Re: [PATCH v9 3/8] mm: Move set/get_pcppage_migratetype to mmzone.h
From: Alexander Duyck
Date: Mon Sep 09 2019 - 14:12:43 EST
On Mon, 2019-09-09 at 11:01 -0700, Alexander Duyck wrote:
> On Mon, 2019-09-09 at 12:56 +0300, Kirill A. Shutemov wrote:
> > On Sat, Sep 07, 2019 at 10:25:28AM -0700, Alexander Duyck wrote:
> > > From: Alexander Duyck <alexander.h.duyck@xxxxxxxxxxxxxxx>
> > >
> > > In order to support page reporting it will be necessary to store and
> > > retrieve the migratetype of a page. To enable that I am moving the set and
> > > get operations for pcppage_migratetype into the mm/internal.h header so
> > > that they can be used outside of the page_alloc.c file.
> > >
> > > Reviewed-by: Dan Williams <dan.j.williams@xxxxxxxxx>
> > > Signed-off-by: Alexander Duyck <alexander.h.duyck@xxxxxxxxxxxxxxx>
> >
> > I'm not sure that it's great idea to export this functionality beyond
> > mm/page_alloc.c without any additional safeguards. How would we avoid to
> > messing with ->index when the page is not in the right state of its
> > life-cycle. Can we add some VM_BUG_ON()s here?
>
> I am not sure what we would need to check on though. There are essentially
> 2 cases where we are using this. The first is the percpu page lists so the
> value is set either as a result of __rmqueue_smallest or
> free_unref_page_prepare. The second one which hasn't been added yet is for
> the Reported pages case which I add with patch 6.
>
> When I use it for page reporting I am essentially using the Reported flag
> to identify what pages in the buddy list will have this value set versus
> those that may not. I didn't explicitly define it that way, but that is
> how I am using it so that the value cannot be trusted unless the Reported
> flag is set.
I guess the alternative would be to just treat the ->index value as the
index within the boundary array, and not use the per-cpu list functions.
Doing that might make things a bit more clear since all we are really
doing is storing the index into the boundary list the page is contained
in. I could probably combine the value of order and migratetype and save
myself a few cycles in the process by just saving the index into the array
directly.