Re: [LSF/MM/BPF TOPIC][RFC PATCH v4 00/27] Private Memory Nodes (w/ Compressed RAM)

[Balbir Singh <balbirs@nvidia.com>]

On Sun, May 24, 2026 at 09:50:06PM -0400, Gregory Price wrote:
> On Thu, May 21, 2026 at 04:23:28PM +1000, Balbir Singh wrote:
> > On Sun, Feb 22, 2026 at 03:48:15AM -0500, Gregory Price wrote:
> > > Topic type: MM
> > > 
> > > Presenter: Gregory Price <gourry@gourry.net>
> > > 
> > > This series introduces N_MEMORY_PRIVATE, a NUMA node state for memory
> > > managed by the buddy allocator but excluded from normal allocations.
> > > 
> > > I present it with an end-to-end Compressed RAM service (mm/cram.c)
> > > that would otherwise not be possible (or would be considerably more
> > > difficult, be device-specific, and add to the ZONE_DEVICE boondoggle).
> > > 
> > 
> > Do we have updates/notes from the meeting?
> > 
> 
> I have been on leave since LSF, but I do have some notes posted:
> 
> https://lore.kernel.org/linux-mm/af9i7dkNvGGxPHzu@gourry-fedora-PF4VCD3F/
> https://lore.kernel.org/linux-mm/agYJcRgOHho8upVv@gourry-fedora-PF4VCD3F/
> 
> I will be trying to post an updated set stripped down without the GFP
> flag as a first pass w/o RFC tags and no UAPI implications so that
> device folks can play with this upstream.
> 
> I'm debating on whether to include OPS_MEMPOLICY in the initial version
> if only because it's not intuitive how it interacts with pagecache. That
> needs more time to bake.
>

It makes sense to look at it and then decide if it makes sense.

> > > 
> > > page = alloc_pages_node(nid, __GFP_PRIVATE, 0);
> > 
> > Do we want to provide kernel level control over allocation of private
> > pages, I assumed that only user space applications? I would assume
> > node affinity would be the way to do so, unless we have multiple
> > 
> 
> alloc_pages_node() is the kernel interface

I was think we wouldn't need explicit flags and that allocations would
happen from user space using __GFP_THISNODE to the node or via a nodemask
based on nodes of interest. Is there a reason to add this flag, a system
might have more than one source of N_MEMORY_PRIVATE?

> 
> > > 
> > > /* Ok but I want to do something useful with it */
> > > static const struct node_private_ops ops = {
> > >         .migrate_to     = my_migrate_to,
> > >         .folio_migrate  = my_folio_migrate,
> > >         .flags = NP_OPS_MIGRATION | NP_OPS_MEMPOLICY,
> > > };
> > > node_private_set_ops(nid, &ops);
> > >
> > 
> > Could you explain this further? Why does OPS_MIGRATION
> > and OPS_MEMPOLICY needs to be set explictly?
> >
> 
> Both of these have been removed from the upcoming version, but in this
> RFC version i was testing OPS_MIGRATION as an explicit flag that meant
> "migrate.c can touch the folios" while OPS_MEMPOLICY meant "mempolicy.c
> can touch the folios".
> 
> As it turns out, OPS_MIGRATION is not a useful filter, as it doesn't
> actually filter anything (anything using OPS_MIGRATION would also need
> its own filter flag, so better to just drop it and do per-server
> opt-ins).
> 

Thanks,
Balbir