xref: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c (revision 3e44c471a2dab210f7e9b1e5f7d4d54d52df59eb)
1 /*
2  * Copyright 2013 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  * Authors: Christian König <christian.koenig@amd.com>
26  */
27 
28 #include <linux/firmware.h>
29 #include <linux/module.h>
30 #include <drm/drmP.h>
31 #include <drm/drm.h>
32 
33 #include "amdgpu.h"
34 #include "amdgpu_pm.h"
35 #include "amdgpu_vce.h"
36 #include "cikd.h"
37 
38 /* 1 second timeout */
39 #define VCE_IDLE_TIMEOUT_MS	1000
40 
41 /* Firmware Names */
42 #ifdef CONFIG_DRM_AMDGPU_CIK
43 #define FIRMWARE_BONAIRE	"radeon/bonaire_vce.bin"
44 #define FIRMWARE_KABINI 	"radeon/kabini_vce.bin"
45 #define FIRMWARE_KAVERI 	"radeon/kaveri_vce.bin"
46 #define FIRMWARE_HAWAII 	"radeon/hawaii_vce.bin"
47 #define FIRMWARE_MULLINS	"radeon/mullins_vce.bin"
48 #endif
49 #define FIRMWARE_TONGA		"amdgpu/tonga_vce.bin"
50 #define FIRMWARE_CARRIZO	"amdgpu/carrizo_vce.bin"
51 
52 #ifdef CONFIG_DRM_AMDGPU_CIK
53 MODULE_FIRMWARE(FIRMWARE_BONAIRE);
54 MODULE_FIRMWARE(FIRMWARE_KABINI);
55 MODULE_FIRMWARE(FIRMWARE_KAVERI);
56 MODULE_FIRMWARE(FIRMWARE_HAWAII);
57 MODULE_FIRMWARE(FIRMWARE_MULLINS);
58 #endif
59 MODULE_FIRMWARE(FIRMWARE_TONGA);
60 MODULE_FIRMWARE(FIRMWARE_CARRIZO);
61 
62 static void amdgpu_vce_idle_work_handler(struct work_struct *work);
63 
64 /**
65  * amdgpu_vce_init - allocate memory, load vce firmware
66  *
67  * @adev: amdgpu_device pointer
68  *
69  * First step to get VCE online, allocate memory and load the firmware
70  */
71 int amdgpu_vce_sw_init(struct amdgpu_device *adev, unsigned long size)
72 {
73 	const char *fw_name;
74 	const struct common_firmware_header *hdr;
75 	unsigned ucode_version, version_major, version_minor, binary_id;
76 	int i, r;
77 
78 	INIT_DELAYED_WORK(&adev->vce.idle_work, amdgpu_vce_idle_work_handler);
79 
80 	switch (adev->asic_type) {
81 #ifdef CONFIG_DRM_AMDGPU_CIK
82 	case CHIP_BONAIRE:
83 		fw_name = FIRMWARE_BONAIRE;
84 		break;
85 	case CHIP_KAVERI:
86 		fw_name = FIRMWARE_KAVERI;
87 		break;
88 	case CHIP_KABINI:
89 		fw_name = FIRMWARE_KABINI;
90 		break;
91 	case CHIP_HAWAII:
92 		fw_name = FIRMWARE_HAWAII;
93 		break;
94 	case CHIP_MULLINS:
95 		fw_name = FIRMWARE_MULLINS;
96 		break;
97 #endif
98 	case CHIP_TONGA:
99 		fw_name = FIRMWARE_TONGA;
100 		break;
101 	case CHIP_CARRIZO:
102 		fw_name = FIRMWARE_CARRIZO;
103 		break;
104 
105 	default:
106 		return -EINVAL;
107 	}
108 
109 	r = request_firmware(&adev->vce.fw, fw_name, adev->dev);
110 	if (r) {
111 		dev_err(adev->dev, "amdgpu_vce: Can't load firmware \"%s\"\n",
112 			fw_name);
113 		return r;
114 	}
115 
116 	r = amdgpu_ucode_validate(adev->vce.fw);
117 	if (r) {
118 		dev_err(adev->dev, "amdgpu_vce: Can't validate firmware \"%s\"\n",
119 			fw_name);
120 		release_firmware(adev->vce.fw);
121 		adev->vce.fw = NULL;
122 		return r;
123 	}
124 
125 	hdr = (const struct common_firmware_header *)adev->vce.fw->data;
126 
127 	ucode_version = le32_to_cpu(hdr->ucode_version);
128 	version_major = (ucode_version >> 20) & 0xfff;
129 	version_minor = (ucode_version >> 8) & 0xfff;
130 	binary_id = ucode_version & 0xff;
131 	DRM_INFO("Found VCE firmware Version: %hhd.%hhd Binary ID: %hhd\n",
132 		version_major, version_minor, binary_id);
