Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1723905 > unrolled thread

[patch 00/29] lockup_detector: Cure hotplug deadlocks and replace duct tape

Back to article view | Back to linux.kernel

Contents

  [patch 00/29] lockup_detector: Cure hotplug deadlocks and replace  duct tape Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 02/29] perf/x86/intel: Sanitize PMU HT bug workaround Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 25/29] lockup_detector: Implement init time detection of perf Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 19/29] lockup_detector: Cleanup header mess Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 26/29] lockup_detector/perf: Implement CPU enable replacement Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 24/29] lockup_detector/perf: Implement init time perf  validation Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
      Re: [patch 24/29] lockup_detector/perf: Implement init time perf  validation Don Zickus <dzickus@redhat.com> - 2017-09-07 18:00 +0200
    [patch 10/29] lockup_detector/perf: Prevent cpu hotplug deadlock Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
      Re: [patch 10/29] lockup_detector/perf: Prevent cpu hotplug deadlock Don Zickus <dzickus@redhat.com> - 2017-09-01 21:10 +0200
        Re: [patch 10/29] lockup_detector/perf: Prevent cpu hotplug  deadlock Thomas Gleixner <tglx@linutronix.de> - 2017-09-01 21:30 +0200
          Re: [patch 10/29] lockup_detector/perf: Prevent cpu hotplug deadlock Don Zickus <dzickus@redhat.com> - 2017-09-05 17:00 +0200
    [patch 21/29] lockup_detector: Cleanup sysctl variable name space Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 18/29] lockup_detector: Further simplify sysctl handling Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 23/29] lockup_detector: Get rid of the racy update loop Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 27/29] lockup_detector: Use new perf CPU enable mechanism Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 13/29] lockup_detector: Cleanup the ifdef maze Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 06/29] lockup_detector: Rework cpu hotplug locking Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 14/29] lockup_detector: Split out cpumask write function Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 11/29] lockup_detector: Remove park_in_progress hackery Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
      Re: [patch 11/29] lockup_detector: Remove park_in_progress hackery Peter Zijlstra <peterz@infradead.org> - 2017-09-04 14:20 +0200
        Re: [patch 11/29] lockup_detector: Remove park_in_progress hackery Don Zickus <dzickus@redhat.com> - 2017-09-05 17:20 +0200
          Re: [patch 11/29] lockup_detector: Remove park_in_progress hackery Thomas Gleixner <tglx@linutronix.de> - 2017-09-05 17:50 +0200
      Re: [patch 11/29] lockup_detector: Remove park_in_progress hackery Thomas Gleixner <tglx@linutronix.de> - 2017-09-05 16:00 +0200
        [patch V2 11/29] lockup_detector: Remove park_in_progress  obfuscation Thomas Gleixner <tglx@linutronix.de> - 2017-09-05 21:20 +0200
    [patch 04/29] parisc: Use lockup_detector_stop() Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 29/29] lockup_detector: Cleanup hotplug locking mess Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 12/29] lockup_detector: Cleanup stub functions Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 03/29] lockup_detector: Provide interface to stop from  poweroff() Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 20/29] lockup_detector/sysctl: Get rid of the ifdeffery Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 08/29] lockup_detector: Mark hardlockup_detector_disable()  __init Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 17/29] lockup_detector: Get rid of the thread teardown/setup  dance Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
      Re: [patch 17/29] lockup_detector: Get rid of the thread  teardown/setup dance Don Zickus <dzickus@redhat.com> - 2017-09-01 21:10 +0200
        Re: [patch 17/29] lockup_detector: Get rid of the thread teardown/setup  dance Thomas Gleixner <tglx@linutronix.de> - 2017-09-01 21:50 +0200
    [patch 15/29] smpboot/threads: Avoid runtime allocation Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 07/29] lockup_detector: Rename watchdog_proc_mutex Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 16/29] lockup_detector: Create new thread handling  infrastructure Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 22/29] lockup_detector: Make watchdog_nmi_reconfigure() two  stage Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
    [patch 01/29] hardlockup_detector: Provide interface to stop/restart  perf events Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:40 +0200
      Re: [patch 01/29] hardlockup_detector: Provide interface to  stop/restart perf events Borislav Petkov <bp@alien8.de> - 2017-09-06 18:20 +0200
    [patch 05/29] lockup_detector: Remove broken suspend/resume interfaces Thomas Gleixner <tglx@linutronix.de> - 2017-08-31 09:50 +0200
    Re: [patch 00/29] lockup_detector: Cure hotplug deadlocks and  replace duct tape Don Zickus <dzickus@redhat.com> - 2017-09-01 00:20 +0200
      Re: [patch 00/29] lockup_detector: Cure hotplug deadlocks and  replace duct tape Nicholas Piggin <npiggin@gmail.com> - 2017-09-01 06:50 +0200
      Re: [patch 00/29] lockup_detector: Cure hotplug deadlocks and replace  duct tape Thomas Gleixner <tglx@linutronix.de> - 2017-09-01 11:20 +0200
    Re: [patch 00/29] lockup_detector: Cure hotplug deadlocks and  replace duct tape Don Zickus <dzickus@redhat.com> - 2017-09-07 18:10 +0200

Page 1 of 3  [1] 2 3  Next page →

#1723905 — [patch 00/29] lockup_detector: Cure hotplug deadlocks and replace duct tape

The lockup detector is broken is several ways:

    - It's deadlock prone vs. CPU hotplug in various ways. Some of these
      are due to recursive cpus_read_lock() others are due to
      cpus_read_lock() from CPU hotplug callbacks which immediately lock
      the machine because cpus are write locked.

    - The handling of the cpu hotplug threads happens sideways to the
      smpboot thread infrastructure, which is racy and pointless

    - The handling of the user space sysctl interface is a complete
      trainwreck as it fiddles directly with variables which can be
      modified or evaluated by the running watchdogs.

    - The perf event initialization is a steaming pile of duct tape as it
      idiotically tries to create perf events over and over even if perf is
      not functional (no hardware, ....). To avoid excessive dmesg spam it
      contains magic printk ratelimiting along with either wrong or useless
      messages.

    - The code structure is horrible as ifdef sections are scattered all
      over the place which makes it unreadable

    - There is more wreckage, but see the changelogs for the ugly details.

Before I get utterly grumpy, I just pretend that I don't give a sh*t!

The following series sanitizes the facility and addresses the problems.

Thanks,

	tglx
---
 arch/parisc/kernel/process.c   |    2 
 arch/powerpc/kernel/watchdog.c |   22 -
 arch/x86/events/intel/core.c   |   11 
 include/linux/nmi.h            |  121 +++----
 include/linux/smpboot.h        |    4 
 kernel/cpu.c                   |    6 
 kernel/smpboot.c               |   22 -
 kernel/sysctl.c                |   22 -
 kernel/watchdog.c              |  638 ++++++++++++++---------------------------
 kernel/watchdog_hld.c          |  193 ++++++------
 10 files changed, 433 insertions(+), 608 deletions(-)

      

[toc] | [next] | [standalone]

#1723906 — [patch 02/29] perf/x86/intel: Sanitize PMU HT bug workaround

From: Peter Zijlstra <peterz@infradead.org>

The lockup_detector_suspend/resume() interface is broken in several ways
especially as it results in recursive locking of the CPU hotplug lock.

Use the new stop/restart interface in the perf NMI watchdog to temporarily
disable and reenable the already active watchdog events. That's enough to
handle it.

Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 arch/x86/events/intel/core.c |   11 ++++-------
 1 file changed, 4 insertions(+), 7 deletions(-)

--- a/arch/x86/events/intel/core.c
+++ b/arch/x86/events/intel/core.c
@@ -4318,10 +4318,9 @@ static __init int fixup_ht_bug(void)
 		return 0;
 	}
 
-	if (lockup_detector_suspend() != 0) {
-		pr_debug("failed to disable PMU erratum BJ122, BV98, HSD29 workaround\n");
-		return 0;
-	}
+	cpus_read_lock();
+
+	hardlockup_detector_perf_stop();
 
 	x86_pmu.flags &= ~(PMU_FL_EXCL_CNTRS | PMU_FL_EXCL_ENABLED);
 
@@ -4329,9 +4328,7 @@ static __init int fixup_ht_bug(void)
 	x86_pmu.commit_scheduling = NULL;
 	x86_pmu.stop_scheduling = NULL;
 
-	lockup_detector_resume();
-
-	cpus_read_lock();
+	hardlockup_detector_perf_restart();
 
 	for_each_online_cpu(c)
 		free_excl_cntrs(c);

[toc] | [prev] | [next] | [standalone]

#1723908 — [patch 25/29] lockup_detector: Implement init time detection of perf

Use the init time detection of the perf NMI watchdog to determine whether
the perf NMI watchdog is functional. If not disable it permanentely. It
won't come back magically at runtime.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 kernel/watchdog.c |   12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

