xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/vcn_v4_0.c (revision 0cb4228f)
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
72  *
73  * @handle: amdgpu_device pointer
74  *
75  * Set ring and irq function pointers
76  */
77 static int vcn_v4_0_early_init(void *handle)
78 {
79 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
80 
81 	if (amdgpu_sriov_vf(adev))
82 		adev->vcn.harvest_config = VCN_HARVEST_MMSCH;
83 
84 	/* re-use enc ring as unified ring */
85 	adev->vcn.num_enc_rings = 1;
86 
87 	vcn_v4_0_set_unified_ring_funcs(adev);
88 	vcn_v4_0_set_irq_funcs(adev);
89 	vcn_v4_0_set_ras_funcs(adev);
90 
91 	return 0;
92 }
93 
94 /**
95  * vcn_v4_0_sw_init - sw init for VCN block
96  *
97  * @handle: amdgpu_device pointer
98  *
99  * Load firmware and sw initialization
100  */
101 static int vcn_v4_0_sw_init(void *handle)
102 {
103 	struct amdgpu_ring *ring;
104 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
105 	int i, r;
106 
107 	r = amdgpu_vcn_sw_init(adev);
108 	if (r)
109 		return r;
110 
111 	amdgpu_vcn_setup_ucode(adev);
112 
113 	r = amdgpu_vcn_resume(adev);
114 	if (r)
115 		return r;
116 
117 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
118 		volatile struct amdgpu_vcn4_fw_shared *fw_shared;
119 
120 		if (adev->vcn.harvest_config & (1 << i))
121 			continue;
122 
123 		atomic_set(&adev->vcn.inst[i].sched_score, 0);
124 
125 		/* VCN UNIFIED TRAP */
126 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
127 				VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);
128 		if (r)
129 			return r;
130 
131 		/* VCN POISON TRAP */
132 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
133 				VCN_4_0__SRCID_UVD_POISON, &adev->vcn.inst[i].irq);
134 		if (r)
135 			return r;
136 
137 		ring = &adev->vcn.inst[i].ring_enc[0];
138 		ring->use_doorbell = true;
139 		if (amdgpu_sriov_vf(adev))
140 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + i * (adev->vcn.num_enc_rings + 1) + 1;
141 		else
142 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + 8 * i;
143 
144 		sprintf(ring->name, "vcn_unified_%d", i);
145 
146 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
147 						AMDGPU_RING_PRIO_0, &adev->vcn.inst[i].sched_score);
148 		if (r)
149 			return r;
150 
151 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
152 		fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_FW_SHARED_FLAG_0_UNIFIED_QUEUE);
153 		fw_shared->sq.is_enabled = 1;
154 
155 		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_SMU_DPM_INTERFACE_FLAG);
156 		fw_shared->smu_dpm_interface.smu_interface_type = (adev->flags & AMD_IS_APU) ?
157 			AMDGPU_VCN_SMU_DPM_INTERFACE_APU : AMDGPU_VCN_SMU_DPM_INTERFACE_DGPU;
158 
159 		if (amdgpu_sriov_vf(adev))
160 			fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
161 
162 		if (amdgpu_vcnfw_log)
163 			amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
164 	}
165 
166 	if (amdgpu_sriov_vf(adev)) {
167 		r = amdgpu_virt_alloc_mm_table(adev);
168 		if (r)
169 			return r;
170 	}
171 
172 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
173 		adev->vcn.pause_dpg_mode = vcn_v4_0_pause_dpg_mode;
174 
175 	return 0;
176 }
177 
178 /**
179  * vcn_v4_0_sw_fini - sw fini for VCN block
180  *
181  * @handle: amdgpu_device pointer
182  *
183  * VCN suspend and free up sw allocation
184  */
185 static int vcn_v4_0_sw_fini(void *handle)
186 {
187 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
188 	int i, r, idx;
189 
190 	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
191 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
192 			volatile struct amdgpu_vcn4_fw_shared *fw_shared;
193 
194 			if (adev->vcn.harvest_config & (1 << i))
195 				continue;
196 
197 			fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
198 			fw_shared->present_flag_0 = 0;
199 			fw_shared->sq.is_enabled = 0;
200 		}
201 
202 		drm_dev_exit(idx);
203 	}
204 
205 	if (amdgpu_sriov_vf(adev))
206 		amdgpu_virt_free_mm_table(adev);
207 
208 	r = amdgpu_vcn_suspend(adev);
209 	if (r)
210 		return r;
211 
212 	r = amdgpu_vcn_sw_fini(adev);
213 
214 	return r;
215 }
216 
217 /**
218  * vcn_v4_0_hw_init - start and test VCN block
219  *
220  * @handle: amdgpu_device pointer
221  *
222  * Initialize the hardware, boot up the VCPU and do some testing
223  */
224 static int vcn_v4_0_hw_init(void *handle)
225 {
226 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
227 	struct amdgpu_ring *ring;
228 	int i, r;
229 
230 	if (amdgpu_sriov_vf(adev)) {
231 		r = vcn_v4_0_start_sriov(adev);
232 		if (r)
233 			goto done;
234 
235 		for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
236 			if (adev->vcn.harvest_config & (1 << i))
237 				continue;
238 
239 			ring = &adev->vcn.inst[i].ring_enc[0];
240 			if (amdgpu_vcn_is_disabled_vcn(adev, VCN_ENCODE_RING, i)) {
241 				ring->sched.ready = false;
242 				ring->no_scheduler = true;
243 				dev_info(adev->dev, "ring %s is disabled by hypervisor\n", ring->name);
244 			} else {
245 				ring->wptr = 0;
246 				ring->wptr_old = 0;
247 				vcn_v4_0_unified_ring_set_wptr(ring);
248 				ring->sched.ready = true;
249 			}
250 		}
251 	} else {
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 
258 			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
259 					((adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i), i);
260 
261 			r = amdgpu_ring_test_helper(ring);
262 			if (r)
263 				goto done;
264 
265 		}
266 	}
267 
268 done:
269 	if (!r)
270 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
271 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
272 
273 	return r;
274 }
275 
276 /**
277  * vcn_v4_0_hw_fini - stop the hardware block
278  *
279  * @handle: amdgpu_device pointer
280  *
281  * Stop the VCN block, mark ring as not ready any more
282  */
283 static int vcn_v4_0_hw_fini(void *handle)
284 {
285 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
286 	int i;
287 
288 	cancel_delayed_work_sync(&adev->vcn.idle_work);
289 
290 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
291 		if (adev->vcn.harvest_config & (1 << i))
292 			continue;
293 		if (!amdgpu_sriov_vf(adev)) {
294 			if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
295                         (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
296                                 RREG32_SOC15(VCN, i, regUVD_STATUS))) {
297                         vcn_v4_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
298 			}
299 		}
300 
301 		amdgpu_irq_put(adev, &adev->vcn.inst[i].irq, 0);
302 	}
303 
304 	return 0;
305 }
306 
307 /**
308  * vcn_v4_0_suspend - suspend VCN block
309  *
310  * @handle: amdgpu_device pointer
311  *
312  * HW fini and suspend VCN block
313  */
314 static int vcn_v4_0_suspend(void *handle)
315 {
316 	int r;
317 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
318 
319 	r = vcn_v4_0_hw_fini(adev);
320 	if (r)
321 		return r;
322 
323 	r = amdgpu_vcn_suspend(adev);
324 
325 	return r;
326 }
327 
328 /**
329  * vcn_v4_0_resume - resume VCN block
330  *
331  * @handle: amdgpu_device pointer
332  *
333  * Resume firmware and hw init VCN block
334  */
335 static int vcn_v4_0_resume(void *handle)
336 {
337 	int r;
338 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
339 
340 	r = amdgpu_vcn_resume(adev);
341 	if (r)
342 		return r;
343 
344 	r = vcn_v4_0_hw_init(adev);
345 
346 	return r;
347 }
348 
349 /**
350  * vcn_v4_0_mc_resume - memory controller programming
351  *
352  * @adev: amdgpu_device pointer
353  * @inst: instance number
354  *
355  * Let the VCN memory controller know it's offsets
356  */
357 static void vcn_v4_0_mc_resume(struct amdgpu_device *adev, int inst)
358 {
359 	uint32_t offset, size;
360 	const struct common_firmware_header *hdr;
361 
362 	hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
363 	size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
364 
365 	/* cache window 0: fw */
366 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
367 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
368 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo));
369 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
370 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi));
371 		WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, 0);
372 		offset = 0;
373 	} else {
374 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
375 			lower_32_bits(adev->vcn.inst[inst].gpu_addr));
376 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
377 			upper_32_bits(adev->vcn.inst[inst].gpu_addr));
378 		offset = size;
379                 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
380 	}
381 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE0, size);
382 
383 	/* cache window 1: stack */
384 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
385 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
386 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
387 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
388 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET1, 0);
389 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
