The patch titled
hrt/dynticks hotplug fix
has been removed from the -mm tree. Its filename was
hrt-dynticks-hotplug-fix.patch
This patch was dropped because it is obsolete
------------------------------------------------------
Subject: hrt/dynticks hotplug fix
From: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
The clockevents code didn't do CPU hotplug (and thus SMP resume/suspend)
logistics properly. [read: resume under highres+dynticks didnt work] There
were multiple problems:
- the local APIC clockevents driver was not unregistered on CPU-down,
but the CPU bootup sequence reinitialized it. This was mostly latent
before but now that it's list managed it became a double list_add():
[<c023b8fb>] list_add+0xa/0xf
[<c01420ff>] clockevents_register_device+0x4e/0x6b
[<c0119793>] setup_APIC_timer+0x47/0x4c
[<c01197a0>] setup_secondary_APIC_clock+0x8/0xa
[<c0118964>] start_secondary+0xe4/0x346
- secondly, when bringing a CPU down we didnt clear its scheduler-tick
hrtimer status, which on one box resulted in a NULL pointer
dereference during CPU hot-unplug:
EIP is at clockevents_program_event+0x38/0xac
Call Trace:
[<c0142df1>] tick_program_event+0x2d/0x55
[<c013ed19>] hrtimer_reprogram+0x60/0x84
[<c013edd5>] enqueue_hrtimer+0x98/0x11e
[<c013f866>] hrtimer_start+0xdd/0xf4
[<c01435a4>] tick_nohz_stop_sched_tick+0x179/0x1f2
[<c012e83d>] irq_exit+0x76/0x83
[<c010671b>] do_IRQ+0xf3/0x10c
[<c0104a76>] common_interrupt+0x2e/0x34
[<c011f88f>] native_safe_halt+0x5/0x7
[<c0102e9c>] cpu_idle+0xca/0x143
[<c0118bbe>] start_secondary+0x33e/0x346
- thirdly, the double hrtimer init also resulted in the sched-tick
hrtimer expiry being executed in softirq context, which is bad
locking-wise and lockdep warned about that:
=================================
[ INFO: inconsistent lock state ]
[ 2.6.20-rc6-rt2 #26
---------------------------------
inconsistent {in-hardirq-W} -> {hardirq-on-W} usage.
softirq-timer/1/2031 [HC0[0]:SC0[0]:HE1:SE1] takes:
(xtime_lock){+...}, at: [<c0143217>] tick_do_update_jiffies64+0x17/0xd0
All these problems are gone with this patch, we now properly get notified of
CPU dead events and properly use/unuse register/unregister the clockevent
devices. The suspend/resume bits of the clockevents code survived from a
really old version and only worked by miracle. It's now pretty OK.
Suspend/resume with high-res and dynticks enabled was successfully tested on
multiple boxes, including SMP ones.
