1 /*
2  * Copyright 2016 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 
62 MODULE_FIRMWARE(FIRMWARE_RAVEN);
63 MODULE_FIRMWARE(FIRMWARE_PICASSO);
64 MODULE_FIRMWARE(FIRMWARE_RAVEN2);
65 MODULE_FIRMWARE(FIRMWARE_ARCTURUS);
66 MODULE_FIRMWARE(FIRMWARE_RENOIR);
67 MODULE_FIRMWARE(FIRMWARE_GREEN_SARDINE);
68 MODULE_FIRMWARE(FIRMWARE_ALDEBARAN);
69 MODULE_FIRMWARE(FIRMWARE_NAVI10);
70 MODULE_FIRMWARE(FIRMWARE_NAVI14);
71 MODULE_FIRMWARE(FIRMWARE_NAVI12);
72 MODULE_FIRMWARE(FIRMWARE_SIENNA_CICHLID);
73 MODULE_FIRMWARE(FIRMWARE_NAVY_FLOUNDER);
74 MODULE_FIRMWARE(FIRMWARE_VANGOGH);
75 MODULE_FIRMWARE(FIRMWARE_DIMGREY_CAVEFISH);
76 MODULE_FIRMWARE(FIRMWARE_BEIGE_GOBY);
77 MODULE_FIRMWARE(FIRMWARE_YELLOW_CARP);
78 MODULE_FIRMWARE(FIRMWARE_VCN_3_1_2);
79 MODULE_FIRMWARE(FIRMWARE_VCN4_0_0);
80 MODULE_FIRMWARE(FIRMWARE_VCN4_0_2);
81 MODULE_FIRMWARE(FIRMWARE_VCN4_0_3);
82 MODULE_FIRMWARE(FIRMWARE_VCN4_0_4);
83 
84 static void amdgpu_vcn_idle_work_handler(struct work_struct *work);
85 
86 int amdgpu_vcn_early_init(struct amdgpu_device *adev)
87 {
88 	char ucode_prefix[30];
89 	char fw_name[40];
90 	int r;
91 
92 	amdgpu_ucode_ip_version_decode(adev, UVD_HWIP, ucode_prefix, sizeof(ucode_prefix));
93 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s.bin", ucode_prefix);
94 	r = amdgpu_ucode_request(adev, &adev->vcn.fw, fw_name);
95 	if (r)
96 		amdgpu_ucode_release(&adev->vcn.fw);
97 
98 	return r;
99 }
100 
101 int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
102 {
103 	unsigned long bo_size;
104 	const struct common_firmware_header *hdr;
105 	unsigned char fw_check;
106 	unsigned int fw_shared_size, log_offset;
107 	int i, r;
108 
109 	INIT_DELAYED_WORK(&adev->vcn.idle_work, amdgpu_vcn_idle_work_handler);
110 	mutex_init(&adev->vcn.vcn_pg_lock);
111 	mutex_init(&adev->vcn.vcn1_jpeg1_workaround);
112 	atomic_set(&adev->vcn.total_submission_cnt, 0);
113 	for (i = 0; i < adev->vcn.num_vcn_inst; i++)
114 		atomic_set(&adev->vcn.inst[i].dpg_enc_submission_cnt, 0);
115 
116 	if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
117 	    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
118 		adev->vcn.indirect_sram = true;
119 
120 	/*
121 	 * Some Steam Deck's BIOS versions are incompatible with the
122 	 * indirect SRAM mode, leading to amdgpu being unable to get
123 	 * properly probed (and even potentially crashing the kernel).
124 	 * Hence, check for these versions here - notice this is
125 	 * restricted to Vangogh (Deck's APU).
126 	 */
127 	if (adev->ip_versions[UVD_HWIP][0] == IP_VERSION(3, 0, 2)) {
128 		const char *bios_ver = dmi_get_system_info(DMI_BIOS_VERSION);
129 
130 		if (bios_ver && (!strncmp("F7A0113", bios_ver, 7) ||
131 		     !strncmp("F7A0114", bios_ver, 7))) {
132 			adev->vcn.indirect_sram = false;
133 			dev_info(adev->dev,
134 				"Steam Deck quirk: indirect SRAM disabled on BIOS %s\n", bios_ver);
135 		}
136 	}
137 
138 	hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
139 	adev->vcn.fw_version = le32_to_cpu(hdr->ucode_version);
140 
141 	/* Bit 20-23, it is encode major and non-zero for new naming convention.
142 	 * This field is part of version minor and DRM_DISABLED_FLAG in old naming
143 	 * convention. Since the l:wq!atest version minor is 0x5B and DRM_DISABLED_FLAG
144 	 * is zero in old naming convention, this field is always zero so far.
145 	 * These four bits are used to tell which naming convention is present.