390 
391 	/* cache window 2: context */
392 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
393 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
394 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
395 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
396 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET2, 0);
397 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
398 
399 	/* non-cache window */
400 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
401 		lower_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
402 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
403 		upper_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
404 	WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_OFFSET0, 0);
405 	WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_SIZE0,
406 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
407 }
408 
409 /**
410  * vcn_v4_0_mc_resume_dpg_mode - memory controller programming for dpg mode
411  *
412  * @adev: amdgpu_device pointer
413  * @inst_idx: instance number index
414  * @indirect: indirectly write sram
415  *
416  * Let the VCN memory controller know it's offsets with dpg mode
417  */
418 static void vcn_v4_0_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
419 {
420 	uint32_t offset, size;
421 	const struct common_firmware_header *hdr;
422 	hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
423 	size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
424 
425 	/* cache window 0: fw */
426 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
427 		if (!indirect) {
428 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
429 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
430 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
431 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
432 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
433 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
434 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
435 				VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
436 		} else {
437 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
438 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
439 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
440 				VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
441 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
442 				VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
443 		}
444 		offset = 0;
445 	} else {
446 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
447 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
448 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
449 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
450 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
451 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
452 		offset = size;
453 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
454 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0),
455 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
456 	}
457 
458 	if (!indirect)
459 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
460 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
461 	else
462 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
463 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
464 
465 	/* cache window 1: stack */
466 	if (!indirect) {
467 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
468 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
469 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
470 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
471 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
472 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
473 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
474 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
475 	} else {
476 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
477 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
478 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
479 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
480 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
481 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
482 	}
483 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
484 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
485 
486 	/* cache window 2: context */
487 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
488 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
489 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
490 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
491 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
492 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
493 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
494 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
495 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
496 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
497 
498 	/* non-cache window */
499 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
500 			VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
501 			lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
502 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
503 			VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
504 			upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
505 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
506 			VCN, inst_idx, regUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
507 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
508 			VCN, inst_idx, regUVD_VCPU_NONCACHE_SIZE0),
509 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)), 0, indirect);
510 
511 	/* VCN global tiling registers */
512 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
513 		VCN, 0, regUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
514 }
515 
516 /**
517  * vcn_v4_0_disable_static_power_gating - disable VCN static power gating
518  *
519  * @adev: amdgpu_device pointer
520  * @inst: instance number
521  *
522  * Disable static power gating for VCN block
523  */
524 static void vcn_v4_0_disable_static_power_gating(struct amdgpu_device *adev, int inst)
525 {
526 	uint32_t data = 0;
527 
528 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
529 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
530 			| 1 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
531 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
532 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
533 			| 2 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
534 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
535 			| 2 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
536 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
537 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
538 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
539 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
540 			| 2 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
541 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
542 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
543 
544 		WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
545 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS,
546 			UVD_PGFSM_STATUS__UVDM_UVDU_UVDLM_PWR_ON_3_0, 0x3F3FFFFF);
547 	} else {
548 		uint32_t value;
549 
550 		value = (inst) ? 0x2200800 : 0;
551 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
552 			| 1 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
553 			| 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
554 			| 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
555 			| 1 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
556 			| 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
557 			| 1 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
558 			| 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
559 			| 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
560 			| 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
561 			| 1 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
562 			| 1 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
563 			| 1 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
564 			| 1 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
565 
566                 WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
567                 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS, value,  0x3F3FFFFF);
568         }
569 
570         data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
571         data &= ~0x103;
572         if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
573                 data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
574                         UVD_POWER_STATUS__UVD_PG_EN_MASK;
575 
576         WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
577 
578         return;
579 }
580 
581 /**
582  * vcn_v4_0_enable_static_power_gating - enable VCN static power gating
583  *
584  * @adev: amdgpu_device pointer
585  * @inst: instance number
586  *
587  * Enable static power gating for VCN block
588  */