133 	adev->vce.fw_version = ((version_major << 24) | (version_minor << 16) |
134 				(binary_id << 8));
135 
136 	/* allocate firmware, stack and heap BO */
137 
138 	r = amdgpu_bo_create(adev, size, PAGE_SIZE, true,
139 			     AMDGPU_GEM_DOMAIN_VRAM, 0, NULL, &adev->vce.vcpu_bo);
140 	if (r) {
141 		dev_err(adev->dev, "(%d) failed to allocate VCE bo\n", r);
142 		return r;
143 	}
144 
145 	r = amdgpu_bo_reserve(adev->vce.vcpu_bo, false);
146 	if (r) {
147 		amdgpu_bo_unref(&adev->vce.vcpu_bo);
148 		dev_err(adev->dev, "(%d) failed to reserve VCE bo\n", r);
149 		return r;
150 	}
151 
152 	r = amdgpu_bo_pin(adev->vce.vcpu_bo, AMDGPU_GEM_DOMAIN_VRAM,
153 			  &adev->vce.gpu_addr);
154 	amdgpu_bo_unreserve(adev->vce.vcpu_bo);
155 	if (r) {
156 		amdgpu_bo_unref(&adev->vce.vcpu_bo);
157 		dev_err(adev->dev, "(%d) VCE bo pin failed\n", r);
158 		return r;
159 	}
160 
161 	for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) {
162 		atomic_set(&adev->vce.handles[i], 0);
163 		adev->vce.filp[i] = NULL;
164 	}
165 
166 	return 0;
167 }
168 
169 /**
170  * amdgpu_vce_fini - free memory
171  *
172  * @adev: amdgpu_device pointer
173  *
174  * Last step on VCE teardown, free firmware memory
175  */
176 int amdgpu_vce_sw_fini(struct amdgpu_device *adev)
177 {
178 	if (adev->vce.vcpu_bo == NULL)
179 		return 0;
180 
181 	amdgpu_bo_unref(&adev->vce.vcpu_bo);
182 
183 	amdgpu_ring_fini(&adev->vce.ring[0]);
184 	amdgpu_ring_fini(&adev->vce.ring[1]);
185 
186 	release_firmware(adev->vce.fw);
187 
188 	return 0;
189 }
190 
191 /**
192  * amdgpu_vce_suspend - unpin VCE fw memory
193  *
194  * @adev: amdgpu_device pointer
195  *
196  */
197 int amdgpu_vce_suspend(struct amdgpu_device *adev)
198 {
199 	int i;
200 
201 	if (adev->vce.vcpu_bo == NULL)
202 		return 0;
203 
204 	for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i)
205 		if (atomic_read(&adev->vce.handles[i]))
206 			break;
207 
208 	if (i == AMDGPU_MAX_VCE_HANDLES)
209 		return 0;
210 
211 	/* TODO: suspending running encoding sessions isn't supported */
212 	return -EINVAL;
213 }
214 
215 /**
216  * amdgpu_vce_resume - pin VCE fw memory
217  *
218  * @adev: amdgpu_device pointer
219  *
220  */
221 int amdgpu_vce_resume(struct amdgpu_device *adev)
222 {
223 	void *cpu_addr;
224 	const struct common_firmware_header *hdr;
225 	unsigned offset;
226 	int r;
227 
228 	if (adev->vce.vcpu_bo == NULL)
229 		return -EINVAL;
230 
231 	r = amdgpu_bo_reserve(adev->vce.vcpu_bo, false);
232 	if (r) {
233 		dev_err(adev->dev, "(%d) failed to reserve VCE bo\n", r);
234 		return r;
235 	}
236 
237 	r = amdgpu_bo_kmap(adev->vce.vcpu_bo, &cpu_addr);
238 	if (r) {
239 		amdgpu_bo_unreserve(adev->vce.vcpu_bo);
240 		dev_err(adev->dev, "(%d) VCE map failed\n", r);
241 		return r;
242 	}
243 
244 	hdr = (const struct common_firmware_header *)adev->vce.fw->data;
245 	offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
246 	memcpy(cpu_addr, (adev->vce.fw->data) + offset,
247 		(adev->vce.fw->size) - offset);
248 
249 	amdgpu_bo_kunmap(adev->vce.vcpu_bo);
250 
251 	amdgpu_bo_unreserve(adev->vce.vcpu_bo);
252 
253 	return 0;
254 }
255 
256 /**
257  * amdgpu_vce_idle_work_handler - power off VCE
258  *
259  * @work: pointer to work structure
260  *
261  * power of VCE when it's not used any more
262  */
263 static void amdgpu_vce_idle_work_handler(struct work_struct *work)
