return 0;
}
+struct amdgpu_cs_parser *amdgpu_cs_parser_create(struct amdgpu_device *adev,
+ struct drm_file *filp,
+ struct amdgpu_ctx *ctx,
+ struct amdgpu_ib *ibs,
+ uint32_t num_ibs)
+{
+ struct amdgpu_cs_parser *parser;
+ int i;
+
+ parser = kzalloc(sizeof(struct amdgpu_cs_parser), GFP_KERNEL);
+ if (!parser)
+ return NULL;
+
+ parser->adev = adev;
+ parser->filp = filp;
+ parser->ctx = ctx;
+ parser->ibs = ibs;
+ parser->num_ibs = num_ibs;
+ for (i = 0; i < num_ibs; i++)
+ ibs[i].ctx = ctx;
+
+ return parser;
+}
+
int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
{
union drm_amdgpu_cs *cs = data;
/* get chunks */
INIT_LIST_HEAD(&p->validated);
- chunk_array = kcalloc(cs->in.num_chunks, sizeof(uint64_t), GFP_KERNEL);
+ chunk_array = kmalloc_array(cs->in.num_chunks, sizeof(uint64_t), GFP_KERNEL);
if (chunk_array == NULL) {
r = -ENOMEM;
goto out;
}
- chunk_array_user = (uint64_t *)(unsigned long)(cs->in.chunks);
+ chunk_array_user = (uint64_t __user *)(cs->in.chunks);
if (copy_from_user(chunk_array, chunk_array_user,
sizeof(uint64_t)*cs->in.num_chunks)) {
r = -EFAULT;
}
p->nchunks = cs->in.num_chunks;
- p->chunks = kcalloc(p->nchunks, sizeof(struct amdgpu_cs_chunk),
+ p->chunks = kmalloc_array(p->nchunks, sizeof(struct amdgpu_cs_chunk),
GFP_KERNEL);
if (p->chunks == NULL) {
r = -ENOMEM;
struct drm_amdgpu_cs_chunk user_chunk;
uint32_t __user *cdata;
- chunk_ptr = (void __user *)(unsigned long)chunk_array[i];
+ chunk_ptr = (void __user *)chunk_array[i];
if (copy_from_user(&user_chunk, chunk_ptr,
sizeof(struct drm_amdgpu_cs_chunk))) {
r = -EFAULT;
p->chunks[i].length_dw = user_chunk.length_dw;
size = p->chunks[i].length_dw;
- cdata = (void __user *)(unsigned long)user_chunk.chunk_data;
+ cdata = (void __user *)user_chunk.chunk_data;
p->chunks[i].user_ptr = cdata;
p->chunks[i].kdata = drm_malloc_ab(size, sizeof(uint32_t));
}
}
+
p->ibs = kcalloc(p->num_ibs, sizeof(struct amdgpu_ib), GFP_KERNEL);
- if (!p->ibs) {
+ if (!p->ibs)
r = -ENOMEM;
- goto out;
- }
out:
kfree(chunk_array);
* into account. We don't want to disallow buffer moves
* completely.
*/
- if (current_domain != AMDGPU_GEM_DOMAIN_CPU &&
+ if ((lobj->allowed_domains & current_domain) != 0 &&
(domain & current_domain) == 0 && /* will be moved */
bytes_moved > bytes_moved_threshold) {
/* don't move it */
return (int)la->robj->tbo.num_pages - (int)lb->robj->tbo.num_pages;
}
-/**
- * cs_parser_fini() - clean parser states
- * @parser: parser structure holding parsing context.
- * @error: error number
- *
- * If error is set than unvalidate buffer, otherwise just free memory
- * used by parsing context.
