• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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  */
radeon_vce_init(struct radeon_device * rdev)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  */
radeon_vce_fini(struct radeon_device * rdev)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  */
radeon_vce_suspend(struct radeon_device * rdev)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  */
radeon_vce_resume(struct radeon_device * rdev)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  */
radeon_vce_idle_work_handler(struct work_struct * work)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  */
radeon_vce_note_usage(struct radeon_device * rdev)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  */
radeon_vce_free_handles(struct radeon_device * rdev,struct drm_file * filp)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  */
radeon_vce_get_create_msg(struct radeon_device * rdev,int ring,uint32_t handle,struct radeon_fence ** fence)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  */
radeon_vce_get_destroy_msg(struct radeon_device * rdev,int ring,uint32_t handle,struct radeon_fence ** fence)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  */
radeon_vce_cs_reloc(struct radeon_cs_parser * p,int lo,int hi,unsigned size)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_cs_reloc *reloc;
457 	uint64_t start, end, offset;
458 	unsigned idx;
459 
460 	relocs_chunk = &p->chunks[p->chunk_relocs_idx];
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_ptr[(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  * @allocated: allocated a new handle?
497  *
498  * Validates the handle and return the found session index or -EINVAL
499  * we we don't have another free session index.
500  */
radeon_vce_validate_handle(struct radeon_cs_parser * p,uint32_t handle,bool * allocated)501 static int radeon_vce_validate_handle(struct radeon_cs_parser *p,
502 				      uint32_t handle, bool *allocated)
503 {
504 	unsigned i;
505 
506 	*allocated = false;
507 
508 	/* validate the handle */
509 	for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) {
510 		if (atomic_read(&p->rdev->vce.handles[i]) == handle) {
511 			if (p->rdev->vce.filp[i] != p->filp) {
512 				DRM_ERROR("VCE handle collision detected!\n");
513 				return -EINVAL;
514 			}
515 			return i;
516 		}
517 	}
518 
519 	/* handle not found try to alloc a new one */
520 	for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) {
521 		if (!atomic_cmpxchg(&p->rdev->vce.handles[i], 0, handle)) {
522 			p->rdev->vce.filp[i] = p->filp;
523 			p->rdev->vce.img_size[i] = 0;
524 			*allocated = true;
525 			return i;
526 		}
527 	}
528 
529 	DRM_ERROR("No more free VCE handles!\n");
530 	return -EINVAL;
531 }
532 
533 /**
534  * radeon_vce_cs_parse - parse and validate the command stream
535  *
536  * @p: parser context
537  *
538  */
radeon_vce_cs_parse(struct radeon_cs_parser * p)539 int radeon_vce_cs_parse(struct radeon_cs_parser *p)
540 {
541 	int session_idx = -1;
542 	bool destroyed = false, created = false, allocated = false;
543 	uint32_t tmp, handle = 0;
544 	uint32_t *size = &tmp;
545 	int i, r = 0;
546 
547 	while (p->idx < p->chunks[p->chunk_ib_idx].length_dw) {
548 		uint32_t len = radeon_get_ib_value(p, p->idx);
549 		uint32_t cmd = radeon_get_ib_value(p, p->idx + 1);
550 
551 		if ((len < 8) || (len & 3)) {
552 			DRM_ERROR("invalid VCE command length (%d)!\n", len);
553 			r = -EINVAL;
554 			goto out;
555 		}
556 
557 		if (destroyed) {