589 static void vcn_v4_0_enable_static_power_gating(struct amdgpu_device *adev, int inst)
590 {
591 	uint32_t data;
592 
593 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
594 		/* Before power off, this indicator has to be turned on */
595 		data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
596 		data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
597 		data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
598 		WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
599 
600 		data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
601 			| 2 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
602 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
603 			| 2 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
604 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
605 			| 2 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
606 			| 2 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
607 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
608 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
609 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
610 			| 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
611 			| 2 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
612 			| 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
613 			| 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
614 		WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
615 
616 		data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
617 			| 2 << UVD_PGFSM_STATUS__UVDS_PWR_STATUS__SHIFT
618 			| 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
619 			| 2 << UVD_PGFSM_STATUS__UVDTC_PWR_STATUS__SHIFT
620 			| 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
621 			| 2 << UVD_PGFSM_STATUS__UVDTA_PWR_STATUS__SHIFT
622 			| 2 << UVD_PGFSM_STATUS__UVDLM_PWR_STATUS__SHIFT
623 			| 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
624 			| 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
625 			| 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT
626 			| 2 << UVD_PGFSM_STATUS__UVDAB_PWR_STATUS__SHIFT
627 			| 2 << UVD_PGFSM_STATUS__UVDTB_PWR_STATUS__SHIFT
628 			| 2 << UVD_PGFSM_STATUS__UVDNA_PWR_STATUS__SHIFT
629 			| 2 << UVD_PGFSM_STATUS__UVDNB_PWR_STATUS__SHIFT);
630 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS, data, 0x3F3FFFFF);
631 	}
632 
633         return;
634 }
635 
636 /**
637  * vcn_v4_0_disable_clock_gating - disable VCN clock gating
638  *
639  * @adev: amdgpu_device pointer
640  * @inst: instance number
641  *
642  * Disable clock gating for VCN block
643  */
644 static void vcn_v4_0_disable_clock_gating(struct amdgpu_device *adev, int inst)
645 {
646 	uint32_t data;
647 
648 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
649 		return;
650 
651 	/* VCN disable CGC */
652 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
653 	data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
654 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
655 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
656 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
657 
658 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_GATE);
659 	data &= ~(UVD_CGC_GATE__SYS_MASK
660 		| UVD_CGC_GATE__UDEC_MASK
661 		| UVD_CGC_GATE__MPEG2_MASK
662 		| UVD_CGC_GATE__REGS_MASK
663 		| UVD_CGC_GATE__RBC_MASK
664 		| UVD_CGC_GATE__LMI_MC_MASK
665 		| UVD_CGC_GATE__LMI_UMC_MASK
666 		| UVD_CGC_GATE__IDCT_MASK
667 		| UVD_CGC_GATE__MPRD_MASK
668 		| UVD_CGC_GATE__MPC_MASK
669 		| UVD_CGC_GATE__LBSI_MASK
670 		| UVD_CGC_GATE__LRBBM_MASK
671 		| UVD_CGC_GATE__UDEC_RE_MASK
672 		| UVD_CGC_GATE__UDEC_CM_MASK
673 		| UVD_CGC_GATE__UDEC_IT_MASK
674 		| UVD_CGC_GATE__UDEC_DB_MASK
675 		| UVD_CGC_GATE__UDEC_MP_MASK
676 		| UVD_CGC_GATE__WCB_MASK
677 		| UVD_CGC_GATE__VCPU_MASK
678 		| UVD_CGC_GATE__MMSCH_MASK);
679 
680 	WREG32_SOC15(VCN, inst, regUVD_CGC_GATE, data);
681 	SOC15_WAIT_ON_RREG(VCN, inst, regUVD_CGC_GATE, 0,  0xFFFFFFFF);
682 
683 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
684 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
685 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
686 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
687 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
688 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
689 		| UVD_CGC_CTRL__SYS_MODE_MASK
690 		| UVD_CGC_CTRL__UDEC_MODE_MASK
691 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
692 		| UVD_CGC_CTRL__REGS_MODE_MASK
693 		| UVD_CGC_CTRL__RBC_MODE_MASK
694 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
695 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
696 		| UVD_CGC_CTRL__IDCT_MODE_MASK
697 		| UVD_CGC_CTRL__MPRD_MODE_MASK
698 		| UVD_CGC_CTRL__MPC_MODE_MASK
699 		| UVD_CGC_CTRL__LBSI_MODE_MASK
700 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
701 		| UVD_CGC_CTRL__WCB_MODE_MASK
702 		| UVD_CGC_CTRL__VCPU_MODE_MASK
703 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
704 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
705 
706 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE);
707 	data |= (UVD_SUVD_CGC_GATE__SRE_MASK
708 		| UVD_SUVD_CGC_GATE__SIT_MASK
709 		| UVD_SUVD_CGC_GATE__SMP_MASK
710 		| UVD_SUVD_CGC_GATE__SCM_MASK
711 		| UVD_SUVD_CGC_GATE__SDB_MASK
712 		| UVD_SUVD_CGC_GATE__SRE_H264_MASK
713 		| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
714 		| UVD_SUVD_CGC_GATE__SIT_H264_MASK
715 		| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
716 		| UVD_SUVD_CGC_GATE__SCM_H264_MASK
717 		| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
718 		| UVD_SUVD_CGC_GATE__SDB_H264_MASK
719 		| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
720 		| UVD_SUVD_CGC_GATE__SCLR_MASK
721 		| UVD_SUVD_CGC_GATE__UVD_SC_MASK
722 		| UVD_SUVD_CGC_GATE__ENT_MASK
723 		| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
724 		| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
725 		| UVD_SUVD_CGC_GATE__SITE_MASK
726 		| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
727 		| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
728 		| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
729 		| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
730 		| UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
731 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE, data);
732 
733 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
734 	data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
735 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
736 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
737 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
738 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
739 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
740 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
741 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
742 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
743 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
744 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
745 }
746 
747 /**
748  * vcn_v4_0_disable_clock_gating_dpg_mode - disable VCN clock gating dpg mode
749  *
750  * @adev: amdgpu_device pointer
751  * @sram_sel: sram select
752  * @inst_idx: instance number index
753  * @indirect: indirectly write sram
754  *
755  * Disable clock gating for VCN block with dpg mode
756  */
757 static void vcn_v4_0_disable_clock_gating_dpg_mode(struct amdgpu_device *adev, uint8_t sram_sel,
758       int inst_idx, uint8_t indirect)
759 {
760 	uint32_t reg_data = 0;
761 
762 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
763 		return;
764 
765 	/* enable sw clock gating control */
766 	reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
767 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
768 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
769 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
770 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
771 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
772 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
773 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
774 		 UVD_CGC_CTRL__SYS_MODE_MASK |
775 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
776 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
777 		 UVD_CGC_CTRL__REGS_MODE_MASK |
778 		 UVD_CGC_CTRL__RBC_MODE_MASK |
779 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
780 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
781 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
782 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
783 		 UVD_CGC_CTRL__MPC_MODE_MASK |
784 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
785 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
786 		 UVD_CGC_CTRL__WCB_MODE_MASK |
787 		 UVD_CGC_CTRL__VCPU_MODE_MASK);
788 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
789 		VCN, inst_idx, regUVD_CGC_CTRL), reg_data, sram_sel, indirect);
790 
791 	/* turn off clock gating */
792 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
