Re: [PATCH 6/7] btrfs: Introduce fast path for checking if a block group is done

[Sun Yangkai <sunk67188@gmail.com>]

在 2026/1/13 00:17, Martin Raiber 写道:
> A block group cannot switch away from BTRFS_CACHE_FINISHED
> once it enters that state. Therefore we can introduce
> a fast path that checks for the likely case
> that the block group is already cached, avoiding
> a full memory barrier in the likely fast path.
> 
> Signed-off-by: Martin Raiber <martin@urbackup.org>
> ---
>  fs/btrfs/block-group.h | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/fs/btrfs/block-group.h b/fs/btrfs/block-group.h
> index cf877747fd56..73bdf7091d49 100644
> --- a/fs/btrfs/block-group.h
> +++ b/fs/btrfs/block-group.h
> @@ -380,6 +380,9 @@ static inline u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
>  
>  static inline int btrfs_block_group_done(const struct btrfs_block_group *cache)
>  {
> +	if (likely(cache->cached == BTRFS_CACHE_FINISHED))
> +		return 1;
> +
>  	smp_mb();
>  	return cache->cached == BTRFS_CACHE_FINISHED ||
>  		cache->cached == BTRFS_CACHE_ERROR;

This function returns bool so maybe we can also fix its return type here.

Thanks,
Sun YangKai