watchdog/hardlockup/perf: Prevent CPU hotplug deadlock

The following deadlock is possible in the watchdog hotplug code:

  cpus_write_lock()
    ...
      takedown_cpu()
        smpboot_park_threads()
          smpboot_park_thread()
            kthread_park()
              ->park() := watchdog_disable()
                watchdog_nmi_disable()
                  perf_event_release_kernel();
                    put_event()
                      _free_event()
                        ->destroy() := hw_perf_event_destroy()
                          x86_release_hardware()
                            release_ds_buffers()
                              get_online_cpus()

when a per cpu watchdog perf event is destroyed which drops the last
reference to the PMU hardware. The cleanup code there invokes
get_online_cpus() which instantly deadlocks because the hotplug percpu
rwsem is write locked.

To solve this add a deferring mechanism:

  cpus_write_lock()
			   kthread_park()
			    watchdog_nmi_disable(deferred)
			      perf_event_disable(event);
			      move_event_to_deferred(event);
			   ....
  cpus_write_unlock()
  cleaup_deferred_events()
    perf_event_release_kernel()

This is still properly serialized against concurrent hotplug via the
cpu_add_remove_lock, which is held by the task which initiated the hotplug
event.

This is also used to handle event destruction when the watchdog threads are
parked via other mechanisms than CPU hotplug.

Analyzed-by: Peter Zijlstra <peterz@infradead.org>

Reported-by: Borislav Petkov <bp@alien8.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Don Zickus <dzickus@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Chris Metcalf <cmetcalf@mellanox.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Siewior <bigeasy@linutronix.de>
Cc: Ulrich Obergfell <uobergfe@redhat.com>
Link: http://lkml.kernel.org/r/20170912194146.884469246@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>

3 files changed, 59 insertions, 6 deletions

diff --git a/kernel/cpu.c b/kernel/cpu.c
index acf5308fad51..a96b348591df 100644
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -24,6 +24,7 @@
 #include <linux/lockdep.h>
 #include <linux/tick.h>
 #include <linux/irq.h>
+#include <linux/nmi.h>
 #include <linux/smpboot.h>
 #include <linux/relay.h>
 #include <linux/slab.h>
@@ -734,6 +735,11 @@ static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
 
 out:
 	cpus_write_unlock();
+	/*
+	 * Do post unplug cleanup. This is still protected against
+	 * concurrent CPU hotplug via cpu_add_remove_lock.
+	 */
+	lockup_detector_cleanup();
 	return ret;
 }
 
diff --git a/kernel/watchdog.c b/kernel/watchdog.c
index af000956286c..dd1fd59683c5 100644
--- a/kernel/watchdog.c
+++ b/kernel/watchdog.c
@@ -109,8 +109,10 @@ int __weak watchdog_nmi_enable(unsigned int cpu)
 {
 	return 0;
 }
+
 void __weak watchdog_nmi_disable(unsigned int cpu)
 {
+	hardlockup_detector_perf_disable();
 }
 
 /*
@@ -193,6 +195,8 @@ __setup("hardlockup_all_cpu_backtrace=", hardlockup_all_cpu_backtrace_setup);
 #endif
 #endif
 
+static void __lockup_detector_cleanup(void);
+
 /*
  * Hard-lockup warnings should be triggered after just a few seconds. Soft-
  * lockups can have false positives under extreme conditions. So we generally
@@ -631,6 +635,24 @@ static void set_sample_period(void)
 }
 #endif /* SOFTLOCKUP */
 
+static void __lockup_detector_cleanup(void)
+{
+	lockdep_assert_held(&watchdog_mutex);
+	hardlockup_detector_perf_cleanup();
+}
+
+/**
+ * lockup_detector_cleanup - Cleanup after cpu hotplug or sysctl changes
+ *
+ * Caller must not hold the cpu hotplug rwsem.
+ */
+void lockup_detector_cleanup(void)
+{
+	mutex_lock(&watchdog_mutex);
+	__lockup_detector_cleanup();
+	mutex_unlock(&watchdog_mutex);
+}
+
 /**
  * lockup_detector_soft_poweroff - Interface to stop lockup detector(s)
  *
@@ -665,6 +687,8 @@ static int proc_watchdog_update(void)
 
 	watchdog_nmi_reconfigure();
 
+	__lockup_detector_cleanup();
+
 	return err;
 
 }
@@ -837,6 +861,7 @@ int proc_watchdog_cpumask(struct ctl_table *table, int write,
 		}
 
 		watchdog_nmi_reconfigure();
+		__lockup_detector_cleanup();
 	}
 
 	mutex_unlock(&watchdog_mutex);
diff --git a/kernel/watchdog_hld.c b/kernel/watchdog_hld.c
index 7b602714ea53..94111ccb09b5 100644
--- a/kernel/watchdog_hld.c
+++ b/kernel/watchdog_hld.c
@@ -21,6 +21,8 @@
 static DEFINE_PER_CPU(bool, hard_watchdog_warn);
 static DEFINE_PER_CPU(bool, watchdog_nmi_touch);
 static DEFINE_PER_CPU(struct perf_event *, watchdog_ev);
+static DEFINE_PER_CPU(struct perf_event *, dead_event);
+static struct cpumask dead_events_mask;
 
 static unsigned long hardlockup_allcpu_dumped;
 static bool hardlockup_detector_disabled;
@@ -239,16 +241,18 @@ out:
 	return 0;
 }
 
-void watchdog_nmi_disable(unsigned int cpu)
+/**
+ * hardlockup_detector_perf_disable - Disable the local event
+ */
+void hardlockup_detector_perf_disable(void)
 {
-	struct perf_event *event = per_cpu(watchdog_ev, cpu);
+	struct perf_event *event = this_cpu_read(watchdog_ev);
 
 	if (event) {
 		perf_event_disable(event);
-		per_cpu(watchdog_ev, cpu) = NULL;
-
-		/* should be in cleanup, but blocks oprofile */
-		perf_event_release_kernel(event);
+		this_cpu_write(watchdog_ev, NULL);
+		this_cpu_write(dead_event, event);
+		cpumask_set_cpu(smp_processor_id(), &dead_events_mask);
 
 		/* watchdog_nmi_enable() expects this to be zero initially. */
 		if (atomic_dec_and_test(&watchdog_cpus))
@@ -257,6 +261,24 @@ void watchdog_nmi_disable(unsigned int cpu)
 }
 
 /**
+ * hardlockup_detector_perf_cleanup - Cleanup disabled events and destroy them
+ *
+ * Called from lockup_detector_cleanup(). Serialized by the caller.
+ */
+void hardlockup_detector_perf_cleanup(void)
+{
+	int cpu;
+
+	for_each_cpu(cpu, &dead_events_mask) {
+		struct perf_event *event = per_cpu(dead_event, cpu);
+
+		per_cpu(dead_event, cpu) = NULL;
+		perf_event_release_kernel(event);
+	}
+	cpumask_clear(&dead_events_mask);
+}
+
+/**
  * hardlockup_detector_perf_stop - Globally stop watchdog events
  *
  * Special interface for x86 to handle the perf HT bug.