This source file includes following definitions.
- radeon_cs_buckets_init
- radeon_cs_buckets_add
- radeon_cs_buckets_get_list
- radeon_cs_parser_relocs
- radeon_cs_get_ring
- radeon_cs_sync_rings
- radeon_cs_parser_init
- cmp_size_smaller_first
- radeon_cs_parser_fini
- radeon_cs_ib_chunk
- radeon_bo_vm_update_pte
- radeon_cs_ib_vm_chunk
- radeon_cs_handle_lockup
- radeon_cs_ib_fill
- radeon_cs_ioctl
- radeon_cs_packet_parse
- radeon_cs_packet_next_is_pkt3_nop
- radeon_cs_dump_packet
- radeon_cs_packet_next_reloc
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28 #include <linux/list_sort.h>
29 #include <linux/uaccess.h>
30
31 #include <drm/drm_device.h>
32 #include <drm/drm_file.h>
33 #include <drm/drm_pci.h>
34 #include <drm/radeon_drm.h>
35
36 #include "radeon.h"
37 #include "radeon_reg.h"
38 #include "radeon_trace.h"
39
40 #define RADEON_CS_MAX_PRIORITY 32u
41 #define RADEON_CS_NUM_BUCKETS (RADEON_CS_MAX_PRIORITY + 1)
42
43
44
45
46
47 struct radeon_cs_buckets {
48 struct list_head bucket[RADEON_CS_NUM_BUCKETS];
49 };
50
51 static void radeon_cs_buckets_init(struct radeon_cs_buckets *b)
52 {
53 unsigned i;
54
55 for (i = 0; i < RADEON_CS_NUM_BUCKETS; i++)
56 INIT_LIST_HEAD(&b->bucket[i]);
57 }
58
59 static void radeon_cs_buckets_add(struct radeon_cs_buckets *b,
60 struct list_head *item, unsigned priority)
61 {
62
63
64
65
66
67 list_add_tail(item, &b->bucket[min(priority, RADEON_CS_MAX_PRIORITY)]);
68 }
69
70 static void radeon_cs_buckets_get_list(struct radeon_cs_buckets *b,
71 struct list_head *out_list)
72 {
73 unsigned i;
74
75
76 for (i = 0; i < RADEON_CS_NUM_BUCKETS; i++) {
77 list_splice(&b->bucket[i], out_list);
78 }
79 }
80
81 static int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
82 {
83 struct radeon_cs_chunk *chunk;
84 struct radeon_cs_buckets buckets;
85 unsigned i;
86 bool need_mmap_lock = false;
87 int r;
88
89 if (p->chunk_relocs == NULL) {
90 return 0;
91 }
92 chunk = p->chunk_relocs;
93 p->dma_reloc_idx = 0;
94
95 p->nrelocs = chunk->length_dw / 4;
96 p->relocs = kvmalloc_array(p->nrelocs, sizeof(struct radeon_bo_list),
97 GFP_KERNEL | __GFP_ZERO);
98 if (p->relocs == NULL) {
99 return -ENOMEM;
100 }
101
102 radeon_cs_buckets_init(&buckets);
103
104 for (i = 0; i < p->nrelocs; i++) {
105 struct drm_radeon_cs_reloc *r;
106 struct drm_gem_object *gobj;
107 unsigned priority;
108
109 r = (struct drm_radeon_cs_reloc *)&chunk->kdata[i*4];
110 gobj = drm_gem_object_lookup(p->filp, r->handle);
111 if (gobj == NULL) {
112 DRM_ERROR("gem object lookup failed 0x%x\n",
113 r->handle);
114 return -ENOENT;
115 }
116 p->relocs[i].robj = gem_to_radeon_bo(gobj);
117
118
119
120
121
122
123
124 priority = (r->flags & RADEON_RELOC_PRIO_MASK) * 2
125 + !!r->write_domain;
126
127
128
129
130
131
132 if (p->ring == R600_RING_TYPE_UVD_INDEX &&
133 (i <= 0 || pci_find_capability(p->rdev->ddev->pdev,
134 PCI_CAP_ID_AGP) ||
135 p->rdev->family == CHIP_RS780 ||
136 p->rdev->family == CHIP_RS880)) {
137
138
139 p->relocs[i].preferred_domains =
140 RADEON_GEM_DOMAIN_VRAM;
141
142 p->relocs[i].allowed_domains =
143 RADEON_GEM_DOMAIN_VRAM;
144
145
146 priority = RADEON_CS_MAX_PRIORITY;
147 } else {
148 uint32_t domain = r->write_domain ?
