Skip to contentSkip to conversation

lib/sbitmap: kill 'depth' from sbitmap_word

PATCH Applied Ming Lei 4 participants 5 messages Jan 10, 2022 – Feb 8, 2022

Subsystems (2)
Block layer Library code
lore
Share & tools
lore

Ming Lei

author
#0 lore raw

PATCH v2 1/1 lib/sbitmap: kill 'depth' from sbitmap_word

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 <[email protected]>
Signed-off-by Ming Lei <[email protected]>

---
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(-)
include/linux/sbitmap.h +10 -7
index fc0357a6e19b..3754dc45f890 100644
@@ -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)
+		return sb->depth - (index << sb->shift);
+	return 1U << sb->shift;
+}
+
 /**
  * sbitmap_free() - Free memory used by a &struct sbitmap.
  * @sb: Bitmap to free.

@@ -251,7 +254,7 @@ static inline void __sbitmap_for_each_set(struct sbitmap *sb,
 	while (scanned < sb->depth) {
 		unsigned long word;
 		unsigned int depth = min_t(unsigned int,
-					   sb->map[index].depth - nr,
+					   __map_depth(sb, index) - nr,
 					   sb->depth - scanned);
 
 		scanned += depth;
lib/sbitmap.c +14 -20
index 2709ab825499..e9a51337a2a3 100644
@@ -85,7 +85,6 @@ int sbitmap_init_node(struct sbitmap *sb, unsigned int depth, int shift,
 		      bool alloc_hint)
 {
 	unsigned int bits_per_word;
-	unsigned int i;
 
 	if (shift < 0)
 		shift = sbitmap_calculate_shift(depth);

@@ -117,10 +116,6 @@ int sbitmap_init_node(struct sbitmap *sb, unsigned int depth, int shift,
 		return -ENOMEM;
 	}
 
-	for (i = 0; i < sb->map_nr; i++) {
-		sb->map[i].depth = min(depth, bits_per_word);
-		depth -= sb->map[i].depth;
-	}
 	return 0;
 }
 EXPORT_SYMBOL_GPL(sbitmap_init_node);

@@ -135,11 +130,6 @@ void sbitmap_resize(struct sbitmap *sb, unsigned int depth)
 
 	sb->depth = depth;
 	sb->map_nr = DIV_ROUND_UP(sb->depth, bits_per_word);
-
-	for (i = 0; i < sb->map_nr; i++) {
-		sb->map[i].depth = min(depth, bits_per_word);
-		depth -= sb->map[i].depth;
-	}
 }
 EXPORT_SYMBOL_GPL(sbitmap_resize);
 

@@ -184,8 +174,8 @@ static int sbitmap_find_bit_in_index(struct sbitmap *sb, int index,
 	int nr;
 
 	do {
-		nr = __sbitmap_get_word(&map->word, map->depth, alloc_hint,
-					!sb->round_robin);
+		nr = __sbitmap_get_word(&map->word, __map_depth(sb, index),
+					alloc_hint, !sb->round_robin);
 		if (nr != -1)
 			break;
 		if (!sbitmap_deferred_clear(map))

@@ -257,7 +247,9 @@ static int __sbitmap_get_shallow(struct sbitmap *sb,
 	for (i = 0; i < sb->map_nr; i++) {
 again:
 		nr = __sbitmap_get_word(&sb->map[index].word,
-					min(sb->map[index].depth, shallow_depth),
+					min_t(unsigned int,
+					      __map_depth(sb, index),
+					      shallow_depth),
 					SB_NR_TO_BIT(sb, alloc_hint), true);
 		if (nr != -1) {
 			nr += index << sb->shift;

@@ -315,11 +307,12 @@ static unsigned int __sbitmap_weight(const struct sbitmap *sb, bool set)
 
 	for (i = 0; i < sb->map_nr; i++) {
 		const struct sbitmap_word *word = &sb->map[i];
+		unsigned int word_depth = __map_depth(sb, i);
 
 		if (set)
-			weight += bitmap_weight(&word->word, word->depth);
+			weight += bitmap_weight(&word->word, word_depth);
 		else
-			weight += bitmap_weight(&word->cleared, word->depth);
+			weight += bitmap_weight(&word->cleared, word_depth);
 	}
 	return weight;
 }

@@ -367,7 +360,7 @@ void sbitmap_bitmap_show(struct sbitmap *sb, struct seq_file *m)
 	for (i = 0; i < sb->map_nr; i++) {
 		unsigned long word = READ_ONCE(sb->map[i].word);
 		unsigned long cleared = READ_ONCE(sb->map[i].cleared);
-		unsigned int word_bits = READ_ONCE(sb->map[i].depth);
+		unsigned int word_bits = __map_depth(sb, i);
 
 		word &= ~cleared;
 

@@ -508,15 +501,16 @@ unsigned long __sbitmap_queue_get_batch(struct sbitmap_queue *sbq, int nr_tags,
 	for (i = 0; i < sb->map_nr; i++) {
 		struct sbitmap_word *map = &sb->map[index];
 		unsigned long get_mask;
+		unsigned int map_depth = __map_depth(sb, index);
 
 		sbitmap_deferred_clear(map);
-		if (map->word == (1UL << (map->depth - 1)) - 1)
+		if (map->word == (1UL << (map_depth - 1)) - 1)
 			continue;
 
-		nr = find_first_zero_bit(&map->word, map->depth);
-		if (nr + nr_tags <= map->depth) {
+		nr = find_first_zero_bit(&map->word, map_depth);
+		if (nr + nr_tags <= map_depth) {
 			atomic_long_t *ptr = (atomic_long_t *) &map->word;
-			int map_tags = min_t(int, nr_tags, map->depth);
+			int map_tags = min_t(int, nr_tags, map_depth);
 			unsigned long val, ret;
 
 			get_mask = ((1UL << map_tags) - 1) << nr;
Signature · 1 line
2.31.1
author
#1 lore raw
Quoted 8 lines · Ming Lei

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.

Quoted 7 lines

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 <[email protected]>
Signed-off-by Ming Lei <[email protected]>

Hello Guys,

Ping...

Thanks,
Ming

Reviewed-by on v2 1/1
#2 lore raw
In reply to Ming Lei
Quoted 11 lines · Ming Lei

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.

Quoted 8 lines

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 <[email protected]>
Signed-off-by Ming Lei <[email protected]>
Reviewed-by Martin Wilck <[email protected]>
Reviewed-by on v2 1/1
#3 lore raw
In reply to Ming Lei
Quoted 8 lines · Ming Lei

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.

Quoted 7 lines

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 <[email protected]>
Signed-off-by Ming Lei <[email protected]>
Reviewed-by John Garry <[email protected]>
Quoted 31 lines

---
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(-)
include/linux/sbitmap.h +1 -6
index fc0357a6e19b..3754dc45f890 100644
@@ -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

include/linux/sbitmap.h struct sbitmap_queue { · line 165

Quoted 7 lines
include/linux/sbitmap.h +4 -0
@@ -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?

Quoted 4 lines
include/linux/sbitmap.h +4 -0
+		return sb->depth - (index << sb->shift);
+	return 1U << sb->shift;
+}
+
maintainer Applied
#4 lore raw
In reply to Ming Lei
Quoted 9 lines · Ming Lei

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.

[...]

Applied, thanks!

[1/1] lib/sbitmap: kill 'depth' from sbitmap_word

      commit: 3301bc53358a0eb0a0db65fd7a513cd4cb50c83a

Best regards,

Signature · 1 line
Jens Axboe