xref: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c (revision 906fd46a65383cd639e5eec72a047efc33045d86)
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 #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 
66 MODULE_FIRMWARE(FIRMWARE_RAVEN);
67 MODULE_FIRMWARE(FIRMWARE_PICASSO);
68 MODULE_FIRMWARE(FIRMWARE_RAVEN2);
69 MODULE_FIRMWARE(FIRMWARE_ARCTURUS);
70 MODULE_FIRMWARE(FIRMWARE_RENOIR);
71 MODULE_FIRMWARE(FIRMWARE_GREEN_SARDINE);
72 MODULE_FIRMWARE(FIRMWARE_ALDEBARAN);
73 MODULE_FIRMWARE(FIRMWARE_NAVI10);
74 MODULE_FIRMWARE(FIRMWARE_NAVI14);
75 MODULE_FIRMWARE(FIRMWARE_NAVI12);
76 MODULE_FIRMWARE(FIRMWARE_SIENNA_CICHLID);
77 MODULE_FIRMWARE(FIRMWARE_NAVY_FLOUNDER);
78 MODULE_FIRMWARE(FIRMWARE_VANGOGH);
79 MODULE_FIRMWARE(FIRMWARE_DIMGREY_CAVEFISH);
80 MODULE_FIRMWARE(FIRMWARE_BEIGE_GOBY);
81 MODULE_FIRMWARE(FIRMWARE_YELLOW_CARP);
82 MODULE_FIRMWARE(FIRMWARE_VCN_3_1_2);
83 MODULE_FIRMWARE(FIRMWARE_VCN4_0_0);
84 MODULE_FIRMWARE(FIRMWARE_VCN4_0_2);
85 MODULE_FIRMWARE(FIRMWARE_VCN4_0_3);
86 MODULE_FIRMWARE(FIRMWARE_VCN4_0_4);
87 MODULE_FIRMWARE(FIRMWARE_VCN4_0_5);
88 MODULE_FIRMWARE(FIRMWARE_VCN4_0_6);
89 MODULE_FIRMWARE(FIRMWARE_VCN4_0_6_1);
90 MODULE_FIRMWARE(FIRMWARE_VCN5_0_0);
91 
92 static void amdgpu_vcn_idle_work_handler(struct work_struct *work);
93 
94 int amdgpu_vcn_early_init(struct amdgpu_device *adev)
95 {
96 	char ucode_prefix[25];
97 	int r, i;
98 
99 	amdgpu_ucode_ip_version_decode(adev, UVD_HWIP, ucode_prefix, sizeof(ucode_prefix));
100 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
101 		if (i == 1 && amdgpu_ip_version(adev, UVD_HWIP, 0) ==  IP_VERSION(4, 0, 6))
102 			r = amdgpu_ucode_request(adev, &adev->vcn.fw[i], "amdgpu/%s_%d.bin", ucode_prefix, i);
103 		else
104 			r = amdgpu_ucode_request(adev, &adev->vcn.fw[i], "amdgpu/%s.bin", ucode_prefix);
105 		if (r) {
106 			amdgpu_ucode_release(&adev->vcn.fw[i]);
107 			return r;
108 		}
109 	}
110 	return r;
111 }
112 
113 int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
114 {
115 	unsigned long bo_size;
116 	const struct common_firmware_header *hdr;
117 	unsigned char fw_check;
118 	unsigned int fw_shared_size, log_offset;
119 	int i, r;
120 
121 	INIT_DELAYED_WORK(&adev->vcn.idle_work, amdgpu_vcn_idle_work_handler);
122 	mutex_init(&adev->vcn.vcn_pg_lock);
123 	mutex_init(&adev->vcn.vcn1_jpeg1_workaround);
124 	atomic_set(&adev->vcn.total_submission_cnt, 0);
125 	for (i = 0; i < adev->vcn.num_vcn_inst; i++)
126 		atomic_set(&adev->vcn.inst[i].dpg_enc_submission_cnt, 0);
127 
128 	if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
129 	    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
130 		adev->vcn.indirect_sram = true;
131 
132 	/*
133 	 * Some Steam Deck's BIOS versions are incompatible with the
134 	 * indirect SRAM mode, leading to amdgpu being unable to get
135 	 * properly probed (and even potentially crashing the kernel).
136 	 * Hence, check for these versions here - notice this is
137 	 * restricted to Vangogh (Deck's APU).
138 	 */
139 	if (amdgpu_ip_version(adev, UVD_HWIP, 0) == IP_VERSION(3, 0, 2)) {
140 		const char *bios_ver = dmi_get_system_info(DMI_BIOS_VERSION);
141 
142 		if (bios_ver && (!strncmp("F7A0113", bios_ver, 7) ||
143 		     !strncmp("F7A0114", bios_ver, 7))) {
144 			adev->vcn.indirect_sram = false;
145 			dev_info(adev->dev,
146 				"Steam Deck quirk: indirect SRAM disabled on BIOS %s\n", bios_ver);
147 		}
148 	}
149 
150 	/* from vcn4 and above, only unified queue is used */
151 	adev->vcn.using_unified_queue =
152 		amdgpu_ip_version(adev, UVD_HWIP, 0) >= IP_VERSION(4, 0, 0);
153 
154 	hdr = (const struct common_firmware_header *)adev->vcn.fw[0]->data;
155 	adev->vcn.fw_version = le32_to_cpu(hdr->ucode_version);
156 
157 	/* Bit 20-23, it is encode major and non-zero for new naming convention.
158 	 * This field is part of version minor and DRM_DISABLED_FLAG in old naming
159 	 * convention. Since the l:wq!atest version minor is 0x5B and DRM_DISABLED_FLAG
160 	 * is zero in old naming convention, this field is always zero so far.
161 	 * These four bits are used to tell which naming convention is present.
