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 "radeon.h"
34 #include "radeon_asic.h"
35 #include "sid.h"
36 
37 /* 1 second timeout */
38 #define VCE_IDLE_TIMEOUT_MS	1000
39 
40 /* Firmware Names */
41 #define FIRMWARE_BONAIRE	"radeon/BONAIRE_vce.bin"
42 
43 MODULE_FIRMWARE(FIRMWARE_BONAIRE);
44 
45 static void radeon_vce_idle_work_handler(struct work_struct *work);
46 
47 /**
48  * radeon_vce_init - allocate memory, load vce firmware
49  *
50  * @rdev: radeon_device pointer
51  *
52  * First step to get VCE online, allocate memory and load the firmware
53  */
54 int radeon_vce_init(struct radeon_device *rdev)
55 {
56 	static const char *fw_version = "[ATI LIB=VCEFW,";
57 	static const char *fb_version = "[ATI LIB=VCEFWSTATS,";
58 	unsigned long size;
59 	const char *fw_name, *c;
60 	uint8_t start, mid, end;
61 	int i, r;
62 
63 	INIT_DELAYED_WORK(&rdev->vce.idle_work, radeon_vce_idle_work_handler);
64 
65 	switch (rdev->family) {
66 	case CHIP_BONAIRE:
67 	case CHIP_KAVERI:
68 	case CHIP_KABINI:
69 	case CHIP_HAWAII:
70 	case CHIP_MULLINS:
71 		fw_name = FIRMWARE_BONAIRE;
72 		break;
73 
74 	default:
75 		return -EINVAL;
76 	}
77 
78 	r = request_firmware(&rdev->vce_fw, fw_name, rdev->dev);
79 	if (r) {
80 		dev_err(rdev->dev, "radeon_vce: Can't load firmware \"%s\"\n",
81 			fw_name);
82 		return r;
83 	}
84 
85 	/* search for firmware version */
86 
87 	size = rdev->vce_fw->size - strlen(fw_version) - 9;
88 	c = rdev->vce_fw->data;
89 	for (;size > 0; --size, ++c)
90 		if (strncmp(c, fw_version, strlen(fw_version)) == 0)
91 			break;
92 
93 	if (size == 0)
94 		return -EINVAL;
95 
96 	c += strlen(fw_version);
97 	if (sscanf(c, "%2hhd.%2hhd.%2hhd]", &start, &mid, &end) != 3)
98 		return -EINVAL;
99 
100 	/* search for feedback version */
101 
102 	size = rdev->vce_fw->size - strlen(fb_version) - 3;
103 	c = rdev->vce_fw->data;
104 	for (;size > 0; --size, ++c)
105 		if (strncmp(c, fb_version, strlen(fb_version)) == 0)
106 			break;
107 
108 	if (size == 0)
109 		return -EINVAL;
110 
111 	c += strlen(fb_version);
112 	if (sscanf(c, "%2u]", &rdev->vce.fb_version) != 1)
113 		return -EINVAL;
114 
115 	DRM_INFO("Found VCE firmware/feedback version %hhd.%hhd.%hhd / %d!\n",
116 		 start, mid, end, rdev->vce.fb_version);
117 
118 	rdev->vce.fw_version = (start << 24) | (mid << 16) | (end << 8);
119 
120 	/* we can only work with this fw version for now */
121 	if (rdev->vce.fw_version != ((40 << 24) | (2 << 16) | (2 << 8)))
122 		return -EINVAL;
123 
124 	/* allocate firmware, stack and heap BO */
125 
126 	size = RADEON_GPU_PAGE_ALIGN(rdev->vce_fw->size) +
127 	       RADEON_VCE_STACK_SIZE + RADEON_VCE_HEAP_SIZE;
128 	r = radeon_bo_create(rdev, size, PAGE_SIZE, true,
129 			     RADEON_GEM_DOMAIN_VRAM, 0, NULL, NULL,
130 			     &rdev->vce.vcpu_bo);
131 	if (r) {
132 		dev_err(rdev->dev, "(%d) failed to allocate VCE bo\n", r);
133 		return r;
134 	}
135 
136 	r = radeon_bo_reserve(rdev->vce.vcpu_bo, false);
