locking/rwsem: Rework zeroing reader waiter->task
authorDavidlohr Bueso <dave@stgolabs.net>
Fri, 13 May 2016 18:56:27 +0000 (11:56 -0700)
committerIngo Molnar <mingo@kernel.org>
Fri, 3 Jun 2016 07:47:12 +0000 (09:47 +0200)
commite38513905eeaae59056eac2c9ac55a43b1fc41b2
tree3f78436bd5eb40437736015f2851cb5765ebbe81
parent133e89ef5ef338e1358b16246521ba17d935c396
locking/rwsem: Rework zeroing reader waiter->task

Readers that are awoken will expect a nil ->task indicating
that a wakeup has occurred. Because of the way readers are
implemented, there's a small chance that the waiter will never
block in the slowpath (rwsem_down_read_failed), and therefore
requires some form of reference counting to avoid the following
scenario:

rwsem_down_read_failed() rwsem_wake()
  get_task_struct();
  spin_lock_irq(&wait_lock);
  list_add_tail(&waiter.list)
  spin_unlock_irq(&wait_lock);
  raw_spin_lock_irqsave(&wait_lock)
  __rwsem_do_wake()
  while (1) {
    set_task_state(TASK_UNINTERRUPTIBLE);
    waiter->task = NULL
    if (!waiter.task) // true
      break;
    schedule() // never reached

   __set_task_state(TASK_RUNNING);
 do_exit();
    wake_up_process(tsk); // boom

... and therefore race with do_exit() when the caller returns.

There is also a mismatch between the smp_mb() and its documentation,
in that the serialization is done between reading the task and the
nil store. Furthermore, in addition to having the overlapping of
loads and stores to waiter->task guaranteed to be ordered within
that CPU, both wake_up_process() originally and now wake_q_add()
already imply barriers upon successful calls, which serves the
comment.

Now, as an alternative to perhaps inverting the checks in the blocker
side (which has its own penalty in that schedule is unavoidable),
with lockless wakeups this situation is naturally addressed and we
can just use the refcount held by wake_q_add(), instead doing so
explicitly. Of course, we must guarantee that the nil store is done
as the _last_ operation in that the task must already be marked for
deletion to not fall into the race above. Spurious wakeups are also
handled transparently in that the task's reference is only removed
when wake_up_q() is actually called _after_ the nil store.

Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman.Long@hpe.com
Cc: dave@stgolabs.net
Cc: jason.low2@hp.com
Cc: peter@hurleysoftware.com
Link: http://lkml.kernel.org/r/1463165787-25937-3-git-send-email-dave@stgolabs.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kernel/locking/rwsem-xadd.c