On Mon, Mar 25 2013, Jan Vesely wrote:
> v2: changed a comment
>
> The original behavior was to refuse all pages after the maximum number of
> segments has been reached. However, some drivers (like st) craft their buffers
> to potentially require exactly max segments and multiple pages in the last
> segment. This patch modifies the check to allow pages that can be merged into
> the last segment.
>
> Fixes EBUSY failures when using large tape block size in high
> memory fragmentation condition.
> This regression was introduced by commit
> 46081b166415acb66d4b3150ecefcd9460bb48a1
> st: Increase success probability in driver buffer allocation
>
> Signed-off-by: Jan Vesely <jves...@redhat.com>
>
> CC: Alexander Viro <v...@zeniv.linux.org.uk>
> CC: FUJITA Tomonori <fujita.tomon...@lab.ntt.co.jp>
> CC: Kai Makisara <kai.makis...@kolumbus.fi>
> CC: James Bottomley <james.bottom...@hansenpartnership.com>
> CC: Jens Axboe <ax...@kernel.dk>
> CC: sta...@vger.kernel.org
> ---
> fs/bio.c | 27 +++++++++++++++++----------
> 1 file changed, 17 insertions(+), 10 deletions(-)
>
> diff --git a/fs/bio.c b/fs/bio.c
> index bb5768f..bc6af71 100644
> --- a/fs/bio.c
> +++ b/fs/bio.c
> @@ -500,7 +500,6 @@ static int __bio_add_page(struct request_queue *q, struct
> bio *bio, struct page
> *page, unsigned int len, unsigned int offset,
> unsigned short max_sectors)
> {
> - int retried_segments = 0;
> struct bio_vec *bvec;
>
> /*
> @@ -551,18 +550,13 @@ static int __bio_add_page(struct request_queue *q,
> struct bio *bio, struct page
> return 0;
>
> /*
> - * we might lose a segment or two here, but rather that than
> - * make this too complex.
> + * The first part of the segment count check,
> + * reduce segment count if possible
> */
>
> - while (bio->bi_phys_segments >= queue_max_segments(q)) {
> -
> - if (retried_segments)
> - return 0;
> -
> - retried_segments = 1;
> + if (bio->bi_phys_segments >= queue_max_segments(q))
> blk_recount_segments(q, bio);
> - }
> +
>
> /*
> * setup the new entry, we might clear it again later if we
> @@ -572,6 +566,19 @@ static int __bio_add_page(struct request_queue *q,
> struct bio *bio, struct page
> bvec->bv_page = page;
> bvec->bv_len = len;
> bvec->bv_offset = offset;
> +
> + /*
> + * the other part of the segment count check, allow mergeable pages
> + */
> + if ((bio->bi_phys_segments > queue_max_segments(q)) ||
> + ( (bio->bi_phys_segments == queue_max_segments(q)) &&
> + !BIOVEC_PHYS_MERGEABLE(bvec - 1, bvec))) {
> + bvec->bv_page = NULL;
> + bvec->bv_len = 0;
> + bvec->bv_offset = 0;
> + return 0;
> + }
> +
This is a bit messy, I think. bi_phys_segments should never be allowed
to go beyond queue_ma_segments(), so the > test does not look right.
Maybe it's an artifact of when we fall through with this patch, we bump
bi_phys_segments even if the segments are physicall contig and
mergeable.
What happens when the segment is physically mergeable, but the resulting
merged segment is too large (bigger than q->limits.max_segment_size)?
--
Jens Axboe
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
