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 
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_TAHITI	"radeon/TAHITI_vce.bin"
42 #define FIRMWARE_BONAIRE	"radeon/BONAIRE_vce.bin"
43 
44 MODULE_FIRMWARE(FIRMWARE_TAHITI);
45 MODULE_FIRMWARE(FIRMWARE_BONAIRE);
46 
47 static void radeon_vce_idle_work_handler(struct work_struct *work);
48 
49 /**
50  * radeon_vce_init - allocate memory, load vce firmware
51  *
52  * @rdev: radeon_device pointer
53  *
54  * First step to get VCE online, allocate memory and load the firmware
55  */
56 int radeon_vce_init(struct radeon_device *rdev)
57 {
58 	static const char *fw_version = "[ATI LIB=VCEFW,";
59 	static const char *fb_version = "[ATI LIB=VCEFWSTATS,";
60 	unsigned long size;
61 	const char *fw_name, *c;
62 	uint8_t start, mid, end;
63 	int i, r;
64 
65 	INIT_DELAYED_WORK(&rdev->vce.idle_work, radeon_vce_idle_work_handler);
66 
67 	switch (rdev->family) {
68 	case CHIP_TAHITI:
69 	case CHIP_PITCAIRN:
70 	case CHIP_VERDE:
71 	case CHIP_OLAND:
72 	case CHIP_ARUBA:
73 		fw_name = FIRMWARE_TAHITI;
74 		break;
75 
76 	case CHIP_BONAIRE:
77 	case CHIP_KAVERI:
78 	case CHIP_KABINI:
79 	case CHIP_HAWAII:
80 	case CHIP_MULLINS:
81 		fw_name = FIRMWARE_BONAIRE;
82 		break;
83 
84 	default:
85 		return -EINVAL;
86 	}
87 
88 	r = request_firmware(&rdev->vce_fw, fw_name, rdev->dev);
89 	if (r) {
90 		dev_err(rdev->dev, "radeon_vce: Can't load firmware \"%s\"\n",
91 			fw_name);
92 		return r;
93 	}
94 
95 	/* search for firmware version */
96 
97 	size = rdev->vce_fw->size - strlen(fw_version) - 9;
98 	c = rdev->vce_fw->data;
99 	for (;size > 0; --size, ++c)
100 		if (strncmp(c, fw_version, strlen(fw_version)) == 0)
101 			break;
102 
103 	if (size == 0)
104 		return -EINVAL;
105 
106 	c += strlen(fw_version);
107 	if (sscanf(c, "%2hhd.%2hhd.%2hhd]", &start, &mid, &end) != 3)
108 		return -EINVAL;
109 
110 	/* search for feedback version */
111 
112 	size = rdev->vce_fw->size - strlen(fb_version) - 3;
113 	c = rdev->vce_fw->data;
114 	for (;size > 0; --size, ++c)
115 		if (strncmp(c, fb_version, strlen(fb_version)) == 0)
116 			break;
117 
118 	if (size == 0)
119 		return -EINVAL;
120 
121 	c += strlen(fb_version);
122 	if (sscanf(c, "%2u]", &rdev->vce.fb_version) != 1)
123 		return -EINVAL;
124 
125 	DRM_INFO("Found VCE firmware/feedback version %hhd.%hhd.%hhd / %d!\n",
126 		 start, mid, end, rdev->vce.fb_version);
127 
128 	rdev->vce.fw_version = (start << 24) | (mid << 16) | (end << 8);
129 
130 	/* we can only work with this fw version for now */
131 	if ((rdev->vce.fw_version != ((40 << 24) | (2 << 16) | (2 << 8))) &&
132 	    (rdev->vce.fw_version != ((50 << 24) | (0 << 16) | (1 << 8))) &&
133 	    (rdev->vce.fw_version != ((50 << 24) | (1 << 16) | (2 << 8))))
134 		return -EINVAL;
135 
136 	/* allocate firmware, stack and heap BO */
137 
138 	if (rdev->family < CHIP_BONAIRE)
139 		size = vce_v1_0_bo_size(rdev);
140 	else
141 		size = vce_v2_0_bo_size(rdev);
142 	r = radeon_bo_create(rdev, size, PAGE_SIZE, true,
143 			     RADEON_GEM_DOMAIN_VRAM, 0, NULL, NULL,
144 			     &rdev->vce.vcpu_bo);
145 	if (r) {
146 		dev_err(rdev->dev, "(%d) failed to allocate VCE bo\n", r);
147 		return r;
148 	}
149 
150 	r = radeon_bo_reserve(rdev->vce.vcpu_bo, false);
151 	if (r) {
152 		radeon_bo_unref(&rdev->vce.vcpu_bo);
153 		dev_err(rdev->dev, "(%d) failed to reserve VCE bo\n", r);
154 		return r;
155 	}
156 
157 	r = radeon_bo_pin(rdev->vce.vcpu_bo, RADEON_GEM_DOMAIN_VRAM,
158 			  &rdev->vce.gpu_addr);
159 	radeon_bo_unreserve(rdev->vce.vcpu_bo);
160 	if (r) {
