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.
27 #include <linux/list_sort.h>
29 #include <drm/amdgpu_drm.h>
31 #include "amdgpu_trace.h"
33 int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type,
34 u32 ip_instance, u32 ring,
35 struct amdgpu_ring **out_ring)
37 /* Right now all IPs have only one instance - multiple rings. */
38 if (ip_instance != 0) {
39 DRM_ERROR("invalid ip instance: %d\n", ip_instance);
45 DRM_ERROR("unknown ip type: %d\n", ip_type);
47 case AMDGPU_HW_IP_GFX:
48 if (ring < adev->gfx.num_gfx_rings) {
49 *out_ring = &adev->gfx.gfx_ring[ring];
51 DRM_ERROR("only %d gfx rings are supported now\n",
52 adev->gfx.num_gfx_rings);
56 case AMDGPU_HW_IP_COMPUTE:
57 if (ring < adev->gfx.num_compute_rings) {
58 *out_ring = &adev->gfx.compute_ring[ring];
60 DRM_ERROR("only %d compute rings are supported now\n",
61 adev->gfx.num_compute_rings);
65 case AMDGPU_HW_IP_DMA:
66 if (ring < adev->sdma.num_instances) {
67 *out_ring = &adev->sdma.instance[ring].ring;
69 DRM_ERROR("only %d SDMA rings are supported\n",
70 adev->sdma.num_instances);
74 case AMDGPU_HW_IP_UVD:
75 *out_ring = &adev->uvd.ring;
77 case AMDGPU_HW_IP_VCE:
79 *out_ring = &adev->vce.ring[ring];
81 DRM_ERROR("only two VCE rings are supported\n");
89 static int amdgpu_cs_user_fence_chunk(struct amdgpu_cs_parser *p,
90 struct drm_amdgpu_cs_chunk_fence *fence_data)
92 struct drm_gem_object *gobj;
95 handle = fence_data->handle;
96 gobj = drm_gem_object_lookup(p->adev->ddev, p->filp,
101 p->uf.bo = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj));
102 p->uf.offset = fence_data->offset;
104 if (amdgpu_ttm_tt_get_usermm(p->uf.bo->tbo.ttm)) {
105 drm_gem_object_unreference_unlocked(gobj);
109 p->uf_entry.robj = amdgpu_bo_ref(p->uf.bo);
110 p->uf_entry.priority = 0;
111 p->uf_entry.tv.bo = &p->uf_entry.robj->tbo;
112 p->uf_entry.tv.shared = true;
114 drm_gem_object_unreference_unlocked(gobj);
118 int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
120 union drm_amdgpu_cs *cs = data;
121 uint64_t *chunk_array_user;
122 uint64_t *chunk_array;
123 struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
128 if (cs->in.num_chunks == 0)
131 chunk_array = kmalloc_array(cs->in.num_chunks, sizeof(uint64_t), GFP_KERNEL);
135 p->ctx = amdgpu_ctx_get(fpriv, cs->in.ctx_id);
142 chunk_array_user = (uint64_t __user *)(unsigned long)(cs->in.chunks);
143 if (copy_from_user(chunk_array, chunk_array_user,
144 sizeof(uint64_t)*cs->in.num_chunks)) {
149 p->nchunks = cs->in.num_chunks;
150 p->chunks = kmalloc_array(p->nchunks, sizeof(struct amdgpu_cs_chunk),
157 for (i = 0; i < p->nchunks; i++) {
158 struct drm_amdgpu_cs_chunk __user **chunk_ptr = NULL;
159 struct drm_amdgpu_cs_chunk user_chunk;
160 uint32_t __user *cdata;
162 chunk_ptr = (void __user *)(unsigned long)chunk_array[i];
163 if (copy_from_user(&user_chunk, chunk_ptr,
164 sizeof(struct drm_amdgpu_cs_chunk))) {
167 goto free_partial_kdata;
169 p->chunks[i].chunk_id = user_chunk.chunk_id;
