[PATCH v2] ftrace based hard lockup detector
From: Frederic Weisbecker
Date: Sun Jan 18 2009 - 20:20:11 EST
[Sorry, I forgot the function tracer dependency]
--
Like the NMI watchdog, this feature try to detect hard lockups by
lurking at the non-progress of the timer interrupts.
You can enable it at boot time by passing the ftrace_hardlockup parameter.
I plan to add a debugfs file to enable/disable at runtime.
When a hardlockup is detected, it will print a backtrace. Perhaps it
would be good to print the locks held from lockdep too?
It only support x86 for the moment, because a kind
Signed-off-by: Frederic Weisbecker <fweisbec@xxxxxxxxx>
---
Documentation/kernel-parameters.txt | 4 +
arch/x86/Kconfig.debug | 15 ++++-
arch/x86/include/asm/nmi.h | 15 +++++
arch/x86/kernel/Makefile | 2 +
arch/x86/kernel/ftrace_hardlockup.c | 115 +++++++++++++++++++++++++++++++++++
arch/x86/kernel/nmi.c | 14 ----
include/linux/seqlock.h | 6 ++
7 files changed, 156 insertions(+), 15 deletions(-)
create mode 100644 arch/x86/kernel/ftrace_hardlockup.c
diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt
index ac613a6..9881426 100644
--- a/Documentation/kernel-parameters.txt
+++ b/Documentation/kernel-parameters.txt
@@ -774,6 +774,10 @@ and is between 256 and 4096 characters. It is defined in the file
ftrace_dump_on_oops
[ftrace] will dump the trace buffers on oops.
+ ftrace_hardlockup
+ [ftrace] will hook most of the kernel functions and
+ try to detect hard lockups.
+
gamecon.map[2|3]=
[HW,JOY] Multisystem joystick and NES/SNES/PSX pad
support via parallel port (up to 5 devices per port)
diff --git a/arch/x86/Kconfig.debug b/arch/x86/Kconfig.debug
index 097d79a..33ae6c5 100644
--- a/arch/x86/Kconfig.debug
+++ b/arch/x86/Kconfig.debug
@@ -367,5 +367,18 @@ config OPTIMIZE_INLINING
If unsure, say N.
-endmenu
+config FTRACE_HARLOCKUPS_DETECT
+ bool "Ftrace based hard-lockup detection"
+ select FUNCTION_TRACER
+ help
+ This options makes ftrace hooking most of the kernel functions, trying to
+ detect a hard lockup and then print a backtrace if so. Its check is
+ similar to the one performed by the nmi watchdog which lurks in the
+ non-progress of the timer interrupts. By default it is not enabled.
+ If you want to launch it, add the ftrace_hardlockup kernel parameter.
+ Enabling this option will bring almost no overhead, assuming you
+ enabled CONFIG_DYNAMIC_FTRACE. But once you activate it through the
+ ftrace_hardlockup parameter, your system will be a bit slower.
+
+endmenu
diff --git a/arch/x86/include/asm/nmi.h b/arch/x86/include/asm/nmi.h
index c45a0a5..9177b60 100644
--- a/arch/x86/include/asm/nmi.h
+++ b/arch/x86/include/asm/nmi.h
@@ -48,6 +48,21 @@ extern int unknown_nmi_panic;
void __trigger_all_cpu_backtrace(void);
#define trigger_all_cpu_backtrace() __trigger_all_cpu_backtrace()
+/*
+ * Take the local apic timer and PIT/HPET into account. We don't
+ * know which one is active, when we have highres/dyntick on
+ */
+static inline unsigned int get_timer_irqs(int cpu)
+{
+#ifdef CONFIG_X86_64
+ return read_pda(apic_timer_irqs) + read_pda(irq0_irqs);
+#else
+ return per_cpu(irq_stat, cpu).apic_timer_irqs +
+ per_cpu(irq_stat, cpu).irq0_irqs;
+#endif
+}
+
+
static inline void localise_nmi_watchdog(void)
{
if (nmi_watchdog == NMI_IO_APIC)
diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile
index eb07453..c488d76 100644
--- a/arch/x86/kernel/Makefile
+++ b/arch/x86/kernel/Makefile
@@ -13,6 +13,7 @@ CFLAGS_REMOVE_rtc.o = -pg
CFLAGS_REMOVE_paravirt-spinlocks.o = -pg
CFLAGS_REMOVE_ftrace.o = -pg
CFLAGS_REMOVE_early_printk.o = -pg
+CFLAGS_REMOVE_ftrace_hardlockup.o = -pg
endif
#
@@ -68,6 +69,7 @@ obj-$(CONFIG_X86_IO_APIC) += io_apic.o
obj-$(CONFIG_X86_REBOOTFIXUPS) += reboot_fixups_32.o
obj-$(CONFIG_DYNAMIC_FTRACE) += ftrace.o
obj-$(CONFIG_FUNCTION_GRAPH_TRACER) += ftrace.o
+obj-$(CONFIG_FTRACE_HARLOCKUPS_DETECT) += ftrace_hardlockup.o
obj-$(CONFIG_KEXEC) += machine_kexec_$(BITS).o
obj-$(CONFIG_KEXEC) += relocate_kernel_$(BITS).o crash.o
obj-$(CONFIG_CRASH_DUMP) += crash_dump_$(BITS).o
diff --git a/arch/x86/kernel/ftrace_hardlockup.c b/arch/x86/kernel/ftrace_hardlockup.c
new file mode 100644
index 0000000..5f61626
--- /dev/null
+++ b/arch/x86/kernel/ftrace_hardlockup.c
@@ -0,0 +1,115 @@
+/*
+ * Ftrace based Hard Lockup detector.
