On 12/15/2014 07:16 PM, Kamezawa Hiroyuki wrote:
> The percpu workqueue pool are persistend and never be freed.
> But cpu<->node relationship can be changed by cpu hotplug and pool->node
> can point to an offlined node.
>
> If pool->node points to an offlined node,
> following allocation failure can happen.
> ==
> SLUB: Unable to allocate memory on node 2 (gfp=0x80d0)
> cache: kmalloc-192, object size: 192, buffer size: 192, default
> order:
> 1, min order: 0
> node 0: slabs: 6172, objs: 259224, free: 245741
> node 1: slabs: 3261, objs: 136962, free: 127656
> ==
>
> This patch clears per-cpu workqueue pool's node affinity at
> cpu offlining and restore it at cpu onlining.
>
> Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hir...@jp.fujitsu.com>
> ---
> kernel/workqueue.c | 11 ++++++++++-
> 1 file changed, 10 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/workqueue.c b/kernel/workqueue.c
> index 7809154..2fd0bd7 100644
> --- a/kernel/workqueue.c
> +++ b/kernel/workqueue.c
> @@ -4586,6 +4586,11 @@ static int workqueue_cpu_up_callback(struct
> notifier_block *nfb,
> case CPU_DOWN_FAILED:
> case CPU_ONLINE:
> mutex_lock(&wq_pool_mutex);
> + /*
> + * now cpu <-> node info is established. update numa node
> + */
> + for_each_cpu_worker_pool(pool, cpu)
> + pool->node = cpu_to_node(cpu);
>
> for_each_pool(pool, pi) {
> mutex_lock(&pool->attach_mutex);
> @@ -4619,6 +4624,7 @@ static int workqueue_cpu_down_callback(struct
> notifier_block *nfb,
> int cpu = (unsigned long)hcpu;
> struct work_struct unbind_work;
> struct workqueue_struct *wq;
> + struct worker_pool *pool;
>
> switch (action & ~CPU_TASKS_FROZEN) {
> case CPU_DOWN_PREPARE:
> @@ -4626,10 +4632,13 @@ static int workqueue_cpu_down_callback(struct
> notifier_block *nfb,
> INIT_WORK_ONSTACK(&unbind_work, wq_unbind_fn);
> queue_work_on(cpu, system_highpri_wq, &unbind_work);
>
> - /* update NUMA affinity of unbound workqueues */
> mutex_lock(&wq_pool_mutex);
> + /* update NUMA affinity of unbound workqueues */
> list_for_each_entry(wq, &workqueues, list)
> wq_update_unbound_numa(wq, cpu, false);
> + /* clear per-cpu workqueues's numa affinity. */
> + for_each_cpu_worker_pool(pool, cpu)
> + pool->node = NUMA_NO_NODE; /* restored at online */
If the node is still online, it is better to keep the original pool->node.
> mutex_unlock(&wq_pool_mutex);
>
> /* wait for per-cpu unbinding to finish */
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
