tick: Remove outgoing CPU from broadcast masks

Valentin reported that unplugging a CPU occasionally results in a warning
in the tick broadcast code which is triggered when an offline CPU is in the
broadcast mask.

This happens because the outgoing CPU is not removing itself from the
broadcast masks, especially not from the broadcast_force_mask. The removal
happens on the control CPU after the outgoing CPU is dead. It's a long
standing issue, but the warning is harmless.

Rework the hotplug mechanism so that the outgoing CPU removes itself from
the broadcast masks after disabling interrupts and removing itself from the
online mask.

Reported-by: Valentin Schneider <valentin.schneider@arm.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Valentin Schneider <valentin.schneider@arm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Link: https://lkml.kernel.org/r/alpine.DEB.2.21.1903211540180.1784@nanos.tec.linutronix.de

1 files changed, 19 insertions, 21 deletions

diff --git a/kernel/time/tick-broadcast.c b/kernel/time/tick-broadcast.c
index ee834d4fb814..0283523de045 100644
--- a/kernel/time/tick-broadcast.c
+++ b/kernel/time/tick-broadcast.c
@@ -36,10 +36,12 @@ static __cacheline_aligned_in_smp DEFINE_RAW_SPINLOCK(tick_broadcast_lock);
 static void tick_broadcast_setup_oneshot(struct clock_event_device *bc);
 static void tick_broadcast_clear_oneshot(int cpu);
 static void tick_resume_broadcast_oneshot(struct clock_event_device *bc);
+static void tick_broadcast_oneshot_offline(unsigned int cpu);
 #else
 static inline void tick_broadcast_setup_oneshot(struct clock_event_device *bc) { BUG(); }
 static inline void tick_broadcast_clear_oneshot(int cpu) { }
 static inline void tick_resume_broadcast_oneshot(struct clock_event_device *bc) { }
+static inline void tick_broadcast_oneshot_offline(unsigned int cpu) { }
 #endif
 
 /*
@@ -433,27 +435,29 @@ void tick_set_periodic_handler(struct clock_event_device *dev, int broadcast)
 }
 
 #ifdef CONFIG_HOTPLUG_CPU
-/*
- * Remove a CPU from broadcasting
- */
-void tick_shutdown_broadcast(unsigned int cpu)
+static void tick_shutdown_broadcast(void)
 {
-	struct clock_event_device *bc;
-	unsigned long flags;
-
-	raw_spin_lock_irqsave(&tick_broadcast_lock, flags);
-
-	bc = tick_broadcast_device.evtdev;
-	cpumask_clear_cpu(cpu, tick_broadcast_mask);
-	cpumask_clear_cpu(cpu, tick_broadcast_on);
+	struct clock_event_device *bc = tick_broadcast_device.evtdev;
 
 	if (tick_broadcast_device.mode == TICKDEV_MODE_PERIODIC) {
 		if (bc && cpumask_empty(tick_broadcast_mask))
 			clockevents_shutdown(bc);
 	}
+}
 
-	raw_spin_unlock_irqrestore(&tick_broadcast_lock, flags);
+/*
+ * Remove a CPU from broadcasting
+ */
+void tick_broadcast_offline(unsigned int cpu)
+{
+	raw_spin_lock(&tick_broadcast_lock);
+	cpumask_clear_cpu(cpu, tick_broadcast_mask);
+	cpumask_clear_cpu(cpu, tick_broadcast_on);
+	tick_broadcast_oneshot_offline(cpu);
+	tick_shutdown_broadcast();
+	raw_spin_unlock(&tick_broadcast_lock);
 }
+
 #endif
 
 void tick_suspend_broadcast(void)
@@ -950,14 +954,10 @@ void hotplug_cpu__broadcast_tick_pull(int deadcpu)
 }
 
 /*
- * Remove a dead CPU from broadcasting
+ * Remove a dying CPU from broadcasting
  */
-void tick_shutdown_broadcast_oneshot(unsigned int cpu)
+static void tick_broadcast_oneshot_offline(unsigned int cpu)
 {
-	unsigned long flags;
-
-	raw_spin_lock_irqsave(&tick_broadcast_lock, flags);
-
 	/*
 	 * Clear the broadcast masks for the dead cpu, but do not stop
 	 * the broadcast device!
@@ -965,8 +965,6 @@ void tick_shutdown_broadcast_oneshot(unsigned int cpu)
 	cpumask_clear_cpu(cpu, tick_broadcast_oneshot_mask);
 	cpumask_clear_cpu(cpu, tick_broadcast_pending_mask);
 	cpumask_clear_cpu(cpu, tick_broadcast_force_mask);
-
-	raw_spin_unlock_irqrestore(&tick_broadcast_lock, flags);
 }
 #endif