558 			DRM_ERROR("No other command allowed after destroy!\n");
559 			r = -EINVAL;
560 			goto out;
561 		}
562 
563 		switch (cmd) {
564 		case 0x00000001: // session
565 			handle = radeon_get_ib_value(p, p->idx + 2);
566 			session_idx = radeon_vce_validate_handle(p, handle,
567 								 &allocated);
568 			if (session_idx < 0)
569 				return session_idx;
570 			size = &p->rdev->vce.img_size[session_idx];
571 			break;
572 
573 		case 0x00000002: // task info
574 			break;
575 
576 		case 0x01000001: // create
577 			created = true;
578 			if (!allocated) {
579 				DRM_ERROR("Handle already in use!\n");
580 				r = -EINVAL;
581 				goto out;
582 			}
583 
584 			*size = radeon_get_ib_value(p, p->idx + 8) *
585 				radeon_get_ib_value(p, p->idx + 10) *
586 				8 * 3 / 2;
587 			break;
588 
589 		case 0x04000001: // config extension
590 		case 0x04000002: // pic control
591 		case 0x04000005: // rate control
592 		case 0x04000007: // motion estimation
593 		case 0x04000008: // rdo
594 			break;
595 
596 		case 0x03000001: // encode
597 			r = radeon_vce_cs_reloc(p, p->idx + 10, p->idx + 9,
598 						*size);
599 			if (r)
600 				goto out;
601 
602 			r = radeon_vce_cs_reloc(p, p->idx + 12, p->idx + 11,
603 						*size / 3);
604 			if (r)
605 				goto out;
606 			break;
607 
608 		case 0x02000001: // destroy
609 			destroyed = true;
610 			break;
611 
612 		case 0x05000001: // context buffer
613 			r = radeon_vce_cs_reloc(p, p->idx + 3, p->idx + 2,
614 						*size * 2);
615 			if (r)
616 				goto out;
617 			break;
618 
619 		case 0x05000004: // video bitstream buffer
620 			tmp = radeon_get_ib_value(p, p->idx + 4);
621 			r = radeon_vce_cs_reloc(p, p->idx + 3, p->idx + 2,
622 						tmp);
623 			if (r)
624 				goto out;
625 			break;
626 
627 		case 0x05000005: // feedback buffer
628 			r = radeon_vce_cs_reloc(p, p->idx + 3, p->idx + 2,
629 						4096);
630 			if (r)
631 				goto out;
632 			break;
633 
634 		default:
635 			DRM_ERROR("invalid VCE command (0x%x)!\n", cmd);
636 			r = -EINVAL;
637 			goto out;
638 		}
639 
640 		if (session_idx == -1) {
641 			DRM_ERROR("no session command at start of IB\n");
642 			r = -EINVAL;
643 			goto out;
644 		}
645 
646 		p->idx += len / 4;
647 	}
648 
649 	if (allocated && !created) {
650 		DRM_ERROR("New session without create command!\n");
651 		r = -ENOENT;
652 	}
653 
654 out:
655 	if ((!r && destroyed) || (r && allocated)) {
656 		/*
657 		 * IB contains a destroy msg or we have allocated an
658 		 * handle and got an error, anyway free the handle
659 		 */
660 		for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i)
661 			atomic_cmpxchg(&p->rdev->vce.handles[i], handle, 0);
662 	}
663 
664 	return r;
665 }
666 
667 /**
668  * radeon_vce_semaphore_emit - emit a semaphore command
669  *
670  * @rdev: radeon_device pointer
671  * @ring: engine to use
672  * @semaphore: address of semaphore
673  * @emit_wait: true=emit wait, false=emit signal
674  *
675  */
radeon_vce_semaphore_emit(struct radeon_device * rdev,struct radeon_ring * ring,struct radeon_semaphore * semaphore,bool emit_wait)676 bool radeon_vce_semaphore_emit(struct radeon_device *rdev,
677 			       struct radeon_ring *ring,
678 			       struct radeon_semaphore *semaphore,
679 			       bool emit_wait)
680 {
681 	uint64_t addr = semaphore->gpu_addr;
682 
683 	radeon_ring_write(ring, VCE_CMD_SEMAPHORE);
684 	radeon_ring_write(ring, (addr >> 3) & 0x000FFFFF);
685 	radeon_ring_write(ring, (addr >> 23) & 0x000FFFFF);
686 	radeon_ring_write(ring, 0x01003000 | (emit_wait ? 1 : 0));
