Re: [PATCH v3 8/8] mm/page_alloc: remove a wrapper for alloc_migration_target()

[Joonsoo Kim]

2020ë 7ì 4ì (í) ìì 1:18, Vlastimil Babka <vbabka@xxxxxxx>ëì ìì:
>
> On 6/23/20 8:13 AM, js1304@xxxxxxxxx wrote:
> > From: Joonsoo Kim <iamjoonsoo.kim@xxxxxxx>
> >
> > There is a well-defined standard migration target callback.
> > Use it directly.
> >
> > Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@xxxxxxx>
>
> Acked-by: Vlastimil Babka <vbabka@xxxxxxx>
>
> But you could move this to patch 5/8 to reduce churn. And do the same with

Yes, I now realize that it is possible to make this change earlier.
However, reordering
the patches would cause additional change so I will not change the
order in the next
version. Result would be the same. :)

> mm/memory-failure.c new_page() there really, to drop the simple wrappers. Only

Okay. As you suggested below, with NUMA_NO_NODE handling, we can remove
the more wrappers. I will do it.

> new_node_page() is complex enough.
> Hm wait, new_node_page() is only called by do_migrate_range() which is only
> called by __offline_pages() with explicit test that all pages are from a single
> zone, so the nmask could also be setup just once and not per each page, making
> it possible to remove the wrapper.

I have tried this suggestion and found that code is not simpler than before.
However, there would be minor performance benefit so I will include
this change, too.

> But for new_page() you would have to define that mtc->nid == NUMA_NO_NODE means
> alloc_migrate_target() does page_to_nid(page) by itself.

Yes, I will use this suggestion.

Thanks.