Re: [PATCH v2 resend] mm/memory_hotplug: Make unpopulated zones PCP structures unreachable during hot remove

[Mel Gorman]

On Tue, Apr 13, 2021 at 11:36:08AM +0200, Vlastimil Babka wrote:
> On 4/12/21 4:08 PM, Mel Gorman wrote:
> > On Mon, Apr 12, 2021 at 02:40:18PM +0200, Vlastimil Babka wrote:
> >> On 4/12/21 2:08 PM, Mel Gorman wrote:
> >
> > the pageset structures in place would be much more straight-forward
> > assuming the structures were not allocated in the zone that is being
> > hot-removed.
> 
> I would expect this is not possible, at least for ZONE_MOVABLE, as the percpu
> allocations should be GFP_KERNEL.

True.

> And it's not realistic to expect offlining to
> succeed at all without using ZONE_MOVABLE.
> 
> AFAIK even Oscar's work on using the node to self-contain its own structures is
> only applicable to struct pages, not percpu allocations?

That I don't know as I didn't check although in general, it would be
somewhat unfortunate if per-cpu structures were remote. It wouldn't be
critical given that they'll be in cache assuming the per-cpu structures
are not straddling cache lines.

-- 
Mel Gorman
SUSE Labs