx, regUVD_MPC_CNTL),
953 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
954 
955 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
956 		VCN, inst_idx, regUVD_MPC_SET_MUXA0),
957 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
958 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
959 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
960 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
961 
962 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
963 		VCN, inst_idx, regUVD_MPC_SET_MUXB0),
964 		 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
965 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
966 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
967 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
968 
969 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
970 		VCN, inst_idx, regUVD_MPC_SET_MUX),
971 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
972 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
973 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
974 
975 	vcn_v4_0_mc_resume_dpg_mode(adev, inst_idx, indirect);
976 
977 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
978 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
979 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
980 		VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
981 
982 	/* enable LMI MC and UMC channels */
983 	tmp = 0x1f << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT;
984 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
985 		VCN, inst_idx, regUVD_LMI_CTRL2), tmp, 0, indirect);
986 
987 	vcn_v4_0_enable_ras(adev, inst_idx, indirect);
988 
989 	/* enable master interrupt */
990 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
991 		VCN, inst_idx, regUVD_MASTINT_EN),
992 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
993 
994 
995 	if (indirect)
996 		psp_update_vcn_sram(adev, inst_idx, adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
997 			(uint32_t)((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
998 				(uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr));
999 
1000 	ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1001 
1002 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_LO, ring->gpu_addr);
1003 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1004 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_SIZE, ring->ring_size / 4);
1005 
1006 	tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
1007 	tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
1008 	WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
1009 	fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
1010 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR, 0);
1011 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, 0);
1012 
1013 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR);
1014 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, tmp);
1015 	ring->wptr = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
1016 
1017 	tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
1018 	tmp |= VCN_RB_ENABLE__RB1_EN_MASK;
1019 	WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
1020 	fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
1021 
1022 	WREG32_SOC15(VCN, inst_idx, regVCN_RB1_DB_CTRL,
1023 			ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
1024 			VCN_RB1_DB_CTRL__EN_MASK);
1025 
1026 	return 0;
1027 }
1028 
1029 
1030 /**
1031  * vcn_v4_0_start - VCN start
1032  *
1033  * @adev: amdgpu_device pointer
1034  *
1035  * Start VCN block
1036  */
1037 static int vcn_v4_0_start(struct amdgpu_device *adev)
1038 {
1039 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1040 	struct amdgpu_ring *ring;
1041 	uint32_t tmp;
1042 	int i, j, k, r;
1043 
1044 	if (adev->pm.dpm_enabled)
1045 		amdgpu_dpm_enable_uvd(adev, true);
1046 
1047 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1048 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1049 
1050 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1051 			r = vcn_v4_0_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
1052 			continue;
1053 		}
1054 
1055 		/* disable VCN power gating */
1056 		vcn_v4_0_disable_static_power_gating(adev, i);
1057 
1058 		/* set VCN status busy */
1059 		tmp = RREG32_SOC15(VCN, i, regUVD_STATUS) | UVD_STATUS__UVD_BUSY;
1060 		WREG32_SOC15(VCN, i, regUVD_STATUS, tmp);
1061 
1062 		/*SW clock gating */
1063 		vcn_v4_0_disable_clock_gating(adev, i);
1064 
1065 		/* enable VCPU clock */
1066 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1067 				UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1068 
1069 		/* disable master interrupt */
1070 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN), 0,
1071 				~UVD_MASTINT_EN__VCPU_EN_MASK);
1072 
1073 		/* enable LMI MC and UMC channels */
1074 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_LMI_CTRL2), 0,
1075 				~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1076 
1077 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1078 		tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1079 		tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1080 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1081 
1082 		/* setup regUVD_LMI_CTRL */
