Re: [PATCH] arch_numa: Restore nid checks before registering a memblock with a node

[Mike Rapoport]

On Fri, Nov 29, 2024 at 08:42:55AM +0000, Marc Zyngier wrote:
> On Fri, 29 Nov 2024 08:24:16 +0000,
> Mike Rapoport <rppt@xxxxxxxxxx> wrote:
> > 
> > On Thu, Nov 28, 2024 at 04:52:14PM +0000, Marc Zyngier wrote:
> > > Hi Mike,
> > > 
> > > On Thu, 28 Nov 2024 07:03:33 +0000,
> > > Mike Rapoport <rppt@xxxxxxxxxx> wrote:
> > > > 
> > > > Hi Marc,
> > > > 
> > > > > diff --git a/drivers/base/arch_numa.c b/drivers/base/arch_numa.c
> > > > > index e187016764265..5457248eb0811 100644
> > > > > --- a/drivers/base/arch_numa.c
> > > > > +++ b/drivers/base/arch_numa.c
> > > > > @@ -207,7 +207,21 @@ static void __init setup_node_data(int nid, u64 start_pfn, u64 end_pfn)
> > > > >  static int __init numa_register_nodes(void)
> > > > >  {
> > > > >  	int nid;
> > > > > -
> > > > > +	struct memblock_region *mblk;
> > > > > +
> > > > > +	/* Check that valid nid is set to memblks */
> > > > > +	for_each_mem_region(mblk) {
> > > > > +		int mblk_nid = memblock_get_region_node(mblk);
> > > > > +		phys_addr_t start = mblk->base;
> > > > > +		phys_addr_t end = mblk->base + mblk->size - 1;
> > > > > +
> > > > > +		if (mblk_nid == NUMA_NO_NODE || mblk_nid >= MAX_NUMNODES) {
> > > > > +			pr_warn("Warning: invalid memblk node %d [mem %pap-%pap]\n",
> > > > > +				mblk_nid, &start, &end);
> > > > > +			return -EINVAL;
> > > > > +		}
> > > > 
> > > > We have memblock_validate_numa_coverage() that checks that amount of memory
> > > > with unset node id is less than a threshold. The loop here can be replaced
> > > > with something like
> > > > 
> > > > 	if (!memblock_validate_numa_coverage(0))
> > > > 		return -EINVAL;
> > > 
> > > Unfortunately, that doesn't seem to result in something that works
> > > (relevant extract only):
> > > 
> > > [    0.000000] NUMA: no nodes coverage for 9MB of 65516MB RAM
> > > [    0.000000] NUMA: Faking a node at [mem 0x0000000000500000-0x0000000fff0fffff]
> > > [    0.000000] NUMA: no nodes coverage for 0MB of 65516MB RAM
> > > [    0.000000] Unable to handle kernel paging request at virtual address 0000000000001d40
> > > 
> > > Any idea?
> > 
> > With 0 as the threshold the validation fails for the fake node, but it
> > should be fine with memblock_validate_numa_coverage(1)
> 
> Huh, subtle. This indeed seems to work. I'll respin the patch next week.

With the patch below memblock_validate_numa_coverage(0) should also work
and it makes more sense.

@Andrew, I can take both this and Marc's new patch via memblock tree if you
prefer.