xfs
[Top] [All Lists]

Re: bio_add_page rw mode check by merge_bvec_fn

To: Christoph Hellwig <hch@xxxxxxxxxxxxx>
Subject: Re: bio_add_page rw mode check by merge_bvec_fn
From: NeilBrown <neilb@xxxxxxx>
Date: Sat, 23 Nov 2013 16:12:34 +1100
Cc: Bernd Schubert <bernd.schubert@xxxxxxxxxxxxxxxxxx>, "linux-fsdevel@xxxxxxxxxxxxxxx" <linux-fsdevel@xxxxxxxxxxxxxxx>, xfs@xxxxxxxxxxx, linux-ext4@xxxxxxxxxxxxxxx, Jens Axboe <axboe@xxxxxxxxx>, Kent Overstreet <kmo@xxxxxxxxxxxxx>
Delivered-to: xfs@xxxxxxxxxxx
In-reply-to: <20131122153618.GB15143@xxxxxxxxxxxxx>
References: <528E4A50.7040609@xxxxxxxxxxxxxxxxxx> <528E4C19.70507@xxxxxxxxxxxxxxxxxx> <20131122153618.GB15143@xxxxxxxxxxxxx>
On Fri, 22 Nov 2013 07:36:18 -0800 Christoph Hellwig <hch@xxxxxxxxxxxxx>
wrote:

> While this is trivial to fix it's also fairly unexpected and easy
> to get wrong for new callers.  Neil, can you explain why you
> desperately need it?

Desperately?  Not at all?
Need?  Not really.  This is just in RAID5 and merge_bvec_fn is purely an
optimisation for RAID5.
Limiting read BIOs to one chunk allows us to bypass the stripe-cache, so can
be good.
Limiting write BIOs is completely unnecessary so we currently don't bother.

So I have no objection to bvm->bi_rw being removed.

Thanks,
NeilBrown

Attachment: signature.asc
Description: PGP signature

<Prev in Thread] Current Thread [Next in Thread>