ftrace: trace irq disabled critical timings
authorSteven Rostedt <srostedt@redhat.com>
Mon, 12 May 2008 19:20:42 +0000 (21:20 +0200)
committerThomas Gleixner <tglx@linutronix.de>
Fri, 23 May 2008 18:32:46 +0000 (20:32 +0200)
This patch adds latency tracing for critical timings
(how long interrupts are disabled for).

 "irqsoff" is added to /debugfs/tracing/available_tracers

Note:
  tracing_max_latency
    also holds the max latency for irqsoff (in usecs).
   (default to large number so one must start latency tracing)

  tracing_thresh
    threshold (in usecs) to always print out if irqs off
    is detected to be longer than stated here.
    If irq_thresh is non-zero, then max_irq_latency
    is ignored.

Here's an example of a trace with ftrace_enabled = 0

=======
preemption latency trace v1.1.5 on 2.6.24-rc7
Signed-off-by: Ingo Molnar <mingo@elte.hu>
--------------------------------------------------------------------
 latency: 100 us, #3/3, CPU#1 | (M:rt VP:0, KP:0, SP:0 HP:0 #P:2)
    -----------------
    | task: swapper-0 (uid:0 nice:0 policy:0 rt_prio:0)
    -----------------
 => started at: _spin_lock_irqsave+0x2a/0xb7
 => ended at:   _spin_unlock_irqrestore+0x32/0x5f

                 _------=> CPU#
                / _-----=> irqs-off
               | / _----=> need-resched
               || / _---=> hardirq/softirq
               ||| / _--=> preempt-depth
               |||| /
               |||||     delay
   cmd     pid ||||| time  |   caller
      \   /    |||||   \   |   /
 swapper-0     1d.s3    0us+: _spin_lock_irqsave+0x2a/0xb7 (e1000_update_stats+0x47/0x64c [e1000])
 swapper-0     1d.s3  100us : _spin_unlock_irqrestore+0x32/0x5f (e1000_update_stats+0x641/0x64c [e1000])
 swapper-0     1d.s3  100us : trace_hardirqs_on_caller+0x75/0x89 (_spin_unlock_irqrestore+0x32/0x5f)

vim:ft=help
=======

And this is a trace with ftrace_enabled == 1

=======
preemption latency trace v1.1.5 on 2.6.24-rc7
--------------------------------------------------------------------
 latency: 102 us, #12/12, CPU#1 | (M:rt VP:0, KP:0, SP:0 HP:0 #P:2)
    -----------------
    | task: swapper-0 (uid:0 nice:0 policy:0 rt_prio:0)
    -----------------
 => started at: _spin_lock_irqsave+0x2a/0xb7
 => ended at:   _spin_unlock_irqrestore+0x32/0x5f

                 _------=> CPU#
                / _-----=> irqs-off
               | / _----=> need-resched
               || / _---=> hardirq/softirq
               ||| / _--=> preempt-depth
               |||| /
               |||||     delay
   cmd     pid ||||| time  |   caller
      \   /    |||||   \   |   /
 swapper-0     1dNs3    0us+: _spin_lock_irqsave+0x2a/0xb7 (e1000_update_stats+0x47/0x64c [e1000])
 swapper-0     1dNs3   46us : e1000_read_phy_reg+0x16/0x225 [e1000] (e1000_update_stats+0x5e2/0x64c [e1000])
 swapper-0     1dNs3   46us : e1000_swfw_sync_acquire+0x10/0x99 [e1000] (e1000_read_phy_reg+0x49/0x225 [e1000])
 swapper-0     1dNs3   46us : e1000_get_hw_eeprom_semaphore+0x12/0xa6 [e1000] (e1000_swfw_sync_acquire+0x36/0x99 [e1000])
 swapper-0     1dNs3   47us : __const_udelay+0x9/0x47 (e1000_read_phy_reg+0x116/0x225 [e1000])
 swapper-0     1dNs3   47us+: __delay+0x9/0x50 (__const_udelay+0x45/0x47)
 swapper-0     1dNs3   97us : preempt_schedule+0xc/0x84 (__delay+0x4e/0x50)
 swapper-0     1dNs3   98us : e1000_swfw_sync_release+0xc/0x55 [e1000] (e1000_read_phy_reg+0x211/0x225 [e1000])
 swapper-0     1dNs3   99us+: e1000_put_hw_eeprom_semaphore+0x9/0x35 [e1000] (e1000_swfw_sync_release+0x50/0x55 [e1000])
 swapper-0     1dNs3  101us : _spin_unlock_irqrestore+0xe/0x5f (e1000_update_stats+0x641/0x64c [e1000])
 swapper-0     1dNs3  102us : _spin_unlock_irqrestore+0x32/0x5f (e1000_update_stats+0x641/0x64c [e1000])
 swapper-0     1dNs3  102us : trace_hardirqs_on_caller+0x75/0x89 (_spin_unlock_irqrestore+0x32/0x5f)

