bio: add allocation cache abstraction

Add a per-cpu bio_set cache for bio allocations, enabling us to quickly
recycle them instead of going through the slab allocator. This cache
isn't IRQ safe, and hence is only really suitable for polled IO.

Very simple - keeps a count of bio's in the cache, and maintains a max
of 512 with a slack of 64. If we get above max + slack, we drop slack
number of bio's.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

4 files changed, 135 insertions, 14 deletions

diff --git a/block/bio.c b/block/bio.c
index 6fa5c653283b..dbb0bc8e1ef7 100644
--- a/block/bio.c
+++ b/block/bio.c
@@ -25,6 +25,11 @@
 #include "blk.h"
 #include "blk-rq-qos.h"
 
+struct bio_alloc_cache {
+	struct bio_list		free_list;
+	unsigned int		nr;
+};
+
 static struct biovec_slab {
 	int nr_vecs;
 	char *name;
@@ -619,6 +624,53 @@ void guard_bio_eod(struct bio *bio)
 	bio_truncate(bio, maxsector << 9);
 }
 
+#define ALLOC_CACHE_MAX		512
+#define ALLOC_CACHE_SLACK	 64
+
+static void bio_alloc_cache_prune(struct bio_alloc_cache *cache,
+				  unsigned int nr)
+{
+	unsigned int i = 0;
+	struct bio *bio;
+
+	while ((bio = bio_list_pop(&cache->free_list)) != NULL) {
+		cache->nr--;
+		bio_free(bio);
+		if (++i == nr)
+			break;
+	}
+}
+
+static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node)
+{
+	struct bio_set *bs;
+
+	bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead);
+	if (bs->cache) {
+		struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu);
+
+		bio_alloc_cache_prune(cache, -1U);
+	}
+	return 0;
+}
+
+static void bio_alloc_cache_destroy(struct bio_set *bs)
+{
+	int cpu;
+
+	if (!bs->cache)
+		return;
+
+	cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
+	for_each_possible_cpu(cpu) {
+		struct bio_alloc_cache *cache;
+
+		cache = per_cpu_ptr(bs->cache, cpu);
+		bio_alloc_cache_prune(cache, -1U);
+	}
+	free_percpu(bs->cache);
+}
+
 /**
  * bio_put - release a reference to a bio
  * @bio:   bio to release reference to
@@ -629,16 +681,23 @@ void guard_bio_eod(struct bio *bio)
  **/
 void bio_put(struct bio *bio)
 {
-	if (!bio_flagged(bio, BIO_REFFED))
-		bio_free(bio);
-	else {
+	if (unlikely(bio_flagged(bio, BIO_REFFED))) {
 		BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
+		if (!atomic_dec_and_test(&bio->__bi_cnt))
+			return;
+	}
 
-		/*
-		 * last put frees it
-		 */
-		if (atomic_dec_and_test(&bio->__bi_cnt))
-			bio_free(bio);
+	if (bio_flagged(bio, BIO_PERCPU_CACHE)) {
+		struct bio_alloc_cache *cache;
+
+		bio_uninit(bio);
+		cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu());
+		bio_list_add_head(&cache->free_list, bio);
+		if (++cache->nr > ALLOC_CACHE_MAX + ALLOC_CACHE_SLACK)
+			bio_alloc_cache_prune(cache, ALLOC_CACHE_SLACK);
+		put_cpu();
+	} else {
+		bio_free(bio);
 	}
 }
 EXPORT_SYMBOL(bio_put);
@@ -1530,6 +1589,7 @@ int biovec_init_pool(mempool_t *pool, int pool_entries)
  */
 void bioset_exit(struct bio_set *bs)
 {
+	bio_alloc_cache_destroy(bs);
 	if (bs->rescue_workqueue)
 		destroy_workqueue(bs->rescue_workqueue);
 	bs->rescue_workqueue = NULL;
@@ -1591,12 +1651,18 @@ int bioset_init(struct bio_set *bs,
 	    biovec_init_pool(&bs->bvec_pool, pool_size))
 		goto bad;
 
-	if (!(flags & BIOSET_NEED_RESCUER))
-		return 0;
-
-	bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
-	if (!bs->rescue_workqueue)
-		goto bad;
+	if (flags & BIOSET_NEED_RESCUER) {
+		bs->rescue_workqueue = alloc_workqueue("bioset",
+							WQ_MEM_RECLAIM, 0);
+		if (!bs->rescue_workqueue)
+			goto bad;
+	}
+	if (flags & BIOSET_PERCPU_CACHE) {
+		bs->cache = alloc_percpu(struct bio_alloc_cache);
+		if (!bs->cache)
+			goto bad;
+		cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
+	}
 
 	return 0;
 bad:
@@ -1623,6 +1689,43 @@ int bioset_init_from_src(struct bio_set *bs, struct bio_set *src)
 }
 EXPORT_SYMBOL(bioset_init_from_src);
 