146 	 */
147 	fw_check = (le32_to_cpu(hdr->ucode_version) >> 20) & 0xf;
148 	if (fw_check) {
149 		unsigned int dec_ver, enc_major, enc_minor, vep, fw_rev;
150 
151 		fw_rev = le32_to_cpu(hdr->ucode_version) & 0xfff;
152 		enc_minor = (le32_to_cpu(hdr->ucode_version) >> 12) & 0xff;
153 		enc_major = fw_check;
154 		dec_ver = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xf;
155 		vep = (le32_to_cpu(hdr->ucode_version) >> 28) & 0xf;
156 		DRM_INFO("Found VCN firmware Version ENC: %u.%u DEC: %u VEP: %u Revision: %u\n",
157 			enc_major, enc_minor, dec_ver, vep, fw_rev);
158 	} else {
159 		unsigned int version_major, version_minor, family_id;
160 
161 		family_id = le32_to_cpu(hdr->ucode_version) & 0xff;
162 		version_major = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xff;
163 		version_minor = (le32_to_cpu(hdr->ucode_version) >> 8) & 0xff;
164 		DRM_INFO("Found VCN firmware Version: %u.%u Family ID: %u\n",
165 			version_major, version_minor, family_id);
166 	}
167 
168 	bo_size = AMDGPU_VCN_STACK_SIZE + AMDGPU_VCN_CONTEXT_SIZE;
169 	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
170 		bo_size += AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
171 
172 	if (adev->ip_versions[UVD_HWIP][0] >= IP_VERSION(4, 0, 0)) {
173 		fw_shared_size = AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared));
174 		log_offset = offsetof(struct amdgpu_vcn4_fw_shared, fw_log);
175 	} else {
176 		fw_shared_size = AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
177 		log_offset = offsetof(struct amdgpu_fw_shared, fw_log);
178 	}
179 
180 	bo_size += fw_shared_size;
181 
182 	if (amdgpu_vcnfw_log)
183 		bo_size += AMDGPU_VCNFW_LOG_SIZE;
184 
185 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
186 		if (adev->vcn.harvest_config & (1 << i))
187 			continue;
188 
189 		r = amdgpu_bo_create_kernel(adev, bo_size, PAGE_SIZE,
190 					    AMDGPU_GEM_DOMAIN_VRAM |
191 					    AMDGPU_GEM_DOMAIN_GTT,
192 					    &adev->vcn.inst[i].vcpu_bo,
193 					    &adev->vcn.inst[i].gpu_addr,
194 					    &adev->vcn.inst[i].cpu_addr);
195 		if (r) {
196 			dev_err(adev->dev, "(%d) failed to allocate vcn bo\n", r);
197 			return r;
198 		}
199 
200 		adev->vcn.inst[i].fw_shared.cpu_addr = adev->vcn.inst[i].cpu_addr +
201 				bo_size - fw_shared_size;
202 		adev->vcn.inst[i].fw_shared.gpu_addr = adev->vcn.inst[i].gpu_addr +
203 				bo_size - fw_shared_size;
204 
205 		adev->vcn.inst[i].fw_shared.mem_size = fw_shared_size;
206 
207 		if (amdgpu_vcnfw_log) {
208 			adev->vcn.inst[i].fw_shared.cpu_addr -= AMDGPU_VCNFW_LOG_SIZE;
209 			adev->vcn.inst[i].fw_shared.gpu_addr -= AMDGPU_VCNFW_LOG_SIZE;
210 			adev->vcn.inst[i].fw_shared.log_offset = log_offset;
211 		}
212 
213 		if (adev->vcn.indirect_sram) {
214 			r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
215 					AMDGPU_GEM_DOMAIN_VRAM |
216 					AMDGPU_GEM_DOMAIN_GTT,
217 					&adev->vcn.inst[i].dpg_sram_bo,
218 					&adev->vcn.inst[i].dpg_sram_gpu_addr,
219 					&adev->vcn.inst[i].dpg_sram_cpu_addr);
220 			if (r) {
221 				dev_err(adev->dev, "VCN %d (%d) failed to allocate DPG bo\n", i, r);
222 				return r;
223 			}
224 		}
225 	}
226 
227 	return 0;
228 }
229 
230 int amdgpu_vcn_sw_fini(struct amdgpu_device *adev)
231 {
232 	int i, j;
233 
234 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
235 		if (adev->vcn.harvest_config & (1 << j))
236 			continue;
237 
238 		amdgpu_bo_free_kernel(
239 			&adev->vcn.inst[j].dpg_sram_bo,
240 			&adev->vcn.inst[j].dpg_sram_gpu_addr,
241 			(void **)&adev->vcn.inst[j].dpg_sram_cpu_addr);
242 
243 		kvfree(adev->vcn.inst[j].saved_bo);
244 
245 		amdgpu_bo_free_kernel(&adev->vcn.inst[j].vcpu_bo,
246 					  &adev->vcn.inst[j].gpu_addr,
247 					  (void **)&adev->vcn.inst[j].cpu_addr);
248 
249 		amdgpu_ring_fini(&adev->vcn.inst[j].ring_dec);
250 
251 		for (i = 0; i < adev->vcn.num_enc_rings; ++i)
252 			amdgpu_ring_fini(&adev->vcn.inst[j].ring_enc[i]);
253 	}
254 
255 	amdgpu_ucode_release(&adev->vcn.fw);
256 	mutex_destroy(&adev->vcn.vcn1_jpeg1_workaround);
257 	mutex_destroy(&adev->vcn.vcn_pg_lock);
258 
259 	return 0;
260 }
261 
262 /* from vcn4 and above, only unified queue is used */
263 static bool amdgpu_vcn_using_unified_queue(struct amdgpu_ring *ring)
264 {
265 	struct amdgpu_device *adev = ring->adev;
266 	bool ret = false;
267 
268 	if (adev->ip_versions[UVD_HWIP][0] >= IP_VERSION(4, 0, 0))
269 		ret = true;
270 
271 	return ret;
272 }
273 
274 bool amdgpu_vcn_is_disabled_vcn(struct amdgpu_device *adev, enum vcn_ring_type type, uint32_t vcn_instance)
275 {
276 	bool ret = false;
277 	int vcn_config = adev->vcn.vcn_config[vcn_instance];
278 