793 		VCN, inst_idx, regUVD_CGC_GATE), 0, sram_sel, indirect);
794 
795 	/* turn on SUVD clock gating */
796 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
797 		VCN, inst_idx, regUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
798 
799 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
800 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
801 		VCN, inst_idx, regUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
802 }
803 
804 /**
805  * vcn_v4_0_enable_clock_gating - enable VCN clock gating
806  *
807  * @adev: amdgpu_device pointer
808  * @inst: instance number
809  *
810  * Enable clock gating for VCN block
811  */
812 static void vcn_v4_0_enable_clock_gating(struct amdgpu_device *adev, int inst)
813 {
814 	uint32_t data;
815 
816 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
817 		return;
818 
819 	/* enable VCN CGC */
820 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
821 	data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
822 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
823 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
824 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
825 
826 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
827 	data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
828 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
829 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
830 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
831 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
832 		| UVD_CGC_CTRL__SYS_MODE_MASK
833 		| UVD_CGC_CTRL__UDEC_MODE_MASK
834 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
835 		| UVD_CGC_CTRL__REGS_MODE_MASK
836 		| UVD_CGC_CTRL__RBC_MODE_MASK
837 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
838 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
839 		| UVD_CGC_CTRL__IDCT_MODE_MASK
840 		| UVD_CGC_CTRL__MPRD_MODE_MASK
841 		| UVD_CGC_CTRL__MPC_MODE_MASK
842 		| UVD_CGC_CTRL__LBSI_MODE_MASK
843 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
844 		| UVD_CGC_CTRL__WCB_MODE_MASK
845 		| UVD_CGC_CTRL__VCPU_MODE_MASK
846 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
847 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
848 
849 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
850 	data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
851 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
852 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
853 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
854 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
855 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
856 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
857 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
858 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
859 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
860 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
861 
862 	return;
863 }
864 
865 static void vcn_v4_0_enable_ras(struct amdgpu_device *adev, int inst_idx,
866 				bool indirect)
867 {
868 	uint32_t tmp;
869 
870 	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN))
871 		return;
872 
873 	tmp = VCN_RAS_CNTL__VCPU_VCODEC_REARM_MASK |
874 	      VCN_RAS_CNTL__VCPU_VCODEC_IH_EN_MASK |
875 	      VCN_RAS_CNTL__VCPU_VCODEC_PMI_EN_MASK |
876 	      VCN_RAS_CNTL__VCPU_VCODEC_STALL_EN_MASK;
877 	WREG32_SOC15_DPG_MODE(inst_idx,
878 			      SOC15_DPG_MODE_OFFSET(VCN, 0, regVCN_RAS_CNTL),
879 			      tmp, 0, indirect);
880 
881 	tmp = UVD_SYS_INT_EN__RASCNTL_VCPU_VCODEC_EN_MASK;
882 	WREG32_SOC15_DPG_MODE(inst_idx,
883 			      SOC15_DPG_MODE_OFFSET(VCN, 0, regUVD_SYS_INT_EN),
884 			      tmp, 0, indirect);
885 }
886 
887 /**
888  * vcn_v4_0_start_dpg_mode - VCN start with dpg mode
889  *
890  * @adev: amdgpu_device pointer
891  * @inst_idx: instance number index
892  * @indirect: indirectly write sram
893  *
894  * Start VCN block with dpg mode
895  */
896 static int vcn_v4_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
897 {
898 	volatile struct amdgpu_vcn4_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
899 	struct amdgpu_ring *ring;
900 	uint32_t tmp;
901 
902 	/* disable register anti-hang mechanism */
903 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 1,
904 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
905 	/* enable dynamic power gating mode */
906 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS);
907 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
908 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
909 	WREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS, tmp);
910 
911 	if (indirect)
912 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
913 
914 	/* enable clock gating */
915 	vcn_v4_0_disable_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
916 
917 	/* enable VCPU clock */
918 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
919 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK | UVD_VCPU_CNTL__BLK_RST_MASK;
920 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
921 		VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
922 
923 	/* disable master interupt */
924 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
925 		VCN, inst_idx, regUVD_MASTINT_EN), 0, 0, indirect);
926 
927 	/* setup regUVD_LMI_CTRL */
928 	tmp = (UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
929 		UVD_LMI_CTRL__REQ_MODE_MASK |
930 		UVD_LMI_CTRL__CRC_RESET_MASK |
931 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
932 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
933 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
934 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
935 		0x00100000L);
936 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
937 		VCN, inst_idx, regUVD_LMI_CTRL), tmp, 0, indirect);
938 
939 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
940 		VCN, inst_idx, regUVD_MPC_CNTL),
941 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
942 
943 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
944 		VCN, inst_idx, regUVD_MPC_SET_MUXA0),
945 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
946 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
947 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
948 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
949 
950 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
951 		VCN, inst_idx, regUVD_MPC_SET_MUXB0),
952 		 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
953 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
954 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
955 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
956 
957 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
958 		VCN, inst_idx, regUVD_MPC_SET_MUX),
959 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
960 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
961 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
962 
963 	vcn_v4_0_mc_resume_dpg_mode(adev, inst_idx, indirect);
964 
965 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
966 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
967 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
968 		VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
969 
970 	/* enable LMI MC and UMC channels */
971 	tmp = 0x1f << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT;
972 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
973 		VCN, inst_idx, regUVD_LMI_CTRL2), tmp, 0, indirect);
974 
975 	vcn_v4_0_enable_ras(adev, inst_idx, indirect);
976 
977 	/* enable master interrupt */
978 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
979 		VCN, inst_idx, regUVD_MASTINT_EN),
980 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
981 
982 
983 	if (indirect)
984 		psp_update_vcn_sram(adev, inst_idx, adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
985 			(uint32_t)((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
986 				(uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr));
987 
988 	ring = &adev->vcn.inst[inst_idx].ring_enc[0];
989 
990 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_LO, ring->gpu_addr);
991 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
992 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_SIZE, ring->ring_size / 4);
993 
994 	tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
995 	tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
996 	WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
997 	fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
998 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR, 0);
999 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, 0);
1000 
1001 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR);
1002 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, tmp);
1003 	ring->wptr = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
1004 
1005 	tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
1006 	tmp |= VCN_RB_ENABLE__RB1_EN_MASK;
1007 	WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
1008 	fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
