drm/amdgpu: user BO priority instead of self coding it (v2)

Keeping groups of BOs on the LRU is to time consuming on command submission.

Instead use the newly added BO priority to give a certain eviction order.

v2: agd: trivial warning fix

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Roger.He <Hongbo.He@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 7aecf26..c43e8c5 100644 (file)
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -386,6 +386,9 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev,
        if (unlikely(r != 0))
                return r;
 
+       bo->tbo.priority = ilog2(bo->tbo.num_pages);
+       bo->tbo.priority = min(bo->tbo.priority, (unsigned)(TTM_MAX_BO_PRIORITY - 1));
+
        if (flags & AMDGPU_GEM_CREATE_VRAM_CLEARED &&
            bo->tbo.mem.placement & TTM_PL_FLAG_VRAM) {
                struct dma_fence *fence;

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index d534450..6b1f85b 100644 (file)
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1048,56 +1048,6 @@ uint32_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct ttm_tt *ttm,
        return flags;
 }
 
-static void amdgpu_ttm_lru_removal(struct ttm_buffer_object *tbo)
-{
-       struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
-       unsigned i, j;
-
-       for (i = 0; i < AMDGPU_TTM_LRU_SIZE; ++i) {
-               struct amdgpu_mman_lru *lru = &adev->mman.log2_size[i];
-
-               for (j = 0; j < TTM_NUM_MEM_TYPES; ++j)
-                       if (&tbo->lru == lru->lru[j])
-                               lru->lru[j] = tbo->lru.prev;
-
-               if (&tbo->swap == lru->swap_lru)
-                       lru->swap_lru = tbo->swap.prev;
-       }
-}
-
-static struct amdgpu_mman_lru *amdgpu_ttm_lru(struct ttm_buffer_object *tbo)
-{
-       struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
-       unsigned log2_size = min(ilog2(tbo->num_pages),
-                                AMDGPU_TTM_LRU_SIZE - 1);
-
-       return &adev->mman.log2_size[log2_size];
-}
-
-static struct list_head *amdgpu_ttm_lru_tail(struct ttm_buffer_object *tbo)
-{
-       struct amdgpu_mman_lru *lru = amdgpu_ttm_lru(tbo);
-       struct list_head *res = lru->lru[tbo->mem.mem_type];
-
-       lru->lru[tbo->mem.mem_type] = &tbo->lru;
-       while ((++lru)->lru[tbo->mem.mem_type] == res)
-               lru->lru[tbo->mem.mem_type] = &tbo->lru;
-
-       return res;
-}
-
-static struct list_head *amdgpu_ttm_swap_lru_tail(struct ttm_buffer_object *tbo)
-{
-       struct amdgpu_mman_lru *lru = amdgpu_ttm_lru(tbo);
-       struct list_head *res = lru->swap_lru;
-
-       lru->swap_lru = &tbo->swap;
-       while ((++lru)->swap_lru == res)
-               lru->swap_lru = &tbo->swap;
-
-       return res;
-}
-
 static bool amdgpu_ttm_bo_eviction_valuable(struct ttm_buffer_object *bo,
                                            const struct ttm_place *place)
 {
@@ -1136,14 +1086,12 @@ static struct ttm_bo_driver amdgpu_bo_driver = {
        .fault_reserve_notify = &amdgpu_bo_fault_reserve_notify,
        .io_mem_reserve = &amdgpu_ttm_io_mem_reserve,
        .io_mem_free = &amdgpu_ttm_io_mem_free,
-       .lru_removal = &amdgpu_ttm_lru_removal,
-       .lru_tail = &amdgpu_ttm_lru_tail,
-       .swap_lru_tail = &amdgpu_ttm_swap_lru_tail,
+       .lru_tail = &ttm_bo_default_lru_tail,
+       .swap_lru_tail = &ttm_bo_default_swap_lru_tail,
 };
 
 int amdgpu_ttm_init(struct amdgpu_device *adev)
 {
-       unsigned i, j;
        int r;
 
        r = amdgpu_ttm_global_init(adev);
@@ -1161,19 +1109,6 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
                DRM_ERROR("failed initializing buffer object driver(%d).\n", r);
                return r;
        }
-
-       for (i = 0; i < AMDGPU_TTM_LRU_SIZE; ++i) {
-               struct amdgpu_mman_lru *lru = &adev->mman.log2_size[i];
-
-               for (j = 0; j < TTM_NUM_MEM_TYPES; ++j)
-                       lru->lru[j] = &adev->mman.bdev.man[j].lru[0];
-               lru->swap_lru = &adev->mman.bdev.glob->swap_lru[0];
-       }
-
-       for (j = 0; j < TTM_NUM_MEM_TYPES; ++j)
-               adev->mman.guard.lru[j] = NULL;
-       adev->mman.guard.swap_lru = NULL;
-
        adev->mman.initialized = true;
        r = ttm_bo_init_mm(&adev->mman.bdev, TTM_PL_VRAM,
                                adev->mc.real_vram_size >> PAGE_SHIFT);

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
index 98ee384..6bdede8 100644 (file)
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
@@ -34,13 +34,6 @@
 #define AMDGPU_PL_FLAG_GWS             (TTM_PL_FLAG_PRIV << 1)
 #define AMDGPU_PL_FLAG_OA              (TTM_PL_FLAG_PRIV << 2)
 
-#define AMDGPU_TTM_LRU_SIZE    20
-
-struct amdgpu_mman_lru {
-       struct list_head                *lru[TTM_NUM_MEM_TYPES];
-       struct list_head                *swap_lru;
-};
-
 struct amdgpu_mman {
        struct ttm_bo_global_ref        bo_global_ref;
        struct drm_global_reference     mem_global_ref;
@@ -58,11 +51,6 @@ struct amdgpu_mman {
        struct amdgpu_ring                      *buffer_funcs_ring;
        /* Scheduler entity for buffer moves */
        struct amd_sched_entity                 entity;
-
-       /* custom LRU management */
-       struct amdgpu_mman_lru                  log2_size[AMDGPU_TTM_LRU_SIZE];
-       /* guard for log2_size array, don't add anything in between */
-       struct amdgpu_mman_lru                  guard;
 };
 
 extern const struct ttm_mem_type_manager_func amdgpu_gtt_mgr_func;