xref: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c (revision 3b669df92c8531040063c78e03e0509d3a5f40b8)
1 /*
2  * Copyright 2016-2024 Advanced Micro Devices, Inc.
3  * All Rights Reserved.
4  *
5  * Permission is hereby granted, free of charge, to any person obtaining a
6  * copy of this software and associated documentation files (the
7  * "Software"), to deal in the Software without restriction, including
8  * without limitation the rights to use, copy, modify, merge, publish,
9  * distribute, sub license, and/or sell copies of the Software, and to
10  * permit persons to whom the Software is furnished to do so, subject to
11  * the following conditions:
12  *
13  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
14  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
16  * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
17  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
18  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
19  * USE OR OTHER DEALINGS IN THE SOFTWARE.
20  *
21  * The above copyright notice and this permission notice (including the
22  * next paragraph) shall be included in all copies or substantial portions
23  * of the Software.
24  *
25  */
26 
27 #include <linux/firmware.h>
28 #include <linux/module.h>
29 #include <linux/dmi.h>
30 #include <linux/pci.h>
31 #include <linux/debugfs.h>
32 #include <drm/drm_drv.h>
33 
34 #include "amdgpu.h"
35 #include "amdgpu_pm.h"
36 #include "amdgpu_vcn.h"
37 #include "soc15d.h"
38 
39 /* Firmware Names */
40 #define FIRMWARE_RAVEN			"amdgpu/raven_vcn.bin"
41 #define FIRMWARE_PICASSO		"amdgpu/picasso_vcn.bin"
42 #define FIRMWARE_RAVEN2			"amdgpu/raven2_vcn.bin"
43 #define FIRMWARE_ARCTURUS		"amdgpu/arcturus_vcn.bin"
44 #define FIRMWARE_RENOIR			"amdgpu/renoir_vcn.bin"
45 #define FIRMWARE_GREEN_SARDINE		"amdgpu/green_sardine_vcn.bin"
46 #define FIRMWARE_NAVI10			"amdgpu/navi10_vcn.bin"
47 #define FIRMWARE_NAVI14			"amdgpu/navi14_vcn.bin"
48 #define FIRMWARE_NAVI12			"amdgpu/navi12_vcn.bin"
49 #define FIRMWARE_SIENNA_CICHLID		"amdgpu/sienna_cichlid_vcn.bin"
50 #define FIRMWARE_NAVY_FLOUNDER		"amdgpu/navy_flounder_vcn.bin"
51 #define FIRMWARE_VANGOGH		"amdgpu/vangogh_vcn.bin"
52 #define FIRMWARE_DIMGREY_CAVEFISH	"amdgpu/dimgrey_cavefish_vcn.bin"
53 #define FIRMWARE_ALDEBARAN		"amdgpu/aldebaran_vcn.bin"
54 #define FIRMWARE_BEIGE_GOBY		"amdgpu/beige_goby_vcn.bin"
55 #define FIRMWARE_YELLOW_CARP		"amdgpu/yellow_carp_vcn.bin"
56 #define FIRMWARE_VCN_3_1_2		"amdgpu/vcn_3_1_2.bin"
57 #define FIRMWARE_VCN4_0_0		"amdgpu/vcn_4_0_0.bin"
58 #define FIRMWARE_VCN4_0_2		"amdgpu/vcn_4_0_2.bin"
59 #define FIRMWARE_VCN4_0_3		"amdgpu/vcn_4_0_3.bin"
60 #define FIRMWARE_VCN4_0_4		"amdgpu/vcn_4_0_4.bin"
61 #define FIRMWARE_VCN4_0_5		"amdgpu/vcn_4_0_5.bin"
62 #define FIRMWARE_VCN4_0_6		"amdgpu/vcn_4_0_6.bin"
63 #define FIRMWARE_VCN4_0_6_1		"amdgpu/vcn_4_0_6_1.bin"
64 #define FIRMWARE_VCN5_0_0		"amdgpu/vcn_5_0_0.bin"
65 #define FIRMWARE_VCN5_0_1		"amdgpu/vcn_5_0_1.bin"
66 
67 MODULE_FIRMWARE(FIRMWARE_RAVEN);
68 MODULE_FIRMWARE(FIRMWARE_PICASSO);
69 MODULE_FIRMWARE(FIRMWARE_RAVEN2);
70 MODULE_FIRMWARE(FIRMWARE_ARCTURUS);
71 MODULE_FIRMWARE(FIRMWARE_RENOIR);
72 MODULE_FIRMWARE(FIRMWARE_GREEN_SARDINE);
73 MODULE_FIRMWARE(FIRMWARE_ALDEBARAN);
74 MODULE_FIRMWARE(FIRMWARE_NAVI10);
75 MODULE_FIRMWARE(FIRMWARE_NAVI14);
76 MODULE_FIRMWARE(FIRMWARE_NAVI12);
77 MODULE_FIRMWARE(FIRMWARE_SIENNA_CICHLID);
78 MODULE_FIRMWARE(FIRMWARE_NAVY_FLOUNDER);
79 MODULE_FIRMWARE(FIRMWARE_VANGOGH);
80 MODULE_FIRMWARE(FIRMWARE_DIMGREY_CAVEFISH);
81 MODULE_FIRMWARE(FIRMWARE_BEIGE_GOBY);
82 MODULE_FIRMWARE(FIRMWARE_YELLOW_CARP);
83 MODULE_FIRMWARE(FIRMWARE_VCN_3_1_2);
84 MODULE_FIRMWARE(FIRMWARE_VCN4_0_0);
85 MODULE_FIRMWARE(FIRMWARE_VCN4_0_2);
86 MODULE_FIRMWARE(FIRMWARE_VCN4_0_3);
87 MODULE_FIRMWARE(FIRMWARE_VCN4_0_4);
88 MODULE_FIRMWARE(FIRMWARE_VCN4_0_5);
89 MODULE_FIRMWARE(FIRMWARE_VCN4_0_6);
90 MODULE_FIRMWARE(FIRMWARE_VCN4_0_6_1);
91 MODULE_FIRMWARE(FIRMWARE_VCN5_0_0);
92 MODULE_FIRMWARE(FIRMWARE_VCN5_0_1);
93 
94 static void amdgpu_vcn_idle_work_handler(struct work_struct *work);
95 
96 int amdgpu_vcn_early_init(struct amdgpu_device *adev, int i)
97 {
98 	char ucode_prefix[25];
99 	int r;
100 
101 	adev->vcn.inst[i].adev = adev;
102 	adev->vcn.inst[i].inst = i;
103 	amdgpu_ucode_ip_version_decode(adev, UVD_HWIP, ucode_prefix, sizeof(ucode_prefix));
104 
105 	if (i != 0 && adev->vcn.per_inst_fw) {
106 		r = amdgpu_ucode_request(adev, &adev->vcn.inst[i].fw,
107 					 AMDGPU_UCODE_REQUIRED,
108 					 "amdgpu/%s_%d.bin", ucode_prefix, i);
109 		if (r)
110 			amdgpu_ucode_release(&adev->vcn.inst[i].fw);
111 	} else {
112 		if (!adev->vcn.inst[0].fw) {
113 			r = amdgpu_ucode_request(adev, &adev->vcn.inst[0].fw,
114 						 AMDGPU_UCODE_REQUIRED,
115 						 "amdgpu/%s.bin", ucode_prefix);
116 			if (r)
117 				amdgpu_ucode_release(&adev->vcn.inst[0].fw);
118 		} else {
119 			r = 0;
120 		}
121 		adev->vcn.inst[i].fw = adev->vcn.inst[0].fw;
122 	}
123 
124 	return r;
125 }
126 
127 int amdgpu_vcn_sw_init(struct amdgpu_device *adev, int i)
128 {
129 	unsigned long bo_size;
130 	const struct common_firmware_header *hdr;
131 	unsigned char fw_check;
132 	unsigned int fw_shared_size, log_offset;
133 	int r;
134 
135 	mutex_init(&adev->vcn.inst[i].vcn1_jpeg1_workaround);
136 	mutex_init(&adev->vcn.inst[i].vcn_pg_lock);
137 	atomic_set(&adev->vcn.inst[i].total_submission_cnt, 0);
138 	INIT_DELAYED_WORK(&adev->vcn.inst[i].idle_work, amdgpu_vcn_idle_work_handler);
139 	atomic_set(&adev->vcn.inst[i].dpg_enc_submission_cnt, 0);
140 	if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
141 	    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
142 		adev->vcn.inst[i].indirect_sram = true;
143 
144 	/*
145 	 * Some Steam Deck's BIOS versions are incompatible with the
146 	 * indirect SRAM mode, leading to amdgpu being unable to get
147 	 * properly probed (and even potentially crashing the kernel).
