Re: [PATCH 2/6] mm/page_io: use a folio in sio_read_complete()

[Matthew Wilcox]

On Tue, Jul 18, 2023 at 08:58:17PM +0800, zhangpeng (AS) wrote:
> On 2023/7/17 21:40, Matthew Wilcox wrote:
> 
> > On Mon, Jul 17, 2023 at 09:25:58PM +0800, Peng Zhang wrote:
> > > +++ b/mm/page_io.c
> > > @@ -406,19 +406,19 @@ static void sio_read_complete(struct kiocb *iocb, long ret)
> > >   	if (ret == sio->len) {
> > >   		for (p = 0; p < sio->pages; p++) {
> > > -			struct page *page = sio->bvec[p].bv_page;
> > > +			struct folio *folio = page_folio(sio->bvec[p].bv_page);
> > > -			SetPageUptodate(page);
> > > -			unlock_page(page);
> > > +			folio_mark_uptodate(folio);
> > > +			folio_unlock(folio);
> > >   		}
> > I'm kind of shocked this works today.  Usually bvecs coalesce adjacent
> > pages into a single entry, so you need to use a real iterator like
> > bio_for_each_folio_all() to extract individual pages from a bvec.
> > Maybe the sio bvec is constructed inefficiently.
> > 
> > I think Kent had some bvec folio iterators in progress?
> 
> I'll convert bio_first_page_all() to bio_first_folio_all() in a v2.

That isn't my point at all.  What I'm saying is that when you call a
function like bio_add_folio(), if @folio is physically adjacent to the
immediately prior folio already in the bio, it will extend the bv_len
instead of adding the new folio to the bvec.  Maybe there's nothing like
that for sio.