watchdog/hardlockup/perf: Use atomics to track in-use cpu counter
authorDon Zickus <dzickus@redhat.com>
Wed, 1 Nov 2017 18:11:27 +0000 (14:11 -0400)
committerThomas Gleixner <tglx@linutronix.de>
Wed, 1 Nov 2017 20:18:40 +0000 (21:18 +0100)
commit42f930da7f00c0ab23df4c7aed36137f35988980
tree893725417db8d2f581994c6baa8cb1b2d2e79f72
parent9c388a5ed1960b2ebbebd3dbe7553092b0c15ec1
watchdog/hardlockup/perf: Use atomics to track in-use cpu counter

Guenter reported:
  There is still a problem. When running
    echo 6 > /proc/sys/kernel/watchdog_thresh
    echo 5 > /proc/sys/kernel/watchdog_thresh
  repeatedly, the message

   NMI watchdog: Enabled. Permanently consumes one hw-PMU counter.

  stops after a while (after ~10-30 iterations, with fluctuations).
  Maybe watchdog_cpus needs to be atomic ?

That's correct as this again is affected by the asynchronous nature of the
smpboot thread unpark mechanism.

CPU 0 CPU1 CPU2
write(watchdog_thresh, 6)
  stop()
    park()
  update()
  start()
    unpark()
thread->unpark()
  cnt++;
write(watchdog_thresh, 5) thread->unpark()
  stop()
    park() thread->park()
   cnt--;   cnt++;
  update()
  start()
    unpark()

That's not a functional problem, it just affects the informational message.

Convert watchdog_cpus to atomic_t to prevent the problem

Reported-and-tested-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Don Zickus <dzickus@redhat.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20171101181126.j727fqjmdthjz4xk@redhat.com
kernel/watchdog_hld.c