137 	if (r) {
138 		radeon_bo_unref(&rdev->vce.vcpu_bo);
139 		dev_err(rdev->dev, "(%d) failed to reserve VCE bo\n", r);
140 		return r;
141 	}
142 
143 	r = radeon_bo_pin(rdev->vce.vcpu_bo, RADEON_GEM_DOMAIN_VRAM,
144 			  &rdev->vce.gpu_addr);
145 	radeon_bo_unreserve(rdev->vce.vcpu_bo);
146 	if (r) {
147 		radeon_bo_unref(&rdev->vce.vcpu_bo);
148 		dev_err(rdev->dev, "(%d) VCE bo pin failed\n", r);
149 		return r;
150 	}
151 
152 	for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) {
153 		atomic_set(&rdev->vce.handles[i], 0);
154 		rdev->vce.filp[i] = NULL;
155         }
156 
157 	return 0;
158 }
159 
160 /**
161  * radeon_vce_fini - free memory
162  *
163  * @rdev: radeon_device pointer
164  *
165  * Last step on VCE teardown, free firmware memory
166  */
167 void radeon_vce_fini(struct radeon_device *rdev)
168 {
169 	if (rdev->vce.vcpu_bo == NULL)
170 		return;
171 
172 	radeon_bo_unref(&rdev->vce.vcpu_bo);
173 
174 	release_firmware(rdev->vce_fw);
175 }
176 
177 /**
178  * radeon_vce_suspend - unpin VCE fw memory
179  *
180  * @rdev: radeon_device pointer
181  *
182  */
183 int radeon_vce_suspend(struct radeon_device *rdev)
184 {
185 	int i;
186 
187 	if (rdev->vce.vcpu_bo == NULL)
188 		return 0;
189 
190 	for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i)
191 		if (atomic_read(&rdev->vce.handles[i]))
192 			break;
193 
194 	if (i == RADEON_MAX_VCE_HANDLES)
195 		return 0;
196 
197 	/* TODO: suspending running encoding sessions isn't supported */
198 	return -EINVAL;
199 }
200 
201 /**
202  * radeon_vce_resume - pin VCE fw memory
203  *
204  * @rdev: radeon_device pointer
205  *
206  */
207 int radeon_vce_resume(struct radeon_device *rdev)
208 {
209 	void *cpu_addr;
210 	int r;
211 
212 	if (rdev->vce.vcpu_bo == NULL)
213 		return -EINVAL;
214 
215 	r = radeon_bo_reserve(rdev->vce.vcpu_bo, false);
216 	if (r) {
217 		dev_err(rdev->dev, "(%d) failed to reserve VCE bo\n", r);
218 		return r;
219 	}
220 
221 	r = radeon_bo_kmap(rdev->vce.vcpu_bo, &cpu_addr);
222 	if (r) {
223 		radeon_bo_unreserve(rdev->vce.vcpu_bo);
224 		dev_err(rdev->dev, "(%d) VCE map failed\n", r);
225 		return r;
226 	}
227 
228 	memcpy(cpu_addr, rdev->vce_fw->data, rdev->vce_fw->size);
229 
230 	radeon_bo_kunmap(rdev->vce.vcpu_bo);
231 
232 	radeon_bo_unreserve(rdev->vce.vcpu_bo);
233 
234 	return 0;
235 }
236 
237 /**
238  * radeon_vce_idle_work_handler - power off VCE
239  *
240  * @work: pointer to work structure
241  *
242  * power of VCE when it's not used any more
243  */
244 static void radeon_vce_idle_work_handler(struct work_struct *work)
245 {
246 	struct radeon_device *rdev =
247 		container_of(work, struct radeon_device, vce.idle_work.work);
248 
249 	if ((radeon_fence_count_emitted(rdev, TN_RING_TYPE_VCE1_INDEX) == 0) &&
250 	    (radeon_fence_count_emitted(rdev, TN_RING_TYPE_VCE2_INDEX) == 0)) {
251 		if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
252 			radeon_dpm_enable_vce(rdev, false);
253 		} else {