- **/
-static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bool backoff)
+static void amdgpu_cs_parser_fini_early(struct amdgpu_cs_parser *parser, int error, bool backoff)
{
- unsigned i;
-
if (!error) {
/* Sort the buffer list from the smallest to largest buffer,
* which affects the order of buffers in the LRU list.
ttm_eu_backoff_reservation(&parser->ticket,
&parser->validated);
}
+}
+static void amdgpu_cs_parser_fini_late(struct amdgpu_cs_parser *parser)
+{
+ unsigned i;
if (parser->ctx)
amdgpu_ctx_put(parser->ctx);
if (parser->bo_list)
amdgpu_bo_list_put(parser->bo_list);
+
drm_free_large(parser->vm_bos);
for (i = 0; i < parser->nchunks; i++)
drm_free_large(parser->chunks[i].kdata);
kfree(parser->chunks);
- if (parser->ibs)
- for (i = 0; i < parser->num_ibs; i++)
- amdgpu_ib_free(parser->adev, &parser->ibs[i]);
- kfree(parser->ibs);
- if (parser->uf.bo)
- drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base);
+ if (!amdgpu_enable_scheduler)
+ {
+ if (parser->ibs)
+ for (i = 0; i < parser->num_ibs; i++)
+ amdgpu_ib_free(parser->adev, &parser->ibs[i]);
+ kfree(parser->ibs);
+ if (parser->uf.bo)
+ drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base);
+ }
+
+ kfree(parser);
+}
+
+/**
+ * cs_parser_fini() - clean parser states
+ * @parser: parser structure holding parsing context.
+ * @error: error number
+ *
+ * If error is set than unvalidate buffer, otherwise just free memory
+ * used by parsing context.
+ **/
+static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bool backoff)
+{
+ amdgpu_cs_parser_fini_early(parser, error, backoff);
+ amdgpu_cs_parser_fini_late(parser);
}
static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
if (r)
return r;
+ r = amdgpu_sync_fence(adev, &p->ibs[0].sync, vm->page_directory_fence);
+ if (r)
+ return r;
+
r = amdgpu_vm_clear_freed(adev, vm);
if (r)
return r;
if (p->bo_list) {
for (i = 0; i < p->bo_list->num_entries; i++) {
+ struct fence *f;
+
/* ignore duplicates */
bo = p->bo_list->array[i].robj;
if (!bo)
if (r)
return r;
- amdgpu_sync_fence(&p->ibs[0].sync, bo_va->last_pt_update);
+ f = bo_va->last_pt_update;
+ r = amdgpu_sync_fence(adev, &p->ibs[0].sync, f);
+ if (r)
+ return r;
}
}
goto out;
}
amdgpu_cs_sync_rings(parser);
-
- r = amdgpu_ib_schedule(adev, parser->num_ibs, parser->ibs,
- parser->filp);
+ if (!amdgpu_enable_scheduler)
+ r = amdgpu_ib_schedule(adev, parser->num_ibs, parser->ibs,
+ parser->filp);
out:
mutex_unlock(&vm->mutex);
ib->oa_size = amdgpu_bo_size(oa);
}
}
-
/* wrap the last IB with user fence */
if (parser->uf.bo) {
struct amdgpu_ib *ib = &parser->ibs[parser->num_ibs - 1];
sizeof(struct drm_amdgpu_cs_chunk_dep);
for (j = 0; j < num_deps; ++j) {
- struct amdgpu_fence *fence;
struct amdgpu_ring *ring;
struct amdgpu_ctx *ctx;
+ struct fence *fence;
r = amdgpu_cs_get_ring(adev, deps[j].ip_type,
deps[j].ip_instance,
if (ctx == NULL)
return -EINVAL;
- r = amdgpu_fence_recreate(ring, p->filp,
- deps[j].handle,
- &fence);
- if (r) {
