2 * Copyright 2008 Jerome Glisse.
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
20 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
22 * DEALINGS IN THE SOFTWARE.
28 #include <linux/list_sort.h>
29 #include <linux/pci.h>
30 #include <linux/uaccess.h>
32 #include <drm/drm_device.h>
33 #include <drm/drm_file.h>
34 #include <drm/radeon_drm.h>
37 #include "radeon_reg.h"
38 #include "radeon_trace.h"
40 #define RADEON_CS_MAX_PRIORITY 32u
41 #define RADEON_CS_NUM_BUCKETS (RADEON_CS_MAX_PRIORITY + 1)
43 /* This is based on the bucket sort with O(n) time complexity.
44 * An item with priority "i" is added to bucket[i]. The lists are then
45 * concatenated in descending order.
47 struct radeon_cs_buckets {
48 struct list_head bucket[RADEON_CS_NUM_BUCKETS];
51 static void radeon_cs_buckets_init(struct radeon_cs_buckets *b)
55 for (i = 0; i < RADEON_CS_NUM_BUCKETS; i++)
56 INIT_LIST_HEAD(&b->bucket[i]);
59 static void radeon_cs_buckets_add(struct radeon_cs_buckets *b,
60 struct list_head *item, unsigned priority)
62 /* Since buffers which appear sooner in the relocation list are
63 * likely to be used more often than buffers which appear later
64 * in the list, the sort mustn't change the ordering of buffers
65 * with the same priority, i.e. it must be stable.
67 list_add_tail(item, &b->bucket[min(priority, RADEON_CS_MAX_PRIORITY)]);
70 static void radeon_cs_buckets_get_list(struct radeon_cs_buckets *b,
71 struct list_head *out_list)
75 /* Connect the sorted buckets in the output list. */
76 for (i = 0; i < RADEON_CS_NUM_BUCKETS; i++) {
77 list_splice(&b->bucket[i], out_list);
81 static int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
83 struct radeon_cs_chunk *chunk;
84 struct radeon_cs_buckets buckets;
86 bool need_mmap_lock = false;
89 if (p->chunk_relocs == NULL) {
92 chunk = p->chunk_relocs;
94 /* FIXME: we assume that each relocs use 4 dwords */
95 p->nrelocs = chunk->length_dw / 4;
96 p->relocs = kvcalloc(p->nrelocs, sizeof(struct radeon_bo_list),
98 if (p->relocs == NULL) {
102 radeon_cs_buckets_init(&buckets);
104 for (i = 0; i < p->nrelocs; i++) {
105 struct drm_radeon_cs_reloc *r;
106 struct drm_gem_object *gobj;
109 r = (struct drm_radeon_cs_reloc *)&chunk->kdata[i*4];
110 gobj = drm_gem_object_lookup(p->filp, r->handle);
112 DRM_ERROR("gem object lookup failed 0x%x\n",
116 p->relocs[i].robj = gem_to_radeon_bo(gobj);
118 /* The userspace buffer priorities are from 0 to 15. A higher
119 * number means the buffer is more important.
120 * Also, the buffers used for write have a higher priority than
121 * the buffers used for read only, which doubles the range
122 * to 0 to 31. 32 is reserved for the kernel driver.
124 priority = (r->flags & RADEON_RELOC_PRIO_MASK) * 2
127 /* The first reloc of an UVD job is the msg and that must be in
128 * VRAM, the second reloc is the DPB and for WMV that must be in
129 * VRAM as well. Also put everything into VRAM on AGP cards and older
130 * IGP chips to avoid image corruptions
132 if (p->ring == R600_RING_TYPE_UVD_INDEX &&
133 (i <= 0 || pci_find_capability(p->rdev->pdev, PCI_CAP_ID_AGP) ||
134 p->rdev->family == CHIP_RS780 ||
135 p->rdev->family == CHIP_RS880)) {
137 /* TODO: is this still needed for NI+ ? */
138 p->relocs[i].preferred_domains =
139 RADEON_GEM_DOMAIN_VRAM;
141 p->relocs[i].allowed_domains =
142 RADEON_GEM_DOMAIN_VRAM;
144 /* prioritize this over any other relocation */
145 priority = RADEON_CS_MAX_PRIORITY;
147 uint32_t domain = r->write_domain ?