254 			radeon_set_vce_clocks(rdev, 0, 0);
255 		}
256 	} else {
257 		schedule_delayed_work(&rdev->vce.idle_work,
258 				      msecs_to_jiffies(VCE_IDLE_TIMEOUT_MS));
259 	}
260 }
261 
262 /**
263  * radeon_vce_note_usage - power up VCE
264  *
265  * @rdev: radeon_device pointer
266  *
267  * Make sure VCE is powerd up when we want to use it
268  */
269 void radeon_vce_note_usage(struct radeon_device *rdev)
270 {
271 	bool streams_changed = false;
272 	bool set_clocks = !cancel_delayed_work_sync(&rdev->vce.idle_work);
273 	set_clocks &= schedule_delayed_work(&rdev->vce.idle_work,
274 					    msecs_to_jiffies(VCE_IDLE_TIMEOUT_MS));
275 
276 	if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
277 		/* XXX figure out if the streams changed */
278 		streams_changed = false;
279 	}
280 
281 	if (set_clocks || streams_changed) {
282 		if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
283 			radeon_dpm_enable_vce(rdev, true);
284 		} else {
285 			radeon_set_vce_clocks(rdev, 53300, 40000);
286 		}
287 	}
288 }
289 
290 /**
291  * radeon_vce_free_handles - free still open VCE handles
292  *
293  * @rdev: radeon_device pointer
294  * @filp: drm file pointer
295  *
296  * Close all VCE handles still open by this file pointer
297  */
298 void radeon_vce_free_handles(struct radeon_device *rdev, struct drm_file *filp)
299 {
300 	int i, r;
301 	for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) {
302 		uint32_t handle = atomic_read(&rdev->vce.handles[i]);
303 		if (!handle || rdev->vce.filp[i] != filp)
304 			continue;
305 
306 		radeon_vce_note_usage(rdev);
307 
308 		r = radeon_vce_get_destroy_msg(rdev, TN_RING_TYPE_VCE1_INDEX,
309 					       handle, NULL);
310 		if (r)
311 			DRM_ERROR("Error destroying VCE handle (%d)!\n", r);
312 
313 		rdev->vce.filp[i] = NULL;
314 		atomic_set(&rdev->vce.handles[i], 0);
315 	}
316 }
317 
318 /**
319  * radeon_vce_get_create_msg - generate a VCE create msg
320  *
321  * @rdev: radeon_device pointer
322  * @ring: ring we should submit the msg to
323  * @handle: VCE session handle to use
324  * @fence: optional fence to return
325  *
326  * Open up a stream for HW test
327  */
328 int radeon_vce_get_create_msg(struct radeon_device *rdev, int ring,
329 			      uint32_t handle, struct radeon_fence **fence)
330 {
331 	const unsigned ib_size_dw = 1024;
332 	struct radeon_ib ib;
333 	uint64_t dummy;
334 	int i, r;
335 
336 	r = radeon_ib_get(rdev, ring, &ib, NULL, ib_size_dw * 4);
337 	if (r) {
338 		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
339 		return r;
340 	}
341 
342 	dummy = ib.gpu_addr + 1024;
343 
344 	/* stitch together an VCE create msg */
345 	ib.length_dw = 0;
346 	ib.ptr[ib.length_dw++] = 0x0000000c; /* len */
347 	ib.ptr[ib.length_dw++] = 0x00000001; /* session cmd */
348 	ib.ptr[ib.length_dw++] = handle;
349 
350 	ib.ptr[ib.length_dw++] = 0x00000030; /* len */
351 	ib.ptr[ib.length_dw++] = 0x01000001; /* create cmd */
352 	ib.ptr[ib.length_dw++] = 0x00000000;