vim:ft=help
=======

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
13 files changed:
arch/x86/kernel/process_64.c
arch/x86/lib/Makefile
arch/x86/lib/thunk_32.S [new file with mode: 0644]
arch/x86/lib/thunk_64.S
include/asm-x86/irqflags.h
include/linux/ftrace.h
include/linux/irqflags.h
kernel/fork.c
kernel/lockdep.c
kernel/printk.c
kernel/trace/Kconfig
kernel/trace/Makefile
kernel/trace/trace_irqsoff.c [new file with mode: 0644]

index e2319f3..dd349c9 100644 (file)
@@ -165,7 +165,10 @@ void cpu_idle(void)
                         */
                        local_irq_disable();
                        enter_idle();
+                       /* Don't trace irqs off for idle */
+                       stop_critical_timings();
                        idle();
+                       start_critical_timings();
                        /* In many cases the interrupt that ended idle
                           has already called exit_idle. But some idle
                           loops can be woken up without interrupt. */
index 76f60f5..84aa288 100644 (file)
@@ -5,6 +5,7 @@
 obj-$(CONFIG_SMP) := msr-on-cpu.o
 
 lib-y := delay_$(BITS).o
+lib-y += thunk_$(BITS).o
 lib-y += usercopy_$(BITS).o getuser_$(BITS).o putuser_$(BITS).o
 lib-y += memcpy_$(BITS).o
 
diff --git a/arch/x86/lib/thunk_32.S b/arch/x86/lib/thunk_32.S
new file mode 100644 (file)
index 0000000..650b11e
--- /dev/null
@@ -0,0 +1,47 @@
+/*
+ * Trampoline to trace irqs off. (otherwise CALLER_ADDR1 might crash)
+ * Copyright 2008 by Steven Rostedt, Red Hat, Inc
+ *  (inspired by Andi Kleen's thunk_64.S)
+ * Subject to the GNU public license, v.2. No warranty of any kind.
+ */
+
+       #include <linux/linkage.h>
+
+#define ARCH_TRACE_IRQS_ON                     \
+       pushl %eax;                             \
+       pushl %ecx;                             \
+       pushl %edx;                             \
+       call trace_hardirqs_on;                 \
+       popl %edx;                              \
+       popl %ecx;                              \
+       popl %eax;
+
+#define ARCH_TRACE_IRQS_OFF                    \
+       pushl %eax;                             \
+       pushl %ecx;                             \
+       pushl %edx;                             \
+       call trace_hardirqs_off;                \
+       popl %edx;                              \
+       popl %ecx;                              \
+       popl %eax;
+
+#ifdef CONFIG_TRACE_IRQFLAGS
+       /* put return address in eax (arg1) */
+       .macro thunk_ra name,func
+       .globl \name
+\name:
+       pushl %eax
+       pushl %ecx
+       pushl %edx
+       /* Place EIP in the arg1 */
+       movl 3*4(%esp), %eax
+       call \func
+       popl %edx
+       popl %ecx
+       popl %eax
+       ret
+       .endm
+
+       thunk_ra trace_hardirqs_on_thunk,trace_hardirqs_on_caller
+       thunk_ra trace_hardirqs_off_thunk,trace_hardirqs_off_caller
+#endif
index e009251..bf9a7d5 100644 (file)
@@ -2,6 +2,7 @@
  * Save registers before calling assembly functions. This avoids
  * disturbance of register allocation in some inline assembly constructs.
  * Copyright 2001,2002 by Andi Kleen, SuSE Labs.
+ * Added trace_hardirqs callers - Copyright 2007 Steven Rostedt, Red Hat, Inc.
  * Subject to the GNU public license, v.2. No warranty of any kind.
  */
 
