Re: [RESEND PATCH V5 00/33] block: support multipage bvec

From: Ming Lei
Date: Sun May 27 2018 - 03:24:18 EST


On Fri, May 25, 2018 at 10:30:46AM -0600, Jens Axboe wrote:
> On 5/24/18 10:53 PM, Kent Overstreet wrote:
> > On Fri, May 25, 2018 at 11:45:48AM +0800, Ming Lei wrote:
> >> Hi,
> >>
> >> This patchset brings multipage bvec into block layer:
> >
> > patch series looks sane to me. goddamn that's a lot of renaming.
>
> Indeed... I actually objected to some of the segment -> page renaming,
> but it's still in there. The foo2() temporary functions also concern me,
> we all know there's nothing more permanent than a temporary fixup.

Jens, I remember I explained the renaming story to you in lsfmm a bit:

1) the current naming of segment is actually wrong, since every segment
only stores one single-page vector

2) the most important part is that once multipage bvec is introduced,
if the old _segment naming is still kept, it can be very confusing,
especially no good name is left for the helpers of dealing with real
segment.

For the foo2() temporary change, that is only for avoiding tree-wide
change in one single tree, with this way, we can change sub-system one
by one, but if you think it is good to do tree-wide conversion in one
patch, I am fine to do it in next version.

>
> > Things are going to get interesting when we start sticking compound pages in the
> > page cache, there'll be some interesting questions of semantics to deal with
> > then but I think getting this will only help w.r.t. plumbing that through and
> > not dealing with 4k pages unnecessarily - but I think even if we were to decide
> > that merging in bio_add_page() is not the way to go when the upper layers are
> > passing compound pages around already, this patch series helps because
> > regardless at some point everything under generic_make_request() is going to
> > have to deal with segments that are more than one page, and this patch series
> > makes that happen. So incremental progress.
> >
> > Jens, any objections to getting this in?
>
> I like most of it, but I'd much rather get this way earlier in the series.
> We're basically just one week away from the merge window, it needs more simmer
> and testing time than that. On top of that, it hasn't received much review
> yet.
>
> So as far as I'm concerned, we can kick off the 4.19 block branch with
> iterated versions of this patchset.

OK, I will post out again once v4.19 is started.

Thanks,
Ming