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.
25 * Jerome Glisse <glisse@freedesktop.org>
27 #include <linux/list_sort.h>
29 #include <drm/radeon_drm.h>
30 #include "radeon_reg.h"
32 #include "radeon_trace.h"
34 #define RADEON_CS_MAX_PRIORITY 32u
35 #define RADEON_CS_NUM_BUCKETS (RADEON_CS_MAX_PRIORITY + 1)
37 /* This is based on the bucket sort with O(n) time complexity.
38 * An item with priority "i" is added to bucket[i]. The lists are then
39 * concatenated in descending order.
41 struct radeon_cs_buckets {
42 struct list_head bucket[RADEON_CS_NUM_BUCKETS];
45 static void radeon_cs_buckets_init(struct radeon_cs_buckets *b)
49 for (i = 0; i < RADEON_CS_NUM_BUCKETS; i++)
50 INIT_LIST_HEAD(&b->bucket[i]);
53 static void radeon_cs_buckets_add(struct radeon_cs_buckets *b,
54 struct list_head *item, unsigned priority)
56 /* Since buffers which appear sooner in the relocation list are
57 * likely to be used more often than buffers which appear later
58 * in the list, the sort mustn't change the ordering of buffers
59 * with the same priority, i.e. it must be stable.
61 list_add_tail(item, &b->bucket[min(priority, RADEON_CS_MAX_PRIORITY)]);
64 static void radeon_cs_buckets_get_list(struct radeon_cs_buckets *b,
65 struct list_head *out_list)
69 /* Connect the sorted buckets in the output list. */
70 for (i = 0; i < RADEON_CS_NUM_BUCKETS; i++) {
71 list_splice(&b->bucket[i], out_list);
75 static int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
77 struct drm_device *ddev = p->rdev->ddev;
78 struct radeon_cs_chunk *chunk;
79 struct radeon_cs_buckets buckets;
81 bool need_mmap_lock = false;
84 if (p->chunk_relocs == NULL) {
87 chunk = p->chunk_relocs;
89 /* FIXME: we assume that each relocs use 4 dwords */
90 p->nrelocs = chunk->length_dw / 4;
91 p->relocs = kcalloc(p->nrelocs, sizeof(struct radeon_bo_list), GFP_KERNEL);
92 if (p->relocs == NULL) {
96 radeon_cs_buckets_init(&buckets);
98 for (i = 0; i < p->nrelocs; i++) {
99 struct drm_radeon_cs_reloc *r;
100 struct drm_gem_object *gobj;
103 r = (struct drm_radeon_cs_reloc *)&chunk->kdata[i*4];
104 gobj = drm_gem_object_lookup(ddev, p->filp, r->handle);
106 DRM_ERROR("gem object lookup failed 0x%x\n",
110 p->relocs[i].robj = gem_to_radeon_bo(gobj);
112 /* The userspace buffer priorities are from 0 to 15. A higher
113 * number means the buffer is more important.
114 * Also, the buffers used for write have a higher priority than
115 * the buffers used for read only, which doubles the range
116 * to 0 to 31. 32 is reserved for the kernel driver.
118 priority = (r->flags & RADEON_RELOC_PRIO_MASK) * 2
121 /* the first reloc of an UVD job is the msg and that must be in
122 VRAM, also but everything into VRAM on AGP cards and older
123 IGP chips to avoid image corruptions */
124 if (p->ring == R600_RING_TYPE_UVD_INDEX &&
125 (i == 0 || drm_pci_device_is_agp(p->rdev->ddev) ||
126 p->rdev->family == CHIP_RS780 ||
127 p->rdev->family == CHIP_RS880)) {
129 /* TODO: is this still needed for NI+ ? */
130 p->relocs[i].prefered_domains =
131 RADEON_GEM_DOMAIN_VRAM;
133 p->relocs[i].allowed_domains =
134 RADEON_GEM_DOMAIN_VRAM;
136 /* prioritize this over any other relocation */
137 priority = RADEON_CS_MAX_PRIORITY;
139 uint32_t domain = r->write_domain ?