Signed-off-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
Signed-off-by: Ingo Molnar <mingo@xxxxxxx>
Cc: john stultz <johnstul@xxxxxxxxxx>
Cc: Roman Zippel <zippel@xxxxxxxxxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxx>
---
include/linux/clockchips.h | 2 -
kernel/hrtimer.c | 1
kernel/time/clockevents.c | 35 ++++++++++++-----------
kernel/time/tick-broadcast.c | 49 ++++++++++++++++++++++-----------
kernel/time/tick-common.c | 28 ++++++++++++++++--
kernel/time/tick-internal.h | 20 +++----------
kernel/time/tick-sched.c | 2 +
kernel/timer.c | 2 -
8 files changed, 86 insertions(+), 53 deletions(-)
diff -puN include/linux/clockchips.h~hrt-dynticks-hotplug-fix include/linux/clockchips.h
--- a/include/linux/clockchips.h~hrt-dynticks-hotplug-fix
+++ a/include/linux/clockchips.h
@@ -34,6 +34,7 @@ enum clock_event_nofitiers {
CLOCK_EVT_NOTIFY_BROADCAST_EXIT,
CLOCK_EVT_NOTIFY_SUSPEND,
CLOCK_EVT_NOTIFY_RESUME,
+ CLOCK_EVT_NOTIFY_CPU_DEAD,
};
/*
@@ -129,7 +130,6 @@ extern void clockevents_unregister_notif
extern int clockevents_program_event(struct clock_event_device *dev,
ktime_t expires);
-extern void clockevents_resume_events(void);
extern void clockevents_notify(unsigned long reason, void *arg);
#else
diff -puN kernel/hrtimer.c~hrt-dynticks-hotplug-fix kernel/hrtimer.c
--- a/kernel/hrtimer.c~hrt-dynticks-hotplug-fix
+++ a/kernel/hrtimer.c
@@ -1384,6 +1384,7 @@ static int __cpuinit hrtimer_cpu_notify(
#ifdef CONFIG_HOTPLUG_CPU
case CPU_DEAD:
+ clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &cpu);
migrate_hrtimers(cpu);
break;
#endif
diff -puN kernel/time/clockevents.c~hrt-dynticks-hotplug-fix kernel/time/clockevents.c
--- a/kernel/time/clockevents.c~hrt-dynticks-hotplug-fix
+++ a/kernel/time/clockevents.c
@@ -248,26 +248,27 @@ void clockevents_notify(unsigned long re
{
spin_lock(&clockevents_lock);
clockevents_do_notify(reason, arg);
- spin_unlock(&clockevents_lock);
-}
-EXPORT_SYMBOL_GPL(clockevents_notify);
-void clockevents_do_resume_events(void *arg)
-{
- spin_lock(&clockevents_lock);
- clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
+ switch (reason) {
+ case CLOCK_EVT_NOTIFY_CPU_DEAD:
+ /*
+ * Unregister the clock event devices which were
+ * released from the users in the notify chain.
+ */
+ while (!list_empty(&clockevents_released)) {
+ struct clock_event_device *dev;
+
+ dev = list_entry(clockevents_released.next,
+ struct clock_event_device, list);
+ list_del(&dev->list);
+ }
+ break;
+ default:
+ break;
+ }
spin_unlock(&clockevents_lock);
}
-
-/**
- * clockevents_resume_events - resume the active clock devices
- *
- * Called after timekeeping is functional again
- */
-void clockevents_resume_events(void)
-{
- on_each_cpu(clockevents_do_resume_events, NULL, 0, 1);
-}
+EXPORT_SYMBOL_GPL(clockevents_notify);
#ifdef CONFIG_SYSFS
diff -puN kernel/timer.c~hrt-dynticks-hotplug-fix kernel/timer.c
--- a/kernel/timer.c~hrt-dynticks-hotplug-fix
+++ a/kernel/timer.c
@@ -995,7 +995,7 @@ static int timekeeping_resume(struct sys
timekeeping_suspended = 0;
write_sequnlock_irqrestore(&xtime_lock, flags);
- clockevents_resume_events();
+ clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
/* Resume hrtimers */
clock_was_set();
diff -puN kernel/time/tick-broadcast.c~hrt-dynticks-hotplug-fix kernel/time/tick-broadcast.c
--- a/kernel/time/tick-broadcast.c~hrt-dynticks-hotplug-fix
+++ a/kernel/time/tick-broadcast.c
@@ -261,17 +261,25 @@ void tick_set_periodic_handler(struct cl
}
/*
- * Called with irqs disabled
+ * Remove a CPU from broadcasting
*/
-void tick_do_resume(int cpu)
+void tick_shutdown_broadcast(unsigned int *cpup)
{
- unsigned long reason;
+ struct clock_event_device *bc;
+ unsigned long flags;
+ unsigned int cpu = *cpup;
+
+ spin_lock_irqsave(&tick_broadcast_lock, flags);
- reason = cpu_isset(cpu, tick_broadcast_mask) ?
- CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
- tick_do_broadcast_on_off(&reason);
+ bc = tick_broadcast_device.evtdev;
+ cpu_clear(cpu, tick_broadcast_mask);
- tick_oneshot_resume(cpu);
+ if (tick_broadcast_device.mode == TICKDEV_MODE_PERIODIC) {
+ if (bc && cpus_empty(tick_broadcast_mask))
+ clockevents_set_mode(bc, CLOCK_EVT_MODE_SHUTDOWN);
+ }
+
+ spin_unlock_irqrestore(&tick_broadcast_lock, flags);
}
#ifdef CONFIG_TICK_ONESHOT
@@ -434,16 +442,27 @@ void tick_broadcast_switch_to_oneshot(vo
spin_unlock_irqrestore(&tick_broadcast_lock, flags);
}
-/**
- * Called with irqs disabled
+
+/*
+ * Remove a dead CPU from broadcasting
*/
-void tick_oneshot_resume(int cpu)
+void tick_shutdown_broadcast_oneshot(unsigned int *cpup)
{
- unsigned long reason;
+ struct clock_event_device *bc;
+ unsigned long flags;
+ unsigned int cpu = *cpup;
+
+ spin_lock_irqsave(&tick_broadcast_lock, flags);
+
+ bc = tick_broadcast_device.evtdev;
+ cpu_clear(cpu, tick_broadcast_oneshot_mask);
+
+ if (tick_broadcast_device.mode == TICKDEV_MODE_ONESHOT) {
+ if (bc && cpus_empty(tick_broadcast_oneshot_mask))
+ clockevents_set_mode(bc, CLOCK_EVT_MODE_SHUTDOWN);
+ }
- reason = cpu_isset(cpu, tick_broadcast_oneshot_mask) ?
- CLOCK_EVT_NOTIFY_BROADCAST_ENTER :
- CLOCK_EVT_NOTIFY_BROADCAST_EXIT;
- tick_broadcast_oneshot_control(reason);
+ spin_unlock_irqrestore(&tick_broadcast_lock, flags);
}
+
#endif
diff -puN kernel/time/tick-common.c~hrt-dynticks-hotplug-fix kernel/time/tick-common.c
--- a/kernel/time/tick-common.c~hrt-dynticks-hotplug-fix
+++ a/kernel/time/tick-common.c
@@ -271,14 +271,29 @@ out:
}
/*
- * Resume tick devices
+ * Shutdown an event device on a given cpu:
+ *
+ * This is called on a life CPU, when a CPU is dead. So we cannot
+ * access the hardware device itself.
+ * We just set the mode and remove it from the lists.
*/
-static void tick_resume(void)
+static void tick_shutdown(unsigned int *cpup)
{
+ struct tick_device *td = &per_cpu(tick_cpu_device, *cpup);
+ struct clock_event_device *dev = td->evtdev;
unsigned long flags;
spin_lock_irqsave(&tick_device_lock, flags);
- tick_do_resume(smp_processor_id());
+ td->mode = TICKDEV_MODE_PERIODIC;
+ if (dev) {
+ /*
+ * Prevent that the clock events layer tries to call
+ * the set mode function!
