We fixed the major cases when the numa mapping is changed. We still have the assumption that when the node<->cpu mapping is changed the original node is offline, and the current code of memory-hutplug also prove this.
This assumption might be changed in future and the orig_node is still online in some cases. And in these cases, the cpumask of the pwqs of the orig_node still contains the onlining CPU which is a CPU of another node, and the worker may run on the onlining CPU (aka run on the wrong node). So we drop this assumption and make the code calls wq_update_unbound_numa() to update the affinity in this case. Cc: Tejun Heo <t...@kernel.org> Cc: Yasuaki Ishimatsu <isimatu.yasu...@jp.fujitsu.com> Cc: "Gu, Zheng" <guz.f...@cn.fujitsu.com> Cc: tangchen <tangc...@cn.fujitsu.com> Cc: Hiroyuki KAMEZAWA <kamezawa.hir...@jp.fujitsu.com> Signed-off-by: Lai Jiangshan <la...@cn.fujitsu.com> --- kernel/workqueue.c | 15 +++++++++++++++ 1 files changed, 15 insertions(+), 0 deletions(-) diff --git a/kernel/workqueue.c b/kernel/workqueue.c index 7fbabf6..29a96c3 100644 --- a/kernel/workqueue.c +++ b/kernel/workqueue.c @@ -4007,6 +4007,21 @@ static void wq_update_numa_mapping(int cpu) if (pool->node != node) pool->node = node; } + + /* Test whether we hit the case where orig_node is still online */ + if (orig_node != NUMA_NO_NODE && + !cpumask_empty(cpumask_of_node(orig_node))) { + struct workqueue_struct *wq; + cpu = cpumask_any(cpumask_of_node(orig_node)); + + /* + * the pwqs of the orig_node are still allowed on the onlining + * CPU but which is belong to new_node, update NUMA affinity + * for orig_node. + */ + list_for_each_entry(wq, &workqueues, list) + wq_update_unbound_numa(wq, cpu, true); + } } static int alloc_and_link_pwqs(struct workqueue_struct *wq) -- 1.7.4.4 -- 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/