1083 		tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL);
1084 		WREG32_SOC15(VCN, i, regUVD_LMI_CTRL, tmp |
1085 				UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
1086 				UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1087 				UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1088 				UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1089 
1090 		/* setup regUVD_MPC_CNTL */
1091 		tmp = RREG32_SOC15(VCN, i, regUVD_MPC_CNTL);
1092 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1093 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1094 		WREG32_SOC15(VCN, i, regUVD_MPC_CNTL, tmp);
1095 
1096 		/* setup UVD_MPC_SET_MUXA0 */
1097 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXA0,
1098 				((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1099 				 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1100 				 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1101 				 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1102 
1103 		/* setup UVD_MPC_SET_MUXB0 */
1104 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXB0,
1105 				((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1106 				 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1107 				 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1108 				 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1109 
1110 		/* setup UVD_MPC_SET_MUX */
1111 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUX,
1112 				((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1113 				 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1114 				 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1115 
1116 		vcn_v4_0_mc_resume(adev, i);
1117 
1118 		/* VCN global tiling registers */
1119 		WREG32_SOC15(VCN, i, regUVD_GFX10_ADDR_CONFIG,
1120 				adev->gfx.config.gb_addr_config);
1121 
1122 		/* unblock VCPU register access */
1123 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL), 0,
1124 				~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1125 
1126 		/* release VCPU reset to boot */
1127 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1128 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1129 
1130 		for (j = 0; j < 10; ++j) {
1131 			uint32_t status;
1132 
1133 			for (k = 0; k < 100; ++k) {
1134 				status = RREG32_SOC15(VCN, i, regUVD_STATUS);
1135 				if (status & 2)
1136 					break;
1137 				mdelay(10);
1138 				if (amdgpu_emu_mode==1)
1139 					msleep(1);
1140 			}
1141 
1142 			if (amdgpu_emu_mode==1) {
1143 				r = -1;
1144 				if (status & 2) {
1145 					r = 0;
1146 					break;
1147 				}
1148 			} else {
1149 				r = 0;
1150 				if (status & 2)
1151 					break;
1152 
1153 				dev_err(adev->dev, "VCN[%d] is not responding, trying to reset the VCPU!!!\n", i);
1154 				WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1155 							UVD_VCPU_CNTL__BLK_RST_MASK,
1156 							~UVD_VCPU_CNTL__BLK_RST_MASK);
1157 				mdelay(10);
1158 				WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1159 						~UVD_VCPU_CNTL__BLK_RST_MASK);
1160 
1161 				mdelay(10);
1162 				r = -1;
1163 			}
1164 		}
1165 
1166 		if (r) {
1167 			dev_err(adev->dev, "VCN[%d] is not responding, giving up!!!\n", i);
1168 			return r;
1169 		}
1170 
1171 		/* enable master interrupt */
1172 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN),
1173 				UVD_MASTINT_EN__VCPU_EN_MASK,
1174 				~UVD_MASTINT_EN__VCPU_EN_MASK);
1175 
1176 		/* clear the busy bit of VCN_STATUS */
1177 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_STATUS), 0,
1178 				~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1179 
1180 		ring = &adev->vcn.inst[i].ring_enc[0];
1181 		WREG32_SOC15(VCN, i, regVCN_RB1_DB_CTRL,
1182 				ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
1183 				VCN_RB1_DB_CTRL__EN_MASK);
1184 
1185 		WREG32_SOC15(VCN, i, regUVD_RB_BASE_LO, ring->gpu_addr);
1186 		WREG32_SOC15(VCN, i, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1187 		WREG32_SOC15(VCN, i, regUVD_RB_SIZE, ring->ring_size / 4);
1188 
1189 		tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1190 		tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
1191 		WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1192 		fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
1193 		WREG32_SOC15(VCN, i, regUVD_RB_RPTR, 0);
1194 		WREG32_SOC15(VCN, i, regUVD_RB_WPTR, 0);
1195 
1196 		tmp = RREG32_SOC15(VCN, i, regUVD_RB_RPTR);
1197 		WREG32_SOC15(VCN, i, regUVD_RB_WPTR, tmp);
1198 		ring->wptr = RREG32_SOC15(VCN, i, regUVD_RB_WPTR);
1199 
1200 		tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1201 		tmp |= VCN_RB_ENABLE__RB1_EN_MASK;
1202 		WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1203 		fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
1204 	}
1205 
1206 	return 0;
1207 }
1208 
1209 static int vcn_v4_0_start_sriov(struct amdgpu_device *adev)
1210 {
1211 	int i;
1212 	struct amdgpu_ring *ring_enc;
1213 	uint64_t cache_addr;
1214 	uint64_t rb_enc_addr;
1215 	uint64_t ctx_addr;
1216 	uint32_t param, resp, expected;
1217 	uint32_t offset, cache_size;
1218 	uint32_t tmp, timeout;
1219 
1220 	struct amdgpu_mm_table *table = &adev->virt.mm_table;
1221 	uint32_t *table_loc;
