Hello, Adrian
I believe the pool->node is being used as a node hint
before workqueue_init() for allocating memory. It is
useful when it is correct.
I think it is better to init it early unless there is a bug
about it in this early stage reported (on same archs).
Thanks
Lai.
On Sun, Nov 1, 2020 at 8:21 PM Adrian Huang <adrianhuang0...@gmail.com> wrote:
>
> From: Adrian Huang <ahuan...@lenovo.com>
>
> The member 'node' of worker_pool struct (per_cpu worker_pool) is
> assigned in workqueue_init_early() and workqueue_init().
> Commit 2186d9f940b6 ("workqueue: move wq_numa_init() to workqueue_init()")
> fixes an issue by moving wq_numa_init() to workqueue_init() in order
> to get the valid 'cpu to node' mapping. So, remove the redundant
> assignment in workqueue_init_early().
>
> Signed-off-by: Adrian Huang <ahuan...@lenovo.com>
> ---
> kernel/workqueue.c | 1 -
> 1 file changed, 1 deletion(-)
>
> diff --git a/kernel/workqueue.c b/kernel/workqueue.c
> index 437935e7a199..cf8c0df2410e 100644
> --- a/kernel/workqueue.c
> +++ b/kernel/workqueue.c
> @@ -5937,7 +5937,6 @@ void __init workqueue_init_early(void)
> pool->cpu = cpu;
> cpumask_copy(pool->attrs->cpumask, cpumask_of(cpu));
> pool->attrs->nice = std_nice[i++];
> - pool->node = cpu_to_node(cpu);
>
> /* alloc pool ID */
> mutex_lock(&wq_pool_mutex);
> --
> 2.17.1
>