162 	 */
163 	fw_check = (le32_to_cpu(hdr->ucode_version) >> 20) & 0xf;
164 	if (fw_check) {
165 		unsigned int dec_ver, enc_major, enc_minor, vep, fw_rev;
166 
167 		fw_rev = le32_to_cpu(hdr->ucode_version) & 0xfff;
168 		enc_minor = (le32_to_cpu(hdr->ucode_version) >> 12) & 0xff;
169 		enc_major = fw_check;
170 		dec_ver = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xf;
171 		vep = (le32_to_cpu(hdr->ucode_version) >> 28) & 0xf;
172 		DRM_INFO("Found VCN firmware Version ENC: %u.%u DEC: %u VEP: %u Revision: %u\n",
173 			enc_major, enc_minor, dec_ver, vep, fw_rev);
174 	} else {
175 		unsigned int version_major, version_minor, family_id;
176 
177 		family_id = le32_to_cpu(hdr->ucode_version) & 0xff;
178 		version_major = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xff;
179 		version_minor = (le32_to_cpu(hdr->ucode_version) >> 8) & 0xff;
180 		DRM_INFO("Found VCN firmware Version: %u.%u Family ID: %u\n",
181 			version_major, version_minor, family_id);
182 	}
183 
184 	bo_size = AMDGPU_VCN_STACK_SIZE + AMDGPU_VCN_CONTEXT_SIZE;
185 	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
186 		bo_size += AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
187 
188 	if (amdgpu_ip_version(adev, UVD_HWIP, 0) >= IP_VERSION(5, 0, 0)) {
189 		fw_shared_size = AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn5_fw_shared));
190 		log_offset = offsetof(struct amdgpu_vcn5_fw_shared, fw_log);
191 	} else if (amdgpu_ip_version(adev, UVD_HWIP, 0) >= IP_VERSION(4, 0, 0)) {
192 		fw_shared_size = AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared));
193 		log_offset = offsetof(struct amdgpu_vcn4_fw_shared, fw_log);
194 	} else {
195 		fw_shared_size = AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
196 		log_offset = offsetof(struct amdgpu_fw_shared, fw_log);
197 	}
198 
199 	bo_size += fw_shared_size;
200 
201 	if (amdgpu_vcnfw_log)
202 		bo_size += AMDGPU_VCNFW_LOG_SIZE;
203 
204 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
205 		if (adev->vcn.harvest_config & (1 << i))
206 			continue;
207 
208 		r = amdgpu_bo_create_kernel(adev, bo_size, PAGE_SIZE,
209 					    AMDGPU_GEM_DOMAIN_VRAM |
210 					    AMDGPU_GEM_DOMAIN_GTT,
211 					    &adev->vcn.inst[i].vcpu_bo,
212 					    &adev->vcn.inst[i].gpu_addr,
213 					    &adev->vcn.inst[i].cpu_addr);
214 		if (r) {
215 			dev_err(adev->dev, "(%d) failed to allocate vcn bo\n", r);
216 			return r;
217 		}
218 
219 		adev->vcn.inst[i].fw_shared.cpu_addr = adev->vcn.inst[i].cpu_addr +
220 				bo_size - fw_shared_size;
221 		adev->vcn.inst[i].fw_shared.gpu_addr = adev->vcn.inst[i].gpu_addr +
222 				bo_size - fw_shared_size;
223 
224 		adev->vcn.inst[i].fw_shared.mem_size = fw_shared_size;
225 
226 		if (amdgpu_vcnfw_log) {
227 			adev->vcn.inst[i].fw_shared.cpu_addr -= AMDGPU_VCNFW_LOG_SIZE;
228 			adev->vcn.inst[i].fw_shared.gpu_addr -= AMDGPU_VCNFW_LOG_SIZE;
229 			adev->vcn.inst[i].fw_shared.log_offset = log_offset;
230 		}
231 
232 		if (adev->vcn.indirect_sram) {
233 			r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
234 					AMDGPU_GEM_DOMAIN_VRAM |
235 					AMDGPU_GEM_DOMAIN_GTT,
236 					&adev->vcn.inst[i].dpg_sram_bo,
237 					&adev->vcn.inst[i].dpg_sram_gpu_addr,
238 					&adev->vcn.inst[i].dpg_sram_cpu_addr);
239 			if (r) {
240 				dev_err(adev->dev, "VCN %d (%d) failed to allocate DPG bo\n", i, r);
241 				return r;
242 			}
243 		}
244 	}
245 
246 	return 0;
247 }
248 
249 int amdgpu_vcn_sw_fini(struct amdgpu_device *adev)
250 {
251 	int i, j;
252 
253 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
254 		if (adev->vcn.harvest_config & (1 << j))
255 			continue;
256 
257 		amdgpu_bo_free_kernel(
258 			&adev->vcn.inst[j].dpg_sram_bo,
259 			&adev->vcn.inst[j].dpg_sram_gpu_addr,
260 			(void **)&adev->vcn.inst[j].dpg_sram_cpu_addr);
261 
262 		kvfree(adev->vcn.inst[j].saved_bo);
263 
264 		amdgpu_bo_free_kernel(&adev->vcn.inst[j].vcpu_bo,
265 					  &adev->vcn.inst[j].gpu_addr,
266 					  (void **)&adev->vcn.inst[j].cpu_addr);
267 
268 		amdgpu_ring_fini(&adev->vcn.inst[j].ring_dec);
269 
270 		for (i = 0; i < adev->vcn.num_enc_rings; ++i)
271 			amdgpu_ring_fini(&adev->vcn.inst[j].ring_enc[i]);
272 
273 		amdgpu_ucode_release(&adev->vcn.fw[j]);
274 	}
275 
276 	mutex_destroy(&adev->vcn.vcn1_jpeg1_workaround);
277 	mutex_destroy(&adev->vcn.vcn_pg_lock);
278 
279 	return 0;
280 }
281 
282 bool amdgpu_vcn_is_disabled_vcn(struct amdgpu_device *adev, enum vcn_ring_type type, uint32_t vcn_instance)
283 {
284 	bool ret = false;
285 	int vcn_config = adev->vcn.vcn_config[vcn_instance];
286 
287 	if ((type == VCN_ENCODE_RING) && (vcn_config & VCN_BLOCK_ENCODE_DISABLE_MASK))
288 		ret = true;
289 	else if ((type == VCN_DECODE_RING) && (vcn_config & VCN_BLOCK_DECODE_DISABLE_MASK))
290 		ret = true;
291 	else if ((type == VCN_UNIFIED_RING) && (vcn_config & VCN_BLOCK_QUEUE_DISABLE_MASK))
292 		ret = true;