264 {
265 	struct amdgpu_device *adev =
266 		container_of(work, struct amdgpu_device, vce.idle_work.work);
267 
268 	if ((amdgpu_fence_count_emitted(&adev->vce.ring[0]) == 0) &&
269 	    (amdgpu_fence_count_emitted(&adev->vce.ring[1]) == 0)) {
270 		if (adev->pm.dpm_enabled) {
271 			amdgpu_dpm_enable_vce(adev, false);
272 		} else {
273 			amdgpu_asic_set_vce_clocks(adev, 0, 0);
274 		}
275 	} else {
276 		schedule_delayed_work(&adev->vce.idle_work,
277 				      msecs_to_jiffies(VCE_IDLE_TIMEOUT_MS));
278 	}
279 }
280 
281 /**
282  * amdgpu_vce_note_usage - power up VCE
283  *
284  * @adev: amdgpu_device pointer
285  *
286  * Make sure VCE is powerd up when we want to use it
287  */
288 static void amdgpu_vce_note_usage(struct amdgpu_device *adev)
289 {
290 	bool streams_changed = false;
291 	bool set_clocks = !cancel_delayed_work_sync(&adev->vce.idle_work);
292 	set_clocks &= schedule_delayed_work(&adev->vce.idle_work,
293 					    msecs_to_jiffies(VCE_IDLE_TIMEOUT_MS));
294 
295 	if (adev->pm.dpm_enabled) {
296 		/* XXX figure out if the streams changed */
297 		streams_changed = false;
298 	}
299 
300 	if (set_clocks || streams_changed) {
301 		if (adev->pm.dpm_enabled) {
302 			amdgpu_dpm_enable_vce(adev, true);
303 		} else {
304 			amdgpu_asic_set_vce_clocks(adev, 53300, 40000);
305 		}
306 	}
307 }
308 
309 /**
310  * amdgpu_vce_free_handles - free still open VCE handles
311  *
312  * @adev: amdgpu_device pointer
313  * @filp: drm file pointer
314  *
315  * Close all VCE handles still open by this file pointer
316  */
317 void amdgpu_vce_free_handles(struct amdgpu_device *adev, struct drm_file *filp)
318 {
319 	struct amdgpu_ring *ring = &adev->vce.ring[0];
320 	int i, r;
321 	for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) {
322 		uint32_t handle = atomic_read(&adev->vce.handles[i]);
323 		if (!handle || adev->vce.filp[i] != filp)
324 			continue;
325 
326 		amdgpu_vce_note_usage(adev);
327 
328 		r = amdgpu_vce_get_destroy_msg(ring, handle, NULL);
329 		if (r)
330 			DRM_ERROR("Error destroying VCE handle (%d)!\n", r);
331 
332 		adev->vce.filp[i] = NULL;
333 		atomic_set(&adev->vce.handles[i], 0);
334 	}
335 }
336 
337 /**
338  * amdgpu_vce_get_create_msg - generate a VCE create msg
339  *
340  * @adev: amdgpu_device pointer
341  * @ring: ring we should submit the msg to
342  * @handle: VCE session handle to use
343  * @fence: optional fence to return
344  *
345  * Open up a stream for HW test
346  */
347 int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
348 			      struct amdgpu_fence **fence)
349 {
350 	const unsigned ib_size_dw = 1024;
351 	struct amdgpu_ib ib;
352 	uint64_t dummy;
353 	int i, r;
354 
355 	r = amdgpu_ib_get(ring, NULL, ib_size_dw * 4, &ib);
356 	if (r) {
357 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
358 		return r;
359 	}
360 
361 	dummy = ib.gpu_addr + 1024;
362 
363 	/* stitch together an VCE create msg */
364 	ib.length_dw = 0;
365 	ib.ptr[ib.length_dw++] = 0x0000000c; /* len */
366 	ib.ptr[ib.length_dw++] = 0x00000001; /* session cmd */
367 	ib.ptr[ib.length_dw++] = handle;
368 
369 	ib.ptr[ib.length_dw++] = 0x00000030; /* len */
370 	ib.ptr[ib.length_dw++] = 0x01000001; /* create cmd */
371 	ib.ptr[ib.length_dw++] = 0x00000000;
372 	ib.ptr[ib.length_dw++] = 0x00000042;