1222 	uint32_t table_size;
1223 	uint32_t size, size_dw;
1224 	uint32_t init_status;
1225 	uint32_t enabled_vcn;
1226 
1227 	struct mmsch_v4_0_cmd_direct_write
1228 		direct_wt = { {0} };
1229 	struct mmsch_v4_0_cmd_direct_read_modify_write
1230 		direct_rd_mod_wt = { {0} };
1231 	struct mmsch_v4_0_cmd_end end = { {0} };
1232 	struct mmsch_v4_0_init_header header;
1233 
1234 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1235 	volatile struct amdgpu_fw_shared_rb_setup *rb_setup;
1236 
1237 	direct_wt.cmd_header.command_type =
1238 		MMSCH_COMMAND__DIRECT_REG_WRITE;
1239 	direct_rd_mod_wt.cmd_header.command_type =
1240 		MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1241 	end.cmd_header.command_type =
1242 		MMSCH_COMMAND__END;
1243 
1244 	header.version = MMSCH_VERSION;
1245 	header.total_size = sizeof(struct mmsch_v4_0_init_header) >> 2;
1246 	for (i = 0; i < MMSCH_V4_0_VCN_INSTANCES; i++) {
1247 		header.inst[i].init_status = 0;
1248 		header.inst[i].table_offset = 0;
1249 		header.inst[i].table_size = 0;
1250 	}
1251 
1252 	table_loc = (uint32_t *)table->cpu_addr;
1253 	table_loc += header.total_size;
1254 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1255 		if (adev->vcn.harvest_config & (1 << i))
1256 			continue;
1257 
1258 		table_size = 0;
1259 
1260 		MMSCH_V4_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(VCN, i,
1261 			regUVD_STATUS),
1262 			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1263 
1264 		cache_size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1265 
1266 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1267 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1268 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1269 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1270 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1271 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1272 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1273 			offset = 0;
1274 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1275 				regUVD_VCPU_CACHE_OFFSET0),
1276 				0);
1277 		} else {
1278 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1279 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1280 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1281 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1282 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1283 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1284 			offset = cache_size;
1285 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1286 				regUVD_VCPU_CACHE_OFFSET0),
1287 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1288 		}
1289 
1290 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1291 			regUVD_VCPU_CACHE_SIZE0),
1292 			cache_size);
1293 
1294 		cache_addr = adev->vcn.inst[i].gpu_addr + offset;
1295 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1296 			regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1297 			lower_32_bits(cache_addr));
1298 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1299 			regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1300 			upper_32_bits(cache_addr));
1301 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1302 			regUVD_VCPU_CACHE_OFFSET1),
1303 			0);
1304 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1305 			regUVD_VCPU_CACHE_SIZE1),
1306 			AMDGPU_VCN_STACK_SIZE);
1307 
1308 		cache_addr = adev->vcn.inst[i].gpu_addr + offset +
1309 			AMDGPU_VCN_STACK_SIZE;
1310 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1311 			regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1312 			lower_32_bits(cache_addr));
1313 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1314 			regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1315 			upper_32_bits(cache_addr));
1316 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1317 			regUVD_VCPU_CACHE_OFFSET2),
1318 			0);
1319 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1320 			regUVD_VCPU_CACHE_SIZE2),
1321 			AMDGPU_VCN_CONTEXT_SIZE);
1322 
1323 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1324 		rb_setup = &fw_shared->rb_setup;
1325 
1326 		ring_enc = &adev->vcn.inst[i].ring_enc[0];
1327 		ring_enc->wptr = 0;
1328 		rb_enc_addr = ring_enc->gpu_addr;
1329 
1330 		rb_setup->is_rb_enabled_flags |= RB_ENABLED;
1331 		rb_setup->rb_addr_lo = lower_32_bits(rb_enc_addr);
1332 		rb_setup->rb_addr_hi = upper_32_bits(rb_enc_addr);
1333 		rb_setup->rb_size = ring_enc->ring_size / 4;
1334 		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
1335 
1336 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1337 			regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
1338 			lower_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
1339 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1340 			regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
1341 			upper_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
1342 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1343 			regUVD_VCPU_NONCACHE_SIZE0),
1344 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
1345 
1346 		/* add end packet */
1347 		MMSCH_V4_0_INSERT_END();
1348 
1349 		/* refine header */
1350 		header.inst[i].init_status = 0;