140 r->write_domain : r->read_domains;
142 if (domain & RADEON_GEM_DOMAIN_CPU) {
143 DRM_ERROR("RADEON_GEM_DOMAIN_CPU is not valid "
144 "for command submission\n");
148 p->relocs[i].prefered_domains = domain;
149 if (domain == RADEON_GEM_DOMAIN_VRAM)
150 domain |= RADEON_GEM_DOMAIN_GTT;
151 p->relocs[i].allowed_domains = domain;
154 if (radeon_ttm_tt_has_userptr(p->relocs[i].robj->tbo.ttm)) {
155 uint32_t domain = p->relocs[i].prefered_domains;
156 if (!(domain & RADEON_GEM_DOMAIN_GTT)) {
157 DRM_ERROR("Only RADEON_GEM_DOMAIN_GTT is "
158 "allowed for userptr BOs\n");
161 need_mmap_lock = true;
162 domain = RADEON_GEM_DOMAIN_GTT;
163 p->relocs[i].prefered_domains = domain;
164 p->relocs[i].allowed_domains = domain;
167 p->relocs[i].tv.bo = &p->relocs[i].robj->tbo;
168 p->relocs[i].tv.shared = !r->write_domain;
170 radeon_cs_buckets_add(&buckets, &p->relocs[i].tv.head,
174 radeon_cs_buckets_get_list(&buckets, &p->validated);
176 if (p->cs_flags & RADEON_CS_USE_VM)
177 p->vm_bos = radeon_vm_get_bos(p->rdev, p->ib.vm,
180 down_read(¤t->mm->mmap_sem);
182 r = radeon_bo_list_validate(p->rdev, &p->ticket, &p->validated, p->ring);
185 up_read(¤t->mm->mmap_sem);
190 static int radeon_cs_get_ring(struct radeon_cs_parser *p, u32 ring, s32 priority)
192 p->priority = priority;
196 DRM_ERROR("unknown ring id: %d\n", ring);
198 case RADEON_CS_RING_GFX:
199 p->ring = RADEON_RING_TYPE_GFX_INDEX;
201 case RADEON_CS_RING_COMPUTE:
202 if (p->rdev->family >= CHIP_TAHITI) {
204 p->ring = CAYMAN_RING_TYPE_CP1_INDEX;
206 p->ring = CAYMAN_RING_TYPE_CP2_INDEX;
208 p->ring = RADEON_RING_TYPE_GFX_INDEX;
210 case RADEON_CS_RING_DMA:
211 if (p->rdev->family >= CHIP_CAYMAN) {
213 p->ring = R600_RING_TYPE_DMA_INDEX;
215 p->ring = CAYMAN_RING_TYPE_DMA1_INDEX;
216 } else if (p->rdev->family >= CHIP_RV770) {
217 p->ring = R600_RING_TYPE_DMA_INDEX;
222 case RADEON_CS_RING_UVD:
223 p->ring = R600_RING_TYPE_UVD_INDEX;
225 case RADEON_CS_RING_VCE:
226 /* TODO: only use the low priority ring for now */
227 p->ring = TN_RING_TYPE_VCE1_INDEX;
233 static int radeon_cs_sync_rings(struct radeon_cs_parser *p)
235 struct radeon_bo_list *reloc;
238 list_for_each_entry(reloc, &p->validated, tv.head) {
239 struct reservation_object *resv;
241 resv = reloc->robj->tbo.resv;
242 r = radeon_sync_resv(p->rdev, &p->ib.sync, resv,
250 /* XXX: note that this is called from the legacy UMS CS ioctl as well */
251 int radeon_cs_parser_init(struct radeon_cs_parser *p, void *data)
253 struct drm_radeon_cs *cs = data;
254 uint64_t *chunk_array_ptr;
256 u32 ring = RADEON_CS_RING_GFX;
259 if (!cs->num_chunks) {
263 INIT_LIST_HEAD(&p->validated);
266 p->const_ib.sa_bo = NULL;
268 p->chunk_relocs = NULL;
269 p->chunk_flags = NULL;
270 p->chunk_const_ib = NULL;
271 p->chunks_array = kcalloc(cs->num_chunks, sizeof(uint64_t), GFP_KERNEL);
272 if (p->chunks_array == NULL) {
275 chunk_array_ptr = (uint64_t *)(unsigned long)(cs->chunks);
276 if (copy_from_user(p->chunks_array, chunk_array_ptr,
277 sizeof(uint64_t)*cs->num_chunks)) {
281 p->nchunks = cs->num_chunks;
282 p->chunks = kcalloc(p->nchunks, sizeof(struct radeon_cs_chunk), GFP_KERNEL);
283 if (p->chunks == NULL) {
286 for (i = 0; i < p->nchunks; i++) {
287 struct drm_radeon_cs_chunk __user **chunk_ptr = NULL;