170 p->chunks[i].length_dw = user_chunk.length_dw;
172 size = p->chunks[i].length_dw;
173 cdata = (void __user *)(unsigned long)user_chunk.chunk_data;
175 p->chunks[i].kdata = drm_malloc_ab(size, sizeof(uint32_t));
176 if (p->chunks[i].kdata == NULL) {
179 goto free_partial_kdata;
181 size *= sizeof(uint32_t);
182 if (copy_from_user(p->chunks[i].kdata, cdata, size)) {
184 goto free_partial_kdata;
187 switch (p->chunks[i].chunk_id) {
188 case AMDGPU_CHUNK_ID_IB:
192 case AMDGPU_CHUNK_ID_FENCE:
193 size = sizeof(struct drm_amdgpu_cs_chunk_fence);
194 if (p->chunks[i].length_dw * sizeof(uint32_t) < size) {
196 goto free_partial_kdata;
199 ret = amdgpu_cs_user_fence_chunk(p, (void *)p->chunks[i].kdata);
201 goto free_partial_kdata;
205 case AMDGPU_CHUNK_ID_DEPENDENCIES:
210 goto free_partial_kdata;
215 p->ibs = kcalloc(p->num_ibs, sizeof(struct amdgpu_ib), GFP_KERNEL);
228 drm_free_large(p->chunks[i].kdata);
231 amdgpu_ctx_put(p->ctx);
238 /* Returns how many bytes TTM can move per IB.
240 static u64 amdgpu_cs_get_threshold_for_moves(struct amdgpu_device *adev)
242 u64 real_vram_size = adev->mc.real_vram_size;
243 u64 vram_usage = atomic64_read(&adev->vram_usage);
245 /* This function is based on the current VRAM usage.
247 * - If all of VRAM is free, allow relocating the number of bytes that
248 * is equal to 1/4 of the size of VRAM for this IB.
250 * - If more than one half of VRAM is occupied, only allow relocating
251 * 1 MB of data for this IB.
253 * - From 0 to one half of used VRAM, the threshold decreases
268 * Note: It's a threshold, not a limit. The threshold must be crossed
269 * for buffer relocations to stop, so any buffer of an arbitrary size
270 * can be moved as long as the threshold isn't crossed before
271 * the relocation takes place. We don't want to disable buffer
272 * relocations completely.
274 * The idea is that buffers should be placed in VRAM at creation time
275 * and TTM should only do a minimum number of relocations during
276 * command submission. In practice, you need to submit at least
277 * a dozen IBs to move all buffers to VRAM if they are in GTT.
279 * Also, things can get pretty crazy under memory pressure and actual
280 * VRAM usage can change a lot, so playing safe even at 50% does
281 * consistently increase performance.
284 u64 half_vram = real_vram_size >> 1;
285 u64 half_free_vram = vram_usage >= half_vram ? 0 : half_vram - vram_usage;
286 u64 bytes_moved_threshold = half_free_vram >> 1;
287 return max(bytes_moved_threshold, 1024*1024ull);
290 int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p,
291 struct list_head *validated)
293 struct amdgpu_bo_list_entry *lobj;
294 u64 initial_bytes_moved;
297 list_for_each_entry(lobj, validated, tv.head) {
298 struct amdgpu_bo *bo = lobj->robj;
299 struct mm_struct *usermm;
302 usermm = amdgpu_ttm_tt_get_usermm(bo->tbo.ttm);
303 if (usermm && usermm != current->mm)
309 /* Avoid moving this one if we have moved too many buffers
310 * for this IB already.
