drm/buddy: fix range bias

commit f41900e upstream.

There is a corner case here where start/end is after/before the block
range we are currently checking. If so we need to be sure that splitting
the block will eventually give use the block size we need. To do that we
should adjust the block range to account for the start/end, and only
continue with the split if the size/alignment will fit the requested
size. Not doing so can result in leaving split blocks unmerged when it
eventually fails.

Fixes: afea229 ("drm: improve drm_buddy_alloc function")
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Arunpravin Paneer Selvam <Arunpravin.PaneerSelvam@amd.com>
Cc: Christian König <christian.koenig@amd.com>
Cc: <stable@vger.kernel.org> # v5.18+
Reviewed-by: Arunpravin Paneer Selvam <Arunpravin.PaneerSelvam@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240219121851.25774-4-matthew.auld@intel.com
Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

drivers/gpu/drm/drm_buddy.c

@@ -332,6 +332,7 @@ alloc_range_bias(struct drm_buddy *mm,
 		 u64 start, u64 end,
 		 unsigned int order)
 {
+	u64 req_size = mm->chunk_size << order;
 	struct drm_buddy_block *block;
 	struct drm_buddy_block *buddy;
 	LIST_HEAD(dfs);
@@ -367,6 +368,15 @@ alloc_range_bias(struct drm_buddy *mm,
 		if (drm_buddy_block_is_allocated(block))
 			continue;
 
+		if (block_start < start || block_end > end) {
+			u64 adjusted_start = max(block_start, start);
+			u64 adjusted_end = min(block_end, end);
+
+			if (round_down(adjusted_end + 1, req_size) <=
+			    round_up(adjusted_start, req_size))
+				continue;
+		}
+
 		if (contains(start, end, block_start, block_end) &&
 		    order == drm_buddy_block_order(block)) {
 			/*