diff --git a/block/bio.c b/block/bio.c
index 265bff6b549a..1319dd2aa0b8 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;
@@ -246,12 +251,40 @@ static void bio_free(struct bio *bio)
 void bio_init(struct bio *bio, struct bio_vec *table,
 	      unsigned short max_vecs)
 {
-	memset(bio, 0, sizeof(*bio));
+	bio->bi_next = NULL;
+	bio->bi_bdev = NULL;
+	bio->bi_opf = 0;
+	bio->bi_flags = 0;
+	bio->bi_ioprio = 0;
+	bio->bi_write_hint = 0;
+	bio->bi_status = 0;
+	bio->bi_iter.bi_sector = 0;
+	bio->bi_iter.bi_size = 0;
+	bio->bi_iter.bi_idx = 0;
+	bio->bi_iter.bi_bvec_done = 0;
+	bio->bi_end_io = NULL;
+	bio->bi_private = NULL;
+#ifdef CONFIG_BLK_CGROUP
+	bio->bi_blkg = NULL;
+	bio->bi_issue.value = 0;
+#ifdef CONFIG_BLK_CGROUP_IOCOST
+	bio->bi_iocost_cost = 0;
+#endif
+#endif
+#ifdef CONFIG_BLK_INLINE_ENCRYPTION
+	bio->bi_crypt_context = NULL;
+#endif
+#ifdef CONFIG_BLK_DEV_INTEGRITY
+	bio->bi_integrity = NULL;
+#endif
+	bio->bi_vcnt = 0;
+
 	atomic_set(&bio->__bi_remaining, 1);
 	atomic_set(&bio->__bi_cnt, 1);
 
-	bio->bi_io_vec = table;
 	bio->bi_max_vecs = max_vecs;
+	bio->bi_io_vec = table;
+	bio->bi_pool = NULL;
 }
 EXPORT_SYMBOL(bio_init);
 
@@ -586,6 +619,53 @@ 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
@@ -596,16 +676,23 @@ 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);
@@ -1496,6 +1583,7 @@ 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;
@@ -1557,12 +1645,18 @@ 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:
@@ -1589,6 +1683,46 @@ 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
+ * @nr_iovecs:	number of iovecs to pre-allocate
+ * @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. On return, the
+ *    bio is marked BIO_PERCPU_CACHEABLE, and the final put of the bio
+ *    MUST be done from process context, not hard/soft IRQ.
+ *
+ */
+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;
@@ -1603,6 +1737,9 @@ 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/block/blk-core.c b/block/blk-core.c
index b5098739f72a..5454db2fa263 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -821,7 +821,7 @@ static noinline_for_stack bool submit_bio_checks(struct bio *bio)
 	}
 
 	if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
-		bio->bi_opf &= ~REQ_HIPRI;
+		bio_clear_hipri(bio);
 
 	switch (bio_op(bio)) {
 	case REQ_OP_DISCARD:
diff --git a/block/blk-merge.c b/block/blk-merge.c
index eeba8422ae82..7a5c81c02c80 100644
--- a/block/blk-merge.c
+++ b/block/blk-merge.c
@@ -285,7 +285,7 @@ split:
 	 * iopoll in direct IO routine. Given performance gain of iopoll for
 	 * big IO can be trival, disable iopoll when split needed.
 	 */
-	bio->bi_opf &= ~REQ_HIPRI;
+	bio_clear_hipri(bio);
 
 	return bio_split(bio, sectors, GFP_NOIO, bs);
 }
diff --git a/block/blk.h b/block/blk.h
index 346d184c82b0..8c96b0c90c48 100644
--- a/block/blk.h
+++ b/block/blk.h
@@ -366,4 +366,11 @@ extern struct device_attribute dev_attr_events;
 extern struct device_attribute dev_attr_events_async;
 extern struct device_attribute dev_attr_events_poll_msecs;
 
+static inline void bio_clear_hipri(struct bio *bio)
+{
+	/* can't support alloc cache if we turn off polling */
+	bio_clear_flag(bio, BIO_PERCPU_CACHE);
+	bio->bi_opf &= ~REQ_HIPRI;
+}
+
 #endif /* BLK_INTERNAL_H */
diff --git a/fs/block_dev.c b/fs/block_dev.c
index 1f21ac984253..45df6cbccf12 100644
--- a/fs/block_dev.c
+++ b/fs/block_dev.c
@@ -386,7 +386,7 @@ static ssize_t __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
 	    (bdev_logical_block_size(bdev) - 1))
 		return -EINVAL;
 
-	bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
+	bio = bio_alloc_kiocb(iocb, nr_pages, &blkdev_dio_pool);
 
 	dio = container_of(bio, struct blkdev_dio, bio);
 	dio->is_sync = is_sync = is_sync_kiocb(iocb);
@@ -514,7 +514,9 @@ blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 
 static __init int blkdev_init(void)
 {
-	return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
+	return bioset_init(&blkdev_dio_pool, 4,
+				offsetof(struct blkdev_dio, bio),
+				BIOSET_NEED_BVECS|BIOSET_PERCPU_CACHE);
 }
 module_init(blkdev_init);
 
diff --git a/fs/io_uring.c b/fs/io_uring.c
index 7cc458e0b636..73928d957691 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -2835,7 +2835,7 @@ static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 		    !kiocb->ki_filp->f_op->iopoll)
 			return -EOPNOTSUPP;
 
-		kiocb->ki_flags |= IOCB_HIPRI;
+		kiocb->ki_flags |= IOCB_HIPRI | IOCB_ALLOC_CACHE;
 		kiocb->ki_complete = io_complete_rw_iopoll;
 		req->iopoll_completed = 0;
 	} else {
diff --git a/include/linux/bio.h b/include/linux/bio.h
index 7b5f65a81f2b..3d67d0fbc868 100644
--- a/include/linux/bio.h
+++ b/include/linux/bio.h
@@ -400,6 +400,7 @@ 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 *);
@@ -408,6 +409,8 @@ 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 *);
 
@@ -657,6 +660,11 @@ 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)
@@ -673,6 +681,11 @@ 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 1335efa8a1db..9e392daa1d7f 100644
--- a/include/linux/blk_types.h
+++ b/include/linux/blk_types.h
@@ -297,6 +297,7 @@ 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 6ac543d33b66..95f88edc8f09 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,
diff --git a/include/linux/fs.h b/include/linux/fs.h
index 7eae53fff7c5..ae29027d2ac8 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -319,6 +319,8 @@ enum rw_hint {
 /* iocb->ki_waitq is valid */
 #define IOCB_WAITQ		(1 << 19)
 #define IOCB_NOIO		(1 << 20)
+/* can use bio alloc cache */
+#define IOCB_ALLOC_CACHE	(1 << 21)
 
 struct kiocb {
 	struct file		*ki_filp;