279 	if ((type == VCN_ENCODE_RING) && (vcn_config & VCN_BLOCK_ENCODE_DISABLE_MASK))
280 		ret = true;
281 	else if ((type == VCN_DECODE_RING) && (vcn_config & VCN_BLOCK_DECODE_DISABLE_MASK))
282 		ret = true;
283 	else if ((type == VCN_UNIFIED_RING) && (vcn_config & VCN_BLOCK_QUEUE_DISABLE_MASK))
284 		ret = true;
285 
286 	return ret;
287 }
288 
289 int amdgpu_vcn_suspend(struct amdgpu_device *adev)
290 {
291 	unsigned int size;
292 	void *ptr;
293 	int i, idx;
294 
295 	cancel_delayed_work_sync(&adev->vcn.idle_work);
296 
297 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
298 		if (adev->vcn.harvest_config & (1 << i))
299 			continue;
300 		if (adev->vcn.inst[i].vcpu_bo == NULL)
301 			return 0;
302 
303 		size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
304 		ptr = adev->vcn.inst[i].cpu_addr;
305 
306 		adev->vcn.inst[i].saved_bo = kvmalloc(size, GFP_KERNEL);
307 		if (!adev->vcn.inst[i].saved_bo)
308 			return -ENOMEM;
309 
310 		if (drm_dev_enter(adev_to_drm(adev), &idx)) {
311 			memcpy_fromio(adev->vcn.inst[i].saved_bo, ptr, size);
312 			drm_dev_exit(idx);
313 		}
314 	}
315 	return 0;
316 }
317 
318 int amdgpu_vcn_resume(struct amdgpu_device *adev)
319 {
320 	unsigned int size;
321 	void *ptr;
322 	int i, idx;
323 
324 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
325 		if (adev->vcn.harvest_config & (1 << i))
326 			continue;
327 		if (adev->vcn.inst[i].vcpu_bo == NULL)
328 			return -EINVAL;
329 
330 		size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
331 		ptr = adev->vcn.inst[i].cpu_addr;
332 
333 		if (adev->vcn.inst[i].saved_bo != NULL) {
334 			if (drm_dev_enter(adev_to_drm(adev), &idx)) {
335 				memcpy_toio(ptr, adev->vcn.inst[i].saved_bo, size);
336 				drm_dev_exit(idx);
337 			}
338 			kvfree(adev->vcn.inst[i].saved_bo);
339 			adev->vcn.inst[i].saved_bo = NULL;
340 		} else {
341 			const struct common_firmware_header *hdr;
342 			unsigned int offset;
343 
344 			hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
345 			if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
346 				offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
347 				if (drm_dev_enter(adev_to_drm(adev), &idx)) {
348 					memcpy_toio(adev->vcn.inst[i].cpu_addr, adev->vcn.fw->data + offset,
349 						    le32_to_cpu(hdr->ucode_size_bytes));
350 					drm_dev_exit(idx);
351 				}
352 				size -= le32_to_cpu(hdr->ucode_size_bytes);
353 				ptr += le32_to_cpu(hdr->ucode_size_bytes);
354 			}
355 			memset_io(ptr, 0, size);
356 		}
357 	}
358 	return 0;
359 }
360 
361 static void amdgpu_vcn_idle_work_handler(struct work_struct *work)
362 {
363 	struct amdgpu_device *adev =
364 		container_of(work, struct amdgpu_device, vcn.idle_work.work);
365 	unsigned int fences = 0, fence[AMDGPU_MAX_VCN_INSTANCES] = {0};
366 	unsigned int i, j;
367 	int r = 0;
368 
369 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
370 		if (adev->vcn.harvest_config & (1 << j))
371 			continue;
372 
373 		for (i = 0; i < adev->vcn.num_enc_rings; ++i)
374 			fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_enc[i]);
375 
376 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)	{
377 			struct dpg_pause_state new_state;
378 
379 			if (fence[j] ||
380 				unlikely(atomic_read(&adev->vcn.inst[j].dpg_enc_submission_cnt)))
381 				new_state.fw_based = VCN_DPG_STATE__PAUSE;
382 			else
383 				new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
384 
385 			adev->vcn.pause_dpg_mode(adev, j, &new_state);
386 		}
387 
388 		fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_dec);
389 		fences += fence[j];
390 	}
391 
392 	if (!fences && !atomic_read(&adev->vcn.total_submission_cnt)) {
393 		amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
394 		       AMD_PG_STATE_GATE);
395 		r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO,
396 				false);
397 		if (r)
398 			dev_warn(adev->dev, "(%d) failed to disable video power profile mode\n", r);
399 	} else {
400 		schedule_delayed_work(&adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
401 	}
402 }
403 
404 void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring)
405 {
406 	struct amdgpu_device *adev = ring->adev;
407 	int r = 0;
408 
409 	atomic_inc(&adev->vcn.total_submission_cnt);
410 
411 	if (!cancel_delayed_work_sync(&adev->vcn.idle_work)) {
412 		r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO,
413 				true);
414 		if (r)
415 			dev_warn(adev->dev, "(%d) failed to switch to video power profile mode\n", r);
416 	}
417 
418 	mutex_lock(&adev->vcn.vcn_pg_lock);
419 	amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
420 	       AMD_PG_STATE_UNGATE);
421 
422 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)	{