149 r->write_domain : r->read_domains;
150
151 if (domain & RADEON_GEM_DOMAIN_CPU) {
152 DRM_ERROR("RADEON_GEM_DOMAIN_CPU is not valid "
153 "for command submission\n");
154 return -EINVAL;
155 }
156
157 p->relocs[i].preferred_domains = domain;
158 if (domain == RADEON_GEM_DOMAIN_VRAM)
159 domain |= RADEON_GEM_DOMAIN_GTT;
160 p->relocs[i].allowed_domains = domain;
161 }
162
163 if (radeon_ttm_tt_has_userptr(p->relocs[i].robj->tbo.ttm)) {
164 uint32_t domain = p->relocs[i].preferred_domains;
165 if (!(domain & RADEON_GEM_DOMAIN_GTT)) {
166 DRM_ERROR("Only RADEON_GEM_DOMAIN_GTT is "
167 "allowed for userptr BOs\n");
168 return -EINVAL;
169 }
170 need_mmap_lock = true;
171 domain = RADEON_GEM_DOMAIN_GTT;
172 p->relocs[i].preferred_domains = domain;
173 p->relocs[i].allowed_domains = domain;
174 }
175
176
177 if (p->relocs[i].robj->prime_shared_count) {
178 p->relocs[i].allowed_domains &= ~RADEON_GEM_DOMAIN_VRAM;
179 if (!p->relocs[i].allowed_domains) {
180 DRM_ERROR("BO associated with dma-buf cannot "
181 "be moved to VRAM\n");
182 return -EINVAL;
183 }
184 }
185
186 p->relocs[i].tv.bo = &p->relocs[i].robj->tbo;
187 p->relocs[i].tv.num_shared = !r->write_domain;
188
189 radeon_cs_buckets_add(&buckets, &p->relocs[i].tv.head,
190 priority);
191 }
192
193 radeon_cs_buckets_get_list(&buckets, &p->validated);
194
195 if (p->cs_flags & RADEON_CS_USE_VM)
196 p->vm_bos = radeon_vm_get_bos(p->rdev, p->ib.vm,
197 &p->validated);
198 if (need_mmap_lock)
199 down_read(¤t->mm->mmap_sem);
200
201 r = radeon_bo_list_validate(p->rdev, &p->ticket, &p->validated, p->ring);
202
203 if (need_mmap_lock)
204 up_read(¤t->mm->mmap_sem);
205
206 return r;
207 }
208
209 static int radeon_cs_get_ring(struct radeon_cs_parser *p, u32 ring, s32 priority)
210 {
211 p->priority = priority;
212
213 switch (ring) {
214 default:
215 DRM_ERROR("unknown ring id: %d\n", ring);
216 return -EINVAL;
217 case RADEON_CS_RING_GFX:
218 p->ring = RADEON_RING_TYPE_GFX_INDEX;
219 break;
220 case RADEON_CS_RING_COMPUTE:
221 if (p->rdev->family >= CHIP_TAHITI) {
222 if (p->priority > 0)
223 p->ring = CAYMAN_RING_TYPE_CP1_INDEX;
224 else
225 p->ring = CAYMAN_RING_TYPE_CP2_INDEX;
226 } else
227 p->ring = RADEON_RING_TYPE_GFX_INDEX;
228 break;
229 case RADEON_CS_RING_DMA:
230 if (p->rdev->family >= CHIP_CAYMAN) {
231 if (p->priority > 0)
232 p->ring = R600_RING_TYPE_DMA_INDEX;
233 else
234 p->ring = CAYMAN_RING_TYPE_DMA1_INDEX;
235 } else if (p->rdev->family >= CHIP_RV770) {
236 p->ring = R600_RING_TYPE_DMA_INDEX;
237 } else {
238 return -EINVAL;
239 }
240 break;
241 case RADEON_CS_RING_UVD:
242 p->ring = R600_RING_TYPE_UVD_INDEX;
243 break;
244 case RADEON_CS_RING_VCE:
245
246 p->ring = TN_RING_TYPE_VCE1_INDEX;
247 break;
248 }
249 return 0;
250 }
251
252 static int radeon_cs_sync_rings(struct radeon_cs_parser *p)
253 {
254 struct radeon_bo_list *reloc;
255 int r;
256
257 list_for_each_entry(reloc, &p->validated, tv.head) {
258 struct dma_resv *resv;
259
260 resv = reloc->robj->tbo.base.resv;
261 r = radeon_sync_resv(p->rdev, &p->ib.sync, resv,
