[PATCH] blk-wbt: Fix data race and avoid possible false sharing

[PATCH] blk-wbt: Fix data race and avoid possible false sharing

[Marco Elver]

The pattern here is to avoid possible false sharing. However, due to
compiler optimizations the code may simply collapse to the write if we
omit READ_ONCE/WRITE_ONCE:
https://github.com/google/ktsan/wiki/READ_ONCE-and-WRITE_ONCE#it-may-improve-performance

==================================================================
BUG: KCSAN: data-race in wbt_wait / wbt_wait

read to 0xffff88821aa6d140 of 8 bytes by task 10372 on cpu 1:
 wb_timestamp block/blk-wbt.c:88 [inline]
 wb_timestamp block/blk-wbt.c:83 [inline]
 wbt_wait+0x1f9/0x250 block/blk-wbt.c:587
 __rq_qos_throttle+0x47/0x70 block/blk-rq-qos.c:72
 rq_qos_throttle block/blk-rq-qos.h:185 [inline]
 blk_mq_make_request+0x29c/0xf60 block/blk-mq.c:1971
 generic_make_request block/blk-core.c:1064 [inline]
 generic_make_request+0x196/0x740 block/blk-core.c:1006
 submit_bio+0x96/0x3c0 block/blk-core.c:1190
 submit_bh_wbc+0x40f/0x460 fs/buffer.c:3095
 submit_bh fs/buffer.c:3101 [inline]
 __bread_slow fs/buffer.c:1177 [inline]
 __bread_gfp+0xe7/0x1e0 fs/buffer.c:1359
 sb_bread include/linux/buffer_head.h:307 [inline]
 fat__get_entry+0x35e/0x4f0 fs/fat/dir.c:100
 fat_get_entry fs/fat/dir.c:128 [inline]
 fat_get_short_entry+0x103/0x200 fs/fat/dir.c:877
 fat_subdirs+0x6b/0x110 fs/fat/dir.c:943
 fat_read_root fs/fat/inode.c:1416 [inline]
 fat_fill_super+0x1552/0x1f50 fs/fat/inode.c:1862
 vfat_fill_super+0x3b/0x50 fs/fat/namei_vfat.c:1050
 mount_bdev+0x262/0x2d0 fs/super.c:1415
 vfat_mount+0x3e/0x60 fs/fat/namei_vfat.c:1057

write to 0xffff88821aa6d140 of 8 bytes by task 10375 on cpu 0:
 wb_timestamp block/blk-wbt.c:89 [inline]
 wb_timestamp block/blk-wbt.c:83 [inline]
 wbt_wait+0x21e/0x250 block/blk-wbt.c:587
 __rq_qos_throttle+0x47/0x70 block/blk-rq-qos.c:72
 rq_qos_throttle block/blk-rq-qos.h:185 [inline]
 blk_mq_make_request+0x29c/0xf60 block/blk-mq.c:1971
 generic_make_request block/blk-core.c:1064 [inline]
 generic_make_request+0x196/0x740 block/blk-core.c:1006
 submit_bio+0x96/0x3c0 block/blk-core.c:1190
 mpage_bio_submit fs/mpage.c:66 [inline]
 mpage_readpages+0x36c/0x3c0 fs/mpage.c:410
 blkdev_readpages+0x36/0x50 fs/block_dev.c:620
 read_pages+0xa2/0x2d0 mm/readahead.c:126
 __do_page_cache_readahead+0x353/0x390 mm/readahead.c:212
 force_page_cache_readahead+0x13a/0x1f0 mm/readahead.c:243
 page_cache_sync_readahead+0x1cf/0x1e0 mm/readahead.c:522
 generic_file_buffered_read mm/filemap.c:2050 [inline]
 generic_file_read_iter+0xeb6/0x1440 mm/filemap.c:2323
 blkdev_read_iter+0xb2/0xe0 fs/block_dev.c:2010

Reported by Kernel Concurrency Sanitizer on:
CPU: 0 PID: 10375 Comm: blkid Not tainted 5.4.0-rc3+ #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
==================================================================

Reported-by: syzbot+ba8947364367f96fe16b@syzkaller.appspotmail.com
Signed-off-by: Marco Elver <elver@google.com>
---
 block/blk-wbt.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/block/blk-wbt.c b/block/blk-wbt.c
index 8641ba9793c5..ce281a9007a6 100644
--- a/block/blk-wbt.c
+++ b/block/blk-wbt.c
@@ -85,8 +85,8 @@ static void wb_timestamp(struct rq_wb *rwb, unsigned long *var)
 	if (rwb_enabled(rwb)) {
 		const unsigned long cur = jiffies;
 
-		if (cur != *var)
-			*var = cur;
+		if (cur != READ_ONCE(*var))
+			WRITE_ONCE(*var, cur);
 	}
 }
 
-- 
2.24.0.rc1.363.gb1bccd3e3d-goog