1009 
1010 	WREG32_SOC15(VCN, inst_idx, regVCN_RB1_DB_CTRL,
1011 			ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
1012 			VCN_RB1_DB_CTRL__EN_MASK);
1013 
1014 	return 0;
1015 }
1016 
1017 
1018 /**
1019  * vcn_v4_0_start - VCN start
1020  *
1021  * @adev: amdgpu_device pointer
1022  *
1023  * Start VCN block
1024  */
1025 static int vcn_v4_0_start(struct amdgpu_device *adev)
1026 {
1027 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1028 	struct amdgpu_ring *ring;
1029 	uint32_t tmp;
1030 	int i, j, k, r;
1031 
1032 	if (adev->pm.dpm_enabled)
1033 		amdgpu_dpm_enable_uvd(adev, true);
1034 
1035 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1036 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1037 
1038 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1039 			r = vcn_v4_0_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
1040 			continue;
1041 		}
1042 
1043 		/* disable VCN power gating */
1044 		vcn_v4_0_disable_static_power_gating(adev, i);
1045 
1046 		/* set VCN status busy */
1047 		tmp = RREG32_SOC15(VCN, i, regUVD_STATUS) | UVD_STATUS__UVD_BUSY;
1048 		WREG32_SOC15(VCN, i, regUVD_STATUS, tmp);
1049 
1050 		/*SW clock gating */
1051 		vcn_v4_0_disable_clock_gating(adev, i);
1052 
1053 		/* enable VCPU clock */
1054 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1055 				UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1056 
1057 		/* disable master interrupt */
1058 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN), 0,
1059 				~UVD_MASTINT_EN__VCPU_EN_MASK);
1060 
1061 		/* enable LMI MC and UMC channels */
1062 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_LMI_CTRL2), 0,
1063 				~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1064 
1065 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1066 		tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1067 		tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1068 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1069 
1070 		/* setup regUVD_LMI_CTRL */
1071 		tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL);
1072 		WREG32_SOC15(VCN, i, regUVD_LMI_CTRL, tmp |
1073 				UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
1074 				UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1075 				UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1076 				UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1077 
1078 		/* setup regUVD_MPC_CNTL */
1079 		tmp = RREG32_SOC15(VCN, i, regUVD_MPC_CNTL);
1080 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1081 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1082 		WREG32_SOC15(VCN, i, regUVD_MPC_CNTL, tmp);
1083 
1084 		/* setup UVD_MPC_SET_MUXA0 */
1085 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXA0,
1086 				((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1087 				 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1088 				 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1089 				 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1090 
1091 		/* setup UVD_MPC_SET_MUXB0 */
1092 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXB0,
1093 				((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1094 				 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1095 				 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1096 				 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1097 
1098 		/* setup UVD_MPC_SET_MUX */
1099 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUX,
1100 				((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1101 				 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1102 				 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1103 
1104 		vcn_v4_0_mc_resume(adev, i);
1105 
1106 		/* VCN global tiling registers */
1107 		WREG32_SOC15(VCN, i, regUVD_GFX10_ADDR_CONFIG,
1108 				adev->gfx.config.gb_addr_config);
1109 
1110 		/* unblock VCPU register access */
1111 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL), 0,
1112 				~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1113 
1114 		/* release VCPU reset to boot */
1115 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1116 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1117 
1118 		for (j = 0; j < 10; ++j) {
1119 			uint32_t status;
1120 
1121 			for (k = 0; k < 100; ++k) {
1122 				status = RREG32_SOC15(VCN, i, regUVD_STATUS);
1123 				if (status & 2)
1124 					break;
1125 				mdelay(10);
1126 				if (amdgpu_emu_mode==1)
1127 					msleep(1);
1128 			}
1129 
1130 			if (amdgpu_emu_mode==1) {
1131 				r = -1;
1132 				if (status & 2) {
1133 					r = 0;
1134 					break;
1135 				}
1136 			} else {
1137 				r = 0;
1138 				if (status & 2)
1139 					break;
1140 
1141 				dev_err(adev->dev, "VCN[%d] is not responding, trying to reset the VCPU!!!\n", i);
1142 				WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1143 							UVD_VCPU_CNTL__BLK_RST_MASK,
1144 							~UVD_VCPU_CNTL__BLK_RST_MASK);
1145 				mdelay(10);
1146 				WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1147 						~UVD_VCPU_CNTL__BLK_RST_MASK);
1148 
1149 				mdelay(10);
1150 				r = -1;
1151 			}
1152 		}
1153 
1154 		if (r) {
1155 			dev_err(adev->dev, "VCN[%d] is not responding, giving up!!!\n", i);
1156 			return r;
1157 		}
1158 
1159 		/* enable master interrupt */
1160 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN),
1161 				UVD_MASTINT_EN__VCPU_EN_MASK,
1162 				~UVD_MASTINT_EN__VCPU_EN_MASK);
1163 
1164 		/* clear the busy bit of VCN_STATUS */
1165 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_STATUS), 0,
1166 				~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1167 
1168 		ring = &adev->vcn.inst[i].ring_enc[0];
1169 		WREG32_SOC15(VCN, i, regVCN_RB1_DB_CTRL,
1170 				ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
1171 				VCN_RB1_DB_CTRL__EN_MASK);
1172 
1173 		WREG32_SOC15(VCN, i, regUVD_RB_BASE_LO, ring->gpu_addr);
1174 		WREG32_SOC15(VCN, i, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1175 		WREG32_SOC15(VCN, i, regUVD_RB_SIZE, ring->ring_size / 4);
1176 
1177 		tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1178 		tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
1179 		WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1180 		fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
1181 		WREG32_SOC15(VCN, i, regUVD_RB_RPTR, 0);
1182 		WREG32_SOC15(VCN, i, regUVD_RB_WPTR, 0);
1183 
1184 		tmp = RREG32_SOC15(VCN, i, regUVD_RB_RPTR);
1185 		WREG32_SOC15(VCN, i, regUVD_RB_WPTR, tmp);
1186 		ring->wptr = RREG32_SOC15(VCN, i, regUVD_RB_WPTR);
1187 
1188 		tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1189 		tmp |= VCN_RB_ENABLE__RB1_EN_MASK;
1190 		WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1191 		fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
1192 	}
1193 
1194 	return 0;
1195 }
1196 
1197 static int vcn_v4_0_start_sriov(struct amdgpu_device *adev)
1198 {
1199 	int i;
1200 	struct amdgpu_ring *ring_enc;
1201 	uint64_t cache_addr;
1202 	uint64_t rb_enc_addr;
1203 	uint64_t ctx_addr;
1204 	uint32_t param, resp, expected;
1205 	uint32_t offset, cache_size;
1206 	uint32_t tmp, timeout;
1207 
1208 	struct amdgpu_mm_table *table = &adev->virt.mm_table;
1209 	uint32_t *table_loc;
1210 	uint32_t table_size;
1211 	uint32_t size, size_dw;
1212 	uint32_t init_status;
1213 	uint32_t enabled_vcn;
1214 
1215 	struct mmsch_v4_0_cmd_direct_write
1216 		direct_wt = { {0} };
1217 	struct mmsch_v4_0_cmd_direct_read_modify_write
1218 		direct_rd_mod_wt = { {0} };
1219 	struct mmsch_v4_0_cmd_end end = { {0} };
1220 	struct mmsch_v4_0_init_header header;
1221 
1222 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1223 	volatile struct amdgpu_fw_shared_rb_setup *rb_setup;
1224 
1225 	direct_wt.cmd_header.command_type =
1226 		MMSCH_COMMAND__DIRECT_REG_WRITE;
1227 	direct_rd_mod_wt.cmd_header.command_type =
1228 		MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1229 	end.cmd_header.command_type =
1230 		MMSCH_COMMAND__END;
1231 
1232 	header.version = MMSCH_VERSION;
1233 	header.total_size = sizeof(struct mmsch_v4_0_init_header) >> 2;
1234 	for (i = 0; i < AMDGPU_MAX_VCN_INSTANCES; i++) {
1235 		header.inst[i].init_status = 0;
1236 		header.inst[i].table_offset = 0;
1237 		header.inst[i].table_size = 0;
1238 	}
1239 
1240 	table_loc = (uint32_t *)table->cpu_addr;
1241 	table_loc += header.total_size;
1242 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1243 		if (adev->vcn.harvest_config & (1 << i))
1244 			continue;
1245 
1246 		table_size = 0;
1247 
1248 		MMSCH_V4_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(VCN, i,
1249 			regUVD_STATUS),
1250 			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1251 
1252 		cache_size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1253 
1254 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1255 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1256 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1257 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1258 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1259 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1260 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1261 			offset = 0;
