Re: [PATCH 2/2] xen: don't use page->lru for ZONE_DEVICE memory

[Jason Andryuk <jandryuk@xxxxxxxxx>]

On Thu, Dec 10, 2020 at 6:40 AM Jürgen Groß <jgross@xxxxxxxx> wrote:
>
> On 10.12.20 12:14, Roger Pau Monné wrote:
> > On Tue, Dec 08, 2020 at 07:45:00AM +0100, Jürgen Groß wrote:
> >> On 07.12.20 21:48, Jason Andryuk wrote:
> >>> On Mon, Dec 7, 2020 at 8:30 AM Juergen Gross <jgross@xxxxxxxx> wrote:
> >>>>
> >>>> Commit 9e2369c06c8a18 ("xen: add helpers to allocate unpopulated
> >>>> memory") introduced usage of ZONE_DEVICE memory for foreign memory
> >>>> mappings.
> >>>>
> >>>> Unfortunately this collides with using page->lru for Xen backend
> >>>> private page caches.
> >>>>
> >>>> Fix that by using page->zone_device_data instead.
> >>>>
> >>>> Fixes: 9e2369c06c8a18 ("xen: add helpers to allocate unpopulated memory")
> >>>> Signed-off-by: Juergen Gross <jgross@xxxxxxxx>
> >>>
> >>> Would it make sense to add BUG_ON(is_zone_device_page(page)) and the
> >>> opposite as appropriate to cache_enq?
> >>
> >> No, I don't think so. At least in the CONFIG_ZONE_DEVICE case the
> >> initial list in a PV dom0 is populated from extra memory (basically
> >> the same, but not marked as zone device memory explicitly).
> >
> > I assume it's fine for us to then use page->zone_device_data even if
> > the page is not explicitly marked as ZONE_DEVICE memory?
>
> I think so, yes, as we are owner of that page and we were fine to use
> lru, too.

I think memremap_pages or devm_memremap_pages (which calls
memremap_pages) is how you mark memory as ZONE_DEVICE.  i.e. they are
explicitly marked.

memremap_pages
  memmap_init_zone_device (with ZONE_DEVICE)
    __init_single_page
      set_page_links
        set_page_zone

grep only finds a few uses of ZONE_DEVICE in the whole tree.

Regards,
Jason