--- a/kernel/watchdog.c
+++ b/kernel/watchdog.c
@@ -44,6 +44,7 @@ int __read_mostly watchdog_user_enabled
 int __read_mostly nmi_watchdog_user_enabled = NMI_WATCHDOG_DEFAULT;
 int __read_mostly soft_watchdog_user_enabled = 1;
 int __read_mostly watchdog_thresh = 10;
+int __read_mostly nmi_watchdog_available;
 
 struct cpumask watchdog_allowed_mask __read_mostly;
 static bool softlockup_threads_initialized __read_mostly;
@@ -138,7 +139,7 @@ static void lockup_detector_update_enabl
 	watchdog_enabled = 0;
 	if (!watchdog_user_enabled)
 		return;
-	if (nmi_watchdog_user_enabled)
+	if (nmi_watchdog_available && nmi_watchdog_user_enabled)
 		watchdog_enabled |= NMI_WATCHDOG_ENABLED;
 	if (soft_watchdog_user_enabled)
 		watchdog_enabled |= SOFT_WATCHDOG_ENABLED;
@@ -686,6 +687,8 @@ int proc_watchdog(struct ctl_table *tabl
 int proc_nmi_watchdog(struct ctl_table *table, int write,
 		      void __user *buffer, size_t *lenp, loff_t *ppos)
 {
+	if (!nmi_watchdog_available && write)
+		return -ENOTSUPP;
 	return proc_watchdog_common(NMI_WATCHDOG_ENABLED,
 				    table, write, buffer, lenp, ppos);
 }
@@ -746,6 +749,12 @@ int proc_watchdog_cpumask(struct ctl_tab
 }
 #endif /* CONFIG_SYSCTL */
 
+static __init void detect_nmi_watchdog(void)
+{
+	if (!hardlockup_detector_perf_init())
+		nmi_watchdog_available = true;
+}
+
 void __init lockup_detector_init(void)
 {
 #ifdef CONFIG_NO_HZ_FULL
@@ -758,5 +767,6 @@ void __init lockup_detector_init(void)
 	cpumask_copy(&watchdog_cpumask, cpu_possible_mask);
 #endif
 
+	detect_nmi_watchdog();
 	softlockup_init_threads();
 }

[toc] | [prev] | [next] | [standalone]

#1723909 — [patch 19/29] lockup_detector: Cleanup header mess

Having the same #ifdef in various places does not make it more
readable. Collect stuff into one place.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 include/linux/nmi.h |   60 +++++++++++++++++++++-------------------------------
 1 file changed, 25 insertions(+), 35 deletions(-)

--- a/include/linux/nmi.h
+++ b/include/linux/nmi.h
@@ -14,11 +14,29 @@
 void lockup_detector_init(void);
 void lockup_detector_soft_poweroff(void);
 void lockup_detector_cleanup(void);
+bool is_hardlockup(void);
+
+extern int watchdog_user_enabled;
+extern int nmi_watchdog_enabled;
+extern int soft_watchdog_enabled;
+extern int watchdog_thresh;
+extern unsigned long watchdog_enabled;
+
+extern struct cpumask watchdog_cpumask;
+extern unsigned long *watchdog_cpumask_bits;
+#ifdef CONFIG_SMP
+extern int sysctl_softlockup_all_cpu_backtrace;
+extern int sysctl_hardlockup_all_cpu_backtrace;
 #else
+#define sysctl_softlockup_all_cpu_backtrace 0
+#define sysctl_hardlockup_all_cpu_backtrace 0
+#endif /* !CONFIG_SMP */
+
+#else /* CONFIG_LOCKUP_DETECTOR */
 static inline void lockup_detector_init(void) { }
 static inline void lockup_detector_soft_poweroff(void) { }
 static inline void lockup_detector_cleanup(void) { }
-#endif
+#endif /* !CONFIG_LOCKUP_DETECTOR */
 
 #ifdef CONFIG_SOFTLOCKUP_DETECTOR
 extern void touch_softlockup_watchdog_sched(void);
@@ -26,28 +44,17 @@ extern void touch_softlockup_watchdog(vo
 extern void touch_softlockup_watchdog_sync(void);
 extern void touch_all_softlockup_watchdogs(void);
 extern unsigned int  softlockup_panic;
-extern int soft_watchdog_enabled;
 #else
-static inline void touch_softlockup_watchdog_sched(void)
-{
-}
-static inline void touch_softlockup_watchdog(void)
-{
-}
-static inline void touch_softlockup_watchdog_sync(void)
-{
-}
-static inline void touch_all_softlockup_watchdogs(void)
-{
-}
+static inline void touch_softlockup_watchdog_sched(void) { }
+static inline void touch_softlockup_watchdog(void) { }
+static inline void touch_softlockup_watchdog_sync(void) { }
+static inline void touch_all_softlockup_watchdogs(void) { }
 #endif
 
 #ifdef CONFIG_DETECT_HUNG_TASK
 void reset_hung_task_detector(void);
 #else
-static inline void reset_hung_task_detector(void)
-{
-}
+static inline void reset_hung_task_detector(void) { }
 #endif
 
 /*
@@ -92,7 +99,7 @@ static inline void arch_touch_nmi_watchd
 
 /**
  * touch_nmi_watchdog - restart NMI watchdog timeout.
- * 
+ *
  * If the architecture supports the NMI watchdog, touch_nmi_watchdog()
  * may be used to reset the timeout - for code which intentionally
  * disables interrupts for a long time. This call is stateless.
@@ -162,21 +169,6 @@ static inline bool trigger_single_cpu_ba
 u64 hw_nmi_get_sample_period(int watchdog_thresh);
 #endif
 
-#ifdef CONFIG_LOCKUP_DETECTOR
-extern int nmi_watchdog_enabled;
-extern int watchdog_user_enabled;
-extern int watchdog_thresh;
-extern unsigned long watchdog_enabled;
-extern struct cpumask watchdog_cpumask;
-extern unsigned long *watchdog_cpumask_bits;
-#ifdef CONFIG_SMP
-extern int sysctl_softlockup_all_cpu_backtrace;
-extern int sysctl_hardlockup_all_cpu_backtrace;
-#else
-#define sysctl_softlockup_all_cpu_backtrace 0
-#define sysctl_hardlockup_all_cpu_backtrace 0
-#endif
-
 #if defined(CONFIG_HARDLOCKUP_CHECK_TIMESTAMP) && \
     defined(CONFIG_HARDLOCKUP_DETECTOR)
 void watchdog_update_hrtimer_threshold(u64 period);
@@ -184,7 +176,6 @@ void watchdog_update_hrtimer_threshold(u
 static inline void watchdog_update_hrtimer_threshold(u64 period) { }
 #endif
 
-extern bool is_hardlockup(void);
 struct ctl_table;
 extern int proc_watchdog(struct ctl_table *, int ,
 			 void __user *, size_t *, loff_t *);
@@ -196,7 +187,6 @@ extern int proc_watchdog_thresh(struct c
 				void __user *, size_t *, loff_t *);
 extern int proc_watchdog_cpumask(struct ctl_table *, int,
 				 void __user *, size_t *, loff_t *);
-#endif
 
 #ifdef CONFIG_HAVE_ACPI_APEI_NMI
 #include <asm/nmi.h>

[toc] | [prev] | [next] | [standalone]

#1723910 — [patch 26/29] lockup_detector/perf: Implement CPU enable replacement

watchdog_nmi_enable() is an unparseable mess, Provide a clean perf specific
implementation, which will be used when the existing setup/teardown mess is
replaced.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 include/linux/nmi.h   |    2 ++
 kernel/watchdog_hld.c |   11 +++++++++++
 2 files changed, 13 insertions(+)

--- a/include/linux/nmi.h
+++ b/include/linux/nmi.h
@@ -92,12 +92,14 @@ extern void arch_touch_nmi_watchdog(void
 extern void hardlockup_detector_perf_stop(void);
 extern void hardlockup_detector_perf_restart(void);
 extern void hardlockup_detector_perf_disable(void);
+extern void hardlockup_detector_perf_enable(void);
 extern void hardlockup_detector_perf_cleanup(void);
 extern int hardlockup_detector_perf_init(void);
 #else
 static inline void hardlockup_detector_perf_stop(void) { }
 static inline void hardlockup_detector_perf_restart(void) { }
 static inline void hardlockup_detector_perf_disable(void) { }
+static inline void hardlockup_detector_perf_enable(void) { }
 static inline void hardlockup_detector_perf_cleanup(void) { }
 # if !defined(CONFIG_HAVE_NMI_WATCHDOG)
 static int hardlockup_detector_perf_init(void) { return -ENODEV; }
--- a/kernel/watchdog_hld.c
+++ b/kernel/watchdog_hld.c
@@ -260,6 +260,17 @@ static int hardlockup_detector_event_cre
 }
 
 /**
+ * hardlockup_detector_perf_enable - Enable the local event
+ */
+void hardlockup_detector_perf_enable(void)
+{
+	if (hardlockup_detector_event_create())
+		return;
+
+	perf_event_enable(this_cpu_read(watchdog_ev));
+}
+
+/**
  * hardlockup_detector_perf_disable - Disable the local event
  */
 void hardlockup_detector_perf_disable(void)

[toc] | [prev] | [next] | [standalone]

#1723911 — [patch 24/29] lockup_detector/perf: Implement init time perf validation

The watchdog tries to create perf events even after it figured out that
perf is not functional or the requested event is not supported.

That's braindead as this can be done once at init time and if not supported
the NMI watchdog can be turned off unconditonally.