 #endif 
        
 #ifdef CONFIG_TRACE_IRQFLAGS
-       thunk trace_hardirqs_on_thunk,trace_hardirqs_on
-       thunk trace_hardirqs_off_thunk,trace_hardirqs_off
+       /* put return address in rdi (arg1) */
+       .macro thunk_ra name,func
+       .globl \name
+\name:
+       CFI_STARTPROC
+       SAVE_ARGS
+       /* SAVE_ARGS pushs 9 elements */
+       /* the next element would be the rip */
+       movq 9*8(%rsp), %rdi
+       call \func
+       jmp  restore
+       CFI_ENDPROC
+       .endm
+
+       thunk_ra trace_hardirqs_on_thunk,trace_hardirqs_on_caller
+       thunk_ra trace_hardirqs_off_thunk,trace_hardirqs_off_caller
 #endif
 
 #ifdef CONFIG_DEBUG_LOCK_ALLOC
index c242527..24d71b1 100644 (file)
@@ -179,8 +179,6 @@ static inline void trace_hardirqs_fixup(void)
  * have a reliable stack. x86_64 only.
  */
 #define SWAPGS_UNSAFE_STACK    swapgs
-#define ARCH_TRACE_IRQS_ON             call trace_hardirqs_on_thunk
-#define ARCH_TRACE_IRQS_OFF            call trace_hardirqs_off_thunk
 #define ARCH_LOCKDEP_SYS_EXIT          call lockdep_sys_exit_thunk
 #define ARCH_LOCKDEP_SYS_EXIT_IRQ      \
        TRACE_IRQS_ON; \
@@ -192,24 +190,6 @@ static inline void trace_hardirqs_fixup(void)
        TRACE_IRQS_OFF;
 
 #else
-#define ARCH_TRACE_IRQS_ON                     \
-       pushl %eax;                             \
-       pushl %ecx;                             \
-       pushl %edx;                             \
-       call trace_hardirqs_on;                 \
-       popl %edx;                              \
-       popl %ecx;                              \
-       popl %eax;
-
-#define ARCH_TRACE_IRQS_OFF                    \
-       pushl %eax;                             \
-       pushl %ecx;                             \
-       pushl %edx;                             \
-       call trace_hardirqs_off;                \
-       popl %edx;                              \
-       popl %ecx;                              \
-       popl %eax;
-
 #define ARCH_LOCKDEP_SYS_EXIT                  \
        pushl %eax;                             \
        pushl %ecx;                             \
@@ -223,8 +203,8 @@ static inline void trace_hardirqs_fixup(void)
 #endif
 
 #ifdef CONFIG_TRACE_IRQFLAGS
-#  define TRACE_IRQS_ON                ARCH_TRACE_IRQS_ON
-#  define TRACE_IRQS_OFF       ARCH_TRACE_IRQS_OFF
+#  define TRACE_IRQS_ON                call trace_hardirqs_on_thunk;
+#  define TRACE_IRQS_OFF       call trace_hardirqs_off_thunk;
 #else
 #  define TRACE_IRQS_ON
 #  define TRACE_IRQS_OFF
index db8a5e7..0a20445 100644 (file)
@@ -50,4 +50,12 @@ extern void mcount(void);
 # define CALLER_ADDR5 0UL
 #endif
 
+#ifdef CONFIG_IRQSOFF_TRACER
+  extern void notrace time_hardirqs_on(unsigned long a0, unsigned long a1);
+  extern void notrace time_hardirqs_off(unsigned long a0, unsigned long a1);
+#else
+# define time_hardirqs_on(a0, a1)              do { } while (0)
+# define time_hardirqs_off(a0, a1)             do { } while (0)
+#endif
+
 #endif /* _LINUX_FTRACE_H */
index e600c4e..5b711d4 100644 (file)
 #define _LINUX_TRACE_IRQFLAGS_H
 
