Re: [PATCH v1 3/5] mm/page_alloc: always move pages to the tail of the freelist in unset_migratetype_isolate()

[Michal Hocko]

On Fri 02-10-20 17:20:09, David Hildenbrand wrote:
> On 02.10.20 15:24, Michal Hocko wrote:
> > On Mon 28-09-20 20:21:08, David Hildenbrand wrote:
> >> Page isolation doesn't actually touch the pages, it simply isolates
> >> pageblocks and moves all free pages to the MIGRATE_ISOLATE freelist.
> >>
> >> We already place pages to the tail of the freelists when undoing
> >> isolation via __putback_isolated_page(), let's do it in any case
> >> (e.g., if order <= pageblock_order) and document the behavior.
> >>
> >> Add a "to_tail" parameter to move_freepages_block() but introduce a
> >> a new move_to_free_list_tail() - similar to add_to_free_list_tail().
> >>
> >> This change results in all pages getting onlined via online_pages() to
> >> be placed to the tail of the freelist.
> > 
> > Is there anything preventing to do this unconditionally? Or in other
> > words is any of the existing callers of move_freepages_block benefiting
> > from adding to the head?
> 
> 1. mm/page_isolation.c:set_migratetype_isolate()
> 
> We move stuff to the MIGRATE_ISOLATE list, we don't care about the order
> there.
> 
> 2. steal_suitable_fallback():
> 
> I don't think we care too much about the order when already stealing
> pageblocks ... and the freelist is empty I guess?
> 
> 3. reserve_highatomic_pageblock()/unreserve_highatomic_pageblock()
> 
> Not sure if we really care.

Honestly, I have no idea. I can imagine that some atomic high order
workloads (e.g. in net) might benefit from cache line hot pages but I am
not sure this is really observable. If yes it would likely be better to
have this documented than relying on wild guess. If we do not have any
evidence then I would vote for simplicity first and go with
unconditional add_to_tail which would simply your patch a bit.

Maybe Vlastimil or Mel would have a better picture.

-- 
Michal Hocko
SUSE Labs