Re: [PATCH mm-unstable] mm: clarify folio_set_compound_order() zero support

[John Hubbard]

On 12/8/22 14:12, Sidhartha Kumar wrote:
> On 12/8/22 2:01 PM, John Hubbard wrote:
> > On 12/8/22 13:58, Sidhartha Kumar wrote:
> > > Thanks John, Mike, Matthew, and Muchun for the feedback.
> > >
> > > To summarize this discussion and outline the next version of this patch, the changes I'll make include:
> > >
> > > 1) change the name of folio_set_compound_order() to folio_set_order()
> > > 2) change the placement of this function from mm.h to mm/internal.h
> > > 3) folio_set_order() will set both _folio_order and _folio_nr_pages and handle the zero order case correctly.
> > > 4) remove the comment about hugetlb's specific use for zero orders
> > > 5) improve the style of folio_set_order() by removing ifdefs from inside the function to doing
> > >
> > > #ifdef CONFIG_64BIT
> > >   static inline void folio_set_order(struct folio *folio,
> > >                   unsigned int order)
> > >   {
> > >       VM_BUG_ON_FOLIO(!folio_test_large(folio), folio);
> >
> > Sounds good, except for this part: why is a function named
> > folio_set_order() BUG-ing on a non-large folio? The naming
> > is still wrong, perhaps?
>
> This is because the _folio_nr_pages and _folio_order fields are part of the first tail page in the folio. folio_test_large returns if the folio is larger than one page which would be required for setting the fields.

OK, but then as I said, the name is wrong. One can either:

a) handle the non-large case, or

b) rename the function to indicate that it only works on large folios.


thanks,
--
John Hubbard
NVIDIA