Per cpu bio cache was only used in the io_uring + raw block device,
after commit 12e4e8c7ab59 ("io_uring/rw: enable bio caches for IRQ
rw"),  bio_put is safe for task and irq context, bio_alloc_bioset is
safe for task context and no one calls in irq context, so we can enable
per cpu bio cache by default.

Benchmarked with t/io_uring and ext4+nvme:
taskset -c 6 /root/fio/t/io_uring  -p0 -d128 -b4096 -s1 -c1 -F1 -B1 -R1
-X1 -n1 -P1  /mnt/testfile
base IOPS is 562K, patch IOPS is 574K. The CPU usage of bio_alloc_bioset
decrease from 1.42% to 1.22%.

The worst case is allocate bio in CPU A but free in CPU B, still use
t/io_uring and ext4+nvme:
base IOPS is 648K, patch IOPS is 647K.

Also use fio test ext4/xfs with libaio/sync/io_uring on null_blk and
nvme, no obvious performance regression.

Signed-off-by: Fengnan Chang <changfengnan@bytedance.com>
---
 block/bio.c        | 26 ++++++++++++--------------
 block/blk-map.c    |  4 ++++
 block/fops.c       |  4 ----
 include/linux/fs.h |  3 ---
 io_uring/rw.c      |  1 -
 5 files changed, 16 insertions(+), 22 deletions(-)

diff --git a/block/bio.c b/block/bio.c
index 3b371a5da159..16b20c10cab7 100644
--- a/block/bio.c
+++ b/block/bio.c
@@ -513,20 +513,18 @@ struct bio *bio_alloc_bioset(struct block_device *bdev, unsigned short nr_vecs,
 	if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_vecs > 0))
 		return NULL;
 
-	if (opf & REQ_ALLOC_CACHE) {
-		if (bs->cache && nr_vecs <= BIO_INLINE_VECS) {
-			bio = bio_alloc_percpu_cache(bdev, nr_vecs, opf,
-						     gfp_mask, bs);
-			if (bio)
-				return bio;
-			/*
-			 * No cached bio available, bio returned below marked with
-			 * REQ_ALLOC_CACHE to particpate in per-cpu alloc cache.
-			 */
-		} else {
-			opf &= ~REQ_ALLOC_CACHE;
-		}
-	}
+	opf |= REQ_ALLOC_CACHE;
+	if (bs->cache && nr_vecs <= BIO_INLINE_VECS) {
+		bio = bio_alloc_percpu_cache(bdev, nr_vecs, opf,
+					     gfp_mask, bs);
+		if (bio)
+			return bio;
+		/*
+		 * No cached bio available, bio returned below marked with
+		 * REQ_ALLOC_CACHE to participate in per-cpu alloc cache.
+		 */
+	} else
+		opf &= ~REQ_ALLOC_CACHE;
 
 	/*
 	 * submit_bio_noacct() converts recursion to iteration; this means if
diff --git a/block/blk-map.c b/block/blk-map.c
index 23e5d5ebe59e..570a7ca6edd1 100644
--- a/block/blk-map.c
+++ b/block/blk-map.c
@@ -255,6 +255,10 @@ static struct bio *blk_rq_map_bio_alloc(struct request *rq,
 {
 	struct bio *bio;
 
+	/*
+	 * Even REQ_ALLOC_CACHE is enabled by default, we still need this to
+	 * mark bio is allocated by bio_alloc_bioset.
+	 */
 	if (rq->cmd_flags & REQ_ALLOC_CACHE && (nr_vecs <= BIO_INLINE_VECS)) {
 		bio = bio_alloc_bioset(NULL, nr_vecs, rq->cmd_flags, gfp_mask,
 					&fs_bio_set);
diff --git a/block/fops.c b/block/fops.c
index ddbc69c0922b..090562a91b4c 100644
--- a/block/fops.c
+++ b/block/fops.c
@@ -177,8 +177,6 @@ static ssize_t __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
 	loff_t pos = iocb->ki_pos;
 	int ret = 0;
 
-	if (iocb->ki_flags & IOCB_ALLOC_CACHE)
-		opf |= REQ_ALLOC_CACHE;
 	bio = bio_alloc_bioset(bdev, nr_pages, opf, GFP_KERNEL,
 			       &blkdev_dio_pool);
 	dio = container_of(bio, struct blkdev_dio, bio);
@@ -326,8 +324,6 @@ static ssize_t __blkdev_direct_IO_async(struct kiocb *iocb,
 	loff_t pos = iocb->ki_pos;
 	int ret = 0;
 
-	if (iocb->ki_flags & IOCB_ALLOC_CACHE)
-		opf |= REQ_ALLOC_CACHE;
 	bio = bio_alloc_bioset(bdev, nr_pages, opf, GFP_KERNEL,
 			       &blkdev_dio_pool);
 	dio = container_of(bio, struct blkdev_dio, bio);
diff --git a/include/linux/fs.h b/include/linux/fs.h
index 601d036a6c78..18ec41732186 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -365,8 +365,6 @@ struct readahead_control;
 /* 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)
 /*
  * IOCB_DIO_CALLER_COMP can be set by the iocb owner, to indicate that the
  * iocb completion can be passed back to the owner for execution from a safe
@@ -399,7 +397,6 @@ struct readahead_control;
 	{ IOCB_WRITE,		"WRITE" }, \
 	{ IOCB_WAITQ,		"WAITQ" }, \
 	{ IOCB_NOIO,		"NOIO" }, \
-	{ IOCB_ALLOC_CACHE,	"ALLOC_CACHE" }, \
 	{ IOCB_DIO_CALLER_COMP,	"CALLER_COMP" }, \
 	{ IOCB_AIO_RW,		"AIO_RW" }, \
 	{ IOCB_HAS_METADATA,	"AIO_HAS_METADATA" }
diff --git a/io_uring/rw.c b/io_uring/rw.c
index af5a54b5db12..fa7655ab9097 100644
--- a/io_uring/rw.c
+++ b/io_uring/rw.c
@@ -856,7 +856,6 @@ static int io_rw_init_file(struct io_kiocb *req, fmode_t mode, int rw_type)
 	ret = kiocb_set_rw_flags(kiocb, rw->flags, rw_type);
 	if (unlikely(ret))
 		return ret;
-	kiocb->ki_flags |= IOCB_ALLOC_CACHE;
 
 	/*
 	 * If the file is marked O_NONBLOCK, still allow retry for it if it
-- 
2.39.5 (Apple Git-154)