293 
294 	return ret;
295 }
296 
297 int amdgpu_vcn_suspend(struct amdgpu_device *adev)
298 {
299 	unsigned int size;
300 	void *ptr;
301 	int i, idx;
302 
303 	bool in_ras_intr = amdgpu_ras_intr_triggered();
304 
305 	cancel_delayed_work_sync(&adev->vcn.idle_work);
306 
307 	/* err_event_athub will corrupt VCPU buffer, so we need to
308 	 * restore fw data and clear buffer in amdgpu_vcn_resume() */
309 	if (in_ras_intr)
310 		return 0;
311 
312 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
313 		if (adev->vcn.harvest_config & (1 << i))
314 			continue;
315 		if (adev->vcn.inst[i].vcpu_bo == NULL)
316 			return 0;
317 
318 		size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
319 		ptr = adev->vcn.inst[i].cpu_addr;
320 
321 		adev->vcn.inst[i].saved_bo = kvmalloc(size, GFP_KERNEL);
322 		if (!adev->vcn.inst[i].saved_bo)
323 			return -ENOMEM;
324 
325 		if (drm_dev_enter(adev_to_drm(adev), &idx)) {
326 			memcpy_fromio(adev->vcn.inst[i].saved_bo, ptr, size);
327 			drm_dev_exit(idx);
328 		}
329 	}
330 	return 0;
331 }
332 
333 int amdgpu_vcn_resume(struct amdgpu_device *adev)
334 {
335 	unsigned int size;
336 	void *ptr;
337 	int i, idx;
338 
339 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
340 		if (adev->vcn.harvest_config & (1 << i))
341 			continue;
342 		if (adev->vcn.inst[i].vcpu_bo == NULL)
343 			return -EINVAL;
344 
345 		size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
346 		ptr = adev->vcn.inst[i].cpu_addr;
347 
348 		if (adev->vcn.inst[i].saved_bo != NULL) {
349 			if (drm_dev_enter(adev_to_drm(adev), &idx)) {
350 				memcpy_toio(ptr, adev->vcn.inst[i].saved_bo, size);
351 				drm_dev_exit(idx);
352 			}
353 			kvfree(adev->vcn.inst[i].saved_bo);
354 			adev->vcn.inst[i].saved_bo = NULL;
355 		} else {
356 			const struct common_firmware_header *hdr;
357 			unsigned int offset;
358 
359 			hdr = (const struct common_firmware_header *)adev->vcn.fw[i]->data;
360 			if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
361 				offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
362 				if (drm_dev_enter(adev_to_drm(adev), &idx)) {
363 					memcpy_toio(adev->vcn.inst[i].cpu_addr,
364 						    adev->vcn.fw[i]->data + offset,
365 						    le32_to_cpu(hdr->ucode_size_bytes));
366 					drm_dev_exit(idx);
367 				}
368 				size -= le32_to_cpu(hdr->ucode_size_bytes);
369 				ptr += le32_to_cpu(hdr->ucode_size_bytes);
370 			}
371 			memset_io(ptr, 0, size);
372 		}
373 	}
374 	return 0;
375 }
376 
377 static void amdgpu_vcn_idle_work_handler(struct work_struct *work)
378 {
379 	struct amdgpu_device *adev =
380 		container_of(work, struct amdgpu_device, vcn.idle_work.work);
381 	unsigned int fences = 0, fence[AMDGPU_MAX_VCN_INSTANCES] = {0};
382 	unsigned int i, j;
383 	int r = 0;
384 
385 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
386 		if (adev->vcn.harvest_config & (1 << j))
387 			continue;
388 
389 		for (i = 0; i < adev->vcn.num_enc_rings; ++i)
390 			fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_enc[i]);
391 
392 		/* Only set DPG pause for VCN3 or below, VCN4 and above will be handled by FW */
393 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG &&
394 		    !adev->vcn.using_unified_queue) {
395 			struct dpg_pause_state new_state;
396 
397 			if (fence[j] ||
398 				unlikely(atomic_read(&adev->vcn.inst[j].dpg_enc_submission_cnt)))
399 				new_state.fw_based = VCN_DPG_STATE__PAUSE;
400 			else
401 				new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
402 
403 			adev->vcn.pause_dpg_mode(adev, j, &new_state);
404 		}
405 
406 		fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_dec);
407 		fences += fence[j];
408 	}
409 
410 	if (!fences && !atomic_read(&adev->vcn.total_submission_cnt)) {
411 		amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
412 		       AMD_PG_STATE_GATE);
413 		r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO,
414 				false);
415 		if (r)
416 			dev_warn(adev->dev, "(%d) failed to disable video power profile mode\n", r);
417 	} else {
418 		schedule_delayed_work(&adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
419 	}
420 }
421 
422 void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring)
423 {
424 	struct amdgpu_device *adev = ring->adev;
425 	int r = 0;
426 
427 	atomic_inc(&adev->vcn.total_submission_cnt);
428 
429 	if (!cancel_delayed_work_sync(&adev->vcn.idle_work)) {
430 		r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO,
431 				true);
432 		if (r)
433 			dev_warn(adev->dev, "(%d) failed to switch to video power profile mode\n", r);
434 	}
435 
436 	mutex_lock(&adev->vcn.vcn_pg_lock);
437 	amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
438 	       AMD_PG_STATE_UNGATE);
439 
440 	/* Only set DPG pause for VCN3 or below, VCN4 and above will be handled by FW */
441 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG &&
