timer: Write timer->flags atomically

lock_timer_base() cannot prevent the following :

CPU1 ( in __mod_timer()
timer->flags |= TIMER_MIGRATING;
spin_unlock(&base->lock);
base = new_base;
spin_lock(&base->lock);
// The next line clears TIMER_MIGRATING
timer->flags &= ~TIMER_BASEMASK;
                                  CPU2 (in lock_timer_base())
                                  see timer base is cpu0 base
                                  spin_lock_irqsave(&base->lock, *flags);
                                  if (timer->flags == tf)
                                       return base; // oops, wrong base
timer->flags |= base->cpu // too late

We must write timer->flags in one go, otherwise we can fool other cpus.

Fixes: bc7a34b ("timer: Reduce timer migration overhead if disabled")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Jon Christopherson <jon@jons.org>
Cc: David Miller <davem@davemloft.net>
Cc: xen-devel@lists.xen.org
Cc: david.vrabel@citrix.com
Cc: Sander Eikelenboom <linux@eikelenboom.it>
Link: http://lkml.kernel.org/r/1439831928.32680.11.camel@edumazet-glaptop2.roam.corp.google.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Thomas Gleixner <tglx@linutronix.de>

kernel/time/timer.c

@@ -807,8 +807,8 @@ __mod_timer(struct timer_list *timer, unsigned long expires,
 			spin_unlock(&base->lock);
 			base = new_base;
 			spin_lock(&base->lock);
-			timer->flags &= ~TIMER_BASEMASK;
-			timer->flags |= base->cpu;
+			WRITE_ONCE(timer->flags,
+				   (timer->flags & ~TIMER_BASEMASK) | base->cpu);
 		}
 	}