1351 		header.inst[i].table_offset = header.total_size;
1352 		header.inst[i].table_size = table_size;
1353 		header.total_size += table_size;
1354 	}
1355 
1356 	/* Update init table header in memory */
1357 	size = sizeof(struct mmsch_v4_0_init_header);
1358 	table_loc = (uint32_t *)table->cpu_addr;
1359 	memcpy((void *)table_loc, &header, size);
1360 
1361 	/* message MMSCH (in VCN[0]) to initialize this client
1362 	 * 1, write to mmsch_vf_ctx_addr_lo/hi register with GPU mc addr
1363 	 * of memory descriptor location
1364 	 */
1365 	ctx_addr = table->gpu_addr;
1366 	WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_ADDR_LO, lower_32_bits(ctx_addr));
1367 	WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_ADDR_HI, upper_32_bits(ctx_addr));
1368 
1369 	/* 2, update vmid of descriptor */
1370 	tmp = RREG32_SOC15(VCN, 0, regMMSCH_VF_VMID);
1371 	tmp &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1372 	/* use domain0 for MM scheduler */
1373 	tmp |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1374 	WREG32_SOC15(VCN, 0, regMMSCH_VF_VMID, tmp);
1375 
1376 	/* 3, notify mmsch about the size of this descriptor */
1377 	size = header.total_size;
1378 	WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_SIZE, size);
1379 
1380 	/* 4, set resp to zero */
1381 	WREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_RESP, 0);
1382 
1383 	/* 5, kick off the initialization and wait until
1384 	 * MMSCH_VF_MAILBOX_RESP becomes non-zero
1385 	 */
1386 	param = 0x00000001;
1387 	WREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_HOST, param);
1388 	tmp = 0;
1389 	timeout = 1000;
1390 	resp = 0;
1391 	expected = MMSCH_VF_MAILBOX_RESP__OK;
1392 	while (resp != expected) {
1393 		resp = RREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_RESP);
1394 		if (resp != 0)
1395 			break;
1396 
1397 		udelay(10);
1398 		tmp = tmp + 10;
1399 		if (tmp >= timeout) {
1400 			DRM_ERROR("failed to init MMSCH. TIME-OUT after %d usec"\
1401 				" waiting for regMMSCH_VF_MAILBOX_RESP "\
1402 				"(expected=0x%08x, readback=0x%08x)\n",
1403 				tmp, expected, resp);
1404 			return -EBUSY;
1405 		}
1406 	}
1407 	enabled_vcn = amdgpu_vcn_is_disabled_vcn(adev, VCN_DECODE_RING, 0) ? 1 : 0;
1408 	init_status = ((struct mmsch_v4_0_init_header *)(table_loc))->inst[enabled_vcn].init_status;
1409 	if (resp != expected && resp != MMSCH_VF_MAILBOX_RESP__INCOMPLETE
1410 	&& init_status != MMSCH_VF_ENGINE_STATUS__PASS)
1411 		DRM_ERROR("MMSCH init status is incorrect! readback=0x%08x, header init "\
1412 			"status for VCN%x: 0x%x\n", resp, enabled_vcn, init_status);
1413 
1414 	return 0;
1415 }
1416 
1417 /**
1418  * vcn_v4_0_stop_dpg_mode - VCN stop with dpg mode
1419  *
1420  * @adev: amdgpu_device pointer
1421  * @inst_idx: instance number index
1422  *
1423  * Stop VCN block with dpg mode
1424  */
1425 static void vcn_v4_0_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1426 {
1427 	struct dpg_pause_state state = {.fw_based = VCN_DPG_STATE__UNPAUSE};
1428 	uint32_t tmp;
1429 
1430 	vcn_v4_0_pause_dpg_mode(adev, inst_idx, &state);
1431 	/* Wait for power status to be 1 */
1432 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1433 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1434 
1435 	/* wait for read ptr to be equal to write ptr */
1436 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
1437 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1438 
1439 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1440 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1441 
1442 	/* disable dynamic power gating mode */
1443 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 0,
1444 		~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1445 }
1446 
1447 /**
1448  * vcn_v4_0_stop - VCN stop
1449  *
1450  * @adev: amdgpu_device pointer
1451  *
1452  * Stop VCN block
1453  */
1454 static int vcn_v4_0_stop(struct amdgpu_device *adev)
1455 {
1456 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1457 	uint32_t tmp;
1458 	int i, r = 0;
1459 
1460 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1461 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1462 		fw_shared->sq.queue_mode |= FW_QUEUE_DPG_HOLD_OFF;
1463 
1464 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1465 			vcn_v4_0_stop_dpg_mode(adev, i);
1466 			continue;
1467 		}
1468 
1469 		/* wait for vcn idle */
1470 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1471 		if (r)
1472 			return r;
1473 
1474 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1475 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1476 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1477 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1478 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1479 		if (r)
1480 			return r;
1481 
1482 		/* disable LMI UMC channel */
1483 		tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL2);
1484 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1485 		WREG32_SOC15(VCN, i, regUVD_LMI_CTRL2, tmp);
1486 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK |
1487 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1488 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1489 		if (r)
1490 			return r;
1491 