1262 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1263 				regUVD_VCPU_CACHE_OFFSET0),
1264 				0);
1265 		} else {
1266 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1267 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1268 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1269 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1270 				regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1271 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1272 			offset = cache_size;
1273 			MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1274 				regUVD_VCPU_CACHE_OFFSET0),
1275 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1276 		}
1277 
1278 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1279 			regUVD_VCPU_CACHE_SIZE0),
1280 			cache_size);
1281 
1282 		cache_addr = adev->vcn.inst[i].gpu_addr + offset;
1283 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1284 			regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1285 			lower_32_bits(cache_addr));
1286 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1287 			regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1288 			upper_32_bits(cache_addr));
1289 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1290 			regUVD_VCPU_CACHE_OFFSET1),
1291 			0);
1292 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1293 			regUVD_VCPU_CACHE_SIZE1),
1294 			AMDGPU_VCN_STACK_SIZE);
1295 
1296 		cache_addr = adev->vcn.inst[i].gpu_addr + offset +
1297 			AMDGPU_VCN_STACK_SIZE;
1298 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1299 			regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1300 			lower_32_bits(cache_addr));
1301 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1302 			regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1303 			upper_32_bits(cache_addr));
1304 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1305 			regUVD_VCPU_CACHE_OFFSET2),
1306 			0);
1307 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1308 			regUVD_VCPU_CACHE_SIZE2),
1309 			AMDGPU_VCN_CONTEXT_SIZE);
1310 
1311 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1312 		rb_setup = &fw_shared->rb_setup;
1313 
1314 		ring_enc = &adev->vcn.inst[i].ring_enc[0];
1315 		ring_enc->wptr = 0;
1316 		rb_enc_addr = ring_enc->gpu_addr;
1317 
1318 		rb_setup->is_rb_enabled_flags |= RB_ENABLED;
1319 		rb_setup->rb_addr_lo = lower_32_bits(rb_enc_addr);
1320 		rb_setup->rb_addr_hi = upper_32_bits(rb_enc_addr);
1321 		rb_setup->rb_size = ring_enc->ring_size / 4;
1322 		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
1323 
1324 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1325 			regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
1326 			lower_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
1327 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1328 			regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
1329 			upper_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
1330 		MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1331 			regUVD_VCPU_NONCACHE_SIZE0),
1332 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
1333 
1334 		/* add end packet */
1335 		MMSCH_V4_0_INSERT_END();
1336 
1337 		/* refine header */
1338 		header.inst[i].init_status = 0;
1339 		header.inst[i].table_offset = header.total_size;
1340 		header.inst[i].table_size = table_size;
1341 		header.total_size += table_size;
1342 	}
1343 
1344 	/* Update init table header in memory */
1345 	size = sizeof(struct mmsch_v4_0_init_header);
1346 	table_loc = (uint32_t *)table->cpu_addr;
1347 	memcpy((void *)table_loc, &header, size);
1348 
1349 	/* message MMSCH (in VCN[0]) to initialize this client
1350 	 * 1, write to mmsch_vf_ctx_addr_lo/hi register with GPU mc addr
1351 	 * of memory descriptor location
1352 	 */
1353 	ctx_addr = table->gpu_addr;
1354 	WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_ADDR_LO, lower_32_bits(ctx_addr));
1355 	WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_ADDR_HI, upper_32_bits(ctx_addr));
1356 
1357 	/* 2, update vmid of descriptor */
1358 	tmp = RREG32_SOC15(VCN, 0, regMMSCH_VF_VMID);
1359 	tmp &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1360 	/* use domain0 for MM scheduler */
1361 	tmp |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1362 	WREG32_SOC15(VCN, 0, regMMSCH_VF_VMID, tmp);
1363 
1364 	/* 3, notify mmsch about the size of this descriptor */
1365 	size = header.total_size;
1366 	WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_SIZE, size);
1367 
1368 	/* 4, set resp to zero */
1369 	WREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_RESP, 0);
1370 
1371 	/* 5, kick off the initialization and wait until
1372 	 * MMSCH_VF_MAILBOX_RESP becomes non-zero
1373 	 */
1374 	param = 0x00000001;
1375 	WREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_HOST, param);
1376 	tmp = 0;
1377 	timeout = 1000;
1378 	resp = 0;
1379 	expected = MMSCH_VF_MAILBOX_RESP__OK;
1380 	while (resp != expected) {
1381 		resp = RREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_RESP);
1382 		if (resp != 0)
1383 			break;
1384 
1385 		udelay(10);
1386 		tmp = tmp + 10;
1387 		if (tmp >= timeout) {
1388 			DRM_ERROR("failed to init MMSCH. TIME-OUT after %d usec"\
1389 				" waiting for regMMSCH_VF_MAILBOX_RESP "\
1390 				"(expected=0x%08x, readback=0x%08x)\n",
1391 				tmp, expected, resp);
1392 			return -EBUSY;
1393 		}
1394 	}
1395 	enabled_vcn = amdgpu_vcn_is_disabled_vcn(adev, VCN_DECODE_RING, 0) ? 1 : 0;
1396 	init_status = ((struct mmsch_v4_0_init_header *)(table_loc))->inst[enabled_vcn].init_status;
1397 	if (resp != expected && resp != MMSCH_VF_MAILBOX_RESP__INCOMPLETE
1398 	&& init_status != MMSCH_VF_ENGINE_STATUS__PASS)
1399 		DRM_ERROR("MMSCH init status is incorrect! readback=0x%08x, header init "\
1400 			"status for VCN%x: 0x%x\n", resp, enabled_vcn, init_status);
1401 
1402 	return 0;
1403 }
1404 
1405 /**
1406  * vcn_v4_0_stop_dpg_mode - VCN stop with dpg mode
1407  *
1408  * @adev: amdgpu_device pointer
1409  * @inst_idx: instance number index
1410  *
1411  * Stop VCN block with dpg mode
1412  */
1413 static void vcn_v4_0_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1414 {
1415 	uint32_t tmp;
1416 
1417 	/* Wait for power status to be 1 */
1418 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1419 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1420 
1421 	/* wait for read ptr to be equal to write ptr */
1422 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
1423 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1424 
1425 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1426 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1427 
1428 	/* disable dynamic power gating mode */
1429 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 0,
1430 		~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1431 }
1432 
1433 /**
1434  * vcn_v4_0_stop - VCN stop
1435  *
1436  * @adev: amdgpu_device pointer
1437  *
1438  * Stop VCN block
1439  */
1440 static int vcn_v4_0_stop(struct amdgpu_device *adev)
1441 {
1442 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1443 	uint32_t tmp;
1444 	int i, r = 0;
1445 
1446 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1447 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1448 		fw_shared->sq.queue_mode |= FW_QUEUE_DPG_HOLD_OFF;
1449 
1450 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1451 			vcn_v4_0_stop_dpg_mode(adev, i);
1452 			continue;
1453 		}
1454 
1455 		/* wait for vcn idle */
1456 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1457 		if (r)
1458 			return r;
1459 
1460 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1461 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1462 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1463 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1464 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1465 		if (r)
1466 			return r;
1467 
1468 		/* disable LMI UMC channel */
1469 		tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL2);
1470 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1471 		WREG32_SOC15(VCN, i, regUVD_LMI_CTRL2, tmp);
1472 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK |
1473 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1474 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1475 		if (r)
1476 			return r;
1477 
1478 		/* block VCPU register access */
1479 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL),
1480 				UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1481 				~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1482 
1483 		/* reset VCPU */
1484 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1485 				UVD_VCPU_CNTL__BLK_RST_MASK,
1486 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1487 
1488 		/* disable VCPU clock */
1489 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1490 				~(UVD_VCPU_CNTL__CLK_EN_MASK));
1491 
1492 		/* apply soft reset */