288 struct drm_radeon_cs_chunk user_chunk;
289 uint32_t __user *cdata;
291 chunk_ptr = (void __user*)(unsigned long)p->chunks_array[i];
292 if (copy_from_user(&user_chunk, chunk_ptr,
293 sizeof(struct drm_radeon_cs_chunk))) {
296 p->chunks[i].length_dw = user_chunk.length_dw;
297 if (user_chunk.chunk_id == RADEON_CHUNK_ID_RELOCS) {
298 p->chunk_relocs = &p->chunks[i];
300 if (user_chunk.chunk_id == RADEON_CHUNK_ID_IB) {
301 p->chunk_ib = &p->chunks[i];
302 /* zero length IB isn't useful */
303 if (p->chunks[i].length_dw == 0)
306 if (user_chunk.chunk_id == RADEON_CHUNK_ID_CONST_IB) {
307 p->chunk_const_ib = &p->chunks[i];
308 /* zero length CONST IB isn't useful */
309 if (p->chunks[i].length_dw == 0)
312 if (user_chunk.chunk_id == RADEON_CHUNK_ID_FLAGS) {
313 p->chunk_flags = &p->chunks[i];
314 /* zero length flags aren't useful */
315 if (p->chunks[i].length_dw == 0)
319 size = p->chunks[i].length_dw;
320 cdata = (void __user *)(unsigned long)user_chunk.chunk_data;
321 p->chunks[i].user_ptr = cdata;
322 if (user_chunk.chunk_id == RADEON_CHUNK_ID_CONST_IB)
325 if (user_chunk.chunk_id == RADEON_CHUNK_ID_IB) {
326 if (!p->rdev || !(p->rdev->flags & RADEON_IS_AGP))
330 p->chunks[i].kdata = drm_malloc_ab(size, sizeof(uint32_t));
331 size *= sizeof(uint32_t);
332 if (p->chunks[i].kdata == NULL) {
335 if (copy_from_user(p->chunks[i].kdata, cdata, size)) {
338 if (user_chunk.chunk_id == RADEON_CHUNK_ID_FLAGS) {
339 p->cs_flags = p->chunks[i].kdata[0];
340 if (p->chunks[i].length_dw > 1)
341 ring = p->chunks[i].kdata[1];
342 if (p->chunks[i].length_dw > 2)
343 priority = (s32)p->chunks[i].kdata[2];
347 /* these are KMS only */
349 if ((p->cs_flags & RADEON_CS_USE_VM) &&
350 !p->rdev->vm_manager.enabled) {
351 DRM_ERROR("VM not active on asic!\n");
355 if (radeon_cs_get_ring(p, ring, priority))
358 /* we only support VM on some SI+ rings */
359 if ((p->cs_flags & RADEON_CS_USE_VM) == 0) {
360 if (p->rdev->asic->ring[p->ring]->cs_parse == NULL) {
361 DRM_ERROR("Ring %d requires VM!\n", p->ring);
365 if (p->rdev->asic->ring[p->ring]->ib_parse == NULL) {
366 DRM_ERROR("VM not supported on ring %d!\n",
376 static int cmp_size_smaller_first(void *priv, struct list_head *a,
379 struct radeon_bo_list *la = list_entry(a, struct radeon_bo_list, tv.head);
380 struct radeon_bo_list *lb = list_entry(b, struct radeon_bo_list, tv.head);
382 /* Sort A before B if A is smaller. */
383 return (int)la->robj->tbo.num_pages - (int)lb->robj->tbo.num_pages;
387 * cs_parser_fini() - clean parser states
388 * @parser: parser structure holding parsing context.
389 * @error: error number
391 * If error is set than unvalidate buffer, otherwise just free memory
392 * used by parsing context.
394 static void radeon_cs_parser_fini(struct radeon_cs_parser *parser, int error, bool backoff)
399 /* Sort the buffer list from the smallest to largest buffer,
400 * which affects the order of buffers in the LRU list.
401 * This assures that the smallest buffers are added first
402 * to the LRU list, so they are likely to be later evicted
403 * first, instead of large buffers whose eviction is more
406 * This slightly lowers the number of bytes moved by TTM
407 * per frame under memory pressure.