423 		struct dpg_pause_state new_state;
424 
425 		if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC) {
426 			atomic_inc(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
427 			new_state.fw_based = VCN_DPG_STATE__PAUSE;
428 		} else {
429 			unsigned int fences = 0;
430 			unsigned int i;
431 
432 			for (i = 0; i < adev->vcn.num_enc_rings; ++i)
433 				fences += amdgpu_fence_count_emitted(&adev->vcn.inst[ring->me].ring_enc[i]);
434 
435 			if (fences || atomic_read(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt))
436 				new_state.fw_based = VCN_DPG_STATE__PAUSE;
437 			else
438 				new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
439 		}
440 
441 		adev->vcn.pause_dpg_mode(adev, ring->me, &new_state);
442 	}
443 	mutex_unlock(&adev->vcn.vcn_pg_lock);
444 }
445 
446 void amdgpu_vcn_ring_end_use(struct amdgpu_ring *ring)
447 {
448 	if (ring->adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG &&
449 		ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC)
450 		atomic_dec(&ring->adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
451 
452 	atomic_dec(&ring->adev->vcn.total_submission_cnt);
453 
454 	schedule_delayed_work(&ring->adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
455 }
456 
457 int amdgpu_vcn_dec_ring_test_ring(struct amdgpu_ring *ring)
458 {
459 	struct amdgpu_device *adev = ring->adev;
460 	uint32_t tmp = 0;
461 	unsigned int i;
462 	int r;
463 
464 	/* VCN in SRIOV does not support direct register read/write */
465 	if (amdgpu_sriov_vf(adev))
466 		return 0;
467 
468 	WREG32(adev->vcn.inst[ring->me].external.scratch9, 0xCAFEDEAD);
469 	r = amdgpu_ring_alloc(ring, 3);
470 	if (r)
471 		return r;
472 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.scratch9, 0));
473 	amdgpu_ring_write(ring, 0xDEADBEEF);
474 	amdgpu_ring_commit(ring);
475 	for (i = 0; i < adev->usec_timeout; i++) {
476 		tmp = RREG32(adev->vcn.inst[ring->me].external.scratch9);
477 		if (tmp == 0xDEADBEEF)
478 			break;
479 		udelay(1);
480 	}
481 
482 	if (i >= adev->usec_timeout)
483 		r = -ETIMEDOUT;
484 
485 	return r;
486 }
487 
488 int amdgpu_vcn_dec_sw_ring_test_ring(struct amdgpu_ring *ring)
489 {
490 	struct amdgpu_device *adev = ring->adev;
491 	uint32_t rptr;
492 	unsigned int i;
493 	int r;
494 
495 	if (amdgpu_sriov_vf(adev))
496 		return 0;
497 
498 	r = amdgpu_ring_alloc(ring, 16);
499 	if (r)
500 		return r;
501 
502 	rptr = amdgpu_ring_get_rptr(ring);
503 
504 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_END);
505 	amdgpu_ring_commit(ring);
506 
507 	for (i = 0; i < adev->usec_timeout; i++) {
508 		if (amdgpu_ring_get_rptr(ring) != rptr)
509 			break;
510 		udelay(1);
511 	}
512 
513 	if (i >= adev->usec_timeout)
514 		r = -ETIMEDOUT;
515 
516 	return r;
517 }
518 
519 static int amdgpu_vcn_dec_send_msg(struct amdgpu_ring *ring,
520 				   struct amdgpu_ib *ib_msg,
521 				   struct dma_fence **fence)
522 {
523 	u64 addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
524 	struct amdgpu_device *adev = ring->adev;
525 	struct dma_fence *f = NULL;
526 	struct amdgpu_job *job;
527 	struct amdgpu_ib *ib;
528 	int i, r;
529 
530 	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
531 				     64, AMDGPU_IB_POOL_DIRECT,
532 				     &job);
533 	if (r)
534 		goto err;
535 
536 	ib = &job->ibs[0];
537 	ib->ptr[0] = PACKET0(adev->vcn.internal.data0, 0);
538 	ib->ptr[1] = addr;
539 	ib->ptr[2] = PACKET0(adev->vcn.internal.data1, 0);
540 	ib->ptr[3] = addr >> 32;
541 	ib->ptr[4] = PACKET0(adev->vcn.internal.cmd, 0);
542 	ib->ptr[5] = 0;
543 	for (i = 6; i < 16; i += 2) {
544 		ib->ptr[i] = PACKET0(adev->vcn.internal.nop, 0);
545 		ib->ptr[i+1] = 0;
546 	}
547 	ib->length_dw = 16;
548 
549 	r = amdgpu_job_submit_direct(job, ring, &f);
550 	if (r)
551 		goto err_free;
552 
553 	amdgpu_ib_free(adev, ib_msg, f);
554 
555 	if (fence)
556 		*fence = dma_fence_get(f);
557 	dma_fence_put(f);
558 
559 	return 0;
560 
561 err_free:
562 	amdgpu_job_free(job);
563 err:
564 	amdgpu_ib_free(adev, ib_msg, f);
565 	return r;
566 }
567 
568 static int amdgpu_vcn_dec_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
569 		struct amdgpu_ib *ib)
570 {
571 	struct amdgpu_device *adev = ring->adev;
572 	uint32_t *msg;
573 	int r, i;
574 
575 	memset(ib, 0, sizeof(*ib));
576 	r = amdgpu_ib_get(adev, NULL, AMDGPU_GPU_PAGE_SIZE * 2,
577 			AMDGPU_IB_POOL_DIRECT,
578 			ib);
579 	if (r)
580 		return r;
581 
582 	msg = (uint32_t *)AMDGPU_GPU_PAGE_ALIGN((unsigned long)ib->ptr);
583 	msg[0] = cpu_to_le32(0x00000028);
584 	msg[1] = cpu_to_le32(0x00000038);
585 	msg[2] = cpu_to_le32(0x00000001);
586 	msg[3] = cpu_to_le32(0x00000000);
587 	msg[4] = cpu_to_le32(handle);
