Re: [PATCH] block: respect virtual boundary mask in bvecs

[Keith Busch]

On Mon, Nov 05, 2018 at 11:23:01AM +0100, Johannes Thumshirn wrote:
> With drivers like iSer we are seeing a lot of bio splitting and smaller I/Os
> being submitted to the driver.
> 
> The root cause of this issue that the virtual boundary mask code does not take
> into consideration that some of the memory segments in the SG list may have
> come from a huge memory page that is being managed in the SG list as 4K
> blocks.  This means that many of the segments in the SG list will have an
> offset into the page that is not 0 but will be a multiple of 4K.

I probably got this wrong, but I thought a 2M huge page was 512 regular
pages with a compound head, and offsets were from those regular pages
rather than from the head.

Overall though, the patch makes sense to me for this and other reasons.

Acked-by: Keith Busch <keith.busch@xxxxxxxxx>

 
> Cc: Jan Kara <jack@xxxxxxx>
> Cc: Sagi Grimberg <sagi@xxxxxxxxxxx>
> Signed-off-by: Johannes Thumshirn <jthumshirn@xxxxxxx>
> ---
>  block/blk-merge.c | 2 +-
>  block/blk.h       | 2 +-
>  2 files changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/block/blk-merge.c b/block/blk-merge.c
> index 6b5ad275ed56..208658a901c6 100644
> --- a/block/blk-merge.c
> +++ b/block/blk-merge.c
> @@ -46,7 +46,7 @@ static inline bool bio_will_gap(struct request_queue *q,
>  		bio_get_first_bvec(prev_rq->bio, &pb);
>  	else
>  		bio_get_first_bvec(prev, &pb);
> -	if (pb.bv_offset)
> +	if (pb.bv_offset & queue_virt_boundary(q))
>  		return true;
>  
>  	/*
> diff --git a/block/blk.h b/block/blk.h
> index a1841b8ff129..c85e53f21cdd 100644
> --- a/block/blk.h
> +++ b/block/blk.h
> @@ -169,7 +169,7 @@ static inline bool biovec_phys_mergeable(struct request_queue *q,
>  static inline bool __bvec_gap_to_prev(struct request_queue *q,
>  		struct bio_vec *bprv, unsigned int offset)
>  {
> -	return offset ||
> +	return (offset & queue_virt_boundary(q)) ||
>  		((bprv->bv_offset + bprv->bv_len) & queue_virt_boundary(q));
>  }
>  
> --