There is an undocumented requirement for create_worker() that it can only be called from existing worker (aka. manager) except the first call.
The reason is that the current create_worker() queues the new worker to idle_list at first and then wake up it. But the new worker is not guaranteed to be migrated until it is waken up. Thus the wq_worker_sleeping() may see the new non-local worker from the idle_list if this block of code is not executed on the local CPU to disable wq_worker_sleeping(). Existing worker can guarantee to run on local CPU when !DISASSOCIATED, so create_worker() is required to be called from existing worker/manager only currently. But we are planning to allow create_worker() to be called out side from its workers and the requirement should be alleviated. So we exchange the order of the code, the new worker is woken up before queued to idle_list. Signed-off-by: Lai Jiangshan <la...@cn.fujitsu.com> --- kernel/workqueue.c | 7 ++++++- 1 files changed, 6 insertions(+), 1 deletions(-) diff --git a/kernel/workqueue.c b/kernel/workqueue.c index 370f947..1d44d8d 100644 --- a/kernel/workqueue.c +++ b/kernel/workqueue.c @@ -1708,8 +1708,13 @@ static struct worker *create_worker(struct worker_pool *pool) /* start the newly created worker */ spin_lock_irq(&pool->lock); worker->pool->nr_workers++; - worker_enter_idle(worker); + /* + * Wake up the worker at first and then queue it to the idle_list, + * so that it is ensued that the wq_worker_sleeping() sees the worker + * had been migrated properly when sees this worker in the idle_list. + */ wake_up_process(worker->task); + worker_enter_idle(worker); spin_unlock_irq(&pool->lock); return worker; -- 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/