drm/ttm: Fix memory type compatibility check

Also check the busy placements before deciding to move a buffer object.
Failing to do this may result in a completely unneccessary move within a
single memory type.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
Cc: stable@vger.kernel.org

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 6c1a38f..8d5a646 100644 (file)
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -998,24 +998,32 @@ out_unlock:
        return ret;
 }
 
-static int ttm_bo_mem_compat(struct ttm_placement *placement,
-                            struct ttm_mem_reg *mem)
+static bool ttm_bo_mem_compat(struct ttm_placement *placement,
+                             struct ttm_mem_reg *mem,
+                             uint32_t *new_flags)
 {
        int i;
 
        if (mem->mm_node && placement->lpfn != 0 &&
            (mem->start < placement->fpfn ||
             mem->start + mem->num_pages > placement->lpfn))
-               return -1;
+               return false;
 
        for (i = 0; i < placement->num_placement; i++) {
-               if ((placement->placement[i] & mem->placement &
-                       TTM_PL_MASK_CACHING) &&
-                       (placement->placement[i] & mem->placement &
-                       TTM_PL_MASK_MEM))
-                       return i;
+               *new_flags = placement->placement[i];
+               if ((*new_flags & mem->placement & TTM_PL_MASK_CACHING) &&
+                   (*new_flags & mem->placement & TTM_PL_MASK_MEM))
+                       return true;
+       }
+
+       for (i = 0; i < placement->num_busy_placement; i++) {
+               *new_flags = placement->busy_placement[i];
+               if ((*new_flags & mem->placement & TTM_PL_MASK_CACHING) &&
+                   (*new_flags & mem->placement & TTM_PL_MASK_MEM))
+                       return true;
        }
-       return -1;
+
+       return false;
 }
 
 int ttm_bo_validate(struct ttm_buffer_object *bo,
@@ -1024,6 +1032,7 @@ int ttm_bo_validate(struct ttm_buffer_object *bo,
                        bool no_wait_gpu)
 {
        int ret;
+       uint32_t new_flags;
 
        lockdep_assert_held(&bo->resv->lock.base);
        /* Check that range is valid */
@@ -1034,8 +1043,7 @@ int ttm_bo_validate(struct ttm_buffer_object *bo,
        /*
         * Check whether we need to move buffer.
         */
-       ret = ttm_bo_mem_compat(placement, &bo->mem);
-       if (ret < 0) {
+       if (!ttm_bo_mem_compat(placement, &bo->mem, &new_flags)) {
                ret = ttm_bo_move_buffer(bo, placement, interruptible,
                                         no_wait_gpu);
                if (ret)
@@ -1045,7 +1053,7 @@ int ttm_bo_validate(struct ttm_buffer_object *bo,
                 * Use the access and other non-mapping-related flag bits from
                 * the compatible memory placement flags to the active flags
                 */
-               ttm_flag_masked(&bo->mem.placement, placement->placement[ret],
+               ttm_flag_masked(&bo->mem.placement, new_flags,
                                ~TTM_PL_MASK_MEMTYPE);
        }
        /*