hrtimer: Force clock_was_set() handling for the HIGHRES=n, NOHZ=y case

When CONFIG_HIGH_RES_TIMERS is disabled, but NOHZ is enabled then
clock_was_set() is not doing anything. With HIGHRES=n the kernel relies on
the periodic tick to update the clock offsets, but when NOHZ is enabled and
active then CPUs which are in a deep idle sleep do not have a periodic tick
which means the expiry of timers affected by clock_was_set() can be
arbitrarily delayed up to the point where the CPUs are brought out of idle
again.

Make the clock_was_set() logic unconditionaly available so that idle CPUs
are kicked out of idle to handle the update.

Signed-off-by: Thomas Gleixner <[email protected]>
Link: https://lore.kernel.org/r/[email protected]

kernel/time/hrtimer.c

@@ -739,23 +739,7 @@ static inline int hrtimer_is_hres_enabled(void)
 	return hrtimer_hres_enabled;
 }
 
-/*
- * Retrigger next event is called after clock was set
- *
- * Called with interrupts disabled via on_each_cpu()
- */
-static void retrigger_next_event(void *arg)
-{
-	struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
-
-	if (!__hrtimer_hres_active(base))
-		return;
-
-	raw_spin_lock(&base->lock);
-	hrtimer_update_base(base);
-	hrtimer_force_reprogram(base, 0);
-	raw_spin_unlock(&base->lock);
-}
+static void retrigger_next_event(void *arg);
 
 /*
  * Switch to high resolution mode
@@ -781,9 +765,50 @@ static void hrtimer_switch_to_hres(void)
 
 static inline int hrtimer_is_hres_enabled(void) { return 0; }
 static inline void hrtimer_switch_to_hres(void) { }
-static inline void retrigger_next_event(void *arg) { }
 
 #endif /* CONFIG_HIGH_RES_TIMERS */
+/*
+ * Retrigger next event is called after clock was set with interrupts
+ * disabled through an SMP function call or directly from low level
+ * resume code.
+ *
+ * This is only invoked when:
+ *	- CONFIG_HIGH_RES_TIMERS is enabled.
+ *	- CONFIG_NOHZ_COMMON is enabled
+ *
+ * For the other cases this function is empty and because the call sites
+ * are optimized out it vanishes as well, i.e. no need for lots of
+ * #ifdeffery.
+ */
+static void retrigger_next_event(void *arg)
+{
+	struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
+
+	/*
+	 * When high resolution mode or nohz is active, then the offsets of
+	 * CLOCK_REALTIME/TAI/BOOTTIME have to be updated. Otherwise the
+	 * next tick will take care of that.
+	 *
+	 * If high resolution mode is active then the next expiring timer
+	 * must be reevaluated and the clock event device reprogrammed if
+	 * necessary.
+	 *
+	 * In the NOHZ case the update of the offset and the reevaluation
+	 * of the next expiring timer is enough. The return from the SMP
+	 * function call will take care of the reprogramming in case the
+	 * CPU was in a NOHZ idle sleep.
+	 */
+	if (!__hrtimer_hres_active(base) && !tick_nohz_active)
+		return;
+
+	raw_spin_lock(&base->lock);
+	hrtimer_update_base(base);
+	if (__hrtimer_hres_active(base))
+		hrtimer_force_reprogram(base, 0);
+	else
+		hrtimer_update_next_event(base);
+	raw_spin_unlock(&base->lock);
+}
 
 /*
  * When a timer is enqueued and expires earlier than the already enqueued
@@ -842,22 +867,28 @@ static void hrtimer_reprogram(struct hrtimer *timer, bool reprogram)
 }
 
 /*
- * Clock realtime was set
- *
- * Change the offset of the realtime clock vs. the monotonic
- * clock.
+ * Clock was set. This might affect CLOCK_REALTIME, CLOCK_TAI and
+ * CLOCK_BOOTTIME (for late sleep time injection).
  *
- * We might have to reprogram the high resolution timer interrupt. On
- * SMP we call the architecture specific code to retrigger _all_ high
- * resolution timer interrupts. On UP we just disable interrupts and
- * call the high resolution interrupt code.
+ * This requires to update the offsets for these clocks
+ * vs. CLOCK_MONOTONIC. When high resolution timers are enabled, then this
+ * also requires to eventually reprogram the per CPU clock event devices
+ * when the change moves an affected timer ahead of the first expiring
+ * timer on that CPU. Obviously remote per CPU clock event devices cannot
+ * be reprogrammed. The other reason why an IPI has to be sent is when the
+ * system is in !HIGH_RES and NOHZ mode. The NOHZ mode updates the offsets
+ * in the tick, which obviously might be stopped, so this has to bring out
+ * the remote CPU which might sleep in idle to get this sorted.
  */
 void clock_was_set(void)
 {
-#ifdef CONFIG_HIGH_RES_TIMERS
+	if (!hrtimer_hres_active() && !tick_nohz_active)
+		goto out_timerfd;
+
 	/* Retrigger the CPU local events everywhere */
 	on_each_cpu(retrigger_next_event, NULL, 1);
-#endif
+
+out_timerfd:
 	timerfd_clock_was_set();
 }