1493 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1494 		tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1495 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1496 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1497 		tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1498 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1499 
1500 		/* clear status */
1501 		WREG32_SOC15(VCN, i, regUVD_STATUS, 0);
1502 
1503 		/* apply HW clock gating */
1504 		vcn_v4_0_enable_clock_gating(adev, i);
1505 
1506 		/* enable VCN power gating */
1507 		vcn_v4_0_enable_static_power_gating(adev, i);
1508 	}
1509 
1510 	if (adev->pm.dpm_enabled)
1511 		amdgpu_dpm_enable_uvd(adev, false);
1512 
1513 	return 0;
1514 }
1515 
1516 /**
1517  * vcn_v4_0_pause_dpg_mode - VCN pause with dpg mode
1518  *
1519  * @adev: amdgpu_device pointer
1520  * @inst_idx: instance number index
1521  * @new_state: pause state
1522  *
1523  * Pause dpg mode for VCN block
1524  */
1525 static int vcn_v4_0_pause_dpg_mode(struct amdgpu_device *adev, int inst_idx,
1526       struct dpg_pause_state *new_state)
1527 {
1528 	uint32_t reg_data = 0;
1529 	int ret_code;
1530 
1531 	/* pause/unpause if state is changed */
1532 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1533 		DRM_DEV_DEBUG(adev->dev, "dpg pause state changed %d -> %d",
1534 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1535 		reg_data = RREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE) &
1536 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1537 
1538 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1539 			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 0x1,
1540 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1541 
1542 			if (!ret_code) {
1543 				/* pause DPG */
1544 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1545 				WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1546 
1547 				/* wait for ACK */
1548 				SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_DPG_PAUSE,
1549 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1550 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1551 
1552 				SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS,
1553 					UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1554 			}
1555 		} else {
1556 			/* unpause dpg, no need to wait */
1557 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1558 			WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1559 		}
1560 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1561 	}
1562 
1563 	return 0;
1564 }
1565 
1566 /**
1567  * vcn_v4_0_unified_ring_get_rptr - get unified read pointer
1568  *
1569  * @ring: amdgpu_ring pointer
1570  *
1571  * Returns the current hardware unified read pointer
1572  */
1573 static uint64_t vcn_v4_0_unified_ring_get_rptr(struct amdgpu_ring *ring)
1574 {
1575 	struct amdgpu_device *adev = ring->adev;
1576 
1577 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1578 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1579 
1580 	return RREG32_SOC15(VCN, ring->me, regUVD_RB_RPTR);
1581 }
1582 
1583 /**
1584  * vcn_v4_0_unified_ring_get_wptr - get unified write pointer
1585  *
1586  * @ring: amdgpu_ring pointer
1587  *
1588  * Returns the current hardware unified write pointer
1589  */
1590 static uint64_t vcn_v4_0_unified_ring_get_wptr(struct amdgpu_ring *ring)
1591 {
1592 	struct amdgpu_device *adev = ring->adev;
1593 
1594 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1595 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1596 
1597 	if (ring->use_doorbell)
1598 		return *ring->wptr_cpu_addr;
1599 	else
1600 		return RREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR);
1601 }
1602 
1603 /**
1604  * vcn_v4_0_unified_ring_set_wptr - set enc write pointer
1605  *
1606  * @ring: amdgpu_ring pointer
1607  *
1608  * Commits the enc write pointer to the hardware
1609  */
1610 static void vcn_v4_0_unified_ring_set_wptr(struct amdgpu_ring *ring)
1611 {
1612 	struct amdgpu_device *adev = ring->adev;
1613 
1614 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1615 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1616 
1617 	if (ring->use_doorbell) {
1618 		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1619 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1620 	} else {
1621 		WREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR, lower_32_bits(ring->wptr));
1622 	}
1623 }
1624 
1625 static int vcn_v4_0_limit_sched(struct amdgpu_cs_parser *p,
1626 				struct amdgpu_job *job)
1627 {
1628 	struct drm_gpu_scheduler **scheds;
1629 
1630 	/* The create msg must be in the first IB submitted */
1631 	if (atomic_read(&job->base.entity->fence_seq))
1632 		return -EINVAL;
1633 
1634 	scheds = p->adev->gpu_sched[AMDGPU_HW_IP_VCN_ENC]
1635 		[AMDGPU_RING_PRIO_0].sched;
1636 	drm_sched_entity_modify_sched(job->base.entity, scheds, 1);
1637 	return 0;
1638 }
1639 
1640 static int vcn_v4_0_dec_msg(struct amdgpu_cs_parser *p, struct amdgpu_job *job,
1641 			    uint64_t addr)
1642 {
1643 	struct ttm_operation_ctx ctx = { false, false };
1644 	struct amdgpu_bo_va_mapping *map;
1645 	uint32_t *msg, num_buffers;
1646 	struct amdgpu_bo *bo;
1647 	uint64_t start, end;
1648 	unsigned int i;
1649 	void *ptr;
1650 	int r;
1651 
1652 	addr &= AMDGPU_GMC_HOLE_MASK;
1653 	r = amdgpu_cs_find_mapping(p, addr, &bo, &map);
1654 	if (r) {
1655 		DRM_ERROR("Can't find BO for addr 0x%08llx\n", addr);
1656 		return r;
1657 	}
1658 
1659 	start = map->start * AMDGPU_GPU_PAGE_SIZE;
1660 	end = (map->last + 1) * AMDGPU_GPU_PAGE_SIZE;
1661 	if (addr & 0x7) {
1662 		DRM_ERROR("VCN messages must be 8 byte aligned!\n");
1663 		return -EINVAL;
1664 	}
1665 
1666 	bo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
1667 	amdgpu_bo_placement_from_domain(bo, bo->allowed_domains);
1668 	r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
1669 	if (r) {
1670 		DRM_ERROR("Failed validating the VCN message BO (%d)!\n", r);
1671 		return r;
1672 	}
1673 
1674 	r = amdgpu_bo_kmap(bo, &ptr);
1675 	if (r) {
1676 		DRM_ERROR("Failed mapping the VCN message (%d)!\n", r);
1677 		return r;
1678 	}
1679 
1680 	msg = ptr + addr - start;
1681 
1682 	/* Check length */
1683 	if (msg[1] > end - addr) {
1684 		r = -EINVAL;
1685 		goto out;
1686 	}
1687 
1688 	if (msg[3] != RDECODE_MSG_CREATE)
1689 		goto out;
1690 
1691 	num_buffers = msg[2];
1692 	for (i = 0, msg = &msg[6]; i < num_buffers; ++i, msg += 4) {
1693 		uint32_t offset, size, *create;
1694 
1695 		if (msg[0] != RDECODE_MESSAGE_CREATE)
1696 			continue;
1697 
1698 		offset = msg[1];
1699 		size = msg[2];
1700 
1701 		if (offset + size > end) {
1702 			r = -EINVAL;
1703 			goto out;
1704 		}
1705 
1706 		create = ptr + addr + offset - start;
1707 
1708 		/* H246, HEVC and VP9 can run on any instance */
1709 		if (create[0] == 0x7 || create[0] == 0x10 || create[0] == 0x11)
1710 			continue;
1711 
1712 		r = vcn_v4_0_limit_sched(p, job);
1713 		if (r)
1714 			goto out;
1715 	}
1716 
1717 out:
1718 	amdgpu_bo_kunmap(bo);
1719 	return r;
1720 }
1721 
1722 #define RADEON_VCN_ENGINE_TYPE_DECODE                                 (0x00000003)
1723 
1724 static int vcn_v4_0_ring_patch_cs_in_place(struct amdgpu_cs_parser *p,
1725 					   struct amdgpu_job *job,
1726 					   struct amdgpu_ib *ib)
1727 {
1728 	struct amdgpu_ring *ring = amdgpu_job_ring(job);
1729 	struct amdgpu_vcn_decode_buffer *decode_buffer;
1730 	uint64_t addr;
1731 	uint32_t val;
1732 
1733 	/* The first instance can decode anything */
1734 	if (!ring->me)
1735 		return 0;
1736 
1737 	/* unified queue ib header has 8 double words. */
1738 	if (ib->length_dw < 8)
1739 		return 0;
1740 
1741 	val = amdgpu_ib_get_value(ib, 6); //RADEON_VCN_ENGINE_TYPE
1742 	if (val != RADEON_VCN_ENGINE_TYPE_DECODE)
1743 		return 0;
1744 
1745 	decode_buffer = (struct amdgpu_vcn_decode_buffer *)&ib->ptr[10];
1746 
1747 	if (!(decode_buffer->valid_buf_flag  & 0x1))
1748 		return 0;
1749 
1750 	addr = ((u64)decode_buffer->msg_buffer_address_hi) << 32 |
1751 		decode_buffer->msg_buffer_address_lo;
1752 	return vcn_v4_0_dec_msg(p, job, addr);
1753 }
1754 
1755 static const struct amdgpu_ring_funcs vcn_v4_0_unified_ring_vm_funcs = {
1756 	.type = AMDGPU_RING_TYPE_VCN_ENC,
1757 	.align_mask = 0x3f,
1758 	.nop = VCN_ENC_CMD_NO_OP,
1759 	.vmhub = AMDGPU_MMHUB_0,
1760 	.get_rptr = vcn_v4_0_unified_ring_get_rptr,
1761 	.get_wptr = vcn_v4_0_unified_ring_get_wptr,
1762 	.set_wptr = vcn_v4_0_unified_ring_set_wptr,
1763 	.patch_cs_in_place = vcn_v4_0_ring_patch_cs_in_place,
1764 	.emit_frame_size =
1765 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1766 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1767 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1768 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1769 		1, /* vcn_v2_0_enc_ring_insert_end */
1770 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1771 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
1772 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
1773 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1774 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
