On 10/01/2022 07:29, Ming Lei wrote:
Only the last sbitmap_word can have different depth, and all the others must have same depth of 1U << sb->shift, so not necessary to store it in sbitmap_word, and it can be retrieved easily and efficiently by adding one internal helper of __map_depth(sb, index). Remove 'depth' field from sbitmap_word, then the annotation of ____cacheline_aligned_in_smp for 'word' isn't needed any more. Not see performance effect when running high parallel IOPS test on null_blk. This way saves us one cacheline(usually 64 words) per each sbitmap_word. Cc: Martin Wilck <martin.wilck@xxxxxxxx> Signed-off-by: Ming Lei <ming.lei@xxxxxxxxxx>
Reviewed-by: John Garry <john.garry@xxxxxxxxxx>
--- V2: - remove the annotation of ____cacheline_aligned_in_smp for 'word' as suggested by Jens include/linux/sbitmap.h | 17 ++++++++++------- lib/sbitmap.c | 34 ++++++++++++++-------------------- 2 files changed, 24 insertions(+), 27 deletions(-) diff --git a/include/linux/sbitmap.h b/include/linux/sbitmap.h index fc0357a6e19b..3754dc45f890 100644 --- a/include/linux/sbitmap.h +++ b/include/linux/sbitmap.h @@ -27,15 +27,10 @@ struct seq_file; * struct sbitmap_word - Word in a &struct sbitmap. */ struct sbitmap_word { - /** - * @depth: Number of bits being used in @word/@cleared - */ - unsigned long depth; - /** * @word: word holding free bits */ - unsigned long word ____cacheline_aligned_in_smp; + unsigned long word;/*** @cleared: word holding cleared bits @@ -164,6 +159,14 @@ struct sbitmap_queue { int sbitmap_init_node(struct sbitmap *sb, unsigned int depth, int shift, gfp_t flags, int node, bool round_robin, bool alloc_hint);+/* sbitmap internal helper */+static inline unsigned int __map_depth(const struct sbitmap *sb, int index) +{ + if (index == sb->map_nr - 1)
Do you think that unlikely may be useful?
+ return sb->depth - (index << sb->shift); + return 1U << sb->shift; +} +
