sched/core: Fix an SMP ordering race in try_to_wake_up() vs. schedule()

Peter Zijlstra · Ingo Molnar · commit ecf7d01c229d · 2015-12-04T10:26:43.000+01:00
Oleg noticed that its possible to falsely observe p-&gt;on_cpu == 0 such
that we'll prematurely continue with the wakeup and effectively run p on
two CPUs at the same time.

Even though the overlap is very limited; the task is in the middle of
being scheduled out; it could still result in corruption of the
scheduler data structures.

        CPU0                            CPU1

        set_current_state(...)

        &lt;preempt_schedule&gt;
          context_switch(X, Y)
            prepare_lock_switch(Y)
              Y-&gt;on_cpu = 1;
            finish_lock_switch(X)
              store_release(X-&gt;on_cpu, 0);

                                        try_to_wake_up(X)
                                          LOCK(p-&gt;pi_lock);

                                          t = X-&gt;on_cpu; // 0

          context_switch(Y, X)
            prepare_lock_switch(X)
              X-&gt;on_cpu = 1;
            finish_lock_switch(Y)
              store_release(Y-&gt;on_cpu, 0);
        &lt;/preempt_schedule&gt;

        schedule();
          deactivate_task(X);
          X-&gt;on_rq = 0;

                                          if (X-&gt;on_rq) // false

                                          if (t) while (X-&gt;on_cpu)
                                            cpu_relax();

          context_switch(X, ..)
            finish_lock_switch(X)
              store_release(X-&gt;on_cpu, 0);

Avoid the load of X-&gt;on_cpu being hoisted over the X-&gt;on_rq load.

Reported-by: Oleg Nesterov &lt;oleg@redhat.com&gt;
Signed-off-by: Peter Zijlstra (Intel) &lt;peterz@infradead.org&gt;
Cc: Linus Torvalds &lt;torvalds@linux-foundation.org&gt;
Cc: Mike Galbraith &lt;efault@gmx.de&gt;
Cc: Peter Zijlstra &lt;peterz@infradead.org&gt;
Cc: Thomas Gleixner &lt;tglx@linutronix.de&gt;
Signed-off-by: Ingo Molnar &lt;mingo@kernel.org&gt;
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
@@ -1946,6 +1946,25 @@ try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
 		goto stat;
 
 #ifdef CONFIG_SMP
+	/*
+	 * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be
+	 * possible to, falsely, observe p->on_cpu == 0.
+	 *
+	 * One must be running (->on_cpu == 1) in order to remove oneself
+	 * from the runqueue.
+	 *
+	 *  [S] ->on_cpu = 1;	[L] ->on_rq
+	 *      UNLOCK rq->lock
+	 *			RMB
+	 *      LOCK   rq->lock
+	 *  [S] ->on_rq = 0;    [L] ->on_cpu
+	 *
+	 * Pairs with the full barrier implied in the UNLOCK+LOCK on rq->lock
+	 * from the consecutive calls to schedule(); the first switching to our
+	 * task, the second putting it to sleep.
+	 */
+	smp_rmb();
+
 	/*
 	 * If the owning (remote) cpu is still in the middle of schedule() with
 	 * this task as prev, wait until its done referencing the task.