1775 	.test_ib = amdgpu_vcn_unified_ring_test_ib,
1776 	.insert_nop = amdgpu_ring_insert_nop,
1777 	.insert_end = vcn_v2_0_enc_ring_insert_end,
1778 	.pad_ib = amdgpu_ring_generic_pad_ib,
1779 	.begin_use = amdgpu_vcn_ring_begin_use,
1780 	.end_use = amdgpu_vcn_ring_end_use,
1781 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1782 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1783 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1784 };
1785 
1786 /**
1787  * vcn_v4_0_set_unified_ring_funcs - set unified ring functions
1788  *
1789  * @adev: amdgpu_device pointer
1790  *
1791  * Set unified ring functions
1792  */
1793 static void vcn_v4_0_set_unified_ring_funcs(struct amdgpu_device *adev)
1794 {
1795 	int i;
1796 
1797 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1798 		if (adev->vcn.harvest_config & (1 << i))
1799 			continue;
1800 
1801 		adev->vcn.inst[i].ring_enc[0].funcs = &vcn_v4_0_unified_ring_vm_funcs;
1802 		adev->vcn.inst[i].ring_enc[0].me = i;
1803 
1804 		DRM_INFO("VCN(%d) encode/decode are enabled in VM mode\n", i);
1805 	}
1806 }
1807 
1808 /**
1809  * vcn_v4_0_is_idle - check VCN block is idle
1810  *
1811  * @handle: amdgpu_device pointer
1812  *
1813  * Check whether VCN block is idle
1814  */
1815 static bool vcn_v4_0_is_idle(void *handle)
1816 {
1817 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1818 	int i, ret = 1;
1819 
1820 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1821 		if (adev->vcn.harvest_config & (1 << i))
1822 			continue;
1823 
1824 		ret &= (RREG32_SOC15(VCN, i, regUVD_STATUS) == UVD_STATUS__IDLE);
1825 	}
1826 
1827 	return ret;
1828 }
1829 
1830 /**
1831  * vcn_v4_0_wait_for_idle - wait for VCN block idle
1832  *
1833  * @handle: amdgpu_device pointer
1834  *
1835  * Wait for VCN block idle
1836  */
1837 static int vcn_v4_0_wait_for_idle(void *handle)
1838 {
1839 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1840 	int i, ret = 0;
1841 
1842 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1843 		if (adev->vcn.harvest_config & (1 << i))
1844 			continue;
1845 
1846 		ret = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE,
1847 			UVD_STATUS__IDLE);
1848 		if (ret)
1849 			return ret;
1850 	}
1851 
1852 	return ret;
1853 }
1854 
1855 /**
1856  * vcn_v4_0_set_clockgating_state - set VCN block clockgating state
1857  *
1858  * @handle: amdgpu_device pointer
1859  * @state: clock gating state
1860  *
1861  * Set VCN block clockgating state
1862  */
1863 static int vcn_v4_0_set_clockgating_state(void *handle, enum amd_clockgating_state state)
1864 {
1865 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1866 	bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
1867 	int i;
1868 
1869 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1870 		if (adev->vcn.harvest_config & (1 << i))
1871 			continue;
1872 
1873 		if (enable) {
1874 			if (RREG32_SOC15(VCN, i, regUVD_STATUS) != UVD_STATUS__IDLE)
1875 				return -EBUSY;
1876 			vcn_v4_0_enable_clock_gating(adev, i);
1877 		} else {
1878 			vcn_v4_0_disable_clock_gating(adev, i);
1879 		}
1880 	}
1881 
1882 	return 0;
1883 }
1884 
1885 /**
1886  * vcn_v4_0_set_powergating_state - set VCN block powergating state
1887  *
1888  * @handle: amdgpu_device pointer
1889  * @state: power gating state
1890  *
1891  * Set VCN block powergating state
1892  */
1893 static int vcn_v4_0_set_powergating_state(void *handle, enum amd_powergating_state state)
1894 {
1895 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1896 	int ret;
1897 
1898 	/* for SRIOV, guest should not control VCN Power-gating
1899 	 * MMSCH FW should control Power-gating and clock-gating
1900 	 * guest should avoid touching CGC and PG
1901 	 */
1902 	if (amdgpu_sriov_vf(adev)) {
1903 		adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
1904 		return 0;
1905 	}
1906 
1907 	if(state == adev->vcn.cur_state)
1908 		return 0;
1909 
1910 	if (state == AMD_PG_STATE_GATE)
1911 		ret = vcn_v4_0_stop(adev);
1912 	else
1913 		ret = vcn_v4_0_start(adev);
1914 
1915 	if(!ret)
1916 		adev->vcn.cur_state = state;
1917 
1918 	return ret;
1919 }
1920 
1921 /**
1922  * vcn_v4_0_set_interrupt_state - set VCN block interrupt state
1923  *
1924  * @adev: amdgpu_device pointer
1925  * @source: interrupt sources
1926  * @type: interrupt types
1927  * @state: interrupt states
1928  *
1929  * Set VCN block interrupt state
1930  */
1931 static int vcn_v4_0_set_interrupt_state(struct amdgpu_device *adev, struct amdgpu_irq_src *source,
1932       unsigned type, enum amdgpu_interrupt_state state)
1933 {
1934 	return 0;
1935 }
1936 
1937 /**
1938  * vcn_v4_0_process_interrupt - process VCN block interrupt
1939  *
1940  * @adev: amdgpu_device pointer
1941  * @source: interrupt sources
1942  * @entry: interrupt entry from clients and sources
1943  *
1944  * Process VCN block interrupt
1945  */
1946 static int vcn_v4_0_process_interrupt(struct amdgpu_device *adev, struct amdgpu_irq_src *source,
1947       struct amdgpu_iv_entry *entry)
1948 {
1949 	uint32_t ip_instance;
1950 
1951 	switch (entry->client_id) {
1952 	case SOC15_IH_CLIENTID_VCN:
1953 		ip_instance = 0;
1954 		break;
1955 	case SOC15_IH_CLIENTID_VCN1:
1956 		ip_instance = 1;
1957 		break;
1958 	default:
1959 		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
1960 		return 0;
1961 	}
1962 
1963 	DRM_DEBUG("IH: VCN TRAP\n");
1964 
1965 	switch (entry->src_id) {
1966 	case VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
1967 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
1968 		break;
1969 	case VCN_4_0__SRCID_UVD_POISON:
1970 		amdgpu_vcn_process_poison_irq(adev, source, entry);
1971 		break;
1972 	default:
1973 		DRM_ERROR("Unhandled interrupt: %d %d\n",
1974 			  entry->src_id, entry->src_data[0]);
1975 		break;
1976 	}
1977 
1978 	return 0;
1979 }
1980 
1981 static const struct amdgpu_irq_src_funcs vcn_v4_0_irq_funcs = {
1982 	.set = vcn_v4_0_set_interrupt_state,
1983 	.process = vcn_v4_0_process_interrupt,
1984 };
1985 
1986 /**
1987  * vcn_v4_0_set_irq_funcs - set VCN block interrupt irq functions
1988  *
1989  * @adev: amdgpu_device pointer
1990  *
1991  * Set VCN block interrupt irq functions
1992  */
1993 static void vcn_v4_0_set_irq_funcs(struct amdgpu_device *adev)
1994 {
1995 	int i;
1996 
1997 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1998 		if (adev->vcn.harvest_config & (1 << i))
1999 			continue;
2000 
2001 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
2002 		adev->vcn.inst[i].irq.funcs = &vcn_v4_0_irq_funcs;
2003 	}
2004 }
2005 
2006 static const struct amd_ip_funcs vcn_v4_0_ip_funcs = {
2007 	.name = "vcn_v4_0",
2008 	.early_init = vcn_v4_0_early_init,
2009 	.late_init = NULL,
2010 	.sw_init = vcn_v4_0_sw_init,
2011 	.sw_fini = vcn_v4_0_sw_fini,
2012 	.hw_init = vcn_v4_0_hw_init,
2013 	.hw_fini = vcn_v4_0_hw_fini,
2014 	.suspend = vcn_v4_0_suspend,
2015 	.resume = vcn_v4_0_resume,
2016 	.is_idle = vcn_v4_0_is_idle,
2017 	.wait_for_idle = vcn_v4_0_wait_for_idle,
2018 	.check_soft_reset = NULL,
2019 	.pre_soft_reset = NULL,
2020 	.soft_reset = NULL,
2021 	.post_soft_reset = NULL,
2022 	.set_clockgating_state = vcn_v4_0_set_clockgating_state,
2023 	.set_powergating_state = vcn_v4_0_set_powergating_state,
2024 };
2025 
2026 const struct amdgpu_ip_block_version vcn_v4_0_ip_block =
2027 {
2028 	.type = AMD_IP_BLOCK_TYPE_VCN,
2029 	.major = 4,
2030 	.minor = 0,
2031 	.rev = 0,
2032 	.funcs = &vcn_v4_0_ip_funcs,
2033 };
2034 
2035 static uint32_t vcn_v4_0_query_poison_by_instance(struct amdgpu_device *adev,
2036 			uint32_t instance, uint32_t sub_block)
2037 {
2038 	uint32_t poison_stat = 0, reg_value = 0;
2039 
2040 	switch (sub_block) {
2041 	case AMDGPU_VCN_V4_0_VCPU_VCODEC:
2042 		reg_value = RREG32_SOC15(VCN, instance, regUVD_RAS_VCPU_VCODEC_STATUS);
2043 		poison_stat = REG_GET_FIELD(reg_value, UVD_RAS_VCPU_VCODEC_STATUS, POISONED_PF);
2044 		break;
2045 	default:
2046 		break;
2047 	}
2048 
2049 	if (poison_stat)
2050 		dev_info(adev->dev, "Poison detected in VCN%d, sub_block%d\n",
2051 			instance, sub_block);
2052 
2053 	return poison_stat;
2054 }
2055 
2056 static bool vcn_v4_0_query_ras_poison_status(struct amdgpu_device *adev)
2057 {
2058 	uint32_t inst, sub;
2059 	uint32_t poison_stat = 0;
2060 
2061 	for (inst = 0; inst < adev->vcn.num_vcn_inst; inst++)
2062 		for (sub = 0; sub < AMDGPU_VCN_V4_0_MAX_SUB_BLOCK; sub++)
2063 			poison_stat +=
2064 				vcn_v4_0_query_poison_by_instance(adev, inst, sub);
2065 
2066 	return !!poison_stat;
2067 }
2068 
2069 const struct amdgpu_ras_block_hw_ops vcn_v4_0_ras_hw_ops = {
2070 	.query_poison_status = vcn_v4_0_query_ras_poison_status,
2071 };
2072 
2073 static struct amdgpu_vcn_ras vcn_v4_0_ras = {
2074 	.ras_block = {
2075 		.hw_ops = &vcn_v4_0_ras_hw_ops,
2076 	},
2077 };
2078 
2079 static void vcn_v4_0_set_ras_funcs(struct amdgpu_device *adev)
2080 {
2081 	switch (adev->ip_versions[VCN_HWIP][0]) {
2082 	case IP_VERSION(4, 0, 0):
2083 		adev->vcn.ras = &vcn_v4_0_ras;
2084 		break;
2085 	default:
2086 		break;
2087 	}
2088 
2089 	amdgpu_vcn_set_ras_funcs(adev);
2090 }
2091