Thank you for your contribution! Sashiko AI review found 1 potential issue to consider: - [Medium] gpu/buddy: u64 underflow on inverted range in __alloc_range_bias() --
commit 84539d9e7a89604b1f5d66dfb390e667e4fccb69 Author: William Palacek <[email protected]> gpu/buddy: find range-biased blocks in the free trees This commit replaces the root-to-leaf block tree walk in __alloc_range_bias() with augmented RB-tree lookups in the per-order free trees, adding a subtree_has_dirty augment to skip all-clear subtrees on dirty requests. > diff --git a/drivers/gpu/buddy.c b/drivers/gpu/buddy.c > index e7d2212bcec95..baf17700ffba5 100644 > --- a/drivers/gpu/buddy.c > +++ b/drivers/gpu/buddy.c [ ... ] > @@ -1247,96 +1262,155 @@ static bool block_incompatible(struct > gpu_buddy_block *block, unsigned int flags > gpu_buddy_block_is_clear(block); > } > > -static struct gpu_buddy_block * > -__alloc_range_bias(struct gpu_buddy *mm, > - u64 start, u64 end, > - unsigned int order, > - unsigned long flags, > - bool fallback) [ ... ] > +static struct gpu_buddy_block * > +__alloc_range_bias(struct gpu_buddy *mm, > + u64 start, u64 end, > + unsigned int order, > + unsigned long flags, > + bool fallback) > +{ > + u64 req_size = mm->chunk_size << order; > + struct gpu_buddy_block *block = NULL; > + u64 best = 0; > + unsigned int i; > + int err; > > + for (i = order; i <= mm->max_order; i++) { > + struct rb_root *root = &mm->free_tree[i]; > + u64 size = mm->chunk_size << i; > + struct gpu_buddy_block *found; > + u64 offset = 0; > > + if (!fallback) { > + if (end - start < size) > + break; > > + found = rbtree_last_range_block(root, end - size + 1, > + flags, false); > + if (!found || gpu_buddy_block_offset(found) < start) > + continue; [Severity: Medium] Can end - start and end - size + 1 underflow when start > end, causing __alloc_range_bias() to allocate a block outside [start, end)? gpu_buddy_alloc_blocks() checks end <= mm->size and range_overflows(start, size, mm->size), which only checks start + size <= mm->size, so an inverted range (for example, when AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED clamps lpfn to visible_pfn below a memory partition's fpfn in amdgpu_bo_placement_from_domain()) reaches __alloc_range_bias() with start > end. Previously, the adjusted_end vs adjusted_start check rejected inverted ranges with -ENOSPC. In the new strict pass, end - start wraps in u64 arithmetic past the size check, and when size > end, end - size + 1 wraps to near U64_MAX, so rbtree_last_range_block() matches and returns a block at offset >= start > end. -- Sashiko AI review ยท https://sashiko.dev/#/patchset/17f2a6366ba845edc799dbeb712c12f8d060afad.1794845083.git.william.pala...@amd.com?part=1
