Re: [PATCH v1 2/6] mm/rmap: move SetPageAnonExclusive out of __page_set_anon_rmap()

[David Hildenbrand]

On 13.09.23 16:32, Matthew Wilcox wrote:
> On Wed, Sep 13, 2023 at 02:51:09PM +0200, David Hildenbrand wrote:
> > @@ -1246,11 +1241,13 @@ void page_add_anon_rmap(struct page *page, struct vm_area_struct *vma,
> >   
> >   	if (likely(!folio_test_ksm(folio))) {
> >   		if (first)
> > -			__page_set_anon_rmap(folio, page, vma, address,
> > -					     !!(flags & RMAP_EXCLUSIVE));
> > +			__folio_set_anon(folio, vma, address,
> > +					 !!(flags & RMAP_EXCLUSIVE));
> >   		else
> >   			__page_check_anon_rmap(folio, page, vma, address);
> >   	}
> > +	if (flags & RMAP_EXCLUSIVE)
> > +		SetPageAnonExclusive(page);
>
> Won't we end up setting AnonExclusive on ksm pages, or do we make sure
> to never pass RMAP_EXCLUSIVE for ksm pages?

Not if there is a bug and someone passes RMAP_EXCLUSIVE for these. :)

Fortunately, we do have

VM_BUG_ON_PGFLAGS(!PageAnon(page) || PageKsm(page), page);

in SetPageAnonExclusive() to catch such bugs.

> Maybe better to move these last two lines inside the previous test,
> just to avoid the question.

That could end up hiding another BUG, so I'd rather let 
SetPageAnonExclusive() catch it.

--
Cheers,

David / dhildenb