OSDN Git Service

workqueue: Remove the warning in wq_worker_sleeping()
authorSebastian Andrzej Siewior <bigeasy@linutronix.de>
Fri, 27 Mar 2020 23:29:59 +0000 (00:29 +0100)
committerIngo Molnar <mingo@kernel.org>
Wed, 8 Apr 2020 09:35:20 +0000 (11:35 +0200)
The kernel test robot triggered a warning with the following race:
   task-ctx A                            interrupt-ctx B
 worker
  -> process_one_work()
    -> work_item()
      -> schedule();
         -> sched_submit_work()
           -> wq_worker_sleeping()
             -> ->sleeping = 1
               atomic_dec_and_test(nr_running)
         __schedule();                *interrupt*
                                       async_page_fault()
                                       -> local_irq_enable();
                                       -> schedule();
                                          -> sched_submit_work()
                                            -> wq_worker_sleeping()
                                               -> if (WARN_ON(->sleeping)) return
                                          -> __schedule()
                                            ->  sched_update_worker()
                                              -> wq_worker_running()
                                                 -> atomic_inc(nr_running);
                                                 -> ->sleeping = 0;

      ->  sched_update_worker()
        -> wq_worker_running()
          if (!->sleeping) return

In this context the warning is pointless everything is fine.
An interrupt before wq_worker_sleeping() will perform the ->sleeping
assignment (0 -> 1 > 0) twice.
An interrupt after wq_worker_sleeping() will trigger the warning and
nr_running will be decremented (by A) and incremented once (only by B, A
will skip it). This is the case until the ->sleeping is zeroed again in
wq_worker_running().

Remove the WARN statement because this condition may happen. Document
that preemption around wq_worker_sleeping() needs to be disabled to
protect ->sleeping and not just as an optimisation.

Fixes: 6d25be5782e48 ("sched/core, workqueues: Distangle worker accounting from rq lock")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Link: https://lkml.kernel.org/r/20200327074308.GY11705@shao2-debian
kernel/sched/core.c
kernel/workqueue.c

index f6b329b..c3d12e3 100644 (file)
@@ -4120,7 +4120,8 @@ static inline void sched_submit_work(struct task_struct *tsk)
         * it wants to wake up a task to maintain concurrency.
         * As this function is called inside the schedule() context,
         * we disable preemption to avoid it calling schedule() again
-        * in the possible wakeup of a kworker.
+        * in the possible wakeup of a kworker and because wq_worker_sleeping()
+        * requires it.
         */
        if (tsk->flags & (PF_WQ_WORKER | PF_IO_WORKER)) {
                preempt_disable();
index 3816a18..891ccad 100644 (file)
@@ -858,7 +858,8 @@ void wq_worker_running(struct task_struct *task)
  * @task: task going to sleep
  *
  * This function is called from schedule() when a busy worker is
- * going to sleep.
+ * going to sleep. Preemption needs to be disabled to protect ->sleeping
+ * assignment.
  */
 void wq_worker_sleeping(struct task_struct *task)
 {
@@ -875,7 +876,8 @@ void wq_worker_sleeping(struct task_struct *task)
 
        pool = worker->pool;
 
-       if (WARN_ON_ONCE(worker->sleeping))
+       /* Return if preempted before wq_worker_running() was reached */
+       if (worker->sleeping)
                return;
 
        worker->sleeping = 1;