Re: [PATCH next] bio: zero inlined bio_vec

From: Hugh Dickins
Date: Tue Dec 23 2008 - 06:20:45 EST


On Tue, 23 Dec 2008, Jens Axboe wrote:
> On Tue, Dec 23 2008, Jens Axboe wrote:
> >
> > That is, it iterates *bio_orig but indexes bio since it knows it has the
> > same number of segments. So this code is the odd one out, I'd be
> > surprised if we had more such cases. And since it would be nice to get
> > rid of the need to memset in general, can you try with the below patch?
>
> Nope, that still wont be enough, since we leave entries 0..i-1
> uninitialized. Lets just do this instead.

Yes, that second version worked for me. But if __blk_queue_bounce()
is indeed the odd one out (likely but not certain), then the extension
below makes more sense - I don't like how your bio.c returns a cleared
bio_vec in some cases but not in others. This is running fine for me
on my test machines, but my test coverage is probably very poor (am I
ever using more than the inlined bio_vec?), and I've not really tried
to understand all the ways through bvec_alloc_bs().

Hugh

--- a/fs/bio.c
+++ b/fs/bio.c
@@ -180,7 +180,7 @@ struct bio_vec *bvec_alloc_bs(gfp_t gfp_
* kzalloc() for the exact number of vecs right away.
*/
if (!bs)
- bvl = kzalloc(nr * sizeof(struct bio_vec), gfp_mask);
+ bvl = kmalloc(nr * sizeof(struct bio_vec), gfp_mask);

/*
* see comment near bvec_array define!
@@ -237,9 +237,6 @@ fallback:
}
}

- if (bvl)
- memset(bvl, 0, bvec_nr_vecs(*idx) * sizeof(struct bio_vec));
-
return bvl;
}

--- a/mm/bounce.c
+++ b/mm/bounce.c
@@ -198,11 +198,14 @@ static void __blk_queue_bounce(struct re
/*
* irk, bounce it
*/
- if (!bio)
- bio = bio_alloc(GFP_NOIO, (*bio_orig)->bi_vcnt);
+ if (!bio) {
+ unsigned int vcnt = (*bio_orig)->bi_vcnt;

- to = bio->bi_io_vec + i;
+ bio = bio_alloc(GFP_NOIO, vcnt);
+ memset(bio->bi_io_vec, 0,vcnt * sizeof(struct bio_vec));
+ }

+ to = bio->bi_io_vec + i;
to->bv_page = mempool_alloc(pool, q->bounce_gfp);
to->bv_len = from->bv_len;
to->bv_offset = from->bv_offset;
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/