312 * Note that this allows moving at least one buffer of
313 * any size, because it doesn't take the current "bo"
314 * into account. We don't want to disallow buffer moves
317 if (p->bytes_moved <= p->bytes_moved_threshold)
318 domain = bo->prefered_domains;
320 domain = bo->allowed_domains;
323 amdgpu_ttm_placement_from_domain(bo, domain);
324 initial_bytes_moved = atomic64_read(&bo->adev->num_bytes_moved);
325 r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false);
326 p->bytes_moved += atomic64_read(&bo->adev->num_bytes_moved) -
330 if (r != -ERESTARTSYS && domain != bo->allowed_domains) {
331 domain = bo->allowed_domains;
340 static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
341 union drm_amdgpu_cs *cs)
343 struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
344 struct list_head duplicates;
345 bool need_mmap_lock = false;
348 INIT_LIST_HEAD(&p->validated);
350 p->bo_list = amdgpu_bo_list_get(fpriv, cs->in.bo_list_handle);
352 need_mmap_lock = p->bo_list->has_userptr;
353 amdgpu_bo_list_get_list(p->bo_list, &p->validated);
356 INIT_LIST_HEAD(&duplicates);
357 amdgpu_vm_get_pd_bo(&fpriv->vm, &p->validated, &p->vm_pd);
360 list_add(&p->uf_entry.tv.head, &p->validated);
363 down_read(¤t->mm->mmap_sem);
365 r = ttm_eu_reserve_buffers(&p->ticket, &p->validated, true, &duplicates);
366 if (unlikely(r != 0))
369 amdgpu_vm_get_pt_bos(&fpriv->vm, &duplicates);
371 p->bytes_moved_threshold = amdgpu_cs_get_threshold_for_moves(p->adev);
374 r = amdgpu_cs_list_validate(p, &duplicates);
378 r = amdgpu_cs_list_validate(p, &p->validated);
383 struct amdgpu_vm *vm = &fpriv->vm;
386 for (i = 0; i < p->bo_list->num_entries; i++) {
387 struct amdgpu_bo *bo = p->bo_list->array[i].robj;
389 p->bo_list->array[i].bo_va = amdgpu_vm_bo_find(vm, bo);
395 amdgpu_vm_move_pt_bos_in_lru(p->adev, &fpriv->vm);
396 ttm_eu_backoff_reservation(&p->ticket, &p->validated);
401 up_read(¤t->mm->mmap_sem);
406 static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p)
408 struct amdgpu_bo_list_entry *e;
411 list_for_each_entry(e, &p->validated, tv.head) {
412 struct reservation_object *resv = e->robj->tbo.resv;
413 r = amdgpu_sync_resv(p->adev, &p->ibs[0].sync, resv, p->filp);
421 static int cmp_size_smaller_first(void *priv, struct list_head *a,
424 struct amdgpu_bo_list_entry *la = list_entry(a, struct amdgpu_bo_list_entry, tv.head);
425 struct amdgpu_bo_list_entry *lb = list_entry(b, struct amdgpu_bo_list_entry, tv.head);
427 /* Sort A before B if A is smaller. */
428 return (int)la->robj->tbo.num_pages - (int)lb->robj->tbo.num_pages;
432 * cs_parser_fini() - clean parser states
433 * @parser: parser structure holding parsing context.
434 * @error: error number
436 * If error is set than unvalidate buffer, otherwise just free memory
437 * used by parsing context.
439 static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bool backoff)
441 struct amdgpu_fpriv *fpriv = parser->filp->driver_priv;
445 amdgpu_vm_move_pt_bos_in_lru(parser->adev, &fpriv->vm);
447 /* Sort the buffer list from the smallest to largest buffer,
448 * which affects the order of buffers in the LRU list.
449 * This assures that the smallest buffers are added first
450 * to the LRU list, so they are likely to be later evicted
451 * first, instead of large buffers whose eviction is more
454 * This slightly lowers the number of bytes moved by TTM
455 * per frame under memory pressure.