+ */
+ dev->mode = CLOCK_EVT_MODE_UNUSED;
+ clockevents_exchange_device(dev, NULL);
+ td->evtdev = NULL;
+ }
spin_unlock_irqrestore(&tick_device_lock, flags);
}
@@ -305,7 +320,12 @@ static int tick_notify(struct notifier_b
case CLOCK_EVT_NOTIFY_RESUME:
tick_resume_jiffy_update();
- tick_resume();
+ break;
+
+ case CLOCK_EVT_NOTIFY_CPU_DEAD:
+ tick_shutdown_broadcast_oneshot(dev);
+ tick_shutdown_broadcast(dev);
+ tick_shutdown(dev);
break;
default:
diff -puN kernel/time/tick-internal.h~hrt-dynticks-hotplug-fix kernel/time/tick-internal.h
--- a/kernel/time/tick-internal.h~hrt-dynticks-hotplug-fix
+++ a/kernel/time/tick-internal.h
@@ -20,12 +20,12 @@ extern void tick_resume_jiffy_update(voi
extern int tick_program_event(ktime_t expires, int force);
extern void tick_oneshot_notify(void);
extern int tick_switch_to_oneshot(void (*handler)(struct clock_event_device *));
-extern void tick_oneshot_resume(int cpu);
# ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
extern void tick_broadcast_setup_oneshot(struct clock_event_device *bc);
extern void tick_broadcast_oneshot_control(unsigned long reason);
extern void tick_broadcast_switch_to_oneshot(void);
+extern void tick_shutdown_broadcast_oneshot(unsigned int *cpup);
# else /* BROADCAST */
static inline void tick_broadcast_setup_oneshot(struct clock_event_device *bc)
{
@@ -33,6 +33,7 @@ static inline void tick_broadcast_setup_
}
static inline void tick_broadcast_oneshot_control(unsigned long reason) { }
static inline void tick_broadcast_switch_to_oneshot(void) { }
+static inline void tick_shutdown_broadcast_oneshot(unsigned int *cpup) { }
# endif /* !BROADCAST */
#else /* !ONESHOT */
@@ -48,13 +49,13 @@ static inline int tick_program_event(kti
return 0;
}
static inline void tick_resume_jiffy_update(void) { }
-static inline void tick_oneshot_resume(int cpu) { }
static inline void tick_oneshot_notify(void) { }
static inline void tick_broadcast_setup_oneshot(struct clock_event_device *bc)
{
BUG();
}
static inline void tick_broadcast_oneshot_control(unsigned long reason) { }
+static inline void tick_shutdown_broadcast_oneshot(unsigned int *cpup) { }
#endif /* !TICK_ONESHOT */
/*
@@ -67,10 +68,10 @@ extern int tick_device_uses_broadcast(st
extern int tick_check_broadcast_device(struct clock_event_device *dev);
extern int tick_is_broadcast_device(struct clock_event_device *dev);
extern void tick_broadcast_on_off(unsigned long reason, int *oncpu);
+extern void tick_shutdown_broadcast(unsigned int *cpup);
extern void
tick_set_periodic_handler(struct clock_event_device *dev, int broadcast);
-extern void tick_do_resume(int cpu);
#else /* !BROADCAST */
@@ -90,6 +91,7 @@ static inline int tick_device_uses_broad
}
static inline void tick_do_periodic_broadcast(struct clock_event_device *d) { }
static inline void tick_broadcast_on_off(unsigned long reason, int *oncpu) { }
+static inline void tick_shutdown_broadcast(unsigned int *cpup) { }
/*
* Set the periodic handler in non broadcast mode
@@ -99,18 +101,6 @@ static inline void tick_set_periodic_han
{
dev->event_handler = tick_handle_periodic;
}
-/*
- * Called with irqs disabled
- */
-static inline void tick_do_resume(int cpu)
-{
- struct tick_device *td = &per_cpu(tick_cpu_device, cpu);
-
- if (td->mode == TICKDEV_MODE_PERIODIC)
- tick_setup_periodic(td->evtdev, 0);
- else
- tick_oneshot_resume(cpu);
-}
#endif /* !BROADCAST */
/*
diff -puN kernel/time/tick-sched.c~hrt-dynticks-hotplug-fix kernel/time/tick-sched.c
--- a/kernel/time/tick-sched.c~hrt-dynticks-hotplug-fix
+++ a/kernel/time/tick-sched.c
@@ -502,6 +502,8 @@ void tick_cancel_sched_timer(int cpu)
if (ts->sched_timer.base)
