Re: [RFC PATCH v3 0/8] mm,numa: N_PRIVATE node isolation for device-managed memory

[Gregory Price <gourry@gourry.net>]

On Mon, Jan 12, 2026 at 07:12:10PM -0800, dan.j.williams@intel.com wrote:
> Gregory Price wrote:
> > On Mon, Jan 12, 2026 at 05:17:53PM -0800, dan.j.williams@intel.com wrote:
> > > 
> > > I think what Balbir is saying is that the _PUBLIC is implied and can be
> > > omitted. It is true that N_MEMORY[_PUBLIC] already indicates multi-zone
> > > support. So N_MEMORY_PRIVATE makes sense to me as something that it is
> > > distinct from N_{HIGH,NORMAL}_MEMORY which are subsets of N_MEMORY.
> > > Distinct to prompt "go read the documentation to figure out why this
> > > thing looks not like the others".
> > 
> > Ah, ack.  Will update for v4 once i give some thought to the compression
> > stuff and the cgroups notes.
> > 
> > I would love if the ZONE_DEVICE folks could also chime in on whether the
> > callback structures for pgmap and hmm might be re-usable here, but might
> > take a few more versions to get the attention of everyone.
> 
> page->pgmap clobbers page->lru, i.e. they share the same union, so you
> could not directly use the current ZONE_DEVICE scheme. That is because
> current ZONE_DEVICE scheme needs to support ZONE_DEVICE mixed with
> ZONE_NORMAL + ZONE_MOVABLE in the same node.
> 
> However, with N_MEMORY_PRIVATE effectively enabling a "node per device"
> construct, you could move 'struct dev_pagemap' to node scope. I.e.
> rather than annotate each page with which device it belongs teach
> pgmap->ops callers to consider that the dev_pagemap instance may come
> from the node instead.

Hmmmmmmm... this is interesting.

should be able to do that cleanly with page_pgmap() and/or folio_pgmap()
and update direct accessors.

probably we'd want mildly different patterns for N_PRIVATE that does
something like

if (is_private_page(page)) {
	... send to private router ...
}

bool is_private_page(page) {
	pgdat = NODE_DATA(page_to_nid(page));
	return pgdat && pgdat->pgmap;

	/* or this, but seems less efficient */
	return node_state(page_to_nid, N_PRIVATE);
}

Then we can add all the callbacks to pgmap instead of dumping them in
node.c.  Shouldn't affect any existing users, since this doesn't
intersect with ZONE_DEVICE.

Technically you COULD have ZONE_DEVICE in N_PRIVATE, but that would be
per-page pgmap, and probably you'd have to have the private router
handle the is_device_page() pattern like everyone else does.

(Seems pointless though, feels like N_PRIVATE replaces ZONE_DEVICE for
 some use cases)

~Gregory