353 	ib.ptr[ib.length_dw++] = 0x00000042;
354 	ib.ptr[ib.length_dw++] = 0x0000000a;
355 	ib.ptr[ib.length_dw++] = 0x00000001;
356 	ib.ptr[ib.length_dw++] = 0x00000080;
357 	ib.ptr[ib.length_dw++] = 0x00000060;
358 	ib.ptr[ib.length_dw++] = 0x00000100;
359 	ib.ptr[ib.length_dw++] = 0x00000100;
360 	ib.ptr[ib.length_dw++] = 0x0000000c;
361 	ib.ptr[ib.length_dw++] = 0x00000000;
362 
363 	ib.ptr[ib.length_dw++] = 0x00000014; /* len */
364 	ib.ptr[ib.length_dw++] = 0x05000005; /* feedback buffer */
365 	ib.ptr[ib.length_dw++] = upper_32_bits(dummy);
366 	ib.ptr[ib.length_dw++] = dummy;
367 	ib.ptr[ib.length_dw++] = 0x00000001;
368 
369 	for (i = ib.length_dw; i < ib_size_dw; ++i)
370 		ib.ptr[i] = 0x0;
371 
372 	r = radeon_ib_schedule(rdev, &ib, NULL, false);
373 	if (r) {
374 	        DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
375 	}
376 
377 	if (fence)
378 		*fence = radeon_fence_ref(ib.fence);
379 
380 	radeon_ib_free(rdev, &ib);
381 
382 	return r;
383 }
384 
385 /**
386  * radeon_vce_get_destroy_msg - generate a VCE destroy msg
387  *
388  * @rdev: radeon_device pointer
389  * @ring: ring we should submit the msg to
390  * @handle: VCE session handle to use
391  * @fence: optional fence to return
392  *
393  * Close up a stream for HW test or if userspace failed to do so
394  */
395 int radeon_vce_get_destroy_msg(struct radeon_device *rdev, int ring,
396 			       uint32_t handle, struct radeon_fence **fence)
397 {
398 	const unsigned ib_size_dw = 1024;
399 	struct radeon_ib ib;
400 	uint64_t dummy;
401 	int i, r;
402 
403 	r = radeon_ib_get(rdev, ring, &ib, NULL, ib_size_dw * 4);
404 	if (r) {
405 		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
406 		return r;
407 	}
408 
409 	dummy = ib.gpu_addr + 1024;
410 
411 	/* stitch together an VCE destroy msg */
412 	ib.length_dw = 0;
413 	ib.ptr[ib.length_dw++] = 0x0000000c; /* len */
414 	ib.ptr[ib.length_dw++] = 0x00000001; /* session cmd */
415 	ib.ptr[ib.length_dw++] = handle;
416 
417 	ib.ptr[ib.length_dw++] = 0x00000014; /* len */
418 	ib.ptr[ib.length_dw++] = 0x05000005; /* feedback buffer */
419 	ib.ptr[ib.length_dw++] = upper_32_bits(dummy);
420 	ib.ptr[ib.length_dw++] = dummy;
421 	ib.ptr[ib.length_dw++] = 0x00000001;
422 
423 	ib.ptr[ib.length_dw++] = 0x00000008; /* len */
424 	ib.ptr[ib.length_dw++] = 0x02000001; /* destroy cmd */
425 
426 	for (i = ib.length_dw; i < ib_size_dw; ++i)
427 		ib.ptr[i] = 0x0;
428 
429 	r = radeon_ib_schedule(rdev, &ib, NULL, false);
430 	if (r) {
431 	        DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
432 	}
433 
434 	if (fence)
435 		*fence = radeon_fence_ref(ib.fence);
436 
437 	radeon_ib_free(rdev, &ib);
438 
439 	return r;
440 }
441 
442 /**
443  * radeon_vce_cs_reloc - command submission relocation
444  *
445  * @p: parser context
446  * @lo: address of lower dword
447  * @hi: address of higher dword
448  * @size: size of checker for relocation buffer
449  *
