[PATCH v2] sched/topology: Fix for_each_node_numadist() lockup with !CONFIG_NUMA

[PATCH v2] sched/topology: Fix for_each_node_numadist() lockup with !CONFIG_NUMA

[Andrea Righi]

for_each_node_numadist() can lead to hard lockups on kernels built
without CONFIG_NUMA. For instance, the following was triggered by
sched_ext:

  watchdog: CPU5: Watchdog detected hard LOCKUP on cpu 5
  ...
  RIP: 0010:_find_first_and_bit+0x8/0x60
  ...
  Call Trace:
  <TASK>
   cpumask_any_and_distribute+0x49/0x80
   pick_idle_cpu_in_node+0xcf/0x140
   scx_bpf_pick_idle_cpu_node+0xaa/0x110
   bpf_prog_16ee5b1f077af006_pick_idle_cpu+0x57f/0x5de
   bpf_prog_df2ce5cfac58ce09_bpfland_select_cpu+0x37/0xf4
   bpf__sched_ext_ops_select_cpu+0x4b/0xb3

This happens because nearest_node_nodemask() always returns NUMA_NO_NODE
(-1) when CONFIG_NUMA is disabled, causing the loop to never terminate,
as the condition node >= MAX_NUMNODES is never satisfied.

Prevent this by providing a stub implementation based on
for_each_node_mask() when CONFIG_NUMA is disabled, which can safely
processes the single available node while still honoring the unvisited
nodemask.

Fixes: f09177ca5f242 ("sched/topology: Introduce for_each_node_numadist() iterator")
Signed-off-by: Andrea Righi <arighi@nvidia.com>
---
 include/linux/topology.h | 5 +++++
 1 file changed, 5 insertions(+)

Changes in v2:
 - Provide a stub implementation for the !CONFIG_NUMA case
 - Link to v1: https://lore.kernel.org/all/20250603080402.170601-1-arighi@nvidia.com/

diff --git a/include/linux/topology.h b/include/linux/topology.h
index 33b7fda97d390..97c4f5fc75038 100644
--- a/include/linux/topology.h
+++ b/include/linux/topology.h
@@ -304,12 +304,17 @@ sched_numa_hop_mask(unsigned int node, unsigned int hops)
  *
  * Requires rcu_lock to be held.
  */
+#ifdef CONFIG_NUMA
 #define for_each_node_numadist(node, unvisited)					\
 	for (int __start = (node),						\
 	     (node) = nearest_node_nodemask((__start), &(unvisited));		\
 	     (node) < MAX_NUMNODES;						\
 	     node_clear((node), (unvisited)),					\
 	     (node) = nearest_node_nodemask((__start), &(unvisited)))
+#else
+#define for_each_node_numadist(node, unvisited)					\
+	for_each_node_mask((node), (unvisited))
+#endif
 
 /**
  * for_each_numa_hop_mask - iterate over cpumasks of increasing NUMA distance
-- 
2.49.0

Re: [PATCH v2] sched/topology: Fix for_each_node_numadist() lockup with !CONFIG_NUMA

[Yury Norov]

On Mon, Jun 09, 2025 at 01:35:36PM +0200, Andrea Righi wrote:
> for_each_node_numadist() can lead to hard lockups on kernels built
> without CONFIG_NUMA. For instance, the following was triggered by
> sched_ext:
> 
>   watchdog: CPU5: Watchdog detected hard LOCKUP on cpu 5
>   ...
>   RIP: 0010:_find_first_and_bit+0x8/0x60
>   ...
>   Call Trace:
>   <TASK>
>    cpumask_any_and_distribute+0x49/0x80
>    pick_idle_cpu_in_node+0xcf/0x140
>    scx_bpf_pick_idle_cpu_node+0xaa/0x110
>    bpf_prog_16ee5b1f077af006_pick_idle_cpu+0x57f/0x5de
>    bpf_prog_df2ce5cfac58ce09_bpfland_select_cpu+0x37/0xf4
>    bpf__sched_ext_ops_select_cpu+0x4b/0xb3
> 
> This happens because nearest_node_nodemask() always returns NUMA_NO_NODE
> (-1) when CONFIG_NUMA is disabled, causing the loop to never terminate,
> as the condition node >= MAX_NUMNODES is never satisfied.
> 
> Prevent this by providing a stub implementation based on
> for_each_node_mask() when CONFIG_NUMA is disabled, which can safely
> processes the single available node while still honoring the unvisited
> nodemask.
> 
> Fixes: f09177ca5f242 ("sched/topology: Introduce for_each_node_numadist() iterator")
> Signed-off-by: Andrea Righi <arighi@nvidia.com>
> ---
>  include/linux/topology.h | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> Changes in v2:
>  - Provide a stub implementation for the !CONFIG_NUMA case
>  - Link to v1: https://lore.kernel.org/all/20250603080402.170601-1-arighi@nvidia.com/
> 
> diff --git a/include/linux/topology.h b/include/linux/topology.h
> index 33b7fda97d390..97c4f5fc75038 100644
> --- a/include/linux/topology.h
> +++ b/include/linux/topology.h
> @@ -304,12 +304,17 @@ sched_numa_hop_mask(unsigned int node, unsigned int hops)
>   *
>   * Requires rcu_lock to be held.
>   */
> +#ifdef CONFIG_NUMA