148 r->write_domain : r->read_domains;
150 if (domain & RADEON_GEM_DOMAIN_CPU) {
151 DRM_ERROR("RADEON_GEM_DOMAIN_CPU is not valid "
152 "for command submission\n");
156 p->relocs[i].preferred_domains = domain;
157 if (domain == RADEON_GEM_DOMAIN_VRAM)
158 domain |= RADEON_GEM_DOMAIN_GTT;
159 p->relocs[i].allowed_domains = domain;
162 if (radeon_ttm_tt_has_userptr(p->rdev, p->relocs[i].robj->tbo.ttm)) {
163 uint32_t domain = p->relocs[i].preferred_domains;
164 if (!(domain & RADEON_GEM_DOMAIN_GTT)) {
165 DRM_ERROR("Only RADEON_GEM_DOMAIN_GTT is "
166 "allowed for userptr BOs\n");
169 need_mmap_lock = true;
170 domain = RADEON_GEM_DOMAIN_GTT;
171 p->relocs[i].preferred_domains = domain;
172 p->relocs[i].allowed_domains = domain;
175 /* Objects shared as dma-bufs cannot be moved to VRAM */
176 if (p->relocs[i].robj->prime_shared_count) {
177 p->relocs[i].allowed_domains &= ~RADEON_GEM_DOMAIN_VRAM;
178 if (!p->relocs[i].allowed_domains) {
179 DRM_ERROR("BO associated with dma-buf cannot "
180 "be moved to VRAM\n");
185 p->relocs[i].tv.bo = &p->relocs[i].robj->tbo;
186 p->relocs[i].tv.num_shared = !r->write_domain;
188 radeon_cs_buckets_add(&buckets, &p->relocs[i].tv.head,
192 radeon_cs_buckets_get_list(&buckets, &p->validated);
194 if (p->cs_flags & RADEON_CS_USE_VM)
195 p->vm_bos = radeon_vm_get_bos(p->rdev, p->ib.vm,
198 mmap_read_lock(current->mm);
200 r = radeon_bo_list_validate(p->rdev, &p->ticket, &p->validated, p->ring);
203 mmap_read_unlock(current->mm);
208 static int radeon_cs_get_ring(struct radeon_cs_parser *p, u32 ring, s32 priority)
210 p->priority = priority;
214 DRM_ERROR("unknown ring id: %d\n", ring);
216 case RADEON_CS_RING_GFX:
217 p->ring = RADEON_RING_TYPE_GFX_INDEX;
219 case RADEON_CS_RING_COMPUTE:
220 if (p->rdev->family >= CHIP_TAHITI) {
222 p->ring = CAYMAN_RING_TYPE_CP1_INDEX;
224 p->ring = CAYMAN_RING_TYPE_CP2_INDEX;
226 p->ring = RADEON_RING_TYPE_GFX_INDEX;
228 case RADEON_CS_RING_DMA:
229 if (p->rdev->family >= CHIP_CAYMAN) {
231 p->ring = R600_RING_TYPE_DMA_INDEX;
233 p->ring = CAYMAN_RING_TYPE_DMA1_INDEX;
234 } else if (p->rdev->family >= CHIP_RV770) {
235 p->ring = R600_RING_TYPE_DMA_INDEX;
240 case RADEON_CS_RING_UVD:
241 p->ring = R600_RING_TYPE_UVD_INDEX;
243 case RADEON_CS_RING_VCE:
244 /* TODO: only use the low priority ring for now */
245 p->ring = TN_RING_TYPE_VCE1_INDEX;
251 static int radeon_cs_sync_rings(struct radeon_cs_parser *p)
253 struct radeon_bo_list *reloc;
256 list_for_each_entry(reloc, &p->validated, tv.head) {
257 struct dma_resv *resv;
259 resv = reloc->robj->tbo.base.resv;
260 r = radeon_sync_resv(p->rdev, &p->ib.sync, resv,
261 reloc->tv.num_shared);
268 /* XXX: note that this is called from the legacy UMS CS ioctl as well */
269 int radeon_cs_parser_init(struct radeon_cs_parser *p, void *data)
271 struct drm_radeon_cs *cs = data;
272 uint64_t *chunk_array_ptr;
275 u32 ring = RADEON_CS_RING_GFX;
278 INIT_LIST_HEAD(&p->validated);
280 if (!cs->num_chunks) {