450  * Patch relocation inside command stream with real buffer address
451  */
452 int radeon_vce_cs_reloc(struct radeon_cs_parser *p, int lo, int hi,
453 			unsigned size)
454 {
455 	struct radeon_cs_chunk *relocs_chunk;
456 	struct radeon_bo_list *reloc;
457 	uint64_t start, end, offset;
458 	unsigned idx;
459 
460 	relocs_chunk = p->chunk_relocs;
461 	offset = radeon_get_ib_value(p, lo);
462 	idx = radeon_get_ib_value(p, hi);
463 
464 	if (idx >= relocs_chunk->length_dw) {
465 		DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
466 			  idx, relocs_chunk->length_dw);
467 		return -EINVAL;
468 	}
469 
470 	reloc = &p->relocs[(idx / 4)];
471 	start = reloc->gpu_offset;
472 	end = start + radeon_bo_size(reloc->robj);
473 	start += offset;
474 
475 	p->ib.ptr[lo] = start & 0xFFFFFFFF;
476 	p->ib.ptr[hi] = start >> 32;
477 
478 	if (end <= start) {
479 		DRM_ERROR("invalid reloc offset %llX!\n", offset);
480 		return -EINVAL;
481 	}
482 	if ((end - start) < size) {
483 		DRM_ERROR("buffer to small (%d / %d)!\n",
484 			(unsigned)(end - start), size);
485 		return -EINVAL;
486 	}
487 
488 	return 0;
489 }
490 
491 /**
492  * radeon_vce_validate_handle - validate stream handle
493  *
494  * @p: parser context
495  * @handle: handle to validate
496  *
497  * Validates the handle and return the found session index or -EINVAL
498  * we we don't have another free session index.
499  */
500 int radeon_vce_validate_handle(struct radeon_cs_parser *p, uint32_t handle)
501 {
502 	unsigned i;
503 
504 	/* validate the handle */
505 	for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) {
506 		if (atomic_read(&p->rdev->vce.handles[i]) == handle)
507 			return i;
508 	}
509 
510 	/* handle not found try to alloc a new one */
511 	for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) {
512 		if (!atomic_cmpxchg(&p->rdev->vce.handles[i], 0, handle)) {
513 			p->rdev->vce.filp[i] = p->filp;
514 			p->rdev->vce.img_size[i] = 0;
515 			return i;
516 		}
517 	}
518 
519 	DRM_ERROR("No more free VCE handles!\n");
520 	return -EINVAL;
521 }
522 
523 /**
524  * radeon_vce_cs_parse - parse and validate the command stream
525  *
526  * @p: parser context
527  *
528  */
529 int radeon_vce_cs_parse(struct radeon_cs_parser *p)
530 {
531 	int session_idx = -1;
532 	bool destroyed = false;
533 	uint32_t tmp, handle = 0;
534 	uint32_t *size = &tmp;
535 	int i, r;
536 
537 	while (p->idx < p->chunk_ib->length_dw) {
538 		uint32_t len = radeon_get_ib_value(p, p->idx);
539 		uint32_t cmd = radeon_get_ib_value(p, p->idx + 1);
540 
541 		if ((len < 8) || (len & 3)) {
542 			DRM_ERROR("invalid VCE command length (%d)!\n", len);
543                 	return -EINVAL;
544 		}
545 
546 		if (destroyed) {
547 			DRM_ERROR("No other command allowed after destroy!\n");
548 			return -EINVAL;
549 		}
550 
551 		switch (cmd) {
552 		case 0x00000001: // session
553 			handle = radeon_get_ib_value(p, p->idx + 2);
554 			session_idx = radeon_vce_validate_handle(p, handle);