Implement the perf hardlockup detector functionality for that. This creates
a new event create function, which will replace the unholy mess of the
existing one in later patches.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 include/linux/nmi.h   |    8 ++++++--
 kernel/watchdog_hld.c |   37 +++++++++++++++++++++++++++++++++++++
 2 files changed, 43 insertions(+), 2 deletions(-)

--- a/include/linux/nmi.h
+++ b/include/linux/nmi.h
@@ -93,14 +93,18 @@ extern void hardlockup_detector_perf_sto
 extern void hardlockup_detector_perf_restart(void);
 extern void hardlockup_detector_perf_disable(void);
 extern void hardlockup_detector_perf_cleanup(void);
+extern int hardlockup_detector_perf_init(void);
 #else
 static inline void hardlockup_detector_perf_stop(void) { }
 static inline void hardlockup_detector_perf_restart(void) { }
 static inline void hardlockup_detector_perf_disable(void) { }
 static inline void hardlockup_detector_perf_cleanup(void) { }
-#if !defined(CONFIG_HAVE_NMI_WATCHDOG)
+# if !defined(CONFIG_HAVE_NMI_WATCHDOG)
+static int hardlockup_detector_perf_init(void) { return -ENODEV; }
 static inline void arch_touch_nmi_watchdog(void) {}
-#endif
+# else
+static int hardlockup_detector_perf_init(void) { return 0; }
+# endif
 #endif
 
 void watchdog_nmi_reconfigure(bool run);
--- a/kernel/watchdog_hld.c
+++ b/kernel/watchdog_hld.c
@@ -238,6 +238,27 @@ int watchdog_nmi_enable(unsigned int cpu
 	return 0;
 }
 
+static int hardlockup_detector_event_create(void)
+{
+	unsigned int cpu = smp_processor_id();
+	struct perf_event_attr *wd_attr;
+	struct perf_event *evt;
+
+	wd_attr = &wd_hw_attr;
+	wd_attr->sample_period = hw_nmi_get_sample_period(watchdog_thresh);
+
+	/* Try to register using hardware perf events */
+	evt = perf_event_create_kernel_counter(wd_attr, cpu, NULL,
+					       watchdog_overflow_callback, NULL);
+	if (IS_ERR(evt)) {
+		pr_info("Perf event create on CPU %d failed with %ld\n", cpu,
+			PTR_ERR(evt));
+		return PTR_ERR(evt);
+	}
+	this_cpu_write(watchdog_ev, evt);
+	return 0;
+}
+
 /**
  * hardlockup_detector_perf_disable - Disable the local event
  */
@@ -315,3 +336,19 @@ void __init hardlockup_detector_perf_res
 			perf_event_enable(event);
 	}
 }
+
+/**
+ * hardlockup_detector_perf_init - Probe whether NMI event is available at all
+ */
+int __init hardlockup_detector_perf_init(void)
+{
+	int ret = hardlockup_detector_event_create();
+
+	if (ret) {
+		pr_info("Perf NMI watchdog permanetely disabled\n");
+	} else {
+		perf_event_release_kernel(this_cpu_read(watchdog_ev));
+		this_cpu_write(watchdog_ev, NULL);
+	}
+	return ret;
+}

[toc] | [prev] | [next] | [standalone]

#1728282 — Re: [patch 24/29] lockup_detector/perf: Implement init time perf validation

On Thu, Aug 31, 2017 at 09:16:22AM +0200, Thomas Gleixner wrote:
> The watchdog tries to create perf events even after it figured out that
> perf is not functional or the requested event is not supported.
> 
> That's braindead as this can be done once at init time and if not supported
> the NMI watchdog can be turned off unconditonally.
> 
> Implement the perf hardlockup detector functionality for that. This creates
> a new event create function, which will replace the unholy mess of the
> existing one in later patches.
> 
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> ---
>  include/linux/nmi.h   |    8 ++++++--
>  kernel/watchdog_hld.c |   37 +++++++++++++++++++++++++++++++++++++
>  2 files changed, 43 insertions(+), 2 deletions(-)
> 
> --- a/include/linux/nmi.h
> +++ b/include/linux/nmi.h
> @@ -93,14 +93,18 @@ extern void hardlockup_detector_perf_sto
>  extern void hardlockup_detector_perf_restart(void);
>  extern void hardlockup_detector_perf_disable(void);
>  extern void hardlockup_detector_perf_cleanup(void);
> +extern int hardlockup_detector_perf_init(void);
>  #else
>  static inline void hardlockup_detector_perf_stop(void) { }
>  static inline void hardlockup_detector_perf_restart(void) { }
>  static inline void hardlockup_detector_perf_disable(void) { }
>  static inline void hardlockup_detector_perf_cleanup(void) { }
> -#if !defined(CONFIG_HAVE_NMI_WATCHDOG)
> +# if !defined(CONFIG_HAVE_NMI_WATCHDOG)
> +static int hardlockup_detector_perf_init(void) { return -ENODEV; }
>  static inline void arch_touch_nmi_watchdog(void) {}
> -#endif
> +# else
> +static int hardlockup_detector_perf_init(void) { return 0; }

hardlockup_detector_perf_init needs to be 'inline' otherwise fails to
compile with CONFIG_HARDLOCKUP_DETECTOR_PERF turned off.


Cheers,
Don

> +# endif
>  #endif
>  
>  void watchdog_nmi_reconfigure(bool run);
> --- a/kernel/watchdog_hld.c
> +++ b/kernel/watchdog_hld.c
> @@ -238,6 +238,27 @@ int watchdog_nmi_enable(unsigned int cpu
>  	return 0;
>  }
>  
> +static int hardlockup_detector_event_create(void)
> +{
> +	unsigned int cpu = smp_processor_id();
> +	struct perf_event_attr *wd_attr;
> +	struct perf_event *evt;
> +
> +	wd_attr = &wd_hw_attr;
> +	wd_attr->sample_period = hw_nmi_get_sample_period(watchdog_thresh);
> +
> +	/* Try to register using hardware perf events */
> +	evt = perf_event_create_kernel_counter(wd_attr, cpu, NULL,
> +					       watchdog_overflow_callback, NULL);
> +	if (IS_ERR(evt)) {
> +		pr_info("Perf event create on CPU %d failed with %ld\n", cpu,
> +			PTR_ERR(evt));
> +		return PTR_ERR(evt);
> +	}
> +	this_cpu_write(watchdog_ev, evt);
> +	return 0;
> +}
> +
>  /**
>   * hardlockup_detector_perf_disable - Disable the local event
>   */
> @@ -315,3 +336,19 @@ void __init hardlockup_detector_perf_res
>  			perf_event_enable(event);
>  	}
>  }
> +
> +/**
> + * hardlockup_detector_perf_init - Probe whether NMI event is available at all
> + */
> +int __init hardlockup_detector_perf_init(void)
> +{
> +	int ret = hardlockup_detector_event_create();
> +
> +	if (ret) {
> +		pr_info("Perf NMI watchdog permanetely disabled\n");
> +	} else {
> +		perf_event_release_kernel(this_cpu_read(watchdog_ev));
> +		this_cpu_write(watchdog_ev, NULL);
> +	}
> +	return ret;
> +}
> 
> 

[toc] | [prev] | [next] | [standalone]

#1723913 — [patch 10/29] lockup_detector/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.

Reported-by: Borislav Petkov <bp@alien8.de>
Analyzed-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 include/linux/nmi.h   |    6 ++++++
 kernel/cpu.c          |    6 ++++++
 kernel/watchdog.c     |   24 ++++++++++++++++++++++++
 kernel/watchdog_hld.c |   34 ++++++++++++++++++++++++++++------
 4 files changed, 64 insertions(+), 6 deletions(-)

--- a/include/linux/nmi.h
+++ b/include/linux/nmi.h
@@ -13,9 +13,11 @@
 #ifdef CONFIG_LOCKUP_DETECTOR
 void lockup_detector_init(void);
 void lockup_detector_soft_poweroff(void);
+void lockup_detector_cleanup(void);
 #else
 static inline void lockup_detector_init(void) { }
 static inline void lockup_detector_soft_poweroff(void) { }
+static inline void lockup_detector_cleanup(void) { }
 #endif
 
 #ifdef CONFIG_SOFTLOCKUP_DETECTOR
@@ -77,9 +79,13 @@ static inline void hardlockup_detector_d
 extern void arch_touch_nmi_watchdog(void);
 extern void hardlockup_detector_perf_stop(void);
 extern void hardlockup_detector_perf_restart(void);
+extern void hardlockup_detector_perf_disable(void);
+extern void hardlockup_detector_perf_cleanup(void);
 #else
 static inline void hardlockup_detector_perf_stop(void) { }
 static inline void hardlockup_detector_perf_restart(void) { }
+static inline void hardlockup_detector_perf_disable(void) { }
+static inline void hardlockup_detector_perf_cleanup(void) { }
 #if !defined(CONFIG_HAVE_NMI_WATCHDOG)
 static inline void arch_touch_nmi_watchdog(void) {}
 #endif
--- 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>
@@ -733,6 +734,11 @@ static int __ref _cpu_down(unsigned int
 
 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;
 }
 
--- a/kernel/watchdog.c
+++ b/kernel/watchdog.c
@@ -193,6 +193,8 @@ static int __init hardlockup_all_cpu_bac
 #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
@@ -459,6 +461,7 @@ static void watchdog_disable(unsigned in
 	hrtimer_cancel(hrtimer);
 	/* disable the perf event */
 	watchdog_nmi_disable(cpu);
+	hardlockup_detector_perf_disable();
 }
 
 static void watchdog_cleanup(unsigned int cpu, bool online)
@@ -631,6 +634,24 @@ 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 +686,8 @@ static int proc_watchdog_update(void)
 
 	watchdog_nmi_reconfigure();
 
+	__lockup_detector_cleanup();
+
 	return err;
 
 }
@@ -837,6 +860,7 @@ int proc_watchdog_cpumask(struct ctl_tab
 		}
 
 		watchdog_nmi_reconfigure();
+		__lockup_detector_cleanup();
 	}
 
 	mutex_unlock(&watchdog_mutex);
--- 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 @@ int watchdog_nmi_enable(unsigned int cpu
 	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 c
 }
 
 /**
+ * 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.