287 p->const_ib.sa_bo = NULL;
289 p->chunk_relocs = NULL;
290 p->chunk_flags = NULL;
291 p->chunk_const_ib = NULL;
292 p->chunks_array = kvmalloc_array(cs->num_chunks, sizeof(uint64_t), GFP_KERNEL);
293 if (p->chunks_array == NULL) {
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)) {
302 p->nchunks = cs->num_chunks;
303 p->chunks = kvcalloc(p->nchunks, sizeof(struct radeon_cs_chunk), GFP_KERNEL);
304 if (p->chunks == NULL) {
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;
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))) {
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];
321 if (user_chunk.chunk_id == RADEON_CHUNK_ID_IB) {
322 p->chunk_ib = &p->chunks[i];
323 /* zero length IB isn't useful */
324 if (p->chunks[i].length_dw == 0)
327 if (user_chunk.chunk_id == RADEON_CHUNK_ID_CONST_IB) {
328 p->chunk_const_ib = &p->chunks[i];
329 /* zero length CONST IB isn't useful */
330 if (p->chunks[i].length_dw == 0)
333 if (user_chunk.chunk_id == RADEON_CHUNK_ID_FLAGS) {
334 p->chunk_flags = &p->chunks[i];
335 /* zero length flags aren't useful */
336 if (p->chunks[i].length_dw == 0)
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)
346 if (user_chunk.chunk_id == RADEON_CHUNK_ID_IB) {
347 if (!p->rdev || !(p->rdev->flags & RADEON_IS_AGP))
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) {
356 if (copy_from_user(p->chunks[i].kdata, cdata, size)) {
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];
368 /* these are KMS only */
370 if ((p->cs_flags & RADEON_CS_USE_VM) &&
371 !p->rdev->vm_manager.enabled) {
372 DRM_ERROR("VM not active on asic!\n");
376 if (radeon_cs_get_ring(p, ring, priority))
379 /* we only support VM on some SI+ rings */
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);
386 if (p->rdev->asic->ring[p->ring]->ib_parse == NULL) {
387 DRM_ERROR("VM not supported on ring %d!\n",
397 static int cmp_size_smaller_first(void *priv, const struct list_head *a,
398 const struct list_head *b)
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);
403 /* Sort A before B if A is smaller. */
404 if (la->robj->tbo.base.size > lb->robj->tbo.base.size)
406 if (la->robj->tbo.base.size < lb->robj->tbo.base.size)
412 * radeon_cs_parser_fini() - clean parser states
413 * @parser: parser structure holding parsing context.
414 * @error: error number
415 * @backoff: indicator to backoff the reservation
417 * If error is set than unvalidate buffer, otherwise just free memory
418 * used by parsing context.
420 static void radeon_cs_parser_fini(struct radeon_cs_parser *parser, int error, bool backoff)
425 /* Sort the buffer list from the smallest to largest buffer,
426 * which affects the order of buffers in the LRU list.
427 * This assures that the smallest buffers are added first
428 * to the LRU list, so they are likely to be later evicted
429 * first, instead of large buffers whose eviction is more
432 * This slightly lowers the number of bytes moved by TTM
433 * per frame under memory pressure.
