static int amdgpu_amdkfd_remove_eviction_fence(struct amdgpu_bo *bo,
struct amdgpu_amdkfd_fence *ef)
{
- struct dma_resv *resv = bo->tbo.base.resv;
- struct dma_resv_list *old, *new;
- unsigned int i, j, k;
+ struct dma_fence *replacement;
if (!ef)
return -EINVAL;
- old = dma_resv_shared_list(resv);
- if (!old)
- return 0;
-
- new = kmalloc(struct_size(new, shared, old->shared_max), GFP_KERNEL);
- if (!new)
- return -ENOMEM;
-
- /* Go through all the shared fences in the resevation object and sort
- * the interesting ones to the end of the list.
+ /* TODO: Instead of block before we should use the fence of the page
+ * table update and TLB flush here directly.
*/
- for (i = 0, j = old->shared_count, k = 0; i < old->shared_count; ++i) {
- struct dma_fence *f;
-
- f = rcu_dereference_protected(old->shared[i],
- dma_resv_held(resv));
-
- if (f->context == ef->base.context)
- RCU_INIT_POINTER(new->shared[--j], f);
- else
- RCU_INIT_POINTER(new->shared[k++], f);
- }
- new->shared_max = old->shared_max;
- new->shared_count = k;
-
- /* Install the new fence list, seqcount provides the barriers */
- write_seqcount_begin(&resv->seq);
- RCU_INIT_POINTER(resv->fence, new);
- write_seqcount_end(&resv->seq);
-
- /* Drop the references to the removed fences or move them to ef_list */
- for (i = j; i < old->shared_count; ++i) {
- struct dma_fence *f;
-
- f = rcu_dereference_protected(new->shared[i],
- dma_resv_held(resv));
- dma_fence_put(f);
- }
- kfree_rcu(old, rcu);
-
+ replacement = dma_fence_get_stub();
+ dma_resv_replace_fences(bo->tbo.base.resv, ef->base.context,
+ replacement, DMA_RESV_USAGE_READ);
+ dma_fence_put(replacement);
return 0;
}
static int update_gpuvm_pte(struct kgd_mem *mem,
struct kfd_mem_attachment *entry,
- struct amdgpu_sync *sync,
- bool *table_freed)
+ struct amdgpu_sync *sync)
{
struct amdgpu_bo_va *bo_va = entry->bo_va;
struct amdgpu_device *adev = entry->adev;
return ret;
/* Update the page tables */
- ret = amdgpu_vm_bo_update(adev, bo_va, false, table_freed);
+ ret = amdgpu_vm_bo_update(adev, bo_va, false);
if (ret) {
pr_err("amdgpu_vm_bo_update failed\n");
return ret;
static int map_bo_to_gpuvm(struct kgd_mem *mem,
struct kfd_mem_attachment *entry,
struct amdgpu_sync *sync,
- bool no_update_pte,
- bool *table_freed)
+ bool no_update_pte)
{
int ret;
if (no_update_pte)
return 0;
- ret = update_gpuvm_pte(mem, entry, sync, table_freed);
+ ret = update_gpuvm_pte(mem, entry, sync);
if (ret) {
pr_err("update_gpuvm_pte() failed\n");
goto update_gpuvm_pte_failed;
AMDGPU_FENCE_OWNER_KFD, false);
if (ret)
goto wait_pd_fail;
- ret = dma_resv_reserve_shared(vm->root.bo->tbo.base.resv, 1);
+ ret = dma_resv_reserve_fences(vm->root.bo->tbo.base.resv, 1);
if (ret)
goto reserve_shared_fail;
amdgpu_bo_fence(vm->root.bo,
int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
struct amdgpu_device *adev, struct kgd_mem *mem,
- void *drm_priv, bool *table_freed)
+ void *drm_priv)
{
struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv);
int ret;
entry->va, entry->va + bo_size, entry);
ret = map_bo_to_gpuvm(mem, entry, ctx.sync,
- is_invalid_userptr, table_freed);
+ is_invalid_userptr);
if (ret) {
pr_err("Failed to map bo to gpuvm\n");
goto out_unreserve;
continue;
kfd_mem_dmaunmap_attachment(mem, attachment);
- ret = update_gpuvm_pte(mem, attachment, &sync, NULL);
+ ret = update_gpuvm_pte(mem, attachment, &sync);
if (ret) {
pr_err("%s: update PTE failed\n", __func__);
/* make sure this gets validated again */
struct amdgpu_bo *bo = mem->bo;
uint32_t domain = mem->domain;
struct kfd_mem_attachment *attachment;
+ struct dma_resv_iter cursor;
+ struct dma_fence *fence;
total_size += amdgpu_bo_size(bo);
goto validate_map_fail;
}
}
- ret = amdgpu_sync_fence(&sync_obj, bo->tbo.moving);
- if (ret) {
- pr_debug("Memory eviction: Sync BO fence failed. Try again\n");
- goto validate_map_fail;
+ dma_resv_for_each_fence(&cursor, bo->tbo.base.resv,
+ DMA_RESV_USAGE_KERNEL, fence) {
+ ret = amdgpu_sync_fence(&sync_obj, fence);
+ if (ret) {
+ pr_debug("Memory eviction: Sync BO fence failed. Try again\n");
+ goto validate_map_fail;
+ }
}
list_for_each_entry(attachment, &mem->attachments, list) {
if (!attachment->is_mapped)
continue;
kfd_mem_dmaunmap_attachment(mem, attachment);
- ret = update_gpuvm_pte(mem, attachment, &sync_obj, NULL);
+ ret = update_gpuvm_pte(mem, attachment, &sync_obj);
if (ret) {
pr_debug("Memory eviction: update PTE failed. Try again\n");
goto validate_map_fail;
* Add process eviction fence to bo so they can
* evict each other.
*/
- ret = dma_resv_reserve_shared(gws_bo->tbo.base.resv, 1);
+ ret = dma_resv_reserve_fences(gws_bo->tbo.base.resv, 1);
if (ret)
goto reserve_shared_fail;
amdgpu_bo_fence(gws_bo, &process_info->eviction_fence->base, true);
bo = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj));
p->uf_entry.priority = 0;
p->uf_entry.tv.bo = &bo->tbo;
- /* One for TTM and one for the CS job */
- p->uf_entry.tv.num_shared = 2;
+ /* One for TTM and two for the CS job */
+ p->uf_entry.tv.num_shared = 3;
drm_gem_object_put(gobj);
struct amdgpu_bo *bo = ttm_to_amdgpu_bo(e->tv.bo);
e->bo_va = amdgpu_vm_bo_find(vm, bo);
-
- if (bo->tbo.base.dma_buf && !amdgpu_bo_explicit_sync(bo)) {
- e->chain = dma_fence_chain_alloc();
- if (!e->chain) {
- r = -ENOMEM;
- goto error_validate;
- }
- }
}
/* Move fence waiting after getting reservation lock of
}
error_validate:
- if (r) {
- amdgpu_bo_list_for_each_entry(e, p->bo_list) {
- dma_fence_chain_free(e->chain);
- e->chain = NULL;
- }
+ if (r)
ttm_eu_backoff_reservation(&p->ticket, &p->validated);
- }
out:
return r;
}
{
unsigned i;
- if (error && backoff) {
- struct amdgpu_bo_list_entry *e;
-
- amdgpu_bo_list_for_each_entry(e, parser->bo_list) {
- dma_fence_chain_free(e->chain);
- e->chain = NULL;
- }
-
+ if (error && backoff)
ttm_eu_backoff_reservation(&parser->ticket,
&parser->validated);
- }
for (i = 0; i < parser->num_post_deps; i++) {
drm_syncobj_put(parser->post_deps[i].syncobj);
if (r)
return r;
- r = amdgpu_vm_bo_update(adev, fpriv->prt_va, false, NULL);
+ r = amdgpu_vm_bo_update(adev, fpriv->prt_va, false);
if (r)
return r;
- r = amdgpu_sync_vm_fence(&p->job->sync, fpriv->prt_va->last_pt_update);
+ r = amdgpu_sync_fence(&p->job->sync, fpriv->prt_va->last_pt_update);
if (r)
return r;
if (amdgpu_mcbp || amdgpu_sriov_vf(adev)) {
bo_va = fpriv->csa_va;
BUG_ON(!bo_va);
- r = amdgpu_vm_bo_update(adev, bo_va, false, NULL);
+ r = amdgpu_vm_bo_update(adev, bo_va, false);
if (r)
return r;
- r = amdgpu_sync_vm_fence(&p->job->sync, bo_va->last_pt_update);
+ r = amdgpu_sync_fence(&p->job->sync, bo_va->last_pt_update);
if (r)
return r;
}
if (bo_va == NULL)
continue;
- r = amdgpu_vm_bo_update(adev, bo_va, false, NULL);
+ r = amdgpu_vm_bo_update(adev, bo_va, false);
if (r)
return r;
- r = amdgpu_sync_vm_fence(&p->job->sync, bo_va->last_pt_update);
+ r = amdgpu_sync_fence(&p->job->sync, bo_va->last_pt_update);
if (r)
return r;
}
if (r)
return r;
- r = amdgpu_sync_vm_fence(&p->job->sync, vm->last_update);
+ r = amdgpu_sync_fence(&p->job->sync, vm->last_update);
if (r)
return r;
amdgpu_vm_move_to_lru_tail(p->adev, &fpriv->vm);
- amdgpu_bo_list_for_each_entry(e, p->bo_list) {
- struct dma_resv *resv = e->tv.bo->base.resv;
- struct dma_fence_chain *chain = e->chain;
-
- if (!chain)
- continue;
-
- /*
- * Work around dma_resv shortcomings by wrapping up the
- * submission in a dma_fence_chain and add it as exclusive
- * fence.
- */
- dma_fence_chain_init(chain, dma_resv_excl_fence(resv),
- dma_fence_get(p->fence), 1);
-
- rcu_assign_pointer(resv->fence_excl, &chain->base);
- e->chain = NULL;
- }
+ /* Make sure all BOs are remembered as writers */
+ amdgpu_bo_list_for_each_entry(e, p->bo_list)
+ e->tv.num_shared = 0;
ttm_eu_fence_buffer_objects(&p->ticket, &p->validated, p->fence);
mutex_unlock(&p->adev->notifier_lock);
#include <drm/drm_fourcc.h>
#include <drm/drm_vblank.h>
+ static int amdgpu_display_framebuffer_init(struct drm_device *dev,
+ struct amdgpu_framebuffer *rfb,
+ const struct drm_mode_fb_cmd2 *mode_cmd,
+ struct drm_gem_object *obj);
+
static void amdgpu_display_flip_callback(struct dma_fence *f,
struct dma_fence_cb *cb)
{
spin_unlock_irqrestore(&crtc->dev->event_lock, flags);
- DRM_DEBUG_DRIVER("crtc:%d[%p], pflip_stat:AMDGPU_FLIP_SUBMITTED, work: %p,\n",
- amdgpu_crtc->crtc_id, amdgpu_crtc, work);
+ drm_dbg_vbl(adev_to_drm(adev),
+ "crtc:%d[%p], pflip_stat:AMDGPU_FLIP_SUBMITTED, work: %p,\n",
+ amdgpu_crtc->crtc_id, amdgpu_crtc, work);
}
goto unpin;
}
- /* TODO: Unify this with other drivers */
- r = dma_resv_get_fences(new_abo->tbo.base.resv, true,
+ r = dma_resv_get_fences(new_abo->tbo.base.resv, DMA_RESV_USAGE_WRITE,
&work->shared_count,
&work->shared);
if (unlikely(r != 0)) {
return r;
}
- int amdgpu_display_gem_fb_init(struct drm_device *dev,
- struct amdgpu_framebuffer *rfb,
- const struct drm_mode_fb_cmd2 *mode_cmd,
- struct drm_gem_object *obj)
- {
- int ret;
-
- rfb->base.obj[0] = obj;
- drm_helper_mode_fill_fb_struct(dev, &rfb->base, mode_cmd);
-
- ret = amdgpu_display_framebuffer_init(dev, rfb, mode_cmd, obj);
- if (ret)
- goto err;
-
- ret = drm_framebuffer_init(dev, &rfb->base, &amdgpu_fb_funcs);
- if (ret)
- goto err;
-
- return 0;
- err:
- drm_dbg_kms(dev, "Failed to init gem fb: %d\n", ret);
- rfb->base.obj[0] = NULL;
- return ret;
- }
-
- int amdgpu_display_gem_fb_verify_and_init(
- struct drm_device *dev, struct amdgpu_framebuffer *rfb,
- struct drm_file *file_priv, const struct drm_mode_fb_cmd2 *mode_cmd,
- struct drm_gem_object *obj)
+ static int amdgpu_display_gem_fb_verify_and_init(struct drm_device *dev,
+ struct amdgpu_framebuffer *rfb,
+ struct drm_file *file_priv,
+ const struct drm_mode_fb_cmd2 *mode_cmd,
+ struct drm_gem_object *obj)
{
int ret;
return ret;
}
- int amdgpu_display_framebuffer_init(struct drm_device *dev,
- struct amdgpu_framebuffer *rfb,
- const struct drm_mode_fb_cmd2 *mode_cmd,
- struct drm_gem_object *obj)
+ static int amdgpu_display_framebuffer_init(struct drm_device *dev,
+ struct amdgpu_framebuffer *rfb,
+ const struct drm_mode_fb_cmd2 *mode_cmd,
+ struct drm_gem_object *obj)
{
struct amdgpu_device *adev = drm_to_adev(dev);
int ret, i;
return -ENOENT;
}
robj = gem_to_amdgpu_bo(gobj);
- ret = dma_resv_wait_timeout(robj->tbo.base.resv, true, true, timeout);
+ ret = dma_resv_wait_timeout(robj->tbo.base.resv, DMA_RESV_USAGE_READ,
+ true, timeout);
/* ret == 0 means not signaled,
* ret > 0 means signaled
if (operation == AMDGPU_VA_OP_MAP ||
operation == AMDGPU_VA_OP_REPLACE) {
- r = amdgpu_vm_bo_update(adev, bo_va, false, NULL);
+ r = amdgpu_vm_bo_update(adev, bo_va, false);
if (r)
goto error;
}
void amdgpu_pasid_free_delayed(struct dma_resv *resv,
u32 pasid)
{
- struct dma_fence *fence, **fences;
struct amdgpu_pasid_cb *cb;
- unsigned count;
+ struct dma_fence *fence;
int r;
- r = dma_resv_get_fences(resv, true, &count, &fences);
+ r = dma_resv_get_singleton(resv, DMA_RESV_USAGE_BOOKKEEP, &fence);
if (r)
goto fallback;
- if (count == 0) {
+ if (!fence) {
amdgpu_pasid_free(pasid);
return;
}
- if (count == 1) {
- fence = fences[0];
- kfree(fences);
- } else {
- uint64_t context = dma_fence_context_alloc(1);
- struct dma_fence_array *array;
-
- array = dma_fence_array_create(count, fences, context,
- 1, false);
- if (!array) {
- kfree(fences);
- goto fallback;
- }
- fence = &array->base;
- }
-
cb = kmalloc(sizeof(*cb), GFP_KERNEL);
if (!cb) {
/* Last resort when we are OOM */
/* Not enough memory for the delayed delete, as last resort
* block for all the fences to complete.