555 			if (session_idx < 0)
556 				return session_idx;
557 			size = &p->rdev->vce.img_size[session_idx];
558 			break;
559 
560 		case 0x00000002: // task info
561 			break;
562 
563 		case 0x01000001: // create
564 			*size = radeon_get_ib_value(p, p->idx + 8) *
565 				radeon_get_ib_value(p, p->idx + 10) *
566 				8 * 3 / 2;
567 			break;
568 
569 		case 0x04000001: // config extension
570 		case 0x04000002: // pic control
571 		case 0x04000005: // rate control
572 		case 0x04000007: // motion estimation
573 		case 0x04000008: // rdo
574 		case 0x04000009: // vui
575 			break;
576 
577 		case 0x03000001: // encode
578 			r = radeon_vce_cs_reloc(p, p->idx + 10, p->idx + 9,
579 						*size);
580 			if (r)
581 				return r;
582 
583 			r = radeon_vce_cs_reloc(p, p->idx + 12, p->idx + 11,
584 						*size / 3);
585 			if (r)
586 				return r;
587 			break;
588 
589 		case 0x02000001: // destroy
590 			destroyed = true;
591 			break;
592 
593 		case 0x05000001: // context buffer
594 			r = radeon_vce_cs_reloc(p, p->idx + 3, p->idx + 2,
595 						*size * 2);
596 			if (r)
597 				return r;
598 			break;
599 
600 		case 0x05000004: // video bitstream buffer
601 			tmp = radeon_get_ib_value(p, p->idx + 4);
602 			r = radeon_vce_cs_reloc(p, p->idx + 3, p->idx + 2,
603 						tmp);
604 			if (r)
605 				return r;
606 			break;
607 
608 		case 0x05000005: // feedback buffer
609 			r = radeon_vce_cs_reloc(p, p->idx + 3, p->idx + 2,
610 						4096);
611 			if (r)
612 				return r;
613 			break;
614 
615 		default:
616 			DRM_ERROR("invalid VCE command (0x%x)!\n", cmd);
617 			return -EINVAL;
618 		}
619 
620 		if (session_idx == -1) {
621 			DRM_ERROR("no session command at start of IB\n");
622 			return -EINVAL;
623 		}
624 
625 		p->idx += len / 4;
626 	}
627 
628 	if (destroyed) {
629 		/* IB contains a destroy msg, free the handle */
630 		for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i)
631 			atomic_cmpxchg(&p->rdev->vce.handles[i], handle, 0);
632 	}
633 
634 	return 0;
635 }
636 
637 /**
638  * radeon_vce_semaphore_emit - emit a semaphore command
639  *
640  * @rdev: radeon_device pointer
641  * @ring: engine to use
642  * @semaphore: address of semaphore
643  * @emit_wait: true=emit wait, false=emit signal
644  *
645  */
646 bool radeon_vce_semaphore_emit(struct radeon_device *rdev,
647 			       struct radeon_ring *ring,
648 			       struct radeon_semaphore *semaphore,
649 			       bool emit_wait)
650 {
651 	uint64_t addr = semaphore->gpu_addr;
652 
653 	radeon_ring_write(ring, VCE_CMD_SEMAPHORE);
654 	radeon_ring_write(ring, (addr >> 3) & 0x000FFFFF);
655 	radeon_ring_write(ring, (addr >> 23) & 0x000FFFFF);
656 	radeon_ring_write(ring, 0x01003000 | (emit_wait ? 1 : 0));
657 	if (!emit_wait)
658 		radeon_ring_write(ring, VCE_CMD_END);
659 
660 	return true;
661 }
662 
663 /**
664  * radeon_vce_ib_execute - execute indirect buffer
665  *
666  * @rdev: radeon_device pointer
667  * @ib: the IB to execute
668  *
