Re: [PATCH v3 mm-new] fs/proc/page: avoid anon folio checks on typed pages except hugetlb

[Harry Yoo <harry.yoo@oracle.com>]

On Tue, Jul 15, 2025 at 02:31:45PM +0200, David Hildenbrand wrote:
> On 15.07.25 14:07, Harry Yoo wrote:
> > folio_test_anon() and folio_test_ksm() may return false positives when
> > invoked on typed pages (except hugetlb folios), because lower bits of
> > folio->mapping may be set even when they do not indicate
> > FOLIO_MAPPING_* flags.
> > 
> > This leads to tools/mm/page-types reporting pages with
> > KPF_SLAB, KPF_ANON and KPF_KSM (with flags, page-counts, MB omitted):
> >    $ sudo ./page-types | grep slab
> >    _______S___________________________________   slab
> >    _______S____a________x_____________________   slab,anonymous,ksm
> > 
> > Currently, and going forward (The New York interpretation),
> > typed pages except hugetlb do not have FOLIO_MAPPING_* flags.
> > In the future, they won't even cast to folios.
> > 
> > For now, avoid checking FOLIO_MAPPING_* flags on pages if they are
> > typed pages that are not hugetlb folios.
> > 
> > Update the comment in FOLIO_MAPPING_* flags accordingly.
> > 
> > Fixes: 130d4df57390 ("mm/sl[au]b: rearrange struct slab fields to allow larger rcu_head")
> > Suggested-by: Vlastimil Babka <vbabka@suse.cz>
> > Signed-off-by: Harry Yoo <harry.yoo@oracle.com>
> > ---
> > 
> > v2->v3: Did not introduce folio_has_mapcount() per David's suggestion
> > v2: https://lore.kernel.org/linux-mm/20250707120740.4413-1-harry.yoo@oracle.com
> > 
> >   fs/proc/page.c             | 19 +++++++++++--------
> >   include/linux/page-flags.h | 15 +++++++++------
> >   2 files changed, 20 insertions(+), 14 deletions(-)
> > 
> > diff --git a/fs/proc/page.c b/fs/proc/page.c
> > index 0cdc78c0d23f..07582dbdef45 100644
> > --- a/fs/proc/page.c
> > +++ b/fs/proc/page.c
> > @@ -148,18 +148,21 @@ u64 stable_page_flags(const struct page *page)
> >   	folio = page_folio(page);
> >   	k = folio->flags;
> > -	mapping = (unsigned long)folio->mapping;
> > -	is_anon = mapping & FOLIO_MAPPING_ANON;
> >   	/*
> >   	 * pseudo flags for the well known (anonymous) memory mapped pages
> >   	 */
> > -	if (page_mapped(page))
> > -		u |= 1 << KPF_MMAP;
> > -	if (is_anon) {
> > -		u |= 1 << KPF_ANON;
> > -		if (mapping & FOLIO_MAPPING_KSM)
> > -			u |= 1 << KPF_KSM;
> > +	if (!page_has_type(&folio->page) || folio_test_hugetlb(folio)) {
> > +		mapping = (unsigned long)folio->mapping;
> > +		is_anon = mapping & FOLIO_MAPPING_ANON;
> > +
> > +		if (page_mapped(page))
> > +			u |= 1 << KPF_MMAP;
> 
> Note: Luiz switches to folio_mapped() in his patch that is in mm-new.

Right, but in mm-new Luiz's patch already depends on my v2 patch,
which still leaves me confused about what I should do... :)

> > +		if (is_anon) {
> > +			u |= 1 << KPF_ANON;
> > +			if (mapping & FOLIO_MAPPING_KSM)
> > +				u |= 1 << KPF_KSM;
> 
> Can we just switch to folio_test_anon() and folio_test_ksm() ?
> 
> I don't really see a reason to not do that. Willy converted these
> checks from page -> mapping, but we should really just use the folio_test_* functions
> I think.
> 
> And looking at it, I think Willy introduced an issue in:
> 
> commit dee3d0bef2b00772be430425832ead6aa9d707f9
> Author: Matthew Wilcox (Oracle) <willy@infradead.org>
> Date:   Tue Mar 26 17:10:32 2024 +0000
> 
>     proc: rewrite stable_page_flags()
> 
> 
> We replaced
> 
> 	if (PageKsm(page))
> 
> essentially by
> 
> 	if (mapping & PAGE_MAPPING_KSM)
> 
> But
> 
> 	#define PAGE_MAPPING_KSM       (PAGE_MAPPING_ANON | PAGE_MAPPING_ANON_KSM)
> 
> So wouldn't we just indicate *all* anon pages as ... KSM pages?

Oops, I didn't notice that!

I think you're right. I don't see any problem with going back to
using folio_test_anon/ksm().

> > +		}
> >   	}
> >   	/*
> > diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
> > index 8e4d6eda8a8d..26ae4c7cf8dd 100644
> > --- a/include/linux/page-flags.h
> > +++ b/include/linux/page-flags.h
> > @@ -706,12 +706,15 @@ PAGEFLAG_FALSE(VmemmapSelfHosted, vmemmap_self_hosted)
> >    * address_space which maps the folio from disk; whereas "folio_mapped"
> >    * refers to user virtual address space into which the folio is mapped.
> >    *
> > - * For slab pages, since slab reuses the bits in struct page to store its
> > - * internal states, the folio->mapping does not exist as such, nor do
> > - * these flags below.  So in order to avoid testing non-existent bits,
> > - * please make sure that folio_test_slab(folio) actually evaluates to
> > - * false before calling the following functions (e.g., folio_test_anon).
> > - * See mm/slab.h.
> > + * For certain typed pages like slabs, since they reuse bits in struct page
> > + * to store internal states, folio->mapping does not point to a valid
> > + * mapping, nor do these flags exist. To avoid testing non-existent bits,
> > + * make sure
> 
> I would write that explicitly:
> 
> that you are dealing with an actual folio: either page has no type or the type indicates a folio (hugetlb).

Will adjust, thanks!

-- 
Cheers,
Harry / Hyeonggon