373 	ib.ptr[ib.length_dw++] = 0x0000000a;
374 	ib.ptr[ib.length_dw++] = 0x00000001;
375 	ib.ptr[ib.length_dw++] = 0x00000080;
376 	ib.ptr[ib.length_dw++] = 0x00000060;
377 	ib.ptr[ib.length_dw++] = 0x00000100;
378 	ib.ptr[ib.length_dw++] = 0x00000100;
379 	ib.ptr[ib.length_dw++] = 0x0000000c;
380 	ib.ptr[ib.length_dw++] = 0x00000000;
381 
382 	ib.ptr[ib.length_dw++] = 0x00000014; /* len */
383 	ib.ptr[ib.length_dw++] = 0x05000005; /* feedback buffer */
384 	ib.ptr[ib.length_dw++] = upper_32_bits(dummy);
385 	ib.ptr[ib.length_dw++] = dummy;
386 	ib.ptr[ib.length_dw++] = 0x00000001;
387 
388 	for (i = ib.length_dw; i < ib_size_dw; ++i)
389 		ib.ptr[i] = 0x0;
390 
391 	r = amdgpu_ib_schedule(ring->adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
392 	if (r) {
393 		DRM_ERROR("amdgpu: failed to schedule ib (%d).\n", r);
394 	}
395 
396 	if (fence)
397 		*fence = amdgpu_fence_ref(ib.fence);
398 
399 	amdgpu_ib_free(ring->adev, &ib);
400 
401 	return r;
402 }
403 
404 /**
405  * amdgpu_vce_get_destroy_msg - generate a VCE destroy msg
406  *
407  * @adev: amdgpu_device pointer
408  * @ring: ring we should submit the msg to
409  * @handle: VCE session handle to use
410  * @fence: optional fence to return
411  *
412  * Close up a stream for HW test or if userspace failed to do so
413  */
414 int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
415 			       struct amdgpu_fence **fence)
416 {
417 	const unsigned ib_size_dw = 1024;
418 	struct amdgpu_ib ib;
419 	uint64_t dummy;
420 	int i, r;
421 
422 	r = amdgpu_ib_get(ring, NULL, ib_size_dw * 4, &ib);
423 	if (r) {
424 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
425 		return r;
426 	}
427 
428 	dummy = ib.gpu_addr + 1024;
429 
430 	/* stitch together an VCE destroy msg */
431 	ib.length_dw = 0;
432 	ib.ptr[ib.length_dw++] = 0x0000000c; /* len */
433 	ib.ptr[ib.length_dw++] = 0x00000001; /* session cmd */
434 	ib.ptr[ib.length_dw++] = handle;
435 
436 	ib.ptr[ib.length_dw++] = 0x00000014; /* len */
437 	ib.ptr[ib.length_dw++] = 0x05000005; /* feedback buffer */
438 	ib.ptr[ib.length_dw++] = upper_32_bits(dummy);
439 	ib.ptr[ib.length_dw++] = dummy;
440 	ib.ptr[ib.length_dw++] = 0x00000001;
441 
442 	ib.ptr[ib.length_dw++] = 0x00000008; /* len */
443 	ib.ptr[ib.length_dw++] = 0x02000001; /* destroy cmd */
444 
445 	for (i = ib.length_dw; i < ib_size_dw; ++i)
446 		ib.ptr[i] = 0x0;
447 
448 	r = amdgpu_ib_schedule(ring->adev, 1, &ib, AMDGPU_FENCE_OWNER_UNDEFINED);
449 	if (r) {
450 		DRM_ERROR("amdgpu: failed to schedule ib (%d).\n", r);
451 	}
452 
453 	if (fence)
454 		*fence = amdgpu_fence_ref(ib.fence);
455 
456 	amdgpu_ib_free(ring->adev, &ib);
457 
458 	return r;
459 }
460 
461 /**
462  * amdgpu_vce_cs_reloc - command submission relocation
463  *
464  * @p: parser context
465  * @lo: address of lower dword
466  * @hi: address of higher dword
467  *
468  * Patch relocation inside command stream with real buffer address
469  */
470 int amdgpu_vce_cs_reloc(struct amdgpu_cs_parser *p, uint32_t ib_idx, int lo, int hi)
471 {
472 	struct amdgpu_bo_va_mapping *mapping;
473 	struct amdgpu_ib *ib = &p->ibs[ib_idx];
474 	struct amdgpu_bo *bo;
475 	uint64_t addr;
476 
477 	addr = ((uint64_t)amdgpu_get_ib_value(p, ib_idx, lo)) |
478 	       ((uint64_t)amdgpu_get_ib_value(p, ib_idx, hi)) << 32;
479 