262 reloc->tv.num_shared);
263 if (r)
264 return r;
265 }
266 return 0;
267 }
268
269
270 int radeon_cs_parser_init(struct radeon_cs_parser *p, void *data)
271 {
272 struct drm_radeon_cs *cs = data;
273 uint64_t *chunk_array_ptr;
274 unsigned size, i;
275 u32 ring = RADEON_CS_RING_GFX;
276 s32 priority = 0;
277
278 INIT_LIST_HEAD(&p->validated);
279
280 if (!cs->num_chunks) {
281 return 0;
282 }
283
284
285 p->idx = 0;
286 p->ib.sa_bo = NULL;
287 p->const_ib.sa_bo = NULL;
288 p->chunk_ib = NULL;
289 p->chunk_relocs = NULL;
290 p->chunk_flags = NULL;
291 p->chunk_const_ib = NULL;
292 p->chunks_array = kcalloc(cs->num_chunks, sizeof(uint64_t), GFP_KERNEL);
293 if (p->chunks_array == NULL) {
294 return -ENOMEM;
295 }
296 chunk_array_ptr = (uint64_t *)(unsigned long)(cs->chunks);
297 if (copy_from_user(p->chunks_array, chunk_array_ptr,
298 sizeof(uint64_t)*cs->num_chunks)) {
299 return -EFAULT;
300 }
301 p->cs_flags = 0;
302 p->nchunks = cs->num_chunks;
303 p->chunks = kcalloc(p->nchunks, sizeof(struct radeon_cs_chunk), GFP_KERNEL);
304 if (p->chunks == NULL) {
305 return -ENOMEM;
306 }
307 for (i = 0; i < p->nchunks; i++) {
308 struct drm_radeon_cs_chunk __user **chunk_ptr = NULL;
309 struct drm_radeon_cs_chunk user_chunk;
310 uint32_t __user *cdata;
311
312 chunk_ptr = (void __user*)(unsigned long)p->chunks_array[i];
313 if (copy_from_user(&user_chunk, chunk_ptr,
314 sizeof(struct drm_radeon_cs_chunk))) {
315 return -EFAULT;
316 }
317 p->chunks[i].length_dw = user_chunk.length_dw;
318 if (user_chunk.chunk_id == RADEON_CHUNK_ID_RELOCS) {
319 p->chunk_relocs = &p->chunks[i];
320 }
321 if (user_chunk.chunk_id == RADEON_CHUNK_ID_IB) {
322 p->chunk_ib = &p->chunks[i];
323
324 if (p->chunks[i].length_dw == 0)
325 return -EINVAL;
326 }
327 if (user_chunk.chunk_id == RADEON_CHUNK_ID_CONST_IB) {
328 p->chunk_const_ib = &p->chunks[i];
329
330 if (p->chunks[i].length_dw == 0)
331 return -EINVAL;
332 }
333 if (user_chunk.chunk_id == RADEON_CHUNK_ID_FLAGS) {
334 p->chunk_flags = &p->chunks[i];
335
336 if (p->chunks[i].length_dw == 0)
337 return -EINVAL;
338 }
339
340 size = p->chunks[i].length_dw;
341 cdata = (void __user *)(unsigned long)user_chunk.chunk_data;
342 p->chunks[i].user_ptr = cdata;
343 if (user_chunk.chunk_id == RADEON_CHUNK_ID_CONST_IB)
344 continue;
345
346 if (user_chunk.chunk_id == RADEON_CHUNK_ID_IB) {
347 if (!p->rdev || !(p->rdev->flags & RADEON_IS_AGP))
348 continue;
349 }
350
351 p->chunks[i].kdata = kvmalloc_array(size, sizeof(uint32_t), GFP_KERNEL);
352 size *= sizeof(uint32_t);
353 if (p->chunks[i].kdata == NULL) {
354 return -ENOMEM;
355 }
356 if (copy_from_user(p->chunks[i].kdata, cdata, size)) {
357 return -EFAULT;
358 }
359 if (user_chunk.chunk_id == RADEON_CHUNK_ID_FLAGS) {
360 p->cs_flags = p->chunks[i].kdata[0];
361 if (p->chunks[i].length_dw > 1)
362 ring = p->chunks[i].kdata[1];
363 if (p->chunks[i].length_dw > 2)
364 priority = (s32)p->chunks[i].kdata[2];
365 }
366 }
367
368
369 if (p->rdev) {
370 if ((p->cs_flags & RADEON_CS_USE_VM) &&
371 !p->rdev->vm_manager.enabled) {
