Re: [PATCH 2/6] mm: Optimise find_subpage for !THP

[Matthew Wilcox <willy@infradead.org>]

On Tue, Mar 03, 2020 at 01:28:13PM -0800, Alexander Duyck wrote:
> On Mon, Mar 2, 2020 at 8:11 PM Matthew Wilcox <willy@infradead.org> wrote:
> > From: "Matthew Wilcox (Oracle)" <willy@infradead.org>
> >
> > If THP is disabled, find_subpage can become a no-op.
> >
> > Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
> > Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
> > ---
> >  include/linux/pagemap.h | 15 ++++++++++-----
> >  1 file changed, 10 insertions(+), 5 deletions(-)
> >
> > diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
> > index b4ea3a5d00e5..8785f60b05f8 100644
> > --- a/include/linux/pagemap.h
> > +++ b/include/linux/pagemap.h
> > @@ -333,14 +333,19 @@ static inline struct page *grab_cache_page_nowait(struct address_space *mapping,
> >                         mapping_gfp_mask(mapping));
> >  }
> >
> > -static inline struct page *find_subpage(struct page *page, pgoff_t offset)
> > +/*
> > + * Given the page we found in the page cache, return the page corresponding
> > + * to this offset in the file
> > + */
> > +static inline struct page *find_subpage(struct page *head, pgoff_t offset)
> >  {
> > -       if (PageHuge(page))
> > -               return page;
> > +       /* HugeTLBfs wants the head page regardless */
> > +       if (PageHuge(head))
> > +               return head;
> >
> > -       VM_BUG_ON_PAGE(PageTail(page), page);
> > +       VM_BUG_ON_PAGE(PageTail(head), head);
> 
> Is there any specific reason for renaming page to head? Just wondering

Christoph Hellwig asked for the rename in an earlier version of this
patchset.

> since it adds some noise to the patch that wasn't really called out in
> the patch description. From what I can tell none of the changes above
> this point have any explanation to them in the patch description, and
> until I noticed the change below I thought maybe you had the wrong
> patch description for this patch.
> 
> > -       return page + (offset & (compound_nr(page) - 1));
> > +       return head + (offset & (hpage_nr_pages(head) - 1));
> >  }
> 
> So the patch description refers to this line here, correct?

Yes, that's the actual change.

> One thing I am noticing is that it looks like the VM_BUG_ON_PAGE is
> now redundant. If I am not not mistaken hpage_nr_pages will call
> PageTransHuge which also performs the same check. So do you still need
> the VM_BUG_ON_PAGE call in this function?

Huh, I didn't know PageTransHuge had that.  I suppose it can go, although
there's only a call to PageTransHuge() in the CONFIG_TRANSPARENT_HUGEPAGE
enabled case.  You might ask how a tail page could get into the page
cache if CONFIG_TRANSPARENT_HUGEPAGE is not enabled, and I would not
have a good answer for that ... although if one has, I think we'd want
to know about it.  So maybe that's a good reason to keep the explicit
check, or maybe it's a good reason to make hpage_nr_pages() check that
assertion if !CONFIG_TRANSPARENT_HUGEPAGE.