457 list_sort(NULL, &parser->validated, cmp_size_smaller_first);
459 ttm_eu_fence_buffer_objects(&parser->ticket,
462 } else if (backoff) {
463 ttm_eu_backoff_reservation(&parser->ticket,
466 fence_put(parser->fence);
469 amdgpu_ctx_put(parser->ctx);
471 amdgpu_bo_list_put(parser->bo_list);
473 for (i = 0; i < parser->nchunks; i++)
474 drm_free_large(parser->chunks[i].kdata);
475 kfree(parser->chunks);
477 for (i = 0; i < parser->num_ibs; i++)
478 amdgpu_ib_free(parser->adev, &parser->ibs[i]);
480 amdgpu_bo_unref(&parser->uf.bo);
481 amdgpu_bo_unref(&parser->uf_entry.robj);
484 static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
485 struct amdgpu_vm *vm)
487 struct amdgpu_device *adev = p->adev;
488 struct amdgpu_bo_va *bo_va;
489 struct amdgpu_bo *bo;
492 r = amdgpu_vm_update_page_directory(adev, vm);
496 r = amdgpu_sync_fence(adev, &p->ibs[0].sync, vm->page_directory_fence);
500 r = amdgpu_vm_clear_freed(adev, vm);
505 for (i = 0; i < p->bo_list->num_entries; i++) {
508 /* ignore duplicates */
509 bo = p->bo_list->array[i].robj;
513 bo_va = p->bo_list->array[i].bo_va;
517 r = amdgpu_vm_bo_update(adev, bo_va, &bo->tbo.mem);
521 f = bo_va->last_pt_update;
522 r = amdgpu_sync_fence(adev, &p->ibs[0].sync, f);
529 r = amdgpu_vm_clear_invalids(adev, vm, &p->ibs[0].sync);
531 if (amdgpu_vm_debug && p->bo_list) {
532 /* Invalidate all BOs to test for userspace bugs */
533 for (i = 0; i < p->bo_list->num_entries; i++) {
534 /* ignore duplicates */
535 bo = p->bo_list->array[i].robj;
539 amdgpu_vm_bo_invalidate(adev, bo);
546 static int amdgpu_cs_ib_vm_chunk(struct amdgpu_device *adev,
547 struct amdgpu_cs_parser *parser)
549 struct amdgpu_fpriv *fpriv = parser->filp->driver_priv;
550 struct amdgpu_vm *vm = &fpriv->vm;
551 struct amdgpu_ring *ring;
554 if (parser->num_ibs == 0)
557 /* Only for UVD/VCE VM emulation */
558 for (i = 0; i < parser->num_ibs; i++) {
559 ring = parser->ibs[i].ring;
560 if (ring->funcs->parse_cs) {
561 r = amdgpu_ring_parse_cs(ring, parser, i);
567 r = amdgpu_bo_vm_update_pte(parser, vm);
569 amdgpu_cs_sync_rings(parser);
574 static int amdgpu_cs_handle_lockup(struct amdgpu_device *adev, int r)
577 r = amdgpu_gpu_reset(adev);
584 static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
585 struct amdgpu_cs_parser *parser)
587 struct amdgpu_fpriv *fpriv = parser->filp->driver_priv;
588 struct amdgpu_vm *vm = &fpriv->vm;
592 for (i = 0, j = 0; i < parser->nchunks && j < parser->num_ibs; i++) {
593 struct amdgpu_cs_chunk *chunk;
594 struct amdgpu_ib *ib;
595 struct drm_amdgpu_cs_chunk_ib *chunk_ib;
596 struct amdgpu_ring *ring;
598 chunk = &parser->chunks[i];
599 ib = &parser->ibs[j];
600 chunk_ib = (struct drm_amdgpu_cs_chunk_ib *)chunk->kdata;
602 if (chunk->chunk_id != AMDGPU_CHUNK_ID_IB)
605 r = amdgpu_cs_get_ring(adev, chunk_ib->ip_type,
606 chunk_ib->ip_instance, chunk_ib->ring,
611 if (ring->funcs->parse_cs) {
612 struct amdgpu_bo_va_mapping *m;
613 struct amdgpu_bo *aobj = NULL;
617 m = amdgpu_cs_find_mapping(parser, chunk_ib->va_start,
620 DRM_ERROR("IB va_start is invalid\n");
624 if ((chunk_ib->va_start + chunk_ib->ib_bytes) >
625 (m->it.last + 1) * AMDGPU_GPU_PAGE_SIZE) {
