Re: [PATCH v2] gpu/buddy: add range-restricted contiguous allocation fallback

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] gpu/buddy: range-restricted alloc fails when free space exceeds range_end
--

commit 2ed4c9d1a70c51390968b65e939095ce52c8ee07
Author: Arunpravin Paneer Selvam <Arunpravin.PaneerSelvam@amd.com>

gpu/buddy: add range-restricted contiguous allocation fallback

A range + contiguous allocation rounds its size up to a power of two
and requires a naturally aligned free block of that size. The
non-range contiguous path already recovers from fragmentation via
__alloc_contig_try_harder(), which is now given a range window to
support the range + contiguous case.

> diff --git a/drivers/gpu/buddy.c b/drivers/gpu/buddy.c
> index 2f2aaadafe351..5265e1f6a3183 100644
> --- a/drivers/gpu/buddy.c
> +++ b/drivers/gpu/buddy.c
[ ... ]
> @@ -1734,20 +1722,24 @@ static int __alloc_contig_try_harder(struct gpu_buddy *mm,
[ ... ]
>  		if (!err) {
>  			/* Allocate the unaligned RHS offset using round_down */
>  			gpu_buddy_free_list_internal(mm, blocks);
> -			err = __alloc_contig_aligned_retry(mm, rhs_offset,
> -							   size,
> -							   min_block_size,
> -							   flags, blocks);
> -			if (!err)
> -				return 0;
> -			if (err != -ENOSPC) {
> -				gpu_buddy_free_list_internal(mm, blocks);
> -				return err;
> +
> +			aligned = round_down(rhs_offset, min_block_size);
> +			if (aligned >= range_start &&
> +			    aligned + size <= range_end) {
> +				err = __gpu_buddy_alloc_range(mm, aligned, size,
> +							      flags, NULL, blocks);
> +				if (!err)
> +					return 0;
> +				if (err != -ENOSPC) {
> +					gpu_buddy_free_list_internal(mm, blocks);
> +					return err;
> +				}
>  			}
>  			goto next;
>  		}

[Severity: High]
Does this skip the left-hand side (LHS) fallback when contiguous free space
exceeds range_end?

If __gpu_buddy_alloc_range() succeeds at finding free space starting at
rhs_offset, err is 0. If the free space extends beyond range_end, the newly
added check (aligned >= range_start && aligned + size <= range_end) might
fail. But since err is still 0, it hits the unconditional goto next.

By skipping to next, doesn't this bypass the LHS window-sliding logic below?
If there was valid free space to the left of rhs_offset that perfectly fits
the window, it appears it would be ignored, causing the allocator to return
-ENOSPC despite sufficient space existing within the requested range.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260930062117.585089-1-arunpravin.paneerselvam@amd.com?part=1