On Wed, May 29, 2024 at 09:39:10AM +0200, Jan Beulich wrote:
> On an (old) x86 system with SRAT just covering space above 4Gb:
>
> ACPI: SRAT: Node 0 PXM 0 [mem 0x100000000-0xfffffffff] hotplug
>
> the commit referenced below leads to this NUMA configuration no longer
> being refused by a CONFIG_NUMA=y kernel (previously
>
> NUMA: nodes only cover 6144MB of your 8185MB e820 RAM. Not used.
> No NUMA configuration found
> Faking a node at [mem 0x0000000000000000-0x000000027fffffff]
>
> was seen in the log directly after the message quoted above), because of
> memblock_validate_numa_coverage() checking for NUMA_NO_NODE (only). This
> in turn led to memblock_alloc_range_nid()'s warning about MAX_NUMNODES
> triggering, followed by a NULL deref in memmap_init() when trying to
> access node 64's (NODE_SHIFT=6) node data.
>
> To compensate said change, make memblock_set_node() warn on and adjust
> a passed in value of MAX_NUMNODES, just like various other functions
> already do.
>
> Fixes: ff6c3d81f2e8 ("NUMA: optimize detection of memory with no node id assigned by firmware")
> Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
> Cc: stable@xxxxxxxxxxxxxxx
> ---
> This still leaves MAX_NUMNODES uses in various other places.
> Interestingly
> https://lore.kernel.org/lkml/20170309034415.GA16588@WeideMacBook-Pro.local/T/#t
> was a more complete patch which, for an unclear reason, looks to never
> have made it anywhere. IOW the two memblock_set_node() invocations from x86'es
> numa_init() likely also want adjusting, among others.
They do. And I think that actually would be the right fix.
The warning and nid adjustment in memblock can be added for robustness, but
the calls to memblock_set_node() in x86 should be fixed regardless.
> --- a/mm/memblock.c
> +++ b/mm/memblock.c
> @@ -1339,6 +1339,10 @@ int __init_memblock memblock_set_node(ph
> int start_rgn, end_rgn;
> int i, ret;
>
> + if (WARN_ONCE(nid == MAX_NUMNODES,
> + "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
> + nid = NUMA_NO_NODE;
> +
> ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
> if (ret)
> return ret;
--
Sincerely yours,
Mike.
