Re: [PATCH v2 2/2] mm/memcg: set memcg when split page

[Matthew Wilcox <willy@infradead.org>]

On Thu, Mar 11, 2021 at 10:21:39AM -0500, Johannes Weiner wrote:
> On Thu, Mar 11, 2021 at 09:37:02AM +0100, Michal Hocko wrote:
> > Johannes, Hugh,
> > 
> > what do you think about this approach? If we want to stick with
> > split_page approach then we need to update the missing place Matthew has
> > pointed out.
> 
> I find the __free_pages() code quite tricky as well. But for that
> reason I would actually prefer to initiate the splitting in there,
> since that's the place where we actually split the page, rather than
> spread the handling of this situation further out.

Mmm.  The thing is, we don't actually split the page because it was
never compound.  I don't know whether anybody actually does this,
but it's legitimate to write:

	struct page *p = alloc_pages(GFP_KERNEL, 2);

	free_unref_page(p + 1);
	free_unref_page(p + 3);
	free_unref_page(p + 2);
	__free_page(p);

The good news is that I recently made free_unref_page() local to
mm/internal.h, so we don't need to worry about device drivers doing this.
As far as I can tell, we don't have any exposure to this kind of thing
today through functions exported from mm, but I might have missed
something.

I'd really like to get rid of non-compound high-order pages.  Slab,
filesystems and anonymous memory all use compound pages.  I think
it's just crusty old device drivers that don't.  And alloc_pages_exact(),
of course, but that's kind of internal.

> The race condition shouldn't be hot, so I don't think we need to be as
> efficient about setting page->memcg_data only on the higher-order
> buddies as in Willy's scratch patch. We can call split_page_memcg(),
> which IMO should actually help document what's happening to the page.

I'm cool with that.  I agree, this is not a performance case!

> I think that function could also benefit a bit more from step-by-step
> documentation about what's going on. The kerneldoc is helpful, but I
> don't think it does justice to how tricky this race condition is.

Always good to have other people read over your explanation ...
the kernel-doc could probably be simplified as a result.