During eviction we do want to trigger the OOM killer.
Only while doing new allocations we should try to avoid that and return -ENOMEM to the application.
v2: rename the flag to gfp_retry_mayfail.
Signed-off-by: Christian König christian.koenig@amd.com Reviewed-by: Daniel Vetter daniel.vetter@ffwll.ch --- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 2 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 3 --- drivers/gpu/drm/ttm/ttm_pool.c | 2 +- drivers/gpu/drm/ttm/ttm_tt.c | 7 ------- include/drm/ttm/ttm_bo_api.h | 2 ++ include/drm/ttm/ttm_bo_driver.h | 3 --- 6 files changed, 5 insertions(+), 14 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index 1aa516429c80..4e9dfbea31c6 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -516,6 +516,8 @@ static int amdgpu_bo_do_create(struct amdgpu_device *adev, struct ttm_operation_ctx ctx = { .interruptible = (bp->type != ttm_bo_type_kernel), .no_wait_gpu = bp->no_wait_gpu, + /* We opt to avoid OOM on system pages allocations */ + .gfp_retry_mayfail = true, .resv = bp->resv, .flags = bp->type != ttm_bo_type_kernel ? TTM_OPT_FLAG_ALLOW_RES_EVICT : 0 diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c index bd6e6641c3fc..c01c060e4ac5 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -1914,9 +1914,6 @@ int amdgpu_ttm_init(struct amdgpu_device *adev) } adev->mman.initialized = true;
- /* We opt to avoid OOM on system pages allocations */ - adev->mman.bdev.no_retry = true; - /* Initialize VRAM pool with all of VRAM divided into pages */ r = amdgpu_vram_mgr_init(adev); if (r) { diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c index 1e50deefb5d5..44ec41aa78d6 100644 --- a/drivers/gpu/drm/ttm/ttm_pool.c +++ b/drivers/gpu/drm/ttm/ttm_pool.c @@ -367,7 +367,7 @@ int ttm_pool_alloc(struct ttm_pool *pool, struct ttm_tt *tt, if (tt->page_flags & TTM_PAGE_FLAG_ZERO_ALLOC) gfp_flags |= __GFP_ZERO;
- if (tt->page_flags & TTM_PAGE_FLAG_NO_RETRY) + if (ctx->gfp_retry_mayfail) gfp_flags |= __GFP_RETRY_MAYFAIL;
if (pool->use_dma32) diff --git a/drivers/gpu/drm/ttm/ttm_tt.c b/drivers/gpu/drm/ttm/ttm_tt.c index 8861a74ac335..cfd633c7e764 100644 --- a/drivers/gpu/drm/ttm/ttm_tt.c +++ b/drivers/gpu/drm/ttm/ttm_tt.c @@ -51,9 +51,6 @@ int ttm_tt_create(struct ttm_buffer_object *bo, bool zero_alloc) if (bo->ttm) return 0;
- if (bdev->no_retry) - page_flags |= TTM_PAGE_FLAG_NO_RETRY; - switch (bo->type) { case ttm_bo_type_device: if (zero_alloc) @@ -211,8 +208,6 @@ int ttm_tt_swapin(struct ttm_tt *ttm)
swap_space = swap_storage->f_mapping; gfp_mask = mapping_gfp_mask(swap_space); - if (ttm->page_flags & TTM_PAGE_FLAG_NO_RETRY) - gfp_mask |= __GFP_RETRY_MAYFAIL;
for (i = 0; i < ttm->num_pages; ++i) { from_page = shmem_read_mapping_page_gfp(swap_space, i, @@ -260,8 +255,6 @@ int ttm_tt_swapout(struct ttm_bo_device *bdev, struct ttm_tt *ttm)
swap_space = swap_storage->f_mapping; gfp_mask = mapping_gfp_mask(swap_space); - if (ttm->page_flags & TTM_PAGE_FLAG_NO_RETRY) - gfp_mask |= __GFP_RETRY_MAYFAIL;
for (i = 0; i < ttm->num_pages; ++i) { from_page = ttm->pages[i]; diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h index 37102e45e496..4637357ba856 100644 --- a/include/drm/ttm/ttm_bo_api.h +++ b/include/drm/ttm/ttm_bo_api.h @@ -195,6 +195,7 @@ struct ttm_bo_kmap_obj { * * @interruptible: Sleep interruptible if sleeping. * @no_wait_gpu: Return immediately if the GPU is busy. + * @gfp_retry_mayfail: Set the __GFP_RETRY_MAYFAIL when allocation pages. * @resv: Reservation object to allow reserved evictions with. * @flags: Including the following flags * @@ -204,6 +205,7 @@ struct ttm_bo_kmap_obj { struct ttm_operation_ctx { bool interruptible; bool no_wait_gpu; + bool gfp_retry_mayfail; struct dma_resv *resv; uint64_t bytes_moved; uint32_t flags; diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h index e9f683fa72dc..da8208f43378 100644 --- a/include/drm/ttm/ttm_bo_driver.h +++ b/include/drm/ttm/ttm_bo_driver.h @@ -276,7 +276,6 @@ extern struct ttm_bo_global { * @dev_mapping: A pointer to the struct address_space representing the * device address space. * @wq: Work queue structure for the delayed delete workqueue. - * @no_retry: Don't retry allocation if it fails * */
@@ -314,8 +313,6 @@ struct ttm_bo_device { */
struct delayed_work wq; - - bool no_retry; };
static inline struct ttm_resource_manager *ttm_manager_type(struct ttm_bo_device *bdev,
The ttm_operation_ctx structure has a mixture of flags and bools. Drop the flags and replace them with bools as well.