442 	    !adev->vcn.using_unified_queue) {
443 		struct dpg_pause_state new_state;
444 
445 		if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC) {
446 			atomic_inc(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
447 			new_state.fw_based = VCN_DPG_STATE__PAUSE;
448 		} else {
449 			unsigned int fences = 0;
450 			unsigned int i;
451 
452 			for (i = 0; i < adev->vcn.num_enc_rings; ++i)
453 				fences += amdgpu_fence_count_emitted(&adev->vcn.inst[ring->me].ring_enc[i]);
454 
455 			if (fences || atomic_read(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt))
456 				new_state.fw_based = VCN_DPG_STATE__PAUSE;
457 			else
458 				new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
459 		}
460 
461 		adev->vcn.pause_dpg_mode(adev, ring->me, &new_state);
462 	}
463 	mutex_unlock(&adev->vcn.vcn_pg_lock);
464 }
465 
466 void amdgpu_vcn_ring_end_use(struct amdgpu_ring *ring)
467 {
468 	struct amdgpu_device *adev = ring->adev;
469 
470 	/* Only set DPG pause for VCN3 or below, VCN4 and above will be handled by FW */
471 	if (ring->adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG &&
472 	    ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC &&
473 	    !adev->vcn.using_unified_queue)
474 		atomic_dec(&ring->adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
475 
476 	atomic_dec(&ring->adev->vcn.total_submission_cnt);
477 
478 	schedule_delayed_work(&ring->adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
479 }
480 
481 int amdgpu_vcn_dec_ring_test_ring(struct amdgpu_ring *ring)
482 {
483 	struct amdgpu_device *adev = ring->adev;
484 	uint32_t tmp = 0;
485 	unsigned int i;
486 	int r;
487 
488 	/* VCN in SRIOV does not support direct register read/write */
489 	if (amdgpu_sriov_vf(adev))
490 		return 0;
491 
492 	WREG32(adev->vcn.inst[ring->me].external.scratch9, 0xCAFEDEAD);
493 	r = amdgpu_ring_alloc(ring, 3);
494 	if (r)
495 		return r;
496 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.scratch9, 0));
497 	amdgpu_ring_write(ring, 0xDEADBEEF);
498 	amdgpu_ring_commit(ring);
499 	for (i = 0; i < adev->usec_timeout; i++) {
500 		tmp = RREG32(adev->vcn.inst[ring->me].external.scratch9);
501 		if (tmp == 0xDEADBEEF)
502 			break;
503 		udelay(1);
504 	}
505 
506 	if (i >= adev->usec_timeout)
507 		r = -ETIMEDOUT;
508 
509 	return r;
510 }
511 
512 int amdgpu_vcn_dec_sw_ring_test_ring(struct amdgpu_ring *ring)
513 {
514 	struct amdgpu_device *adev = ring->adev;
515 	uint32_t rptr;
516 	unsigned int i;
517 	int r;
518 
519 	if (amdgpu_sriov_vf(adev))
520 		return 0;
521 
522 	r = amdgpu_ring_alloc(ring, 16);
523 	if (r)
524 		return r;
525 
526 	rptr = amdgpu_ring_get_rptr(ring);
527 
528 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_END);
529 	amdgpu_ring_commit(ring);
530 
531 	for (i = 0; i < adev->usec_timeout; i++) {
532 		if (amdgpu_ring_get_rptr(ring) != rptr)
533 			break;
534 		udelay(1);
535 	}
536 
537 	if (i >= adev->usec_timeout)
538 		r = -ETIMEDOUT;
539 
540 	return r;
541 }
542 
543 static int amdgpu_vcn_dec_send_msg(struct amdgpu_ring *ring,
544 				   struct amdgpu_ib *ib_msg,
545 				   struct dma_fence **fence)
546 {
547 	u64 addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
548 	struct amdgpu_device *adev = ring->adev;
549 	struct dma_fence *f = NULL;
550 	struct amdgpu_job *job;
551 	struct amdgpu_ib *ib;
552 	int i, r;
553 
554 	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
555 				     64, AMDGPU_IB_POOL_DIRECT,
556 				     &job);
557 	if (r)
558 		goto err;
559 
560 	ib = &job->ibs[0];
561 	ib->ptr[0] = PACKET0(adev->vcn.internal.data0, 0);
562 	ib->ptr[1] = addr;
563 	ib->ptr[2] = PACKET0(adev->vcn.internal.data1, 0);
564 	ib->ptr[3] = addr >> 32;
565 	ib->ptr[4] = PACKET0(adev->vcn.internal.cmd, 0);
566 	ib->ptr[5] = 0;
567 	for (i = 6; i < 16; i += 2) {
568 		ib->ptr[i] = PACKET0(adev->vcn.internal.nop, 0);
569 		ib->ptr[i+1] = 0;
570 	}
571 	ib->length_dw = 16;
572 
573 	r = amdgpu_job_submit_direct(job, ring, &f);
574 	if (r)
575 		goto err_free;
576 
577 	amdgpu_ib_free(adev, ib_msg, f);
578 
579 	if (fence)
580 		*fence = dma_fence_get(f);
581 	dma_fence_put(f);
582 
583 	return 0;
584 
585 err_free:
586 	amdgpu_job_free(job);
587 err:
588 	amdgpu_ib_free(adev, ib_msg, f);
589 	return r;
590 }
591 
592 static int amdgpu_vcn_dec_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
593 		struct amdgpu_ib *ib)
594 {
595 	struct amdgpu_device *adev = ring->adev;
596 	uint32_t *msg;
597 	int r, i;
598 
599 	memset(ib, 0, sizeof(*ib));
600 	r = amdgpu_ib_get(adev, NULL, AMDGPU_GPU_PAGE_SIZE * 2,
601 			AMDGPU_IB_POOL_DIRECT,
602 			ib);
603 	if (r)
604 		return r;
605 
606 	msg = (uint32_t *)AMDGPU_GPU_PAGE_ALIGN((unsigned long)ib->ptr);
607 	msg[0] = cpu_to_le32(0x00000028);
608 	msg[1] = cpu_to_le32(0x00000038);
609 	msg[2] = cpu_to_le32(0x00000001);
610 	msg[3] = cpu_to_le32(0x00000000);
611 	msg[4] = cpu_to_le32(handle);
