hulk inclusion category: perf bugzilla: https://gitee.com/openeuler/kernel/issues/I90ZB5 CVE: NA
--------------------------------
In DIO overwriting case, there is no need to convert unwritten exntents and ext4_handle_inode_extension() can be ignored, which means that endio process can be executed under irq context. Since commit 240930fb7e6b5 ("ext4: dio take shared inode lock when overwriting preallocated blocks") has provided a method to judge whether overwriting is happening, just do nothing in endio process if DIO overwriting happens. This patch enables ext4 processing endio under irq context in DIO overwriting case, which brings a performance improvement in the following fio test on a x86 physical machine with nvme when irq and fio run on the same cpu:
Test: fio -direct=1 -iodepth=128 -rw=randwrite -ioengine=libaio -bs=4k -size=2G -numjobs=1 -overwrite=1 -time_based -runtime=60 -group_reporting -filename=/test/test -name=Rand_write_Testing --cpus_allowed=1
before: 953 MiB/s after: 1350 MiB/s, ~41% perf improvement.
Suggested-by: Zhang Yi yi.zhang@huawei.com Signed-off-by: Zhihao Cheng chengzhihao1@huawei.com --- fs/ext4/file.c | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-)
diff --git a/fs/ext4/file.c b/fs/ext4/file.c index a3d179bc89fc..535ad9d23d7e 100644 --- a/fs/ext4/file.c +++ b/fs/ext4/file.c @@ -489,8 +489,10 @@ static ssize_t ext4_dio_write_iter(struct kiocb *iocb, struct iov_iter *from) loff_t offset = iocb->ki_pos; size_t count = iov_iter_count(from); const struct iomap_ops *iomap_ops = &ext4_iomap_ops; + const struct iomap_dio_ops *iomap_dops = &ext4_dio_write_ops; bool extend = false, unaligned_io = false, unwritten = false; bool ilock_shared = true; + int dio_flags = 0;
/* * We initially start with shared inode lock unless it is @@ -577,10 +579,13 @@ static ssize_t ext4_dio_write_iter(struct kiocb *iocb, struct iov_iter *from) ext4_journal_stop(handle); }
- if (ilock_shared && !unwritten) + if (ilock_shared && !unwritten) { iomap_ops = &ext4_iomap_overwrite_ops; - ret = iomap_dio_rw(iocb, from, iomap_ops, &ext4_dio_write_ops, - (unaligned_io || extend) ? IOMAP_DIO_FORCE_WAIT : 0); + iomap_dops = NULL; + dio_flags = IOMAP_DIO_MAY_INLINE_COMP; + } else if (unaligned_io || extend) + dio_flags |= IOMAP_DIO_FORCE_WAIT; + ret = iomap_dio_rw(iocb, from, iomap_ops, iomap_dops, dio_flags); if (ret == -ENOTBLK) ret = 0;