1492 		/* block VCPU register access */
1493 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL),
1494 				UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1495 				~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1496 
1497 		/* reset VCPU */
1498 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1499 				UVD_VCPU_CNTL__BLK_RST_MASK,
1500 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1501 
1502 		/* disable VCPU clock */
1503 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1504 				~(UVD_VCPU_CNTL__CLK_EN_MASK));
1505 
1506 		/* apply soft reset */
1507 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1508 		tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1509 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1510 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1511 		tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1512 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1513 
1514 		/* clear status */
1515 		WREG32_SOC15(VCN, i, regUVD_STATUS, 0);
1516 
1517 		/* apply HW clock gating */
1518 		vcn_v4_0_enable_clock_gating(adev, i);
1519 
1520 		/* enable VCN power gating */
1521 		vcn_v4_0_enable_static_power_gating(adev, i);
1522 	}
1523 
1524 	if (adev->pm.dpm_enabled)
1525 		amdgpu_dpm_enable_uvd(adev, false);
1526 
1527 	return 0;
1528 }
1529 
1530 /**
1531  * vcn_v4_0_pause_dpg_mode - VCN pause with dpg mode
1532  *
1533  * @adev: amdgpu_device pointer
1534  * @inst_idx: instance number index
1535  * @new_state: pause state
1536  *
1537  * Pause dpg mode for VCN block
1538  */
1539 static int vcn_v4_0_pause_dpg_mode(struct amdgpu_device *adev, int inst_idx,
1540       struct dpg_pause_state *new_state)
1541 {
1542 	uint32_t reg_data = 0;
1543 	int ret_code;
1544 
1545 	/* pause/unpause if state is changed */
1546 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1547 		DRM_DEV_DEBUG(adev->dev, "dpg pause state changed %d -> %d",
1548 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1549 		reg_data = RREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE) &
1550 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1551 
1552 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1553 			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 0x1,
1554 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1555 
1556 			if (!ret_code) {
1557 				/* pause DPG */
1558 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1559 				WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1560 
1561 				/* wait for ACK */
1562 				SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_DPG_PAUSE,
1563 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1564 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1565 
1566 				SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS,
1567 					UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1568 			}
1569 		} else {
1570 			/* unpause dpg, no need to wait */
1571 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1572 			WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1573 		}
1574 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1575 	}
1576 
1577 	return 0;
1578 }
1579 
1580 /**
1581  * vcn_v4_0_unified_ring_get_rptr - get unified read pointer
1582  *
1583  * @ring: amdgpu_ring pointer
1584  *
1585  * Returns the current hardware unified read pointer
1586  */
1587 static uint64_t vcn_v4_0_unified_ring_get_rptr(struct amdgpu_ring *ring)
1588 {
1589 	struct amdgpu_device *adev = ring->adev;
1590 
1591 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1592 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1593 
1594 	return RREG32_SOC15(VCN, ring->me, regUVD_RB_RPTR);
1595 }
1596 
1597 /**
1598  * vcn_v4_0_unified_ring_get_wptr - get unified write pointer
1599  *
1600  * @ring: amdgpu_ring pointer
1601  *
1602  * Returns the current hardware unified write pointer
1603  */
1604 static uint64_t vcn_v4_0_unified_ring_get_wptr(struct amdgpu_ring *ring)
1605 {
1606 	struct amdgpu_device *adev = ring->adev;
1607 
1608 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1609 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1610 
1611 	if (ring->use_doorbell)
1612 		return *ring->wptr_cpu_addr;
1613 	else
1614 		return RREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR);
1615 }
1616 
1617 /**
1618  * vcn_v4_0_unified_ring_set_wptr - set enc write pointer
1619  *
1620  * @ring: amdgpu_ring pointer
1621  *
1622  * Commits the enc write pointer to the hardware
1623  */
1624 static void vcn_v4_0_unified_ring_set_wptr(struct amdgpu_ring *ring)
1625 {
1626 	struct amdgpu_device *adev = ring->adev;
1627 
1628 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1629 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1630 
1631 	if (ring->use_doorbell) {
1632 		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1633 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1634 	} else {
1635 		WREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR, lower_32_bits(ring->wptr));
1636 	}
1637 }
1638 
1639 static int vcn_v4_0_limit_sched(struct amdgpu_cs_parser *p,
1640 				struct amdgpu_job *job)
1641 {
1642 	struct drm_gpu_scheduler **scheds;
1643 
1644 	/* The create msg must be in the first IB submitted */