612 	msg[5] = cpu_to_le32(0x00000000);
613 	msg[6] = cpu_to_le32(0x00000001);
614 	msg[7] = cpu_to_le32(0x00000028);
615 	msg[8] = cpu_to_le32(0x00000010);
616 	msg[9] = cpu_to_le32(0x00000000);
617 	msg[10] = cpu_to_le32(0x00000007);
618 	msg[11] = cpu_to_le32(0x00000000);
619 	msg[12] = cpu_to_le32(0x00000780);
620 	msg[13] = cpu_to_le32(0x00000440);
621 	for (i = 14; i < 1024; ++i)
622 		msg[i] = cpu_to_le32(0x0);
623 
624 	return 0;
625 }
626 
627 static int amdgpu_vcn_dec_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
628 					  struct amdgpu_ib *ib)
629 {
630 	struct amdgpu_device *adev = ring->adev;
631 	uint32_t *msg;
632 	int r, i;
633 
634 	memset(ib, 0, sizeof(*ib));
635 	r = amdgpu_ib_get(adev, NULL, AMDGPU_GPU_PAGE_SIZE * 2,
636 			AMDGPU_IB_POOL_DIRECT,
637 			ib);
638 	if (r)
639 		return r;
640 
641 	msg = (uint32_t *)AMDGPU_GPU_PAGE_ALIGN((unsigned long)ib->ptr);
642 	msg[0] = cpu_to_le32(0x00000028);
643 	msg[1] = cpu_to_le32(0x00000018);
644 	msg[2] = cpu_to_le32(0x00000000);
645 	msg[3] = cpu_to_le32(0x00000002);
646 	msg[4] = cpu_to_le32(handle);
647 	msg[5] = cpu_to_le32(0x00000000);
648 	for (i = 6; i < 1024; ++i)
649 		msg[i] = cpu_to_le32(0x0);
650 
651 	return 0;
652 }
653 
654 int amdgpu_vcn_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout)
655 {
656 	struct dma_fence *fence = NULL;
657 	struct amdgpu_ib ib;
658 	long r;
659 
660 	r = amdgpu_vcn_dec_get_create_msg(ring, 1, &ib);
661 	if (r)
662 		goto error;
663 
664 	r = amdgpu_vcn_dec_send_msg(ring, &ib, NULL);
665 	if (r)
666 		goto error;
667 	r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &ib);
668 	if (r)
669 		goto error;
670 
671 	r = amdgpu_vcn_dec_send_msg(ring, &ib, &fence);
672 	if (r)
673 		goto error;
674 
675 	r = dma_fence_wait_timeout(fence, false, timeout);
676 	if (r == 0)
677 		r = -ETIMEDOUT;
678 	else if (r > 0)
679 		r = 0;
680 
681 	dma_fence_put(fence);
682 error:
683 	return r;
684 }
685 
686 static uint32_t *amdgpu_vcn_unified_ring_ib_header(struct amdgpu_ib *ib,
687 						uint32_t ib_pack_in_dw, bool enc)
688 {
689 	uint32_t *ib_checksum;
690 
691 	ib->ptr[ib->length_dw++] = 0x00000010; /* single queue checksum */
692 	ib->ptr[ib->length_dw++] = 0x30000002;
693 	ib_checksum = &ib->ptr[ib->length_dw++];
694 	ib->ptr[ib->length_dw++] = ib_pack_in_dw;
695 
696 	ib->ptr[ib->length_dw++] = 0x00000010; /* engine info */
697 	ib->ptr[ib->length_dw++] = 0x30000001;
698 	ib->ptr[ib->length_dw++] = enc ? 0x2 : 0x3;
699 	ib->ptr[ib->length_dw++] = ib_pack_in_dw * sizeof(uint32_t);
700 
701 	return ib_checksum;
702 }
703 
704 static void amdgpu_vcn_unified_ring_ib_checksum(uint32_t **ib_checksum,
705 						uint32_t ib_pack_in_dw)
706 {
707 	uint32_t i;
708 	uint32_t checksum = 0;
709 
710 	for (i = 0; i < ib_pack_in_dw; i++)
711 		checksum += *(*ib_checksum + 2 + i);
712 
713 	**ib_checksum = checksum;
714 }
715 
716 static int amdgpu_vcn_dec_sw_send_msg(struct amdgpu_ring *ring,
717 				      struct amdgpu_ib *ib_msg,
718 				      struct dma_fence **fence)
719 {
720 	struct amdgpu_vcn_decode_buffer *decode_buffer = NULL;
721 	unsigned int ib_size_dw = 64;
722 	struct amdgpu_device *adev = ring->adev;
723 	struct dma_fence *f = NULL;
724 	struct amdgpu_job *job;
725 	struct amdgpu_ib *ib;
726 	uint64_t addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
727 	uint32_t *ib_checksum;
728 	uint32_t ib_pack_in_dw;
729 	int i, r;
730 
731 	if (adev->vcn.using_unified_queue)
732 		ib_size_dw += 8;
733 
734 	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
735 				     ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
736 				     &job);
737 	if (r)
738 		goto err;
739 
740 	ib = &job->ibs[0];
741 	ib->length_dw = 0;
742 
743 	/* single queue headers */
744 	if (adev->vcn.using_unified_queue) {
745 		ib_pack_in_dw = sizeof(struct amdgpu_vcn_decode_buffer) / sizeof(uint32_t)
746 						+ 4 + 2; /* engine info + decoding ib in dw */
747 		ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, ib_pack_in_dw, false);
748 	}
749 
750 	ib->ptr[ib->length_dw++] = sizeof(struct amdgpu_vcn_decode_buffer) + 8;
751 	ib->ptr[ib->length_dw++] = cpu_to_le32(AMDGPU_VCN_IB_FLAG_DECODE_BUFFER);
752 	decode_buffer = (struct amdgpu_vcn_decode_buffer *)&(ib->ptr[ib->length_dw]);
753 	ib->length_dw += sizeof(struct amdgpu_vcn_decode_buffer) / 4;
754 	memset(decode_buffer, 0, sizeof(struct amdgpu_vcn_decode_buffer));
755 
756 	decode_buffer->valid_buf_flag |= cpu_to_le32(AMDGPU_VCN_CMD_FLAG_MSG_BUFFER);
757 	decode_buffer->msg_buffer_address_hi = cpu_to_le32(addr >> 32);
758 	decode_buffer->msg_buffer_address_lo = cpu_to_le32(addr);
759 
760 	for (i = ib->length_dw; i < ib_size_dw; ++i)
761 		ib->ptr[i] = 0x0;
762 
763 	if (adev->vcn.using_unified_queue)
764 		amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, ib_pack_in_dw);
765 
766 	r = amdgpu_job_submit_direct(job, ring, &f);
