linux-kernel - Re: [PATCH V11 03/19] block: introduce bio_for_each

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Wed, 21 Nov 2018 23:31:36 +0800
From:   Ming Lei <ming.lei@...hat.com>
To:     Christoph Hellwig <hch@....de>
Cc:     Jens Axboe <axboe@...nel.dk>, linux-block@...r.kernel.org,
        linux-kernel@...r.kernel.org, linux-mm@...ck.org,
        Theodore Ts'o <tytso@....edu>, Omar Sandoval <osandov@...com>,
        Sagi Grimberg <sagi@...mberg.me>,
        Dave Chinner <dchinner@...hat.com>,
        Kent Overstreet <kent.overstreet@...il.com>,
        Mike Snitzer <snitzer@...hat.com>, dm-devel@...hat.com,
        Alexander Viro <viro@...iv.linux.org.uk>,
        linux-fsdevel@...r.kernel.org, Shaohua Li <shli@...nel.org>,
        linux-raid@...r.kernel.org, David Sterba <dsterba@...e.com>,
        linux-btrfs@...r.kernel.org,
        "Darrick J . Wong" <darrick.wong@...cle.com>,
        linux-xfs@...r.kernel.org, Gao Xiang <gaoxiang25@...wei.com>,
        linux-ext4@...r.kernel.org, Coly Li <colyli@...e.de>,
        linux-bcache@...r.kernel.org, Boaz Harrosh <ooo@...ctrozaur.com>,
        Bob Peterson <rpeterso@...hat.com>, cluster-devel@...hat.com
Subject: Re: [PATCH V11 03/19] block: introduce bio_for_each_bvec()

On Wed, Nov 21, 2018 at 02:32:44PM +0100, Christoph Hellwig wrote:
> > +#define bio_iter_mp_iovec(bio, iter)				\
> > +	segment_iter_bvec((bio)->bi_io_vec, (iter))
> 
> Besides the mp naming we'd like to get rid off there also is just
> a single user of this macro, please just expand it there.

OK.

> 
> > +#define segment_iter_bvec(bvec, iter)				\
> > +((struct bio_vec) {							\
> > +	.bv_page	= segment_iter_page((bvec), (iter)),	\
> > +	.bv_len		= segment_iter_len((bvec), (iter)),	\
> > +	.bv_offset	= segment_iter_offset((bvec), (iter)),	\
> > +})
> 
> And for this one please keep the segment vs bvec versions of these
> macros close together in the file please, right now it follow the
> bvec_iter_bvec variant closely.

OK.

> 
> > +static inline void __bio_advance_iter(struct bio *bio, struct bvec_iter *iter,
> > +				      unsigned bytes, unsigned max_seg_len)
> >  {
> >  	iter->bi_sector += bytes >> 9;
> >  
> >  	if (bio_no_advance_iter(bio))
> >  		iter->bi_size -= bytes;
> >  	else
> > -		bvec_iter_advance(bio->bi_io_vec, iter, bytes);
> > +		__bvec_iter_advance(bio->bi_io_vec, iter, bytes, max_seg_len);
> >  		/* TODO: It is reasonable to complete bio with error here. */
> >  }
> >  
> > +static inline void bio_advance_iter(struct bio *bio, struct bvec_iter *iter,
> > +				    unsigned bytes)
> > +{
> > +	__bio_advance_iter(bio, iter, bytes, PAGE_SIZE);
> > +}
> 
> Btw, I think the remaining users of bio_advance_iter() in bio.h
> should probably switch to using __bio_advance_iter to make them a little
> more clear to read.

Good point.

> 
> > +/* returns one real segment(multi-page bvec) each time */
> 
> space before the brace, please.

OK.

> 
> > +#define BVEC_MAX_LEN  ((unsigned int)-1)
> 
> >  	while (bytes) {
> > +		unsigned segment_len = segment_iter_len(bv, *iter);
> >  
> > -		iter->bi_bvec_done += len;
> > +		if (max_seg_len < BVEC_MAX_LEN)
> > +			segment_len = min_t(unsigned, segment_len,
> > +					    max_seg_len -
> > +					    bvec_iter_offset(bv, *iter));
> > +
> > +		segment_len = min(bytes, segment_len);
> 
> Please stick to passing the magic zero here as can often generate more
> efficient code.

But zero may decrease the code readability. Actually the passed
'max_seg_len' is just a constant, and complier should have generated
same efficient code for any constant, either 0 or other.

> 
> Talking about efficent code - I wonder how much code size we'd save
> by moving this function out of line..

That is good point, see the following diff:

[mingl@hp kernel]$ diff -u inline.size non_inline.size
--- inline.size	2018-11-21 23:24:52.305312076 +0800
+++ non_inline.size	2018-11-21 23:24:59.908393010 +0800
@@ -1,2 +1,2 @@
    text	   data	    bss	    dec	    hex	filename
-13429213	6893922	4292692	24615827	1779b93	vmlinux.inline
+13429153	6893346	4292692	24615191	1779917	vmlinux.non_inline

vmlinux(non_inline) is built by just moving/exporting __bvec_iter_advance()
into block/bio.c.

The difference is about 276bytes.

> 
> But while looking over this I wonder why we even need the max_seg_len
> here.  The only thing __bvec_iter_advance does it to move bi_bvec_done
> and bi_idx forward, with corresponding decrements of bi_size.  As far
> as I can tell the only thing that max_seg_len does is that we need
> to more iterations of the while loop to archive the same thing.
> 
> And actual bvec used by the caller will be obtained using
> bvec_iter_bvec or segment_iter_bvec depending on if they want multi-page
> or single-page variants.

Right, we let __bvec_iter_advance() serve for both multi-page and single-page
case, then we have to tell it via one way or another, now we use the constant
of 'max_seg_len'.

Or you suggest to implement two versions of __bvec_iter_advance()?

Thanks,
Ming