409 list_sort(NULL, &parser->validated, cmp_size_smaller_first);
411 ttm_eu_fence_buffer_objects(&parser->ticket,
413 &parser->ib.fence->base);
414 } else if (backoff) {
415 ttm_eu_backoff_reservation(&parser->ticket,
419 if (parser->relocs != NULL) {
420 for (i = 0; i < parser->nrelocs; i++) {
421 struct radeon_bo *bo = parser->relocs[i].robj;
425 drm_gem_object_unreference_unlocked(&bo->gem_base);
428 kfree(parser->track);
429 kfree(parser->relocs);
430 drm_free_large(parser->vm_bos);
431 for (i = 0; i < parser->nchunks; i++)
432 drm_free_large(parser->chunks[i].kdata);
433 kfree(parser->chunks);
434 kfree(parser->chunks_array);
435 radeon_ib_free(parser->rdev, &parser->ib);
436 radeon_ib_free(parser->rdev, &parser->const_ib);
439 static int radeon_cs_ib_chunk(struct radeon_device *rdev,
440 struct radeon_cs_parser *parser)
444 if (parser->chunk_ib == NULL)
447 if (parser->cs_flags & RADEON_CS_USE_VM)
450 r = radeon_cs_parse(rdev, parser->ring, parser);
451 if (r || parser->parser_error) {
452 DRM_ERROR("Invalid command stream !\n");
456 r = radeon_cs_sync_rings(parser);
458 if (r != -ERESTARTSYS)
459 DRM_ERROR("Failed to sync rings: %i\n", r);
463 if (parser->ring == R600_RING_TYPE_UVD_INDEX)
464 radeon_uvd_note_usage(rdev);
465 else if ((parser->ring == TN_RING_TYPE_VCE1_INDEX) ||
466 (parser->ring == TN_RING_TYPE_VCE2_INDEX))
467 radeon_vce_note_usage(rdev);
469 r = radeon_ib_schedule(rdev, &parser->ib, NULL, true);
471 DRM_ERROR("Failed to schedule IB !\n");
476 static int radeon_bo_vm_update_pte(struct radeon_cs_parser *p,
477 struct radeon_vm *vm)
479 struct radeon_device *rdev = p->rdev;
480 struct radeon_bo_va *bo_va;
483 r = radeon_vm_update_page_directory(rdev, vm);
487 r = radeon_vm_clear_freed(rdev, vm);
491 if (vm->ib_bo_va == NULL) {
492 DRM_ERROR("Tmp BO not in VM!\n");
496 r = radeon_vm_bo_update(rdev, vm->ib_bo_va,
497 &rdev->ring_tmp_bo.bo->tbo.mem);
501 for (i = 0; i < p->nrelocs; i++) {
502 struct radeon_bo *bo;
504 bo = p->relocs[i].robj;
505 bo_va = radeon_vm_bo_find(vm, bo);
507 dev_err(rdev->dev, "bo %p not in vm %p\n", bo, vm);
511 r = radeon_vm_bo_update(rdev, bo_va, &bo->tbo.mem);
515 radeon_sync_fence(&p->ib.sync, bo_va->last_pt_update);
518 return radeon_vm_clear_invalids(rdev, vm);
521 static int radeon_cs_ib_vm_chunk(struct radeon_device *rdev,
522 struct radeon_cs_parser *parser)
524 struct radeon_fpriv *fpriv = parser->filp->driver_priv;
525 struct radeon_vm *vm = &fpriv->vm;
528 if (parser->chunk_ib == NULL)
530 if ((parser->cs_flags & RADEON_CS_USE_VM) == 0)
533 if (parser->const_ib.length_dw) {
534 r = radeon_ring_ib_parse(rdev, parser->ring, &parser->const_ib);
540 r = radeon_ring_ib_parse(rdev, parser->ring, &parser->ib);
545 if (parser->ring == R600_RING_TYPE_UVD_INDEX)
546 radeon_uvd_note_usage(rdev);
548 mutex_lock(&vm->mutex);
549 r = radeon_bo_vm_update_pte(parser, vm);
554 r = radeon_cs_sync_rings(parser);
556 if (r != -ERESTARTSYS)
557 DRM_ERROR("Failed to sync rings: %i\n", r);
561 if ((rdev->family >= CHIP_TAHITI) &&
562 (parser->chunk_const_ib != NULL)) {
563 r = radeon_ib_schedule(rdev, &parser->ib, &parser->const_ib, true);