767 	if (r)
768 		goto err_free;
769 
770 	amdgpu_ib_free(adev, ib_msg, f);
771 
772 	if (fence)
773 		*fence = dma_fence_get(f);
774 	dma_fence_put(f);
775 
776 	return 0;
777 
778 err_free:
779 	amdgpu_job_free(job);
780 err:
781 	amdgpu_ib_free(adev, ib_msg, f);
782 	return r;
783 }
784 
785 int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout)
786 {
787 	struct dma_fence *fence = NULL;
788 	struct amdgpu_ib ib;
789 	long r;
790 
791 	r = amdgpu_vcn_dec_get_create_msg(ring, 1, &ib);
792 	if (r)
793 		goto error;
794 
795 	r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, NULL);
796 	if (r)
797 		goto error;
798 	r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &ib);
799 	if (r)
800 		goto error;
801 
802 	r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, &fence);
803 	if (r)
804 		goto error;
805 
806 	r = dma_fence_wait_timeout(fence, false, timeout);
807 	if (r == 0)
808 		r = -ETIMEDOUT;
809 	else if (r > 0)
810 		r = 0;
811 
812 	dma_fence_put(fence);
813 error:
814 	return r;
815 }
816 
817 int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring)
818 {
819 	struct amdgpu_device *adev = ring->adev;
820 	uint32_t rptr;
821 	unsigned int i;
822 	int r;
823 
824 	if (amdgpu_sriov_vf(adev))
825 		return 0;
826 
827 	r = amdgpu_ring_alloc(ring, 16);
828 	if (r)
829 		return r;
830 
831 	rptr = amdgpu_ring_get_rptr(ring);
832 
833 	amdgpu_ring_write(ring, VCN_ENC_CMD_END);
834 	amdgpu_ring_commit(ring);
835 
836 	for (i = 0; i < adev->usec_timeout; i++) {
837 		if (amdgpu_ring_get_rptr(ring) != rptr)
838 			break;
839 		udelay(1);
840 	}
841 
842 	if (i >= adev->usec_timeout)
843 		r = -ETIMEDOUT;
844 
845 	return r;
846 }
847 
848 static int amdgpu_vcn_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
849 					 struct amdgpu_ib *ib_msg,
850 					 struct dma_fence **fence)
851 {
852 	unsigned int ib_size_dw = 16;
853 	struct amdgpu_device *adev = ring->adev;
854 	struct amdgpu_job *job;
855 	struct amdgpu_ib *ib;
856 	struct dma_fence *f = NULL;
857 	uint32_t *ib_checksum = NULL;
858 	uint64_t addr;
859 	int i, r;
860 
861 	if (adev->vcn.using_unified_queue)
862 		ib_size_dw += 8;
863 
864 	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
865 				     ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
866 				     &job);
867 	if (r)
868 		return r;
869 
870 	ib = &job->ibs[0];
871 	addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
872 
873 	ib->length_dw = 0;
874 
875 	if (adev->vcn.using_unified_queue)
876 		ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, 0x11, true);
877 
878 	ib->ptr[ib->length_dw++] = 0x00000018;
879 	ib->ptr[ib->length_dw++] = 0x00000001; /* session info */
880 	ib->ptr[ib->length_dw++] = handle;
881 	ib->ptr[ib->length_dw++] = upper_32_bits(addr);
882 	ib->ptr[ib->length_dw++] = addr;
883 	ib->ptr[ib->length_dw++] = 0x00000000;
884 
885 	ib->ptr[ib->length_dw++] = 0x00000014;
886 	ib->ptr[ib->length_dw++] = 0x00000002; /* task info */
887 	ib->ptr[ib->length_dw++] = 0x0000001c;
888 	ib->ptr[ib->length_dw++] = 0x00000000;
889 	ib->ptr[ib->length_dw++] = 0x00000000;
890 
891 	ib->ptr[ib->length_dw++] = 0x00000008;
892 	ib->ptr[ib->length_dw++] = 0x08000001; /* op initialize */
893 
894 	for (i = ib->length_dw; i < ib_size_dw; ++i)
895 		ib->ptr[i] = 0x0;
896 
897 	if (adev->vcn.using_unified_queue)
898 		amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, 0x11);
899 
900 	r = amdgpu_job_submit_direct(job, ring, &f);
901 	if (r)
902 		goto err;
903 
904 	if (fence)
905 		*fence = dma_fence_get(f);
906 	dma_fence_put(f);
907 
908 	return 0;
909 
910 err:
911 	amdgpu_job_free(job);
912 	return r;
913 }
914 
915 static int amdgpu_vcn_enc_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
916 					  struct amdgpu_ib *ib_msg,
917 					  struct dma_fence **fence)
918 {
919 	unsigned int ib_size_dw = 16;
920 	struct amdgpu_device *adev = ring->adev;
921 	struct amdgpu_job *job;
922 	struct amdgpu_ib *ib;
923 	struct dma_fence *f = NULL;
924 	uint32_t *ib_checksum = NULL;
925 	uint64_t addr;
926 	int i, r;
927 
928 	if (adev->vcn.using_unified_queue)
929 		ib_size_dw += 8;
930 
931 	r = amdgpu_job_alloc_with_ib(ring->adev, NULL, NULL,
932 				     ib_size_dw * 4, AMDGPU_IB_POOL_DIRECT,
933 				     &job);
934 	if (r)
935 		return r;
936 
937 	ib = &job->ibs[0];
938 	addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
939 
940 	ib->length_dw = 0;
941 
942 	if (adev->vcn.using_unified_queue)
943 		ib_checksum = amdgpu_vcn_unified_ring_ib_header(ib, 0x11, true);
944 
945 	ib->ptr[ib->length_dw++] = 0x00000018;
946 	ib->ptr[ib->length_dw++] = 0x00000001;
947 	ib->ptr[ib->length_dw++] = handle;
948 	ib->ptr[ib->length_dw++] = upper_32_bits(addr);
949 	ib->ptr[ib->length_dw++] = addr;
950 	ib->ptr[ib->length_dw++] = 0x00000000;
951 
952 	ib->ptr[ib->length_dw++] = 0x00000014;
953 	ib->ptr[ib->length_dw++] = 0x00000002;