1645 	if (atomic_read(&job->base.entity->fence_seq))
1646 		return -EINVAL;
1647 
1648 	/* if VCN0 is harvested, we can't support AV1 */
1649 	if (p->adev->vcn.harvest_config & AMDGPU_VCN_HARVEST_VCN0)
1650 		return -EINVAL;
1651 
1652 	scheds = p->adev->gpu_sched[AMDGPU_HW_IP_VCN_ENC]
1653 		[AMDGPU_RING_PRIO_0].sched;
1654 	drm_sched_entity_modify_sched(job->base.entity, scheds, 1);
1655 	return 0;
1656 }
1657 
1658 static int vcn_v4_0_dec_msg(struct amdgpu_cs_parser *p, struct amdgpu_job *job,
1659 			    uint64_t addr)
1660 {
1661 	struct ttm_operation_ctx ctx = { false, false };
1662 	struct amdgpu_bo_va_mapping *map;
1663 	uint32_t *msg, num_buffers;
1664 	struct amdgpu_bo *bo;
1665 	uint64_t start, end;
1666 	unsigned int i;
1667 	void *ptr;
1668 	int r;
1669 
1670 	addr &= AMDGPU_GMC_HOLE_MASK;
1671 	r = amdgpu_cs_find_mapping(p, addr, &bo, &map);
1672 	if (r) {
1673 		DRM_ERROR("Can't find BO for addr 0x%08llx\n", addr);
1674 		return r;
1675 	}
1676 
1677 	start = map->start * AMDGPU_GPU_PAGE_SIZE;
1678 	end = (map->last + 1) * AMDGPU_GPU_PAGE_SIZE;
1679 	if (addr & 0x7) {
1680 		DRM_ERROR("VCN messages must be 8 byte aligned!\n");
1681 		return -EINVAL;
1682 	}
1683 
1684 	bo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
1685 	amdgpu_bo_placement_from_domain(bo, bo->allowed_domains);
1686 	r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
1687 	if (r) {
1688 		DRM_ERROR("Failed validating the VCN message BO (%d)!\n", r);
1689 		return r;
1690 	}
1691 
1692 	r = amdgpu_bo_kmap(bo, &ptr);
1693 	if (r) {
1694 		DRM_ERROR("Failed mapping the VCN message (%d)!\n", r);
1695 		return r;
1696 	}
1697 
1698 	msg = ptr + addr - start;
1699 
1700 	/* Check length */
1701 	if (msg[1] > end - addr) {
1702 		r = -EINVAL;
1703 		goto out;
1704 	}
1705 
1706 	if (msg[3] != RDECODE_MSG_CREATE)
1707 		goto out;
1708 
1709 	num_buffers = msg[2];
1710 	for (i = 0, msg = &msg[6]; i < num_buffers; ++i, msg += 4) {
1711 		uint32_t offset, size, *create;
1712 
1713 		if (msg[0] != RDECODE_MESSAGE_CREATE)
1714 			continue;
1715 
1716 		offset = msg[1];
1717 		size = msg[2];
1718 
1719 		if (offset + size > end) {
1720 			r = -EINVAL;
1721 			goto out;
1722 		}
1723 
1724 		create = ptr + addr + offset - start;
1725 
1726 		/* H264, HEVC and VP9 can run on any instance */
1727 		if (create[0] == 0x7 || create[0] == 0x10 || create[0] == 0x11)
1728 			continue;
1729 
1730 		r = vcn_v4_0_limit_sched(p, job);
1731 		if (r)
1732 			goto out;
1733 	}
1734 
1735 out:
1736 	amdgpu_bo_kunmap(bo);
1737 	return r;
1738 }
1739 
1740 #define RADEON_VCN_ENGINE_TYPE_ENCODE			(0x00000002)
1741 #define RADEON_VCN_ENGINE_TYPE_DECODE			(0x00000003)
1742 
1743 #define RADEON_VCN_ENGINE_INFO				(0x30000001)
1744 #define RADEON_VCN_ENGINE_INFO_MAX_OFFSET		16
1745 
1746 #define RENCODE_ENCODE_STANDARD_AV1			2
1747 #define RENCODE_IB_PARAM_SESSION_INIT			0x00000003
1748 #define RENCODE_IB_PARAM_SESSION_INIT_MAX_OFFSET	64
1749 
1750 /* return the offset in ib if id is found, -1 otherwise
1751  * to speed up the searching we only search upto max_offset
1752  */
1753 static int vcn_v4_0_enc_find_ib_param(struct amdgpu_ib *ib, uint32_t id, int max_offset)
1754 {
1755 	int i;
1756 
1757 	for (i = 0; i < ib->length_dw && i < max_offset && ib->ptr[i] >= 8; i += ib->ptr[i]/4) {
1758 		if (ib->ptr[i + 1] == id)
1759 			return i;
1760 	}
1761 	return -1;
1762 }
1763 
1764 static int vcn_v4_0_ring_patch_cs_in_place(struct amdgpu_cs_parser *p,
1765 					   struct amdgpu_job *job,
1766 					   struct amdgpu_ib *ib)
1767 {
1768 	struct amdgpu_ring *ring = amdgpu_job_ring(job);
1769 	struct amdgpu_vcn_decode_buffer *decode_buffer;
1770 	uint64_t addr;
1771 	uint32_t val;
1772 	int idx;
1773 
1774 	/* The first instance can decode anything */
1775 	if (!ring->me)
1776 		return 0;
1777 
1778 	/* RADEON_VCN_ENGINE_INFO is at the top of ib block */
1779 	idx = vcn_v4_0_enc_find_ib_param(ib, RADEON_VCN_ENGINE_INFO,
1780 			RADEON_VCN_ENGINE_INFO_MAX_OFFSET);
1781 	if (idx < 0) /* engine info is missing */
1782 		return 0;
1783 
1784 	val = amdgpu_ib_get_value(ib, idx + 2); /* RADEON_VCN_ENGINE_TYPE */
1785 	if (val == RADEON_VCN_ENGINE_TYPE_DECODE) {
1786 		decode_buffer = (struct amdgpu_vcn_decode_buffer *)&ib->ptr[idx + 6];
1787 
1788 		if (!(decode_buffer->valid_buf_flag  & 0x1))
1789 			return 0;
1790 
1791 		addr = ((u64)decode_buffer->msg_buffer_address_hi) << 32 |
1792 			decode_buffer->msg_buffer_address_lo;
1793 		return vcn_v4_0_dec_msg(p, job, addr);
1794 	} else if (val == RADEON_VCN_ENGINE_TYPE_ENCODE) {
1795 		idx = vcn_v4_0_enc_find_ib_param(ib, RENCODE_IB_PARAM_SESSION_INIT,
1796 			RENCODE_IB_PARAM_SESSION_INIT_MAX_OFFSET);
1797 		if (idx >= 0 && ib->ptr[idx + 2] == RENCODE_ENCODE_STANDARD_AV1)
1798 			return vcn_v4_0_limit_sched(p, job);
1799 	}
1800 	return 0;
1801 }
1802 
1803 static const struct amdgpu_ring_funcs vcn_v4_0_unified_ring_vm_funcs = {
1804 	.type = AMDGPU_RING_TYPE_VCN_ENC,
1805 	.align_mask = 0x3f,
1806 	.nop = VCN_ENC_CMD_NO_OP,
1807 	.get_rptr = vcn_v4_0_unified_ring_get_rptr,
1808 	.get_wptr = vcn_v4_0_unified_ring_get_wptr,
1809 	.set_wptr = vcn_v4_0_unified_ring_set_wptr,
1810 	.patch_cs_in_place = vcn_v4_0_ring_patch_cs_in_place,