161 		radeon_bo_unref(&rdev->vce.vcpu_bo);
162 		dev_err(rdev->dev, "(%d) VCE bo pin failed\n", r);
163 		return r;
164 	}
165 
166 	for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) {
167 		atomic_set(&rdev->vce.handles[i], 0);
168 		rdev->vce.filp[i] = NULL;
169 	}
170 
171 	return 0;
172 }
173 
174 /**
175  * radeon_vce_fini - free memory
176  *
177  * @rdev: radeon_device pointer
178  *
179  * Last step on VCE teardown, free firmware memory
180  */
181 void radeon_vce_fini(struct radeon_device *rdev)
182 {
183 	if (rdev->vce.vcpu_bo == NULL)
184 		return;
185 
186 	radeon_bo_unref(&rdev->vce.vcpu_bo);
187 
188 	release_firmware(rdev->vce_fw);
189 }
190 
191 /**
192  * radeon_vce_suspend - unpin VCE fw memory
193  *
194  * @rdev: radeon_device pointer
195  *
196  */
197 int radeon_vce_suspend(struct radeon_device *rdev)
198 {
199 	int i;
200 
201 	if (rdev->vce.vcpu_bo == NULL)
202 		return 0;
203 
204 	for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i)
205 		if (atomic_read(&rdev->vce.handles[i]))
206 			break;
207 
208 	if (i == RADEON_MAX_VCE_HANDLES)
209 		return 0;
210 
211 	/* TODO: suspending running encoding sessions isn't supported */
212 	return -EINVAL;
213 }
214 
215 /**
216  * radeon_vce_resume - pin VCE fw memory
217  *
218  * @rdev: radeon_device pointer
219  *
220  */
221 int radeon_vce_resume(struct radeon_device *rdev)
222 {
223 	void *cpu_addr;
224 	int r;
225 
226 	if (rdev->vce.vcpu_bo == NULL)
227 		return -EINVAL;
228 
229 	r = radeon_bo_reserve(rdev->vce.vcpu_bo, false);
230 	if (r) {
231 		dev_err(rdev->dev, "(%d) failed to reserve VCE bo\n", r);
232 		return r;
233 	}
234 
235 	r = radeon_bo_kmap(rdev->vce.vcpu_bo, &cpu_addr);
236 	if (r) {
237 		radeon_bo_unreserve(rdev->vce.vcpu_bo);
238 		dev_err(rdev->dev, "(%d) VCE map failed\n", r);
239 		return r;
240 	}
241 
242 	memset(cpu_addr, 0, radeon_bo_size(rdev->vce.vcpu_bo));
243 	if (rdev->family < CHIP_BONAIRE)
244 		r = vce_v1_0_load_fw(rdev, cpu_addr);
245 	else
246 		memcpy(cpu_addr, rdev->vce_fw->data, rdev->vce_fw->size);
247 
248 	radeon_bo_kunmap(rdev->vce.vcpu_bo);
249 
250 	radeon_bo_unreserve(rdev->vce.vcpu_bo);
251 
252 	return r;
253 }
254 
255 /**
256  * radeon_vce_idle_work_handler - power off VCE
257  *
258  * @work: pointer to work structure
259  *
260  * power of VCE when it's not used any more
261  */
262 static void radeon_vce_idle_work_handler(struct work_struct *work)
263 {
264 	struct radeon_device *rdev =
265 		container_of(work, struct radeon_device, vce.idle_work.work);
266 
267 	if ((radeon_fence_count_emitted(rdev, TN_RING_TYPE_VCE1_INDEX) == 0) &&
268 	    (radeon_fence_count_emitted(rdev, TN_RING_TYPE_VCE2_INDEX) == 0)) {
269 		if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
270 			radeon_dpm_enable_vce(rdev, false);
271 		} else {
272 			radeon_set_vce_clocks(rdev, 0, 0);
273 		}
274 	} else {
275 		schedule_delayed_work(&rdev->vce.idle_work,
276 				      msecs_to_jiffies(VCE_IDLE_TIMEOUT_MS));
277 	}
278 }
279 
280 /**
281  * radeon_vce_note_usage - power up VCE
282  *
283  * @rdev: radeon_device pointer
284  *
285  * Make sure VCE is powerd up when we want to use it
286  */
287 void radeon_vce_note_usage(struct radeon_device *rdev)
288 {
289 	bool streams_changed = false;
290 	bool set_clocks = !cancel_delayed_work_sync(&rdev->vce.idle_work);
291 	set_clocks &= schedule_delayed_work(&rdev->vce.idle_work,
292 					    msecs_to_jiffies(VCE_IDLE_TIMEOUT_MS));
293 
294 	if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
295 		/* XXX figure out if the streams changed */