v2: fix typos, improve comments
Signed-off-by: Christian König christian.koenig@amd.com Reviewed-by: Daniel Vetter daniel.vetter@ffwll.ch --- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 3 +-- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 5 ++--- drivers/gpu/drm/ttm/ttm_bo.c | 2 +- drivers/gpu/drm/ttm/ttm_bo_vm.c | 3 +-- drivers/gpu/drm/ttm/ttm_memory.c | 3 ++- drivers/gpu/drm/ttm/ttm_resource.c | 2 +- include/drm/ttm/ttm_bo_api.h | 13 ++++++------- 7 files changed, 14 insertions(+), 17 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index d50b63a93d37..8466558d0d93 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -404,8 +404,7 @@ static int amdgpu_cs_bo_validate(struct amdgpu_cs_parser *p, struct ttm_operation_ctx ctx = { .interruptible = true, .no_wait_gpu = false, - .resv = bo->tbo.base.resv, - .flags = 0 + .resv = bo->tbo.base.resv }; uint32_t domain; int r; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index 4e9dfbea31c6..e1f64ef8c765 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -518,9 +518,8 @@ static int amdgpu_bo_do_create(struct amdgpu_device *adev, .no_wait_gpu = bp->no_wait_gpu, /* We opt to avoid OOM on system pages allocations */ .gfp_retry_mayfail = true, - .resv = bp->resv, - .flags = bp->type != ttm_bo_type_kernel ? - TTM_OPT_FLAG_ALLOW_RES_EVICT : 0 + .allow_res_evict = bp->type != ttm_bo_type_kernel, + .resv = bp->resv }; struct amdgpu_bo *bo; unsigned long page_align, size = bp->size; diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index c63b7ea1cd5d..e2a124b3affb 100644 --- a/drivers/gpu/drm/ttm/ttm_bo.c +++ b/drivers/gpu/drm/ttm/ttm_bo.c @@ -637,7 +637,7 @@ static bool ttm_bo_evict_swapout_allowable(struct ttm_buffer_object *bo,
if (bo->base.resv == ctx->resv) { dma_resv_assert_held(bo->base.resv); - if (ctx->flags & TTM_OPT_FLAG_ALLOW_RES_EVICT) + if (ctx->allow_res_evict) ret = true; *locked = false; if (busy) diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c b/drivers/gpu/drm/ttm/ttm_bo_vm.c index eeaca5d1efe3..2944fa0af493 100644 --- a/drivers/gpu/drm/ttm/ttm_bo_vm.c +++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c @@ -315,8 +315,7 @@ vm_fault_t ttm_bo_vm_fault_reserved(struct vm_fault *vmf, struct ttm_operation_ctx ctx = { .interruptible = false, .no_wait_gpu = false, - .flags = TTM_OPT_FLAG_FORCE_ALLOC - + .force_alloc = true };
ttm = bo->ttm; diff --git a/drivers/gpu/drm/ttm/ttm_memory.c b/drivers/gpu/drm/ttm/ttm_memory.c index f9a90bfaa3c1..5ed1fc8f2ace 100644 --- a/drivers/gpu/drm/ttm/ttm_memory.c +++ b/drivers/gpu/drm/ttm/ttm_memory.c @@ -542,7 +542,8 @@ ttm_check_under_lowerlimit(struct ttm_mem_global *glob, { int64_t available;
- if (ctx->flags & TTM_OPT_FLAG_FORCE_ALLOC) + /* We allow over commit during suspend */ + if (ctx->force_alloc) return false;
available = get_nr_swap_pages() + si_mem_available(); diff --git a/drivers/gpu/drm/ttm/ttm_resource.c b/drivers/gpu/drm/ttm/ttm_resource.c index 4ebc043e2867..b60699bf4816 100644 --- a/drivers/gpu/drm/ttm/ttm_resource.c +++ b/drivers/gpu/drm/ttm/ttm_resource.c @@ -89,7 +89,7 @@ int ttm_resource_manager_evict_all(struct ttm_bo_device *bdev, struct ttm_operation_ctx ctx = { .interruptible = false, .no_wait_gpu = false, - .flags = TTM_OPT_FLAG_FORCE_ALLOC + .force_alloc = true }; struct ttm_bo_global *glob = &ttm_bo_glob; struct dma_fence *fence; diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h index 4637357ba856..5ddad88ae6ed 100644 --- a/include/drm/ttm/ttm_bo_api.h +++ b/include/drm/ttm/ttm_bo_api.h @@ -196,8 +196,11 @@ struct ttm_bo_kmap_obj { * @interruptible: Sleep interruptible if sleeping. * @no_wait_gpu: Return immediately if the GPU is busy. * @gfp_retry_mayfail: Set the __GFP_RETRY_MAYFAIL when allocation pages. + * @allow_res_evict: Allow eviction of reserved BOs. Can be used when multiple + * BOs share the same reservation object. + * @force_alloc: Don't check the memory account during suspend or CPU page + * faults. Should only be used by TTM internally. * @resv: Reservation object to allow reserved evictions with. - * @flags: Including the following flags * * Context for TTM operations like changing buffer placement or general memory * allocation. @@ -206,16 +209,12 @@ struct ttm_operation_ctx { bool interruptible; bool no_wait_gpu; bool gfp_retry_mayfail; + bool allow_res_evict; + bool force_alloc; struct dma_resv *resv; uint64_t bytes_moved; - uint32_t flags; };
-/* Allow eviction of reserved BOs */ -#define TTM_OPT_FLAG_ALLOW_RES_EVICT 0x1 -/* when serving page fault or suspend, allow alloc anyway */ -#define TTM_OPT_FLAG_FORCE_ALLOC 0x2 - /** * ttm_bo_get - reference a struct ttm_buffer_object *
dri-devel@lists.freedesktop.org
-
Christian König