588 	msg[5] = cpu_to_le32(0x00000000);
589 	msg[6] = cpu_to_le32(0x00000001);
590 	msg[7] = cpu_to_le32(0x00000028);
591 	msg[8] = cpu_to_le32(0x00000010);
592 	msg[9] = cpu_to_le32(0x00000000);
593 	msg[10] = cpu_to_le32(0x00000007);
594 	msg[11] = cpu_to_le32(0x00000000);
595 	msg[12] = cpu_to_le32(0x00000780);
596 	msg[13] = cpu_to_le32(0x00000440);
597 	for (i = 14; i < 1024; ++i)
598 		msg[i] = cpu_to_le32(0x0);
599 
600 	return 0;
601 }
602 
603 static int amdgpu_vcn_dec_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
604 					  struct amdgpu_ib *ib)
605 {
606 	struct amdgpu_device *adev = ring->adev;
607 	uint32_t *msg;
608 	int r, i;
609 
610 	memset(ib, 0, sizeof(*ib));
611 	r = amdgpu_ib_get(adev, NULL, AMDGPU_GPU_PAGE_SIZE * 2,
612 			AMDGPU_IB_POOL_DIRECT,
613 			ib);
614 	if (r)
615 		return r;
616 
617 	msg = (uint32_t *)AMDGPU_GPU_PAGE_ALIGN((unsigned long)ib->ptr);
618 	msg[0] = cpu_to_le32(0x00000028);
619 	msg[1] = cpu_to_le32(0x00000018);
620 	msg[2] = cpu_to_le32(0x00000000);
621 	msg[3] = cpu_to_le32(0x00000002);
622 	msg[4] = cpu_to_le32(handle);
623 	msg[5] = cpu_to_le32(0x00000000);
624 	for (i = 6; i < 1024; ++i)
625 		msg[i] = cpu_to_le32(0x0);
626 
627 	return 0;
628 }
629 
630 int amdgpu_vcn_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout)
631 {
632 	struct dma_fence *fence = NULL;
633 	struct amdgpu_ib ib;
634 	long r;
635 
636 	r = amdgpu_vcn_dec_get_create_msg(ring, 1, &ib);
637 	if (r)
638 		goto error;
639 
640 	r = amdgpu_vcn_dec_send_msg(ring, &ib, NULL);
641 	if (r)
642 		goto error;
643 	r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &ib);
644 	if (r)
645 		goto error;
646 
647 	r = amdgpu_vcn_dec_send_msg(ring, &ib, &fence);
648 	if (r)
649 		goto error;
650 
651 	r = dma_fence_wait_timeout(fence, false, timeout);
652 	if (r == 0)
653 		r = -ETIMEDOUT;
654 	else if (r > 0)
655 		r = 0;
656 
657 	dma_fence_put(fence);
658 error:
659 	return r;
660 }
661 
662 static uint32_t *amdgpu_vcn_unified_ring_ib_header(struct amdgpu_ib *ib,
663 						uint32_t ib_pack_in_dw, bool enc)
664 {
665 	uint32_t *ib_checksum;
666 
667 	ib->ptr[ib->length_dw++] = 0x00000010; /* single queue checksum */
668 	ib->ptr[ib->length_dw++] = 0x30000002;
669 	ib_checksum = &ib->ptr[ib->length_dw++];
670 	ib->ptr[ib->length_dw++] = ib_pack_in_dw;
671 
672 	ib->ptr[ib->length_dw++] = 0x00000010; /* engine info */
673 	ib->ptr[ib->length_dw++] = 0x30000001;
674 	ib->ptr[ib->length_dw++] = enc ? 0x2 : 0x3;
675 	ib->ptr[ib->length_dw++] = ib_pack_in_dw * sizeof(uint32_t);
676 
677 	return ib_checksum;
678 }
679 
680 static void amdgpu_vcn_unified_ring_ib_checksum(uint32_t **ib_checksum,
681 						uint32_t ib_pack_in_dw)
682 {
683 	uint32_t i;
684 	uint32_t checksum = 0;
685 
686 	for (i = 0; i < ib_pack_in_dw; i++)
687 		checksum += *(*ib_checksum + 2 + i);
688 
689 	**ib_checksum = checksum;
690 }
691 
692 static int amdgpu_vcn_dec_sw_send_msg(struct amdgpu_ring *ring,
693 				      struct amdgpu_ib *ib_msg,
694 				      struct dma_fence **fence)
695 {
696 	struct amdgpu_vcn_decode_buffer *decode_buffer = NULL;
697 	unsigned int ib_size_dw = 64;
698 	struct amdgpu_device *adev = ring->adev;
699 	struct dma_fence *f = NULL;
700 	struct amdgpu_job *job;
701 	struct amdgpu_ib *ib;
702 	uint64_t addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
703 	bool sq = amdgpu_vcn_using_unified_queue(ring);
704 	uint32_t *ib_checksum;
705 	uint32_t ib_pack_in_dw;
706 	int i, r;
707 
708 	if (sq)
709 		ib_size_dw += 8;
710 
711 	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
712 				     ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
713 				     &job);
714 	if (r)
715 		goto err;
716 
717 	ib = &job->ibs[0];
718 	ib->length_dw = 0;
719 
720 	/* single queue headers */
721 	if (sq) {
722 		ib_pack_in_dw = sizeof(struct amdgpu_vcn_decode_buffer) / sizeof(uint32_t)
723 						+ 4 + 2; /* engine info + decoding ib in dw */
724 		ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, ib_pack_in_dw, false);
725 	}
726 
727 	ib->ptr[ib->length_dw++] = sizeof(struct amdgpu_vcn_decode_buffer) + 8;
728 	ib->ptr[ib->length_dw++] = cpu_to_le32(AMDGPU_VCN_IB_FLAG_DECODE_BUFFER);
729 	decode_buffer = (struct amdgpu_vcn_decode_buffer *)&(ib->ptr[ib->length_dw]);
730 	ib->length_dw += sizeof(struct amdgpu_vcn_decode_buffer) / 4;
731 	memset(decode_buffer, 0, sizeof(struct amdgpu_vcn_decode_buffer));
732 
733 	decode_buffer->valid_buf_flag |= cpu_to_le32(AMDGPU_VCN_CMD_FLAG_MSG_BUFFER);
734 	decode_buffer->msg_buffer_address_hi = cpu_to_le32(addr >> 32);
735 	decode_buffer->msg_buffer_address_lo = cpu_to_le32(addr);
736 
737 	for (i = ib->length_dw; i < ib_size_dw; ++i)
738 		ib->ptr[i] = 0x0;
