Re: [PATCH v2] btrfs: introduce sync_csum_mode to tweak sync checksum behavior

[David Sterba <dsterba@suse.cz>]

On Wed, Jan 31, 2024 at 02:15:32PM +0000, Johannes Thumshirn wrote:
> On 31.01.24 08:15, Naohiro Aota wrote:
> > We disable offloading checksum to workqueues and do it synchronously when
> > the checksum algorithm is fast. However, as reported in the link below,
> > RAID0 with multiple devices may suffer from the sync checksum, because
> > "fast checksum" is still not fast enough to catch up RAID0 writing.
> > 
> > To measure the effectiveness of sync checksum for developers, it would be
> > better to have a switch for the sync checksum under CONFIG_BTRFS_DEBUG
> > hood.
> > 
> > This commit introduces fs_devices->sync_csum_mode for CONFIG_BTRFS_DEBUG,
> > so that a btrfs developer can change the behavior by writing to
> > /sys/fs/btrfs/<uuid>/sync_csum. The default is "auto" which is the same as
> > the previous behavior. Or, you can set "on" or "off" to always/never use
> > sync checksum.
> > 
> > More benchmark should be collected with this knob to implement a proper
> > criteria to enable/disable sync checksum.
> > 
> > Link: https://lore.kernel.org/linux-btrfs/20230731152223.4EFB.409509F4@e16-tech.com/
> > Link: https://lore.kernel.org/linux-btrfs/p3vo3g7pqn664mhmdhlotu5dzcna6vjtcoc2hb2lsgo2fwct7k@xzaxclba5tae/
> > Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
> 
> As access to sysfs and fs_info can happen concurrently, should the 
> read/write of fs_devices->sync_csum_mode be guarded by a 
> READ_ONCE()/WRITE_ONCE()?

Yes we use the *_ONCE helpers for values set from sysfs in cases where
it's without bad effects to race and change the value in the middle.
Here it would only skip one checksum offload/sync. It's not really a
guard but a note that there's something special about the value.