[PATCH] mempolicy: fix div-by-zero in alloc_pages_bulk_interleave

Liu Jing posted 1 patch 3 weeks, 3 days ago
mm/mempolicy.c | 3 +++
1 file changed, 3 insertion(+), 0 deletion(-)
[PATCH] mempolicy: fix div-by-zero in alloc_pages_bulk_interleave
Posted by Liu Jing 3 weeks, 3 days ago
In alloc_pages_bulk_interleave(), nodes_weight(pol->nodes) may return
0 if all nodes in the policy mask have been offlined at runtime. The
subsequent division "nr_pages / nodes" triggers a divide-by-zero
panic.

Add a zero check returning 0 (no pages allocated) when the nodemask
is empty, consistent with the guard in
alloc_pages_bulk_weighted_interleave().

Signed-off-by: Liu Jing <liujing@cmss.chinamobile.com>
---
 mm/mempolicy.c | 3 +++
 1 file changed, 3 insertion(+), 0 deletion(-)

--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -2600,6 +2600,9 @@
 	unsigned long total_allocated = 0;
 
 	nodes = nodes_weight(pol->nodes);
+	if (!nodes)
+		return 0;
+
 	nr_pages_per_node = nr_pages / nodes;
 	delta = nr_pages - nodes * nr_pages_per_node;
 

--
2.43.0
Re: [PATCH] mempolicy: fix div-by-zero in alloc_pages_bulk_interleave
Posted by Gregory Price 3 weeks, 2 days ago
On Wed, Sep 02, 2026 at 05:02:45PM +0800, Liu Jing wrote:
> In alloc_pages_bulk_interleave(), nodes_weight(pol->nodes) may return
> 0 if all nodes in the policy mask have been offlined at runtime. The
> subsequent division "nr_pages / nodes" triggers a divide-by-zero
> panic.
> 

now i'm positive this is LLM produced without understanding the
component.

A policy cannot have an empty nodemask, it can either
1) have no nodemask (default policy), or
2) be perceived as empty during a torn-read during a policy remap event

If you are running LLMs to search for surface level bugs and try to
hot-fix them without taking time to understand the underlying component,
please at least:

1) tell us that
2) generate reproducers

these "quick fixes" based on surface level observations are going to hide
real bugs and just degrade the overall state of the system.

> Add a zero check returning 0 (no pages allocated) when the nodemask
> is empty, consistent with the guard in
> alloc_pages_bulk_weighted_interleave().
> 
> Signed-off-by: Liu Jing <liujing@cmss.chinamobile.com>
> ---
>  mm/mempolicy.c | 3 +++
>  1 file changed, 3 insertion(+), 0 deletion(-)
> 
> --- a/mm/mempolicy.c
> +++ b/mm/mempolicy.c
> @@ -2600,6 +2600,9 @@
>  	unsigned long total_allocated = 0;
>  
>  	nodes = nodes_weight(pol->nodes);
> +	if (!nodes)
> +		return 0;
> +
>  	nr_pages_per_node = nr_pages / nodes;
>  	delta = nr_pages - nodes * nr_pages_per_node;
>  
> 
> --
> 2.43.0
> 
>