626 DRM_ERROR("IB va_start+ib_bytes is invalid\n");
630 /* the IB should be reserved at this point */
631 r = amdgpu_bo_kmap(aobj, (void **)&kptr);
636 offset = ((uint64_t)m->it.start) * AMDGPU_GPU_PAGE_SIZE;
637 kptr += chunk_ib->va_start - offset;
639 r = amdgpu_ib_get(ring, NULL, chunk_ib->ib_bytes, ib);
641 DRM_ERROR("Failed to get ib !\n");
645 memcpy(ib->ptr, kptr, chunk_ib->ib_bytes);
646 amdgpu_bo_kunmap(aobj);
648 r = amdgpu_ib_get(ring, vm, 0, ib);
650 DRM_ERROR("Failed to get ib !\n");
654 ib->gpu_addr = chunk_ib->va_start;
657 ib->length_dw = chunk_ib->ib_bytes / 4;
658 ib->flags = chunk_ib->flags;
659 ib->ctx = parser->ctx;
663 if (!parser->num_ibs)
666 /* add GDS resources to first IB */
667 if (parser->bo_list) {
668 struct amdgpu_bo *gds = parser->bo_list->gds_obj;
669 struct amdgpu_bo *gws = parser->bo_list->gws_obj;
670 struct amdgpu_bo *oa = parser->bo_list->oa_obj;
671 struct amdgpu_ib *ib = &parser->ibs[0];
674 ib->gds_base = amdgpu_bo_gpu_offset(gds);
675 ib->gds_size = amdgpu_bo_size(gds);
678 ib->gws_base = amdgpu_bo_gpu_offset(gws);
679 ib->gws_size = amdgpu_bo_size(gws);
682 ib->oa_base = amdgpu_bo_gpu_offset(oa);
683 ib->oa_size = amdgpu_bo_size(oa);
686 /* wrap the last IB with user fence */
688 struct amdgpu_ib *ib = &parser->ibs[parser->num_ibs - 1];
690 /* UVD & VCE fw doesn't support user fences */
691 if (ib->ring->type == AMDGPU_RING_TYPE_UVD ||
692 ib->ring->type == AMDGPU_RING_TYPE_VCE)
695 ib->user = &parser->uf;
701 static int amdgpu_cs_dependencies(struct amdgpu_device *adev,
702 struct amdgpu_cs_parser *p)
704 struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
705 struct amdgpu_ib *ib;
711 /* Add dependencies to first IB */
713 for (i = 0; i < p->nchunks; ++i) {
714 struct drm_amdgpu_cs_chunk_dep *deps;
715 struct amdgpu_cs_chunk *chunk;
718 chunk = &p->chunks[i];
720 if (chunk->chunk_id != AMDGPU_CHUNK_ID_DEPENDENCIES)
723 deps = (struct drm_amdgpu_cs_chunk_dep *)chunk->kdata;
724 num_deps = chunk->length_dw * 4 /
725 sizeof(struct drm_amdgpu_cs_chunk_dep);
727 for (j = 0; j < num_deps; ++j) {
728 struct amdgpu_ring *ring;
729 struct amdgpu_ctx *ctx;
732 r = amdgpu_cs_get_ring(adev, deps[j].ip_type,
734 deps[j].ring, &ring);
738 ctx = amdgpu_ctx_get(fpriv, deps[j].ctx_id);
742 fence = amdgpu_ctx_get_fence(ctx, ring,
750 r = amdgpu_sync_fence(adev, &ib->sync, fence);
762 static int amdgpu_cs_free_job(struct amdgpu_job *job)
766 for (i = 0; i < job->num_ibs; i++)
767 amdgpu_ib_free(job->adev, &job->ibs[i]);
770 amdgpu_bo_unref(&job->uf.bo);
774 static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
775 union drm_amdgpu_cs *cs)
777 struct amdgpu_ring * ring = p->ibs->ring;
778 struct amd_sched_fence *fence;
779 struct amdgpu_job *job;
781 job = kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL);
785 job->base.sched = &ring->sched;
786 job->base.s_entity = &p->ctx->rings[ring->idx].entity;
788 job->owner = p->filp;
789 job->free_job = amdgpu_cs_free_job;
792 job->num_ibs = p->num_ibs;
796 if (job->ibs[job->num_ibs - 1].user) {
798 job->ibs[job->num_ibs - 1].user = &job->uf;
802 fence = amd_sched_fence_create(job->base.s_entity, p->filp);
804 amdgpu_cs_free_job(job);