565 r = radeon_ib_schedule(rdev, &parser->ib, NULL, true);
569 mutex_unlock(&vm->mutex);
573 static int radeon_cs_handle_lockup(struct radeon_device *rdev, int r)
576 r = radeon_gpu_reset(rdev);
583 static int radeon_cs_ib_fill(struct radeon_device *rdev, struct radeon_cs_parser *parser)
585 struct radeon_cs_chunk *ib_chunk;
586 struct radeon_vm *vm = NULL;
589 if (parser->chunk_ib == NULL)
592 if (parser->cs_flags & RADEON_CS_USE_VM) {
593 struct radeon_fpriv *fpriv = parser->filp->driver_priv;
596 if ((rdev->family >= CHIP_TAHITI) &&
597 (parser->chunk_const_ib != NULL)) {
598 ib_chunk = parser->chunk_const_ib;
599 if (ib_chunk->length_dw > RADEON_IB_VM_MAX_SIZE) {
600 DRM_ERROR("cs IB CONST too big: %d\n", ib_chunk->length_dw);
603 r = radeon_ib_get(rdev, parser->ring, &parser->const_ib,
604 vm, ib_chunk->length_dw * 4);
606 DRM_ERROR("Failed to get const ib !\n");
609 parser->const_ib.is_const_ib = true;
610 parser->const_ib.length_dw = ib_chunk->length_dw;
611 if (copy_from_user(parser->const_ib.ptr,
613 ib_chunk->length_dw * 4))
617 ib_chunk = parser->chunk_ib;
618 if (ib_chunk->length_dw > RADEON_IB_VM_MAX_SIZE) {
619 DRM_ERROR("cs IB too big: %d\n", ib_chunk->length_dw);
623 ib_chunk = parser->chunk_ib;
625 r = radeon_ib_get(rdev, parser->ring, &parser->ib,
626 vm, ib_chunk->length_dw * 4);
628 DRM_ERROR("Failed to get ib !\n");
631 parser->ib.length_dw = ib_chunk->length_dw;
633 memcpy(parser->ib.ptr, ib_chunk->kdata, ib_chunk->length_dw * 4);
634 else if (copy_from_user(parser->ib.ptr, ib_chunk->user_ptr, ib_chunk->length_dw * 4))
639 int radeon_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
641 struct radeon_device *rdev = dev->dev_private;
642 struct radeon_cs_parser parser;
645 down_read(&rdev->exclusive_lock);
646 if (!rdev->accel_working) {
647 up_read(&rdev->exclusive_lock);
650 if (rdev->in_reset) {
651 up_read(&rdev->exclusive_lock);
652 r = radeon_gpu_reset(rdev);
657 /* initialize parser */
658 memset(&parser, 0, sizeof(struct radeon_cs_parser));
661 parser.dev = rdev->dev;
662 parser.family = rdev->family;
663 r = radeon_cs_parser_init(&parser, data);
665 DRM_ERROR("Failed to initialize parser !\n");
666 radeon_cs_parser_fini(&parser, r, false);
667 up_read(&rdev->exclusive_lock);
668 r = radeon_cs_handle_lockup(rdev, r);
672 r = radeon_cs_ib_fill(rdev, &parser);
674 r = radeon_cs_parser_relocs(&parser);
675 if (r && r != -ERESTARTSYS)
676 DRM_ERROR("Failed to parse relocation %d!\n", r);
680 radeon_cs_parser_fini(&parser, r, false);
681 up_read(&rdev->exclusive_lock);
682 r = radeon_cs_handle_lockup(rdev, r);
686 trace_radeon_cs(&parser);
688 r = radeon_cs_ib_chunk(rdev, &parser);
692 r = radeon_cs_ib_vm_chunk(rdev, &parser);
697 radeon_cs_parser_fini(&parser, r, true);
698 up_read(&rdev->exclusive_lock);
699 r = radeon_cs_handle_lockup(rdev, r);
704 * radeon_cs_packet_parse() - parse cp packet and point ib index to next packet
705 * @parser: parser structure holding parsing context.
706 * @pkt: where to store packet information
708 * Assume that chunk_ib_index is properly set. Will return -EINVAL
709 * if packet is bigger than remaining ib size. or if packets is unknown.