*/
- dma_resv_wait_timeout(resv, true, false, MAX_SCHEDULE_TIMEOUT);
+ dma_resv_wait_timeout(resv, DMA_RESV_USAGE_BOOKKEEP,
+ false, MAX_SCHEDULE_TIMEOUT);
amdgpu_pasid_free(pasid);
}
struct amdgpu_device *adev = ring->adev;
unsigned vmhub = ring->funcs->vmhub;
uint64_t fence_context = adev->fence_context + ring->idx;
- struct dma_fence *updates = sync->last_vm_update;
bool needs_flush = vm->use_cpu_for_update;
- int r = 0;
+ uint64_t updates = amdgpu_vm_tlb_seq(vm);
+ int r;
*id = vm->reserved_vmid[vmhub];
- if (updates && (*id)->flushed_updates &&
- updates->context == (*id)->flushed_updates->context &&
- !dma_fence_is_later(updates, (*id)->flushed_updates))
- updates = NULL;
-
if ((*id)->owner != vm->immediate.fence_context ||
- job->vm_pd_addr != (*id)->pd_gpu_addr ||
- updates || !(*id)->last_flush ||
+ (*id)->pd_gpu_addr != job->vm_pd_addr ||
+ (*id)->flushed_updates < updates ||
+ !(*id)->last_flush ||
((*id)->last_flush->context != fence_context &&
!dma_fence_is_signaled((*id)->last_flush))) {
struct dma_fence *tmp;
tmp = amdgpu_sync_peek_fence(&(*id)->active, ring);
if (tmp) {
*id = NULL;
- r = amdgpu_sync_fence(sync, tmp);
- return r;
+ return amdgpu_sync_fence(sync, tmp);
}
needs_flush = true;
}
if (r)
return r;
- if (updates) {
- dma_fence_put((*id)->flushed_updates);
- (*id)->flushed_updates = dma_fence_get(updates);
- }
+ (*id)->flushed_updates = updates;
job->vm_needs_flush = needs_flush;
return 0;
}
unsigned vmhub = ring->funcs->vmhub;
struct amdgpu_vmid_mgr *id_mgr = &adev->vm_manager.id_mgr[vmhub];
uint64_t fence_context = adev->fence_context + ring->idx;
- struct dma_fence *updates = sync->last_vm_update;
+ uint64_t updates = amdgpu_vm_tlb_seq(vm);
int r;
job->vm_needs_flush = vm->use_cpu_for_update;
/* Check if we can use a VMID already assigned to this VM */
list_for_each_entry_reverse((*id), &id_mgr->ids_lru, list) {
bool needs_flush = vm->use_cpu_for_update;
- struct dma_fence *flushed;
/* Check all the prerequisites to using this VMID */
if ((*id)->owner != vm->immediate.fence_context)
!dma_fence_is_signaled((*id)->last_flush)))
needs_flush = true;
- flushed = (*id)->flushed_updates;
- if (updates && (!flushed || dma_fence_is_later(updates, flushed)))
+ if ((*id)->flushed_updates < updates)
needs_flush = true;
if (needs_flush && !adev->vm_manager.concurrent_flush)
if (r)
return r;
- if (updates && (!flushed || dma_fence_is_later(updates, flushed))) {
- dma_fence_put((*id)->flushed_updates);
- (*id)->flushed_updates = dma_fence_get(updates);
- }
-
+ (*id)->flushed_updates = updates;
job->vm_needs_flush |= needs_flush;
return 0;
}
goto error;
if (!id) {
- struct dma_fence *updates = sync->last_vm_update;
-
/* Still no ID to use? Then use the idle one found earlier */
id = idle;
if (r)
goto error;
- dma_fence_put(id->flushed_updates);
- id->flushed_updates = dma_fence_get(updates);
+ id->flushed_updates = amdgpu_vm_tlb_seq(vm);
job->vm_needs_flush = true;
}
struct amdgpu_vmid *id = &id_mgr->ids[j];
amdgpu_sync_free(&id->active);
- dma_fence_put(id->flushed_updates);
dma_fence_put(id->last_flush);
dma_fence_put(id->pasid_mapping);
}
fail:
DRM_DEBUG("BO size %lu > total memory in domain: %llu\n", size,
- man->size << PAGE_SHIFT);
+ man->size);
return false;
}
if (unlikely(r))
goto fail_unreserve;
- amdgpu_bo_fence(bo, fence, false);
- dma_fence_put(bo->tbo.moving);
- bo->tbo.moving = dma_fence_get(fence);
+ dma_resv_add_fence(bo->tbo.base.resv, fence,
+ DMA_RESV_USAGE_KERNEL);
dma_fence_put(fence);
}
if (!bp->resv)
if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)
return -EPERM;
+ r = dma_resv_wait_timeout(bo->tbo.base.resv, DMA_RESV_USAGE_KERNEL,
+ false, MAX_SCHEDULE_TIMEOUT);
+ if (r < 0)
+ return r;
+
kptr = amdgpu_bo_kptr(bo);
if (kptr) {
if (ptr)
return 0;
}
- r = dma_resv_wait_timeout(bo->tbo.base.resv, false, false,
- MAX_SCHEDULE_TIMEOUT);
- if (r < 0)
- return r;
-
r = ttm_bo_kmap(&bo->tbo, 0, bo->tbo.resource->num_pages, &bo->kmap);
if (r)
return r;
bool shared)
{
struct dma_resv *resv = bo->tbo.base.resv;
+ int r;
- if (shared)
- dma_resv_add_shared_fence(resv, fence);
- else
- dma_resv_add_excl_fence(resv, fence);
+ r = dma_resv_reserve_fences(resv, 1);
+ if (r) {
+ /* As last resort on OOM we block for the fence */
+ dma_fence_wait(fence, false);
+ return;
+ }
+
+ dma_resv_add_fence(resv, fence, shared ? DMA_RESV_USAGE_READ :
+ DMA_RESV_USAGE_WRITE);
}
/**
void amdgpu_sync_create(struct amdgpu_sync *sync)
{
hash_init(sync->fences);
- sync->last_vm_update = NULL;
}
/**
return 0;
}
- /**
- * amdgpu_sync_vm_fence - remember to sync to this VM fence
- *
- * @sync: sync object to add fence to
- * @fence: the VM fence to add
- *
- * Add the fence to the sync object and remember it as VM update.
- */
- int amdgpu_sync_vm_fence(struct amdgpu_sync *sync, struct dma_fence *fence)
- {
- if (!fence)
- return 0;
-
- amdgpu_sync_keep_later(&sync->last_vm_update, fence);
- return amdgpu_sync_fence(sync, fence);
- }
-
/* Determine based on the owner and mode if we should sync to a fence or not */
static bool amdgpu_sync_test_fence(struct amdgpu_device *adev,
enum amdgpu_sync_mode mode,
if (resv == NULL)
return -EINVAL;
- dma_resv_for_each_fence(&cursor, resv, true, f) {
+ /* TODO: Use DMA_RESV_USAGE_READ here */
+ dma_resv_for_each_fence(&cursor, resv, DMA_RESV_USAGE_BOOKKEEP, f) {
dma_fence_chain_for_each(f, f) {
struct dma_fence *tmp = dma_fence_chain_contained(f);
}
}
- dma_fence_put(clone->last_vm_update);
- clone->last_vm_update = dma_fence_get(source->last_vm_update);
-
return 0;
}
dma_fence_put(e->fence);
kmem_cache_free(amdgpu_sync_slab, e);
}
-
- dma_fence_put(sync->last_vm_update);
}
/**
struct dma_fence_cb cb;
};
+ /**
+ * struct amdgpu_vm_tlb_seq_cb - Helper to increment the TLB flush sequence
+ */
+ struct amdgpu_vm_tlb_seq_cb {
+ /**
+ * @vm: pointer to the amdgpu_vm structure to set the fence sequence on
+ */
+ struct amdgpu_vm *vm;
+
+ /**
+ * @cb: callback
+ */
+ struct dma_fence_cb cb;
+ };
+
/**
* amdgpu_vm_set_pasid - manage pasid and vm ptr mapping
*
mutex_unlock(&vm->eviction_lock);
}
- /**
- * amdgpu_vm_level_shift - return the addr shift for each level
- *
- * @adev: amdgpu_device pointer
- * @level: VMPT level
- *
- * Returns:
- * The number of bits the pfn needs to be right shifted for a level.
- */
- static unsigned amdgpu_vm_level_shift(struct amdgpu_device *adev,
- unsigned level)
- {
- switch (level) {
- case AMDGPU_VM_PDB2:
- case AMDGPU_VM_PDB1:
- case AMDGPU_VM_PDB0:
- return 9 * (AMDGPU_VM_PDB0 - level) +
- adev->vm_manager.block_size;
- case AMDGPU_VM_PTB:
- return 0;
- default:
- return ~0;
- }
- }
-
- /**
- * amdgpu_vm_num_entries - return the number of entries in a PD/PT
- *
- * @adev: amdgpu_device pointer
- * @level: VMPT level
- *
- * Returns:
- * The number of entries in a page directory or page table.
- */
- static unsigned amdgpu_vm_num_entries(struct amdgpu_device *adev,
- unsigned level)
- {
- unsigned shift = amdgpu_vm_level_shift(adev,
- adev->vm_manager.root_level);
-
- if (level == adev->vm_manager.root_level)
- /* For the root directory */
- return round_up(adev->vm_manager.max_pfn, 1ULL << shift)
- >> shift;
- else if (level != AMDGPU_VM_PTB)
- /* Everything in between */
- return 512;
- else
- /* For the page tables on the leaves */
- return AMDGPU_VM_PTE_COUNT(adev);
- }
-
- /**
- * amdgpu_vm_num_ats_entries - return the number of ATS entries in the root PD
- *
- * @adev: amdgpu_device pointer
- *
- * Returns:
- * The number of entries in the root page directory which needs the ATS setting.
- */
- static unsigned amdgpu_vm_num_ats_entries(struct amdgpu_device *adev)
- {
- unsigned shift;
-
- shift = amdgpu_vm_level_shift(adev, adev->vm_manager.root_level);
- return AMDGPU_GMC_HOLE_START >> (shift + AMDGPU_GPU_PAGE_SHIFT);
- }
-
- /**
- * amdgpu_vm_entries_mask - the mask to get the entry number of a PD/PT
- *
- * @adev: amdgpu_device pointer
- * @level: VMPT level
- *
- * Returns:
- * The mask to extract the entry number of a PD/PT from an address.
- */
- static uint32_t amdgpu_vm_entries_mask(struct amdgpu_device *adev,
- unsigned int level)
- {
- if (level <= adev->vm_manager.root_level)
- return 0xffffffff;
- else if (level != AMDGPU_VM_PTB)
- return 0x1ff;
- else
- return AMDGPU_VM_PTE_COUNT(adev) - 1;
- }
-
- /**
- * amdgpu_vm_bo_size - returns the size of the BOs in bytes
- *
- * @adev: amdgpu_device pointer
- * @level: VMPT level
- *
- * Returns:
- * The size of the BO for a page directory or page table in bytes.
- */
- static unsigned amdgpu_vm_bo_size(struct amdgpu_device *adev, unsigned level)
- {
- return AMDGPU_GPU_PAGE_ALIGN(amdgpu_vm_num_entries(adev, level) * 8);
- }
-
/**
* amdgpu_vm_bo_evicted - vm_bo is evicted
*
* Initialize a bo_va_base structure and add it to the appropriate lists
*
*/
- static void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base,
- struct amdgpu_vm *vm,
- struct amdgpu_bo *bo)
+ void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base,
+ struct amdgpu_vm *vm, struct amdgpu_bo *bo)
{
base->vm = vm;
base->bo = bo;
dma_resv_assert_held(vm->root.bo->tbo.base.resv);
- vm->bulk_moveable = false;
+ ttm_bo_set_bulk_move(&bo->tbo, &vm->lru_bulk_move);
if (bo->tbo.type == ttm_bo_type_kernel && bo->parent)
amdgpu_vm_bo_relocated(base);
else
amdgpu_vm_bo_evicted(base);
}
- /**
- * amdgpu_vm_pt_parent - get the parent page directory
- *
- * @pt: child page table
- *
- * Helper to get the parent entry for the child page table. NULL if we are at
- * the root page directory.
- */
- static struct amdgpu_vm_bo_base *amdgpu_vm_pt_parent(struct amdgpu_vm_bo_base *pt)
- {
- struct amdgpu_bo *parent = pt->bo->parent;
-
- if (!parent)
- return NULL;
-
- return parent->vm_bo;
- }
-
- /*
- * amdgpu_vm_pt_cursor - state for for_each_amdgpu_vm_pt
- */
- struct amdgpu_vm_pt_cursor {
- uint64_t pfn;
- struct amdgpu_vm_bo_base *parent;
- struct amdgpu_vm_bo_base *entry;
- unsigned level;
- };
-
- /**
- * amdgpu_vm_pt_start - start PD/PT walk
- *
- * @adev: amdgpu_device pointer
- * @vm: amdgpu_vm structure
- * @start: start address of the walk
- * @cursor: state to initialize
- *
- * Initialize a amdgpu_vm_pt_cursor to start a walk.
- */
- static void amdgpu_vm_pt_start(struct amdgpu_device *adev,
- struct amdgpu_vm *vm, uint64_t start,
- struct amdgpu_vm_pt_cursor *cursor)
- {
- cursor->pfn = start;
- cursor->parent = NULL;
- cursor->entry = &vm->root;
- cursor->level = adev->vm_manager.root_level;
- }
-
- /**
- * amdgpu_vm_pt_descendant - go to child node
- *
- * @adev: amdgpu_device pointer
- * @cursor: current state
- *
- * Walk to the child node of the current node.
- * Returns:
- * True if the walk was possible, false otherwise.
- */
- static bool amdgpu_vm_pt_descendant(struct amdgpu_device *adev,
- struct amdgpu_vm_pt_cursor *cursor)
- {
- unsigned mask, shift, idx;
-
- if ((cursor->level == AMDGPU_VM_PTB) || !cursor->entry ||
- !cursor->entry->bo)
- return false;
-
- mask = amdgpu_vm_entries_mask(adev, cursor->level);
- shift = amdgpu_vm_level_shift(adev, cursor->level);
-
- ++cursor->level;
- idx = (cursor->pfn >> shift) & mask;
- cursor->parent = cursor->entry;
- cursor->entry = &to_amdgpu_bo_vm(cursor->entry->bo)->entries[idx];
- return true;
- }
-
- /**
- * amdgpu_vm_pt_sibling - go to sibling node
- *
- * @adev: amdgpu_device pointer
- * @cursor: current state
- *
- * Walk to the sibling node of the current node.
- * Returns:
- * True if the walk was possible, false otherwise.