296 		streams_changed = false;
297 	}
298 
299 	if (set_clocks || streams_changed) {
300 		if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
301 			radeon_dpm_enable_vce(rdev, true);
302 		} else {
303 			radeon_set_vce_clocks(rdev, 53300, 40000);
304 		}
305 	}
306 }
307 
308 /**
309  * radeon_vce_free_handles - free still open VCE handles
310  *
311  * @rdev: radeon_device pointer
312  * @filp: drm file pointer
313  *
314  * Close all VCE handles still open by this file pointer
315  */
316 void radeon_vce_free_handles(struct radeon_device *rdev, struct drm_file *filp)
317 {
318 	int i, r;
319 	for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) {
320 		uint32_t handle = atomic_read(&rdev->vce.handles[i]);
321 		if (!handle || rdev->vce.filp[i] != filp)
322 			continue;
323 
324 		radeon_vce_note_usage(rdev);
325 
326 		r = radeon_vce_get_destroy_msg(rdev, TN_RING_TYPE_VCE1_INDEX,
327 					       handle, NULL);
328 		if (r)
329 			DRM_ERROR("Error destroying VCE handle (%d)!\n", r);
330 
331 		rdev->vce.filp[i] = NULL;
332 		atomic_set(&rdev->vce.handles[i], 0);
333 	}
334 }
335 
336 /**
337  * radeon_vce_get_create_msg - generate a VCE create msg
338  *
339  * @rdev: radeon_device pointer
340  * @ring: ring we should submit the msg to
341  * @handle: VCE session handle to use
342  * @fence: optional fence to return
343  *
344  * Open up a stream for HW test
345  */
346 int radeon_vce_get_create_msg(struct radeon_device *rdev, int ring,
347 			      uint32_t handle, struct radeon_fence **fence)
348 {
349 	const unsigned ib_size_dw = 1024;
350 	struct radeon_ib ib;
351 	uint64_t dummy;
352 	int i, r;
353 
354 	r = radeon_ib_get(rdev, ring, &ib, NULL, ib_size_dw * 4);
355 	if (r) {
356 		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
357 		return r;
358 	}
359 
360 	dummy = ib.gpu_addr + 1024;
361 
362 	/* stitch together an VCE create msg */
363 	ib.length_dw = 0;
364 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x0000000c); /* len */
365 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000001); /* session cmd */
366 	ib.ptr[ib.length_dw++] = cpu_to_le32(handle);
367 
368 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000030); /* len */
369 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x01000001); /* create cmd */
370 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000000);
371 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000042);
372 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x0000000a);
373 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000001);
374 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000080);
375 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000060);
376 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000100);
377 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000100);
378 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x0000000c);
379 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000000);
380 
381 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000014); /* len */
382 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x05000005); /* feedback buffer */
383 	ib.ptr[ib.length_dw++] = cpu_to_le32(upper_32_bits(dummy));
384 	ib.ptr[ib.length_dw++] = cpu_to_le32(dummy);
385 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000001);
386 
387 	for (i = ib.length_dw; i < ib_size_dw; ++i)
388 		ib.ptr[i] = cpu_to_le32(0x0);
389 
390 	r = radeon_ib_schedule(rdev, &ib, NULL, false);
391 	if (r) {
392 		DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
393 	}
394 
395 	if (fence)