+ *
+ * Copyright (C) 2009 Frederic Weisbecker <fweisbec@xxxxxxxxx>
+ * Suggested by Ingo Molnar <mingo@xxxxxxx>
+ *
+ * Based on ideas and code of nmi.c which was
+ * started by Ingo Molnar.
+ *
+ */
+
+
+#include <linux/ftrace.h>
+#include <linux/cpumask.h>
+#include <linux/percpu.h>
+#include <linux/hrtimer.h>
+#include <linux/time.h>
+#include <asm/hardirq.h>
+#include <asm/nmi.h>
+
+
+/* Store the numbers of timer interrupts raised on one CPU. */
+struct last_timer_irq_stat {
+ int nb_irqs;
+ ktime_t last_seen;
+};
+
+static bool hardlockup_detect_enabled __read_mostly;
+
+static DEFINE_PER_CPU(struct last_timer_irq_stat, prev_timer_irqs);
+
+/* Protect from ftrace recursions */
+static DEFINE_PER_CPU(atomic_t, trace_recursion_wall);
+
+/* Mask of Cpus which already had a hard lockup backtrace printed */
+static cpumask_t backtrace_mask;
+
+
+/* The main check function */
+static void check_hardlockup(unsigned long ip, unsigned long parent_ip)
+{
+ struct last_timer_irq_stat *prev_stat;
+ atomic_t *trace_recursion_wall;
+ unsigned int nb_timer_irqs;
+ ktime_t elapsed, now;
+ int cpu;
+
+ /* We use a lot of per-cpu values here */
+ preempt_disable_notrace();
+
+ /* Protect against ftrace recursion */
+ trace_recursion_wall = &__get_cpu_var(trace_recursion_wall);
+ if (atomic_inc_return(trace_recursion_wall) != 1)
+ goto out;
+
+ /*
+ * We use ktime_get() here. And ktime uses a read lock
+ * on xtime_lock. So if we trace a function which holds a
+ * write lock on xtime_lock in this CPU, we will deadlock.
+ * Avoid it with this check.
+ */
+ if (unlikely(write_is_seqlocked(&xtime_lock)))
+ goto out;
+
+ cpu = raw_smp_processor_id();
+
+ nb_timer_irqs = get_timer_irqs(cpu);
+ prev_stat = &__get_cpu_var(prev_timer_irqs);
+
+ if (nb_timer_irqs != prev_stat->nb_irqs) {
+ prev_stat->nb_irqs = nb_timer_irqs;
+ prev_stat->last_seen = ktime_get();
+ goto out;
+ }
+
+ now = ktime_get();
+ elapsed = ktime_sub(now, prev_stat->last_seen);
+
+ /*
+ * Check if the timer interrupts didn't triggered for 5 secs.
+ * Don't spend time with a div, 30 bits right is near 10^9
+ */
+ if (ktime_to_ns(elapsed) >> 30 >= 5) {
+ /* Already printed ? */
+ if (cpu_test_and_set(cpu, backtrace_mask))
+ goto out;
+
+ printk(KERN_WARNING "ftrace: hard lockup detected\n");
+ dump_stack();
+ }
+
+out:
+ atomic_dec(trace_recursion_wall);
+ preempt_enable_no_resched_notrace();
+}
+
+static struct ftrace_ops hardlockup_detect_ops __read_mostly = {
+ .func = check_hardlockup
+};
+
+static __init int enable_ftrace_hardlockup(char *str)
+{
+ hardlockup_detect_enabled = true;
+ return 1;
+}
+__setup("ftrace_hardlockup", enable_ftrace_hardlockup);
+
+static int __init ftrace_hardlockup_init(void)
+{
+ if (hardlockup_detect_enabled)
+ register_ftrace_function(&hardlockup_detect_ops);
+
+ return 0;
+}
+device_initcall(ftrace_hardlockup_init);
diff --git a/arch/x86/kernel/nmi.c b/arch/x86/kernel/nmi.c
index 7228979..86225f8 100644
--- a/arch/x86/kernel/nmi.c
+++ b/arch/x86/kernel/nmi.c
@@ -76,20 +76,6 @@ static inline int mce_in_progress(void)
return 0;
}
-/*
- * Take the local apic timer and PIT/HPET into account. We don't
- * know which one is active, when we have highres/dyntick on
- */
-static inline unsigned int get_timer_irqs(int cpu)
-{
-#ifdef CONFIG_X86_64
- return read_pda(apic_timer_irqs) + read_pda(irq0_irqs);
-#else
- return per_cpu(irq_stat, cpu).apic_timer_irqs +
- per_cpu(irq_stat, cpu).irq0_irqs;
-#endif
-}
-
#ifdef CONFIG_SMP
/*
* The performance counters used by NMI_LOCAL_APIC don't trigger when
diff --git a/include/linux/seqlock.h b/include/linux/seqlock.h
index 632205c..545cb85 100644
--- a/include/linux/seqlock.h
+++ b/include/linux/seqlock.h
@@ -82,6 +82,12 @@ static inline int write_tryseqlock(seqlock_t *sl)
return ret;
}
+static inline int write_is_seqlocked(seqlock_t *sl)
+{
+ return spin_is_locked(&sl->lock);
+}
+
+
/* Start of read calculation -- fetch last complete writer token */
static __always_inline unsigned read_seqbegin(const seqlock_t *sl)
{
--
1.6.0.4
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/