- */
- static bool amdgpu_vm_pt_sibling(struct amdgpu_device *adev,
- struct amdgpu_vm_pt_cursor *cursor)
- {
- unsigned shift, num_entries;
-
- /* Root doesn't have a sibling */
- if (!cursor->parent)
- return false;
-
- /* Go to our parents and see if we got a sibling */
- shift = amdgpu_vm_level_shift(adev, cursor->level - 1);
- num_entries = amdgpu_vm_num_entries(adev, cursor->level - 1);
-
- if (cursor->entry == &to_amdgpu_bo_vm(cursor->parent->bo)->entries[num_entries - 1])
- return false;
-
- cursor->pfn += 1ULL << shift;
- cursor->pfn &= ~((1ULL << shift) - 1);
- ++cursor->entry;
- return true;
- }
-
- /**
- * amdgpu_vm_pt_ancestor - go to parent node
- *
- * @cursor: current state
- *
- * Walk to the parent node of the current node.
- * Returns:
- * True if the walk was possible, false otherwise.
- */
- static bool amdgpu_vm_pt_ancestor(struct amdgpu_vm_pt_cursor *cursor)
- {
- if (!cursor->parent)
- return false;
-
- --cursor->level;
- cursor->entry = cursor->parent;
- cursor->parent = amdgpu_vm_pt_parent(cursor->parent);
- return true;
- }
-
- /**
- * amdgpu_vm_pt_next - get next PD/PT in hieratchy
- *
- * @adev: amdgpu_device pointer
- * @cursor: current state
- *
- * Walk the PD/PT tree to the next node.
- */
- static void amdgpu_vm_pt_next(struct amdgpu_device *adev,
- struct amdgpu_vm_pt_cursor *cursor)
- {
- /* First try a newborn child */
- if (amdgpu_vm_pt_descendant(adev, cursor))
- return;
-
- /* If that didn't worked try to find a sibling */
- while (!amdgpu_vm_pt_sibling(adev, cursor)) {
- /* No sibling, go to our parents and grandparents */
- if (!amdgpu_vm_pt_ancestor(cursor)) {
- cursor->pfn = ~0ll;
- return;
- }
- }
- }
-
- /**
- * amdgpu_vm_pt_first_dfs - start a deep first search
- *
- * @adev: amdgpu_device structure
- * @vm: amdgpu_vm structure
- * @start: optional cursor to start with
- * @cursor: state to initialize
- *
- * Starts a deep first traversal of the PD/PT tree.
- */
- static void amdgpu_vm_pt_first_dfs(struct amdgpu_device *adev,
- struct amdgpu_vm *vm,
- struct amdgpu_vm_pt_cursor *start,
- struct amdgpu_vm_pt_cursor *cursor)
- {
- if (start)
- *cursor = *start;
- else
- amdgpu_vm_pt_start(adev, vm, 0, cursor);
- while (amdgpu_vm_pt_descendant(adev, cursor));
- }
-
- /**
- * amdgpu_vm_pt_continue_dfs - check if the deep first search should continue
- *
- * @start: starting point for the search
- * @entry: current entry
- *
- * Returns:
- * True when the search should continue, false otherwise.
- */
- static bool amdgpu_vm_pt_continue_dfs(struct amdgpu_vm_pt_cursor *start,
- struct amdgpu_vm_bo_base *entry)
- {
- return entry && (!start || entry != start->entry);
- }
-
- /**
- * amdgpu_vm_pt_next_dfs - get the next node for a deep first search
- *
- * @adev: amdgpu_device structure
- * @cursor: current state
- *
- * Move the cursor to the next node in a deep first search.
- */
- static void amdgpu_vm_pt_next_dfs(struct amdgpu_device *adev,
- struct amdgpu_vm_pt_cursor *cursor)
- {
- if (!cursor->entry)
- return;
-
- if (!cursor->parent)
- cursor->entry = NULL;
- else if (amdgpu_vm_pt_sibling(adev, cursor))
- while (amdgpu_vm_pt_descendant(adev, cursor));
- else
- amdgpu_vm_pt_ancestor(cursor);
- }
-
- /*
- * for_each_amdgpu_vm_pt_dfs_safe - safe deep first search of all PDs/PTs
- */
- #define for_each_amdgpu_vm_pt_dfs_safe(adev, vm, start, cursor, entry) \
- for (amdgpu_vm_pt_first_dfs((adev), (vm), (start), &(cursor)), \
- (entry) = (cursor).entry, amdgpu_vm_pt_next_dfs((adev), &(cursor));\
- amdgpu_vm_pt_continue_dfs((start), (entry)); \
- (entry) = (cursor).entry, amdgpu_vm_pt_next_dfs((adev), &(cursor)))
-
/**
* amdgpu_vm_get_pd_bo - add the VM PD to a validation list
*
list_add(&entry->tv.head, validated);
}
-/**
- * amdgpu_vm_del_from_lru_notify - update bulk_moveable flag
- *
- * @bo: BO which was removed from the LRU
- *
- * Make sure the bulk_moveable flag is updated when a BO is removed from the
- * LRU.
- */
-void amdgpu_vm_del_from_lru_notify(struct ttm_buffer_object *bo)
-{
- struct amdgpu_bo *abo;
- struct amdgpu_vm_bo_base *bo_base;
-
- if (!amdgpu_bo_is_amdgpu_bo(bo))
- return;
-
- if (bo->pin_count)
- return;
-
- abo = ttm_to_amdgpu_bo(bo);
- if (!abo->parent)
- return;
- for (bo_base = abo->vm_bo; bo_base; bo_base = bo_base->next) {
- struct amdgpu_vm *vm = bo_base->vm;
-
- if (abo->tbo.base.resv == vm->root.bo->tbo.base.resv)
- vm->bulk_moveable = false;
- }
-
-}
/**
* amdgpu_vm_move_to_lru_tail - move all BOs to the end of LRU
*
void amdgpu_vm_move_to_lru_tail(struct amdgpu_device *adev,
struct amdgpu_vm *vm)
{
- struct amdgpu_vm_bo_base *bo_base;
-
- if (vm->bulk_moveable) {
- spin_lock(&adev->mman.bdev.lru_lock);
- ttm_bo_bulk_move_lru_tail(&vm->lru_bulk_move);
- spin_unlock(&adev->mman.bdev.lru_lock);
- return;
- }
-
- memset(&vm->lru_bulk_move, 0, sizeof(vm->lru_bulk_move));
-
spin_lock(&adev->mman.bdev.lru_lock);
- list_for_each_entry(bo_base, &vm->idle, vm_status) {
- struct amdgpu_bo *bo = bo_base->bo;
- struct amdgpu_bo *shadow = amdgpu_bo_shadowed(bo);
-
- if (!bo->parent)
- continue;
-
- ttm_bo_move_to_lru_tail(&bo->tbo, bo->tbo.resource,
- &vm->lru_bulk_move);
- if (shadow)
- ttm_bo_move_to_lru_tail(&shadow->tbo,
- shadow->tbo.resource,
- &vm->lru_bulk_move);
- }
+ ttm_lru_bulk_move_tail(&vm->lru_bulk_move);
spin_unlock(&adev->mman.bdev.lru_lock);
-
- vm->bulk_moveable = true;
}
/**
struct amdgpu_vm_bo_base *bo_base, *tmp;
int r;
- vm->bulk_moveable &= list_empty(&vm->evicted);
-
list_for_each_entry_safe(bo_base, tmp, &vm->evicted, vm_status) {
struct amdgpu_bo *bo = bo_base->bo;
struct amdgpu_bo *shadow = amdgpu_bo_shadowed(bo);
return ret && list_empty(&vm->evicted);
}
- /**
- * amdgpu_vm_clear_bo - initially clear the PDs/PTs
- *
- * @adev: amdgpu_device pointer
- * @vm: VM to clear BO from
- * @vmbo: BO to clear
- * @immediate: use an immediate update
- *
- * Root PD needs to be reserved when calling this.
- *
- * Returns:
- * 0 on success, errno otherwise.
- */
- static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
- struct amdgpu_vm *vm,
- struct amdgpu_bo_vm *vmbo,
- bool immediate)
- {
- struct ttm_operation_ctx ctx = { true, false };
- unsigned level = adev->vm_manager.root_level;
- struct amdgpu_vm_update_params params;
- struct amdgpu_bo *ancestor = &vmbo->bo;
- struct amdgpu_bo *bo = &vmbo->bo;
- unsigned entries, ats_entries;
- uint64_t addr;
- int r, idx;
-
- /* Figure out our place in the hierarchy */
- if (ancestor->parent) {
- ++level;
- while (ancestor->parent->parent) {
- ++level;
- ancestor = ancestor->parent;
- }
- }
-
- entries = amdgpu_bo_size(bo) / 8;
- if (!vm->pte_support_ats) {
- ats_entries = 0;
-
- } else if (!bo->parent) {
- ats_entries = amdgpu_vm_num_ats_entries(adev);
- ats_entries = min(ats_entries, entries);
- entries -= ats_entries;
-
- } else {
- struct amdgpu_vm_bo_base *pt;
-
- pt = ancestor->vm_bo;
- ats_entries = amdgpu_vm_num_ats_entries(adev);
- if ((pt - to_amdgpu_bo_vm(vm->root.bo)->entries) >= ats_entries) {
- ats_entries = 0;
- } else {
- ats_entries = entries;
- entries = 0;
- }
- }
-
- r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
- if (r)
- return r;
-
- if (vmbo->shadow) {
- struct amdgpu_bo *shadow = vmbo->shadow;
-
- r = ttm_bo_validate(&shadow->tbo, &shadow->placement, &ctx);
- if (r)
- return r;
- }
-
- if (!drm_dev_enter(adev_to_drm(adev), &idx))
- return -ENODEV;
-
- r = vm->update_funcs->map_table(vmbo);
- if (r)
- goto exit;
-
- memset(¶ms, 0, sizeof(params));
- params.adev = adev;
- params.vm = vm;
- params.immediate = immediate;
-
- r = vm->update_funcs->prepare(¶ms, NULL, AMDGPU_SYNC_EXPLICIT);
- if (r)
- goto exit;
-
- addr = 0;
- if (ats_entries) {
- uint64_t value = 0, flags;
-
- flags = AMDGPU_PTE_DEFAULT_ATC;
- if (level != AMDGPU_VM_PTB) {
- /* Handle leaf PDEs as PTEs */
- flags |= AMDGPU_PDE_PTE;
- amdgpu_gmc_get_vm_pde(adev, level, &value, &flags);
- }
-
- r = vm->update_funcs->update(¶ms, vmbo, addr, 0, ats_entries,
- value, flags);
- if (r)
- goto exit;
-
- addr += ats_entries * 8;
- }
-
- if (entries) {
- uint64_t value = 0, flags = 0;
-
- if (adev->asic_type >= CHIP_VEGA10) {
- if (level != AMDGPU_VM_PTB) {
- /* Handle leaf PDEs as PTEs */
- flags |= AMDGPU_PDE_PTE;
- amdgpu_gmc_get_vm_pde(adev, level,
- &value, &flags);
- } else {
- /* Workaround for fault priority problem on GMC9 */
- flags = AMDGPU_PTE_EXECUTABLE;
- }
- }
-
- r = vm->update_funcs->update(¶ms, vmbo, addr, 0, entries,
- value, flags);
- if (r)
- goto exit;
- }
-
- r = vm->update_funcs->commit(¶ms, NULL);
- exit:
- drm_dev_exit(idx);
- return r;
- }
-
- /**
- * amdgpu_vm_pt_create - create bo for PD/PT
- *
- * @adev: amdgpu_device pointer
- * @vm: requesting vm
- * @level: the page table level
- * @immediate: use a immediate update
- * @vmbo: pointer to the buffer object pointer
- */
- static int amdgpu_vm_pt_create(struct amdgpu_device *adev,
- struct amdgpu_vm *vm,
- int level, bool immediate,
- struct amdgpu_bo_vm **vmbo)
- {
- struct amdgpu_bo_param bp;
- struct amdgpu_bo *bo;
- struct dma_resv *resv;
- unsigned int num_entries;
- int r;
-
- memset(&bp, 0, sizeof(bp));
-
- bp.size = amdgpu_vm_bo_size(adev, level);
- bp.byte_align = AMDGPU_GPU_PAGE_SIZE;
- bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
- bp.domain = amdgpu_bo_get_preferred_domain(adev, bp.domain);
- bp.flags = AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS |
- AMDGPU_GEM_CREATE_CPU_GTT_USWC;
-
- if (level < AMDGPU_VM_PTB)
- num_entries = amdgpu_vm_num_entries(adev, level);
- else
- num_entries = 0;
-
- bp.bo_ptr_size = struct_size((*vmbo), entries, num_entries);
-
- if (vm->use_cpu_for_update)
- bp.flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
-
- bp.type = ttm_bo_type_kernel;
- bp.no_wait_gpu = immediate;
- if (vm->root.bo)
- bp.resv = vm->root.bo->tbo.base.resv;
-
- r = amdgpu_bo_create_vm(adev, &bp, vmbo);
- if (r)
- return r;
-
- bo = &(*vmbo)->bo;
- if (vm->is_compute_context || (adev->flags & AMD_IS_APU)) {
- (*vmbo)->shadow = NULL;
- return 0;
- }
-
- if (!bp.resv)
- WARN_ON(dma_resv_lock(bo->tbo.base.resv,
- NULL));
- resv = bp.resv;
- memset(&bp, 0, sizeof(bp));
- bp.size = amdgpu_vm_bo_size(adev, level);
- bp.domain = AMDGPU_GEM_DOMAIN_GTT;
- bp.flags = AMDGPU_GEM_CREATE_CPU_GTT_USWC;
- bp.type = ttm_bo_type_kernel;
- bp.resv = bo->tbo.base.resv;
- bp.bo_ptr_size = sizeof(struct amdgpu_bo);
-
- r = amdgpu_bo_create(adev, &bp, &(*vmbo)->shadow);
-
- if (!resv)
- dma_resv_unlock(bo->tbo.base.resv);
-
- if (r) {
- amdgpu_bo_unref(&bo);
- return r;
- }
-
- (*vmbo)->shadow->parent = amdgpu_bo_ref(bo);
- amdgpu_bo_add_to_shadow_list(*vmbo);
-
- return 0;
- }
-
- /**
- * amdgpu_vm_alloc_pts - Allocate a specific page table
- *
- * @adev: amdgpu_device pointer
- * @vm: VM to allocate page tables for
- * @cursor: Which page table to allocate
- * @immediate: use an immediate update
- *
- * Make sure a specific page table or directory is allocated.
- *
- * Returns:
- * 1 if page table needed to be allocated, 0 if page table was already
- * allocated, negative errno if an error occurred.
- */
- static int amdgpu_vm_alloc_pts(struct amdgpu_device *adev,
- struct amdgpu_vm *vm,
- struct amdgpu_vm_pt_cursor *cursor,
- bool immediate)
- {
- struct amdgpu_vm_bo_base *entry = cursor->entry;
- struct amdgpu_bo *pt_bo;
- struct amdgpu_bo_vm *pt;
- int r;
-
- if (entry->bo)
- return 0;
-
- r = amdgpu_vm_pt_create(adev, vm, cursor->level, immediate, &pt);
- if (r)
- return r;
-
- /* Keep a reference to the root directory to avoid
- * freeing them up in the wrong order.