396 		*fence = radeon_fence_ref(ib.fence);
397 
398 	radeon_ib_free(rdev, &ib);
399 
400 	return r;
401 }
402 
403 /**
404  * radeon_vce_get_destroy_msg - generate a VCE destroy msg
405  *
406  * @rdev: radeon_device pointer
407  * @ring: ring we should submit the msg to
408  * @handle: VCE session handle to use
409  * @fence: optional fence to return
410  *
411  * Close up a stream for HW test or if userspace failed to do so
412  */
413 int radeon_vce_get_destroy_msg(struct radeon_device *rdev, int ring,
414 			       uint32_t handle, struct radeon_fence **fence)
415 {
416 	const unsigned ib_size_dw = 1024;
417 	struct radeon_ib ib;
418 	uint64_t dummy;
419 	int i, r;
420 
421 	r = radeon_ib_get(rdev, ring, &ib, NULL, ib_size_dw * 4);
422 	if (r) {
423 		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
424 		return r;
425 	}
426 
427 	dummy = ib.gpu_addr + 1024;
428 
429 	/* stitch together an VCE destroy msg */
430 	ib.length_dw = 0;
431 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x0000000c); /* len */
432 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000001); /* session cmd */
433 	ib.ptr[ib.length_dw++] = cpu_to_le32(handle);
434 
435 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000014); /* len */
436 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x05000005); /* feedback buffer */
437 	ib.ptr[ib.length_dw++] = cpu_to_le32(upper_32_bits(dummy));
438 	ib.ptr[ib.length_dw++] = cpu_to_le32(dummy);
439 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000001);
440 
441 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000008); /* len */
442 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x02000001); /* destroy cmd */
443 
444 	for (i = ib.length_dw; i < ib_size_dw; ++i)
445 		ib.ptr[i] = cpu_to_le32(0x0);
446 
447 	r = radeon_ib_schedule(rdev, &ib, NULL, false);
448 	if (r) {
449 		DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
450 	}
451 
452 	if (fence)
453 		*fence = radeon_fence_ref(ib.fence);
454 
455 	radeon_ib_free(rdev, &ib);
456 
457 	return r;
458 }
459 
460 /**
461  * radeon_vce_cs_reloc - command submission relocation
462  *
463  * @p: parser context
464  * @lo: address of lower dword
465  * @hi: address of higher dword
466  * @size: size of checker for relocation buffer
467  *
468  * Patch relocation inside command stream with real buffer address
469  */
470 int radeon_vce_cs_reloc(struct radeon_cs_parser *p, int lo, int hi,
471 			unsigned size)
472 {
473 	struct radeon_cs_chunk *relocs_chunk;
474 	struct radeon_bo_list *reloc;
475 	uint64_t start, end, offset;
476 	unsigned idx;
477 
478 	relocs_chunk = p->chunk_relocs;
479 	offset = radeon_get_ib_value(p, lo);
480 	idx = radeon_get_ib_value(p, hi);
481 
482 	if (idx >= relocs_chunk->length_dw) {
483 		DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
484 			  idx, relocs_chunk->length_dw);
485 		return -EINVAL;
486 	}
487 
488 	reloc = &p->relocs[(idx / 4)];
489 	start = reloc->gpu_offset;
490 	end = start + radeon_bo_size(reloc->robj);
491 	start += offset;
492 
493 	p->ib.ptr[lo] = start & 0xFFFFFFFF;
494 	p->ib.ptr[hi] = start >> 32;
495 
496 	if (end <= start) {
497 		DRM_ERROR("invalid reloc offset %llX!\n", offset);
498 		return -EINVAL;
499 	}
500 	if ((end - start) < size) {
501 		DRM_ERROR("buffer to small (%d / %d)!\n",
502 			(unsigned)(end - start), size);
503 		return -EINVAL;
504 	}
505 
506 	return 0;
507 }
508 
509 /**
510  * radeon_vce_validate_handle - validate stream handle
511  *
512  * @p: parser context
513  * @handle: handle to validate
514  * @allocated: allocated a new handle?
