It is to optimize memory mapping latency, and to aviod a page fault in a corner case of changing valid PDE into PTE. Signed-off-by: Eric Huang <jinhuieric.huang@xxxxxxx> --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 2 +- .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 19 ++++++++------ drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 25 +++++++++++-------- drivers/gpu/drm/amd/amdkfd/kfd_process.c | 3 ++- 4 files changed, 28 insertions(+), 21 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h index 2560977760b3..8f2d6711e12f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h @@ -280,7 +280,7 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu( struct kgd_dev *kgd, struct kgd_mem *mem, void *drm_priv, uint64_t *size); int amdgpu_amdkfd_gpuvm_map_memory_to_gpu( - struct kgd_dev *kgd, struct kgd_mem *mem, void *drm_priv); + struct kgd_dev *kgd, struct kgd_mem *mem, void *drm_priv, bool *flush_tlb); int amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu( struct kgd_dev *kgd, struct kgd_mem *mem, void *drm_priv); int amdgpu_amdkfd_gpuvm_sync_memory( diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c index 1fcfa172911a..585b50b6009f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c @@ -1117,7 +1117,8 @@ static void unmap_bo_from_gpuvm(struct kgd_mem *mem, static int update_gpuvm_pte(struct kgd_mem *mem, struct kfd_mem_attachment *entry, - struct amdgpu_sync *sync) + struct amdgpu_sync *sync, + bool *flush_tlb) { struct amdgpu_bo_va *bo_va = entry->bo_va; struct amdgpu_device *adev = entry->adev; @@ -1128,7 +1129,7 @@ static int update_gpuvm_pte(struct kgd_mem *mem, return ret; /* Update the page tables */ - ret = amdgpu_vm_bo_update(adev, bo_va, false); + ret = amdgpu_vm_bo_update(adev, bo_va, false, flush_tlb); if (ret) { pr_err("amdgpu_vm_bo_update failed\n"); return ret; @@ -1140,7 +1141,8 @@ static int update_gpuvm_pte(struct kgd_mem *mem, static int map_bo_to_gpuvm(struct kgd_mem *mem, struct kfd_mem_attachment *entry, struct amdgpu_sync *sync, - bool no_update_pte) + bool no_update_pte, + bool *flush_tlb) { int ret; @@ -1157,7 +1159,7 @@ static int map_bo_to_gpuvm(struct kgd_mem *mem, if (no_update_pte) return 0; - ret = update_gpuvm_pte(mem, entry, sync); + ret = update_gpuvm_pte(mem, entry, sync, flush_tlb); if (ret) { pr_err("update_gpuvm_pte() failed\n"); goto update_gpuvm_pte_failed; @@ -1687,7 +1689,8 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu( } int amdgpu_amdkfd_gpuvm_map_memory_to_gpu( - struct kgd_dev *kgd, struct kgd_mem *mem, void *drm_priv) + struct kgd_dev *kgd, struct kgd_mem *mem, + void *drm_priv, bool *flush_tlb) { struct amdgpu_device *adev = get_amdgpu_device(kgd); struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv); @@ -1775,7 +1778,7 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu( entry->va, entry->va + bo_size, entry); ret = map_bo_to_gpuvm(mem, entry, ctx.sync, - is_invalid_userptr); + is_invalid_userptr, flush_tlb); if (ret) { pr_err("Failed to map bo to gpuvm\n"); goto out_unreserve; @@ -2469,7 +2472,7 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info) continue; kfd_mem_dmaunmap_attachment(mem, attachment); - ret = update_gpuvm_pte(mem, attachment, &sync); + ret = update_gpuvm_pte(mem, attachment, &sync, NULL); if (ret) { pr_err("%s: update PTE failed\n", __func__); /* make sure this gets validated again */ @@ -2675,7 +2678,7 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence **ef) continue; kfd_mem_dmaunmap_attachment(mem, attachment); - ret = update_gpuvm_pte(mem, attachment, &sync_obj); + ret = update_gpuvm_pte(mem, attachment, &sync_obj, NULL); if (ret) { pr_debug("Memory eviction: update PTE failed. Try again\n"); goto validate_map_fail; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c index 98f1d2b586c5..441d9ad955d7 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c @@ -1574,6 +1574,7 @@ static int kfd_ioctl_map_memory_to_gpu(struct file *filep, long err = 0; int i; uint32_t *devices_arr = NULL; + bool *flush_tlb = false; trace_kfd_map_memory_to_gpu_start(p); dev = kfd_device_by_id(GET_GPU_ID(args->handle)); @@ -1637,7 +1638,8 @@ static int kfd_ioctl_map_memory_to_gpu(struct file *filep, goto map_memory_to_gpu_failed; err = amdgpu_amdkfd_gpuvm_map_memory_to_gpu( - peer->kgd, (struct kgd_mem *)mem, peer_pdd->drm_priv); + peer->kgd, (struct kgd_mem *)mem, + peer_pdd->drm_priv, flush_tlb); if (err) { pr_err("Failed to map to gpu %d/%d\n", i, args->n_devices); @@ -1658,19 +1660,20 @@ static int kfd_ioctl_map_memory_to_gpu(struct file *filep, } /* Flush TLBs after waiting for the page table updates to complete */ - for (i = 0; i < args->n_devices; i++) { - peer = kfd_device_by_id(devices_arr[i]); - if (WARN_ON_ONCE(!peer)) - continue; - peer_pdd = kfd_get_process_device_data(peer, p); - if (WARN_ON_ONCE(!peer_pdd)) + if (flush_tlb) { + for (i = 0; i < args->n_devices; i++) { + peer = kfd_device_by_id(devices_arr[i]); + if (WARN_ON_ONCE(!peer)) continue; - if (!amdgpu_read_lock(peer->ddev, true)) { - kfd_flush_tlb(peer_pdd, TLB_FLUSH_LEGACY); - amdgpu_read_unlock(peer->ddev); + peer_pdd = kfd_get_process_device_data(peer, p); + if (WARN_ON_ONCE(!peer_pdd)) + continue; + if (!amdgpu_read_lock(peer->ddev, true)) { + kfd_flush_tlb(peer_pdd, TLB_FLUSH_LEGACY); + amdgpu_read_unlock(peer->ddev); + } } } - kfree(devices_arr); trace_kfd_map_memory_to_gpu_end(p, diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c index 72741f6579d3..9708214116dc 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c @@ -689,7 +689,8 @@ static int kfd_process_alloc_gpuvm(struct kfd_process_device *pdd, if (err) goto err_alloc_mem; - err = amdgpu_amdkfd_gpuvm_map_memory_to_gpu(kdev->kgd, mem, pdd->drm_priv); + err = amdgpu_amdkfd_gpuvm_map_memory_to_gpu(kdev->kgd, mem, + pdd->drm_priv, NULL); if (err) goto err_map_mem; -- 2.25.1 _______________________________________________ amd-gfx mailing list amd-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/amd-gfx