- */
- pt_bo = &pt->bo;
- pt_bo->parent = amdgpu_bo_ref(cursor->parent->bo);
- amdgpu_vm_bo_base_init(entry, vm, pt_bo);
- r = amdgpu_vm_clear_bo(adev, vm, pt, immediate);
- if (r)
- goto error_free_pt;
-
- return 0;
-
- error_free_pt:
- amdgpu_bo_unref(&pt->shadow);
- amdgpu_bo_unref(&pt_bo);
- return r;
- }
-
- /**
- * amdgpu_vm_free_table - fre one PD/PT
- *
- * @entry: PDE to free
- */
- static void amdgpu_vm_free_table(struct amdgpu_vm_bo_base *entry)
- {
- struct amdgpu_bo *shadow;
-
- if (!entry->bo)
- return;
-
- shadow = amdgpu_bo_shadowed(entry->bo);
- if (shadow) {
- ttm_bo_set_bulk_move(&shadow->tbo, NULL);
- amdgpu_bo_unref(&shadow);
- }
-
- ttm_bo_set_bulk_move(&entry->bo->tbo, NULL);
- entry->bo->vm_bo = NULL;
- list_del(&entry->vm_status);
- amdgpu_bo_unref(&entry->bo);
- }
-
- /**
- * amdgpu_vm_free_pts - free PD/PT levels
- *
- * @adev: amdgpu device structure
- * @vm: amdgpu vm structure
- * @start: optional cursor where to start freeing PDs/PTs
- *
- * Free the page directory or page table level and all sub levels.
- */
- static void amdgpu_vm_free_pts(struct amdgpu_device *adev,
- struct amdgpu_vm *vm,
- struct amdgpu_vm_pt_cursor *start)
- {
- struct amdgpu_vm_pt_cursor cursor;
- struct amdgpu_vm_bo_base *entry;
-
- for_each_amdgpu_vm_pt_dfs_safe(adev, vm, start, cursor, entry)
- amdgpu_vm_free_table(entry);
-
- if (start)
- amdgpu_vm_free_table(start->entry);
- }
-
/**
* amdgpu_vm_check_compute_bug - check whether asic has compute vm bug
*
return result;
}
- /**
- * amdgpu_vm_update_pde - update a single level in the hierarchy
- *
- * @params: parameters for the update
- * @vm: requested vm
- * @entry: entry to update
- *
- * Makes sure the requested entry in parent is up to date.
- */
- static int amdgpu_vm_update_pde(struct amdgpu_vm_update_params *params,
- struct amdgpu_vm *vm,
- struct amdgpu_vm_bo_base *entry)
- {
- struct amdgpu_vm_bo_base *parent = amdgpu_vm_pt_parent(entry);
- struct amdgpu_bo *bo = parent->bo, *pbo;
- uint64_t pde, pt, flags;
- unsigned level;
-
- for (level = 0, pbo = bo->parent; pbo; ++level)
- pbo = pbo->parent;
-
- level += params->adev->vm_manager.root_level;
- amdgpu_gmc_get_pde_for_bo(entry->bo, level, &pt, &flags);
- pde = (entry - to_amdgpu_bo_vm(parent->bo)->entries) * 8;
- return vm->update_funcs->update(params, to_amdgpu_bo_vm(bo), pde, pt,
- 1, 0, flags);
- }
-
- /**
- * amdgpu_vm_invalidate_pds - mark all PDs as invalid
- *
- * @adev: amdgpu_device pointer
- * @vm: related vm
- *
- * Mark all PD level as invalid after an error.
- */
- static void amdgpu_vm_invalidate_pds(struct amdgpu_device *adev,
- struct amdgpu_vm *vm)
- {
- struct amdgpu_vm_pt_cursor cursor;
- struct amdgpu_vm_bo_base *entry;
-
- for_each_amdgpu_vm_pt_dfs_safe(adev, vm, NULL, cursor, entry)
- if (entry->bo && !entry->moved)
- amdgpu_vm_bo_relocated(entry);
- }
-
/**
* amdgpu_vm_update_pdes - make sure that all directories are valid
*
struct amdgpu_vm *vm, bool immediate)
{
struct amdgpu_vm_update_params params;
+ struct amdgpu_vm_bo_base *entry;
int r, idx;
if (list_empty(&vm->relocated))
r = vm->update_funcs->prepare(¶ms, NULL, AMDGPU_SYNC_EXPLICIT);
if (r)
- goto exit;
-
- while (!list_empty(&vm->relocated)) {
- struct amdgpu_vm_bo_base *entry;
-
- entry = list_first_entry(&vm->relocated,
- struct amdgpu_vm_bo_base,
- vm_status);
- amdgpu_vm_bo_idle(entry);
+ goto error;
- r = amdgpu_vm_update_pde(¶ms, vm, entry);
+ list_for_each_entry(entry, &vm->relocated, vm_status) {
+ r = amdgpu_vm_pde_update(¶ms, entry);
if (r)
goto error;
}
r = vm->update_funcs->commit(¶ms, &vm->last_update);
if (r)
goto error;
- drm_dev_exit(idx);
- return 0;
+
+ while (!list_empty(&vm->relocated)) {
+ entry = list_first_entry(&vm->relocated,
+ struct amdgpu_vm_bo_base,
+ vm_status);
+ amdgpu_vm_bo_idle(entry);
+ }
error:
- amdgpu_vm_invalidate_pds(adev, vm);
- exit:
drm_dev_exit(idx);
return r;
}
- /*
- * amdgpu_vm_update_flags - figure out flags for PTE updates
- *
- * Make sure to set the right flags for the PTEs at the desired level.
- */
- static void amdgpu_vm_update_flags(struct amdgpu_vm_update_params *params,
- struct amdgpu_bo_vm *pt, unsigned int level,
- uint64_t pe, uint64_t addr,
- unsigned int count, uint32_t incr,
- uint64_t flags)
-
- {
- if (level != AMDGPU_VM_PTB) {
- flags |= AMDGPU_PDE_PTE;
- amdgpu_gmc_get_vm_pde(params->adev, level, &addr, &flags);
-
- } else if (params->adev->asic_type >= CHIP_VEGA10 &&
- !(flags & AMDGPU_PTE_VALID) &&
- !(flags & AMDGPU_PTE_PRT)) {
-
- /* Workaround for fault priority problem on GMC9 */
- flags |= AMDGPU_PTE_EXECUTABLE;
- }
-
- params->vm->update_funcs->update(params, pt, pe, addr, count, incr,
- flags);
- }
-
/**
- * amdgpu_vm_fragment - get fragment for PTEs
+ * amdgpu_vm_tlb_seq_cb - make sure to increment tlb sequence
+ * @fence: unused
+ * @cb: the callback structure
*
- * @params: see amdgpu_vm_update_params definition
- * @start: first PTE to handle
- * @end: last PTE to handle
- * @flags: hw mapping flags
- * @frag: resulting fragment size
- * @frag_end: end of this fragment
- *
- * Returns the first possible fragment for the start and end address.
- */
- static void amdgpu_vm_fragment(struct amdgpu_vm_update_params *params,
- uint64_t start, uint64_t end, uint64_t flags,
- unsigned int *frag, uint64_t *frag_end)
- {
- /**
- * The MC L1 TLB supports variable sized pages, based on a fragment
- * field in the PTE. When this field is set to a non-zero value, page
- * granularity is increased from 4KB to (1 << (12 + frag)). The PTE
- * flags are considered valid for all PTEs within the fragment range
- * and corresponding mappings are assumed to be physically contiguous.
- *
- * The L1 TLB can store a single PTE for the whole fragment,
- * significantly increasing the space available for translation
- * caching. This leads to large improvements in throughput when the
- * TLB is under pressure.
- *
- * The L2 TLB distributes small and large fragments into two
- * asymmetric partitions. The large fragment cache is significantly
- * larger. Thus, we try to use large fragments wherever possible.
- * Userspace can support this by aligning virtual base address and
- * allocation size to the fragment size.
- *
- * Starting with Vega10 the fragment size only controls the L1. The L2
- * is now directly feed with small/huge/giant pages from the walker.
- */
- unsigned max_frag;
-
- if (params->adev->asic_type < CHIP_VEGA10)
- max_frag = params->adev->vm_manager.fragment_size;
- else
- max_frag = 31;
-
- /* system pages are non continuously */
- if (params->pages_addr) {
- *frag = 0;
- *frag_end = end;
- return;
- }
-
- /* This intentionally wraps around if no bit is set */
- *frag = min((unsigned)ffs(start) - 1, (unsigned)fls64(end - start) - 1);
- if (*frag >= max_frag) {
- *frag = max_frag;
- *frag_end = end & ~((1ULL << max_frag) - 1);
- } else {
- *frag_end = start + (1 << *frag);
- }
- }
-
- /**
- * amdgpu_vm_update_ptes - make sure that page tables are valid
- *
- * @params: see amdgpu_vm_update_params definition
- * @start: start of GPU address range
- * @end: end of GPU address range
- * @dst: destination address to map to, the next dst inside the function
- * @flags: mapping flags
- *
- * Update the page tables in the range @start - @end.
- *
- * Returns:
- * 0 for success, -EINVAL for failure.
+ * Increments the tlb sequence to make sure that future CS execute a VM flush.
*/
- static int amdgpu_vm_update_ptes(struct amdgpu_vm_update_params *params,
- uint64_t start, uint64_t end,
- uint64_t dst, uint64_t flags)
+ static void amdgpu_vm_tlb_seq_cb(struct dma_fence *fence,
+ struct dma_fence_cb *cb)
{
- struct amdgpu_device *adev = params->adev;
- struct amdgpu_vm_pt_cursor cursor;
- uint64_t frag_start = start, frag_end;
- unsigned int frag;
- int r;
-
- /* figure out the initial fragment */
- amdgpu_vm_fragment(params, frag_start, end, flags, &frag, &frag_end);
-
- /* walk over the address space and update the PTs */
- amdgpu_vm_pt_start(adev, params->vm, start, &cursor);
- while (cursor.pfn < end) {
- unsigned shift, parent_shift, mask;
- uint64_t incr, entry_end, pe_start;
- struct amdgpu_bo *pt;
-
- if (!params->unlocked) {
- /* make sure that the page tables covering the
- * address range are actually allocated
- */
- r = amdgpu_vm_alloc_pts(params->adev, params->vm,
- &cursor, params->immediate);
- if (r)
- return r;
- }
-
- shift = amdgpu_vm_level_shift(adev, cursor.level);
- parent_shift = amdgpu_vm_level_shift(adev, cursor.level - 1);
- if (params->unlocked) {
- /* Unlocked updates are only allowed on the leaves */
- if (amdgpu_vm_pt_descendant(adev, &cursor))
- continue;
- } else if (adev->asic_type < CHIP_VEGA10 &&
- (flags & AMDGPU_PTE_VALID)) {
- /* No huge page support before GMC v9 */
- if (cursor.level != AMDGPU_VM_PTB) {
- if (!amdgpu_vm_pt_descendant(adev, &cursor))
- return -ENOENT;
- continue;
- }
- } else if (frag < shift) {
- /* We can't use this level when the fragment size is
- * smaller than the address shift. Go to the next
- * child entry and try again.
- */
- if (amdgpu_vm_pt_descendant(adev, &cursor))
- continue;
- } else if (frag >= parent_shift) {
- /* If the fragment size is even larger than the parent
- * shift we should go up one level and check it again.
- */
- if (!amdgpu_vm_pt_ancestor(&cursor))
- return -EINVAL;
- continue;
- }
+ struct amdgpu_vm_tlb_seq_cb *tlb_cb;
- pt = cursor.entry->bo;
- if (!pt) {
- /* We need all PDs and PTs for mapping something, */
- if (flags & AMDGPU_PTE_VALID)
- return -ENOENT;
-
- /* but unmapping something can happen at a higher
- * level.
- */
- if (!amdgpu_vm_pt_ancestor(&cursor))
- return -EINVAL;
-
- pt = cursor.entry->bo;
- shift = parent_shift;
- frag_end = max(frag_end, ALIGN(frag_start + 1,
- 1ULL << shift));
- }
-
- /* Looks good so far, calculate parameters for the update */
- incr = (uint64_t)AMDGPU_GPU_PAGE_SIZE << shift;
- mask = amdgpu_vm_entries_mask(adev, cursor.level);
- pe_start = ((cursor.pfn >> shift) & mask) * 8;
- entry_end = ((uint64_t)mask + 1) << shift;
- entry_end += cursor.pfn & ~(entry_end - 1);
- entry_end = min(entry_end, end);
-
- do {
- struct amdgpu_vm *vm = params->vm;
- uint64_t upd_end = min(entry_end, frag_end);
- unsigned nptes = (upd_end - frag_start) >> shift;
- uint64_t upd_flags = flags | AMDGPU_PTE_FRAG(frag);
-
- /* This can happen when we set higher level PDs to
- * silent to stop fault floods.
- */
- nptes = max(nptes, 1u);
-
- trace_amdgpu_vm_update_ptes(params, frag_start, upd_end,
- min(nptes, 32u), dst, incr, upd_flags,
- vm->task_info.pid,
- vm->immediate.fence_context);
- amdgpu_vm_update_flags(params, to_amdgpu_bo_vm(pt),
- cursor.level, pe_start, dst,
- nptes, incr, upd_flags);
-
- pe_start += nptes * 8;
- dst += nptes * incr;
-
- frag_start = upd_end;
- if (frag_start >= frag_end) {
- /* figure out the next fragment */
- amdgpu_vm_fragment(params, frag_start, end,
- flags, &frag, &frag_end);
- if (frag < shift)
- break;
- }
- } while (frag_start < entry_end);
-
- if (amdgpu_vm_pt_descendant(adev, &cursor)) {
- /* Free all child entries.
- * Update the tables with the flags and addresses and free up subsequent
- * tables in the case of huge pages or freed up areas.
- * This is the maximum you can free, because all other page tables are not
- * completely covered by the range and so potentially still in use.
- */
- while (cursor.pfn < frag_start) {
- /* Make sure previous mapping is freed */
- if (cursor.entry->bo) {
- params->table_freed = true;
- amdgpu_vm_free_pts(adev, params->vm, &cursor);
- }
- amdgpu_vm_pt_next(adev, &cursor);
- }
-
- } else if (frag >= shift) {
- /* or just move on to the next on the same level. */
- amdgpu_vm_pt_next(adev, &cursor);
- }
- }
-
- return 0;
+ tlb_cb = container_of(cb, typeof(*tlb_cb), cb);
+ atomic64_inc(&tlb_cb->vm->tlb_seq);
+ kfree(tlb_cb);
}
/**
- * amdgpu_vm_bo_update_mapping - update a mapping in the vm page table
+ * amdgpu_vm_update_range - update a range in the vm page table
*
- * @adev: amdgpu_device pointer of the VM
- * @bo_adev: amdgpu_device pointer of the mapped BO
- * @vm: requested vm
+ * @adev: amdgpu_device pointer to use for commands
+ * @vm: the VM to update the range
* @immediate: immediate submission in a page fault
* @unlocked: unlocked invalidation during MM callback
+ * @flush_tlb: trigger tlb invalidation after update completed
* @resv: fences we need to sync to
* @start: start of mapped range
* @last: last mapped entry
* @flags: flags for the entries
* @offset: offset into nodes and pages_addr
+ * @vram_base: base for vram mappings
* @res: ttm_resource to map
* @pages_addr: DMA addresses to use for mapping
* @fence: optional resulting fence
- * @table_freed: return true if page table is freed
*
* Fill in the page table entries between @start and @last.