 #ifdef CONFIG_TRACE_IRQFLAGS
-  extern void trace_hardirqs_on(void);
-  extern void trace_hardirqs_off(void);
   extern void trace_softirqs_on(unsigned long ip);
   extern void trace_softirqs_off(unsigned long ip);
+  extern void trace_hardirqs_on(void);
+  extern void trace_hardirqs_off(void);
 # define trace_hardirq_context(p)      ((p)->hardirq_context)
 # define trace_softirq_context(p)      ((p)->softirq_context)
 # define trace_hardirqs_enabled(p)     ((p)->hardirqs_enabled)
 # define INIT_TRACE_IRQFLAGS
 #endif
 
+#ifdef CONFIG_IRQSOFF_TRACER
+ extern void stop_critical_timings(void);
+ extern void start_critical_timings(void);
+#else
+# define stop_critical_timings() do { } while (0)
+# define start_critical_timings() do { } while (0)
+#endif
+
 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
 
 #include <asm/irqflags.h>
index 19908b2..d66d676 100644 (file)
@@ -909,7 +909,7 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 
        rt_mutex_init_task(p);
 
-#ifdef CONFIG_TRACE_IRQFLAGS
+#if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_LOCKDEP)
        DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
        DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
 #endif
index 81a4e4a..e219243 100644 (file)
@@ -39,6 +39,7 @@
 #include <linux/irqflags.h>
 #include <linux/utsname.h>
 #include <linux/hash.h>
+#include <linux/ftrace.h>
 
 #include <asm/sections.h>
 
@@ -982,7 +983,7 @@ check_noncircular(struct lock_class *source, unsigned int depth)
        return 1;
 }
 
-#ifdef CONFIG_TRACE_IRQFLAGS
+#if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING)
 /*
  * Forwards and backwards subgraph searching, for the purposes of
  * proving that two subgraphs can be connected by a new dependency
@@ -1680,7 +1681,7 @@ valid_state(struct task_struct *curr, struct held_lock *this,
 static int mark_lock(struct task_struct *curr, struct held_lock *this,
                     enum lock_usage_bit new_bit);
 
-#ifdef CONFIG_TRACE_IRQFLAGS
+#if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING)
 
 /*
  * print irq inversion bug:
@@ -2013,11 +2014,13 @@ void early_boot_irqs_on(void)
 /*
  * Hardirqs will be enabled:
  */
-void trace_hardirqs_on(void)
+void notrace trace_hardirqs_on_caller(unsigned long a0)
 {
        struct task_struct *curr = current;
        unsigned long ip;
 
+       time_hardirqs_on(CALLER_ADDR0, a0);
+
        if (unlikely(!debug_locks || current->lockdep_recursion))
                return;
 
@@ -2055,16 +2058,23 @@ void trace_hardirqs_on(void)
        curr->hardirq_enable_event = ++curr->irq_events;
        debug_atomic_inc(&hardirqs_on_events);
 }
+EXPORT_SYMBOL(trace_hardirqs_on_caller);
 
+void notrace trace_hardirqs_on(void)
+{
+       trace_hardirqs_on_caller(CALLER_ADDR0);
+}
 EXPORT_SYMBOL(trace_hardirqs_on);
 
 /*
  * Hardirqs were disabled:
  */
-void trace_hardirqs_off(void)
+void notrace trace_hardirqs_off_caller(unsigned long a0)
 {
        struct task_struct *curr = current;
 
+       time_hardirqs_off(CALLER_ADDR0, a0);
+
        if (unlikely(!debug_locks || current->lockdep_recursion))
                return;
 
@@ -2082,7 +2092,12 @@ void trace_hardirqs_off(void)
        } else
                debug_atomic_inc(&redundant_hardirqs_off);
 }
+EXPORT_SYMBOL(trace_hardirqs_off_caller);
 
+void notrace trace_hardirqs_off(void)
+{
+       trace_hardirqs_off_caller(CALLER_ADDR0);
+}
 EXPORT_SYMBOL(trace_hardirqs_off);
 
 /*
index 8fb01c3..ae7d5b9 100644 (file)
@@ -1041,7 +1041,9 @@ void release_console_sem(void)
                _log_end = log_end;
                con_start = log_end;            /* Flush */
                spin_unlock(&logbuf_lock);
+               stop_critical_timings();        /* don't trace print latency */
                call_console_drivers(_con_start, _log_end);
+               start_critical_timings();
                local_irq_restore(flags);
        }
        console_locked = 0;
index 892ecc9..896df1c 100644 (file)
@@ -26,6 +26,24 @@ config FTRACE
          (the bootup default), then the overhead of the instructions is very
          small and not measurable even in micro-benchmarks.
 