1811 	.emit_frame_size =
1812 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1813 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1814 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1815 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1816 		1, /* vcn_v2_0_enc_ring_insert_end */
1817 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1818 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
1819 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
1820 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1821 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
1822 	.test_ib = amdgpu_vcn_unified_ring_test_ib,
1823 	.insert_nop = amdgpu_ring_insert_nop,
1824 	.insert_end = vcn_v2_0_enc_ring_insert_end,
1825 	.pad_ib = amdgpu_ring_generic_pad_ib,
1826 	.begin_use = amdgpu_vcn_ring_begin_use,
1827 	.end_use = amdgpu_vcn_ring_end_use,
1828 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1829 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1830 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1831 };
1832 
1833 /**
1834  * vcn_v4_0_set_unified_ring_funcs - set unified ring functions
1835  *
1836  * @adev: amdgpu_device pointer
1837  *
1838  * Set unified ring functions
1839  */
1840 static void vcn_v4_0_set_unified_ring_funcs(struct amdgpu_device *adev)
1841 {
1842 	int i;
1843 
1844 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1845 		if (adev->vcn.harvest_config & (1 << i))
1846 			continue;
1847 
1848 		adev->vcn.inst[i].ring_enc[0].funcs = &vcn_v4_0_unified_ring_vm_funcs;
1849 		adev->vcn.inst[i].ring_enc[0].me = i;
1850 
1851 		DRM_INFO("VCN(%d) encode/decode are enabled in VM mode\n", i);
1852 	}
1853 }
1854 
1855 /**
1856  * vcn_v4_0_is_idle - check VCN block is idle
1857  *
1858  * @handle: amdgpu_device pointer
1859  *
1860  * Check whether VCN block is idle
1861  */
1862 static bool vcn_v4_0_is_idle(void *handle)
1863 {
1864 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1865 	int i, ret = 1;
1866 
1867 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1868 		if (adev->vcn.harvest_config & (1 << i))
1869 			continue;
1870 
1871 		ret &= (RREG32_SOC15(VCN, i, regUVD_STATUS) == UVD_STATUS__IDLE);
1872 	}
1873 
1874 	return ret;
1875 }
1876 
1877 /**
1878  * vcn_v4_0_wait_for_idle - wait for VCN block idle
1879  *
1880  * @handle: amdgpu_device pointer
1881  *
1882  * Wait for VCN block idle
1883  */
1884 static int vcn_v4_0_wait_for_idle(void *handle)
1885 {
1886 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1887 	int i, ret = 0;
1888 
1889 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1890 		if (adev->vcn.harvest_config & (1 << i))
1891 			continue;
1892 
1893 		ret = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE,
1894 			UVD_STATUS__IDLE);
1895 		if (ret)
1896 			return ret;
1897 	}
1898 
1899 	return ret;
1900 }
1901 
1902 /**
1903  * vcn_v4_0_set_clockgating_state - set VCN block clockgating state
1904  *
1905  * @handle: amdgpu_device pointer
1906  * @state: clock gating state
1907  *
1908  * Set VCN block clockgating state
1909  */
1910 static int vcn_v4_0_set_clockgating_state(void *handle, enum amd_clockgating_state state)
1911 {
1912 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1913 	bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
1914 	int i;
1915 
1916 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1917 		if (adev->vcn.harvest_config & (1 << i))
1918 			continue;
1919 
1920 		if (enable) {
1921 			if (RREG32_SOC15(VCN, i, regUVD_STATUS) != UVD_STATUS__IDLE)
1922 				return -EBUSY;
1923 			vcn_v4_0_enable_clock_gating(adev, i);
1924 		} else {
1925 			vcn_v4_0_disable_clock_gating(adev, i);
1926 		}
1927 	}
1928 
1929 	return 0;
1930 }
1931 
1932 /**
1933  * vcn_v4_0_set_powergating_state - set VCN block powergating state
1934  *
1935  * @handle: amdgpu_device pointer
1936  * @state: power gating state
1937  *
1938  * Set VCN block powergating state
1939  */
1940 static int vcn_v4_0_set_powergating_state(void *handle, enum amd_powergating_state state)
1941 {
1942 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1943 	int ret;
1944 
1945 	/* for SRIOV, guest should not control VCN Power-gating
1946 	 * MMSCH FW should control Power-gating and clock-gating
1947 	 * guest should avoid touching CGC and PG
1948 	 */
1949 	if (amdgpu_sriov_vf(adev)) {
1950 		adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
1951 		return 0;
1952 	}
1953 
1954 	if(state == adev->vcn.cur_state)
1955 		return 0;
1956 
1957 	if (state == AMD_PG_STATE_GATE)
1958 		ret = vcn_v4_0_stop(adev);
1959 	else
1960 		ret = vcn_v4_0_start(adev);
1961 
1962 	if(!ret)
1963 		adev->vcn.cur_state = state;
1964 
1965 	return ret;
1966 }
1967 
1968 /**
1969  * vcn_v4_0_set_interrupt_state - set VCN block interrupt state
1970  *
1971  * @adev: amdgpu_device pointer
1972  * @source: interrupt sources
1973  * @type: interrupt types
1974  * @state: interrupt states
1975  *
1976  * Set VCN block interrupt state
1977  */
1978 static int vcn_v4_0_set_interrupt_state(struct amdgpu_device *adev, struct amdgpu_irq_src *source,
1979       unsigned type, enum amdgpu_interrupt_state state)
1980 {
1981 	return 0;
1982 }
1983 
1984 /**