*
* Returns:
- * 0 for success, -EINVAL for failure.
+ * 0 for success, negative erro code for failure.
*/
- int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
- struct amdgpu_device *bo_adev,
- struct amdgpu_vm *vm, bool immediate,
- bool unlocked, struct dma_resv *resv,
- uint64_t start, uint64_t last,
- uint64_t flags, uint64_t offset,
- struct ttm_resource *res,
- dma_addr_t *pages_addr,
- struct dma_fence **fence,
- bool *table_freed)
+ int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+ bool immediate, bool unlocked, bool flush_tlb,
+ struct dma_resv *resv, uint64_t start, uint64_t last,
+ uint64_t flags, uint64_t offset, uint64_t vram_base,
+ struct ttm_resource *res, dma_addr_t *pages_addr,
+ struct dma_fence **fence)
{
struct amdgpu_vm_update_params params;
+ struct amdgpu_vm_tlb_seq_cb *tlb_cb;
struct amdgpu_res_cursor cursor;
enum amdgpu_sync_mode sync_mode;
int r, idx;
if (!drm_dev_enter(adev_to_drm(adev), &idx))
return -ENODEV;
+ tlb_cb = kmalloc(sizeof(*tlb_cb), GFP_KERNEL);
+ if (!tlb_cb) {
+ r = -ENOMEM;
+ goto error_unlock;
+ }
+
+ /* Vega20+XGMI where PTEs get inadvertently cached in L2 texture cache,
+ * heavy-weight flush TLB unconditionally.
+ */
+ flush_tlb |= adev->gmc.xgmi.num_physical_nodes &&
+ adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 0);
+
memset(¶ms, 0, sizeof(params));
params.adev = adev;
params.vm = vm;
amdgpu_vm_eviction_lock(vm);
if (vm->evicting) {
r = -EBUSY;
- goto error_unlock;
+ goto error_free;
}
if (!unlocked && !dma_fence_is_signaled(vm->last_unlocked)) {
r = vm->update_funcs->prepare(¶ms, resv, sync_mode);
if (r)
- goto error_unlock;
+ goto error_free;
amdgpu_res_first(pages_addr ? NULL : res, offset,
(last - start + 1) * AMDGPU_GPU_PAGE_SIZE, &cursor);
}
} else if (flags & (AMDGPU_PTE_VALID | AMDGPU_PTE_PRT)) {
- addr = bo_adev->vm_manager.vram_base_offset +
- cursor.start;
+ addr = vram_base + cursor.start;
} else {
addr = 0;
}
tmp = start + num_entries;
- r = amdgpu_vm_update_ptes(¶ms, start, tmp, addr, flags);
+ r = amdgpu_vm_ptes_update(¶ms, start, tmp, addr, flags);
if (r)
- goto error_unlock;
+ goto error_free;
amdgpu_res_next(&cursor, num_entries * AMDGPU_GPU_PAGE_SIZE);
start = tmp;
r = vm->update_funcs->commit(¶ms, fence);
- if (table_freed)
- *table_freed = *table_freed || params.table_freed;
+ if (flush_tlb || params.table_freed) {
+ tlb_cb->vm = vm;
+ if (fence && *fence &&
+ !dma_fence_add_callback(*fence, &tlb_cb->cb,
+ amdgpu_vm_tlb_seq_cb)) {
+ dma_fence_put(vm->last_tlb_flush);
+ vm->last_tlb_flush = dma_fence_get(*fence);
+ } else {
+ amdgpu_vm_tlb_seq_cb(NULL, &tlb_cb->cb);
+ }
+ tlb_cb = NULL;
+ }
+
+ error_free:
+ kfree(tlb_cb);
error_unlock:
amdgpu_vm_eviction_unlock(vm);
* @adev: amdgpu_device pointer
* @bo_va: requested BO and VM object
* @clear: if true clear the entries
- * @table_freed: return true if page table is freed
*
* Fill in the page table entries for @bo_va.
*
* 0 for success, -EINVAL for failure.
*/
int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va,
- bool clear, bool *table_freed)
+ bool clear)
{
struct amdgpu_bo *bo = bo_va->base.bo;
struct amdgpu_vm *vm = bo_va->base.vm;
dma_addr_t *pages_addr = NULL;
struct ttm_resource *mem;
struct dma_fence **last_update;
+ bool flush_tlb = clear;
struct dma_resv *resv;
+ uint64_t vram_base;
uint64_t flags;
- struct amdgpu_device *bo_adev = adev;
int r;
if (clear || !bo) {
}
if (bo) {
+ struct amdgpu_device *bo_adev;
+
flags = amdgpu_ttm_tt_pte_flags(adev, bo->tbo.ttm, mem);
if (amdgpu_bo_encrypted(bo))
flags |= AMDGPU_PTE_TMZ;
bo_adev = amdgpu_ttm_adev(bo->tbo.bdev);
+ vram_base = bo_adev->vm_manager.vram_base_offset;
} else {
flags = 0x0;
+ vram_base = 0;
}
if (clear || (bo && bo->tbo.base.resv ==
last_update = &bo_va->last_pt_update;
if (!clear && bo_va->base.moved) {
- bo_va->base.moved = false;
+ flush_tlb = true;
list_splice_init(&bo_va->valids, &bo_va->invalids);
} else if (bo_va->cleared != clear) {
trace_amdgpu_vm_bo_update(mapping);
- r = amdgpu_vm_bo_update_mapping(adev, bo_adev, vm, false, false,
- resv, mapping->start,
- mapping->last, update_flags,
- mapping->offset, mem,
- pages_addr, last_update, table_freed);
+ r = amdgpu_vm_update_range(adev, vm, false, false, flush_tlb,
+ resv, mapping->start, mapping->last,
+ update_flags, mapping->offset,
+ vram_base, mem, pages_addr,
+ last_update);
if (r)
return r;
}
list_splice_init(&bo_va->invalids, &bo_va->valids);
bo_va->cleared = clear;
+ bo_va->base.moved = false;
if (trace_amdgpu_vm_bo_mapping_enabled()) {
list_for_each_entry(mapping, &bo_va->valids, list)
struct dma_resv_iter cursor;
struct dma_fence *fence;
- dma_resv_for_each_fence(&cursor, resv, true, fence) {
+ dma_resv_for_each_fence(&cursor, resv, DMA_RESV_USAGE_BOOKKEEP, fence) {
/* Add a callback for each fence in the reservation object */
amdgpu_vm_prt_get(adev);
amdgpu_vm_add_prt_cb(adev, fence);
mapping->start < AMDGPU_GMC_HOLE_START)
init_pte_value = AMDGPU_PTE_DEFAULT_ATC;
- r = amdgpu_vm_bo_update_mapping(adev, adev, vm, false, false,
- resv, mapping->start,
- mapping->last, init_pte_value,
- 0, NULL, NULL, &f, NULL);
+ r = amdgpu_vm_update_range(adev, vm, false, false, true, resv,
+ mapping->start, mapping->last,
+ init_pte_value, 0, 0, NULL, NULL,
+ &f);
amdgpu_vm_free_mapping(adev, vm, mapping, f);
if (r) {
dma_fence_put(f);
list_for_each_entry_safe(bo_va, tmp, &vm->moved, base.vm_status) {
/* Per VM BOs never need to bo cleared in the page tables */
- r = amdgpu_vm_bo_update(adev, bo_va, false, NULL);
+ r = amdgpu_vm_bo_update(adev, bo_va, false);
if (r)
return r;
}
else
clear = true;
- r = amdgpu_vm_bo_update(adev, bo_va, clear, NULL);
+ r = amdgpu_vm_bo_update(adev, bo_va, clear);
if (r)
return r;
if (bo) {
dma_resv_assert_held(bo->tbo.base.resv);
if (bo->tbo.base.resv == vm->root.bo->tbo.base.resv)
- vm->bulk_moveable = false;
+ ttm_bo_set_bulk_move(&bo->tbo, NULL);
for (base = &bo_va->base.bo->vm_bo; *base;
base = &(*base)->next) {
return true;
/* Don't evict VM page tables while they are busy */
- if (!dma_resv_test_signaled(bo->tbo.base.resv, true))
+ if (!dma_resv_test_signaled(bo->tbo.base.resv, DMA_RESV_USAGE_BOOKKEEP))
return false;
/* Try to block ongoing updates */
*/
long amdgpu_vm_wait_idle(struct amdgpu_vm *vm, long timeout)
{
- timeout = dma_resv_wait_timeout(vm->root.bo->tbo.base.resv, true,
+ timeout = dma_resv_wait_timeout(vm->root.bo->tbo.base.resv,
+ DMA_RESV_USAGE_BOOKKEEP,
true, timeout);
if (timeout <= 0)
return timeout;
vm->update_funcs = &amdgpu_vm_sdma_funcs;
vm->last_update = NULL;
vm->last_unlocked = dma_fence_get_stub();
+ vm->last_tlb_flush = dma_fence_get_stub();
mutex_init(&vm->eviction_lock);
vm->evicting = false;
if (r)
goto error_free_root;
- r = dma_resv_reserve_shared(root_bo->tbo.base.resv, 1);
+ r = dma_resv_reserve_fences(root_bo->tbo.base.resv, 1);
if (r)
goto error_unreserve;
amdgpu_vm_bo_base_init(&vm->root, vm, root_bo);
- r = amdgpu_vm_clear_bo(adev, vm, root, false);
+ r = amdgpu_vm_pt_clear(adev, vm, root, false);
if (r)
goto error_unreserve;
vm->root.bo = NULL;
error_free_delayed:
+ dma_fence_put(vm->last_tlb_flush);
dma_fence_put(vm->last_unlocked);
drm_sched_entity_destroy(&vm->delayed);
return r;
}
- /**
- * amdgpu_vm_check_clean_reserved - check if a VM is clean
- *
- * @adev: amdgpu_device pointer
- * @vm: the VM to check
- *
- * check all entries of the root PD, if any subsequent PDs are allocated,
- * it means there are page table creating and filling, and is no a clean
- * VM
- *
- * Returns:
- * 0 if this VM is clean
- */
- static int amdgpu_vm_check_clean_reserved(struct amdgpu_device *adev,
- struct amdgpu_vm *vm)
- {
- enum amdgpu_vm_level root = adev->vm_manager.root_level;
- unsigned int entries = amdgpu_vm_num_entries(adev, root);
- unsigned int i = 0;
-
- for (i = 0; i < entries; i++) {
- if (to_amdgpu_bo_vm(vm->root.bo)->entries[i].bo)
- return -EINVAL;
- }
-
- return 0;
- }
-
/**
* amdgpu_vm_make_compute - Turn a GFX VM into a compute VM
*
return r;
/* Sanity checks */
- r = amdgpu_vm_check_clean_reserved(adev, vm);
- if (r)
+ if (!amdgpu_vm_pt_is_root_clean(adev, vm)) {
+ r = -EINVAL;
goto unreserve_bo;
+ }
/* Check if PD needs to be reinitialized and do it before
* changing any other state, in case it fails.
*/
if (pte_support_ats != vm->pte_support_ats) {
vm->pte_support_ats = pte_support_ats;
- r = amdgpu_vm_clear_bo(adev, vm,
- to_amdgpu_bo_vm(vm->root.bo),
+ r = amdgpu_vm_pt_clear(adev, vm, to_amdgpu_bo_vm(vm->root.bo),
false);
if (r)
goto unreserve_bo;
struct amdgpu_bo_va_mapping *mapping, *tmp;
bool prt_fini_needed = !!adev->gmc.gmc_funcs->set_prt;
struct amdgpu_bo *root;
+ unsigned long flags;
int i;
amdgpu_amdkfd_gpuvm_destroy_cb(adev, vm);
amdgpu_vm_set_pasid(adev, vm, 0);
dma_fence_wait(vm->last_unlocked, false);
dma_fence_put(vm->last_unlocked);
+ dma_fence_wait(vm->last_tlb_flush, false);
+ /* Make sure that all fence callbacks have completed */
+ spin_lock_irqsave(vm->last_tlb_flush->lock, flags);
+ spin_unlock_irqrestore(vm->last_tlb_flush->lock, flags);
+ dma_fence_put(vm->last_tlb_flush);
list_for_each_entry_safe(mapping, tmp, &vm->freed, list) {
if (mapping->flags & AMDGPU_PTE_PRT && prt_fini_needed) {
amdgpu_vm_free_mapping(adev, vm, mapping, NULL);
}
- amdgpu_vm_free_pts(adev, vm, NULL);
+ amdgpu_vm_pt_free_root(adev, vm);
amdgpu_bo_unreserve(root);
amdgpu_bo_unref(&root);
WARN_ON(vm->root.bo);
value = 0;
}
- r = dma_resv_reserve_shared(root->tbo.base.resv, 1);
+ r = dma_resv_reserve_fences(root->tbo.base.resv, 1);
if (r) {
pr_debug("failed %d to reserve fence slot\n", r);
goto error_unlock;
}
- r = amdgpu_vm_bo_update_mapping(adev, adev, vm, true, false, NULL, addr,
- addr, flags, value, NULL, NULL, NULL,
- NULL);
+ r = amdgpu_vm_update_range(adev, vm, true, false, false, NULL, addr,
+ addr, flags, value, 0, NULL, NULL, NULL);
if (r)
goto error_unlock;
struct drm_sched_entity immediate;
struct drm_sched_entity delayed;
+ /* Last finished delayed update */
+ atomic64_t tlb_seq;
+ struct dma_fence *last_tlb_flush;
+
/* Last unlocked submission to the scheduler entities */
struct dma_fence *last_unlocked;
/* Store positions of group of BOs */
struct ttm_lru_bulk_move lru_bulk_move;
- /* mark whether can do the bulk move */
- bool bulk_moveable;
/* Flag to indicate if VM is used for compute */
bool is_compute_context;
};
struct dma_fence **fence);
int amdgpu_vm_handle_moved(struct amdgpu_device *adev,
struct amdgpu_vm *vm);
- int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
- struct amdgpu_device *bo_adev,
- struct amdgpu_vm *vm, bool immediate,
- bool unlocked, struct dma_resv *resv,
- uint64_t start, uint64_t last,
- uint64_t flags, uint64_t offset,
- struct ttm_resource *res,
- dma_addr_t *pages_addr,
- struct dma_fence **fence, bool *free_table);
+ void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base,
+ struct amdgpu_vm *vm, struct amdgpu_bo *bo);
+ int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+ bool immediate, bool unlocked, bool flush_tlb,
+ struct dma_resv *resv, uint64_t start, uint64_t last,
+ uint64_t flags, uint64_t offset, uint64_t vram_base,
+ struct ttm_resource *res, dma_addr_t *pages_addr,
+ struct dma_fence **fence);
int amdgpu_vm_bo_update(struct amdgpu_device *adev,
struct amdgpu_bo_va *bo_va,
- bool clear, bool *table_freed);
+ bool clear);
bool amdgpu_vm_evictable(struct amdgpu_bo *bo);
void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,
struct amdgpu_bo *bo, bool evicted);
void amdgpu_vm_move_to_lru_tail(struct amdgpu_device *adev,
struct amdgpu_vm *vm);
-void amdgpu_vm_del_from_lru_notify(struct ttm_buffer_object *bo);
void amdgpu_vm_get_memory(struct amdgpu_vm *vm, uint64_t *vram_mem,
uint64_t *gtt_mem, uint64_t *cpu_mem);
+ int amdgpu_vm_pt_clear(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+ struct amdgpu_bo_vm *vmbo, bool immediate);
+ int amdgpu_vm_pt_create(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+ int level, bool immediate, struct amdgpu_bo_vm **vmbo);
+ void amdgpu_vm_pt_free_root(struct amdgpu_device *adev, struct amdgpu_vm *vm);
+ bool amdgpu_vm_pt_is_root_clean(struct amdgpu_device *adev,
+ struct amdgpu_vm *vm);
+
+ int amdgpu_vm_pde_update(struct amdgpu_vm_update_params *params,
+ struct amdgpu_vm_bo_base *entry);
+ int amdgpu_vm_ptes_update(struct amdgpu_vm_update_params *params,
+ uint64_t start, uint64_t end,
+ uint64_t dst, uint64_t flags);
+
#if defined(CONFIG_DEBUG_FS)
void amdgpu_debugfs_vm_bo_info(struct amdgpu_vm *vm, struct seq_file *m);
#endif
+ /**
+ * amdgpu_vm_tlb_seq - return tlb flush sequence number
+ * @vm: the amdgpu_vm structure to query
+ *
+ * Returns the tlb flush sequence number which indicates that the VM TLBs needs
+ * to be invalidated whenever the sequence number change.
