Am 03.04.2018 um 11:24 schrieb Chunming Zhou:
> user cases:
> 1. KFD wraps amdgpu_bo_create, they have no fallback case which is different
> with amdgpu_gem_object_create.
> since upstream branch has no amdgpu_amdkfd_gpuvm.c, which need KFD
> guys add this flag to __alloc_memory_of_gpu:
> + flags |= AMDGPU_GEM_CREATE_NO_FALLBACK;
> 2. UMD can specify this flag for their allocation as well if they like.
Please drop the second comment, we currently block the UMD from using
this flag.
>
> Change-Id: Id3ac1a0bdd48633328d0b22c4c057fab4e94d0b3
> Signed-off-by: Chunming Zhou <david1.zhou at amd.com>
> Cc: felix.kuehling at amd.com
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 3 ++-
> drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 3 ++-
> include/uapi/drm/amdgpu_drm.h | 2 ++
> 3 files changed, 6 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
> index dc34b50e6b29..d7d7ce1507ec 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
> @@ -386,7 +386,8 @@ static int amdgpu_cs_bo_validate(struct amdgpu_cs_parser *p,
> bo->tbo.mem.start < adev->gmc.visible_vram_size >> PAGE_SHIFT)
> p->bytes_moved_vis += ctx.bytes_moved;
>
> - if (unlikely(r == -ENOMEM) && domain != bo->allowed_domains) {
> + if (unlikely(r == -ENOMEM) && domain != bo->allowed_domains &&
> + !(bo->flags & AMDGPU_GEM_CREATE_NO_FALLBACK)) {
Please drop that change, we can control BO placement purely using the
preferred/allowed domains here.
Apart from that the patch looks good to me,
Christian.
> domain = bo->allowed_domains;
> goto retry;
> }
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> index f04bfab1eebd..c70d745073b8 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> @@ -420,7 +420,8 @@ static int amdgpu_bo_do_create(struct amdgpu_device *adev, unsigned long size,
> r = ttm_bo_init_reserved(&adev->mman.bdev, &bo->tbo, size, type,
> &bo->placement, page_align, &ctx, acc_size,
> NULL, resv, &amdgpu_ttm_bo_destroy);
> - if (unlikely(r && r != -ERESTARTSYS) && type == ttm_bo_type_device) {
> + if (unlikely(r && r != -ERESTARTSYS) && type == ttm_bo_type_device &&
> + !(flags & AMDGPU_GEM_CREATE_NO_FALLBACK)) {
> if (flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED) {
> flags &= ~AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
> goto retry;
> diff --git a/include/uapi/drm/amdgpu_drm.h b/include/uapi/drm/amdgpu_drm.h
> index 0087799962cf..80665715e651 100644
> --- a/include/uapi/drm/amdgpu_drm.h
> +++ b/include/uapi/drm/amdgpu_drm.h
> @@ -98,6 +98,8 @@ extern "C" {
> #define AMDGPU_GEM_CREATE_VM_ALWAYS_VALID (1 << 6)
> /* Flag that BO sharing will be explicitly synchronized */
> #define AMDGPU_GEM_CREATE_EXPLICIT_SYNC (1 << 7)
> +/* Flag that BO doesn't need fallback */
> +#define AMDGPU_GEM_CREATE_NO_FALLBACK (1 << 8)
>
> struct drm_amdgpu_gem_create_in {
> /** the requested memory size */
