On Tue, Feb 23, 2016 at 03:32:44PM -0800, Andrew Morton wrote: > On Tue, 23 Feb 2016 16:21:17 +0900 [email protected] wrote: > > > From: Joonsoo Kim <[email protected]> > > > > Success of CMA allocation largely depends on success of migration > > and key factor of it is page reference count. Until now, page reference > > is manipulated by direct calling atomic functions so we cannot follow up > > who and where manipulate it. Then, it is hard to find actual reason > > of CMA allocation failure. CMA allocation should be guaranteed to succeed > > so finding offending place is really important. > > > > In this patch, call sites where page reference is manipulated are converted > > to introduced wrapper function. This is preparation step to add tracepoint > > to each page reference manipulation function. With this facility, we can > > easily find reason of CMA allocation failure. There is no functional change > > in this patch. > > > > ... > > > > --- a/arch/mips/mm/gup.c > > +++ b/arch/mips/mm/gup.c > > @@ -64,7 +64,7 @@ static inline void get_head_page_multiple(struct page > > *page, int nr) > > { > > VM_BUG_ON(page != compound_head(page)); > > VM_BUG_ON(page_count(page) == 0); > > - atomic_add(nr, &page->_count); > > + page_ref_add(page, nr); > > Seems reasonable. Those open-coded refcount manipulations have always > bugged me.
I think so. > > The patches will be a bit of a pain to maintain but surprisingly they > apply OK at present. It's possible that by the time they hit upstream, > some direct ->_count references will still be present and it will > require a second pass to complete the conversion. In fact, the patch doesn't change direct ->_count reference for *read*. That's the reason that it is surprisingly OK at present. It's a good idea to change direct ->_count reference even for read. How about changing it in rc2 after mering this patch in rc1? > After that pass is completed I suggest we rename page._count to > something else (page.ref_count_dont_use_this_directly_you_dope?). That > way, any attempts to later add direct page._count references will > hopefully break, alerting the programmer to the new regime. Agreed. Thanks.