435 list_sort(NULL, &parser->validated, cmp_size_smaller_first);
437 ttm_eu_fence_buffer_objects(&parser->ticket,
439 &parser->ib.fence->base);
440 } else if (backoff) {
441 ttm_eu_backoff_reservation(&parser->ticket,
445 if (parser->relocs != NULL) {
446 for (i = 0; i < parser->nrelocs; i++) {
447 struct radeon_bo *bo = parser->relocs[i].robj;
451 drm_gem_object_put(&bo->tbo.base);
454 kfree(parser->track);
455 kvfree(parser->relocs);
456 kvfree(parser->vm_bos);
457 for (i = 0; i < parser->nchunks; i++)
458 kvfree(parser->chunks[i].kdata);
459 kvfree(parser->chunks);
460 kvfree(parser->chunks_array);
461 radeon_ib_free(parser->rdev, &parser->ib);
462 radeon_ib_free(parser->rdev, &parser->const_ib);
465 static int radeon_cs_ib_chunk(struct radeon_device *rdev,
466 struct radeon_cs_parser *parser)
470 if (parser->chunk_ib == NULL)
473 if (parser->cs_flags & RADEON_CS_USE_VM)
476 r = radeon_cs_parse(rdev, parser->ring, parser);
477 if (r || parser->parser_error) {
478 DRM_ERROR("Invalid command stream !\n");
482 r = radeon_cs_sync_rings(parser);
484 if (r != -ERESTARTSYS)
485 DRM_ERROR("Failed to sync rings: %i\n", r);
489 if (parser->ring == R600_RING_TYPE_UVD_INDEX)
490 radeon_uvd_note_usage(rdev);
491 else if ((parser->ring == TN_RING_TYPE_VCE1_INDEX) ||
492 (parser->ring == TN_RING_TYPE_VCE2_INDEX))
493 radeon_vce_note_usage(rdev);
495 r = radeon_ib_schedule(rdev, &parser->ib, NULL, true);
497 DRM_ERROR("Failed to schedule IB !\n");
502 static int radeon_bo_vm_update_pte(struct radeon_cs_parser *p,
503 struct radeon_vm *vm)
505 struct radeon_device *rdev = p->rdev;
506 struct radeon_bo_va *bo_va;
509 r = radeon_vm_update_page_directory(rdev, vm);
513 r = radeon_vm_clear_freed(rdev, vm);
517 if (vm->ib_bo_va == NULL) {
518 DRM_ERROR("Tmp BO not in VM!\n");
522 r = radeon_vm_bo_update(rdev, vm->ib_bo_va,
523 rdev->ring_tmp_bo.bo->tbo.resource);
527 for (i = 0; i < p->nrelocs; i++) {
528 struct radeon_bo *bo;
530 bo = p->relocs[i].robj;
531 bo_va = radeon_vm_bo_find(vm, bo);
533 dev_err(rdev->dev, "bo %p not in vm %p\n", bo, vm);
537 r = radeon_vm_bo_update(rdev, bo_va, bo->tbo.resource);
541 radeon_sync_fence(&p->ib.sync, bo_va->last_pt_update);
543 r = dma_resv_reserve_fences(bo->tbo.base.resv, 1);
548 return radeon_vm_clear_invalids(rdev, vm);
551 static int radeon_cs_ib_vm_chunk(struct radeon_device *rdev,
552 struct radeon_cs_parser *parser)
554 struct radeon_fpriv *fpriv = parser->filp->driver_priv;
555 struct radeon_vm *vm = &fpriv->vm;
558 if (parser->chunk_ib == NULL)
560 if ((parser->cs_flags & RADEON_CS_USE_VM) == 0)
563 if (parser->const_ib.length_dw) {
564 r = radeon_ring_ib_parse(rdev, parser->ring, &parser->const_ib);
570 r = radeon_ring_ib_parse(rdev, parser->ring, &parser->ib);
575 if (parser->ring == R600_RING_TYPE_UVD_INDEX)
576 radeon_uvd_note_usage(rdev);
578 mutex_lock(&vm->mutex);
579 r = radeon_bo_vm_update_pte(parser, vm);
584 r = radeon_cs_sync_rings(parser);
586 if (r != -ERESTARTSYS)
587 DRM_ERROR("Failed to sync rings: %i\n", r);
591 if ((rdev->family >= CHIP_TAHITI) &&
592 (parser->chunk_const_ib != NULL)) {
593 r = radeon_ib_schedule(rdev, &parser->ib, &parser->const_ib, true);