[toc] | [prev] | [next] | [standalone]

#1725294 — Re: [patch 10/29] lockup_detector/perf: Prevent cpu hotplug deadlock

On Thu, Aug 31, 2017 at 09:16:08AM +0200, Thomas Gleixner wrote:
> 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.

The main reason perf_event_release_kernel is in this path is because the
oprofile folks complained they couldn't use the perf counters when the
nmi_watchdog was disabled on the command line.

I believe your patches only release the perf event on cpu unplug now.

Unless oprofile has been updated, this may still cause issues.  But
detecting this on sysctl changes can probably work around this.

Cheers,
Don

> 
> 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.
> 
> Reported-by: Borislav Petkov <bp@alien8.de>
> Analyzed-by: Peter Zijlstra <peterz@infradead.org>
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> ---
>  include/linux/nmi.h   |    6 ++++++
>  kernel/cpu.c          |    6 ++++++
>  kernel/watchdog.c     |   24 ++++++++++++++++++++++++
>  kernel/watchdog_hld.c |   34 ++++++++++++++++++++++++++++------
>  4 files changed, 64 insertions(+), 6 deletions(-)
> 
> --- a/include/linux/nmi.h
> +++ b/include/linux/nmi.h
> @@ -13,9 +13,11 @@
>  #ifdef CONFIG_LOCKUP_DETECTOR
>  void lockup_detector_init(void);
>  void lockup_detector_soft_poweroff(void);
> +void lockup_detector_cleanup(void);
>  #else
>  static inline void lockup_detector_init(void) { }
>  static inline void lockup_detector_soft_poweroff(void) { }
> +static inline void lockup_detector_cleanup(void) { }
>  #endif
>  
>  #ifdef CONFIG_SOFTLOCKUP_DETECTOR
> @@ -77,9 +79,13 @@ static inline void hardlockup_detector_d
>  extern void arch_touch_nmi_watchdog(void);
>  extern void hardlockup_detector_perf_stop(void);
>  extern void hardlockup_detector_perf_restart(void);
> +extern void hardlockup_detector_perf_disable(void);
> +extern void hardlockup_detector_perf_cleanup(void);
>  #else
>  static inline void hardlockup_detector_perf_stop(void) { }
>  static inline void hardlockup_detector_perf_restart(void) { }
> +static inline void hardlockup_detector_perf_disable(void) { }
> +static inline void hardlockup_detector_perf_cleanup(void) { }
>  #if !defined(CONFIG_HAVE_NMI_WATCHDOG)
>  static inline void arch_touch_nmi_watchdog(void) {}
>  #endif
> --- 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>
> @@ -733,6 +734,11 @@ static int __ref _cpu_down(unsigned int
>  
>  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;
>  }
>  
> --- a/kernel/watchdog.c
> +++ b/kernel/watchdog.c
> @@ -193,6 +193,8 @@ static int __init hardlockup_all_cpu_bac
>  #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
> @@ -459,6 +461,7 @@ static void watchdog_disable(unsigned in
>  	hrtimer_cancel(hrtimer);
>  	/* disable the perf event */
>  	watchdog_nmi_disable(cpu);
> +	hardlockup_detector_perf_disable();
>  }
>  
>  static void watchdog_cleanup(unsigned int cpu, bool online)
> @@ -631,6 +634,24 @@ 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 +686,8 @@ static int proc_watchdog_update(void)
>  
>  	watchdog_nmi_reconfigure();
>  
> +	__lockup_detector_cleanup();
> +
>  	return err;
>  
>  }
> @@ -837,6 +860,7 @@ int proc_watchdog_cpumask(struct ctl_tab
>  		}
>  
>  		watchdog_nmi_reconfigure();
> +		__lockup_detector_cleanup();
>  	}
>  
>  	mutex_unlock(&watchdog_mutex);
> --- 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 @@ int watchdog_nmi_enable(unsigned int cpu
>  	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 c
>  }
>  
>  /**
> + * 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.
> 
> 

[toc] | [prev] | [next] | [standalone]

#1725300 — Re: [patch 10/29] lockup_detector/perf: Prevent cpu hotplug deadlock

On Fri, 1 Sep 2017, Don Zickus wrote:
> On Thu, Aug 31, 2017 at 09:16:08AM +0200, Thomas Gleixner wrote:
> > 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.
> 
> The main reason perf_event_release_kernel is in this path is because the
> oprofile folks complained they couldn't use the perf counters when the
> nmi_watchdog was disabled on the command line.

If the nmi watchdog is disabled on the command line then there are no
counters claimed at all.

Thanks,

	tglx

[toc] | [prev] | [next] | [standalone]

#1726804 — Re: [patch 10/29] lockup_detector/perf: Prevent cpu hotplug deadlock

On Fri, Sep 01, 2017 at 09:29:07PM +0200, Thomas Gleixner wrote:
> On Fri, 1 Sep 2017, Don Zickus wrote:
> > On Thu, Aug 31, 2017 at 09:16:08AM +0200, Thomas Gleixner wrote:
> > > 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.
> > 
> > The main reason perf_event_release_kernel is in this path is because the
> > oprofile folks complained they couldn't use the perf counters when the
> > nmi_watchdog was disabled on the command line.
> 
> If the nmi watchdog is disabled on the command line then there are no
> counters claimed at all.

Ah, I see it now.  When you park all the threads, you clear the cpumask
which then calls the lockup detector_cleanup to release all perf_counters
that are no longer used by the cpumask.

Further reading this code, I see how the code covers various race conditions
we tried solving in the past.

Thanks!

Cheers,
Don

[toc] | [prev] | [next] | [standalone]

#1723914 — [patch 21/29] lockup_detector: Cleanup sysctl variable name space

Reflect that these variables are user interface related and remove the
whitespace damage in the sysctl table while at it.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 include/linux/nmi.h |   16 ++++++++--------
 kernel/sysctl.c     |   16 ++++++++--------
 kernel/watchdog.c   |   41 ++++++++++++++++++++---------------------
 3 files changed, 36 insertions(+), 37 deletions(-)

--- a/include/linux/nmi.h
+++ b/include/linux/nmi.h
@@ -17,8 +17,8 @@ void lockup_detector_cleanup(void);
 bool is_hardlockup(void);
 
 extern int watchdog_user_enabled;
-extern int nmi_watchdog_enabled;
-extern int soft_watchdog_enabled;
+extern int nmi_watchdog_user_enabled;
+extern int soft_watchdog_user_enabled;
 extern int watchdog_thresh;
 extern unsigned long watchdog_enabled;
 
@@ -62,12 +62,12 @@ static inline void reset_hung_task_detec
  * 'watchdog_enabled' variable. Each lockup detector has its dedicated bit -
  * bit 0 for the hard lockup detector and bit 1 for the soft lockup detector.
  *
- * 'watchdog_user_enabled', 'nmi_watchdog_enabled' and 'soft_watchdog_enabled'
- * are variables that are only used as an 'interface' between the parameters
- * in /proc/sys/kernel and the internal state bits in 'watchdog_enabled'. The
- * 'watchdog_thresh' variable is handled differently because its value is not
- * boolean, and the lockup detectors are 'suspended' while 'watchdog_thresh'
- * is equal zero.
+ * 'watchdog_user_enabled', 'nmi_watchdog_user_enabled' and
+ * 'soft_watchdog_user_enabled' are variables that are only used as an
+ * 'interface' between the parameters in /proc/sys/kernel and the internal
+ * state bits in 'watchdog_enabled'. The 'watchdog_thresh' variable is
+ * handled differently because its value is not boolean, and the lockup
+ * detectors are 'suspended' while 'watchdog_thresh' is equal zero.
  */
 #define NMI_WATCHDOG_ENABLED_BIT   0
 #define SOFT_WATCHDOG_ENABLED_BIT  1
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -871,9 +871,9 @@ static struct ctl_table kern_table[] = {
 #if defined(CONFIG_LOCKUP_DETECTOR)
 	{
 		.procname       = "watchdog",
-		.data           = &watchdog_user_enabled,
-		.maxlen         = sizeof (int),
-		.mode           = 0644,
+		.data		= &watchdog_user_enabled,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
 		.proc_handler   = proc_watchdog,
 		.extra1		= &zero,
 		.extra2		= &one,
@@ -889,8 +889,8 @@ static struct ctl_table kern_table[] = {
 	},
 	{
 		.procname       = "nmi_watchdog",
-		.data           = &nmi_watchdog_enabled,
-		.maxlen         = sizeof (int),
+		.data		= &nmi_watchdog_user_enabled,
+		.maxlen		= sizeof(int),
 		.mode		= NMI_WATCHDOG_SYSCTL_PERM,
 		.proc_handler   = proc_nmi_watchdog,
 		.extra1		= &zero,
@@ -906,9 +906,9 @@ static struct ctl_table kern_table[] = {
 #ifdef CONFIG_SOFTLOCKUP_DETECTOR
 	{
 		.procname       = "soft_watchdog",
-		.data           = &soft_watchdog_enabled,
-		.maxlen         = sizeof (int),
-		.mode           = 0644,
+		.data		= &soft_watchdog_user_enabled,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
 		.proc_handler   = proc_soft_watchdog,
 		.extra1		= &zero,
 		.extra2		= &one,
--- a/kernel/watchdog.c
+++ b/kernel/watchdog.c
@@ -31,8 +31,6 @@
 
 static DEFINE_MUTEX(watchdog_mutex);
 
