Re: [PATCH v6 20/37] mm: fix non-compound multi-order memory accounting in __free_pages

[Suren Baghdasaryan]

On Thu, Mar 21, 2024 at 10:19 AM Suren Baghdasaryan <surenb@xxxxxxxxxx> wrote:
>
> On Thu, Mar 21, 2024 at 10:04 AM Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote:
> >
> > On Thu, Mar 21, 2024 at 04:48:53PM +0000, Matthew Wilcox wrote:
> > > On Thu, Mar 21, 2024 at 09:36:42AM -0700, Suren Baghdasaryan wrote:
> > > > +++ b/mm/page_alloc.c
> > > > @@ -4700,12 +4700,15 @@ void __free_pages(struct page *page, unsigned int order)
> > > >  {
> > > >     /* get PageHead before we drop reference */
> > > >     int head = PageHead(page);
> > > > +   struct alloc_tag *tag = pgalloc_tag_get(page);
> > > >
> > > >     if (put_page_testzero(page))
> > > >             free_the_page(page, order);
> > > > -   else if (!head)
> > > > +   else if (!head) {
> > > > +           pgalloc_tag_sub_pages(tag, (1 << order) - 1);
> > > >             while (order-- > 0)
> > > >                     free_the_page(page + (1 << order), order);
> > > > +   }
> > >
> > > Why do you need these new functions instead of just:
> > >
> > > +     else if (!head) {
> > > +             pgalloc_tag_sub(page, (1 << order) - 1);
> > >               while (order-- > 0)
> > >                       free_the_page(page + (1 << order), order);
> > > +     }
> >
> > Actually, I'm not sure this is safe (I don't fully understand codetags,
> > so it may be safe).  What can happen is that the put_page() can come in
> > before the pgalloc_tag_sub(), and then that page can be allocated again.
> > Will that cause confusion?

I indirectly answered your question in the reason #2 but to be clear,
we obtain codetag before we do put_page() here, therefore it's valid.
If another page is allocated and it points to the same codetag, then
it will operate on the same codetag per-cpu counters and that should
not be a problem.

>
> So, there are two reasons I unfortunately can't reuse pgalloc_tag_sub():
>
> 1. We need to subtract `bytes` counter from the codetag but not the
> `calls` counter, otherwise the final accounting will be incorrect.
> This is because we effectively allocated multiple pages with one call
> but freeing them with separate calls here. pgalloc_tag_sub_pages()
> subtracts bytes but keeps calls counter the same. I mentioned this in
> here: https://lore.kernel.org/all/CAJuCfpEgh1OiYNE_uKG-BqW2x97sOL9+AaTX4Jct3=WHzAv+kg@xxxxxxxxxxxxxx/
> 2. The codetag object itself is stable, it's created at build time.
> The exception is when we unload modules and the codetag section gets
> freed but during module unloading we check that all module codetags
> are not referenced anymore and we prevent unloading this section if
> any of them are still referenced (should not normally happen). That
> said, the reference to the codetag (in this case from the page_ext)
> might change from under us and we have to make sure it's valid. We
> ensure that here by getting the codetag itself with pgalloc_tag_get()
> *before* calling put_page_testzero(), which ensures its stability.
>
> >