480 	mapping = amdgpu_cs_find_mapping(p, addr, &bo);
481 	if (mapping == NULL) {
482 		DRM_ERROR("Can't find BO for addr 0x%010Lx %d %d\n",
483 			  addr, lo, hi);
484 		return -EINVAL;
485 	}
486 
487 	addr -= ((uint64_t)mapping->it.start) * AMDGPU_GPU_PAGE_SIZE;
488 	addr += amdgpu_bo_gpu_offset(bo);
489 
490 	ib->ptr[lo] = addr & 0xFFFFFFFF;
491 	ib->ptr[hi] = addr >> 32;
492 
493 	return 0;
494 }
495 
496 /**
497  * amdgpu_vce_cs_parse - parse and validate the command stream
498  *
499  * @p: parser context
500  *
501  */
502 int amdgpu_vce_ring_parse_cs(struct amdgpu_cs_parser *p, uint32_t ib_idx)
503 {
504 	uint32_t handle = 0;
505 	bool destroy = false;
506 	int i, r, idx = 0;
507 	struct amdgpu_ib *ib = &p->ibs[ib_idx];
508 
509 	amdgpu_vce_note_usage(p->adev);
510 
511 	while (idx < ib->length_dw) {
512 		uint32_t len = amdgpu_get_ib_value(p, ib_idx, idx);
513 		uint32_t cmd = amdgpu_get_ib_value(p, ib_idx, idx + 1);
514 
515 		if ((len < 8) || (len & 3)) {
516 			DRM_ERROR("invalid VCE command length (%d)!\n", len);
517 			return -EINVAL;
518 		}
519 
520 		switch (cmd) {
521 		case 0x00000001: // session
522 			handle = amdgpu_get_ib_value(p, ib_idx, idx + 2);
523 			break;
524 
525 		case 0x00000002: // task info
526 		case 0x01000001: // create
527 		case 0x04000001: // config extension
528 		case 0x04000002: // pic control
529 		case 0x04000005: // rate control
530 		case 0x04000007: // motion estimation
531 		case 0x04000008: // rdo
532 		case 0x04000009: // vui
533 		case 0x05000002: // auxiliary buffer
534 			break;
535 
536 		case 0x03000001: // encode
537 			r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 10, idx + 9);
538 			if (r)
539 				return r;
540 
541 			r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 12, idx + 11);
542 			if (r)
543 				return r;
544 			break;
545 
546 		case 0x02000001: // destroy
547 			destroy = true;
548 			break;
549 
550 		case 0x05000001: // context buffer
551 		case 0x05000004: // video bitstream buffer
552 		case 0x05000005: // feedback buffer
553 			r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 3, idx + 2);
554 			if (r)
555 				return r;
556 			break;
557 
558 		default:
559 			DRM_ERROR("invalid VCE command (0x%x)!\n", cmd);
560 			return -EINVAL;
561 		}
562 
563 		idx += len / 4;
564 	}
565 
566 	if (destroy) {
567 		/* IB contains a destroy msg, free the handle */
568 		for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i)
569 			atomic_cmpxchg(&p->adev->vce.handles[i], handle, 0);
570 
571 		return 0;
572 	}
573 
574 	/* create or encode, validate the handle */
575 	for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) {
576 		if (atomic_read(&p->adev->vce.handles[i]) == handle)
577 			return 0;
578 	}
579 
580 	/* handle not found try to alloc a new one */
581 	for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) {
582 		if (!atomic_cmpxchg(&p->adev->vce.handles[i], 0, handle)) {
583 			p->adev->vce.filp[i] = p->filp;
584 			return 0;
585 		}
586 	}
587 
588 	DRM_ERROR("No more free VCE handles!\n");
589 
590 	return -EINVAL;
591 }
592 
593 /**
594  * amdgpu_vce_ring_emit_semaphore - emit a semaphore command
595  *
596  * @ring: engine to use
597  * @semaphore: address of semaphore
598  * @emit_wait: true=emit wait, false=emit signal
599  *
600  */
601 bool amdgpu_vce_ring_emit_semaphore(struct amdgpu_ring *ring,
