sched/fair: Plug hole between hotplug and active_load_balance()
authorPeter Zijlstra <peterz@infradead.org>
Thu, 7 Sep 2017 15:03:51 +0000 (17:03 +0200)
committerIngo Molnar <mingo@kernel.org>
Tue, 12 Sep 2017 15:41:04 +0000 (17:41 +0200)
commitedd8e41d2e3cbd6ebe13ead30eb1adc6f48cbb33
tree3a7dca8c3058d0513b20657e4f1eb30506876f5f
parent2800486ee34825d954f64c6f98037daea328f121
sched/fair: Plug hole between hotplug and active_load_balance()

The load balancer applies cpu_active_mask to whatever sched_domains it
finds, however in the case of active_balance there is a hole between
setting rq->{active_balance,push_cpu} and running the stop_machine
work doing the actual migration.

The @push_cpu can go offline in this window, which would result in us
moving a task onto a dead cpu, which is a fairly bad thing.

Double check the active mask before the stop work does the migration.

  CPU0 CPU1

  <SoftIRQ>
stop_machine(takedown_cpu)
    load_balance() cpu_stopper_thread()
      ...   work = multi_cpu_stop
      stop_one_cpu_nowait(     /* wait for CPU0 */
.func = active_load_balance_cpu_stop
      );
  </SoftIRQ>

  cpu_stopper_thread()
    work = multi_cpu_stop
      /* sync with CPU1 */
    take_cpu_down()
<idle>
  play_dead();

    work = active_load_balance_cpu_stop
      set_task_cpu(p, CPU1); /* oops!! */

Reported-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20170907150614.044460912@infradead.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kernel/sched/fair.c