809 job->base.s_fence = fence;
810 p->fence = fence_get(&fence->base);
812 cs->out.handle = amdgpu_ctx_add_fence(p->ctx, ring,
814 job->ibs[job->num_ibs - 1].sequence = cs->out.handle;
816 trace_amdgpu_cs_ioctl(job);
817 amd_sched_entity_push_job(&job->base);
822 int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
824 struct amdgpu_device *adev = dev->dev_private;
825 union drm_amdgpu_cs *cs = data;
826 struct amdgpu_cs_parser parser = {};
827 bool reserved_buffers = false;
830 if (!adev->accel_working)
836 r = amdgpu_cs_parser_init(&parser, data);
838 DRM_ERROR("Failed to initialize parser !\n");
839 amdgpu_cs_parser_fini(&parser, r, false);
840 r = amdgpu_cs_handle_lockup(adev, r);
843 r = amdgpu_cs_parser_bos(&parser, data);
845 DRM_ERROR("Not enough memory for command submission!\n");
846 else if (r && r != -ERESTARTSYS)
847 DRM_ERROR("Failed to process the buffer list %d!\n", r);
849 reserved_buffers = true;
850 r = amdgpu_cs_ib_fill(adev, &parser);
854 r = amdgpu_cs_dependencies(adev, &parser);
856 DRM_ERROR("Failed in the dependencies handling %d!\n", r);
862 for (i = 0; i < parser.num_ibs; i++)
863 trace_amdgpu_cs(&parser, i);
865 r = amdgpu_cs_ib_vm_chunk(adev, &parser);
870 r = amdgpu_cs_submit(&parser, cs);
873 amdgpu_cs_parser_fini(&parser, r, reserved_buffers);
874 r = amdgpu_cs_handle_lockup(adev, r);
879 * amdgpu_cs_wait_ioctl - wait for a command submission to finish
882 * @data: data from userspace
883 * @filp: file private
885 * Wait for the command submission identified by handle to finish.
887 int amdgpu_cs_wait_ioctl(struct drm_device *dev, void *data,
888 struct drm_file *filp)
890 union drm_amdgpu_wait_cs *wait = data;
891 struct amdgpu_device *adev = dev->dev_private;
892 unsigned long timeout = amdgpu_gem_timeout(wait->in.timeout);
893 struct amdgpu_ring *ring = NULL;
894 struct amdgpu_ctx *ctx;
898 r = amdgpu_cs_get_ring(adev, wait->in.ip_type, wait->in.ip_instance,
899 wait->in.ring, &ring);
903 ctx = amdgpu_ctx_get(filp->driver_priv, wait->in.ctx_id);
907 fence = amdgpu_ctx_get_fence(ctx, ring, wait->in.handle);
911 r = fence_wait_timeout(fence, true, timeout);
920 memset(wait, 0, sizeof(*wait));
921 wait->out.status = (r == 0);
927 * amdgpu_cs_find_bo_va - find bo_va for VM address
929 * @parser: command submission parser context
931 * @bo: resulting BO of the mapping found
933 * Search the buffer objects in the command submission context for a certain
934 * virtual memory address. Returns allocation structure when found, NULL
937 struct amdgpu_bo_va_mapping *
938 amdgpu_cs_find_mapping(struct amdgpu_cs_parser *parser,
939 uint64_t addr, struct amdgpu_bo **bo)
941 struct amdgpu_bo_va_mapping *mapping;
944 if (!parser->bo_list)
947 addr /= AMDGPU_GPU_PAGE_SIZE;
949 for (i = 0; i < parser->bo_list->num_entries; i++) {
950 struct amdgpu_bo_list_entry *lobj;
952 lobj = &parser->bo_list->array[i];
956 list_for_each_entry(mapping, &lobj->bo_va->valids, list) {
957 if (mapping->it.start > addr ||
958 addr > mapping->it.last)
961 *bo = lobj->bo_va->bo;
965 list_for_each_entry(mapping, &lobj->bo_va->invalids, list) {
966 if (mapping->it.start > addr ||
967 addr > mapping->it.last)
970 *bo = lobj->bo_va->bo;