[PATCH] clocksource: register cpu notifier to remove timer from dying CPU

[Viresh Kumar]

clocksource core is using add_timer_on() to run clocksource_watchdog() on all
CPUs one by one. But when a core is brought down, clocksource core doesn't
remove this timer from the dying CPU. And in this case timer core gives this
(Gives this only with unmerged code, anyway in the current code as well timer
core is migrating a pinned timer to other CPUs, which is also wrong:
http://www.gossamer-threads.com/lists/linux/kernel/1898117)

migrate_timer_list: can't migrate pinned timer: ffffffff81f06a60,
timer->function: ffffffff810d7010,deactivating it Modules linked in:

CPU: 0 PID: 1932 Comm: 01-cpu-hotplug Not tainted 3.14.0-rc1-00088-gab3c4fd #4
Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
 0000000000000009 ffff88001d407c38 ffffffff817237bd ffff88001d407c80
 ffff88001d407c70 ffffffff8106a1dd 0000000000000010 ffffffff81f06a60
 ffff88001e04d040 ffffffff81e3d4c0 ffff88001e04d030 ffff88001d407cd0
Call Trace:
 [<ffffffff817237bd>] dump_stack+0x4d/0x66
 [<ffffffff8106a1dd>] warn_slowpath_common+0x7d/0xa0
 [<ffffffff8106a24c>] warn_slowpath_fmt+0x4c/0x50
 [<ffffffff810761c3>] ? __internal_add_timer+0x113/0x130
 [<ffffffff810d7010>] ? clocksource_watchdog_kthread+0x40/0x40
 [<ffffffff8107753b>] migrate_timer_list+0xdb/0xf0
 [<ffffffff810782dc>] timer_cpu_notify+0xfc/0x1f0
 [<ffffffff8173046c>] notifier_call_chain+0x4c/0x70
 [<ffffffff8109340e>] __raw_notifier_call_chain+0xe/0x10
 [<ffffffff8106a3f3>] cpu_notify+0x23/0x50
 [<ffffffff8106a44e>] cpu_notify_nofail+0xe/0x20
 [<ffffffff81712a5d>] _cpu_down+0x1ad/0x2e0
 [<ffffffff81712bc4>] cpu_down+0x34/0x50
 [<ffffffff813fec54>] cpu_subsys_offline+0x14/0x20
 [<ffffffff813f9f65>] device_offline+0x95/0xc0
 [<ffffffff813fa060>] online_store+0x40/0x90
 [<ffffffff813f75d8>] dev_attr_store+0x18/0x30
 [<ffffffff8123309d>] sysfs_kf_write+0x3d/0x50

This patch tries to fix this by registering cpu notifiers from clocksource core,
only when we start clocksource-watchdog. And if on the CPU_DEAD notification it
is found that dying CPU was the CPU on which this timer is queued on, then it is
removed from that CPU and queued to next CPU.

Reported-by: Fengguang Wu <fengguang.wu@xxxxxxxxx>
Reported-and-tested-by: Jet Chen <jet.chen@xxxxxxxxx>
Signed-off-by: Viresh Kumar <viresh.kumar@xxxxxxxxxx>
---
 kernel/time/clocksource.c | 64 +++++++++++++++++++++++++++++++++++++++--------
 1 file changed, 53 insertions(+), 11 deletions(-)

diff --git a/kernel/time/clocksource.c b/kernel/time/clocksource.c
index ba3e502..8054fb1 100644
--- a/kernel/time/clocksource.c
+++ b/kernel/time/clocksource.c
@@ -23,16 +23,21 @@
  *   o Allow clocksource drivers to be unregistered
  */
 
+#include <linux/cpu.h>
 #include <linux/device.h>
 #include <linux/clocksource.h>
 #include <linux/init.h>
 #include <linux/module.h>
+#include <linux/notifier.h>
 #include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */
 #include <linux/tick.h>
 #include <linux/kthread.h>
 
