Re: x86: WARNING: at mm/memblock.c:1339 memblock_set_node - Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead

From: Paul E. McKenney
Date: Thu Jun 06 2024 - 10:42:23 EST


On Thu, Jun 06, 2024 at 08:13:17AM +0200, Jan Beulich wrote:
> On 05.06.2024 22:48, Paul E. McKenney wrote:
> > On Wed, Jun 05, 2024 at 09:46:37PM +0200, Jan Beulich wrote:
> >> On 05.06.2024 21:07, Paul E. McKenney wrote:
> >>> On Mon, Jun 03, 2024 at 07:19:21PM +0530, Naresh Kamboju wrote:
> >>>> The following kernel warnings are noticed on x86 devices while booting
> >>>> the Linux next-20240603 tag and looks like it is expected to warn users to
> >>>> use NUMA_NO_NODE instead.
> >>>>
> >>>> Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead
> >>>>
> >>>> The following config is enabled
> >>>> CONFIG_NUMA=y
> >>>
> >>> I am seeing this as well. Is the following commit premature?
> >>>
> >>> e0eec24e2e19 ("memblock: make memblock_set_node() also warn about use of MAX_NUMNODES")
> >>>
> >>> Maybe old ACPI tables and device trees need to catch up?
> >>>
> >>> Left to myself, I would simply remove the WARN_ON_ONCE() from the above
> >>> commit, but I would guess that there is a better way.
> >>
> >> Well, the warning is issued precisely to make clear that call
> >> sites need to change. A patch to do so for the two instances
> >> on x86 that I'm aware of is already pending maintainer approval.
> >
> > Could you please point me at that patch so that I can stop repeatedly
> > reproducing those two particular issues?
>
> https://lore.kernel.org/lkml/abadb736-a239-49e4-ab42-ace7acdd4278@xxxxxxxx/

Thank you, Jan!

A quick initial test shows that this clears things up. I have started
a longer test to check for additional issues. But in the meantime
for the issues I was already seeing in the initial test:

Tested-by: Paul E. McKenney <paulmck@xxxxxxxxxx>

Thanx, Paul