711 int radeon_cs_packet_parse(struct radeon_cs_parser *p,
712 struct radeon_cs_packet *pkt,
715 struct radeon_cs_chunk *ib_chunk = p->chunk_ib;
716 struct radeon_device *rdev = p->rdev;
720 if (idx >= ib_chunk->length_dw) {
721 DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
722 idx, ib_chunk->length_dw);
725 header = radeon_get_ib_value(p, idx);
727 pkt->type = RADEON_CP_PACKET_GET_TYPE(header);
728 pkt->count = RADEON_CP_PACKET_GET_COUNT(header);
731 case RADEON_PACKET_TYPE0:
732 if (rdev->family < CHIP_R600) {
733 pkt->reg = R100_CP_PACKET0_GET_REG(header);
735 RADEON_CP_PACKET0_GET_ONE_REG_WR(header);
737 pkt->reg = R600_CP_PACKET0_GET_REG(header);
739 case RADEON_PACKET_TYPE3:
740 pkt->opcode = RADEON_CP_PACKET3_GET_OPCODE(header);
742 case RADEON_PACKET_TYPE2:
746 DRM_ERROR("Unknown packet type %d at %d !\n", pkt->type, idx);
750 if ((pkt->count + 1 + pkt->idx) >= ib_chunk->length_dw) {
751 DRM_ERROR("Packet (%d:%d:%d) end after CS buffer (%d) !\n",
752 pkt->idx, pkt->type, pkt->count, ib_chunk->length_dw);
759 for (i = 0; i < ib_chunk->length_dw; i++) {
761 printk("\t0x%08x <---\n", radeon_get_ib_value(p, i));
763 printk("\t0x%08x\n", radeon_get_ib_value(p, i));
769 * radeon_cs_packet_next_is_pkt3_nop() - test if the next packet is P3 NOP
770 * @p: structure holding the parser context.
772 * Check if the next packet is NOP relocation packet3.
774 bool radeon_cs_packet_next_is_pkt3_nop(struct radeon_cs_parser *p)
776 struct radeon_cs_packet p3reloc;
779 r = radeon_cs_packet_parse(p, &p3reloc, p->idx);
782 if (p3reloc.type != RADEON_PACKET_TYPE3)
784 if (p3reloc.opcode != RADEON_PACKET3_NOP)
790 * radeon_cs_dump_packet() - dump raw packet context
791 * @p: structure holding the parser context.
792 * @pkt: structure holding the packet.
794 * Used mostly for debugging and error reporting.
796 void radeon_cs_dump_packet(struct radeon_cs_parser *p,
797 struct radeon_cs_packet *pkt)
799 volatile uint32_t *ib;
805 for (i = 0; i <= (pkt->count + 1); i++, idx++)
806 DRM_INFO("ib[%d]=0x%08X\n", idx, ib[idx]);
810 * radeon_cs_packet_next_reloc() - parse next (should be reloc) packet
811 * @parser: parser structure holding parsing context.
812 * @data: pointer to relocation data
813 * @offset_start: starting offset
814 * @offset_mask: offset mask (to align start offset on)
815 * @reloc: reloc informations
817 * Check if next packet is relocation packet3, do bo validation and compute
818 * GPU offset using the provided start.
820 int radeon_cs_packet_next_reloc(struct radeon_cs_parser *p,
821 struct radeon_bo_list **cs_reloc,
824 struct radeon_cs_chunk *relocs_chunk;
825 struct radeon_cs_packet p3reloc;
829 if (p->chunk_relocs == NULL) {
830 DRM_ERROR("No relocation chunk !\n");
834 relocs_chunk = p->chunk_relocs;
835 r = radeon_cs_packet_parse(p, &p3reloc, p->idx);
838 p->idx += p3reloc.count + 2;
839 if (p3reloc.type != RADEON_PACKET_TYPE3 ||
840 p3reloc.opcode != RADEON_PACKET3_NOP) {
841 DRM_ERROR("No packet3 for relocation for packet at %d.\n",
843 radeon_cs_dump_packet(p, &p3reloc);
846 idx = radeon_get_ib_value(p, p3reloc.idx + 1);
847 if (idx >= relocs_chunk->length_dw) {
848 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
849 idx, relocs_chunk->length_dw);
850 radeon_cs_dump_packet(p, &p3reloc);
853 /* FIXME: we assume reloc size is 4 dwords */
855 *cs_reloc = p->relocs;
856 (*cs_reloc)->gpu_offset =
857 (u64)relocs_chunk->kdata[idx + 3] << 32;
858 (*cs_reloc)->gpu_offset |= relocs_chunk->kdata[idx + 0];
860 *cs_reloc = &p->relocs[(idx / 4)];