+ */
+ static inline uint64_t amdgpu_vm_tlb_seq(struct amdgpu_vm *vm)
+ {
+ return atomic64_read(&vm->tlb_seq);
+ }
+
#endif
--- /dev/null
- vm->bulk_moveable = false;
-
+ // SPDX-License-Identifier: GPL-2.0 OR MIT
+ /*
+ * Copyright 2022 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+ #include <drm/drm_drv.h>
+
+ #include "amdgpu.h"
+ #include "amdgpu_trace.h"
+ #include "amdgpu_vm.h"
+
+ /*
+ * amdgpu_vm_pt_cursor - state for for_each_amdgpu_vm_pt
+ */
+ struct amdgpu_vm_pt_cursor {
+ uint64_t pfn;
+ struct amdgpu_vm_bo_base *parent;
+ struct amdgpu_vm_bo_base *entry;
+ unsigned int level;
+ };
+
+ /**
+ * amdgpu_vm_pt_level_shift - return the addr shift for each level
+ *
+ * @adev: amdgpu_device pointer
+ * @level: VMPT level
+ *
+ * Returns:
+ * The number of bits the pfn needs to be right shifted for a level.
+ */
+ static unsigned int amdgpu_vm_pt_level_shift(struct amdgpu_device *adev,
+ unsigned int level)
+ {
+ switch (level) {
+ case AMDGPU_VM_PDB2:
+ case AMDGPU_VM_PDB1:
+ case AMDGPU_VM_PDB0:
+ return 9 * (AMDGPU_VM_PDB0 - level) +
+ adev->vm_manager.block_size;
+ case AMDGPU_VM_PTB:
+ return 0;
+ default:
+ return ~0;
+ }
+ }
+
+ /**
+ * amdgpu_vm_pt_num_entries - return the number of entries in a PD/PT
+ *
+ * @adev: amdgpu_device pointer
+ * @level: VMPT level
+ *
+ * Returns:
+ * The number of entries in a page directory or page table.
+ */
+ static unsigned int amdgpu_vm_pt_num_entries(struct amdgpu_device *adev,
+ unsigned int level)
+ {
+ unsigned int shift;
+
+ shift = amdgpu_vm_pt_level_shift(adev, adev->vm_manager.root_level);
+ if (level == adev->vm_manager.root_level)
+ /* For the root directory */
+ return round_up(adev->vm_manager.max_pfn, 1ULL << shift)
+ >> shift;
+ else if (level != AMDGPU_VM_PTB)
+ /* Everything in between */
+ return 512;
+
+ /* For the page tables on the leaves */
+ return AMDGPU_VM_PTE_COUNT(adev);
+ }
+
+ /**
+ * amdgpu_vm_pt_num_ats_entries - return the number of ATS entries in the root PD
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Returns:
+ * The number of entries in the root page directory which needs the ATS setting.
+ */
+ static unsigned int amdgpu_vm_pt_num_ats_entries(struct amdgpu_device *adev)
+ {
+ unsigned int shift;
+
+ shift = amdgpu_vm_pt_level_shift(adev, adev->vm_manager.root_level);
+ return AMDGPU_GMC_HOLE_START >> (shift + AMDGPU_GPU_PAGE_SHIFT);
+ }
+
+ /**
+ * amdgpu_vm_pt_entries_mask - the mask to get the entry number of a PD/PT
+ *
+ * @adev: amdgpu_device pointer
+ * @level: VMPT level
+ *
+ * Returns:
+ * The mask to extract the entry number of a PD/PT from an address.
+ */
+ static uint32_t amdgpu_vm_pt_entries_mask(struct amdgpu_device *adev,
+ unsigned int level)
+ {
+ if (level <= adev->vm_manager.root_level)
+ return 0xffffffff;
+ else if (level != AMDGPU_VM_PTB)
+ return 0x1ff;
+ else
+ return AMDGPU_VM_PTE_COUNT(adev) - 1;
+ }
+
+ /**
+ * amdgpu_vm_pt_size - returns the size of the page table in bytes
+ *
+ * @adev: amdgpu_device pointer
+ * @level: VMPT level
+ *
+ * Returns:
+ * The size of the BO for a page directory or page table in bytes.
+ */
+ static unsigned int amdgpu_vm_pt_size(struct amdgpu_device *adev,
+ unsigned int level)
+ {
+ return AMDGPU_GPU_PAGE_ALIGN(amdgpu_vm_pt_num_entries(adev, level) * 8);
+ }
+
+ /**
+ * amdgpu_vm_pt_parent - get the parent page directory
+ *
+ * @pt: child page table
+ *
+ * Helper to get the parent entry for the child page table. NULL if we are at
+ * the root page directory.
+ */
+ static struct amdgpu_vm_bo_base *
+ amdgpu_vm_pt_parent(struct amdgpu_vm_bo_base *pt)
+ {
+ struct amdgpu_bo *parent = pt->bo->parent;
+
+ if (!parent)
+ return NULL;
+
+ return parent->vm_bo;
+ }
+
+ /**
+ * amdgpu_vm_pt_start - start PD/PT walk
+ *
+ * @adev: amdgpu_device pointer
+ * @vm: amdgpu_vm structure
+ * @start: start address of the walk
+ * @cursor: state to initialize
+ *
+ * Initialize a amdgpu_vm_pt_cursor to start a walk.
+ */
+ static void amdgpu_vm_pt_start(struct amdgpu_device *adev,
+ struct amdgpu_vm *vm, uint64_t start,
+ struct amdgpu_vm_pt_cursor *cursor)
+ {
+ cursor->pfn = start;
+ cursor->parent = NULL;
+ cursor->entry = &vm->root;
+ cursor->level = adev->vm_manager.root_level;
+ }
+
+ /**
+ * amdgpu_vm_pt_descendant - go to child node
+ *
+ * @adev: amdgpu_device pointer
+ * @cursor: current state
+ *
+ * Walk to the child node of the current node.
+ * Returns:
+ * True if the walk was possible, false otherwise.
+ */
+ static bool amdgpu_vm_pt_descendant(struct amdgpu_device *adev,
+ struct amdgpu_vm_pt_cursor *cursor)
+ {
+ unsigned int mask, shift, idx;
+
+ if ((cursor->level == AMDGPU_VM_PTB) || !cursor->entry ||
+ !cursor->entry->bo)
+ return false;
+
+ mask = amdgpu_vm_pt_entries_mask(adev, cursor->level);
+ shift = amdgpu_vm_pt_level_shift(adev, cursor->level);
+
+ ++cursor->level;
+ idx = (cursor->pfn >> shift) & mask;
+ cursor->parent = cursor->entry;
+ cursor->entry = &to_amdgpu_bo_vm(cursor->entry->bo)->entries[idx];
+ return true;
+ }
+
+ /**
+ * amdgpu_vm_pt_sibling - go to sibling node
+ *
+ * @adev: amdgpu_device pointer
+ * @cursor: current state
+ *
+ * Walk to the sibling node of the current node.
+ * Returns:
+ * True if the walk was possible, false otherwise.
+ */
+ static bool amdgpu_vm_pt_sibling(struct amdgpu_device *adev,
+ struct amdgpu_vm_pt_cursor *cursor)
+ {
+
+ unsigned int shift, num_entries;
+ struct amdgpu_bo_vm *parent;
+
+ /* Root doesn't have a sibling */
+ if (!cursor->parent)
+ return false;
+
+ /* Go to our parents and see if we got a sibling */
+ shift = amdgpu_vm_pt_level_shift(adev, cursor->level - 1);
+ num_entries = amdgpu_vm_pt_num_entries(adev, cursor->level - 1);
+ parent = to_amdgpu_bo_vm(cursor->parent->bo);
+
+ if (cursor->entry == &parent->entries[num_entries - 1])
+ return false;
+
+ cursor->pfn += 1ULL << shift;
+ cursor->pfn &= ~((1ULL << shift) - 1);
+ ++cursor->entry;
+ return true;
+ }
+
+ /**
+ * amdgpu_vm_pt_ancestor - go to parent node
+ *
+ * @cursor: current state
+ *
+ * Walk to the parent node of the current node.
+ * Returns:
+ * True if the walk was possible, false otherwise.
+ */
+ static bool amdgpu_vm_pt_ancestor(struct amdgpu_vm_pt_cursor *cursor)
+ {
+ if (!cursor->parent)
+ return false;
+
+ --cursor->level;
+ cursor->entry = cursor->parent;
+ cursor->parent = amdgpu_vm_pt_parent(cursor->parent);
+ return true;
+ }
+
+ /**
+ * amdgpu_vm_pt_next - get next PD/PT in hieratchy
+ *
+ * @adev: amdgpu_device pointer
+ * @cursor: current state
+ *
+ * Walk the PD/PT tree to the next node.
+ */
+ static void amdgpu_vm_pt_next(struct amdgpu_device *adev,
+ struct amdgpu_vm_pt_cursor *cursor)
+ {
+ /* First try a newborn child */
+ if (amdgpu_vm_pt_descendant(adev, cursor))
+ return;
+
+ /* If that didn't worked try to find a sibling */
+ while (!amdgpu_vm_pt_sibling(adev, cursor)) {
+ /* No sibling, go to our parents and grandparents */
+ if (!amdgpu_vm_pt_ancestor(cursor)) {
+ cursor->pfn = ~0ll;
+ return;
+ }
+ }
+ }
+
+ /**
+ * amdgpu_vm_pt_first_dfs - start a deep first search
+ *
+ * @adev: amdgpu_device structure
+ * @vm: amdgpu_vm structure
+ * @start: optional cursor to start with
+ * @cursor: state to initialize
+ *
+ * Starts a deep first traversal of the PD/PT tree.
+ */
+ static void amdgpu_vm_pt_first_dfs(struct amdgpu_device *adev,
+ struct amdgpu_vm *vm,
+ struct amdgpu_vm_pt_cursor *start,
+ struct amdgpu_vm_pt_cursor *cursor)
+ {
+ if (start)
+ *cursor = *start;
+ else
+ amdgpu_vm_pt_start(adev, vm, 0, cursor);
+
+ while (amdgpu_vm_pt_descendant(adev, cursor))
+ ;
+ }
+
+ /**
+ * amdgpu_vm_pt_continue_dfs - check if the deep first search should continue
+ *
+ * @start: starting point for the search
+ * @entry: current entry
+ *
+ * Returns:
+ * True when the search should continue, false otherwise.
+ */
+ static bool amdgpu_vm_pt_continue_dfs(struct amdgpu_vm_pt_cursor *start,
+ struct amdgpu_vm_bo_base *entry)
+ {
+ return entry && (!start || entry != start->entry);
+ }
+
+ /**
+ * amdgpu_vm_pt_next_dfs - get the next node for a deep first search
+ *
+ * @adev: amdgpu_device structure
+ * @cursor: current state
+ *
+ * Move the cursor to the next node in a deep first search.
+ */
+ static void amdgpu_vm_pt_next_dfs(struct amdgpu_device *adev,
+ struct amdgpu_vm_pt_cursor *cursor)
+ {
+ if (!cursor->entry)
+ return;
+
+ if (!cursor->parent)
+ cursor->entry = NULL;
+ else if (amdgpu_vm_pt_sibling(adev, cursor))
+ while (amdgpu_vm_pt_descendant(adev, cursor))
+ ;
+ else
+ amdgpu_vm_pt_ancestor(cursor);
+ }
+
+ /*
+ * for_each_amdgpu_vm_pt_dfs_safe - safe deep first search of all PDs/PTs
+ */
+ #define for_each_amdgpu_vm_pt_dfs_safe(adev, vm, start, cursor, entry) \
+ for (amdgpu_vm_pt_first_dfs((adev), (vm), (start), &(cursor)), \
+ (entry) = (cursor).entry, amdgpu_vm_pt_next_dfs((adev), &(cursor));\
+ amdgpu_vm_pt_continue_dfs((start), (entry)); \
+ (entry) = (cursor).entry, amdgpu_vm_pt_next_dfs((adev), &(cursor)))
+
+ /**
+ * amdgpu_vm_pt_clear - initially clear the PDs/PTs
+ *
+ * @adev: amdgpu_device pointer
+ * @vm: VM to clear BO from
+ * @vmbo: BO to clear
+ * @immediate: use an immediate update
+ *
+ * Root PD needs to be reserved when calling this.
+ *
+ * Returns:
+ * 0 on success, errno otherwise.