515  *
516  * Validates the handle and return the found session index or -EINVAL
517  * we we don't have another free session index.
518  */
519 static int radeon_vce_validate_handle(struct radeon_cs_parser *p,
520 				      uint32_t handle, bool *allocated)
521 {
522 	unsigned i;
523 
524 	*allocated = false;
525 
526 	/* validate the handle */
527 	for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) {
528 		if (atomic_read(&p->rdev->vce.handles[i]) == handle) {
529 			if (p->rdev->vce.filp[i] != p->filp) {
530 				DRM_ERROR("VCE handle collision detected!\n");
531 				return -EINVAL;
532 			}
533 			return i;
534 		}
535 	}
536 
537 	/* handle not found try to alloc a new one */
538 	for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) {
539 		if (!atomic_cmpxchg(&p->rdev->vce.handles[i], 0, handle)) {
540 			p->rdev->vce.filp[i] = p->filp;
541 			p->rdev->vce.img_size[i] = 0;
542 			*allocated = true;
543 			return i;
544 		}
545 	}
546 
547 	DRM_ERROR("No more free VCE handles!\n");
548 	return -EINVAL;
549 }
550 
551 /**
552  * radeon_vce_cs_parse - parse and validate the command stream
553  *
554  * @p: parser context
555  *
556  */
557 int radeon_vce_cs_parse(struct radeon_cs_parser *p)
558 {
559 	int session_idx = -1;
560 	bool destroyed = false, created = false, allocated = false;
561 	uint32_t tmp, handle = 0;
562 	uint32_t *size = &tmp;
563 	int i, r = 0;
564 
565 	while (p->idx < p->chunk_ib->length_dw) {
566 		uint32_t len = radeon_get_ib_value(p, p->idx);
567 		uint32_t cmd = radeon_get_ib_value(p, p->idx + 1);
568 
569 		if ((len < 8) || (len & 3)) {
570 			DRM_ERROR("invalid VCE command length (%d)!\n", len);
571 			r = -EINVAL;
572 			goto out;
573 		}
574 
575 		if (destroyed) {
576 			DRM_ERROR("No other command allowed after destroy!\n");
577 			r = -EINVAL;
578 			goto out;
579 		}
580 
581 		switch (cmd) {
582 		case 0x00000001: // session
583 			handle = radeon_get_ib_value(p, p->idx + 2);
584 			session_idx = radeon_vce_validate_handle(p, handle,
585 								 &allocated);
586 			if (session_idx < 0)
587 				return session_idx;
588 			size = &p->rdev->vce.img_size[session_idx];
589 			break;
590 
591 		case 0x00000002: // task info
592 			break;
593 
594 		case 0x01000001: // create
595 			created = true;
596 			if (!allocated) {
597 				DRM_ERROR("Handle already in use!\n");
598 				r = -EINVAL;
599 				goto out;
600 			}
601 
602 			*size = radeon_get_ib_value(p, p->idx + 8) *
603 				radeon_get_ib_value(p, p->idx + 10) *
604 				8 * 3 / 2;
605 			break;
606 
607 		case 0x04000001: // config extension
608 		case 0x04000002: // pic control
609 		case 0x04000005: // rate control
610 		case 0x04000007: // motion estimation
611 		case 0x04000008: // rdo
612 		case 0x04000009: // vui
613 			break;
614 
615 		case 0x03000001: // encode
616 			r = radeon_vce_cs_reloc(p, p->idx + 10, p->idx + 9,
617 						*size);
618 			if (r)