687 	if (!emit_wait)
688 		radeon_ring_write(ring, VCE_CMD_END);
689 
690 	return true;
691 }
692 
693 /**
694  * radeon_vce_ib_execute - execute indirect buffer
695  *
696  * @rdev: radeon_device pointer
697  * @ib: the IB to execute
698  *
699  */
radeon_vce_ib_execute(struct radeon_device * rdev,struct radeon_ib * ib)700 void radeon_vce_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
701 {
702 	struct radeon_ring *ring = &rdev->ring[ib->ring];
703 	radeon_ring_write(ring, VCE_CMD_IB);
704 	radeon_ring_write(ring, ib->gpu_addr);
705 	radeon_ring_write(ring, upper_32_bits(ib->gpu_addr));
706 	radeon_ring_write(ring, ib->length_dw);
707 }
708 
709 /**
710  * radeon_vce_fence_emit - add a fence command to the ring
711  *
712  * @rdev: radeon_device pointer
713  * @fence: the fence
714  *
715  */
radeon_vce_fence_emit(struct radeon_device * rdev,struct radeon_fence * fence)716 void radeon_vce_fence_emit(struct radeon_device *rdev,
717 			   struct radeon_fence *fence)
718 {
719 	struct radeon_ring *ring = &rdev->ring[fence->ring];
720 	uint64_t addr = rdev->fence_drv[fence->ring].gpu_addr;
721 
722 	radeon_ring_write(ring, VCE_CMD_FENCE);
723 	radeon_ring_write(ring, addr);
724 	radeon_ring_write(ring, upper_32_bits(addr));
725 	radeon_ring_write(ring, fence->seq);
726 	radeon_ring_write(ring, VCE_CMD_TRAP);
727 	radeon_ring_write(ring, VCE_CMD_END);
728 }
729 
730 /**
731  * radeon_vce_ring_test - test if VCE ring is working
732  *
733  * @rdev: radeon_device pointer
734  * @ring: the engine to test on
735  *
736  */
radeon_vce_ring_test(struct radeon_device * rdev,struct radeon_ring * ring)737 int radeon_vce_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
738 {
739 	uint32_t rptr = vce_v1_0_get_rptr(rdev, ring);
740 	unsigned i;
741 	int r;
742 
743 	r = radeon_ring_lock(rdev, ring, 16);
744 	if (r) {
745 		DRM_ERROR("radeon: vce failed to lock ring %d (%d).\n",
746 			  ring->idx, r);
747 		return r;
748 	}
749 	radeon_ring_write(ring, VCE_CMD_END);
750 	radeon_ring_unlock_commit(rdev, ring, false);
751 
752 	for (i = 0; i < rdev->usec_timeout; i++) {
753 	        if (vce_v1_0_get_rptr(rdev, ring) != rptr)
754 	                break;
755 	        DRM_UDELAY(1);
756 	}
757 
758 	if (i < rdev->usec_timeout) {
759 	        DRM_INFO("ring test on %d succeeded in %d usecs\n",
760 	                 ring->idx, i);
761 	} else {
762 	        DRM_ERROR("radeon: ring %d test failed\n",
763 	                  ring->idx);
764 	        r = -ETIMEDOUT;
765 	}
766 
767 	return r;
768 }
769 
770 /**
771  * radeon_vce_ib_test - test if VCE IBs are working
772  *
773  * @rdev: radeon_device pointer
774  * @ring: the engine to test on
775  *
776  */
radeon_vce_ib_test(struct radeon_device * rdev,struct radeon_ring * ring)777 int radeon_vce_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
778 {
779 	struct radeon_fence *fence = NULL;
780 	int r;
781 
782 	r = radeon_vce_get_create_msg(rdev, ring->idx, 1, NULL);
783 	if (r) {
784 		DRM_ERROR("radeon: failed to get create msg (%d).\n", r);
785 		goto error;
786 	}
787 
788 	r = radeon_vce_get_destroy_msg(rdev, ring->idx, 1, &fence);
789 	if (r) {
790 		DRM_ERROR("radeon: failed to get destroy ib (%d).\n", r);
791 		goto error;
792 	}
793 
794 	r = radeon_fence_wait(fence, false);
795 	if (r) {
796 		DRM_ERROR("radeon: fence wait failed (%d).\n", r);
797 	} else {
798 	        DRM_INFO("ib test on ring %d succeeded\n", ring->idx);
799 	}
800 error:
801 	radeon_fence_unref(&fence);
802 	return r;
803 }
804