669  */
670 void radeon_vce_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
671 {
672 	struct radeon_ring *ring = &rdev->ring[ib->ring];
673 	radeon_ring_write(ring, VCE_CMD_IB);
674 	radeon_ring_write(ring, ib->gpu_addr);
675 	radeon_ring_write(ring, upper_32_bits(ib->gpu_addr));
676 	radeon_ring_write(ring, ib->length_dw);
677 }
678 
679 /**
680  * radeon_vce_fence_emit - add a fence command to the ring
681  *
682  * @rdev: radeon_device pointer
683  * @fence: the fence
684  *
685  */
686 void radeon_vce_fence_emit(struct radeon_device *rdev,
687 			   struct radeon_fence *fence)
688 {
689 	struct radeon_ring *ring = &rdev->ring[fence->ring];
690 	uint64_t addr = rdev->fence_drv[fence->ring].gpu_addr;
691 
692 	radeon_ring_write(ring, VCE_CMD_FENCE);
693 	radeon_ring_write(ring, addr);
694 	radeon_ring_write(ring, upper_32_bits(addr));
695 	radeon_ring_write(ring, fence->seq);
696 	radeon_ring_write(ring, VCE_CMD_TRAP);
697 	radeon_ring_write(ring, VCE_CMD_END);
698 }
699 
700 /**
701  * radeon_vce_ring_test - test if VCE ring is working
702  *
703  * @rdev: radeon_device pointer
704  * @ring: the engine to test on
705  *
706  */
707 int radeon_vce_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
708 {
709 	uint32_t rptr = vce_v1_0_get_rptr(rdev, ring);
710 	unsigned i;
711 	int r;
712 
713 	r = radeon_ring_lock(rdev, ring, 16);
714 	if (r) {
715 		DRM_ERROR("radeon: vce failed to lock ring %d (%d).\n",
716 			  ring->idx, r);
717 		return r;
718 	}
719 	radeon_ring_write(ring, VCE_CMD_END);
720 	radeon_ring_unlock_commit(rdev, ring, false);
721 
722 	for (i = 0; i < rdev->usec_timeout; i++) {
723 	        if (vce_v1_0_get_rptr(rdev, ring) != rptr)
724 	                break;
725 	        DRM_UDELAY(1);
726 	}
727 
728 	if (i < rdev->usec_timeout) {
729 	        DRM_INFO("ring test on %d succeeded in %d usecs\n",
730 	                 ring->idx, i);
731 	} else {
732 	        DRM_ERROR("radeon: ring %d test failed\n",
733 	                  ring->idx);
734 	        r = -ETIMEDOUT;
735 	}
736 
737 	return r;
738 }
739 
740 /**
741  * radeon_vce_ib_test - test if VCE IBs are working
742  *
743  * @rdev: radeon_device pointer
744  * @ring: the engine to test on
745  *
746  */
747 int radeon_vce_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
748 {
749 	struct radeon_fence *fence = NULL;
750 	int r;
751 
752 	r = radeon_vce_get_create_msg(rdev, ring->idx, 1, NULL);
753 	if (r) {
754 		DRM_ERROR("radeon: failed to get create msg (%d).\n", r);
755 		goto error;
756 	}
757 
758 	r = radeon_vce_get_destroy_msg(rdev, ring->idx, 1, &fence);
759 	if (r) {
760 		DRM_ERROR("radeon: failed to get destroy ib (%d).\n", r);
761 		goto error;
762 	}
763 
764 	r = radeon_fence_wait(fence, false);
765 	if (r) {
766 		DRM_ERROR("radeon: fence wait failed (%d).\n", r);
767 	} else {
768 	        DRM_INFO("ib test on ring %d succeeded\n", ring->idx);
769 	}
770 error:
771 	radeon_fence_unref(&fence);
772 	return r;
773 }
774