372 DRM_ERROR("VM not active on asic!\n");
373 return -EINVAL;
374 }
375
376 if (radeon_cs_get_ring(p, ring, priority))
377 return -EINVAL;
378
379
380 if ((p->cs_flags & RADEON_CS_USE_VM) == 0) {
381 if (p->rdev->asic->ring[p->ring]->cs_parse == NULL) {
382 DRM_ERROR("Ring %d requires VM!\n", p->ring);
383 return -EINVAL;
384 }
385 } else {
386 if (p->rdev->asic->ring[p->ring]->ib_parse == NULL) {
387 DRM_ERROR("VM not supported on ring %d!\n",
388 p->ring);
389 return -EINVAL;
390 }
391 }
392 }
393
394 return 0;
395 }
396
397 static int cmp_size_smaller_first(void *priv, struct list_head *a,
398 struct list_head *b)
399 {
400 struct radeon_bo_list *la = list_entry(a, struct radeon_bo_list, tv.head);
401 struct radeon_bo_list *lb = list_entry(b, struct radeon_bo_list, tv.head);
402
403
404 return (int)la->robj->tbo.num_pages - (int)lb->robj->tbo.num_pages;
405 }
406
407
408
409
410
411
412
413
414
415 static void radeon_cs_parser_fini(struct radeon_cs_parser *parser, int error, bool backoff)
416 {
417 unsigned i;
418
419 if (!error) {
420
421
422
423
424
425
426
427
428
429
430 list_sort(NULL, &parser->validated, cmp_size_smaller_first);
431
432 ttm_eu_fence_buffer_objects(&parser->ticket,
433 &parser->validated,
434 &parser->ib.fence->base);
435 } else if (backoff) {
436 ttm_eu_backoff_reservation(&parser->ticket,
437 &parser->validated);
438 }
439
440 if (parser->relocs != NULL) {
441 for (i = 0; i < parser->nrelocs; i++) {
442 struct radeon_bo *bo = parser->relocs[i].robj;
443 if (bo == NULL)
444 continue;
445
446 drm_gem_object_put_unlocked(&bo->tbo.base);
447 }
448 }
449 kfree(parser->track);
450 kvfree(parser->relocs);
451 kvfree(parser->vm_bos);
452 for (i = 0; i < parser->nchunks; i++)
453 kvfree(parser->chunks[i].kdata);
454 kfree(parser->chunks);
455 kfree(parser->chunks_array);
456 radeon_ib_free(parser->rdev, &parser->ib);
457 radeon_ib_free(parser->rdev, &parser->const_ib);
458 }
459
460 static int radeon_cs_ib_chunk(struct radeon_device *rdev,
461 struct radeon_cs_parser *parser)
462 {
463 int r;
464
465 if (parser->chunk_ib == NULL)
466 return 0;
467
468 if (parser->cs_flags & RADEON_CS_USE_VM)
469 return 0;
470
471 r = radeon_cs_parse(rdev, parser->ring, parser);
472 if (r || parser->parser_error) {
473 DRM_ERROR("Invalid command stream !\n");
474 return r;
475 }
476
477 r = radeon_cs_sync_rings(parser);
478 if (r) {
479 if (r != -ERESTARTSYS)
480 DRM_ERROR("Failed to sync rings: %i\n", r);
481 return r;
482 }
483
484 if (parser->ring == R600_RING_TYPE_UVD_INDEX)
485 radeon_uvd_note_usage(rdev);
486 else if ((parser->ring == TN_RING_TYPE_VCE1_INDEX) ||
487 (parser->ring == TN_RING_TYPE_VCE2_INDEX))
488 radeon_vce_note_usage(rdev);
489
490 r = radeon_ib_schedule(rdev, &parser->ib, NULL, true);
491 if (r) {
492 DRM_ERROR("Failed to schedule IB !\n");
493 }
494 return r;
495 }
496
497 static int radeon_bo_vm_update_pte(struct radeon_cs_parser *p,
498 struct radeon_vm *vm)
499 {
500 struct radeon_device *rdev = p->rdev;
501 struct radeon_bo_va *bo_va;
502 int i, r;
503
504 r = radeon_vm_update_page_directory(rdev, vm);
