Re: [PATCH] block: make sure big bio is splitted into at most 256 bvecs

[Kent Overstreet]

On Wed, Apr 06, 2016 at 10:51:40AM +0800, Ming Lei wrote:
> On Wed, Apr 6, 2016 at 10:40 AM, Kent Overstreet
> <kent.overstreet@xxxxxxxxx> wrote:
> > On Wed, Apr 06, 2016 at 10:37:05AM +0800, Ming Lei wrote:
> >> On Wed, Apr 6, 2016 at 10:34 AM, Kent Overstreet
> >> <kent.overstreet@xxxxxxxxx> wrote:
> >> > On Wed, Apr 06, 2016 at 10:30:22AM +0800, Ming Lei wrote:
> >> >> Now limit.max_segments is for limiting segments from hw view, one this
> >> >> segment may include lots of pages/bvecs.
> >> >>
> >> >> The current bio_clone() issue is that we can't clone from one bio which
> >> >> includes more than 256 bvecs, maybe all these 256 bvecs belong to
> >> >> one same hw segment.
> >> >
> >> > So the distinction is purely a post multipage bvec thing?
> >>
> >> Even after multipage bvec is applied, the limit for max bvecs is still needed
> >> for some cases like bio bounce, in which bio_clone() need to
> >> clone single page bvec.
> >
> > s/max bvecs/max pages/?
> 
> Exactly, :-)
> 
> >
> > What I meant is that until we have multipage bvecs, unless I'm missing something
> > max_segments is exactly what we want. After multipage bvecs, things do get more
> 
> Yes, but now we need to fix current issue and backport the fix. Given bio
> bounce isn't easy to fix, not like other users of bio_clone, I think
> we still need
> to apply the max pages limit globally, what do you think of it?

Ugh, yeah bouncing is an issue, with that it's most of the drivers.

Ok, yeah I guess the global limit makes sense for now.

Please add a giant comment explaining why the global limit is there, precisely
what it's needed for and that we'd like to get rid of it.