 #include "tick-internal.h"
 
+/* Tracks current CPU to queue watchdog timer on */
+static int timer_cpu;
+
 void timecounter_init(struct timecounter *tc,
 		      const struct cyclecounter *cc,
 		      u64 start_tstamp)
@@ -246,12 +251,25 @@ void clocksource_mark_unstable(struct clocksource *cs)
 	spin_unlock_irqrestore(&watchdog_lock, flags);
 }
 
+static void queue_timer_on_next_cpu(void)
+{
+	/*
+	 * Cycle through CPUs to check if the CPUs stay synchronized to each
+	 * other.
+	 */
+	timer_cpu = cpumask_next(timer_cpu, cpu_online_mask);
+	if (timer_cpu >= nr_cpu_ids)
+		timer_cpu = cpumask_first(cpu_online_mask);
+	watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
+	add_timer_on(&watchdog_timer, timer_cpu);
+}
+
 static void clocksource_watchdog(unsigned long data)
 {
 	struct clocksource *cs;
 	cycle_t csnow, wdnow;
 	int64_t wd_nsec, cs_nsec;
-	int next_cpu, reset_pending;
+	int reset_pending;
 
 	spin_lock(&watchdog_lock);
 	if (!watchdog_running)
@@ -336,27 +354,50 @@ static void clocksource_watchdog(unsigned long data)
 	if (reset_pending)
 		atomic_dec(&watchdog_reset_pending);
 
-	/*
-	 * Cycle through CPUs to check if the CPUs stay synchronized
-	 * to each other.
-	 */
-	next_cpu = cpumask_next(raw_smp_processor_id(), cpu_online_mask);
-	if (next_cpu >= nr_cpu_ids)
-		next_cpu = cpumask_first(cpu_online_mask);
-	watchdog_timer.expires += WATCHDOG_INTERVAL;
-	add_timer_on(&watchdog_timer, next_cpu);
+	queue_timer_on_next_cpu();
 out:
 	spin_unlock(&watchdog_lock);
 }
 
+static int clocksource_cpu_notify(struct notifier_block *self,
+				unsigned long action, void *hcpu)
+{
+	long cpu = (long)hcpu;
+
+	spin_lock(&watchdog_lock);
+	if (!watchdog_running)
+		goto notify_out;
+
+	switch (action) {
+	case CPU_DEAD:
+	case CPU_DEAD_FROZEN:
+		if (cpu != timer_cpu)
+			break;
+		del_timer(&watchdog_timer);
+		queue_timer_on_next_cpu();
+		break;
+	}
+
+notify_out:
+	spin_unlock(&watchdog_lock);
+	return NOTIFY_OK;
+}
+
+static struct notifier_block clocksource_nb = {
+	.notifier_call	= clocksource_cpu_notify,
+	.priority = 1,
+};
+
 static inline void clocksource_start_watchdog(void)
 {
 	if (watchdog_running || !watchdog || list_empty(&watchdog_list))
 		return;
+	timer_cpu = cpumask_first(cpu_online_mask);
+	register_cpu_notifier(&clocksource_nb);
 	init_timer(&watchdog_timer);
 	watchdog_timer.function = clocksource_watchdog;
 	watchdog_timer.expires = jiffies + WATCHDOG_INTERVAL;
-	add_timer_on(&watchdog_timer, cpumask_first(cpu_online_mask));
+	add_timer_on(&watchdog_timer, timer_cpu);
 	watchdog_running = 1;
 }
 
@@ -365,6 +406,7 @@ static inline void clocksource_stop_watchdog(void)
 	if (!watchdog_running || (watchdog && !list_empty(&watchdog_list)))
 		return;
 	del_timer(&watchdog_timer);
+	unregister_cpu_notifier(&clocksource_nb);
 	watchdog_running = 0;
 }
 
-- 
1.7.12.rc2.18.g61b472e

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/