739 
740 	if (sq)
741 		amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, ib_pack_in_dw);
742 
743 	r = amdgpu_job_submit_direct(job, ring, &f);
744 	if (r)
745 		goto err_free;
746 
747 	amdgpu_ib_free(adev, ib_msg, f);
748 
749 	if (fence)
750 		*fence = dma_fence_get(f);
751 	dma_fence_put(f);
752 
753 	return 0;
754 
755 err_free:
756 	amdgpu_job_free(job);
757 err:
758 	amdgpu_ib_free(adev, ib_msg, f);
759 	return r;
760 }
761 
762 int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout)
763 {
764 	struct dma_fence *fence = NULL;
765 	struct amdgpu_ib ib;
766 	long r;
767 
768 	r = amdgpu_vcn_dec_get_create_msg(ring, 1, &ib);
769 	if (r)
770 		goto error;
771 
772 	r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, NULL);
773 	if (r)
774 		goto error;
775 	r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &ib);
776 	if (r)
777 		goto error;
778 
779 	r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, &fence);
780 	if (r)
781 		goto error;
782 
783 	r = dma_fence_wait_timeout(fence, false, timeout);
784 	if (r == 0)
785 		r = -ETIMEDOUT;
786 	else if (r > 0)
787 		r = 0;
788 
789 	dma_fence_put(fence);
790 error:
791 	return r;
792 }
793 
794 int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring)
795 {
796 	struct amdgpu_device *adev = ring->adev;
797 	uint32_t rptr;
798 	unsigned int i;
799 	int r;
800 
801 	if (amdgpu_sriov_vf(adev))
802 		return 0;
803 
804 	r = amdgpu_ring_alloc(ring, 16);
805 	if (r)
806 		return r;
807 
808 	rptr = amdgpu_ring_get_rptr(ring);
809 
810 	amdgpu_ring_write(ring, VCN_ENC_CMD_END);
811 	amdgpu_ring_commit(ring);
812 
813 	for (i = 0; i < adev->usec_timeout; i++) {
814 		if (amdgpu_ring_get_rptr(ring) != rptr)
815 			break;
816 		udelay(1);
817 	}
818 
819 	if (i >= adev->usec_timeout)
820 		r = -ETIMEDOUT;
821 
822 	return r;
823 }
824 
825 static int amdgpu_vcn_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
826 					 struct amdgpu_ib *ib_msg,
827 					 struct dma_fence **fence)
828 {
829 	unsigned int ib_size_dw = 16;
830 	struct amdgpu_job *job;
831 	struct amdgpu_ib *ib;
832 	struct dma_fence *f = NULL;
833 	uint32_t *ib_checksum = NULL;
834 	uint64_t addr;
835 	bool sq = amdgpu_vcn_using_unified_queue(ring);
836 	int i, r;
837 
838 	if (sq)
839 		ib_size_dw += 8;
840 
841 	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
842 				     ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
843 				     &job);
844 	if (r)
845 		return r;
846 
847 	ib = &job->ibs[0];
848 	addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
849 
850 	ib->length_dw = 0;
851 
852 	if (sq)
853 		ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, 0x11, true);
854 
855 	ib->ptr[ib->length_dw++] = 0x00000018;
856 	ib->ptr[ib->length_dw++] = 0x00000001; /* session info */
857 	ib->ptr[ib->length_dw++] = handle;
858 	ib->ptr[ib->length_dw++] = upper_32_bits(addr);
859 	ib->ptr[ib->length_dw++] = addr;
860 	ib->ptr[ib->length_dw++] = 0x0000000b;
861 
862 	ib->ptr[ib->length_dw++] = 0x00000014;
863 	ib->ptr[ib->length_dw++] = 0x00000002; /* task info */
864 	ib->ptr[ib->length_dw++] = 0x0000001c;
865 	ib->ptr[ib->length_dw++] = 0x00000000;
866 	ib->ptr[ib->length_dw++] = 0x00000000;
867 
868 	ib->ptr[ib->length_dw++] = 0x00000008;
869 	ib->ptr[ib->length_dw++] = 0x08000001; /* op initialize */
870 
871 	for (i = ib->length_dw; i < ib_size_dw; ++i)
872 		ib->ptr[i] = 0x0;
873 
874 	if (sq)
875 		amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, 0x11);
876 
877 	r = amdgpu_job_submit_direct(job, ring, &f);
878 	if (r)
879 		goto err;
880 
881 	if (fence)
882 		*fence = dma_fence_get(f);
883 	dma_fence_put(f);
884 
885 	return 0;
886 
887 err:
888 	amdgpu_job_free(job);
889 	return r;
890 }
891 
892 static int amdgpu_vcn_enc_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
893 					  struct amdgpu_ib *ib_msg,
894 					  struct dma_fence **fence)
895 {
896 	unsigned int ib_size_dw = 16;
897 	struct amdgpu_job *job;
898 	struct amdgpu_ib *ib;
899 	struct dma_fence *f = NULL;
900 	uint32_t *ib_checksum = NULL;
901 	uint64_t addr;
902 	bool sq = amdgpu_vcn_using_unified_queue(ring);
903 	int i, r;
904 
905 	if (sq)
906 		ib_size_dw += 8;
907 
908 	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
909 				     ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
910 				     &job);
911 	if (r)
912 		return r;
913 
914 	ib = &job->ibs[0];
915 	addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
916 
917 	ib->length_dw = 0;
918 
919 	if (sq)
920 		ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, 0x11, true);
921 
922 	ib->ptr[ib->length_dw++] = 0x00000018;
923 	ib->ptr[ib->length_dw++] = 0x00000001;