+ fence = amdgpu_ctx_get_fence(ctx, ring,
+ deps[j].handle);
+ if (IS_ERR(fence)) {
+ r = PTR_ERR(fence);
amdgpu_ctx_put(ctx);
return r;
- }
- amdgpu_sync_fence(&ib->sync, fence);
- amdgpu_fence_unref(&fence);
- amdgpu_ctx_put(ctx);
+ } else if (fence) {
+ r = amdgpu_sync_fence(adev, &ib->sync, fence);
+ fence_put(fence);
+ amdgpu_ctx_put(ctx);
+ if (r)
+ return r;
+ }
}
}
return 0;
}
+static int amdgpu_cs_free_job(struct amdgpu_job *sched_job)
+{
+ int i;
+ if (sched_job->ibs)
+ for (i = 0; i < sched_job->num_ibs; i++)
+ amdgpu_ib_free(sched_job->adev, &sched_job->ibs[i]);
+ kfree(sched_job->ibs);
+ if (sched_job->uf.bo)
+ drm_gem_object_unreference_unlocked(&sched_job->uf.bo->gem_base);
+ return 0;
+}
+
int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
{
struct amdgpu_device *adev = dev->dev_private;
union drm_amdgpu_cs *cs = data;
- struct amdgpu_cs_parser parser;
- int r, i;
+ struct amdgpu_cs_parser *parser;
bool reserved_buffers = false;
+ int i, r;
down_read(&adev->exclusive_lock);
if (!adev->accel_working) {
up_read(&adev->exclusive_lock);
return -EBUSY;
}
- /* initialize parser */
- memset(&parser, 0, sizeof(struct amdgpu_cs_parser));
- parser.filp = filp;
- parser.adev = adev;
- r = amdgpu_cs_parser_init(&parser, data);
+
+ parser = amdgpu_cs_parser_create(adev, filp, NULL, NULL, 0);
+ if (!parser)
+ return -ENOMEM;
+ r = amdgpu_cs_parser_init(parser, data);
if (r) {
DRM_ERROR("Failed to initialize parser !\n");
- amdgpu_cs_parser_fini(&parser, r, false);
+ amdgpu_cs_parser_fini(parser, r, false);
up_read(&adev->exclusive_lock);
r = amdgpu_cs_handle_lockup(adev, r);
return r;
}
- r = amdgpu_cs_parser_relocs(&parser);
- if (r) {
- if (r != -ERESTARTSYS) {
- if (r == -ENOMEM)
- DRM_ERROR("Not enough memory for command submission!\n");
- else
- DRM_ERROR("Failed to process the buffer list %d!\n", r);
- }
+ r = amdgpu_cs_parser_relocs(parser);
+ if (r == -ENOMEM)
+ DRM_ERROR("Not enough memory for command submission!\n");
+ else if (r && r != -ERESTARTSYS)
+ DRM_ERROR("Failed to process the buffer list %d!\n", r);
+ else if (!r) {
+ reserved_buffers = true;
+ r = amdgpu_cs_ib_fill(adev, parser);
}
if (!r) {
- reserved_buffers = true;
- r = amdgpu_cs_ib_fill(adev, &parser);
+ r = amdgpu_cs_dependencies(adev, parser);
+ if (r)
+ DRM_ERROR("Failed in the dependencies handling %d!\n", r);
}
- if (!r)
- r = amdgpu_cs_dependencies(adev, &parser);
-
- if (r) {
- amdgpu_cs_parser_fini(&parser, r, reserved_buffers);
- up_read(&adev->exclusive_lock);
- r = amdgpu_cs_handle_lockup(adev, r);
- return r;
- }
+ if (r)
+ goto out;
- for (i = 0; i < parser.num_ibs; i++)
- trace_amdgpu_cs(&parser, i);
+ for (i = 0; i < parser->num_ibs; i++)
+ trace_amdgpu_cs(parser, i);
- r = amdgpu_cs_ib_vm_chunk(adev, &parser);
- if (r) {
+ r = amdgpu_cs_ib_vm_chunk(adev, parser);
+ if (r)
goto out;
+
+ if (amdgpu_enable_scheduler && parser->num_ibs) {
+ struct amdgpu_job *job;
+ struct amdgpu_ring * ring = parser->ibs->ring;