-int __read_mostly nmi_watchdog_enabled;
-
 #if defined(CONFIG_HARDLOCKUP_DETECTOR) || defined(CONFIG_HAVE_NMI_WATCHDOG)
 unsigned long __read_mostly watchdog_enabled = SOFT_WATCHDOG_ENABLED |
 						NMI_WATCHDOG_ENABLED;
@@ -40,6 +38,17 @@ unsigned long __read_mostly watchdog_ena
 unsigned long __read_mostly watchdog_enabled = SOFT_WATCHDOG_ENABLED;
 #endif
 
+int __read_mostly nmi_watchdog_user_enabled;
+int __read_mostly soft_watchdog_user_enabled;
+int __read_mostly watchdog_user_enabled;
+int __read_mostly watchdog_thresh = 10;
+
+struct cpumask watchdog_allowed_mask __read_mostly;
+static bool softlockup_threads_initialized __read_mostly;
+
+struct cpumask watchdog_cpumask __read_mostly;
+unsigned long *watchdog_cpumask_bits = cpumask_bits(&watchdog_cpumask);
+
 #ifdef CONFIG_HARDLOCKUP_DETECTOR
 /*
  * Should we panic when a soft-lockup or hard-lockup occurs:
@@ -85,12 +94,6 @@ static int __init hardlockup_all_cpu_bac
 # endif /* CONFIG_SMP */
 #endif /* CONFIG_HARDLOCKUP_DETECTOR */
 
-int __read_mostly watchdog_user_enabled;
-int __read_mostly watchdog_thresh = 10;
-
-struct cpumask watchdog_cpumask __read_mostly;
-unsigned long *watchdog_cpumask_bits = cpumask_bits(&watchdog_cpumask);
-
 /*
  * These functions can be overridden if an architecture implements its
  * own hardlockup detector.
@@ -106,7 +109,7 @@ void __weak watchdog_nmi_disable(unsigne
  * watchdog_nmi_reconfigure can be implemented to be notified after any
  * watchdog configuration change. The arch hardlockup watchdog should
  * respond to the following variables:
- * - nmi_watchdog_enabled
+ * - watchdog_enabled
  * - watchdog_thresh
  * - watchdog_cpumask
  * - sysctl_hardlockup_all_cpu_backtrace
@@ -119,10 +122,6 @@ void __weak watchdog_nmi_reconfigure(voi
 /* Global variables, exported for sysctl */
 unsigned int __read_mostly softlockup_panic =
 			CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE;
-int __read_mostly soft_watchdog_enabled;
-
-struct cpumask watchdog_allowed_mask __read_mostly;
-static bool softlockup_threads_initialized __read_mostly;
 
 static u64 __read_mostly sample_period;
 
@@ -600,14 +599,14 @@ static void proc_watchdog_update(void)
 /*
  * common function for watchdog, nmi_watchdog and soft_watchdog parameter
  *
- * caller             | table->data points to | 'which' contains the flag(s)
- * -------------------|-----------------------|-----------------------------
- * proc_watchdog      | watchdog_user_enabled | NMI_WATCHDOG_ENABLED or'ed
- *                    |                       | with SOFT_WATCHDOG_ENABLED
- * -------------------|-----------------------|-----------------------------
- * proc_nmi_watchdog  | nmi_watchdog_enabled  | NMI_WATCHDOG_ENABLED
- * -------------------|-----------------------|-----------------------------
- * proc_soft_watchdog | soft_watchdog_enabled | SOFT_WATCHDOG_ENABLED
+ * caller             | table->data points to      | 'which'
+ * -------------------|----------------------------|--------------------------
+ * proc_watchdog      | watchdog_user_enabled      | NMI_WATCHDOG_ENABLED |
+ *                    |                            | SOFT_WATCHDOG_ENABLED
+ * -------------------|----------------------------|--------------------------
+ * proc_nmi_watchdog  | nmi_watchdog_user_enabled  | NMI_WATCHDOG_ENABLED
+ * -------------------|----------------------------|--------------------------
+ * proc_soft_watchdog | soft_watchdog_user_enabled | SOFT_WATCHDOG_ENABLED
  */
 static int proc_watchdog_common(int which, struct ctl_table *table, int write,
 				void __user *buffer, size_t *lenp, loff_t *ppos)

[toc] | [prev] | [next] | [standalone]

#1723915 — [patch 18/29] lockup_detector: Further simplify sysctl handling

Use a single function to update sysctl changes. This is not a high
frequency user space interface and it's root only.

Preparatory patch to cleanup the sysctl variable handling.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 kernel/watchdog.c |   27 +++++++--------------------
 1 file changed, 7 insertions(+), 20 deletions(-)

--- a/kernel/watchdog.c
+++ b/kernel/watchdog.c
@@ -501,11 +501,8 @@ static void softlockup_park_all_threads(
 	softlockup_update_smpboot_threads();
 }
 
-/*
- * Park threads which are not longer enabled and unpark threads which have
- * been newly enabled.
- */
-static void softlockup_update_threads(void)
+/* Unpark enabled threads */
+static void softlockup_unpark_threads(void)
 {
 	cpumask_copy(&watchdog_allowed_mask, &watchdog_cpumask);
 	softlockup_update_smpboot_threads();
@@ -516,7 +513,7 @@ static void softlockup_reconfigure_threa
 	softlockup_park_all_threads();
 	set_sample_period();
 	if (enabled)
-		softlockup_update_threads();
+		softlockup_unpark_threads();
 }
 
 /*
@@ -557,7 +554,6 @@ static inline void watchdog_unpark_threa
 static inline int watchdog_enable_all_cpus(void) { return 0; }
 static inline void watchdog_disable_all_cpus(void) { }
 static inline void softlockup_init_threads(void) { }
-static inline void softlockup_update_threads(void) { }
 static inline void softlockup_reconfigure_threads(bool enabled) { }
 #endif /* !CONFIG_SOFTLOCKUP_DETECTOR */
 
@@ -592,11 +588,11 @@ void lockup_detector_soft_poweroff(void)
 
 #ifdef CONFIG_SYSCTL
 
-/*
- * Update the run state of the lockup detectors.
- */
+/* Propagate any changes to the watchdog threads */
 static void proc_watchdog_update(void)
 {
+	/* Remove impossible cpus to keep sysctl output clean. */
+	cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask);
 	softlockup_reconfigure_threads(watchdog_enabled && watchdog_thresh);
 	watchdog_nmi_reconfigure();
 }
@@ -715,15 +711,6 @@ int proc_watchdog_thresh(struct ctl_tabl
 	return err;
 }
 
-static void proc_watchdog_cpumask_update(void)
-{
-	/* Remove impossible cpus to keep sysctl output clean. */
-	cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask);
-
-	softlockup_update_threads();
-	watchdog_nmi_reconfigure();
-}
-
 /*
  * The cpumask is the mask of possible cpus that the watchdog can run
  * on, not the mask of cpus it is actually running on.  This allows the
@@ -740,7 +727,7 @@ int proc_watchdog_cpumask(struct ctl_tab
 
 	err = proc_do_large_bitmap(table, write, buffer, lenp, ppos);
 	if (!err && write)
-		proc_watchdog_cpumask_update();
+		proc_watchdog_update();
 
 	mutex_unlock(&watchdog_mutex);
 	cpu_hotplug_enable();

[toc] | [prev] | [next] | [standalone]

#1723916 — [patch 23/29] lockup_detector: Get rid of the racy update loop

Letting user space poke directly at variables which are used at run time is
stupid and causes a lot of race conditions and other issues.

