Re: [PATCH v3 2/3] mm: move folio LRU code out of swap

[Lorenzo Stoakes]

On Wed, Jun 03, 2026 at 06:49:16PM +0800, Baoquan He wrote:
> On 06/02/26 at 04:57pm, Matthew Wilcox wrote:
> > On Tue, Jun 02, 2026 at 09:01:16AM +0800, Jianyue Wu wrote:
> > > Rename mm/swap.c to mm/folio_lru.c so the filename better matches
> > > the code's main responsibility.
> > >
> > > This keeps the implementation split from swap-specific code without
> > > changing the published LRU helper interfaces.
> >
> > At the same time, we have Tal trying to split folio_wait.c out
> > of filemap.c.  https://lore.kernel.org/linux-mm/20260520-filemap-split-v1-0-c36ddc2b6cf2@xxxxxxxxxxxx/
> >
> > I wonder if instead of folio_lru.c and folio_wait.c, we shouldn't simply
> > have folio.c.  Tal has 700 lines in his diffstat, you'll have about 1100
> > lines, so that's about 1800 llines which seems like a good size.
> > There's some functions from mm/util.c that would fit nicely too --
> > folio_mapping(), folio_anon_vma(), folio_copy(), flush_dcache_folio()
>
> I think folio_lru.c and folio_wait.c are better than a single folio.c.
>
> A single folio.c with all of them would have no clear boundary for what
> belongs there and what doesn't. For comparison, page_alloc.c isn't called
> page.c (we have files called mm/page*.c).

Well page_alloc.c doesn't only contain things that operate on struct page *
right?

Whereas folio.c is strictly operations literally performed on folios which seems
like a good match to me.

I find folio_lru.c and folio_wait.c far more difficult to figure out what goes
where and there'll be a bunch of miscellaneous stuff all over the place.

I vote for folio.c.

>
> >
> > Other people have thoughts?
> >

Thanks, Lorenzo