Re: [PATCH 14/16] mm/migration: fix potential invalid node access for reclaim-based migration

[Baolin Wang]

On 3/7/2022 1:14 PM, Huang, Ying wrote:
> Baolin Wang <baolin.wang@xxxxxxxxxxxxxxxxx> writes:
>
> > On 3/4/2022 5:34 PM, Miaohe Lin wrote:
> > > If we failed to setup hotplug state callbacks for mm/demotion:online in
> > > some corner cases, node_demotion will be left uninitialized. Invalid node
> > > might be returned from the next_demotion_node() when doing reclaim-based
> > > migration. Use kcalloc to allocate node_demotion to fix the issue.
> > > Fixes: ac16ec835314 ("mm: migrate: support multiple target nodes
> > > demotion")
> > > Signed-off-by: Miaohe Lin <linmiaohe@xxxxxxxxxx>
> > > ---
> > >    mm/migrate.c | 6 +++---
> > >    1 file changed, 3 insertions(+), 3 deletions(-)
> > > diff --git a/mm/migrate.c b/mm/migrate.c
> > > index 279940c0c064..7b1c0b988234 100644
> > > --- a/mm/migrate.c
> > > +++ b/mm/migrate.c
> > > @@ -2516,9 +2516,9 @@ static int __init migrate_on_reclaim_init(void)
> > >    {
> > >    	int ret;
> > >    -	node_demotion = kmalloc_array(nr_node_ids,
> > > -				      sizeof(struct demotion_nodes),
> > > -				      GFP_KERNEL);
> > > +	node_demotion = kcalloc(nr_node_ids,
> > > +				sizeof(struct demotion_nodes),
> > > +				GFP_KERNEL);
> >
> > Nit: not sure if this is worthy of this rare corner case, but I think
> > the target demotion nodes' default value should be NUMA_NO_NODE
> > instead of 0.
>
> The "nr" field of "struct demotion_nodes" should be initialized as 0.  I
> think that is checked before "nodes[]" field.

Right, but it will be confusing that if nr = 0, while the nodes[] still 
contains valid node id 0. While we are at this, why not initialize the 
node_demotion structure with a clear default value? Anyway, no strong 
opinion on this :)