+ */
+ int amdgpu_vm_pt_clear(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+ struct amdgpu_bo_vm *vmbo, bool immediate)
+ {
+ unsigned int level = adev->vm_manager.root_level;
+ struct ttm_operation_ctx ctx = { true, false };
+ struct amdgpu_vm_update_params params;
+ struct amdgpu_bo *ancestor = &vmbo->bo;
+ unsigned int entries, ats_entries;
+ struct amdgpu_bo *bo = &vmbo->bo;
+ uint64_t addr;
+ int r, idx;
+
+ /* Figure out our place in the hierarchy */
+ if (ancestor->parent) {
+ ++level;
+ while (ancestor->parent->parent) {
+ ++level;
+ ancestor = ancestor->parent;
+ }
+ }
+
+ entries = amdgpu_bo_size(bo) / 8;
+ if (!vm->pte_support_ats) {
+ ats_entries = 0;
+
+ } else if (!bo->parent) {
+ ats_entries = amdgpu_vm_pt_num_ats_entries(adev);
+ ats_entries = min(ats_entries, entries);
+ entries -= ats_entries;
+
+ } else {
+ struct amdgpu_vm_bo_base *pt;
+
+ pt = ancestor->vm_bo;
+ ats_entries = amdgpu_vm_pt_num_ats_entries(adev);
+ if ((pt - to_amdgpu_bo_vm(vm->root.bo)->entries) >=
+ ats_entries) {
+ ats_entries = 0;
+ } else {
+ ats_entries = entries;
+ entries = 0;
+ }
+ }
+
+ r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+ if (r)
+ return r;
+
+ if (vmbo->shadow) {
+ struct amdgpu_bo *shadow = vmbo->shadow;
+
+ r = ttm_bo_validate(&shadow->tbo, &shadow->placement, &ctx);
+ if (r)
+ return r;
+ }
+
+ if (!drm_dev_enter(adev_to_drm(adev), &idx))
+ return -ENODEV;
+
+ r = vm->update_funcs->map_table(vmbo);
+ if (r)
+ goto exit;
+
+ memset(¶ms, 0, sizeof(params));
+ params.adev = adev;
+ params.vm = vm;
+ params.immediate = immediate;
+
+ r = vm->update_funcs->prepare(¶ms, NULL, AMDGPU_SYNC_EXPLICIT);
+ if (r)
+ goto exit;
+
+ addr = 0;
+ if (ats_entries) {
+ uint64_t value = 0, flags;
+
+ flags = AMDGPU_PTE_DEFAULT_ATC;
+ if (level != AMDGPU_VM_PTB) {
+ /* Handle leaf PDEs as PTEs */
+ flags |= AMDGPU_PDE_PTE;
+ amdgpu_gmc_get_vm_pde(adev, level, &value, &flags);
+ }
+
+ r = vm->update_funcs->update(¶ms, vmbo, addr, 0,
+ ats_entries, value, flags);
+ if (r)
+ goto exit;
+
+ addr += ats_entries * 8;
+ }
+
+ if (entries) {
+ uint64_t value = 0, flags = 0;
+
+ if (adev->asic_type >= CHIP_VEGA10) {
+ if (level != AMDGPU_VM_PTB) {
+ /* Handle leaf PDEs as PTEs */
+ flags |= AMDGPU_PDE_PTE;
+ amdgpu_gmc_get_vm_pde(adev, level,
+ &value, &flags);
+ } else {
+ /* Workaround for fault priority problem on GMC9 */
+ flags = AMDGPU_PTE_EXECUTABLE;
+ }
+ }
+
+ r = vm->update_funcs->update(¶ms, vmbo, addr, 0, entries,
+ value, flags);
+ if (r)
+ goto exit;
+ }
+
+ r = vm->update_funcs->commit(¶ms, NULL);
+ exit:
+ drm_dev_exit(idx);
+ return r;
+ }
+
+ /**
+ * amdgpu_vm_pt_create - create bo for PD/PT
+ *
+ * @adev: amdgpu_device pointer
+ * @vm: requesting vm
+ * @level: the page table level
+ * @immediate: use a immediate update
+ * @vmbo: pointer to the buffer object pointer
+ */
+ int amdgpu_vm_pt_create(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+ int level, bool immediate, struct amdgpu_bo_vm **vmbo)
+ {
+ struct amdgpu_bo_param bp;
+ struct amdgpu_bo *bo;
+ struct dma_resv *resv;
+ unsigned int num_entries;
+ int r;
+
+ memset(&bp, 0, sizeof(bp));
+
+ bp.size = amdgpu_vm_pt_size(adev, level);
+ bp.byte_align = AMDGPU_GPU_PAGE_SIZE;
+ bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
+ bp.domain = amdgpu_bo_get_preferred_domain(adev, bp.domain);
+ bp.flags = AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS |
+ AMDGPU_GEM_CREATE_CPU_GTT_USWC;
+
+ if (level < AMDGPU_VM_PTB)
+ num_entries = amdgpu_vm_pt_num_entries(adev, level);
+ else
+ num_entries = 0;
+
+ bp.bo_ptr_size = struct_size((*vmbo), entries, num_entries);
+
+ if (vm->use_cpu_for_update)
+ bp.flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
+
+ bp.type = ttm_bo_type_kernel;
+ bp.no_wait_gpu = immediate;
+ if (vm->root.bo)
+ bp.resv = vm->root.bo->tbo.base.resv;
+
+ r = amdgpu_bo_create_vm(adev, &bp, vmbo);
+ if (r)
+ return r;
+
+ bo = &(*vmbo)->bo;
+ if (vm->is_compute_context || (adev->flags & AMD_IS_APU)) {
+ (*vmbo)->shadow = NULL;
+ return 0;
+ }
+
+ if (!bp.resv)
+ WARN_ON(dma_resv_lock(bo->tbo.base.resv,
+ NULL));
+ resv = bp.resv;
+ memset(&bp, 0, sizeof(bp));
+ bp.size = amdgpu_vm_pt_size(adev, level);
+ bp.domain = AMDGPU_GEM_DOMAIN_GTT;
+ bp.flags = AMDGPU_GEM_CREATE_CPU_GTT_USWC;
+ bp.type = ttm_bo_type_kernel;
+ bp.resv = bo->tbo.base.resv;
+ bp.bo_ptr_size = sizeof(struct amdgpu_bo);
+
+ r = amdgpu_bo_create(adev, &bp, &(*vmbo)->shadow);
+
+ if (!resv)
+ dma_resv_unlock(bo->tbo.base.resv);
+
+ if (r) {
+ amdgpu_bo_unref(&bo);
+ return r;
+ }
+
+ (*vmbo)->shadow->parent = amdgpu_bo_ref(bo);
+ amdgpu_bo_add_to_shadow_list(*vmbo);
+
+ return 0;
+ }
+
+ /**
+ * amdgpu_vm_pt_alloc - Allocate a specific page table
+ *
+ * @adev: amdgpu_device pointer
+ * @vm: VM to allocate page tables for
+ * @cursor: Which page table to allocate
+ * @immediate: use an immediate update
+ *
+ * Make sure a specific page table or directory is allocated.
+ *
+ * Returns:
+ * 1 if page table needed to be allocated, 0 if page table was already
+ * allocated, negative errno if an error occurred.
+ */
+ static int amdgpu_vm_pt_alloc(struct amdgpu_device *adev,
+ struct amdgpu_vm *vm,
+ struct amdgpu_vm_pt_cursor *cursor,
+ bool immediate)
+ {
+ struct amdgpu_vm_bo_base *entry = cursor->entry;
+ struct amdgpu_bo *pt_bo;
+ struct amdgpu_bo_vm *pt;
+ int r;
+
+ if (entry->bo)
+ return 0;
+
+ r = amdgpu_vm_pt_create(adev, vm, cursor->level, immediate, &pt);
+ if (r)
+ return r;
+
+ /* Keep a reference to the root directory to avoid
+ * freeing them up in the wrong order.
+ */
+ pt_bo = &pt->bo;
+ pt_bo->parent = amdgpu_bo_ref(cursor->parent->bo);
+ amdgpu_vm_bo_base_init(entry, vm, pt_bo);
+ r = amdgpu_vm_pt_clear(adev, vm, pt, immediate);
+ if (r)
+ goto error_free_pt;
+
+ return 0;
+
+ error_free_pt:
+ amdgpu_bo_unref(&pt->shadow);
+ amdgpu_bo_unref(&pt_bo);
+ return r;
+ }
+
+ /**
+ * amdgpu_vm_pt_free - free one PD/PT
+ *
+ * @entry: PDE to free
+ */
+ static void amdgpu_vm_pt_free(struct amdgpu_vm_bo_base *entry)
+ {
+ struct amdgpu_bo *shadow;
+
+ if (!entry->bo)
+ return;
+ shadow = amdgpu_bo_shadowed(entry->bo);
+ entry->bo->vm_bo = NULL;
+ list_del(&entry->vm_status);
+ amdgpu_bo_unref(&shadow);
+ amdgpu_bo_unref(&entry->bo);
+ }
+
+ /**
+ * amdgpu_vm_pt_free_dfs - free PD/PT levels
+ *
+ * @adev: amdgpu device structure
+ * @vm: amdgpu vm structure
+ * @start: optional cursor where to start freeing PDs/PTs
+ *
+ * Free the page directory or page table level and all sub levels.
+ */
+ static void amdgpu_vm_pt_free_dfs(struct amdgpu_device *adev,
+ struct amdgpu_vm *vm,
+ struct amdgpu_vm_pt_cursor *start)
+ {
+ struct amdgpu_vm_pt_cursor cursor;
+ struct amdgpu_vm_bo_base *entry;
+
+ for_each_amdgpu_vm_pt_dfs_safe(adev, vm, start, cursor, entry)
+ amdgpu_vm_pt_free(entry);
+
+ if (start)
+ amdgpu_vm_pt_free(start->entry);
+ }
+
+ /**
+ * amdgpu_vm_pt_free_root - free root PD
+ * @adev: amdgpu device structure
+ * @vm: amdgpu vm structure
+ *
+ * Free the root page directory and everything below it.
+ */
+ void amdgpu_vm_pt_free_root(struct amdgpu_device *adev, struct amdgpu_vm *vm)
+ {
+ amdgpu_vm_pt_free_dfs(adev, vm, NULL);
+ }
+
+ /**
+ * amdgpu_vm_pt_is_root_clean - check if a root PD is clean
+ *
+ * @adev: amdgpu_device pointer
+ * @vm: the VM to check
+ *
+ * Check all entries of the root PD, if any subsequent PDs are allocated,
+ * it means there are page table creating and filling, and is no a clean
+ * VM
+ *
+ * Returns:
+ * 0 if this VM is clean
+ */
+ bool amdgpu_vm_pt_is_root_clean(struct amdgpu_device *adev,
+ struct amdgpu_vm *vm)
+ {
+ enum amdgpu_vm_level root = adev->vm_manager.root_level;
+ unsigned int entries = amdgpu_vm_pt_num_entries(adev, root);
+ unsigned int i = 0;
+
+ for (i = 0; i < entries; i++) {
+ if (to_amdgpu_bo_vm(vm->root.bo)->entries[i].bo)
+ return false;
+ }
+ return true;
+ }
+
+ /**
+ * amdgpu_vm_pde_update - update a single level in the hierarchy
+ *
+ * @params: parameters for the update
+ * @entry: entry to update
+ *
+ * Makes sure the requested entry in parent is up to date.
+ */
+ int amdgpu_vm_pde_update(struct amdgpu_vm_update_params *params,
+ struct amdgpu_vm_bo_base *entry)
+ {
+ struct amdgpu_vm_bo_base *parent = amdgpu_vm_pt_parent(entry);
+ struct amdgpu_bo *bo = parent->bo, *pbo;
+ struct amdgpu_vm *vm = params->vm;
+ uint64_t pde, pt, flags;
+ unsigned int level;
+
+ for (level = 0, pbo = bo->parent; pbo; ++level)
+ pbo = pbo->parent;
+
+ level += params->adev->vm_manager.root_level;
+ amdgpu_gmc_get_pde_for_bo(entry->bo, level, &pt, &flags);
+ pde = (entry - to_amdgpu_bo_vm(parent->bo)->entries) * 8;
+ return vm->update_funcs->update(params, to_amdgpu_bo_vm(bo), pde, pt,
+ 1, 0, flags);
+ }
+
+ /*
+ * amdgpu_vm_pte_update_flags - figure out flags for PTE updates
+ *
+ * Make sure to set the right flags for the PTEs at the desired level.
+ */
+ static void amdgpu_vm_pte_update_flags(struct amdgpu_vm_update_params *params,
+ struct amdgpu_bo_vm *pt,
+ unsigned int level,
+ uint64_t pe, uint64_t addr,
+ unsigned int count, uint32_t incr,
+ uint64_t flags)
+
+ {
+ if (level != AMDGPU_VM_PTB) {
+ flags |= AMDGPU_PDE_PTE;
+ amdgpu_gmc_get_vm_pde(params->adev, level, &addr, &flags);
+
+ } else if (params->adev->asic_type >= CHIP_VEGA10 &&
+ !(flags & AMDGPU_PTE_VALID) &&
+ !(flags & AMDGPU_PTE_PRT)) {
+
+ /* Workaround for fault priority problem on GMC9 */
+ flags |= AMDGPU_PTE_EXECUTABLE;
+ }
+
+ params->vm->update_funcs->update(params, pt, pe, addr, count, incr,
+ flags);
+ }
+
+ /**
+ * amdgpu_vm_pte_fragment - get fragment for PTEs
+ *
+ * @params: see amdgpu_vm_update_params definition
+ * @start: first PTE to handle
+ * @end: last PTE to handle
+ * @flags: hw mapping flags
+ * @frag: resulting fragment size
+ * @frag_end: end of this fragment
+ *
+ * Returns the first possible fragment for the start and end address.
+ */
+ static void amdgpu_vm_pte_fragment(struct amdgpu_vm_update_params *params,
+ uint64_t start, uint64_t end, uint64_t flags,
+ unsigned int *frag, uint64_t *frag_end)
+ {
+ /**
+ * The MC L1 TLB supports variable sized pages, based on a fragment
+ * field in the PTE. When this field is set to a non-zero value, page
+ * granularity is increased from 4KB to (1 << (12 + frag)). The PTE
+ * flags are considered valid for all PTEs within the fragment range
+ * and corresponding mappings are assumed to be physically contiguous.
+ *
+ * The L1 TLB can store a single PTE for the whole fragment,
+ * significantly increasing the space available for translation
+ * caching. This leads to large improvements in throughput when the
+ * TLB is under pressure.
+ *
+ * The L2 TLB distributes small and large fragments into two
+ * asymmetric partitions. The large fragment cache is significantly
+ * larger. Thus, we try to use large fragments wherever possible.
+ * Userspace can support this by aligning virtual base address and
+ * allocation size to the fragment size.
+ *
+ * Starting with Vega10 the fragment size only controls the L1. The L2
+ * is now directly feed with small/huge/giant pages from the walker.
+ */
+ unsigned int max_frag;
+
+ if (params->adev->asic_type < CHIP_VEGA10)
+ max_frag = params->adev->vm_manager.fragment_size;
+ else
+ max_frag = 31;
+
+ /* system pages are non continuously */
+ if (params->pages_addr) {
+ *frag = 0;
+ *frag_end = end;
+ return;
+ }
+
+ /* This intentionally wraps around if no bit is set */
+ *frag = min_t(unsigned int, ffs(start) - 1, fls64(end - start) - 1);
+ if (*frag >= max_frag) {
+ *frag = max_frag;
+ *frag_end = end & ~((1ULL << max_frag) - 1);
+ } else {
+ *frag_end = start + (1 << *frag);
+ }
+ }
+
+ /**
+ * amdgpu_vm_ptes_update - make sure that page tables are valid
+ *
+ * @params: see amdgpu_vm_update_params definition
+ * @start: start of GPU address range
+ * @end: end of GPU address range
+ * @dst: destination address to map to, the next dst inside the function
+ * @flags: mapping flags
+ *
+ * Update the page tables in the range @start - @end.