148 	 * Hence, check for these versions here - notice this is
149 	 * restricted to Vangogh (Deck's APU).
150 	 */
151 	if (amdgpu_ip_version(adev, UVD_HWIP, 0) == IP_VERSION(3, 0, 2)) {
152 		const char *bios_ver = dmi_get_system_info(DMI_BIOS_VERSION);
153 
154 		if (bios_ver && (!strncmp("F7A0113", bios_ver, 7) ||
155 				 !strncmp("F7A0114", bios_ver, 7))) {
156 			adev->vcn.inst[i].indirect_sram = false;
157 			dev_info(adev->dev,
158 				 "Steam Deck quirk: indirect SRAM disabled on BIOS %s\n", bios_ver);
159 		}
160 	}
161 
162 	/* from vcn4 and above, only unified queue is used */
163 	adev->vcn.inst[i].using_unified_queue =
164 		amdgpu_ip_version(adev, UVD_HWIP, 0) >= IP_VERSION(4, 0, 0);
165 
166 	hdr = (const struct common_firmware_header *)adev->vcn.inst[i].fw->data;
167 	adev->vcn.inst[i].fw_version = le32_to_cpu(hdr->ucode_version);
168 	adev->vcn.fw_version = le32_to_cpu(hdr->ucode_version);
169 
170 	/* Bit 20-23, it is encode major and non-zero for new naming convention.
171 	 * This field is part of version minor and DRM_DISABLED_FLAG in old naming
172 	 * convention. Since the l:wq!atest version minor is 0x5B and DRM_DISABLED_FLAG
173 	 * is zero in old naming convention, this field is always zero so far.
174 	 * These four bits are used to tell which naming convention is present.
175 	 */
176 	fw_check = (le32_to_cpu(hdr->ucode_version) >> 20) & 0xf;
177 	if (fw_check) {
178 		unsigned int dec_ver, enc_major, enc_minor, vep, fw_rev;
179 
180 		fw_rev = le32_to_cpu(hdr->ucode_version) & 0xfff;
181 		enc_minor = (le32_to_cpu(hdr->ucode_version) >> 12) & 0xff;
182 		enc_major = fw_check;
183 		dec_ver = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xf;
184 		vep = (le32_to_cpu(hdr->ucode_version) >> 28) & 0xf;
185 		dev_info(adev->dev,
186 			 "Found VCN firmware Version ENC: %u.%u DEC: %u VEP: %u Revision: %u\n",
187 			 enc_major, enc_minor, dec_ver, vep, fw_rev);
188 	} else {
189 		unsigned int version_major, version_minor, family_id;
190 
191 		family_id = le32_to_cpu(hdr->ucode_version) & 0xff;
192 		version_major = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xff;
193 		version_minor = (le32_to_cpu(hdr->ucode_version) >> 8) & 0xff;
194 		dev_info(adev->dev, "Found VCN firmware Version: %u.%u Family ID: %u\n",
195 			 version_major, version_minor, family_id);
196 	}
197 
198 	bo_size = AMDGPU_VCN_STACK_SIZE + AMDGPU_VCN_CONTEXT_SIZE;
199 	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
200 		bo_size += AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
201 
202 	if (amdgpu_ip_version(adev, UVD_HWIP, 0) >= IP_VERSION(5, 0, 0)) {
203 		fw_shared_size = AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn5_fw_shared));
204 		log_offset = offsetof(struct amdgpu_vcn5_fw_shared, fw_log);
205 	} else if (amdgpu_ip_version(adev, UVD_HWIP, 0) >= IP_VERSION(4, 0, 0)) {
206 		fw_shared_size = AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared));
207 		log_offset = offsetof(struct amdgpu_vcn4_fw_shared, fw_log);
208 	} else {
209 		fw_shared_size = AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
210 		log_offset = offsetof(struct amdgpu_fw_shared, fw_log);
211 	}
212 
213 	bo_size += fw_shared_size;
214 
215 	if (amdgpu_vcnfw_log)
216 		bo_size += AMDGPU_VCNFW_LOG_SIZE;
217 
218 	r = amdgpu_bo_create_kernel(adev, bo_size, PAGE_SIZE,
219 				    AMDGPU_GEM_DOMAIN_VRAM |
220 				    AMDGPU_GEM_DOMAIN_GTT,
221 				    &adev->vcn.inst[i].vcpu_bo,
222 				    &adev->vcn.inst[i].gpu_addr,
223 				    &adev->vcn.inst[i].cpu_addr);
224 	if (r) {
225 		dev_err(adev->dev, "(%d) failed to allocate vcn bo\n", r);
226 		return r;
227 	}
228 
229 	adev->vcn.inst[i].fw_shared.cpu_addr = adev->vcn.inst[i].cpu_addr +
230 		bo_size - fw_shared_size;
231 	adev->vcn.inst[i].fw_shared.gpu_addr = adev->vcn.inst[i].gpu_addr +
232 		bo_size - fw_shared_size;
233 
234 	adev->vcn.inst[i].fw_shared.mem_size = fw_shared_size;
235 
236 	if (amdgpu_vcnfw_log) {
237 		adev->vcn.inst[i].fw_shared.cpu_addr -= AMDGPU_VCNFW_LOG_SIZE;
238 		adev->vcn.inst[i].fw_shared.gpu_addr -= AMDGPU_VCNFW_LOG_SIZE;
239 		adev->vcn.inst[i].fw_shared.log_offset = log_offset;
240 	}
241 
242 	if (adev->vcn.inst[i].indirect_sram) {
243 		r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
244 					    AMDGPU_GEM_DOMAIN_VRAM |
245 					    AMDGPU_GEM_DOMAIN_GTT,
246 					    &adev->vcn.inst[i].dpg_sram_bo,
247 					    &adev->vcn.inst[i].dpg_sram_gpu_addr,
248 					    &adev->vcn.inst[i].dpg_sram_cpu_addr);
249 		if (r) {
250 			dev_err(adev->dev, "VCN %d (%d) failed to allocate DPG bo\n", i, r);
251 			return r;
252 		}
253 	}
254 
255 	return 0;
256 }
257 
258 int amdgpu_vcn_sw_fini(struct amdgpu_device *adev, int i)
259 {
260 	int j;
261 
262 	if (adev->vcn.harvest_config & (1 << i))
263 		return 0;
264 
265 	amdgpu_bo_free_kernel(
266 		&adev->vcn.inst[i].dpg_sram_bo,
267 		&adev->vcn.inst[i].dpg_sram_gpu_addr,
268 		(void **)&adev->vcn.inst[i].dpg_sram_cpu_addr);
269 
270 	kvfree(adev->vcn.inst[i].saved_bo);
271 
272 	amdgpu_bo_free_kernel(&adev->vcn.inst[i].vcpu_bo,
273 			      &adev->vcn.inst[i].gpu_addr,
274 			      (void **)&adev->vcn.inst[i].cpu_addr);
275 
276 	amdgpu_ring_fini(&adev->vcn.inst[i].ring_dec);
277 
278 	for (j = 0; j < adev->vcn.inst[i].num_enc_rings; ++j)
279 		amdgpu_ring_fini(&adev->vcn.inst[i].ring_enc[j]);
280 
281 	if (adev->vcn.per_inst_fw) {
282 		amdgpu_ucode_release(&adev->vcn.inst[i].fw);
283 	} else {
284 		amdgpu_ucode_release(&adev->vcn.inst[0].fw);
285 		adev->vcn.inst[i].fw = NULL;
286 	}
287 	mutex_destroy(&adev->vcn.inst[i].vcn_pg_lock);
288 	mutex_destroy(&adev->vcn.inst[i].vcn1_jpeg1_workaround);
289 
290 	return 0;
291 }
292 