924 	ib->ptr[ib->length_dw++] = handle;
925 	ib->ptr[ib->length_dw++] = upper_32_bits(addr);
926 	ib->ptr[ib->length_dw++] = addr;
927 	ib->ptr[ib->length_dw++] = 0x0000000b;
928 
929 	ib->ptr[ib->length_dw++] = 0x00000014;
930 	ib->ptr[ib->length_dw++] = 0x00000002;
931 	ib->ptr[ib->length_dw++] = 0x0000001c;
932 	ib->ptr[ib->length_dw++] = 0x00000000;
933 	ib->ptr[ib->length_dw++] = 0x00000000;
934 
935 	ib->ptr[ib->length_dw++] = 0x00000008;
936 	ib->ptr[ib->length_dw++] = 0x08000002; /* op close session */
937 
938 	for (i = ib->length_dw; i < ib_size_dw; ++i)
939 		ib->ptr[i] = 0x0;
940 
941 	if (sq)
942 		amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, 0x11);
943 
944 	r = amdgpu_job_submit_direct(job, ring, &f);
945 	if (r)
946 		goto err;
947 
948 	if (fence)
949 		*fence = dma_fence_get(f);
950 	dma_fence_put(f);
951 
952 	return 0;
953 
954 err:
955 	amdgpu_job_free(job);
956 	return r;
957 }
958 
959 int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout)
960 {
961 	struct amdgpu_device *adev = ring->adev;
962 	struct dma_fence *fence = NULL;
963 	struct amdgpu_ib ib;
964 	long r;
965 
966 	memset(&ib, 0, sizeof(ib));
967 	r = amdgpu_ib_get(adev, NULL, (128 << 10) + AMDGPU_GPU_PAGE_SIZE,
968 			AMDGPU_IB_POOL_DIRECT,
969 			&ib);
970 	if (r)
971 		return r;
972 
973 	r = amdgpu_vcn_enc_get_create_msg(ring, 1, &ib, NULL);
974 	if (r)
975 		goto error;
976 
977 	r = amdgpu_vcn_enc_get_destroy_msg(ring, 1, &ib, &fence);
978 	if (r)
979 		goto error;
980 
981 	r = dma_fence_wait_timeout(fence, false, timeout);
982 	if (r == 0)
983 		r = -ETIMEDOUT;
984 	else if (r > 0)
985 		r = 0;
986 
987 error:
988 	amdgpu_ib_free(adev, &ib, fence);
989 	dma_fence_put(fence);
990 
991 	return r;
992 }
993 
994 int amdgpu_vcn_unified_ring_test_ib(struct amdgpu_ring *ring, long timeout)
995 {
996 	struct amdgpu_device *adev = ring->adev;
997 	long r;
998 
999 	if (adev->ip_versions[UVD_HWIP][0] != IP_VERSION(4, 0, 3)) {
1000 		r = amdgpu_vcn_enc_ring_test_ib(ring, timeout);
1001 		if (r)
1002 			goto error;
1003 	}
1004 
1005 	r =  amdgpu_vcn_dec_sw_ring_test_ib(ring, timeout);
1006 
1007 error:
1008 	return r;
1009 }
1010 
1011 enum amdgpu_ring_priority_level amdgpu_vcn_get_enc_ring_prio(int ring)
1012 {
1013 	switch (ring) {
1014 	case 0:
1015 		return AMDGPU_RING_PRIO_0;
1016 	case 1:
1017 		return AMDGPU_RING_PRIO_1;
1018 	case 2:
1019 		return AMDGPU_RING_PRIO_2;
1020 	default:
1021 		return AMDGPU_RING_PRIO_0;
1022 	}
1023 }
1024 
1025 void amdgpu_vcn_setup_ucode(struct amdgpu_device *adev)
1026 {
1027 	int i;
1028 	unsigned int idx;
1029 
1030 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1031 		const struct common_firmware_header *hdr;
1032 
1033 		hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
1034 
1035 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1036 			if (adev->vcn.harvest_config & (1 << i))
1037 				continue;
1038 			/* currently only support 2 FW instances */
1039 			if (i >= 2) {
1040 				dev_info(adev->dev, "More then 2 VCN FW instances!\n");
1041 				break;
1042 			}
1043 			idx = AMDGPU_UCODE_ID_VCN + i;
1044 			adev->firmware.ucode[idx].ucode_id = idx;
1045 			adev->firmware.ucode[idx].fw = adev->vcn.fw;
1046 			adev->firmware.fw_size +=
1047 				ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
1048 
1049 			if (adev->ip_versions[UVD_HWIP][0] == IP_VERSION(4, 0, 3))
1050 				break;
1051 		}
1052 		dev_info(adev->dev, "Will use PSP to load VCN firmware\n");
1053 	}
1054 }
1055 
1056 /*
1057  * debugfs for mapping vcn firmware log buffer.
1058  */
1059 #if defined(CONFIG_DEBUG_FS)
1060 static ssize_t amdgpu_debugfs_vcn_fwlog_read(struct file *f, char __user *buf,
1061 					     size_t size, loff_t *pos)
1062 {
1063 	struct amdgpu_vcn_inst *vcn;
1064 	void *log_buf;
1065 	volatile struct amdgpu_vcn_fwlog *plog;
1066 	unsigned int read_pos, write_pos, available, i, read_bytes = 0;
1067 	unsigned int read_num[2] = {0};
1068 
1069 	vcn = file_inode(f)->i_private;
1070 	if (!vcn)
1071 		return -ENODEV;
1072 
1073 	if (!vcn->fw_shared.cpu_addr || !amdgpu_vcnfw_log)
1074 		return -EFAULT;
1075 
1076 	log_buf = vcn->fw_shared.cpu_addr + vcn->fw_shared.mem_size;
1077 
1078 	plog = (volatile struct amdgpu_vcn_fwlog *)log_buf;
1079 	read_pos = plog->rptr;
1080 	write_pos = plog->wptr;
1081 
1082 	if (read_pos > AMDGPU_VCNFW_LOG_SIZE || write_pos > AMDGPU_VCNFW_LOG_SIZE)
1083 		return -EFAULT;
1084 
1085 	if (!size || (read_pos == write_pos))
1086 		return 0;
1087 
1088 	if (write_pos > read_pos) {
1089 		available = write_pos - read_pos;
1090 		read_num[0] = min(size, (size_t)available);
1091 	} else {
1092 		read_num[0] = AMDGPU_VCNFW_LOG_SIZE - read_pos;