954 	ib->ptr[ib->length_dw++] = 0x0000001c;
955 	ib->ptr[ib->length_dw++] = 0x00000000;
956 	ib->ptr[ib->length_dw++] = 0x00000000;
957 
958 	ib->ptr[ib->length_dw++] = 0x00000008;
959 	ib->ptr[ib->length_dw++] = 0x08000002; /* op close session */
960 
961 	for (i = ib->length_dw; i < ib_size_dw; ++i)
962 		ib->ptr[i] = 0x0;
963 
964 	if (adev->vcn.using_unified_queue)
965 		amdgpu_vcn_unified_ring_ib_checksum(&ib_checksum, 0x11);
966 
967 	r = amdgpu_job_submit_direct(job, ring, &f);
968 	if (r)
969 		goto err;
970 
971 	if (fence)
972 		*fence = dma_fence_get(f);
973 	dma_fence_put(f);
974 
975 	return 0;
976 
977 err:
978 	amdgpu_job_free(job);
979 	return r;
980 }
981 
982 int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout)
983 {
984 	struct amdgpu_device *adev = ring->adev;
985 	struct dma_fence *fence = NULL;
986 	struct amdgpu_ib ib;
987 	long r;
988 
989 	memset(&ib, 0, sizeof(ib));
990 	r = amdgpu_ib_get(adev, NULL, (128 << 10) + AMDGPU_GPU_PAGE_SIZE,
991 			AMDGPU_IB_POOL_DIRECT,
992 			&ib);
993 	if (r)
994 		return r;
995 
996 	r = amdgpu_vcn_enc_get_create_msg(ring, 1, &ib, NULL);
997 	if (r)
998 		goto error;
999 
1000 	r = amdgpu_vcn_enc_get_destroy_msg(ring, 1, &ib, &fence);
1001 	if (r)
1002 		goto error;
1003 
1004 	r = dma_fence_wait_timeout(fence, false, timeout);
1005 	if (r == 0)
1006 		r = -ETIMEDOUT;
1007 	else if (r > 0)
1008 		r = 0;
1009 
1010 error:
1011 	amdgpu_ib_free(adev, &ib, fence);
1012 	dma_fence_put(fence);
1013 
1014 	return r;
1015 }
1016 
1017 int amdgpu_vcn_unified_ring_test_ib(struct amdgpu_ring *ring, long timeout)
1018 {
1019 	struct amdgpu_device *adev = ring->adev;
1020 	long r;
1021 
1022 	if (amdgpu_ip_version(adev, UVD_HWIP, 0) != IP_VERSION(4, 0, 3)) {
1023 		r = amdgpu_vcn_enc_ring_test_ib(ring, timeout);
1024 		if (r)
1025 			goto error;
1026 	}
1027 
1028 	r =  amdgpu_vcn_dec_sw_ring_test_ib(ring, timeout);
1029 
1030 error:
1031 	return r;
1032 }
1033 
1034 enum amdgpu_ring_priority_level amdgpu_vcn_get_enc_ring_prio(int ring)
1035 {
1036 	switch (ring) {
1037 	case 0:
1038 		return AMDGPU_RING_PRIO_0;
1039 	case 1:
1040 		return AMDGPU_RING_PRIO_1;
1041 	case 2:
1042 		return AMDGPU_RING_PRIO_2;
1043 	default:
1044 		return AMDGPU_RING_PRIO_0;
1045 	}
1046 }
1047 
1048 void amdgpu_vcn_setup_ucode(struct amdgpu_device *adev)
1049 {
1050 	int i;
1051 	unsigned int idx;
1052 
1053 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1054 		const struct common_firmware_header *hdr;
1055 
1056 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1057 			if (adev->vcn.harvest_config & (1 << i))
1058 				continue;
1059 
1060 			hdr = (const struct common_firmware_header *)adev->vcn.fw[i]->data;
1061 			/* currently only support 2 FW instances */
1062 			if (i >= 2) {
1063 				dev_info(adev->dev, "More then 2 VCN FW instances!\n");
1064 				break;
1065 			}
1066 			idx = AMDGPU_UCODE_ID_VCN + i;
1067 			adev->firmware.ucode[idx].ucode_id = idx;
1068 			adev->firmware.ucode[idx].fw = adev->vcn.fw[i];
1069 			adev->firmware.fw_size +=
1070 				ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
1071 
1072 			if (amdgpu_ip_version(adev, UVD_HWIP, 0) ==
1073 			    IP_VERSION(4, 0, 3))
1074 				break;
1075 		}
1076 	}
1077 }
1078 
1079 /*
1080  * debugfs for mapping vcn firmware log buffer.
1081  */
1082 #if defined(CONFIG_DEBUG_FS)
1083 static ssize_t amdgpu_debugfs_vcn_fwlog_read(struct file *f, char __user *buf,
1084 					     size_t size, loff_t *pos)
1085 {
1086 	struct amdgpu_vcn_inst *vcn;
1087 	void *log_buf;
1088 	volatile struct amdgpu_vcn_fwlog *plog;
1089 	unsigned int read_pos, write_pos, available, i, read_bytes = 0;
1090 	unsigned int read_num[2] = {0};
1091 
1092 	vcn = file_inode(f)->i_private;
1093 	if (!vcn)
1094 		return -ENODEV;
1095 
1096 	if (!vcn->fw_shared.cpu_addr || !amdgpu_vcnfw_log)
1097 		return -EFAULT;
1098 
1099 	log_buf = vcn->fw_shared.cpu_addr + vcn->fw_shared.mem_size;
1100 
1101 	plog = (volatile struct amdgpu_vcn_fwlog *)log_buf;
1102 	read_pos = plog->rptr;
1103 	write_pos = plog->wptr;
1104 
1105 	if (read_pos > AMDGPU_VCNFW_LOG_SIZE || write_pos > AMDGPU_VCNFW_LOG_SIZE)
1106 		return -EFAULT;
1107 
1108 	if (!size || (read_pos == write_pos))
1109 		return 0;
1110 
1111 	if (write_pos > read_pos) {
1112 		available = write_pos - read_pos;
1113 		read_num[0] = min_t(size_t, size, available);
1114 	} else {
1115 		read_num[0] = AMDGPU_VCNFW_LOG_SIZE - read_pos;
1116 		available = read_num[0] + write_pos - plog->header_size;
1117 		if (size > available)
1118 			read_num[1] = write_pos - plog->header_size;
1119 		else if (size > read_num[0])
1120 			read_num[1] = size - read_num[0];
1121 		else
1122 			read_num[0] = size;
1123 	}
1124 
1125 	for (i = 0; i < 2; i++) {
1126 		if (read_num[i]) {
1127 			if (read_pos == AMDGPU_VCNFW_LOG_SIZE)
1128 				read_pos = plog->header_size;