293 bool amdgpu_vcn_is_disabled_vcn(struct amdgpu_device *adev, enum vcn_ring_type type, uint32_t vcn_instance)
294 {
295 	bool ret = false;
296 	int vcn_config = adev->vcn.inst[vcn_instance].vcn_config;
297 
298 	if ((type == VCN_ENCODE_RING) && (vcn_config & VCN_BLOCK_ENCODE_DISABLE_MASK))
299 		ret = true;
300 	else if ((type == VCN_DECODE_RING) && (vcn_config & VCN_BLOCK_DECODE_DISABLE_MASK))
301 		ret = true;
302 	else if ((type == VCN_UNIFIED_RING) && (vcn_config & VCN_BLOCK_QUEUE_DISABLE_MASK))
303 		ret = true;
304 
305 	return ret;
306 }
307 
308 static int amdgpu_vcn_save_vcpu_bo_inst(struct amdgpu_device *adev, int i)
309 {
310 	unsigned int size;
311 	void *ptr;
312 	int idx;
313 
314 	if (adev->vcn.harvest_config & (1 << i))
315 		return 0;
316 	if (adev->vcn.inst[i].vcpu_bo == NULL)
317 		return 0;
318 
319 	size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
320 	ptr = adev->vcn.inst[i].cpu_addr;
321 
322 	adev->vcn.inst[i].saved_bo = kvmalloc(size, GFP_KERNEL);
323 	if (!adev->vcn.inst[i].saved_bo)
324 		return -ENOMEM;
325 
326 	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
327 		memcpy_fromio(adev->vcn.inst[i].saved_bo, ptr, size);
328 		drm_dev_exit(idx);
329 	}
330 
331 	return 0;
332 }
333 
334 int amdgpu_vcn_save_vcpu_bo(struct amdgpu_device *adev)
335 {
336 	int ret, i;
337 
338 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
339 		ret = amdgpu_vcn_save_vcpu_bo_inst(adev, i);
340 		if (ret)
341 			return ret;
342 	}
343 
344 	return 0;
345 }
346 
347 int amdgpu_vcn_suspend(struct amdgpu_device *adev, int i)
348 {
349 	bool in_ras_intr = amdgpu_ras_intr_triggered();
350 
351 	if (adev->vcn.harvest_config & (1 << i))
352 		return 0;
353 
354 	cancel_delayed_work_sync(&adev->vcn.inst[i].idle_work);
355 
356 	/* err_event_athub will corrupt VCPU buffer, so we need to
357 	 * restore fw data and clear buffer in amdgpu_vcn_resume() */
358 	if (in_ras_intr)
359 		return 0;
360 
361 	return amdgpu_vcn_save_vcpu_bo_inst(adev, i);
362 }
363 
364 int amdgpu_vcn_resume(struct amdgpu_device *adev, int i)
365 {
366 	unsigned int size;
367 	void *ptr;
368 	int idx;
369 
370 	if (adev->vcn.harvest_config & (1 << i))
371 		return 0;
372 	if (adev->vcn.inst[i].vcpu_bo == NULL)
373 		return -EINVAL;
374 
375 	size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
376 	ptr = adev->vcn.inst[i].cpu_addr;
377 
378 	if (adev->vcn.inst[i].saved_bo != NULL) {
379 		if (drm_dev_enter(adev_to_drm(adev), &idx)) {
380 			memcpy_toio(ptr, adev->vcn.inst[i].saved_bo, size);
381 			drm_dev_exit(idx);
382 		}
383 		kvfree(adev->vcn.inst[i].saved_bo);
384 		adev->vcn.inst[i].saved_bo = NULL;
385 	} else {
386 		const struct common_firmware_header *hdr;
387 		unsigned int offset;
388 
389 		hdr = (const struct common_firmware_header *)adev->vcn.inst[i].fw->data;
390 		if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
391 			offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
392 			if (drm_dev_enter(adev_to_drm(adev), &idx)) {
393 				memcpy_toio(adev->vcn.inst[i].cpu_addr,
394 					    adev->vcn.inst[i].fw->data + offset,
395 					    le32_to_cpu(hdr->ucode_size_bytes));
396 				drm_dev_exit(idx);
397 			}
398 			size -= le32_to_cpu(hdr->ucode_size_bytes);
399 			ptr += le32_to_cpu(hdr->ucode_size_bytes);
400 		}
401 		memset_io(ptr, 0, size);
402 	}
403 
404 	return 0;
405 }
406 
407 static void amdgpu_vcn_idle_work_handler(struct work_struct *work)
408 {
409 	struct amdgpu_vcn_inst *vcn_inst =
410 		container_of(work, struct amdgpu_vcn_inst, idle_work.work);
411 	struct amdgpu_device *adev = vcn_inst->adev;
412 	unsigned int fences = 0, fence[AMDGPU_MAX_VCN_INSTANCES] = {0};
413 	unsigned int i = vcn_inst->inst, j;
414 	int r = 0;
415 
416 	if (adev->vcn.harvest_config & (1 << i))
417 		return;
418 
419 	for (j = 0; j < adev->vcn.inst[i].num_enc_rings; ++j)
420 		fence[i] += amdgpu_fence_count_emitted(&vcn_inst->ring_enc[j]);
421 
422 	/* Only set DPG pause for VCN3 or below, VCN4 and above will be handled by FW */
423 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG &&
424 	    !adev->vcn.inst[i].using_unified_queue) {
425 		struct dpg_pause_state new_state;
426 
427 		if (fence[i] ||
428 		    unlikely(atomic_read(&vcn_inst->dpg_enc_submission_cnt)))
429 			new_state.fw_based = VCN_DPG_STATE__PAUSE;
430 		else
431 			new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
432 
433 		adev->vcn.inst[i].pause_dpg_mode(vcn_inst, &new_state);
434 	}
435 
436 	fence[i] += amdgpu_fence_count_emitted(&vcn_inst->ring_dec);
437 	fences += fence[i];
438 
439 	if (!fences && !atomic_read(&vcn_inst->total_submission_cnt)) {
440 		vcn_inst->set_pg_state(vcn_inst, AMD_PG_STATE_GATE);
441 		mutex_lock(&adev->vcn.workload_profile_mutex);
442 		if (adev->vcn.workload_profile_active) {
443 			r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO,
444 							    false);
445 			if (r)
446 				dev_warn(adev->dev, "(%d) failed to disable video power profile mode\n", r);
447 			adev->vcn.workload_profile_active = false;
448 		}
449 		mutex_unlock(&adev->vcn.workload_profile_mutex);
450 	} else {
451 		schedule_delayed_work(&vcn_inst->idle_work, VCN_IDLE_TIMEOUT);
452 	}
453 }
454 
455 void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring)
456 {
457 	struct amdgpu_device *adev = ring->adev;
458 	struct amdgpu_vcn_inst *vcn_inst = &adev->vcn.inst[ring->me];
459 	int r = 0;
460 
461 	atomic_inc(&vcn_inst->total_submission_cnt);
462 
463 	cancel_delayed_work_sync(&vcn_inst->idle_work);
464 
465 	/* We can safely return early here because we've cancelled the
466 	 * the delayed work so there is no one else to set it to false
467 	 * and we don't care if someone else sets it to true.