While there, can you expand this optimization to MAX_NUMNODES == 1
case?
        #if defined(CONFIG_NUMA) && (MAX_NUMNODES > 1)

With that:

Acked-by: Yury Norov [NVIDIA] <yury.norov@gmail.com>

Thanks,
Yury

>  #define for_each_node_numadist(node, unvisited)					\
>  	for (int __start = (node),						\
>  	     (node) = nearest_node_nodemask((__start), &(unvisited));		\
>  	     (node) < MAX_NUMNODES;						\
>  	     node_clear((node), (unvisited)),					\
>  	     (node) = nearest_node_nodemask((__start), &(unvisited)))
> +#else
> +#define for_each_node_numadist(node, unvisited)					\
> +	for_each_node_mask((node), (unvisited))
> +#endif
>  
>  /**
>   * for_each_numa_hop_mask - iterate over cpumasks of increasing NUMA distance
> -- 
> 2.49.0

Re: [PATCH v2] sched/topology: Fix for_each_node_numadist() lockup with !CONFIG_NUMA

[Andrea Righi]

Hi Yury,

On Mon, Jun 09, 2025 at 11:03:29AM -0400, Yury Norov wrote:
> On Mon, Jun 09, 2025 at 01:35:36PM +0200, Andrea Righi wrote:
> > for_each_node_numadist() can lead to hard lockups on kernels built
> > without CONFIG_NUMA. For instance, the following was triggered by
> > sched_ext:
> > 
> >   watchdog: CPU5: Watchdog detected hard LOCKUP on cpu 5
> >   ...
> >   RIP: 0010:_find_first_and_bit+0x8/0x60
> >   ...
> >   Call Trace:
> >   <TASK>
> >    cpumask_any_and_distribute+0x49/0x80
> >    pick_idle_cpu_in_node+0xcf/0x140
> >    scx_bpf_pick_idle_cpu_node+0xaa/0x110
> >    bpf_prog_16ee5b1f077af006_pick_idle_cpu+0x57f/0x5de
> >    bpf_prog_df2ce5cfac58ce09_bpfland_select_cpu+0x37/0xf4
> >    bpf__sched_ext_ops_select_cpu+0x4b/0xb3
> > 
> > This happens because nearest_node_nodemask() always returns NUMA_NO_NODE
> > (-1) when CONFIG_NUMA is disabled, causing the loop to never terminate,
> > as the condition node >= MAX_NUMNODES is never satisfied.
> > 
> > Prevent this by providing a stub implementation based on
> > for_each_node_mask() when CONFIG_NUMA is disabled, which can safely
> > processes the single available node while still honoring the unvisited
> > nodemask.
> > 
> > Fixes: f09177ca5f242 ("sched/topology: Introduce for_each_node_numadist() iterator")
> > Signed-off-by: Andrea Righi <arighi@nvidia.com>
> > ---
> >  include/linux/topology.h | 5 +++++
> >  1 file changed, 5 insertions(+)
> > 
> > Changes in v2:
> >  - Provide a stub implementation for the !CONFIG_NUMA case
> >  - Link to v1: https://lore.kernel.org/all/20250603080402.170601-1-arighi@nvidia.com/
> > 
> > diff --git a/include/linux/topology.h b/include/linux/topology.h
> > index 33b7fda97d390..97c4f5fc75038 100644
> > --- a/include/linux/topology.h
> > +++ b/include/linux/topology.h
> > @@ -304,12 +304,17 @@ sched_numa_hop_mask(unsigned int node, unsigned int hops)
> >   *
> >   * Requires rcu_lock to be held.
> >   */
> > +#ifdef CONFIG_NUMA
> 
> While there, can you expand this optimization to MAX_NUMNODES == 1
> case?
>         #if defined(CONFIG_NUMA) && (MAX_NUMNODES > 1)

Makes sense, will send a v3, thanks!

-Andrea

> 
> With that:
> 
> Acked-by: Yury Norov [NVIDIA] <yury.norov@gmail.com>
> 
> Thanks,
> Yury
> 
> >  #define for_each_node_numadist(node, unvisited)					\
> >  	for (int __start = (node),						\
> >  	     (node) = nearest_node_nodemask((__start), &(unvisited));		\
> >  	     (node) < MAX_NUMNODES;						\
> >  	     node_clear((node), (unvisited)),					\
> >  	     (node) = nearest_node_nodemask((__start), &(unvisited)))
> > +#else
> > +#define for_each_node_numadist(node, unvisited)					\
> > +	for_each_node_mask((node), (unvisited))
> > +#endif
> >  
> >  /**
> >   * for_each_numa_hop_mask - iterate over cpumasks of increasing NUMA distance
> > -- 
> > 2.49.0