1129 			if (read_num[i] == copy_to_user((buf + read_bytes),
1130 							(log_buf + read_pos), read_num[i]))
1131 				return -EFAULT;
1132 
1133 			read_bytes += read_num[i];
1134 			read_pos += read_num[i];
1135 		}
1136 	}
1137 
1138 	plog->rptr = read_pos;
1139 	*pos += read_bytes;
1140 	return read_bytes;
1141 }
1142 
1143 static const struct file_operations amdgpu_debugfs_vcnfwlog_fops = {
1144 	.owner = THIS_MODULE,
1145 	.read = amdgpu_debugfs_vcn_fwlog_read,
1146 	.llseek = default_llseek
1147 };
1148 #endif
1149 
1150 void amdgpu_debugfs_vcn_fwlog_init(struct amdgpu_device *adev, uint8_t i,
1151 				   struct amdgpu_vcn_inst *vcn)
1152 {
1153 #if defined(CONFIG_DEBUG_FS)
1154 	struct drm_minor *minor = adev_to_drm(adev)->primary;
1155 	struct dentry *root = minor->debugfs_root;
1156 	char name[32];
1157 
1158 	sprintf(name, "amdgpu_vcn_%d_fwlog", i);
1159 	debugfs_create_file_size(name, S_IFREG | 0444, root, vcn,
1160 				 &amdgpu_debugfs_vcnfwlog_fops,
1161 				 AMDGPU_VCNFW_LOG_SIZE);
1162 #endif
1163 }
1164 
1165 void amdgpu_vcn_fwlog_init(struct amdgpu_vcn_inst *vcn)
1166 {
1167 #if defined(CONFIG_DEBUG_FS)
1168 	volatile uint32_t *flag = vcn->fw_shared.cpu_addr;
1169 	void *fw_log_cpu_addr = vcn->fw_shared.cpu_addr + vcn->fw_shared.mem_size;
1170 	uint64_t fw_log_gpu_addr = vcn->fw_shared.gpu_addr + vcn->fw_shared.mem_size;
1171 	volatile struct amdgpu_vcn_fwlog *log_buf = fw_log_cpu_addr;
1172 	volatile struct amdgpu_fw_shared_fw_logging *fw_log = vcn->fw_shared.cpu_addr
1173 							 + vcn->fw_shared.log_offset;
1174 	*flag |= cpu_to_le32(AMDGPU_VCN_FW_LOGGING_FLAG);
1175 	fw_log->is_enabled = 1;
1176 	fw_log->addr_lo = cpu_to_le32(fw_log_gpu_addr & 0xFFFFFFFF);
1177 	fw_log->addr_hi = cpu_to_le32(fw_log_gpu_addr >> 32);
1178 	fw_log->size = cpu_to_le32(AMDGPU_VCNFW_LOG_SIZE);
1179 
1180 	log_buf->header_size = sizeof(struct amdgpu_vcn_fwlog);
1181 	log_buf->buffer_size = AMDGPU_VCNFW_LOG_SIZE;
1182 	log_buf->rptr = log_buf->header_size;
1183 	log_buf->wptr = log_buf->header_size;
1184 	log_buf->wrapped = 0;
1185 #endif
1186 }
1187 
1188 int amdgpu_vcn_process_poison_irq(struct amdgpu_device *adev,
1189 				struct amdgpu_irq_src *source,
1190 				struct amdgpu_iv_entry *entry)
1191 {
1192 	struct ras_common_if *ras_if = adev->vcn.ras_if;
1193 	struct ras_dispatch_if ih_data = {
1194 		.entry = entry,
1195 	};
1196 
1197 	if (!ras_if)
1198 		return 0;
1199 
1200 	if (!amdgpu_sriov_vf(adev)) {
1201 		ih_data.head = *ras_if;
1202 		amdgpu_ras_interrupt_dispatch(adev, &ih_data);
1203 	} else {
1204 		if (adev->virt.ops && adev->virt.ops->ras_poison_handler)
1205 			adev->virt.ops->ras_poison_handler(adev, ras_if->block);
1206 		else
1207 			dev_warn(adev->dev,
1208 				"No ras_poison_handler interface in SRIOV for VCN!\n");
1209 	}
1210 
1211 	return 0;
1212 }
1213 
1214 int amdgpu_vcn_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block)
1215 {
1216 	int r, i;
1217 
1218 	r = amdgpu_ras_block_late_init(adev, ras_block);
1219 	if (r)
1220 		return r;
1221 
1222 	if (amdgpu_ras_is_supported(adev, ras_block->block)) {
1223 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1224 			if (adev->vcn.harvest_config & (1 << i) ||
1225 			    !adev->vcn.inst[i].ras_poison_irq.funcs)
1226 				continue;
1227 
1228 			r = amdgpu_irq_get(adev, &adev->vcn.inst[i].ras_poison_irq, 0);
1229 			if (r)
1230 				goto late_fini;
1231 		}
1232 	}
1233 	return 0;
1234 
1235 late_fini:
1236 	amdgpu_ras_block_late_fini(adev, ras_block);
1237 	return r;
1238 }
1239 
1240 int amdgpu_vcn_ras_sw_init(struct amdgpu_device *adev)
1241 {
1242 	int err;
1243 	struct amdgpu_vcn_ras *ras;
1244 
1245 	if (!adev->vcn.ras)
1246 		return 0;
1247 
1248 	ras = adev->vcn.ras;
1249 	err = amdgpu_ras_register_ras_block(adev, &ras->ras_block);
1250 	if (err) {
1251 		dev_err(adev->dev, "Failed to register vcn ras block!\n");
1252 		return err;
1253 	}
1254 
1255 	strcpy(ras->ras_block.ras_comm.name, "vcn");
1256 	ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__VCN;
1257 	ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__POISON;
1258 	adev->vcn.ras_if = &ras->ras_block.ras_comm;
1259 
1260 	if (!ras->ras_block.ras_late_init)
1261 		ras->ras_block.ras_late_init = amdgpu_vcn_ras_late_init;
1262 
1263 	return 0;
1264 }
1265 
1266 int amdgpu_vcn_psp_update_sram(struct amdgpu_device *adev, int inst_idx,
1267 			       enum AMDGPU_UCODE_ID ucode_id)
1268 {
1269 	struct amdgpu_firmware_info ucode = {
1270 		.ucode_id = (ucode_id ? ucode_id :
1271 			    (inst_idx ? AMDGPU_UCODE_ID_VCN1_RAM :
1272 					AMDGPU_UCODE_ID_VCN0_RAM)),
1273 		.mc_addr = adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
1274 		.ucode_size = ((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
1275 			      (uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr),
1276 	};
1277 
1278 	return psp_execute_ip_fw_load(&adev->psp, &ucode);
1279 }
1280