468 	 */
469 	if (adev->vcn.workload_profile_active)
470 		goto pg_lock;
471 
472 	mutex_lock(&adev->vcn.workload_profile_mutex);
473 	if (!adev->vcn.workload_profile_active) {
474 		r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO,
475 						    true);
476 		if (r)
477 			dev_warn(adev->dev, "(%d) failed to switch to video power profile mode\n", r);
478 		adev->vcn.workload_profile_active = true;
479 	}
480 	mutex_unlock(&adev->vcn.workload_profile_mutex);
481 
482 pg_lock:
483 	mutex_lock(&vcn_inst->vcn_pg_lock);
484 	vcn_inst->set_pg_state(vcn_inst, AMD_PG_STATE_UNGATE);
485 
486 	/* Only set DPG pause for VCN3 or below, VCN4 and above will be handled by FW */
487 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG &&
488 	    !vcn_inst->using_unified_queue) {
489 		struct dpg_pause_state new_state;
490 
491 		if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC) {
492 			atomic_inc(&vcn_inst->dpg_enc_submission_cnt);
493 			new_state.fw_based = VCN_DPG_STATE__PAUSE;
494 		} else {
495 			unsigned int fences = 0;
496 			unsigned int i;
497 
498 			for (i = 0; i < vcn_inst->num_enc_rings; ++i)
499 				fences += amdgpu_fence_count_emitted(&vcn_inst->ring_enc[i]);
500 
501 			if (fences || atomic_read(&vcn_inst->dpg_enc_submission_cnt))
502 				new_state.fw_based = VCN_DPG_STATE__PAUSE;
503 			else
504 				new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
505 		}
506 
507 		vcn_inst->pause_dpg_mode(vcn_inst, &new_state);
508 	}
509 	mutex_unlock(&vcn_inst->vcn_pg_lock);
510 }
511 
512 void amdgpu_vcn_ring_end_use(struct amdgpu_ring *ring)
513 {
514 	struct amdgpu_device *adev = ring->adev;
515 
516 	/* Only set DPG pause for VCN3 or below, VCN4 and above will be handled by FW */
517 	if (ring->adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG &&
518 	    ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC &&
519 	    !adev->vcn.inst[ring->me].using_unified_queue)
520 		atomic_dec(&ring->adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
521 
522 	atomic_dec(&ring->adev->vcn.inst[ring->me].total_submission_cnt);
523 
524 	schedule_delayed_work(&ring->adev->vcn.inst[ring->me].idle_work,
525 			      VCN_IDLE_TIMEOUT);
526 }
527 
528 int amdgpu_vcn_dec_ring_test_ring(struct amdgpu_ring *ring)
529 {
530 	struct amdgpu_device *adev = ring->adev;
531 	uint32_t tmp = 0;
532 	unsigned int i;
533 	int r;
534 
535 	/* VCN in SRIOV does not support direct register read/write */
536 	if (amdgpu_sriov_vf(adev))
537 		return 0;
538 
539 	WREG32(adev->vcn.inst[ring->me].external.scratch9, 0xCAFEDEAD);
540 	r = amdgpu_ring_alloc(ring, 3);
541 	if (r)
542 		return r;
543 	amdgpu_ring_write(ring, PACKET0(adev->vcn.inst[ring->me].internal.scratch9, 0));
544 	amdgpu_ring_write(ring, 0xDEADBEEF);
545 	amdgpu_ring_commit(ring);
546 	for (i = 0; i < adev->usec_timeout; i++) {
547 		tmp = RREG32(adev->vcn.inst[ring->me].external.scratch9);
548 		if (tmp == 0xDEADBEEF)
549 			break;
550 		udelay(1);
551 	}
552 
553 	if (i >= adev->usec_timeout)
554 		r = -ETIMEDOUT;
555 
556 	return r;
557 }
558 
559 int amdgpu_vcn_dec_sw_ring_test_ring(struct amdgpu_ring *ring)
560 {
561 	struct amdgpu_device *adev = ring->adev;
562 	uint32_t rptr;
563 	unsigned int i;
564 	int r;
565 
566 	if (amdgpu_sriov_vf(adev))
567 		return 0;
568 
569 	r = amdgpu_ring_alloc(ring, 16);
570 	if (r)
571 		return r;
572 
573 	rptr = amdgpu_ring_get_rptr(ring);
574 
575 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_END);
576 	amdgpu_ring_commit(ring);
577 
578 	for (i = 0; i < adev->usec_timeout; i++) {
579 		if (amdgpu_ring_get_rptr(ring) != rptr)
580 			break;
581 		udelay(1);
582 	}
583 
584 	if (i >= adev->usec_timeout)
585 		r = -ETIMEDOUT;
586 
587 	return r;
588 }
589 
590 static int amdgpu_vcn_dec_send_msg(struct amdgpu_ring *ring,
591 				   struct amdgpu_ib *ib_msg,
592 				   struct dma_fence **fence)
593 {
594 	u64 addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
595 	struct amdgpu_device *adev = ring->adev;
596 	struct dma_fence *f = NULL;
597 	struct amdgpu_job *job;
598 	struct amdgpu_ib *ib;
599 	int i, r;
600 
601 	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
602 				     64, AMDGPU_IB_POOL_DIRECT,
603 				     &job);
604 	if (r)
605 		goto err;
606 
607 	ib = &job->ibs[0];
608 	ib->ptr[0] = PACKET0(adev->vcn.inst[ring->me].internal.data0, 0);
609 	ib->ptr[1] = addr;
610 	ib->ptr[2] = PACKET0(adev->vcn.inst[ring->me].internal.data1, 0);
611 	ib->ptr[3] = addr >> 32;
612 	ib->ptr[4] = PACKET0(adev->vcn.inst[ring->me].internal.cmd, 0);
613 	ib->ptr[5] = 0;
614 	for (i = 6; i < 16; i += 2) {
615 		ib->ptr[i] = PACKET0(adev->vcn.inst[ring->me].internal.nop, 0);
616 		ib->ptr[i+1] = 0;
617 	}
618 	ib->length_dw = 16;
619 
620 	r = amdgpu_job_submit_direct(job, ring, &f);
621 	if (r)
622 		goto err_free;
623 
624 	amdgpu_ib_free(ib_msg, f);
625 
626 	if (fence)
627 		*fence = dma_fence_get(f);
628 	dma_fence_put(f);
629 
630 	return 0;
631 
632 err_free:
633 	amdgpu_job_free(job);
634 err:
635 	amdgpu_ib_free(ib_msg, f);
636 	return r;
637 }
638 
639 static int amdgpu_vcn_dec_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
640 		struct amdgpu_ib *ib)
641 {
642 	struct amdgpu_device *adev = ring->adev;
643 	uint32_t *msg;
644 	int r, i;
645 
646 	memset(ib, 0, sizeof(*ib));
647 	r = amdgpu_ib_get(adev, NULL, AMDGPU_GPU_PAGE_SIZE * 2,
648 			AMDGPU_IB_POOL_DIRECT,
649 			ib);
650 	if (r)
651 		return r;
652 
653 	msg = (uint32_t *)AMDGPU_GPU_PAGE_ALIGN((unsigned long)ib->ptr);
654 	msg[0] = cpu_to_le32(0x00000028);
655 	msg[1] = cpu_to_le32(0x00000038);
656 	msg[2] = cpu_to_le32(0x00000001);
657 	msg[3] = cpu_to_le32(0x00000000);
658 	msg[4] = cpu_to_le32(handle);
659 	msg[5] = cpu_to_le32(0x00000000);
660 	msg[6] = cpu_to_le32(0x00000001);
661 	msg[7] = cpu_to_le32(0x00000028);
662 	msg[8] = cpu_to_le32(0x00000010);
663 	msg[9] = cpu_to_le32(0x00000000);
664 	msg[10] = cpu_to_le32(0x00000007);
665 	msg[11] = cpu_to_le32(0x00000000);
666 	msg[12] = cpu_to_le32(0x00000780);
667 	msg[13] = cpu_to_le32(0x00000440);
668 	for (i = 14; i < 1024; ++i)
669 		msg[i] = cpu_to_le32(0x0);
670 
671 	return 0;
672 }
673 
674 static int amdgpu_vcn_dec_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
675 					  struct amdgpu_ib *ib)
676 {
677 	struct amdgpu_device *adev = ring->adev;
678 	uint32_t *msg;
679 	int r, i;
