On Wed, Aug 06, 2025 at 09:44:47AM -0600, Keith Busch wrote:
> On Wed, Aug 06, 2025 at 04:56:21PM +0200, Christoph Hellwig wrote:
> > > index 0a29b20939d17..d0ed28d40fe02 100644
> > > --- a/include/linux/blk_types.h
> > > +++ b/include/linux/blk_types.h
> > > @@ -264,6 +264,8 @@ struct bio {
> > >
> > > unsigned short bi_max_vecs; /* max bvl_vecs we can hold */
> > >
> > > + unsigned int page_gaps; /* a mask of all the vector gaps */
> >
> > Bloating the bio for the gaps, especially as the bio is otherwise not
> > built to hardware limits at all seems like an odd tradeoff.
>
> Maybe, but I don't have anywhere else to put this. We split the bio to
> its hardware limits at some point, which is where this field gets
> initially set.
What we do for nr_segs is to just do it on stack while splitting,
and then only record it in the request. I think you could do the same
here, i.e. replace the nr_segs output parameter to __bio_split_to_limits
and it's helpers with a new struct bio_split_info that in the first
version just contains nr_segs, but can be extended to other easily
derived information like the page gaps.