595 r = radeon_ib_schedule(rdev, &parser->ib, NULL, true);
599 mutex_unlock(&vm->mutex);
603 static int radeon_cs_handle_lockup(struct radeon_device *rdev, int r)
606 r = radeon_gpu_reset(rdev);
613 static int radeon_cs_ib_fill(struct radeon_device *rdev, struct radeon_cs_parser *parser)
615 struct radeon_cs_chunk *ib_chunk;
616 struct radeon_vm *vm = NULL;
619 if (parser->chunk_ib == NULL)
622 if (parser->cs_flags & RADEON_CS_USE_VM) {
623 struct radeon_fpriv *fpriv = parser->filp->driver_priv;
626 if ((rdev->family >= CHIP_TAHITI) &&
627 (parser->chunk_const_ib != NULL)) {
628 ib_chunk = parser->chunk_const_ib;
629 if (ib_chunk->length_dw > RADEON_IB_VM_MAX_SIZE) {
630 DRM_ERROR("cs IB CONST too big: %d\n", ib_chunk->length_dw);
633 r = radeon_ib_get(rdev, parser->ring, &parser->const_ib,
634 vm, ib_chunk->length_dw * 4);
636 DRM_ERROR("Failed to get const ib !\n");
639 parser->const_ib.is_const_ib = true;
640 parser->const_ib.length_dw = ib_chunk->length_dw;
641 if (copy_from_user(parser->const_ib.ptr,
643 ib_chunk->length_dw * 4))
647 ib_chunk = parser->chunk_ib;
648 if (ib_chunk->length_dw > RADEON_IB_VM_MAX_SIZE) {
649 DRM_ERROR("cs IB too big: %d\n", ib_chunk->length_dw);
653 ib_chunk = parser->chunk_ib;
655 r = radeon_ib_get(rdev, parser->ring, &parser->ib,
656 vm, ib_chunk->length_dw * 4);
658 DRM_ERROR("Failed to get ib !\n");
661 parser->ib.length_dw = ib_chunk->length_dw;
663 memcpy(parser->ib.ptr, ib_chunk->kdata, ib_chunk->length_dw * 4);
664 else if (copy_from_user(parser->ib.ptr, ib_chunk->user_ptr, ib_chunk->length_dw * 4))
669 int radeon_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
671 struct radeon_device *rdev = dev->dev_private;
672 struct radeon_cs_parser parser;
675 down_read(&rdev->exclusive_lock);
676 if (!rdev->accel_working) {
677 up_read(&rdev->exclusive_lock);
680 if (rdev->in_reset) {
681 up_read(&rdev->exclusive_lock);
682 r = radeon_gpu_reset(rdev);
687 /* initialize parser */
688 memset(&parser, 0, sizeof(struct radeon_cs_parser));
691 parser.dev = rdev->dev;
692 parser.family = rdev->family;
693 r = radeon_cs_parser_init(&parser, data);
695 DRM_ERROR("Failed to initialize parser !\n");
696 radeon_cs_parser_fini(&parser, r, false);
697 up_read(&rdev->exclusive_lock);
698 r = radeon_cs_handle_lockup(rdev, r);
702 r = radeon_cs_ib_fill(rdev, &parser);
704 r = radeon_cs_parser_relocs(&parser);
705 if (r && r != -ERESTARTSYS)
706 DRM_ERROR("Failed to parse relocation %d!\n", r);
710 radeon_cs_parser_fini(&parser, r, false);
711 up_read(&rdev->exclusive_lock);
712 r = radeon_cs_handle_lockup(rdev, r);
716 trace_radeon_cs(&parser);
718 r = radeon_cs_ib_chunk(rdev, &parser);
722 r = radeon_cs_ib_vm_chunk(rdev, &parser);
727 radeon_cs_parser_fini(&parser, r, true);
728 up_read(&rdev->exclusive_lock);
729 r = radeon_cs_handle_lockup(rdev, r);
734 * radeon_cs_packet_parse() - parse cp packet and point ib index to next packet
735 * @p: parser structure holding parsing context.
736 * @pkt: where to store packet information
739 * Assume that chunk_ib_index is properly set. Will return -EINVAL
740 * if packet is bigger than remaining ib size. or if packets is unknown.