Seperate the user variables and on change invoke the reconfiguration, which
then stops the watchdogs, reevaluates the new user value and restarts the
watchdogs with the new parameters.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 kernel/watchdog.c |   95 ++++++++++++++++++++++++++----------------------------
 1 file changed, 47 insertions(+), 48 deletions(-)

--- a/kernel/watchdog.c
+++ b/kernel/watchdog.c
@@ -32,15 +32,17 @@
 static DEFINE_MUTEX(watchdog_mutex);
 
 #if defined(CONFIG_HARDLOCKUP_DETECTOR) || defined(CONFIG_HAVE_NMI_WATCHDOG)
-unsigned long __read_mostly watchdog_enabled = SOFT_WATCHDOG_ENABLED |
-						NMI_WATCHDOG_ENABLED;
+# define WATCHDOG_DEFAULT	(SOFT_WATCHDOG_ENABLED | NMI_WATCHDOG_ENABLED)
+# define NMI_WATCHDOG_DEFAULT	1
 #else
-unsigned long __read_mostly watchdog_enabled = SOFT_WATCHDOG_ENABLED;
+# define WATCHDOG_DEFAULT	(SOFT_WATCHDOG_ENABLED)
+# define NMI_WATCHDOG_DEFAULT	0
 #endif
 
-int __read_mostly nmi_watchdog_user_enabled;
-int __read_mostly soft_watchdog_user_enabled;
-int __read_mostly watchdog_user_enabled;
+unsigned long __read_mostly watchdog_enabled;
+int __read_mostly watchdog_user_enabled = 1;
+int __read_mostly nmi_watchdog_user_enabled = NMI_WATCHDOG_DEFAULT;
+int __read_mostly soft_watchdog_user_enabled = 1;
 int __read_mostly watchdog_thresh = 10;
 
 struct cpumask watchdog_allowed_mask __read_mostly;
@@ -65,7 +67,7 @@ unsigned int __read_mostly hardlockup_pa
  */
 void __init hardlockup_detector_disable(void)
 {
-	watchdog_enabled &= ~NMI_WATCHDOG_ENABLED;
+	nmi_watchdog_user_enabled = 0;
 }
 
 static int __init hardlockup_panic_setup(char *str)
@@ -75,9 +77,9 @@ static int __init hardlockup_panic_setup
 	else if (!strncmp(str, "nopanic", 7))
 		hardlockup_panic = 0;
 	else if (!strncmp(str, "0", 1))
-		watchdog_enabled &= ~NMI_WATCHDOG_ENABLED;
+		nmi_watchdog_user_enabled = 0;
 	else if (!strncmp(str, "1", 1))
-		watchdog_enabled |= NMI_WATCHDOG_ENABLED;
+		nmi_watchdog_user_enabled = 1;
 	return 1;
 }
 __setup("nmi_watchdog=", hardlockup_panic_setup);
@@ -125,6 +127,23 @@ void __weak watchdog_nmi_disable(unsigne
  */
 void __weak watchdog_nmi_reconfigure(bool stop) { }
 
+/**
+ * lockup_detector_update_enable - Update the sysctl enable bit
+ *
+ * Caller needs to make sure that the NMI/perf watchdogs are off, so this
+ * can't race with hardlockup_detector_disable().
+ */
+static void lockup_detector_update_enable(void)
+{
+	watchdog_enabled = 0;
+	if (!watchdog_user_enabled)
+		return;
+	if (nmi_watchdog_user_enabled)
+		watchdog_enabled |= NMI_WATCHDOG_ENABLED;
+	if (soft_watchdog_user_enabled)
+		watchdog_enabled |= SOFT_WATCHDOG_ENABLED;
+}
+
 #ifdef CONFIG_SOFTLOCKUP_DETECTOR
 
 /* Global variables, exported for sysctl */
@@ -153,14 +172,14 @@ static int __init softlockup_panic_setup
 
 static int __init nowatchdog_setup(char *str)
 {
-	watchdog_enabled = 0;
+	watchdog_user_enabled = 0;
 	return 1;
 }
 __setup("nowatchdog", nowatchdog_setup);
 
 static int __init nosoftlockup_setup(char *str)
 {
-	watchdog_enabled &= ~SOFT_WATCHDOG_ENABLED;
+	soft_watchdog_user_enabled = 0;
 	return 1;
 }
 __setup("nosoftlockup", nosoftlockup_setup);
@@ -515,12 +534,13 @@ static void softlockup_unpark_threads(vo
 	softlockup_update_smpboot_threads();
 }
 
-static void softlockup_reconfigure_threads(bool enabled)
+static void softlockup_reconfigure_threads(void)
 {
 	watchdog_nmi_reconfigure(false);
 	softlockup_park_all_threads();
 	set_sample_period();
-	if (enabled)
+	lockup_detector_update_enable();
+	if (watchdog_enabled && watchdog_thresh)
 		softlockup_unpark_threads();
 	watchdog_nmi_reconfigure(true);
 }
@@ -540,6 +560,8 @@ static __init void softlockup_init_threa
 	 * If sysctl is off and watchdog got disabled on the command line,
 	 * nothing to do here.
 	 */
+	lockup_detector_update_enable();
+
 	if (!IS_ENABLED(CONFIG_SYSCTL) &&
 	    !(watchdog_enabled && watchdog_thresh))
 		return;
@@ -553,7 +575,7 @@ static __init void softlockup_init_threa
 
 	mutex_lock(&watchdog_mutex);
 	softlockup_threads_initialized = true;
-	softlockup_reconfigure_threads(watchdog_enabled && watchdog_thresh);
+	softlockup_reconfigure_threads();
 	mutex_unlock(&watchdog_mutex);
 }
 
@@ -563,9 +585,10 @@ static inline void watchdog_unpark_threa
 static inline int watchdog_enable_all_cpus(void) { return 0; }
 static inline void watchdog_disable_all_cpus(void) { }
 static inline void softlockup_init_threads(void) { }
-static void softlockup_reconfigure_threads(bool enabled)
+static void softlockup_reconfigure_threads(void)
 {
 	watchdog_nmi_reconfigure(false);
+	lockup_detector_update_enable();
 	watchdog_nmi_reconfigure(true);
 }
 #endif /* !CONFIG_SOFTLOCKUP_DETECTOR */
@@ -606,7 +629,7 @@ static void proc_watchdog_update(void)
 {
 	/* Remove impossible cpus to keep sysctl output clean. */
 	cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask);
-	softlockup_reconfigure_threads(watchdog_enabled && watchdog_thresh);
+	softlockup_reconfigure_threads();
 }
 
 /*
@@ -624,48 +647,24 @@ static void proc_watchdog_update(void)
 static int proc_watchdog_common(int which, struct ctl_table *table, int write,
 				void __user *buffer, size_t *lenp, loff_t *ppos)
 {
-	int err, old, new;
-	int *watchdog_param = (int *)table->data;
+	int err, old, *param = table->data;
 
 	cpu_hotplug_disable();
 	mutex_lock(&watchdog_mutex);
 
-	/*
-	 * If the parameter is being read return the state of the corresponding
-	 * bit(s) in 'watchdog_enabled', else update 'watchdog_enabled' and the
-	 * run state of the lockup detectors.
-	 */
 	if (!write) {
-		*watchdog_param = (watchdog_enabled & which) != 0;
+		/*
+		 * On read synchronize the userspace interface. This is a
+		 * racy snapshot.
+		 */
+		*param = (watchdog_enabled & which) != 0;
 		err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
 	} else {
+		old = READ_ONCE(*param);
 		err = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
-		if (err)
-			goto out;
-
-		/*
-		 * There is a race window between fetching the current value
-		 * from 'watchdog_enabled' and storing the new value. During
-		 * this race window, watchdog_nmi_enable() can sneak in and
-		 * clear the NMI_WATCHDOG_ENABLED bit in 'watchdog_enabled'.
-		 * The 'cmpxchg' detects this race and the loop retries.
-		 */
-		do {
-			old = watchdog_enabled;
-			/*
-			 * If the parameter value is not zero set the
-			 * corresponding bit(s), else clear it(them).
-			 */
-			if (*watchdog_param)
-				new = old | which;
-			else
-				new = old & ~which;
-		} while (cmpxchg(&watchdog_enabled, old, new) != old);
-
-		if (old != new)
+		if (!err && old != READ_ONCE(*param))
 			proc_watchdog_update();
 	}
-out:
 	mutex_unlock(&watchdog_mutex);
 	cpu_hotplug_enable();
 	return err;

[toc] | [prev] | [next] | [standalone]

#1723918 — [patch 27/29] lockup_detector: Use new perf CPU enable mechanism

Get rid of the hodgepodge which tries to be smart about perf being
unavailable and error printout rate limiting.