680 
681 	memset(ib, 0, sizeof(*ib));
682 	r = amdgpu_ib_get(adev, NULL, AMDGPU_GPU_PAGE_SIZE * 2,
683 			AMDGPU_IB_POOL_DIRECT,
684 			ib);
685 	if (r)
686 		return r;
687 
688 	msg = (uint32_t *)AMDGPU_GPU_PAGE_ALIGN((unsigned long)ib->ptr);
689 	msg[0] = cpu_to_le32(0x00000028);
690 	msg[1] = cpu_to_le32(0x00000018);
691 	msg[2] = cpu_to_le32(0x00000000);
692 	msg[3] = cpu_to_le32(0x00000002);
693 	msg[4] = cpu_to_le32(handle);
694 	msg[5] = cpu_to_le32(0x00000000);
695 	for (i = 6; i < 1024; ++i)
696 		msg[i] = cpu_to_le32(0x0);
697 
698 	return 0;
699 }
700 
701 int amdgpu_vcn_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout)
702 {
703 	struct dma_fence *fence = NULL;
704 	struct amdgpu_ib ib;
705 	long r;
706 
707 	r = amdgpu_vcn_dec_get_create_msg(ring, 1, &ib);
708 	if (r)
709 		goto error;
710 
711 	r = amdgpu_vcn_dec_send_msg(ring, &ib, NULL);
712 	if (r)
713 		goto error;
714 	r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &ib);
715 	if (r)
716 		goto error;
717 
718 	r = amdgpu_vcn_dec_send_msg(ring, &ib, &fence);
719 	if (r)
720 		goto error;
721 
722 	r = dma_fence_wait_timeout(fence, false, timeout);
723 	if (r == 0)
724 		r = -ETIMEDOUT;
725 	else if (r > 0)
726 		r = 0;
727 
728 	dma_fence_put(fence);
729 error:
730 	return r;
731 }
732 
733 static uint32_t *amdgpu_vcn_unified_ring_ib_header(struct amdgpu_ib *ib,
734 						uint32_t ib_pack_in_dw, bool enc)
735 {
736 	uint32_t *ib_checksum;
737 
738 	ib->ptr[ib->length_dw++] = 0x00000010; /* single queue checksum */
739 	ib->ptr[ib->length_dw++] = 0x30000002;
740 	ib_checksum = &ib->ptr[ib->length_dw++];
741 	ib->ptr[ib->length_dw++] = ib_pack_in_dw;
742 
743 	ib->ptr[ib->length_dw++] = 0x00000010; /* engine info */
744 	ib->ptr[ib->length_dw++] = 0x30000001;
745 	ib->ptr[ib->length_dw++] = enc ? 0x2 : 0x3;
746 	ib->ptr[ib->length_dw++] = ib_pack_in_dw * sizeof(uint32_t);
747 
748 	return ib_checksum;
749 }
750 
751 static void amdgpu_vcn_unified_ring_ib_checksum(uint32_t **ib_checksum,
752 						uint32_t ib_pack_in_dw)
753 {
754 	uint32_t i;
755 	uint32_t checksum = 0;
756 
757 	for (i = 0; i < ib_pack_in_dw; i++)
758 		checksum += *(*ib_checksum + 2 + i);
759 
760 	**ib_checksum = checksum;
761 }
762 
763 static int amdgpu_vcn_dec_sw_send_msg(struct amdgpu_ring *ring,
764 				      struct amdgpu_ib *ib_msg,
765 				      struct dma_fence **fence)
766 {
767 	struct amdgpu_vcn_decode_buffer *decode_buffer = NULL;
768 	unsigned int ib_size_dw = 64;
769 	struct amdgpu_device *adev = ring->adev;
770 	struct dma_fence *f = NULL;
771 	struct amdgpu_job *job;
772 	struct amdgpu_ib *ib;
773 	uint64_t addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
774 	uint32_t *ib_checksum;
775 	uint32_t ib_pack_in_dw;
776 	int i, r;
777 
778 	if (adev->vcn.inst[ring->me].using_unified_queue)
779 		ib_size_dw += 8;
780 
781 	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
782 				     ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
783 				     &job);
784 	if (r)
785 		goto err;
786 
787 	ib = &job->ibs[0];
788 	ib->length_dw = 0;
789 
790 	/* single queue headers */
791 	if (adev->vcn.inst[ring->me].using_unified_queue) {
792 		ib_pack_in_dw = sizeof(struct amdgpu_vcn_decode_buffer) / sizeof(uint32_t)
793 						+ 4 + 2; /* engine info + decoding ib in dw */
794 		ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, ib_pack_in_dw, false);
795 	}
796 
797 	ib->ptr[ib->length_dw++] = sizeof(struct amdgpu_vcn_decode_buffer) + 8;
798 	ib->ptr[ib->length_dw++] = cpu_to_le32(AMDGPU_VCN_IB_FLAG_DECODE_BUFFER);
799 	decode_buffer = (struct amdgpu_vcn_decode_buffer *)&(ib->ptr[ib->length_dw]);
800 	ib->length_dw += sizeof(struct amdgpu_vcn_decode_buffer) / 4;
801 	memset(decode_buffer, 0, sizeof(struct amdgpu_vcn_decode_buffer));
802 
803 	decode_buffer->valid_buf_flag |= cpu_to_le32(AMDGPU_VCN_CMD_FLAG_MSG_BUFFER);
804 	decode_buffer->msg_buffer_address_hi = cpu_to_le32(addr >> 32);
805 	decode_buffer->msg_buffer_address_lo = cpu_to_le32(addr);
806 
807 	for (i = ib->length_dw; i < ib_size_dw; ++i)
808 		ib->ptr[i] = 0x0;
809 
810 	if (adev->vcn.inst[ring->me].using_unified_queue)
811 		amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, ib_pack_in_dw);
812 
813 	r = amdgpu_job_submit_direct(job, ring, &f);
814 	if (r)
815 		goto err_free;
816 
817 	amdgpu_ib_free(ib_msg, f);
818 
819 	if (fence)
820 		*fence = dma_fence_get(f);
821 	dma_fence_put(f);
822 
823 	return 0;
824 
825 err_free:
826 	amdgpu_job_free(job);
827 err:
828 	amdgpu_ib_free(ib_msg, f);
829 	return r;
830 }
831 
832 int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout)
833 {
834 	struct dma_fence *fence = NULL;
835 	struct amdgpu_ib ib;
836 	long r;
837 
838 	r = amdgpu_vcn_dec_get_create_msg(ring, 1, &ib);
839 	if (r)
840 		goto error;
841 
842 	r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, NULL);
843 	if (r)
844 		goto error;
845 	r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &ib);
846 	if (r)
847 		goto error;
848 
849 	r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, &fence);
850 	if (r)
851 		goto error;
852 
853 	r = dma_fence_wait_timeout(fence, false, timeout);
854 	if (r == 0)
855 		r = -ETIMEDOUT;
856 	else if (r > 0)
857 		r = 0;
858 
859 	dma_fence_put(fence);
860 error:
861 	return r;
862 }
863 
864 int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring)
865 {
866 	struct amdgpu_device *adev = ring->adev;
867 	uint32_t rptr;
868 	unsigned int i;
869 	int r;
870 
871 	if (amdgpu_sriov_vf(adev))
872 		return 0;
873 
874 	r = amdgpu_ring_alloc(ring, 16);
875 	if (r)
876 		return r;
877 
878 	rptr = amdgpu_ring_get_rptr(ring);
879 
880 	amdgpu_ring_write(ring, VCN_ENC_CMD_END);
881 	amdgpu_ring_commit(ring);
882 
883 	for (i = 0; i < adev->usec_timeout; i++) {
884 		if (amdgpu_ring_get_rptr(ring) != rptr)
885 			break;
886 		udelay(1);
887 	}
888 
889 	if (i >= adev->usec_timeout)
890 		r = -ETIMEDOUT;
891 
892 	return r;
893 }
894 
895 static int amdgpu_vcn_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
896 					 struct amdgpu_ib *ib_msg,
897 					 struct dma_fence **fence)
898 {
899 	unsigned int ib_size_dw = 16;
900 	struct amdgpu_device *adev = ring->adev;
901 	struct amdgpu_job *job;
902 	struct amdgpu_ib *ib;
903 	struct dma_fence *f = NULL;
904 	uint32_t *ib_checksum = NULL;
905 	uint64_t addr;
906 	int i, r;
907 