505 if (r)
506 return r;
507
508 r = radeon_vm_clear_freed(rdev, vm);
509 if (r)
510 return r;
511
512 if (vm->ib_bo_va == NULL) {
513 DRM_ERROR("Tmp BO not in VM!\n");
514 return -EINVAL;
515 }
516
517 r = radeon_vm_bo_update(rdev, vm->ib_bo_va,
518 &rdev->ring_tmp_bo.bo->tbo.mem);
519 if (r)
520 return r;
521
522 for (i = 0; i < p->nrelocs; i++) {
523 struct radeon_bo *bo;
524
525 bo = p->relocs[i].robj;
526 bo_va = radeon_vm_bo_find(vm, bo);
527 if (bo_va == NULL) {
528 dev_err(rdev->dev, "bo %p not in vm %p\n", bo, vm);
529 return -EINVAL;
530 }
531
532 r = radeon_vm_bo_update(rdev, bo_va, &bo->tbo.mem);
533 if (r)
534 return r;
535
536 radeon_sync_fence(&p->ib.sync, bo_va->last_pt_update);
537 }
538
539 return radeon_vm_clear_invalids(rdev, vm);
540 }
541
542 static int radeon_cs_ib_vm_chunk(struct radeon_device *rdev,
543 struct radeon_cs_parser *parser)
544 {
545 struct radeon_fpriv *fpriv = parser->filp->driver_priv;
546 struct radeon_vm *vm = &fpriv->vm;
547 int r;
548
549 if (parser->chunk_ib == NULL)
550 return 0;
551 if ((parser->cs_flags & RADEON_CS_USE_VM) == 0)
552 return 0;
553
554 if (parser->const_ib.length_dw) {
555 r = radeon_ring_ib_parse(rdev, parser->ring, &parser->const_ib);
556 if (r) {
557 return r;
558 }
559 }
560
561 r = radeon_ring_ib_parse(rdev, parser->ring, &parser->ib);
562 if (r) {
563 return r;
564 }
565
566 if (parser->ring == R600_RING_TYPE_UVD_INDEX)
567 radeon_uvd_note_usage(rdev);
568
569 mutex_lock(&vm->mutex);
570 r = radeon_bo_vm_update_pte(parser, vm);
571 if (r) {
572 goto out;
573 }
574
575 r = radeon_cs_sync_rings(parser);
576 if (r) {
577 if (r != -ERESTARTSYS)
578 DRM_ERROR("Failed to sync rings: %i\n", r);
579 goto out;
580 }
581
582 if ((rdev->family >= CHIP_TAHITI) &&
583 (parser->chunk_const_ib != NULL)) {
584 r = radeon_ib_schedule(rdev, &parser->ib, &parser->const_ib, true);
585 } else {
586 r = radeon_ib_schedule(rdev, &parser->ib, NULL, true);
587 }
588
589 out:
590 mutex_unlock(&vm->mutex);
591 return r;
592 }
593
594 static int radeon_cs_handle_lockup(struct radeon_device *rdev, int r)
595 {
596 if (r == -EDEADLK) {
597 r = radeon_gpu_reset(rdev);
598 if (!r)
599 r = -EAGAIN;
600 }
601 return r;
602 }
603
604 static int radeon_cs_ib_fill(struct radeon_device *rdev, struct radeon_cs_parser *parser)
605 {
606 struct radeon_cs_chunk *ib_chunk;
607 struct radeon_vm *vm = NULL;
608 int r;
609
610 if (parser->chunk_ib == NULL)
611 return 0;
612
613 if (parser->cs_flags & RADEON_CS_USE_VM) {
614 struct radeon_fpriv *fpriv = parser->filp->driver_priv;
615 vm = &fpriv->vm;
616
617 if ((rdev->family >= CHIP_TAHITI) &&
618 (parser->chunk_const_ib != NULL)) {
619 ib_chunk = parser->chunk_const_ib;
620 if (ib_chunk->length_dw > RADEON_IB_VM_MAX_SIZE) {
621 DRM_ERROR("cs IB CONST too big: %d\n", ib_chunk->length_dw);
622 return -EINVAL;
623 }
624 r = radeon_ib_get(rdev, parser->ring, &parser->const_ib,
625 vm, ib_chunk->length_dw * 4);
626 if (r) {
627 DRM_ERROR("Failed to get const ib !\n");
628 return r;
629 }
630 parser->const_ib.is_const_ib = true;