619 				goto out;
620 
621 			r = radeon_vce_cs_reloc(p, p->idx + 12, p->idx + 11,
622 						*size / 3);
623 			if (r)
624 				goto out;
625 			break;
626 
627 		case 0x02000001: // destroy
628 			destroyed = true;
629 			break;
630 
631 		case 0x05000001: // context buffer
632 			r = radeon_vce_cs_reloc(p, p->idx + 3, p->idx + 2,
633 						*size * 2);
634 			if (r)
635 				goto out;
636 			break;
637 
638 		case 0x05000004: // video bitstream buffer
639 			tmp = radeon_get_ib_value(p, p->idx + 4);
640 			r = radeon_vce_cs_reloc(p, p->idx + 3, p->idx + 2,
641 						tmp);
642 			if (r)
643 				goto out;
644 			break;
645 
646 		case 0x05000005: // feedback buffer
647 			r = radeon_vce_cs_reloc(p, p->idx + 3, p->idx + 2,
648 						4096);
649 			if (r)
650 				goto out;
651 			break;
652 
653 		default:
654 			DRM_ERROR("invalid VCE command (0x%x)!\n", cmd);
655 			r = -EINVAL;
656 			goto out;
657 		}
658 
659 		if (session_idx == -1) {
660 			DRM_ERROR("no session command at start of IB\n");
661 			r = -EINVAL;
662 			goto out;
663 		}
664 
665 		p->idx += len / 4;
666 	}
667 
668 	if (allocated && !created) {
669 		DRM_ERROR("New session without create command!\n");
670 		r = -ENOENT;
671 	}
672 
673 out:
674 	if ((!r && destroyed) || (r && allocated)) {
675 		/*
676 		 * IB contains a destroy msg or we have allocated an
677 		 * handle and got an error, anyway free the handle
678 		 */
679 		for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i)
680 			atomic_cmpxchg(&p->rdev->vce.handles[i], handle, 0);
681 	}
682 
683 	return r;
684 }
685 
686 /**
687  * radeon_vce_semaphore_emit - emit a semaphore command
688  *
689  * @rdev: radeon_device pointer
690  * @ring: engine to use
691  * @semaphore: address of semaphore
692  * @emit_wait: true=emit wait, false=emit signal
693  *
694  */
695 bool radeon_vce_semaphore_emit(struct radeon_device *rdev,
696 			       struct radeon_ring *ring,
697 			       struct radeon_semaphore *semaphore,
698 			       bool emit_wait)
699 {
700 	uint64_t addr = semaphore->gpu_addr;
701 
702 	radeon_ring_write(ring, cpu_to_le32(VCE_CMD_SEMAPHORE));
703 	radeon_ring_write(ring, cpu_to_le32((addr >> 3) & 0x000FFFFF));
704 	radeon_ring_write(ring, cpu_to_le32((addr >> 23) & 0x000FFFFF));
705 	radeon_ring_write(ring, cpu_to_le32(0x01003000 | (emit_wait ? 1 : 0)));
706 	if (!emit_wait)
707 		radeon_ring_write(ring, cpu_to_le32(VCE_CMD_END));
708 
709 	return true;
710 }
711 
712 /**
713  * radeon_vce_ib_execute - execute indirect buffer
714  *
715  * @rdev: radeon_device pointer
716  * @ib: the IB to execute
717  *
718  */
719 void radeon_vce_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
720 {
721 	struct radeon_ring *ring = &rdev->ring[ib->ring];
722 	radeon_ring_write(ring, cpu_to_le32(VCE_CMD_IB));
723 	radeon_ring_write(ring, cpu_to_le32(ib->gpu_addr));
724 	radeon_ring_write(ring, cpu_to_le32(upper_32_bits(ib->gpu_addr)));
