Re: [PATCHv2 1/2] block: accumulate segment page gaps per bio

[Christoph Hellwig <hch@lst.de>]

On Wed, Aug 06, 2025 at 07:51:35AM -0700, Keith Busch wrote:
> +static inline unsigned int bvec_seg_gap(struct bio_vec bv, struct bio_vec bp)
> +{
> +	return bv.bv_offset | ((bp.bv_offset + bp.bv_len) & (PAGE_SIZE - 1));
> +}

Can you just pass a pointer to the bio_vec even if the compiler
is probably optimizing away the function either way?

Also bp is a bit of an odd name for a bio_vec.  Note that bvprv flows
well, but it is what the caller uses, so mahybe stick to it?

>  	/* the following two fields are internal, NEVER access directly */
>  	unsigned int __data_len;	/* total data len */
> +	unsigned int __page_gaps;	/* a mask of all the segment gaps */
>  	sector_t __sector;		/* sector cursor */
>  
>  	struct bio *bio;
> @@ -1080,6 +1081,11 @@ static inline sector_t blk_rq_pos(const struct request *rq)
>  	return rq->__sector;
>  }
>  
> +static inline unsigned int blk_rq_page_gaps(const struct request *rq)
> +{
> +	return rq->__page_gaps;
> +}

I don't think we really need the __ and the access helper here.  This was
mostly done for fields where historically drivers accessed the field
directly, but it subtly changed semantics making these direct accesses
unsafe.