timer: Read jiffies once when forwarding base clk

The timer delayed for more than 3 seconds warning was triggered during
testing.

  Workqueue: events_unbound sched_tick_remote
  RIP: 0010:sched_tick_remote+0xee/0x100
  ...
  Call Trace:
   process_one_work+0x18c/0x3a0
   worker_thread+0x30/0x380
   kthread+0x113/0x130
   ret_from_fork+0x22/0x40

The reason is that the code in collect_expired_timers() uses jiffies
unprotected:

    if (next_event > jiffies)
        base->clk = jiffies;

As the compiler is allowed to reload the value base->clk can advance
between the check and the store and in the worst case advance farther than
next event. That causes the timer expiry to be delayed until the wheel
pointer wraps around.

Convert the code to use READ_ONCE()

Fixes: 23696838 ("timers: Optimize collect_expired_timers() for NOHZ")
Signed-off-by: Li RongQing <lirongqing@baidu.com>
Signed-off-by: Liang ZhiCheng <liangzhicheng@baidu.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/1568894687-14499-1-git-send-email-lirongqing@baidu.com

kernel/time/timer.c

@@ -1678,24 +1678,26 @@ void timer_clear_idle(void)
 static int collect_expired_timers(struct timer_base *base,
 				  struct hlist_head *heads)
 {
+	unsigned long now = READ_ONCE(jiffies);
+
 	/*
 	 * NOHZ optimization. After a long idle sleep we need to forward the
 	 * base to current jiffies. Avoid a loop by searching the bitfield for
 	 * the next expiring timer.
 	 */
-	if ((long)(jiffies - base->clk) > 2) {
+	if ((long)(now - base->clk) > 2) {
 		unsigned long next = __next_timer_interrupt(base);
 
 		/*
 		 * If the next timer is ahead of time forward to current
 		 * jiffies, otherwise forward to the next expiry time:
 		 */
-		if (time_after(next, jiffies)) {
+		if (time_after(next, now)) {
 			/*
 			 * The call site will increment base->clk and then
 			 * terminate the expiry loop immediately.
 			 */
-			base->clk = jiffies;
+			base->clk = now;
 			return 0;
 		}
 		base->clk = next;