+config IRQSOFF_TRACER
+       bool "Interrupts-off Latency Tracer"
+       default n
+       depends on TRACE_IRQFLAGS_SUPPORT
+       depends on GENERIC_TIME
+       select TRACE_IRQFLAGS
+       select TRACING
+       select TRACER_MAX_TRACE
+       help
+         This option measures the time spent in irqs-off critical
+         sections, with microsecond accuracy.
+
+         The default measurement method is a maximum search, which is
+         disabled by default and can be runtime (re-)started
+         via:
+
+             echo 0 > /debugfs/tracing/tracing_max_latency
+
 config SCHED_TRACER
        bool "Scheduling Latency Tracer"
        depends on DEBUG_KERNEL
index 5508cdb..46be864 100644 (file)
@@ -3,6 +3,7 @@ obj-$(CONFIG_FTRACE) += libftrace.o
 obj-$(CONFIG_TRACING) += trace.o
 obj-$(CONFIG_CONTEXT_SWITCH_TRACER) += trace_sched_switch.o
 obj-$(CONFIG_FTRACE) += trace_functions.o
+obj-$(CONFIG_IRQSOFF_TRACER) += trace_irqsoff.o
 obj-$(CONFIG_SCHED_TRACER) += trace_sched_wakeup.o
 
 libftrace-y := ftrace.o
diff --git a/kernel/trace/trace_irqsoff.c b/kernel/trace/trace_irqsoff.c
new file mode 100644 (file)
index 0000000..a9131b0
--- /dev/null
@@ -0,0 +1,402 @@
+/*
+ * trace irqs off criticall timings
+ *
+ * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
+ * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
+ *
+ * From code in the latency_tracer, that is:
+ *
+ *  Copyright (C) 2004-2006 Ingo Molnar
+ *  Copyright (C) 2004 William Lee Irwin III
+ */
+#include <linux/kallsyms.h>
+#include <linux/debugfs.h>
+#include <linux/uaccess.h>
+#include <linux/module.h>
+#include <linux/ftrace.h>
+#include <linux/fs.h>
+
+#include "trace.h"
+
+static struct trace_array              *irqsoff_trace __read_mostly;
+static int                             tracer_enabled __read_mostly;
+
+/*
+ * Sequence count - we record it when starting a measurement and
+ * skip the latency if the sequence has changed - some other section
+ * did a maximum and could disturb our measurement with serial console
+ * printouts, etc. Truly coinciding maximum latencies should be rare
+ * and what happens together happens separately as well, so this doesnt
+ * decrease the validity of the maximum found:
+ */
+static __cacheline_aligned_in_smp      unsigned long max_sequence;
+
+#ifdef CONFIG_FTRACE
+/*
+ * irqsoff uses its own tracer function to keep the overhead down:
+ */
+static void notrace
+irqsoff_tracer_call(unsigned long ip, unsigned long parent_ip)
+{
+       struct trace_array *tr = irqsoff_trace;
+       struct trace_array_cpu *data;
+       unsigned long flags;
+       long disabled;
+       int cpu;
+
+       if (likely(!tracer_enabled))
+               return;
+
+       local_save_flags(flags);
+
+       if (!irqs_disabled_flags(flags))
+               return;
+
+       cpu = raw_smp_processor_id();
+       data = tr->data[cpu];
+       disabled = atomic_inc_return(&data->disabled);
+
+       if (likely(disabled == 1))
+               ftrace(tr, data, ip, parent_ip, flags);
+
+       atomic_dec(&data->disabled);
+}
+
+static struct ftrace_ops trace_ops __read_mostly =
+{
+       .func = irqsoff_tracer_call,
+};
+#endif /* CONFIG_FTRACE */
+
+/*
+ * Should this new latency be reported/recorded?
+ */
+static int notrace report_latency(cycle_t delta)
+{
+       if (tracing_thresh) {
+               if (delta < tracing_thresh)
+                       return 0;
+       } else {
+               if (delta <= tracing_max_latency)
+                       return 0;
+       }
+       return 1;
+}
+
+static void notrace
+check_critical_timing(struct trace_array *tr,
+                     struct trace_array_cpu *data,
+                     unsigned long parent_ip,
+                     int cpu)
+{
