When a timer is enqueued with a negative delta (ie: expiry is below
base->clk), it gets added to the wheel as expiring now (base->clk).
Yet the value that gets stored in base->next_expiry, while calling
trigger_dyntick_cpu(), is the initial timer->expires value. The
resulting state becomes:
	base->next_expiry < base->clk
On the next timer enqueue, forward_timer_base() may accidentally
rewind base->clk. As a possible outcome, timers may expire way too
early, the worst case being that the highest wheel levels get spuriously
processed again.
To prevent from that, make sure that base->next_expiry doesn't get below
base->clk.
Fixes: a683f390b93f ("timers: Forward the wheel clock whenever possible")
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Tested-by: Juri Lelli <juri.lelli@redhat.com>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/20200703010657.2302-1-frederic@kernel.org
 
         * Set the next expiry time and kick the CPU so it can reevaluate the
         * wheel:
         */
-       base->next_expiry = timer->expires;
+       if (time_before(timer->expires, base->clk)) {
+               /*
+                * Prevent from forward_timer_base() moving the base->clk
+                * backward
+                */
+               base->next_expiry = base->clk;
+       } else {
+               base->next_expiry = timer->expires;
+       }
        wake_up_nohz_cpu(base->cpu);
 }
 
         * If the next expiry value is > jiffies, then we fast forward to
         * jiffies otherwise we forward to the next expiry value.
         */
-       if (time_after(base->next_expiry, jnow))
+       if (time_after(base->next_expiry, jnow)) {
                base->clk = jnow;
-       else
+       } else {
+               if (WARN_ON_ONCE(time_before(base->next_expiry, base->clk)))
+                       return;
                base->clk = base->next_expiry;
+       }
 #endif
 }