+ job = kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL);
+ if (!job)
+ return -ENOMEM;
+ job->base.sched = ring->scheduler;
+ job->base.s_entity = &parser->ctx->rings[ring->idx].entity;
+ job->adev = parser->adev;
+ job->ibs = parser->ibs;
+ job->num_ibs = parser->num_ibs;
+ job->base.owner = parser->filp;
+ mutex_init(&job->job_lock);
+ if (job->ibs[job->num_ibs - 1].user) {
+ memcpy(&job->uf, &parser->uf,
+ sizeof(struct amdgpu_user_fence));
+ job->ibs[job->num_ibs - 1].user = &job->uf;
+ }
+
+ job->free_job = amdgpu_cs_free_job;
+ mutex_lock(&job->job_lock);
+ r = amd_sched_entity_push_job((struct amd_sched_job *)job);
+ if (r) {
+ mutex_unlock(&job->job_lock);
+ amdgpu_cs_free_job(job);
+ kfree(job);
+ goto out;
+ }
+ cs->out.handle =
+ amdgpu_ctx_add_fence(parser->ctx, ring,
+ &job->base.s_fence->base);
+ parser->ibs[parser->num_ibs - 1].sequence = cs->out.handle;
+
+ list_sort(NULL, &parser->validated, cmp_size_smaller_first);
+ ttm_eu_fence_buffer_objects(&parser->ticket,
+ &parser->validated,
+ &job->base.s_fence->base);
+
+ mutex_unlock(&job->job_lock);
+ amdgpu_cs_parser_fini_late(parser);
+ up_read(&adev->exclusive_lock);
+ return 0;
}
- cs->out.handle = parser.ibs[parser.num_ibs - 1].fence->seq;
+ cs->out.handle = parser->ibs[parser->num_ibs - 1].sequence;
out:
- amdgpu_cs_parser_fini(&parser, r, true);
+ amdgpu_cs_parser_fini(parser, r, reserved_buffers);
up_read(&adev->exclusive_lock);
r = amdgpu_cs_handle_lockup(adev, r);
return r;
union drm_amdgpu_wait_cs *wait = data;
struct amdgpu_device *adev = dev->dev_private;
unsigned long timeout = amdgpu_gem_timeout(wait->in.timeout);
- struct amdgpu_fence *fence = NULL;
struct amdgpu_ring *ring = NULL;
struct amdgpu_ctx *ctx;
+ struct fence *fence;
long r;
- ctx = amdgpu_ctx_get(filp->driver_priv, wait->in.ctx_id);
- if (ctx == NULL)
- return -EINVAL;
-
r = amdgpu_cs_get_ring(adev, wait->in.ip_type, wait->in.ip_instance,
wait->in.ring, &ring);
- if (r) {
- amdgpu_ctx_put(ctx);
+ if (r)
return r;
- }
- r = amdgpu_fence_recreate(ring, filp, wait->in.handle, &fence);
- if (r) {
- amdgpu_ctx_put(ctx);
- return r;
- }
+ ctx = amdgpu_ctx_get(filp->driver_priv, wait->in.ctx_id);
+ if (ctx == NULL)
+ return -EINVAL;
+
+ fence = amdgpu_ctx_get_fence(ctx, ring, wait->in.handle);
+ if (IS_ERR(fence))
+ r = PTR_ERR(fence);
+ else if (fence) {
+ r = fence_wait_timeout(fence, true, timeout);
+ fence_put(fence);
+ } else
+ r = 1;
- r = fence_wait_timeout(&fence->base, true, timeout);
- amdgpu_fence_unref(&fence);
amdgpu_ctx_put(ctx);
if (r < 0)
return r;
if (!reloc->bo_va)
continue;
- list_for_each_entry(mapping, &reloc->bo_va->mappings, list) {
+ list_for_each_entry(mapping, &reloc->bo_va->valids, list) {
+ if (mapping->it.start > addr ||
+ addr > mapping->it.last)
+ continue;
+
+ *bo = reloc->bo_va->bo;
+ return mapping;
+ }
+
+ list_for_each_entry(mapping, &reloc->bo_va->invalids, list) {
if (mapping->it.start > addr ||
addr > mapping->it.last)
continue;
projects / linux.git / blobdiff