725 	radeon_ring_write(ring, cpu_to_le32(ib->length_dw));
726 }
727 
728 /**
729  * radeon_vce_fence_emit - add a fence command to the ring
730  *
731  * @rdev: radeon_device pointer
732  * @fence: the fence
733  *
734  */
735 void radeon_vce_fence_emit(struct radeon_device *rdev,
736 			   struct radeon_fence *fence)
737 {
738 	struct radeon_ring *ring = &rdev->ring[fence->ring];
739 	uint64_t addr = rdev->fence_drv[fence->ring].gpu_addr;
740 
741 	radeon_ring_write(ring, cpu_to_le32(VCE_CMD_FENCE));
742 	radeon_ring_write(ring, cpu_to_le32(addr));
743 	radeon_ring_write(ring, cpu_to_le32(upper_32_bits(addr)));
744 	radeon_ring_write(ring, cpu_to_le32(fence->seq));
745 	radeon_ring_write(ring, cpu_to_le32(VCE_CMD_TRAP));
746 	radeon_ring_write(ring, cpu_to_le32(VCE_CMD_END));
747 }
748 
749 /**
750  * radeon_vce_ring_test - test if VCE ring is working
751  *
752  * @rdev: radeon_device pointer
753  * @ring: the engine to test on
754  *
755  */
756 int radeon_vce_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
757 {
758 	uint32_t rptr = vce_v1_0_get_rptr(rdev, ring);
759 	unsigned i;
760 	int r;
761 
762 	r = radeon_ring_lock(rdev, ring, 16);
763 	if (r) {
764 		DRM_ERROR("radeon: vce failed to lock ring %d (%d).\n",
765 			  ring->idx, r);
766 		return r;
767 	}
768 	radeon_ring_write(ring, cpu_to_le32(VCE_CMD_END));
769 	radeon_ring_unlock_commit(rdev, ring, false);
770 
771 	for (i = 0; i < rdev->usec_timeout; i++) {
772 		if (vce_v1_0_get_rptr(rdev, ring) != rptr)
773 			break;
774 		udelay(1);
775 	}
776 
777 	if (i < rdev->usec_timeout) {
778 		DRM_INFO("ring test on %d succeeded in %d usecs\n",
779 			 ring->idx, i);
780 	} else {
781 		DRM_ERROR("radeon: ring %d test failed\n",
782 			 ring->idx);
783 		r = -ETIMEDOUT;
784 	}
785 
786 	return r;
787 }
788 
789 /**
790  * radeon_vce_ib_test - test if VCE IBs are working
791  *
792  * @rdev: radeon_device pointer
793  * @ring: the engine to test on
794  *
795  */
796 int radeon_vce_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
797 {
798 	struct radeon_fence *fence = NULL;
799 	int r;
800 
801 	r = radeon_vce_get_create_msg(rdev, ring->idx, 1, NULL);
802 	if (r) {
803 		DRM_ERROR("radeon: failed to get create msg (%d).\n", r);
804 		goto error;
805 	}
806 
807 	r = radeon_vce_get_destroy_msg(rdev, ring->idx, 1, &fence);
808 	if (r) {
809 		DRM_ERROR("radeon: failed to get destroy ib (%d).\n", r);
810 		goto error;
811 	}
812 
813 	r = radeon_fence_wait_timeout(fence, false, usecs_to_jiffies(
814 		RADEON_USEC_IB_TEST_TIMEOUT));
815 	if (r < 0) {
816 		DRM_ERROR("radeon: fence wait failed (%d).\n", r);
817 	} else if (r == 0) {
818 		DRM_ERROR("radeon: fence wait timed out.\n");
819 		r = -ETIMEDOUT;
820 	} else {
821 		DRM_INFO("ib test on ring %d succeeded\n", ring->idx);
822 		r = 0;
823 	}
824 error:
825 	radeon_fence_unref(&fence);
826 	return r;
827 }
828