Re: trying to understand READ_META, READ_SYNC, WRITE_SYNC & co

[Christoph Hellwig <hch@lst.de>]

On Mon, Jun 21, 2010 at 08:56:55PM +0200, Jens Axboe wrote:
> FWIW, Windows marks meta data writes and they go out with FUA set
> on SATA disks. And SATA firmware prioritizes FUA writes, it's essentially
> a priority bit as well as a platter access bit. So at least we have some
> one else using a meta data boost. I agree that it would be a lot more
> trivial to add the annotations if they didn't have scheduler impact
> as well, but I still think it's a sane thing.

And we still disable the FUA bit in libata unless people set a
non-standard module option..

> >> Reads are sync by nature in the block layer, so they don't get that
> >> special annotation.
> > 
> > Well, we do give them this special annotation in a few places, but we
> > don't actually use it.
> 
> For unplugging?

We use the explicit unplugging, yes - but READ_META also includes
REQ_SYNC which is not used anywhere.

> > But that leaves the question why disabling the idling logical for
> > data integrity ->writepage is fine?  This gets called from ->fsync
> > or O_SYNC writes and will have the same impact as O_DIRECT writes.
> 
> We have never enabled idling for those. O_SYNC should get a nice
> boost too, it just needs to be benchmarked and tested and then
> there would be no reason not to add it.

We've only started using any kind of sync tag last year in ->writepage in
commit a64c8610bd3b753c6aff58f51c04cdf0ae478c18 "block_write_full_page:
Use synchronous writes for WBC_SYNC_ALL writebacks", switching from
WRITE_SYNC to WRITE_SYNC_PLUG a bit later in commit
6e34eeddf7deec1444bbddab533f03f520d8458c "block_write_full_page: switch
synchronous writes to use WRITE_SYNC_PLUG"

Before that we used plain WRITE, which had the normal idling logic.