908 	if (adev->vcn.inst[ring->me].using_unified_queue)
909 		ib_size_dw += 8;
910 
911 	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
912 				     ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
913 				     &job);
914 	if (r)
915 		return r;
916 
917 	ib = &job->ibs[0];
918 	addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
919 
920 	ib->length_dw = 0;
921 
922 	if (adev->vcn.inst[ring->me].using_unified_queue)
923 		ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, 0x11, true);
924 
925 	ib->ptr[ib->length_dw++] = 0x00000018;
926 	ib->ptr[ib->length_dw++] = 0x00000001; /* session info */
927 	ib->ptr[ib->length_dw++] = handle;
928 	ib->ptr[ib->length_dw++] = upper_32_bits(addr);
929 	ib->ptr[ib->length_dw++] = addr;
930 	ib->ptr[ib->length_dw++] = 0x00000000;
931 
932 	ib->ptr[ib->length_dw++] = 0x00000014;
933 	ib->ptr[ib->length_dw++] = 0x00000002; /* task info */
934 	ib->ptr[ib->length_dw++] = 0x0000001c;
935 	ib->ptr[ib->length_dw++] = 0x00000000;
936 	ib->ptr[ib->length_dw++] = 0x00000000;
937 
938 	ib->ptr[ib->length_dw++] = 0x00000008;
939 	ib->ptr[ib->length_dw++] = 0x08000001; /* op initialize */
940 
941 	for (i = ib->length_dw; i < ib_size_dw; ++i)
942 		ib->ptr[i] = 0x0;
943 
944 	if (adev->vcn.inst[ring->me].using_unified_queue)
945 		amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, 0x11);
946 
947 	r = amdgpu_job_submit_direct(job, ring, &f);
948 	if (r)
949 		goto err;
950 
951 	if (fence)
952 		*fence = dma_fence_get(f);
953 	dma_fence_put(f);
954 
955 	return 0;
956 
957 err:
958 	amdgpu_job_free(job);
959 	return r;
960 }
961 
962 static int amdgpu_vcn_enc_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
963 					  struct amdgpu_ib *ib_msg,
964 					  struct dma_fence **fence)
965 {
966 	unsigned int ib_size_dw = 16;
967 	struct amdgpu_device *adev = ring->adev;
968 	struct amdgpu_job *job;
969 	struct amdgpu_ib *ib;
970 	struct dma_fence *f = NULL;
971 	uint32_t *ib_checksum = NULL;
972 	uint64_t addr;
973 	int i, r;
974 
975 	if (adev->vcn.inst[ring->me].using_unified_queue)
976 		ib_size_dw += 8;
977 
978 	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
979 				     ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
980 				     &job);
981 	if (r)
982 		return r;
983 
984 	ib = &job->ibs[0];
985 	addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
986 
987 	ib->length_dw = 0;
988 
989 	if (adev->vcn.inst[ring->me].using_unified_queue)
990 		ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, 0x11, true);
991 
992 	ib->ptr[ib->length_dw++] = 0x00000018;
993 	ib->ptr[ib->length_dw++] = 0x00000001;
994 	ib->ptr[ib->length_dw++] = handle;
995 	ib->ptr[ib->length_dw++] = upper_32_bits(addr);
996 	ib->ptr[ib->length_dw++] = addr;
997 	ib->ptr[ib->length_dw++] = 0x00000000;
998 
999 	ib->ptr[ib->length_dw++] = 0x00000014;
1000 	ib->ptr[ib->length_dw++] = 0x00000002;
1001 	ib->ptr[ib->length_dw++] = 0x0000001c;
1002 	ib->ptr[ib->length_dw++] = 0x00000000;
1003 	ib->ptr[ib->length_dw++] = 0x00000000;
1004 
1005 	ib->ptr[ib->length_dw++] = 0x00000008;
1006 	ib->ptr[ib->length_dw++] = 0x08000002; /* op close session */
1007 
1008 	for (i = ib->length_dw; i < ib_size_dw; ++i)
1009 		ib->ptr[i] = 0x0;
1010 
1011 	if (adev->vcn.inst[ring->me].using_unified_queue)
1012 		amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, 0x11);
1013 
1014 	r = amdgpu_job_submit_direct(job, ring, &f);
1015 	if (r)
1016 		goto err;
1017 
1018 	if (fence)
1019 		*fence = dma_fence_get(f);
1020 	dma_fence_put(f);
1021 
1022 	return 0;
1023 
1024 err:
1025 	amdgpu_job_free(job);
1026 	return r;
1027 }
1028 
1029 int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout)
1030 {
1031 	struct amdgpu_device *adev = ring->adev;
1032 	struct dma_fence *fence = NULL;
1033 	struct amdgpu_ib ib;
1034 	long r;
1035 
1036 	memset(&ib, 0, sizeof(ib));
1037 	r = amdgpu_ib_get(adev, NULL, (128 << 10) + AMDGPU_GPU_PAGE_SIZE,
1038 			AMDGPU_IB_POOL_DIRECT,
1039 			&ib);
1040 	if (r)
1041 		return r;
1042 
1043 	r = amdgpu_vcn_enc_get_create_msg(ring, 1, &ib, NULL);
1044 	if (r)
1045 		goto error;
1046 
1047 	r = amdgpu_vcn_enc_get_destroy_msg(ring, 1, &ib, &fence);
1048 	if (r)
1049 		goto error;
1050 
1051 	r = dma_fence_wait_timeout(fence, false, timeout);
1052 	if (r == 0)
1053 		r = -ETIMEDOUT;
1054 	else if (r > 0)
1055 		r = 0;
1056 
1057 error:
1058 	amdgpu_ib_free(&ib, fence);
1059 	dma_fence_put(fence);
1060 
1061 	return r;
1062 }
1063 
1064 int amdgpu_vcn_unified_ring_test_ib(struct amdgpu_ring *ring, long timeout)
1065 {
1066 	struct amdgpu_device *adev = ring->adev;
1067 	long r;
1068 
1069 	if ((amdgpu_ip_version(adev, UVD_HWIP, 0) != IP_VERSION(4, 0, 3)) &&
1070 	    (amdgpu_ip_version(adev, UVD_HWIP, 0) != IP_VERSION(5, 0, 1))) {
1071 		r = amdgpu_vcn_enc_ring_test_ib(ring, timeout);
1072 		if (r)
1073 			goto error;
1074 	}
1075 
1076 	r =  amdgpu_vcn_dec_sw_ring_test_ib(ring, timeout);
1077 
1078 error:
1079 	return r;
1080 }
1081 
1082 enum amdgpu_ring_priority_level amdgpu_vcn_get_enc_ring_prio(int ring)
1083 {
1084 	switch (ring) {
1085 	case 0:
1086 		return AMDGPU_RING_PRIO_0;
1087 	case 1:
1088 		return AMDGPU_RING_PRIO_1;
1089 	case 2:
1090 		return AMDGPU_RING_PRIO_2;
1091 	default:
1092 		return AMDGPU_RING_PRIO_0;
1093 	}
1094 }
1095 
1096 void amdgpu_vcn_setup_ucode(struct amdgpu_device *adev, int i)
1097 {
1098 	unsigned int idx;
1099 
1100 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1101 		const struct common_firmware_header *hdr;
1102 
1103 		if (adev->vcn.harvest_config & (1 << i))
1104 			return;
1105 
1106 		if ((amdgpu_ip_version(adev, UVD_HWIP, 0) == IP_VERSION(4, 0, 3) ||
1107 		     amdgpu_ip_version(adev, UVD_HWIP, 0) == IP_VERSION(5, 0, 1))
1108 		    && (i > 0))
1109 			return;
1110 
1111 		hdr = (const struct common_firmware_header *)adev->vcn.inst[i].fw->data;
1112 		/* currently only support 2 FW instances */
1113 		if (i >= 2) {
1114 			dev_info(adev->dev, "More then 2 VCN FW instances!\n");
1115 			return;
1116 		}
1117 		idx = AMDGPU_UCODE_ID_VCN + i;
1118 		adev->firmware.ucode[idx].ucode_id = idx;
1119 		adev->firmware.ucode[idx].fw = adev->vcn.inst[i].fw;
1120 		adev->firmware.fw_size +=
1121 			ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
1122 	}
1123 }
1124 
1125 /*
1126  * debugfs for mapping vcn firmware log buffer.