631 parser->const_ib.length_dw = ib_chunk->length_dw;
632 if (copy_from_user(parser->const_ib.ptr,
633 ib_chunk->user_ptr,
634 ib_chunk->length_dw * 4))
635 return -EFAULT;
636 }
637
638 ib_chunk = parser->chunk_ib;
639 if (ib_chunk->length_dw > RADEON_IB_VM_MAX_SIZE) {
640 DRM_ERROR("cs IB too big: %d\n", ib_chunk->length_dw);
641 return -EINVAL;
642 }
643 }
644 ib_chunk = parser->chunk_ib;
645
646 r = radeon_ib_get(rdev, parser->ring, &parser->ib,
647 vm, ib_chunk->length_dw * 4);
648 if (r) {
649 DRM_ERROR("Failed to get ib !\n");
650 return r;
651 }
652 parser->ib.length_dw = ib_chunk->length_dw;
653 if (ib_chunk->kdata)
654 memcpy(parser->ib.ptr, ib_chunk->kdata, ib_chunk->length_dw * 4);
655 else if (copy_from_user(parser->ib.ptr, ib_chunk->user_ptr, ib_chunk->length_dw * 4))
656 return -EFAULT;
657 return 0;
658 }
659
660 int radeon_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
661 {
662 struct radeon_device *rdev = dev->dev_private;
663 struct radeon_cs_parser parser;
664 int r;
665
666 down_read(&rdev->exclusive_lock);
667 if (!rdev->accel_working) {
668 up_read(&rdev->exclusive_lock);
669 return -EBUSY;
670 }
671 if (rdev->in_reset) {
672 up_read(&rdev->exclusive_lock);
673 r = radeon_gpu_reset(rdev);
674 if (!r)
675 r = -EAGAIN;
676 return r;
677 }
678
679 memset(&parser, 0, sizeof(struct radeon_cs_parser));
680 parser.filp = filp;
681 parser.rdev = rdev;
682 parser.dev = rdev->dev;
683 parser.family = rdev->family;
684 r = radeon_cs_parser_init(&parser, data);
685 if (r) {
686 DRM_ERROR("Failed to initialize parser !\n");
687 radeon_cs_parser_fini(&parser, r, false);
688 up_read(&rdev->exclusive_lock);
689 r = radeon_cs_handle_lockup(rdev, r);
690 return r;
691 }
692
693 r = radeon_cs_ib_fill(rdev, &parser);
694 if (!r) {
695 r = radeon_cs_parser_relocs(&parser);
696 if (r && r != -ERESTARTSYS)
697 DRM_ERROR("Failed to parse relocation %d!\n", r);
698 }
699
700 if (r) {
701 radeon_cs_parser_fini(&parser, r, false);
702 up_read(&rdev->exclusive_lock);
703 r = radeon_cs_handle_lockup(rdev, r);
704 return r;
705 }
706
707 trace_radeon_cs(&parser);
708
709 r = radeon_cs_ib_chunk(rdev, &parser);
710 if (r) {
711 goto out;
712 }
713 r = radeon_cs_ib_vm_chunk(rdev, &parser);
714 if (r) {
715 goto out;
716 }
717 out:
718 radeon_cs_parser_fini(&parser, r, true);
719 up_read(&rdev->exclusive_lock);
720 r = radeon_cs_handle_lockup(rdev, r);
721 return r;
722 }
723
724
725
726
727
728
729
730
731
732 int radeon_cs_packet_parse(struct radeon_cs_parser *p,
733 struct radeon_cs_packet *pkt,
734 unsigned idx)
735 {
736 struct radeon_cs_chunk *ib_chunk = p->chunk_ib;
737 struct radeon_device *rdev = p->rdev;
738 uint32_t header;
739 int ret = 0, i;
740
741 if (idx >= ib_chunk->length_dw) {
742 DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
743 idx, ib_chunk->length_dw);
744 return -EINVAL;
745 }
746 header = radeon_get_ib_value(p, idx);
747 pkt->idx = idx;
748 pkt->type = RADEON_CP_PACKET_GET_TYPE(header);
749 pkt->count = RADEON_CP_PACKET_GET_COUNT(header);
750 pkt->one_reg_wr = 0;