hrtimer_cancel(&ts->sched_timer);
+ ts->tick_stopped = 0;
+ ts->nohz_mode = NOHZ_MODE_INACTIVE;
}
#endif /* HIGH_RES_TIMERS */
_
Patches currently in -mm which might be from tglx@xxxxxxxxxxxxx are
bugfixes-pci-devices-get-assigned-redundant-irqs.patch
use-cycle_t-instead-of-u64-in-struct-time_interpolator.patch
proc-remove-useless-and-buggy-nlink-settings.patch
add-irq-flag-to-disable-balancing-for-an-interrupt.patch
add-a-functions-to-handle-interrupt-affinity-setting.patch
add-a-functions-to-handle-interrupt-affinity-setting-alpha-fix.patch
hz-free-ntp.patch
uninline-jiffiesh-functions.patch
fix-multiple-conversion-bugs-in-msecs_to_jiffies.patch
fix-timeout-overflow-with-jiffies.patch
gtod-persistent-clock-support.patch
i386-use-gtod-persistent-clock-support.patch
i386-remove-useless-code-in-tscc.patch
simplify-the-registration-of-clocksources.patch
x86-rewrite-smp-tsc-sync-code.patch
clocksource-replace-is_continuous-by-a-flag-field.patch
clocksource-replace-is_continuous-by-a-flag-field-fix.patch
clocksource-fixup-is_continous-changes-on-arm.patch
clocksource-fixup-is_continous-changes-on-avr32.patch
clocksource-fixup-is_continous-changes-on-s390.patch
clocksource-fixup-is_continous-changes-on-mips.patch
clocksource-remove-the-update-callback.patch
clocksource-add-verification-watchdog-helper.patch
mark-tsc-on-geodelx-reliable.patch
uninline-irq_enter.patch
fix-cascade-lookup-of-next_timer_interrupt.patch
extend-next_timer_interrupt-to-use-a-reference-jiffie.patch
hrtimers-namespace-and-enum-cleanup.patch
hrtimers-namespace-and-enum-cleanup-vs-git-input.patch
hrtimers-cleanup-locking.patch
hrtimers-add-state-tracking.patch
hrtimers-clean-up-callback-tracking.patch
hrtimers-move-and-add-documentation.patch
acpi-fix-missing-include-for-up.patch
acpi-keep-track-of-timer-broadcasting.patch
allow-early-access-to-the-power-management-timer.patch
i386-apic-clean-up-the-apic-code.patch
clockevents-add-core-functionality.patch
tick-management-core-functionality.patch
tick-management-broadcast-functionality.patch
tick-management-dyntick--highres-functionality.patch
clockevents-i383-drivers.patch
i386-rework-local-apic-timer-calibration.patch
i386-prepare-for-dyntick.patch
i386-prepare-nmi-watchdog-for-dynticks.patch
hrtimers-add-high-resolution-timer-support.patch
hrtimers-prevent-possible-itimer-dos.patch
add-debugging-feature-proc-timer_stat.patch
add-debugging-feature-proc-timer_list.patch
add-sysrq-q-to-print-timer_list-debug-info.patch
hrt-dynticks-hotplug-fix.patch
hrt-dynticks-hotplug-fix-fix.patch
generic-vsyscall-gtod-support-for-generic_time.patch
generic-vsyscall-gtod-support-for-generic_time-tidy.patch
time-x86_64-hpet_address-cleanup.patch
revert-x86_64-mm-ignore-long-smi-interrupts-in-clock-calibration.patch
time-x86_64-split-x86_64-kernel-timec-up.patch
time-x86_64-split-x86_64-kernel-timec-up-tidy.patch
time-x86_64-split-x86_64-kernel-timec-up-fix.patch
reapply-x86_64-mm-ignore-long-smi-interrupts-in-clock-calibration.patch
time-x86_64-convert-x86_64-to-use-generic_time.patch
time-x86_64-convert-x86_64-to-use-generic_time-fix.patch
time-x86_64-convert-x86_64-to-use-generic_time-tidy.patch
time-x86_64-hpet-fixup-clocksource-changes.patch
time-x86_64-tsc-fixup-clocksource-changes.patch
time-x86_64-re-enable-vsyscall-support-for-x86_64.patch
time-x86_64-re-enable-vsyscall-support-for-x86_64-tidy.patch
scheduled-removal-of-sa_xxx-interrupt-flags-fixups.patch
scheduled-removal-of-sa_xxx-interrupt-flags-fixups-2.patch
scheduled-removal-of-sa_xxx-interrupt-flags.patch
-
To unsubscribe from this list: send the line "unsubscribe mm-commits" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
