diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 65 |
1 files changed, 48 insertions, 17 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c index 2d6b2d77b738..71acd577803e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c @@ -207,7 +207,7 @@ static void unreserve_mem_limit(struct amdgpu_device *adev, spin_unlock(&kfd_mem_limit.mem_limit_lock); } -void amdgpu_amdkfd_unreserve_memory_limit(struct amdgpu_bo *bo) +void amdgpu_amdkfd_release_notify(struct amdgpu_bo *bo) { struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); u32 domain = bo->preferred_domains; @@ -219,6 +219,8 @@ void amdgpu_amdkfd_unreserve_memory_limit(struct amdgpu_bo *bo) } unreserve_mem_limit(adev, amdgpu_bo_size(bo), domain, sg); + + kfree(bo->kfd_bo); } @@ -563,6 +565,7 @@ kfd_mem_dmaunmap_userptr(struct kgd_mem *mem, dma_unmap_sgtable(adev->dev, ttm->sg, direction, 0); sg_free_table(ttm->sg); + kfree(ttm->sg); ttm->sg = NULL; } @@ -733,14 +736,19 @@ static int kfd_mem_attach(struct amdgpu_device *adev, struct kgd_mem *mem, } /* Add BO to VM internal data structures */ + ret = amdgpu_bo_reserve(bo[i], false); + if (ret) { + pr_debug("Unable to reserve BO during memory attach"); + goto unwind; + } attachment[i]->bo_va = amdgpu_vm_bo_add(adev, vm, bo[i]); + amdgpu_bo_unreserve(bo[i]); if (unlikely(!attachment[i]->bo_va)) { ret = -ENOMEM; pr_err("Failed to add BO object to VM. ret == %d\n", ret); goto unwind; } - attachment[i]->va = va; attachment[i]->pte_flags = get_pte_flags(adev, mem); attachment[i]->adev = adev; @@ -756,7 +764,9 @@ unwind: if (!attachment[i]) continue; if (attachment[i]->bo_va) { + amdgpu_bo_reserve(bo[i], true); amdgpu_vm_bo_rmv(adev, attachment[i]->bo_va); + amdgpu_bo_unreserve(bo[i]); list_del(&attachment[i]->list); } if (bo[i]) @@ -1502,7 +1512,7 @@ allocate_init_user_pages_failed: remove_kgd_mem_from_kfd_bo_list(*mem, avm->process_info); drm_vma_node_revoke(&gobj->vma_node, drm_priv); err_node_allow: - amdgpu_bo_unref(&bo); + drm_gem_object_put(gobj); /* Don't unreserve system mem limit twice */ goto err_reserve_limit; err_bo_create: @@ -1567,12 +1577,12 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu( pr_debug("Release VA 0x%llx - 0x%llx\n", mem->va, mem->va + bo_size * (1 + mem->aql_queue)); - ret = unreserve_bo_and_vms(&ctx, false, false); - /* Remove from VM internal data structures */ list_for_each_entry_safe(entry, tmp, &mem->attachments, list) kfd_mem_detach(entry); + ret = unreserve_bo_and_vms(&ctx, false, false); + /* Free the sync object */ amdgpu_sync_free(&mem->sync); @@ -1599,9 +1609,13 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu( drm_vma_node_revoke(&mem->bo->tbo.base.vma_node, drm_priv); if (mem->dmabuf) dma_buf_put(mem->dmabuf); - drm_gem_object_put(&mem->bo->tbo.base); mutex_destroy(&mem->lock); - kfree(mem); + + /* If this releases the last reference, it will end up calling + * amdgpu_amdkfd_release_notify and kfree the mem struct. That's why + * this needs to be the last call here. + */ + drm_gem_object_put(&mem->bo->tbo.base); return ret; } @@ -1870,6 +1884,16 @@ bo_reserve_failed: return ret; } +void amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel(struct kgd_dev *kgd, struct kgd_mem *mem) +{ + struct amdgpu_bo *bo = mem->bo; + + amdgpu_bo_reserve(bo, true); + amdgpu_bo_kunmap(bo); + amdgpu_bo_unpin(bo); + amdgpu_bo_unreserve(bo); +} + int amdgpu_amdkfd_gpuvm_get_vm_fault_info(struct kgd_dev *kgd, struct kfd_vm_fault_info *mem) { @@ -2040,19 +2064,26 @@ static int update_invalid_user_pages(struct amdkfd_process_info *process_info, /* Get updated user pages */ ret = amdgpu_ttm_tt_get_user_pages(bo, bo->tbo.ttm->pages); if (ret) { - pr_debug("%s: Failed to get user pages: %d\n", - __func__, ret); + pr_debug("Failed %d to get user pages\n", ret); + + /* Return -EFAULT bad address error as success. It will + * fail later with a VM fault if the GPU tries to access + * it. Better than hanging indefinitely with stalled + * user mode queues. + * + * Return other error -EBUSY or -ENOMEM to retry restore + */ + if (ret != -EFAULT) + return ret; + } else { - /* Return error -EBUSY or -ENOMEM, retry restore */ - return ret; + /* + * FIXME: Cannot ignore the return code, must hold + * notifier_lock + */ + amdgpu_ttm_tt_get_user_pages_done(bo->tbo.ttm); } - /* - * FIXME: Cannot ignore the return code, must hold - * notifier_lock - */ - amdgpu_ttm_tt_get_user_pages_done(bo->tbo.ttm); - /* Mark the BO as valid unless it was invalidated * again concurrently. */ |