On Sat, Mar 03, 2018 at 12:00:42AM +0100, Christoph Hellwig wrote:
> Oh, and another thing: I think you want to make this new code dependent
> on the block devie actually supporting REQ_FUA natively. Otherwise
> you'll cause a flush for every emulated FUA write, which is only going
> make things worse, especially for ATA where FLUSH is not queued. And
> last time I check libata still disabled FUA by default.
Yup, but the issue we have right now is that for pure RWF_DSYNC data
overwrites we are already doing a post-flush on every IO. It's being
issued as a separate zero-length IO, which is why REQ_FUA is faster
and results in lower overall IOPS. The flush comes from this path:
generic_write_sync
vfs_fsync_range
xfs_file_fsync
....
/*
* If we only have a single device, and the log force about was
* a no-op we might have to flush the data device cache here.
* This can only happen for fdatasync/O_DSYNC if we were overwriting
* an already allocated file and thus do not have any metadata to
* commit.
*/
if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) &&
mp->m_logdev_targp == mp->m_ddev_targp)
xfs_blkdev_issue_flush(mp->m_ddev_targp);
So the end result of using REQ_FUA and not calling
generic_write_sync() on devices that don't support REQ_FUA is that
we get a post-flush attached to the IO rather than separately
issuing the post-flush via generic_write_sync().
Hence I think we end up with the same behaviour on devices that
don't support REQ_FUA, just via a slightly different mechanism.
Cheers,
Dave.
--
Dave Chinner
david@xxxxxxxxxxxxx
