Re: [PATCH v3] sched/numa: add statistics of numa balance task migration

[Venkat Rao Bagalkote <venkat88@linux.ibm.com>]

On 05/05/25 8:33 pm, Chen, Yu C wrote:
> On 5/5/2025 2:43 PM, Jain, Ayush wrote:
>>
>> Hello,
>>
>> Hitting Kernel Panic on latest-next while running rcutorture tests
>>
>> 37ff6e9a2ce3 ("Add linux-next specific files for 20250502")
>>
>> reverting this patch fixes it
>> 3b2339eeb032 
>> ("sched-numa-add-statistics-of-numa-balance-task-migration-v3")
>> https://web.git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/kernel/sched/core.c?id=3b2339eeb032627e9329daf70a4ba8cd62c9cc8d 
>>
>>
>> by looking at RIP pointer
>>
>> $ ./scripts/faddr2line vmlinux __migrate_swap_task+0x2e/0x180
>> __migrate_swap_task+0x2e/0x180:
>> count_memcg_events_mm at include/linux/memcontrol.h:987
>> (inlined by) count_memcg_events_mm at include/linux/memcontrol.h:978
>> (inlined by) __migrate_swap_task at kernel/sched/core.c:3356
>>
>> memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
>> mm->owner -> NULL
>>
>> Attaching kernel logs below:
>>
>> [ 1070.635450] rcu-torture: rcu_torture_read_exit: End of episode
>> [ 1074.047617] BUG: kernel NULL pointer dereference, address:
>> 0000000000000498
>
> Thanks Ayush,
>
> According to this address,
>    4c 8b af 50 09 00 00    mov    0x950(%rdi),%r13  <--- r13 = p->mm;
>    49 8b bd 98 04 00 00    mov    0x498(%r13),%rdi  <--- p->mm->owner
> It seems that this task to be swapped has NULL mm_struct.
>
> Does the following help?
>
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index 96db6947bc92..0cb8cc4d551d 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -3353,7 +3353,8 @@ void set_task_cpu(struct task_struct *p, 
> unsigned int new_cpu)
>  static void __migrate_swap_task(struct task_struct *p, int cpu)
>  {
>         __schedstat_inc(p->stats.numa_task_swapped);
> -       count_memcg_event_mm(p->mm, NUMA_TASK_SWAP);
> +       if (p->mm)
> +               count_memcg_event_mm(p->mm, NUMA_TASK_SWAP);
>
>         if (task_on_rq_queued(p)) {
>                 struct rq *src_rq, *dst_rq;
>

Hello Chenyu,


This issue is reported even on IBM Power servers.

Proposed fix works fine. Hence,


Tested-by: Venkat Rao Bagalkote <venkat88@linux.ibm.com>


Regards,

Venkat.

>
>
> Hi Andrew,
> May I know if we can hold this patch and not merge it for now,
> besides this regression, Libo has another comment related to
> this patch and I'll address it in next version. Sorry for
> inconvenience.
>
> thanks,
> Chenyu
>