1985  * vcn_v4_0_set_ras_interrupt_state - set VCN block RAS interrupt state
1986  *
1987  * @adev: amdgpu_device pointer
1988  * @source: interrupt sources
1989  * @type: interrupt types
1990  * @state: interrupt states
1991  *
1992  * Set VCN block RAS interrupt state
1993  */
1994 static int vcn_v4_0_set_ras_interrupt_state(struct amdgpu_device *adev,
1995 	struct amdgpu_irq_src *source,
1996 	unsigned int type,
1997 	enum amdgpu_interrupt_state state)
1998 {
1999 	return 0;
2000 }
2001 
2002 /**
2003  * vcn_v4_0_process_interrupt - process VCN block interrupt
2004  *
2005  * @adev: amdgpu_device pointer
2006  * @source: interrupt sources
2007  * @entry: interrupt entry from clients and sources
2008  *
2009  * Process VCN block interrupt
2010  */
2011 static int vcn_v4_0_process_interrupt(struct amdgpu_device *adev, struct amdgpu_irq_src *source,
2012       struct amdgpu_iv_entry *entry)
2013 {
2014 	uint32_t ip_instance;
2015 
2016 	switch (entry->client_id) {
2017 	case SOC15_IH_CLIENTID_VCN:
2018 		ip_instance = 0;
2019 		break;
2020 	case SOC15_IH_CLIENTID_VCN1:
2021 		ip_instance = 1;
2022 		break;
2023 	default:
2024 		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
2025 		return 0;
2026 	}
2027 
2028 	DRM_DEBUG("IH: VCN TRAP\n");
2029 
2030 	switch (entry->src_id) {
2031 	case VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
2032 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
2033 		break;
2034 	default:
2035 		DRM_ERROR("Unhandled interrupt: %d %d\n",
2036 			  entry->src_id, entry->src_data[0]);
2037 		break;
2038 	}
2039 
2040 	return 0;
2041 }
2042 
2043 static const struct amdgpu_irq_src_funcs vcn_v4_0_irq_funcs = {
2044 	.set = vcn_v4_0_set_interrupt_state,
2045 	.process = vcn_v4_0_process_interrupt,
2046 };
2047 
2048 static const struct amdgpu_irq_src_funcs vcn_v4_0_ras_irq_funcs = {
2049 	.set = vcn_v4_0_set_ras_interrupt_state,
2050 	.process = amdgpu_vcn_process_poison_irq,
2051 };
2052 
2053 /**
2054  * vcn_v4_0_set_irq_funcs - set VCN block interrupt irq functions
2055  *
2056  * @adev: amdgpu_device pointer
2057  *
2058  * Set VCN block interrupt irq functions
2059  */
2060 static void vcn_v4_0_set_irq_funcs(struct amdgpu_device *adev)
2061 {
2062 	int i;
2063 
2064 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2065 		if (adev->vcn.harvest_config & (1 << i))
2066 			continue;
2067 
2068 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
2069 		adev->vcn.inst[i].irq.funcs = &vcn_v4_0_irq_funcs;
2070 
2071 		adev->vcn.inst[i].ras_poison_irq.num_types = adev->vcn.num_enc_rings + 1;
2072 		adev->vcn.inst[i].ras_poison_irq.funcs = &vcn_v4_0_ras_irq_funcs;
2073 	}
2074 }
2075 
2076 static const struct amd_ip_funcs vcn_v4_0_ip_funcs = {
2077 	.name = "vcn_v4_0",
2078 	.early_init = vcn_v4_0_early_init,
2079 	.late_init = NULL,
2080 	.sw_init = vcn_v4_0_sw_init,
2081 	.sw_fini = vcn_v4_0_sw_fini,
2082 	.hw_init = vcn_v4_0_hw_init,
2083 	.hw_fini = vcn_v4_0_hw_fini,
2084 	.suspend = vcn_v4_0_suspend,
2085 	.resume = vcn_v4_0_resume,
2086 	.is_idle = vcn_v4_0_is_idle,
2087 	.wait_for_idle = vcn_v4_0_wait_for_idle,
2088 	.check_soft_reset = NULL,
2089 	.pre_soft_reset = NULL,
2090 	.soft_reset = NULL,
2091 	.post_soft_reset = NULL,
2092 	.set_clockgating_state = vcn_v4_0_set_clockgating_state,
2093 	.set_powergating_state = vcn_v4_0_set_powergating_state,
2094 };
2095 
2096 const struct amdgpu_ip_block_version vcn_v4_0_ip_block =
2097 {
2098 	.type = AMD_IP_BLOCK_TYPE_VCN,
2099 	.major = 4,
2100 	.minor = 0,
2101 	.rev = 0,
2102 	.funcs = &vcn_v4_0_ip_funcs,
2103 };
2104 
2105 static uint32_t vcn_v4_0_query_poison_by_instance(struct amdgpu_device *adev,
2106 			uint32_t instance, uint32_t sub_block)
2107 {
2108 	uint32_t poison_stat = 0, reg_value = 0;
2109 
2110 	switch (sub_block) {
2111 	case AMDGPU_VCN_V4_0_VCPU_VCODEC:
2112 		reg_value = RREG32_SOC15(VCN, instance, regUVD_RAS_VCPU_VCODEC_STATUS);
2113 		poison_stat = REG_GET_FIELD(reg_value, UVD_RAS_VCPU_VCODEC_STATUS, POISONED_PF);
2114 		break;
2115 	default:
2116 		break;
2117 	}
2118 
2119 	if (poison_stat)
2120 		dev_info(adev->dev, "Poison detected in VCN%d, sub_block%d\n",
2121 			instance, sub_block);
2122 
2123 	return poison_stat;
2124 }
2125 
2126 static bool vcn_v4_0_query_ras_poison_status(struct amdgpu_device *adev)
2127 {
2128 	uint32_t inst, sub;
2129 	uint32_t poison_stat = 0;
2130 
2131 	for (inst = 0; inst < adev->vcn.num_vcn_inst; inst++)
2132 		for (sub = 0; sub < AMDGPU_VCN_V4_0_MAX_SUB_BLOCK; sub++)
2133 			poison_stat +=
2134 				vcn_v4_0_query_poison_by_instance(adev, inst, sub);
2135 
2136 	return !!poison_stat;
2137 }
2138 
2139 const struct amdgpu_ras_block_hw_ops vcn_v4_0_ras_hw_ops = {
2140 	.query_poison_status = vcn_v4_0_query_ras_poison_status,
2141 };
2142 
2143 static struct amdgpu_vcn_ras vcn_v4_0_ras = {
2144 	.ras_block = {
2145 		.hw_ops = &vcn_v4_0_ras_hw_ops,
2146 		.ras_late_init = amdgpu_vcn_ras_late_init,
2147 	},
2148 };
2149 
2150 static void vcn_v4_0_set_ras_funcs(struct amdgpu_device *adev)
2151 {
2152 	switch (adev->ip_versions[VCN_HWIP][0]) {
2153 	case IP_VERSION(4, 0, 0):
2154 		adev->vcn.ras = &vcn_v4_0_ras;
2155 		break;
2156 	default:
2157 		break;
2158 	}
2159 }
2160