Re: [PATCH] mm: move PG_slab flag to page_type

[David Hildenbrand <david@redhat.com>]

On 25.09.22 01:04, Matthew Wilcox wrote:
> On Mon, Sep 19, 2022 at 09:57:08PM +0900, Hyeonggon Yoo wrote:
>> For now, only SLAB uses _mapcount field as a number of active objects in
>> a slab, and other slab allocators do not use it. As 16 bits are enough
>> for that, use remaining 16 bits of _mapcount as page_type even when
>> SLAB is used. And then move PG_slab flag to page_type!
>>
>> Note that page_type is always placed in upper 16 bits of _mapcount to
>> avoid confusing normal _mapcount as page_type. As underflow (actually
>> I mean, yeah, overflow) is not a concern anymore, use more lower bits
>> except bit zero.
>>
>> Add more folio helpers for PAGE_TYPE_OPS() not to break existing
>> slab implementations.
>>
>> Remove PG_slab check from PAGE_FLAGS_CHECK_AT_FREE. buddy will still
>> check if _mapcount is properly set at free.
>>
>> Exclude PG_slab from hwpoison and show_page_flags() for now.
>>
>> Note that with this patch, page_mapped() and folio_mapped() always return
>> false for slab page.
> 
> This is an interesting approach.  It raises some questions.
> 
> First, you say that folio_mapped() returns false for slab pages.  That's
> only true for order-0 slab pages.  For larger pages,
> 
>          if (!folio_test_large(folio))
>                  return atomic_read(&folio->_mapcount) >= 0;
>          if (atomic_read(folio_mapcount_ptr(folio)) >= 0)
>                  return true;
> 
> so that's going to depend what folio_mapcount_ptr() aliases with.
> 
> Second, this patch changes the behaviour of PageSlab() when applied to
> tail pages.  Which raises the further question of what PageBuddy(),
> PageTable(), PageGuard() and PageIsolated() should do for multi-page
> folios, if that is even possible.

IIRC, these flags never apply on real compound pages so far. For 
example, PageBuddy() is only set on the first page of a (budy-aligned) 
free chunk of memory, and all "remaining" (tail) pages have a refcount 
of zero and don't have the flag set.

There are page tables on some systems (e.g., s390x) that span multiple 
pages (pmd_alloc_one()). I think the behavior is similar -- no compound 
page, and only the first page has the flag set.

-- 
Thanks,

David / dhildenb