742 int radeon_cs_packet_parse(struct radeon_cs_parser *p,
743 struct radeon_cs_packet *pkt,
746 struct radeon_cs_chunk *ib_chunk = p->chunk_ib;
747 struct radeon_device *rdev = p->rdev;
751 if (idx >= ib_chunk->length_dw) {
752 DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
753 idx, ib_chunk->length_dw);
756 header = radeon_get_ib_value(p, idx);
758 pkt->type = RADEON_CP_PACKET_GET_TYPE(header);
759 pkt->count = RADEON_CP_PACKET_GET_COUNT(header);
762 case RADEON_PACKET_TYPE0:
763 if (rdev->family < CHIP_R600) {
764 pkt->reg = R100_CP_PACKET0_GET_REG(header);
766 RADEON_CP_PACKET0_GET_ONE_REG_WR(header);
768 pkt->reg = R600_CP_PACKET0_GET_REG(header);
770 case RADEON_PACKET_TYPE3:
771 pkt->opcode = RADEON_CP_PACKET3_GET_OPCODE(header);
773 case RADEON_PACKET_TYPE2:
777 DRM_ERROR("Unknown packet type %d at %d !\n", pkt->type, idx);
781 if ((pkt->count + 1 + pkt->idx) >= ib_chunk->length_dw) {
782 DRM_ERROR("Packet (%d:%d:%d) end after CS buffer (%d) !\n",
783 pkt->idx, pkt->type, pkt->count, ib_chunk->length_dw);
790 for (i = 0; i < ib_chunk->length_dw; i++) {
792 printk("\t0x%08x <---\n", radeon_get_ib_value(p, i));
794 printk("\t0x%08x\n", radeon_get_ib_value(p, i));
800 * radeon_cs_packet_next_is_pkt3_nop() - test if the next packet is P3 NOP
801 * @p: structure holding the parser context.
803 * Check if the next packet is NOP relocation packet3.
805 bool radeon_cs_packet_next_is_pkt3_nop(struct radeon_cs_parser *p)
807 struct radeon_cs_packet p3reloc;
810 r = radeon_cs_packet_parse(p, &p3reloc, p->idx);
813 if (p3reloc.type != RADEON_PACKET_TYPE3)
815 if (p3reloc.opcode != RADEON_PACKET3_NOP)
821 * radeon_cs_dump_packet() - dump raw packet context
822 * @p: structure holding the parser context.
823 * @pkt: structure holding the packet.
825 * Used mostly for debugging and error reporting.
827 void radeon_cs_dump_packet(struct radeon_cs_parser *p,
828 struct radeon_cs_packet *pkt)
830 volatile uint32_t *ib;
836 for (i = 0; i <= (pkt->count + 1); i++, idx++)
837 DRM_INFO("ib[%d]=0x%08X\n", idx, ib[idx]);
841 * radeon_cs_packet_next_reloc() - parse next (should be reloc) packet
842 * @p: parser structure holding parsing context.
843 * @cs_reloc: reloc informations
844 * @nomm: no memory management for debugging
846 * Check if next packet is relocation packet3, do bo validation and compute
847 * GPU offset using the provided start.
849 int radeon_cs_packet_next_reloc(struct radeon_cs_parser *p,
850 struct radeon_bo_list **cs_reloc,
853 struct radeon_cs_chunk *relocs_chunk;
854 struct radeon_cs_packet p3reloc;
858 if (p->chunk_relocs == NULL) {
859 DRM_ERROR("No relocation chunk !\n");
863 relocs_chunk = p->chunk_relocs;
864 r = radeon_cs_packet_parse(p, &p3reloc, p->idx);
867 p->idx += p3reloc.count + 2;
868 if (p3reloc.type != RADEON_PACKET_TYPE3 ||
869 p3reloc.opcode != RADEON_PACKET3_NOP) {
870 DRM_ERROR("No packet3 for relocation for packet at %d.\n",
872 radeon_cs_dump_packet(p, &p3reloc);
875 idx = radeon_get_ib_value(p, p3reloc.idx + 1);
876 if (idx >= relocs_chunk->length_dw) {
877 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
878 idx, relocs_chunk->length_dw);
879 radeon_cs_dump_packet(p, &p3reloc);
882 /* FIXME: we assume reloc size is 4 dwords */
884 *cs_reloc = p->relocs;
885 (*cs_reloc)->gpu_offset =
886 (u64)relocs_chunk->kdata[idx + 3] << 32;
887 (*cs_reloc)->gpu_offset |= relocs_chunk->kdata[idx + 0];
889 *cs_reloc = &p->relocs[(idx / 4)];