1093 		available = read_num[0] + write_pos - plog->header_size;
1094 		if (size > available)
1095 			read_num[1] = write_pos - plog->header_size;
1096 		else if (size > read_num[0])
1097 			read_num[1] = size - read_num[0];
1098 		else
1099 			read_num[0] = size;
1100 	}
1101 
1102 	for (i = 0; i < 2; i++) {
1103 		if (read_num[i]) {
1104 			if (read_pos == AMDGPU_VCNFW_LOG_SIZE)
1105 				read_pos = plog->header_size;
1106 			if (read_num[i] == copy_to_user((buf + read_bytes),
1107 							(log_buf + read_pos), read_num[i]))
1108 				return -EFAULT;
1109 
1110 			read_bytes += read_num[i];
1111 			read_pos += read_num[i];
1112 		}
1113 	}
1114 
1115 	plog->rptr = read_pos;
1116 	*pos += read_bytes;
1117 	return read_bytes;
1118 }
1119 
1120 static const struct file_operations amdgpu_debugfs_vcnfwlog_fops = {
1121 	.owner = THIS_MODULE,
1122 	.read = amdgpu_debugfs_vcn_fwlog_read,
1123 	.llseek = default_llseek
1124 };
1125 #endif
1126 
1127 void amdgpu_debugfs_vcn_fwlog_init(struct amdgpu_device *adev, uint8_t i,
1128 				   struct amdgpu_vcn_inst *vcn)
1129 {
1130 #if defined(CONFIG_DEBUG_FS)
1131 	struct drm_minor *minor = adev_to_drm(adev)->primary;
1132 	struct dentry *root = minor->debugfs_root;
1133 	char name[32];
1134 
1135 	sprintf(name, "amdgpu_vcn_%d_fwlog", i);
1136 	debugfs_create_file_size(name, S_IFREG | 0444, root, vcn,
1137 				 &amdgpu_debugfs_vcnfwlog_fops,
1138 				 AMDGPU_VCNFW_LOG_SIZE);
1139 #endif
1140 }
1141 
1142 void amdgpu_vcn_fwlog_init(struct amdgpu_vcn_inst *vcn)
1143 {
1144 #if defined(CONFIG_DEBUG_FS)
1145 	volatile uint32_t *flag = vcn->fw_shared.cpu_addr;
1146 	void *fw_log_cpu_addr = vcn->fw_shared.cpu_addr + vcn->fw_shared.mem_size;
1147 	uint64_t fw_log_gpu_addr = vcn->fw_shared.gpu_addr + vcn->fw_shared.mem_size;
1148 	volatile struct amdgpu_vcn_fwlog *log_buf = fw_log_cpu_addr;
1149 	volatile struct amdgpu_fw_shared_fw_logging *fw_log = vcn->fw_shared.cpu_addr
1150 							 + vcn->fw_shared.log_offset;
1151 	*flag |= cpu_to_le32(AMDGPU_VCN_FW_LOGGING_FLAG);
1152 	fw_log->is_enabled = 1;
1153 	fw_log->addr_lo = cpu_to_le32(fw_log_gpu_addr & 0xFFFFFFFF);
1154 	fw_log->addr_hi = cpu_to_le32(fw_log_gpu_addr >> 32);
1155 	fw_log->size = cpu_to_le32(AMDGPU_VCNFW_LOG_SIZE);
1156 
1157 	log_buf->header_size = sizeof(struct amdgpu_vcn_fwlog);
1158 	log_buf->buffer_size = AMDGPU_VCNFW_LOG_SIZE;
1159 	log_buf->rptr = log_buf->header_size;
1160 	log_buf->wptr = log_buf->header_size;
1161 	log_buf->wrapped = 0;
1162 #endif
1163 }
1164 
1165 int amdgpu_vcn_process_poison_irq(struct amdgpu_device *adev,
1166 				struct amdgpu_irq_src *source,
1167 				struct amdgpu_iv_entry *entry)
1168 {
1169 	struct ras_common_if *ras_if = adev->vcn.ras_if;
1170 	struct ras_dispatch_if ih_data = {
1171 		.entry = entry,
1172 	};
1173 
1174 	if (!ras_if)
1175 		return 0;
1176 
1177 	if (!amdgpu_sriov_vf(adev)) {
1178 		ih_data.head = *ras_if;
1179 		amdgpu_ras_interrupt_dispatch(adev, &ih_data);
1180 	} else {
1181 		if (adev->virt.ops && adev->virt.ops->ras_poison_handler)
1182 			adev->virt.ops->ras_poison_handler(adev);
1183 		else
1184 			dev_warn(adev->dev,
1185 				"No ras_poison_handler interface in SRIOV for VCN!\n");
1186 	}
1187 
1188 	return 0;
1189 }
1190 
1191 int amdgpu_vcn_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block)
1192 {
1193 	int r, i;
1194 
1195 	r = amdgpu_ras_block_late_init(adev, ras_block);
1196 	if (r)
1197 		return r;
1198 
1199 	if (amdgpu_ras_is_supported(adev, ras_block->block)) {
1200 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1201 			if (adev->vcn.harvest_config & (1 << i) ||
1202 			    !adev->vcn.inst[i].ras_poison_irq.funcs)
1203 				continue;
1204 
1205 			r = amdgpu_irq_get(adev, &adev->vcn.inst[i].ras_poison_irq, 0);
1206 			if (r)
1207 				goto late_fini;
1208 		}
1209 	}
1210 	return 0;
1211 
1212 late_fini:
1213 	amdgpu_ras_block_late_fini(adev, ras_block);
1214 	return r;
1215 }
1216 
1217 int amdgpu_vcn_ras_sw_init(struct amdgpu_device *adev)
1218 {
1219 	int err;
1220 	struct amdgpu_vcn_ras *ras;
1221 
1222 	if (!adev->vcn.ras)
1223 		return 0;
1224 
1225 	ras = adev->vcn.ras;
1226 	err = amdgpu_ras_register_ras_block(adev, &ras->ras_block);
1227 	if (err) {
1228 		dev_err(adev->dev, "Failed to register vcn ras block!\n");
1229 		return err;
1230 	}
1231 
1232 	strcpy(ras->ras_block.ras_comm.name, "vcn");
1233 	ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__VCN;
1234 	ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__POISON;
1235 	adev->vcn.ras_if = &ras->ras_block.ras_comm;
1236 
1237 	if (!ras->ras_block.ras_late_init)
1238 		ras->ras_block.ras_late_init = amdgpu_vcn_ras_late_init;
1239 
1240 	return 0;
1241 }
1242