751 switch (pkt->type) {
752 case RADEON_PACKET_TYPE0:
753 if (rdev->family < CHIP_R600) {
754 pkt->reg = R100_CP_PACKET0_GET_REG(header);
755 pkt->one_reg_wr =
756 RADEON_CP_PACKET0_GET_ONE_REG_WR(header);
757 } else
758 pkt->reg = R600_CP_PACKET0_GET_REG(header);
759 break;
760 case RADEON_PACKET_TYPE3:
761 pkt->opcode = RADEON_CP_PACKET3_GET_OPCODE(header);
762 break;
763 case RADEON_PACKET_TYPE2:
764 pkt->count = -1;
765 break;
766 default:
767 DRM_ERROR("Unknown packet type %d at %d !\n", pkt->type, idx);
768 ret = -EINVAL;
769 goto dump_ib;
770 }
771 if ((pkt->count + 1 + pkt->idx) >= ib_chunk->length_dw) {
772 DRM_ERROR("Packet (%d:%d:%d) end after CS buffer (%d) !\n",
773 pkt->idx, pkt->type, pkt->count, ib_chunk->length_dw);
774 ret = -EINVAL;
775 goto dump_ib;
776 }
777 return 0;
778
779 dump_ib:
780 for (i = 0; i < ib_chunk->length_dw; i++) {
781 if (i == idx)
782 printk("\t0x%08x <---\n", radeon_get_ib_value(p, i));
783 else
784 printk("\t0x%08x\n", radeon_get_ib_value(p, i));
785 }
786 return ret;
787 }
788
789
790
791
792
793
794
795 bool radeon_cs_packet_next_is_pkt3_nop(struct radeon_cs_parser *p)
796 {
797 struct radeon_cs_packet p3reloc;
798 int r;
799
800 r = radeon_cs_packet_parse(p, &p3reloc, p->idx);
801 if (r)
802 return false;
803 if (p3reloc.type != RADEON_PACKET_TYPE3)
804 return false;
805 if (p3reloc.opcode != RADEON_PACKET3_NOP)
806 return false;
807 return true;
808 }
809
810
811
812
813
814
815
816
817 void radeon_cs_dump_packet(struct radeon_cs_parser *p,
818 struct radeon_cs_packet *pkt)
819 {
820 volatile uint32_t *ib;
821 unsigned i;
822 unsigned idx;
823
824 ib = p->ib.ptr;
825 idx = pkt->idx;
826 for (i = 0; i <= (pkt->count + 1); i++, idx++)
827 DRM_INFO("ib[%d]=0x%08X\n", idx, ib[idx]);
828 }
829
830
831
832
833
834
835
836
837
838
839
840
841 int radeon_cs_packet_next_reloc(struct radeon_cs_parser *p,
842 struct radeon_bo_list **cs_reloc,
843 int nomm)
844 {
845 struct radeon_cs_chunk *relocs_chunk;
846 struct radeon_cs_packet p3reloc;
847 unsigned idx;
848 int r;
849
850 if (p->chunk_relocs == NULL) {
851 DRM_ERROR("No relocation chunk !\n");
852 return -EINVAL;
853 }
854 *cs_reloc = NULL;
855 relocs_chunk = p->chunk_relocs;
856 r = radeon_cs_packet_parse(p, &p3reloc, p->idx);
857 if (r)
858 return r;
859 p->idx += p3reloc.count + 2;
860 if (p3reloc.type != RADEON_PACKET_TYPE3 ||
861 p3reloc.opcode != RADEON_PACKET3_NOP) {
862 DRM_ERROR("No packet3 for relocation for packet at %d.\n",
863 p3reloc.idx);
864 radeon_cs_dump_packet(p, &p3reloc);
865 return -EINVAL;
866 }
867 idx = radeon_get_ib_value(p, p3reloc.idx + 1);
868 if (idx >= relocs_chunk->length_dw) {
869 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
870 idx, relocs_chunk->length_dw);
871 radeon_cs_dump_packet(p, &p3reloc);
872 return -EINVAL;
873 }
874
875 if (nomm) {
876 *cs_reloc = p->relocs;
877 (*cs_reloc)->gpu_offset =
878 (u64)relocs_chunk->kdata[idx + 3] << 32;
879 (*cs_reloc)->gpu_offset |= relocs_chunk->kdata[idx + 0];
880 } else
881 *cs_reloc = &p->relocs[(idx / 4)];
882 return 0;
883 }