That's all not required simply because this is never invoked when the perf
NMI watchdog is not functional.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 kernel/watchdog.c     |    2 +
 kernel/watchdog_hld.c |   88 ++------------------------------------------------
 2 files changed, 7 insertions(+), 83 deletions(-)

--- a/kernel/watchdog.c
+++ b/kernel/watchdog.c
@@ -452,6 +452,8 @@ static void watchdog_enable(unsigned int
 	/* Initialize timestamp */
 	__touch_watchdog();
 	/* Enable the perf event */
+	if (watchdog_enabled & NMI_WATCHDOG_ENABLED)
+		hardlockup_detector_perf_enable();
 	watchdog_nmi_enable(cpu);
 
 	watchdog_set_prio(SCHED_FIFO, MAX_RT_PRIO - 1);
--- a/kernel/watchdog_hld.c
+++ b/kernel/watchdog_hld.c
@@ -25,7 +25,7 @@ static DEFINE_PER_CPU(struct perf_event
 static struct cpumask dead_events_mask;
 
 static unsigned long hardlockup_allcpu_dumped;
-static bool hardlockup_detector_disabled;
+static unsigned int watchdog_cpus;
 
 void arch_touch_nmi_watchdog(void)
 {
@@ -160,84 +160,6 @@ static void watchdog_overflow_callback(s
 	return;
 }
 
-/*
- * People like the simple clean cpu node info on boot.
- * Reduce the watchdog noise by only printing messages
- * that are different from what cpu0 displayed.
- */
-static unsigned long firstcpu_err;
-static atomic_t watchdog_cpus;
-
-int watchdog_nmi_enable(unsigned int cpu)
-{
-	struct perf_event_attr *wd_attr;
-	struct perf_event *event = per_cpu(watchdog_ev, cpu);
-	int firstcpu = 0;
-
-	/* nothing to do if the hard lockup detector is disabled */
-	if (!(watchdog_enabled & NMI_WATCHDOG_ENABLED))
-		goto out;
-
-	/* A failure disabled the hardlockup detector permanently */
-	if (hardlockup_detector_disabled)
-		return -ENODEV;
-
-	/* is it already setup and enabled? */
-	if (event && event->state > PERF_EVENT_STATE_OFF)
-		goto out;
-
-	/* it is setup but not enabled */
-	if (event != NULL)
-		goto out_enable;
-
-	if (atomic_inc_return(&watchdog_cpus) == 1)
-		firstcpu = 1;
-
-	wd_attr = &wd_hw_attr;
-	wd_attr->sample_period = hw_nmi_get_sample_period(watchdog_thresh);
-
-	/* Try to register using hardware perf events */
-	event = perf_event_create_kernel_counter(wd_attr, cpu, NULL, watchdog_overflow_callback, NULL);
-
-	/* save the first cpu's error for future comparision */
-	if (firstcpu && IS_ERR(event))
-		firstcpu_err = PTR_ERR(event);
-
-	if (!IS_ERR(event)) {
-		/* only print for the first cpu initialized */
-		if (firstcpu || firstcpu_err)
-			pr_info("enabled on all CPUs, permanently consumes one hw-PMU counter.\n");
-		goto out_save;
-	}
-
-	/* skip displaying the same error again */
-	if (!firstcpu && (PTR_ERR(event) == firstcpu_err))
-		return PTR_ERR(event);
-
-	/* vary the KERN level based on the returned errno */
-	if (PTR_ERR(event) == -EOPNOTSUPP)
-		pr_info("disabled (cpu%i): not supported (no LAPIC?)\n", cpu);
-	else if (PTR_ERR(event) == -ENOENT)
-		pr_warn("disabled (cpu%i): hardware events not enabled\n",
-			 cpu);
-	else
-		pr_err("disabled (cpu%i): unable to create perf event: %ld\n",
-			cpu, PTR_ERR(event));
-
-	pr_info("Disabling hard lockup detector permanently\n");
-	hardlockup_detector_disabled = true;
-
-	return PTR_ERR(event);
-
-	/* success path */
-out_save:
-	per_cpu(watchdog_ev, cpu) = event;
-out_enable:
-	perf_event_enable(per_cpu(watchdog_ev, cpu));
-out:
-	return 0;
-}
-
 static int hardlockup_detector_event_create(void)
 {
 	unsigned int cpu = smp_processor_id();
@@ -267,6 +189,9 @@ void hardlockup_detector_perf_enable(voi
 	if (hardlockup_detector_event_create())
 		return;
 
+	if (!watchdog_cpus++)
+		pr_info("Enabled. Permanently consumes one hw-PMU counter.\n");
+
 	perf_event_enable(this_cpu_read(watchdog_ev));
 }
 
@@ -282,10 +207,7 @@ void hardlockup_detector_perf_disable(vo
 		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))
-			firstcpu_err = 0;
+		watchdog_cpus--;
 	}
 }
 

[toc] | [prev] | [next] | [standalone]

#1723919 — [patch 13/29] lockup_detector: Cleanup the ifdef maze

The #ifdef maze in this file is horrible, Group stuff at least a bit so one
can figure out what belongs to what.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 kernel/watchdog.c |   33 +++++++++++++--------------------
 1 file changed, 13 insertions(+), 20 deletions(-)

--- a/kernel/watchdog.c
+++ b/kernel/watchdog.c
@@ -41,7 +41,6 @@ unsigned long __read_mostly watchdog_ena
 #endif
 
 #ifdef CONFIG_HARDLOCKUP_DETECTOR
-/* boot commands */
 /*
  * Should we panic when a soft-lockup or hard-lockup occurs:
  */
@@ -74,19 +73,21 @@ static int __init hardlockup_panic_setup
 }
 __setup("nmi_watchdog=", hardlockup_panic_setup);
 
-#endif
+# ifdef CONFIG_SMP
+int __read_mostly sysctl_hardlockup_all_cpu_backtrace;
 
-#ifdef CONFIG_SOFTLOCKUP_DETECTOR
-int __read_mostly soft_watchdog_enabled;
-#endif
+static int __init hardlockup_all_cpu_backtrace_setup(char *str)
+{
+	sysctl_hardlockup_all_cpu_backtrace = !!simple_strtol(str, NULL, 0);
+	return 1;
+}
+__setup("hardlockup_all_cpu_backtrace=", hardlockup_all_cpu_backtrace_setup);
+# endif /* CONFIG_SMP */
+#endif /* CONFIG_HARDLOCKUP_DETECTOR */
 
 int __read_mostly watchdog_user_enabled;
 int __read_mostly watchdog_thresh = 10;
 
-#ifdef CONFIG_SMP
-int __read_mostly sysctl_softlockup_all_cpu_backtrace;
-int __read_mostly sysctl_hardlockup_all_cpu_backtrace;
-#endif
 struct cpumask watchdog_cpumask __read_mostly;
 unsigned long *watchdog_cpumask_bits = cpumask_bits(&watchdog_cpumask);
 
@@ -166,22 +167,14 @@ static int __init nosoftlockup_setup(cha
 __setup("nosoftlockup", nosoftlockup_setup);
 
 #ifdef CONFIG_SMP
+int __read_mostly sysctl_softlockup_all_cpu_backtrace;
+
 static int __init softlockup_all_cpu_backtrace_setup(char *str)
 {
-	sysctl_softlockup_all_cpu_backtrace =
-		!!simple_strtol(str, NULL, 0);
+	sysctl_softlockup_all_cpu_backtrace = !!simple_strtol(str, NULL, 0);
 	return 1;
 }
 __setup("softlockup_all_cpu_backtrace=", softlockup_all_cpu_backtrace_setup);
-#ifdef CONFIG_HARDLOCKUP_DETECTOR
-static int __init hardlockup_all_cpu_backtrace_setup(char *str)
-{
-	sysctl_hardlockup_all_cpu_backtrace =
-		!!simple_strtol(str, NULL, 0);
-	return 1;
-}
-__setup("hardlockup_all_cpu_backtrace=", hardlockup_all_cpu_backtrace_setup);
-#endif
 #endif
 
 static void __lockup_detector_cleanup(void);

[toc] | [prev] | [next] | [standalone]

#1723920 — [patch 06/29] lockup_detector: Rework cpu hotplug locking

The watchdog proc interface causes extensive recursive locking of the cpu
hotplug percpu rwsem, which is deadlock prone.

Replace the get/put_online_cpus() pairs with cpu_hotplug_disable()/enable()
calls for now. Later patches will remove that requirement completely.

Reported-by: Borislav Petkov <bp@alien8.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 kernel/watchdog.c |   12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

--- a/kernel/watchdog.c
+++ b/kernel/watchdog.c
@@ -703,7 +703,7 @@ static int proc_watchdog_common(int whic
 	int err, old, new;
 	int *watchdog_param = (int *)table->data;
 
-	get_online_cpus();
+	cpu_hotplug_disable();
 	mutex_lock(&watchdog_proc_mutex);
 
 	/*
@@ -752,7 +752,7 @@ static int proc_watchdog_common(int whic
 	}
 out:
 	mutex_unlock(&watchdog_proc_mutex);
-	put_online_cpus();
+	cpu_hotplug_enable();
 	return err;
 }
 
@@ -794,7 +794,7 @@ int proc_watchdog_thresh(struct ctl_tabl
 {
 	int err, old, new;
 
-	get_online_cpus();
+	cpu_hotplug_disable();
 	mutex_lock(&watchdog_proc_mutex);
 
 	old = ACCESS_ONCE(watchdog_thresh);
@@ -818,7 +818,7 @@ int proc_watchdog_thresh(struct ctl_tabl
 	}
 out:
 	mutex_unlock(&watchdog_proc_mutex);
-	put_online_cpus();
+	cpu_hotplug_enable();
 	return err;
 }
 
@@ -833,7 +833,7 @@ int proc_watchdog_cpumask(struct ctl_tab
 {
 	int err;
 
-	get_online_cpus();
+	cpu_hotplug_disable();
 	mutex_lock(&watchdog_proc_mutex);
 
 	err = proc_do_large_bitmap(table, write, buffer, lenp, ppos);
@@ -856,7 +856,7 @@ int proc_watchdog_cpumask(struct ctl_tab
 	}
 
 	mutex_unlock(&watchdog_proc_mutex);
-	put_online_cpus();
+	cpu_hotplug_enable();
 	return err;
 }
 

[toc] | [prev] | [next] | [standalone]