1127  */
1128 #if defined(CONFIG_DEBUG_FS)
1129 static ssize_t amdgpu_debugfs_vcn_fwlog_read(struct file *f, char __user *buf,
1130 					     size_t size, loff_t *pos)
1131 {
1132 	struct amdgpu_vcn_inst *vcn;
1133 	void *log_buf;
1134 	volatile struct amdgpu_vcn_fwlog *plog;
1135 	unsigned int read_pos, write_pos, available, i, read_bytes = 0;
1136 	unsigned int read_num[2] = {0};
1137 
1138 	vcn = file_inode(f)->i_private;
1139 	if (!vcn)
1140 		return -ENODEV;
1141 
1142 	if (!vcn->fw_shared.cpu_addr || !amdgpu_vcnfw_log)
1143 		return -EFAULT;
1144 
1145 	log_buf = vcn->fw_shared.cpu_addr + vcn->fw_shared.mem_size;
1146 
1147 	plog = (volatile struct amdgpu_vcn_fwlog *)log_buf;
1148 	read_pos = plog->rptr;
1149 	write_pos = plog->wptr;
1150 
1151 	if (read_pos > AMDGPU_VCNFW_LOG_SIZE || write_pos > AMDGPU_VCNFW_LOG_SIZE)
1152 		return -EFAULT;
1153 
1154 	if (!size || (read_pos == write_pos))
1155 		return 0;
1156 
1157 	if (write_pos > read_pos) {
1158 		available = write_pos - read_pos;
1159 		read_num[0] = min_t(size_t, size, available);
1160 	} else {
1161 		read_num[0] = AMDGPU_VCNFW_LOG_SIZE - read_pos;
1162 		available = read_num[0] + write_pos - plog->header_size;
1163 		if (size > available)
1164 			read_num[1] = write_pos - plog->header_size;
1165 		else if (size > read_num[0])
1166 			read_num[1] = size - read_num[0];
1167 		else
1168 			read_num[0] = size;
1169 	}
1170 
1171 	for (i = 0; i < 2; i++) {
1172 		if (read_num[i]) {
1173 			if (read_pos == AMDGPU_VCNFW_LOG_SIZE)
1174 				read_pos = plog->header_size;
1175 			if (read_num[i] == copy_to_user((buf + read_bytes),
1176 							(log_buf + read_pos), read_num[i]))
1177 				return -EFAULT;
1178 
1179 			read_bytes += read_num[i];
1180 			read_pos += read_num[i];
1181 		}
1182 	}
1183 
1184 	plog->rptr = read_pos;
1185 	*pos += read_bytes;
1186 	return read_bytes;
1187 }
1188 
1189 static const struct file_operations amdgpu_debugfs_vcnfwlog_fops = {
1190 	.owner = THIS_MODULE,
1191 	.read = amdgpu_debugfs_vcn_fwlog_read,
1192 	.llseek = default_llseek
1193 };
1194 #endif
1195 
1196 void amdgpu_debugfs_vcn_fwlog_init(struct amdgpu_device *adev, uint8_t i,
1197 				   struct amdgpu_vcn_inst *vcn)
1198 {
1199 #if defined(CONFIG_DEBUG_FS)
1200 	struct drm_minor *minor = adev_to_drm(adev)->primary;
1201 	struct dentry *root = minor->debugfs_root;
1202 	char name[32];
1203 
1204 	sprintf(name, "amdgpu_vcn_%d_fwlog", i);
1205 	debugfs_create_file_size(name, S_IFREG | 0444, root, vcn,
1206 				 &amdgpu_debugfs_vcnfwlog_fops,
1207 				 AMDGPU_VCNFW_LOG_SIZE);
1208 #endif
1209 }
1210 
1211 void amdgpu_vcn_fwlog_init(struct amdgpu_vcn_inst *vcn)
1212 {
1213 #if defined(CONFIG_DEBUG_FS)
1214 	volatile uint32_t *flag = vcn->fw_shared.cpu_addr;
1215 	void *fw_log_cpu_addr = vcn->fw_shared.cpu_addr + vcn->fw_shared.mem_size;
1216 	uint64_t fw_log_gpu_addr = vcn->fw_shared.gpu_addr + vcn->fw_shared.mem_size;
1217 	volatile struct amdgpu_vcn_fwlog *log_buf = fw_log_cpu_addr;
1218 	volatile struct amdgpu_fw_shared_fw_logging *fw_log = vcn->fw_shared.cpu_addr
1219 							 + vcn->fw_shared.log_offset;
1220 	*flag |= cpu_to_le32(AMDGPU_VCN_FW_LOGGING_FLAG);
1221 	fw_log->is_enabled = 1;
1222 	fw_log->addr_lo = cpu_to_le32(fw_log_gpu_addr & 0xFFFFFFFF);
1223 	fw_log->addr_hi = cpu_to_le32(fw_log_gpu_addr >> 32);
1224 	fw_log->size = cpu_to_le32(AMDGPU_VCNFW_LOG_SIZE);
1225 
1226 	log_buf->header_size = sizeof(struct amdgpu_vcn_fwlog);
1227 	log_buf->buffer_size = AMDGPU_VCNFW_LOG_SIZE;
1228 	log_buf->rptr = log_buf->header_size;
1229 	log_buf->wptr = log_buf->header_size;
1230 	log_buf->wrapped = 0;
1231 #endif
1232 }
1233 
1234 int amdgpu_vcn_process_poison_irq(struct amdgpu_device *adev,
1235 				struct amdgpu_irq_src *source,
1236 				struct amdgpu_iv_entry *entry)
1237 {
1238 	struct ras_common_if *ras_if = adev->vcn.ras_if;
1239 	struct ras_dispatch_if ih_data = {
1240 		.entry = entry,
1241 	};
1242 
1243 	if (!ras_if)
1244 		return 0;
1245 
1246 	if (!amdgpu_sriov_vf(adev)) {
1247 		ih_data.head = *ras_if;
1248 		amdgpu_ras_interrupt_dispatch(adev, &ih_data);
1249 	} else {
1250 		if (adev->virt.ops && adev->virt.ops->ras_poison_handler)
1251 			adev->virt.ops->ras_poison_handler(adev, ras_if->block);
1252 		else
1253 			dev_warn(adev->dev,
1254 				"No ras_poison_handler interface in SRIOV for VCN!\n");
1255 	}
1256 
1257 	return 0;
1258 }
1259 
1260 int amdgpu_vcn_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block)
1261 {
1262 	int r, i;
1263 
1264 	r = amdgpu_ras_block_late_init(adev, ras_block);
1265 	if (r)
1266 		return r;
1267 
1268 	if (amdgpu_ras_is_supported(adev, ras_block->block)) {
1269 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1270 			if (adev->vcn.harvest_config & (1 << i) ||
1271 			    !adev->vcn.inst[i].ras_poison_irq.funcs)
1272 				continue;
1273 
1274 			r = amdgpu_irq_get(adev, &adev->vcn.inst[i].ras_poison_irq, 0);
1275 			if (r)
1276 				goto late_fini;
1277 		}
1278 	}
1279 	return 0;
1280 
1281 late_fini:
1282 	amdgpu_ras_block_late_fini(adev, ras_block);
1283 	return r;
1284 }
1285 