+ *
+ * Returns:
+ * 0 for success, -EINVAL for failure.
+ */
+ int amdgpu_vm_ptes_update(struct amdgpu_vm_update_params *params,
+ uint64_t start, uint64_t end,
+ uint64_t dst, uint64_t flags)
+ {
+ struct amdgpu_device *adev = params->adev;
+ struct amdgpu_vm_pt_cursor cursor;
+ uint64_t frag_start = start, frag_end;
+ unsigned int frag;
+ int r;
+
+ /* figure out the initial fragment */
+ amdgpu_vm_pte_fragment(params, frag_start, end, flags, &frag,
+ &frag_end);
+
+ /* walk over the address space and update the PTs */
+ amdgpu_vm_pt_start(adev, params->vm, start, &cursor);
+ while (cursor.pfn < end) {
+ unsigned int shift, parent_shift, mask;
+ uint64_t incr, entry_end, pe_start;
+ struct amdgpu_bo *pt;
+
+ if (!params->unlocked) {
+ /* make sure that the page tables covering the
+ * address range are actually allocated
+ */
+ r = amdgpu_vm_pt_alloc(params->adev, params->vm,
+ &cursor, params->immediate);
+ if (r)
+ return r;
+ }
+
+ shift = amdgpu_vm_pt_level_shift(adev, cursor.level);
+ parent_shift = amdgpu_vm_pt_level_shift(adev, cursor.level - 1);
+ if (params->unlocked) {
+ /* Unlocked updates are only allowed on the leaves */
+ if (amdgpu_vm_pt_descendant(adev, &cursor))
+ continue;
+ } else if (adev->asic_type < CHIP_VEGA10 &&
+ (flags & AMDGPU_PTE_VALID)) {
+ /* No huge page support before GMC v9 */
+ if (cursor.level != AMDGPU_VM_PTB) {
+ if (!amdgpu_vm_pt_descendant(adev, &cursor))
+ return -ENOENT;
+ continue;
+ }
+ } else if (frag < shift) {
+ /* We can't use this level when the fragment size is
+ * smaller than the address shift. Go to the next
+ * child entry and try again.
+ */
+ if (amdgpu_vm_pt_descendant(adev, &cursor))
+ continue;
+ } else if (frag >= parent_shift) {
+ /* If the fragment size is even larger than the parent
+ * shift we should go up one level and check it again.
+ */
+ if (!amdgpu_vm_pt_ancestor(&cursor))
+ return -EINVAL;
+ continue;
+ }
+
+ pt = cursor.entry->bo;
+ if (!pt) {
+ /* We need all PDs and PTs for mapping something, */
+ if (flags & AMDGPU_PTE_VALID)
+ return -ENOENT;
+
+ /* but unmapping something can happen at a higher
+ * level.
+ */
+ if (!amdgpu_vm_pt_ancestor(&cursor))
+ return -EINVAL;
+
+ pt = cursor.entry->bo;
+ shift = parent_shift;
+ frag_end = max(frag_end, ALIGN(frag_start + 1,
+ 1ULL << shift));
+ }
+
+ /* Looks good so far, calculate parameters for the update */
+ incr = (uint64_t)AMDGPU_GPU_PAGE_SIZE << shift;
+ mask = amdgpu_vm_pt_entries_mask(adev, cursor.level);
+ pe_start = ((cursor.pfn >> shift) & mask) * 8;
+ entry_end = ((uint64_t)mask + 1) << shift;
+ entry_end += cursor.pfn & ~(entry_end - 1);
+ entry_end = min(entry_end, end);
+
+ do {
+ struct amdgpu_vm *vm = params->vm;
+ uint64_t upd_end = min(entry_end, frag_end);
+ unsigned int nptes = (upd_end - frag_start) >> shift;
+ uint64_t upd_flags = flags | AMDGPU_PTE_FRAG(frag);
+
+ /* This can happen when we set higher level PDs to
+ * silent to stop fault floods.
+ */
+ nptes = max(nptes, 1u);
+
+ trace_amdgpu_vm_update_ptes(params, frag_start, upd_end,
+ min(nptes, 32u), dst, incr,
+ upd_flags,
+ vm->task_info.pid,
+ vm->immediate.fence_context);
+ amdgpu_vm_pte_update_flags(params, to_amdgpu_bo_vm(pt),
+ cursor.level, pe_start, dst,
+ nptes, incr, upd_flags);
+
+ pe_start += nptes * 8;
+ dst += nptes * incr;
+
+ frag_start = upd_end;
+ if (frag_start >= frag_end) {
+ /* figure out the next fragment */
+ amdgpu_vm_pte_fragment(params, frag_start, end,
+ flags, &frag, &frag_end);
+ if (frag < shift)
+ break;
+ }
+ } while (frag_start < entry_end);
+
+ if (amdgpu_vm_pt_descendant(adev, &cursor)) {
+ /* Free all child entries.
+ * Update the tables with the flags and addresses and free up subsequent
+ * tables in the case of huge pages or freed up areas.
+ * This is the maximum you can free, because all other page tables are not
+ * completely covered by the range and so potentially still in use.
+ */
+ while (cursor.pfn < frag_start) {
+ /* Make sure previous mapping is freed */
+ if (cursor.entry->bo) {
+ params->table_freed = true;
+ amdgpu_vm_pt_free_dfs(adev, params->vm,
+ &cursor);
+ }
+ amdgpu_vm_pt_next(adev, &cursor);
+ }
+
+ } else if (frag >= shift) {
+ /* or just move on to the next on the same level. */
+ amdgpu_vm_pt_next(adev, &cursor);
+ }
+ }
+
+ return 0;
+ }
if (p->unlocked) {
struct dma_fence *tmp = dma_fence_get(f);
- swap(p->vm->last_unlocked, f);
+ swap(p->vm->last_unlocked, tmp);
dma_fence_put(tmp);
} else {
amdgpu_bo_fence(p->vm->root.bo, f, true);
struct amdgpu_bo *bo = &vmbo->bo;
enum amdgpu_ib_pool_type pool = p->immediate ? AMDGPU_IB_POOL_IMMEDIATE
: AMDGPU_IB_POOL_DELAYED;
+ struct dma_resv_iter cursor;
unsigned int i, ndw, nptes;
+ struct dma_fence *fence;
uint64_t *pte;
int r;
/* Wait for PD/PT moves to be completed */
- r = amdgpu_sync_fence(&p->job->sync, bo->tbo.moving);
- if (r)
- return r;
+ dma_resv_for_each_fence(&cursor, bo->tbo.base.resv,
+ DMA_RESV_USAGE_KERNEL, fence) {
+ r = amdgpu_sync_fence(&p->job->sync, fence);
+ if (r)
+ return r;
+ }
do {
ndw = p->num_dw_left;
#include <linux/hashtable.h>
#include <linux/mmu_notifier.h>
+#include <linux/memremap.h>
#include <linux/mutex.h>
#include <linux/types.h>
#include <linux/atomic.h>
/* VM context for GPUVM allocations */
struct file *drm_file;
void *drm_priv;
+ atomic64_t tlb_seq;
/* GPUVM allocations storage */
struct idr alloc_idr;
int kfd_topology_remove_device(struct kfd_dev *gpu);
struct kfd_topology_device *kfd_topology_device_by_proximity_domain(
uint32_t proximity_domain);
+ struct kfd_topology_device *kfd_topology_device_by_proximity_domain_no_lock(
+ uint32_t proximity_domain);
struct kfd_topology_device *kfd_topology_device_by_id(uint32_t gpu_id);
struct kfd_dev *kfd_device_by_id(uint32_t gpu_id);
struct kfd_dev *kfd_device_by_pci_dev(const struct pci_dev *pdev);
goto reserve_bo_failed;
}
- r = dma_resv_reserve_shared(bo->tbo.base.resv, 1);
+ r = dma_resv_reserve_fences(bo->tbo.base.resv, 1);
if (r) {
pr_debug("failed %d to reserve bo\n", r);
amdgpu_bo_unreserve(bo);
pr_debug("[0x%llx 0x%llx]\n", start, last);
- return amdgpu_vm_bo_update_mapping(adev, adev, vm, false, true, NULL,
- start, last, init_pte_value, 0,
- NULL, NULL, fence, NULL);
+ return amdgpu_vm_update_range(adev, vm, false, true, true, NULL, start,
+ last, init_pte_value, 0, 0, NULL, NULL,
+ fence);
}
static int
{
struct amdgpu_device *adev = pdd->dev->adev;
struct amdgpu_vm *vm = drm_priv_to_vm(pdd->drm_priv);
- bool table_freed = false;
uint64_t pte_flags;
unsigned long last_start;
int last_domain;
(last_domain == SVM_RANGE_VRAM_DOMAIN) ? 1 : 0,
pte_flags);
- r = amdgpu_vm_bo_update_mapping(adev, bo_adev, vm, false, false,
- NULL, last_start,
- prange->start + i, pte_flags,
- last_start - prange->start,
- NULL, dma_addr,
- &vm->last_update,
- &table_freed);
+ r = amdgpu_vm_update_range(adev, vm, false, false, false, NULL,
+ last_start, prange->start + i,
+ pte_flags,
+ last_start - prange->start,
+ bo_adev ? bo_adev->vm_manager.vram_base_offset : 0,
+ NULL, dma_addr, &vm->last_update);
for (j = last_start - prange->start; j <= i; j++)
dma_addr[j] |= last_domain;
if (fence)
*fence = dma_fence_get(vm->last_update);
- if (table_freed)
- kfd_flush_tlb(pdd, TLB_FLUSH_LEGACY);
out:
return r;
}
break;
}
}
+
+ kfd_flush_tlb(pdd, TLB_FLUSH_LEGACY);
}
return r;
struct kfd_process *process;
struct svm_range *prange;
bool intr;
- unsigned long bitmap[MAX_GPU_INSTANCE];
+ DECLARE_BITMAP(bitmap, MAX_GPU_INSTANCE);
struct ttm_validate_buffer tv[MAX_GPU_INSTANCE];
struct list_head validate_list;
struct ww_acquire_ctx ticket;
pr_debug("kfd process not founded pasid 0x%x\n", pasid);
return 0;
}
- if (!p->xnack_enabled) {
- pr_debug("XNACK not enabled for pasid 0x%x\n", pasid);
- r = -EFAULT;
- goto out;
- }
svms = &p->svms;
pr_debug("restoring svms 0x%p fault address 0x%llx\n", svms, addr);
goto out;
}
+ if (!p->xnack_enabled) {
+ pr_debug("XNACK not enabled for pasid 0x%x\n", pasid);
+ r = -EFAULT;
+ goto out;
+ }
+
/* p->lead_thread is available as kfd_process_wq_release flush the work
* before releasing task ref.
*/
* deadlock during GPU reset when this fence will not signal
* but we hold reservation lock for the BO.
*/
- r = dma_resv_wait_timeout(abo->tbo.base.resv, true, false,
+ r = dma_resv_wait_timeout(abo->tbo.base.resv,
+ DMA_RESV_USAGE_WRITE, false,
msecs_to_jiffies(5000));
if (unlikely(r <= 0))
DRM_ERROR("Waiting for fences timed out!");
&bundle->flip_addrs[planes_count].address,
afb->tmz_surface, false);
- DRM_DEBUG_ATOMIC("plane: id=%d dcc_en=%d\n",
+ drm_dbg_state(state->dev, "plane: id=%d dcc_en=%d\n",
new_plane_state->plane->index,
bundle->plane_infos[planes_count].dcc.enable);
dc_plane,
bundle->flip_addrs[planes_count].flip_timestamp_in_us);
- DRM_DEBUG_ATOMIC("%s Flipping to hi: 0x%x, low: 0x%x\n",
+ drm_dbg_state(state->dev, "%s Flipping to hi: 0x%x, low: 0x%x\n",
__func__,
bundle->flip_addrs[planes_count].address.grph.addr.high_part,
bundle->flip_addrs[planes_count].address.grph.addr.low_part);
dm_new_crtc_state = to_dm_crtc_state(new_crtc_state);
dm_old_crtc_state = to_dm_crtc_state(old_crtc_state);
- DRM_DEBUG_ATOMIC(
+ drm_dbg_state(state->dev,
"amdgpu_crtc id:%d crtc_state_flags: enable:%d, active:%d, "
"planes_changed:%d, mode_changed:%d,active_changed:%d,"
"connectors_changed:%d\n",
if (!drm_atomic_crtc_needs_modeset(new_crtc_state))
goto skip_modeset;
- DRM_DEBUG_ATOMIC(
+ drm_dbg_state(state->dev,
"amdgpu_crtc id:%d crtc_state_flags: enable:%d, active:%d, "
"planes_changed:%d, mode_changed:%d,active_changed:%d,"
"connectors_changed:%d\n",
# define DP_PSR_IS_SUPPORTED 1
# define DP_PSR2_IS_SUPPORTED 2 /* eDP 1.4 */
# define DP_PSR2_WITH_Y_COORD_IS_SUPPORTED 3 /* eDP 1.4a */
+ # define DP_PSR2_WITH_Y_COORD_ET_SUPPORTED 4 /* eDP 1.5, adopted eDP 1.4b SCR */
#define DP_PSR_CAPS 0x071 /* XXX 1.2? */
# define DP_PSR_NO_TRAIN_ON_EXIT 1
# define DP_PSR_SETUP_TIME_SHIFT 1
# define DP_PSR2_SU_Y_COORDINATE_REQUIRED (1 << 4) /* eDP 1.4a */
# define DP_PSR2_SU_GRANULARITY_REQUIRED (1 << 5) /* eDP 1.4b */
+ # define DP_PSR2_SU_AUX_FRAME_SYNC_NOT_NEEDED (1 << 6)/* eDP 1.5, adopted eDP 1.4b SCR */
#define DP_PSR2_SU_X_GRANULARITY 0x072 /* eDP 1.4b */
#define DP_PSR2_SU_Y_GRANULARITY 0x074 /* eDP 1.4b */
bool is_remote;
};
+int drm_dp_dpcd_probe(struct drm_dp_aux *aux, unsigned int offset);
ssize_t drm_dp_dpcd_read(struct drm_dp_aux *aux, unsigned int offset,
void *buffer, size_t size);
ssize_t drm_dp_dpcd_write(struct drm_dp_aux *aux, unsigned int offset,
int drm_dp_read_sink_count(struct drm_dp_aux *aux);
int drm_dp_read_lttpr_common_caps(struct drm_dp_aux *aux,
+ const u8 dpcd[DP_RECEIVER_CAP_SIZE],
u8 caps[DP_LTTPR_COMMON_CAP_SIZE]);
int drm_dp_read_lttpr_phy_caps(struct drm_dp_aux *aux,
+ const u8 dpcd[DP_RECEIVER_CAP_SIZE],
enum drm_dp_phy dp_phy,
u8 caps[DP_LTTPR_PHY_CAP_SIZE]);
int drm_dp_lttpr_count(const u8 cap[DP_LTTPR_COMMON_CAP_SIZE]);