#1723921 — [patch 14/29] lockup_detector: Split out cpumask write function

Split the write part of the cpumask proc handler out into a separate helper
to avoid deep indentation. This also reduces the patch complexity in the
following cleanups.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 kernel/watchdog.c |   40 +++++++++++++++++++++-------------------
 1 file changed, 21 insertions(+), 19 deletions(-)

--- a/kernel/watchdog.c
+++ b/kernel/watchdog.c
@@ -786,10 +786,29 @@ static int watchdog_update_cpus(void)
 	if (IS_ENABLED(CONFIG_SOFTLOCKUP_DETECTOR)) {
 		return smpboot_update_cpumask_percpu_thread(&watchdog_threads,
 							    &watchdog_cpumask);
+		__lockup_detector_cleanup();
 	}
 	return 0;
 }
 
+static void proc_watchdog_cpumask_update(void)
+{
+	/* Remove impossible cpus to keep sysctl output clean. */
+	cpumask_and(&watchdog_cpumask, &watchdog_cpumask, cpu_possible_mask);
+
+	if (watchdog_running) {
+		/*
+		 * Failure would be due to being unable to allocate a
+		 * temporary cpumask, so we are likely not in a position to
+		 * do much else to make things better.
+		 */
+		if (watchdog_update_cpus() != 0)
+			pr_err("cpumask update failed\n");
+	}
+
+	watchdog_nmi_reconfigure();
+}
+
 /*
  * The cpumask is the mask of possible cpus that the watchdog can run
  * on, not the mask of cpus it is actually running on.  This allows the
@@ -805,30 +824,13 @@ int proc_watchdog_cpumask(struct ctl_tab
 	mutex_lock(&watchdog_mutex);
 
 	err = proc_do_large_bitmap(table, write, buffer, lenp, ppos);
-	if (!err && write) {
-		/* Remove impossible cpus to keep sysctl output cleaner. */
-		cpumask_and(&watchdog_cpumask, &watchdog_cpumask,
-			    cpu_possible_mask);
-
-		if (watchdog_running) {
-			/*
-			 * Failure would be due to being unable to allocate
-			 * a temporary cpumask, so we are likely not in a
-			 * position to do much else to make things better.
-			 */
-			if (watchdog_update_cpus() != 0)
-				pr_err("cpumask update failed\n");
-		}
-
-		watchdog_nmi_reconfigure();
-		__lockup_detector_cleanup();
-	}
+	if (!err && write)
+		proc_watchdog_cpumask_update();
 
 	mutex_unlock(&watchdog_mutex);
 	cpu_hotplug_enable();
 	return err;
 }
-
 #endif /* CONFIG_SYSCTL */
 
 void __init lockup_detector_init(void)

[toc] | [prev] | [next] | [standalone]

#1723923 — [patch 11/29] lockup_detector: Remove park_in_progress hackery

b94f51183b06 ("kernel/watchdog: prevent false hardlockup on overloaded
system") tries to fix the following issue:

 watchdog_stop()
	hrtimer_cancel()
	perf_nmi_event_stop()

If the task gets preempted after canceling the hrtimer or the VM gets
scheduled out long enough, then there is a chance that the next NMI will
see a stale hrtimer interrupt count and trigger a false positive hard
lockup splat.

That commit added a complete abstrusity with a atomic variable telling the
NMI watchdog function that stop is in progress. This is so stupid, that
it's not funny anymore.

Of course the patch missed that the same issue can happen in start:

 watchdog_start()
	perf_nmi_event_start()
	hrtimer_start()

The same effect can be achieved by reversing the start/stop order:

 watchdog_stop()
	perf_nmi_event_stop()
	hrtimer_cancel()

 watchdog_start()
	hrtimer_start()
	perf_nmi_event_start()

Get rid of the nonsense.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 include/linux/nmi.h   |    1 -
 kernel/watchdog.c     |   39 ++++++++++++++++++---------------------
 kernel/watchdog_hld.c |    7 ++-----
 3 files changed, 20 insertions(+), 27 deletions(-)

--- a/include/linux/nmi.h
+++ b/include/linux/nmi.h
@@ -27,7 +27,6 @@ extern void touch_softlockup_watchdog_sy
 extern void touch_all_softlockup_watchdogs(void);
 extern unsigned int  softlockup_panic;
 extern int soft_watchdog_enabled;
-extern atomic_t watchdog_park_in_progress;
 #else
 static inline void touch_softlockup_watchdog_sched(void)
 {
--- a/kernel/watchdog.c
+++ b/kernel/watchdog.c
@@ -134,8 +134,6 @@ void __weak watchdog_nmi_reconfigure(voi
 #define for_each_watchdog_cpu(cpu) \
 	for_each_cpu_and((cpu), cpu_online_mask, &watchdog_cpumask)
 
-atomic_t watchdog_park_in_progress = ATOMIC_INIT(0);
-
 static u64 __read_mostly sample_period;
 
 static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts);
@@ -320,8 +318,7 @@ static enum hrtimer_restart watchdog_tim
 	int duration;
 	int softlockup_all_cpu_backtrace = sysctl_softlockup_all_cpu_backtrace;
 
-	if (!watchdog_enabled ||
-	    atomic_read(&watchdog_park_in_progress) != 0)
+	if (!watchdog_enabled)
 		return HRTIMER_NORESTART;
 
 	/* kick the hardlockup detector */
@@ -435,33 +432,38 @@ static void watchdog_set_prio(unsigned i
 
 static void watchdog_enable(unsigned int cpu)
 {
-	struct hrtimer *hrtimer = raw_cpu_ptr(&watchdog_hrtimer);
+	struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
 
-	/* kick off the timer for the hardlockup detector */
+	/*
+	 * Start the timer first to prevent the NMI watchdog triggering
+	 * before the timer has a chance to fire.
+	 */
 	hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 	hrtimer->function = watchdog_timer_fn;
+	hrtimer_start(hrtimer, ns_to_ktime(sample_period),
+		      HRTIMER_MODE_REL_PINNED);
 
+	/* Initialize timestamp */
+	__touch_watchdog();
 	/* Enable the perf event */
 	watchdog_nmi_enable(cpu);
 
-	/* done here because hrtimer_start can only pin to smp_processor_id() */
-	hrtimer_start(hrtimer, ns_to_ktime(sample_period),
-		      HRTIMER_MODE_REL_PINNED);
-
-	/* initialize timestamp */
 	watchdog_set_prio(SCHED_FIFO, MAX_RT_PRIO - 1);
-	__touch_watchdog();
 }
 
 static void watchdog_disable(unsigned int cpu)
 {
-	struct hrtimer *hrtimer = raw_cpu_ptr(&watchdog_hrtimer);
+	struct hrtimer *hrtimer = this_cpu_ptr(&watchdog_hrtimer);
 
 	watchdog_set_prio(SCHED_NORMAL, 0);
-	hrtimer_cancel(hrtimer);
-	/* disable the perf event */
-	watchdog_nmi_disable(cpu);
+	/*
+	 * Disable the perf event first. That prevents that a large delay
+	 * between disabling the timer and disabling the perf event causes
+	 * the perf NMI to detect a false positive.
+	 */
 	hardlockup_detector_perf_disable();
+	watchdog_nmi_disable(cpu);
+	hrtimer_cancel(hrtimer);
 }
 
 static void watchdog_cleanup(unsigned int cpu, bool online)
@@ -517,16 +519,11 @@ static int watchdog_park_threads(void)
 {
 	int cpu, ret = 0;
 
-	atomic_set(&watchdog_park_in_progress, 1);
-
 	for_each_watchdog_cpu(cpu) {
 		ret = kthread_park(per_cpu(softlockup_watchdog, cpu));
 		if (ret)
 			break;
 	}
-
-	atomic_set(&watchdog_park_in_progress, 0);
-
 	return ret;
 }
 
--- a/kernel/watchdog_hld.c
+++ b/kernel/watchdog_hld.c
@@ -106,15 +106,12 @@ static struct perf_event_attr wd_hw_attr
 
 /* Callback function for perf event subsystem */
 static void watchdog_overflow_callback(struct perf_event *event,
-		 struct perf_sample_data *data,
-		 struct pt_regs *regs)
+				       struct perf_sample_data *data,
+				       struct pt_regs *regs)
 {
 	/* Ensure the watchdog never gets throttled */
 	event->hw.interrupts = 0;
 
-	if (atomic_read(&watchdog_park_in_progress) != 0)
-		return;
-
 	if (__this_cpu_read(watchdog_nmi_touch) == true) {
 		__this_cpu_write(watchdog_nmi_touch, false);
 		return;

[toc] | [prev] | [next] | [standalone]

#1726027 — Re: [patch 11/29] lockup_detector: Remove park_in_progress hackery

On Mon, Sep 04, 2017 at 01:09:06PM +0200, Ulrich Obergfell wrote:

> - A thread hogs CPU N (soft lockup) so that watchdog/N is unable to run.
> - A user re-configures 'watchdog_thresh' on the fly. The reconfiguration
>   requires parking/unparking of all watchdog threads.

This is where you fail, its silly to require parking for
reconfiguration.

[toc] | [prev] | [next] | [standalone]

Page 1 of 3  [1] 2 3  Next page →

Back to top | Article view | linux.kernel

csiph-web