1286 int amdgpu_vcn_ras_sw_init(struct amdgpu_device *adev)
1287 {
1288 	int err;
1289 	struct amdgpu_vcn_ras *ras;
1290 
1291 	if (!adev->vcn.ras)
1292 		return 0;
1293 
1294 	ras = adev->vcn.ras;
1295 	err = amdgpu_ras_register_ras_block(adev, &ras->ras_block);
1296 	if (err) {
1297 		dev_err(adev->dev, "Failed to register vcn ras block!\n");
1298 		return err;
1299 	}
1300 
1301 	strcpy(ras->ras_block.ras_comm.name, "vcn");
1302 	ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__VCN;
1303 	ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__POISON;
1304 	adev->vcn.ras_if = &ras->ras_block.ras_comm;
1305 
1306 	if (!ras->ras_block.ras_late_init)
1307 		ras->ras_block.ras_late_init = amdgpu_vcn_ras_late_init;
1308 
1309 	return 0;
1310 }
1311 
1312 int amdgpu_vcn_psp_update_sram(struct amdgpu_device *adev, int inst_idx,
1313 			       enum AMDGPU_UCODE_ID ucode_id)
1314 {
1315 	struct amdgpu_firmware_info ucode = {
1316 		.ucode_id = (ucode_id ? ucode_id :
1317 			    (inst_idx ? AMDGPU_UCODE_ID_VCN1_RAM :
1318 					AMDGPU_UCODE_ID_VCN0_RAM)),
1319 		.mc_addr = adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
1320 		.ucode_size = ((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
1321 			      (uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr),
1322 	};
1323 
1324 	return psp_execute_ip_fw_load(&adev->psp, &ucode);
1325 }
1326 
1327 static ssize_t amdgpu_get_vcn_reset_mask(struct device *dev,
1328 						struct device_attribute *attr,
1329 						char *buf)
1330 {
1331 	struct drm_device *ddev = dev_get_drvdata(dev);
1332 	struct amdgpu_device *adev = drm_to_adev(ddev);
1333 
1334 	if (!adev)
1335 		return -ENODEV;
1336 
1337 	return amdgpu_show_reset_mask(buf, adev->vcn.supported_reset);
1338 }
1339 
1340 static DEVICE_ATTR(vcn_reset_mask, 0444,
1341 		   amdgpu_get_vcn_reset_mask, NULL);
1342 
1343 int amdgpu_vcn_sysfs_reset_mask_init(struct amdgpu_device *adev)
1344 {
1345 	int r = 0;
1346 
1347 	if (adev->vcn.num_vcn_inst) {
1348 		r = device_create_file(adev->dev, &dev_attr_vcn_reset_mask);
1349 		if (r)
1350 			return r;
1351 	}
1352 
1353 	return r;
1354 }
1355 
1356 void amdgpu_vcn_sysfs_reset_mask_fini(struct amdgpu_device *adev)
1357 {
1358 	if (adev->dev->kobj.sd) {
1359 		if (adev->vcn.num_vcn_inst)
1360 			device_remove_file(adev->dev, &dev_attr_vcn_reset_mask);
1361 	}
1362 }
1363 
1364 /*
1365  * debugfs to enable/disable vcn job submission to specific core or
1366  * instance. It is created only if the queue type is unified.
1367  */
1368 #if defined(CONFIG_DEBUG_FS)
1369 static int amdgpu_debugfs_vcn_sched_mask_set(void *data, u64 val)
1370 {
1371 	struct amdgpu_device *adev = (struct amdgpu_device *)data;
1372 	u32 i;
1373 	u64 mask;
1374 	struct amdgpu_ring *ring;
1375 
1376 	if (!adev)
1377 		return -ENODEV;
1378 
1379 	mask = (1ULL << adev->vcn.num_vcn_inst) - 1;
1380 	if ((val & mask) == 0)
1381 		return -EINVAL;
1382 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1383 		ring = &adev->vcn.inst[i].ring_enc[0];
1384 		if (val & (1ULL << i))
1385 			ring->sched.ready = true;
1386 		else
1387 			ring->sched.ready = false;
1388 	}
1389 	/* publish sched.ready flag update effective immediately across smp */
1390 	smp_rmb();
1391 	return 0;
1392 }
1393 
1394 static int amdgpu_debugfs_vcn_sched_mask_get(void *data, u64 *val)
1395 {
1396 	struct amdgpu_device *adev = (struct amdgpu_device *)data;
1397 	u32 i;
1398 	u64 mask = 0;
1399 	struct amdgpu_ring *ring;
1400 
1401 	if (!adev)
1402 		return -ENODEV;
1403 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1404 		ring = &adev->vcn.inst[i].ring_enc[0];
1405 		if (ring->sched.ready)
1406 			mask |= 1ULL << i;
1407 		}
1408 	*val = mask;
1409 	return 0;
1410 }
1411 
1412 DEFINE_DEBUGFS_ATTRIBUTE(amdgpu_debugfs_vcn_sched_mask_fops,
1413 			 amdgpu_debugfs_vcn_sched_mask_get,
1414 			 amdgpu_debugfs_vcn_sched_mask_set, "%llx\n");
1415 #endif
1416 
1417 void amdgpu_debugfs_vcn_sched_mask_init(struct amdgpu_device *adev)
1418 {
1419 #if defined(CONFIG_DEBUG_FS)
1420 	struct drm_minor *minor = adev_to_drm(adev)->primary;
1421 	struct dentry *root = minor->debugfs_root;
1422 	char name[32];
1423 
1424 	if (adev->vcn.num_vcn_inst <= 1 || !adev->vcn.inst[0].using_unified_queue)
1425 		return;
1426 	sprintf(name, "amdgpu_vcn_sched_mask");
1427 	debugfs_create_file(name, 0600, root, adev,
1428 			    &amdgpu_debugfs_vcn_sched_mask_fops);
1429 #endif
1430 }
1431 
1432 /**
1433  * vcn_set_powergating_state - set VCN block powergating state
1434  *
1435  * @ip_block: amdgpu_ip_block pointer
1436  * @state: power gating state
1437  *
1438  * Set VCN block powergating state
1439  */
1440 int vcn_set_powergating_state(struct amdgpu_ip_block *ip_block,
1441 			      enum amd_powergating_state state)
1442 {
1443 	struct amdgpu_device *adev = ip_block->adev;
1444 	int ret = 0, i;
1445 
1446 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1447 		struct amdgpu_vcn_inst *vinst = &adev->vcn.inst[i];
1448 
1449 		ret |= vinst->set_pg_state(vinst, state);
1450 	}
1451 
1452 	return ret;
1453 }
1454