602 				    struct amdgpu_semaphore *semaphore,
603 				    bool emit_wait)
604 {
605 	uint64_t addr = semaphore->gpu_addr;
606 
607 	amdgpu_ring_write(ring, VCE_CMD_SEMAPHORE);
608 	amdgpu_ring_write(ring, (addr >> 3) & 0x000FFFFF);
609 	amdgpu_ring_write(ring, (addr >> 23) & 0x000FFFFF);
610 	amdgpu_ring_write(ring, 0x01003000 | (emit_wait ? 1 : 0));
611 	if (!emit_wait)
612 		amdgpu_ring_write(ring, VCE_CMD_END);
613 
614 	return true;
615 }
616 
617 /**
618  * amdgpu_vce_ring_emit_ib - execute indirect buffer
619  *
620  * @ring: engine to use
621  * @ib: the IB to execute
622  *
623  */
624 void amdgpu_vce_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
625 {
626 	amdgpu_ring_write(ring, VCE_CMD_IB);
627 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
628 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
629 	amdgpu_ring_write(ring, ib->length_dw);
630 }
631 
632 /**
633  * amdgpu_vce_ring_emit_fence - add a fence command to the ring
634  *
635  * @ring: engine to use
636  * @fence: the fence
637  *
638  */
639 void amdgpu_vce_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
640 				unsigned flags)
641 {
642 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
643 
644 	amdgpu_ring_write(ring, VCE_CMD_FENCE);
645 	amdgpu_ring_write(ring, addr);
646 	amdgpu_ring_write(ring, upper_32_bits(addr));
647 	amdgpu_ring_write(ring, seq);
648 	amdgpu_ring_write(ring, VCE_CMD_TRAP);
649 	amdgpu_ring_write(ring, VCE_CMD_END);
650 }
651 
652 /**
653  * amdgpu_vce_ring_test_ring - test if VCE ring is working
654  *
655  * @ring: the engine to test on
656  *
657  */
658 int amdgpu_vce_ring_test_ring(struct amdgpu_ring *ring)
659 {
660 	struct amdgpu_device *adev = ring->adev;
661 	uint32_t rptr = amdgpu_ring_get_rptr(ring);
662 	unsigned i;
663 	int r;
664 
665 	r = amdgpu_ring_lock(ring, 16);
666 	if (r) {
667 		DRM_ERROR("amdgpu: vce failed to lock ring %d (%d).\n",
668 			  ring->idx, r);
669 		return r;
670 	}
671 	amdgpu_ring_write(ring, VCE_CMD_END);
672 	amdgpu_ring_unlock_commit(ring);
673 
674 	for (i = 0; i < adev->usec_timeout; i++) {
675 		if (amdgpu_ring_get_rptr(ring) != rptr)
676 			break;
677 		DRM_UDELAY(1);
678 	}
679 
680 	if (i < adev->usec_timeout) {
681 		DRM_INFO("ring test on %d succeeded in %d usecs\n",
682 			 ring->idx, i);
683 	} else {
684 		DRM_ERROR("amdgpu: ring %d test failed\n",
685 			  ring->idx);
686 		r = -ETIMEDOUT;
687 	}
688 
689 	return r;
690 }
691 
692 /**
693  * amdgpu_vce_ring_test_ib - test if VCE IBs are working
694  *
695  * @ring: the engine to test on
696  *
697  */
698 int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring)
699 {
700 	struct amdgpu_fence *fence = NULL;
701 	int r;
702 
703 	r = amdgpu_vce_get_create_msg(ring, 1, NULL);
704 	if (r) {
705 		DRM_ERROR("amdgpu: failed to get create msg (%d).\n", r);
706 		goto error;
707 	}
708 
709 	r = amdgpu_vce_get_destroy_msg(ring, 1, &fence);
710 	if (r) {
711 		DRM_ERROR("amdgpu: failed to get destroy ib (%d).\n", r);
712 		goto error;
713 	}
714 
715 	r = amdgpu_fence_wait(fence, false);
716 	if (r) {
717 		DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
718 	} else {
719 		DRM_INFO("ib test on ring %d succeeded\n", ring->idx);
720 	}
721 error:
722 	amdgpu_fence_unref(&fence);
723 	return r;
724 }
725