On Thu, Apr 09, 2020 at 05:09:14PM +0300, Kirill A. Shutemov wrote:
> On Wed, Apr 08, 2020 at 08:01:44AM -0700, Matthew Wilcox wrote:
> > diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
> > index 4aba6c0c2ba8..a8c3fa076f43 100644
> > --- a/include/linux/mm_types.h
> > +++ b/include/linux/mm_types.h
> > @@ -221,15 +221,8 @@ struct page {
> > #endif
> > } _struct_page_alignment;
> >
> > -static inline atomic_t *compound_mapcount_ptr(struct page *page)
> > -{
> > - return &page[1].compound_mapcount;
> > -}
> > -
> > -static inline atomic_t *compound_pincount_ptr(struct page *page)
> > -{
> > - return &page[2].hpage_pinned_refcount;
> > -}
> > +#define compound_mapcount_ptr(page) (&(page)[1].compound_mapcount)
> > +#define compound_pincount_ptr(page) (&(page)[2].hpage_pinned_refcount)
>
> Looks like this conversion is not covered by the reason given in the
> commit message. Hm?
They were supposed to be covered by "That means some inline functions
have to become macros so they can return a struct page which is the same
const-ness as their argument."
They're not quite covered since they need to return an atomic_t that
is the same constness as their argument.
> > +#define compound_head(page) ({ \
> > + __typeof__(page) _page = page; \
> > + unsigned long head = READ_ONCE(_page->compound_head); \
> > + if (unlikely(head & 1)) \
> > + _page = (void *)(head - 1); \
> > + _page; \
> > +})
>
> Ugh..
>
> Other option would be to make compound_head() take 'const struct page *'
> and return 'void *'. It arguably would provide better type safety, no?
We have a few places that do things like
mm/filemap.c: if (unlikely(compound_head(page)->mapping != mapping)) {