+/**
+ * bio_alloc_kiocb - Allocate a bio from bio_set based on kiocb
+ * @kiocb:	kiocb describing the IO
+ * @bs:		bio_set to allocate from
+ *
+ * Description:
+ *    Like @bio_alloc_bioset, but pass in the kiocb. The kiocb is only
+ *    used to check if we should dip into the per-cpu bio_set allocation
+ *    cache. The allocation uses GFP_KERNEL internally.
+ *
+ */
+struct bio *bio_alloc_kiocb(struct kiocb *kiocb, unsigned short nr_vecs,
+			    struct bio_set *bs)
+{
+	struct bio_alloc_cache *cache;
+	struct bio *bio;
+
+	if (!(kiocb->ki_flags & IOCB_ALLOC_CACHE) || nr_vecs > BIO_INLINE_VECS)
+		return bio_alloc_bioset(GFP_KERNEL, nr_vecs, bs);
+
+	cache = per_cpu_ptr(bs->cache, get_cpu());
+	bio = bio_list_pop(&cache->free_list);
+	if (bio) {
+		cache->nr--;
+		put_cpu();
+		bio_init(bio, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs);
+		bio->bi_pool = bs;
+		bio_set_flag(bio, BIO_PERCPU_CACHE);
+		return bio;
+	}
+	put_cpu();
+	bio = bio_alloc_bioset(GFP_KERNEL, nr_vecs, bs);
+	bio_set_flag(bio, BIO_PERCPU_CACHE);
+	return bio;
+}
+EXPORT_SYMBOL_GPL(bio_alloc_kiocb);
+
 static int __init init_bio(void)
 {
 	int i;
@@ -1637,6 +1740,9 @@ static int __init init_bio(void)
 				SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
 	}
 
+	cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL,
+					bio_cpu_dead);
+
 	if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS))
 		panic("bio: can't allocate bios\n");
 
diff --git a/include/linux/bio.h b/include/linux/bio.h
index 2203b686e1f0..89ad28213b1d 100644
--- a/include/linux/bio.h
+++ b/include/linux/bio.h
@@ -401,6 +401,7 @@ static inline struct bio *bio_next_split(struct bio *bio, int sectors,
 enum {
 	BIOSET_NEED_BVECS = BIT(0),
 	BIOSET_NEED_RESCUER = BIT(1),
+	BIOSET_PERCPU_CACHE = BIT(2),
 };
 extern int bioset_init(struct bio_set *, unsigned int, unsigned int, int flags);
 extern void bioset_exit(struct bio_set *);
@@ -409,6 +410,8 @@ extern int bioset_init_from_src(struct bio_set *bs, struct bio_set *src);
 
 struct bio *bio_alloc_bioset(gfp_t gfp, unsigned short nr_iovecs,
 		struct bio_set *bs);
+struct bio *bio_alloc_kiocb(struct kiocb *kiocb, unsigned short nr_vecs,
+		struct bio_set *bs);
 struct bio *bio_kmalloc(gfp_t gfp_mask, unsigned short nr_iovecs);
 extern void bio_put(struct bio *);
 
@@ -699,6 +702,11 @@ struct bio_set {
 	struct kmem_cache *bio_slab;
 	unsigned int front_pad;
 
+	/*
+	 * per-cpu bio alloc cache
+	 */
+	struct bio_alloc_cache __percpu *cache;
+
 	mempool_t bio_pool;
 	mempool_t bvec_pool;
 #if defined(CONFIG_BLK_DEV_INTEGRITY)
@@ -715,6 +723,11 @@ struct bio_set {
 	struct bio_list		rescue_list;
 	struct work_struct	rescue_work;
 	struct workqueue_struct	*rescue_workqueue;
+
+	/*
+	 * Hot un-plug notifier for the per-cpu cache, if used
+	 */
+	struct hlist_node cpuhp_dead;
 };
 
 static inline bool bioset_initialized(struct bio_set *bs)
diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
index 290f9061b29a..f68d4e8c775e 100644
--- a/include/linux/blk_types.h
+++ b/include/linux/blk_types.h
@@ -301,6 +301,7 @@ enum {
 	BIO_TRACKED,		/* set if bio goes through the rq_qos path */
 	BIO_REMAPPED,
 	BIO_ZONE_WRITE_LOCKED,	/* Owns a zoned device zone write lock */
+	BIO_PERCPU_CACHE,	/* can participate in per-cpu alloc cache */
 	BIO_FLAG_LAST
 };
 
diff --git a/include/linux/cpuhotplug.h b/include/linux/cpuhotplug.h
index f39b34b13871..fe72c8d6c980 100644
--- a/include/linux/cpuhotplug.h
+++ b/include/linux/cpuhotplug.h
@@ -46,6 +46,7 @@ enum cpuhp_state {
 	CPUHP_ARM_OMAP_WAKE_DEAD,
 	CPUHP_IRQ_POLL_DEAD,
 	CPUHP_BLOCK_SOFTIRQ_DEAD,
+	CPUHP_BIO_DEAD,
 	CPUHP_ACPI_CPUDRV_DEAD,
 	CPUHP_S390_PFAULT_DEAD,
 	CPUHP_BLK_MQ_DEAD,