+       unsigned long latency, t0, t1;
+       cycle_t T0, T1, T2, delta;
+       unsigned long flags;
+
+       /*
+        * usecs conversion is slow so we try to delay the conversion
+        * as long as possible:
+        */
+       T0 = data->preempt_timestamp;
+       T1 = now(cpu);
+       delta = T1-T0;
+
+       local_save_flags(flags);
+
+       if (!report_latency(delta))
+               goto out;
+
+       ftrace(tr, data, CALLER_ADDR0, parent_ip, flags);
+       /*
+        * Update the timestamp, because the trace entry above
+        * might change it (it can only get larger so the latency
+        * is fair to be reported):
+        */
+       T2 = now(cpu);
+
+       delta = T2-T0;
+
+       latency = nsecs_to_usecs(delta);
+
+       if (data->critical_sequence != max_sequence)
+               goto out;
+
+       tracing_max_latency = delta;
+       t0 = nsecs_to_usecs(T0);
+       t1 = nsecs_to_usecs(T1);
+
+       data->critical_end = parent_ip;
+
+       update_max_tr_single(tr, current, cpu);
+
+       if (tracing_thresh)
+               printk(KERN_INFO "(%16s-%-5d|#%d): %lu us critical section "
+                      "violates %lu us threshold.\n"
+                      " => started at timestamp %lu: ",
+                               current->comm, current->pid,
+                               raw_smp_processor_id(),
+                               latency, nsecs_to_usecs(tracing_thresh), t0);
+       else
+               printk(KERN_INFO "(%16s-%-5d|#%d):"
+                      " new %lu us maximum-latency "
+                      "critical section.\n => started at timestamp %lu: ",
+                               current->comm, current->pid,
+                               raw_smp_processor_id(),
+                               latency, t0);
+
+       print_symbol(KERN_CONT "<%s>\n", data->critical_start);
+       printk(KERN_CONT " =>   ended at timestamp %lu: ", t1);
+       print_symbol(KERN_CONT "<%s>\n", data->critical_end);
+       dump_stack();
+       t1 = nsecs_to_usecs(now(cpu));
+       printk(KERN_CONT " =>   dump-end timestamp %lu\n\n", t1);
+
+       max_sequence++;
+
+out:
+       data->critical_sequence = max_sequence;
+       data->preempt_timestamp = now(cpu);
+       tracing_reset(data);
+       ftrace(tr, data, CALLER_ADDR0, parent_ip, flags);
+}
+
+static inline void notrace
+start_critical_timing(unsigned long ip, unsigned long parent_ip)
+{
+       int cpu;
+       struct trace_array *tr = irqsoff_trace;
+       struct trace_array_cpu *data;
+       unsigned long flags;
+
+       if (likely(!tracer_enabled))
+               return;
+
+       cpu = raw_smp_processor_id();
+       data = tr->data[cpu];
+
+       if (unlikely(!data) || unlikely(!data->trace) ||
+           data->critical_start || atomic_read(&data->disabled))
+               return;
+
+       atomic_inc(&data->disabled);
+
+       data->critical_sequence = max_sequence;
+       data->preempt_timestamp = now(cpu);
+       data->critical_start = parent_ip;
+       tracing_reset(data);
+
+       local_save_flags(flags);
+       ftrace(tr, data, ip, parent_ip, flags);
+
+       atomic_dec(&data->disabled);
+}
+
+static inline void notrace
+stop_critical_timing(unsigned long ip, unsigned long parent_ip)
+{
+       int cpu;
+       struct trace_array *tr = irqsoff_trace;
+       struct trace_array_cpu *data;
+       unsigned long flags;
+
+       if (likely(!tracer_enabled))
+               return;
+
+       cpu = raw_smp_processor_id();
+       data = tr->data[cpu];
+
+       if (unlikely(!data) || unlikely(!data->trace) ||
+           !data->critical_start || atomic_read(&data->disabled))
+               return;
+
+       atomic_inc(&data->disabled);
+       local_save_flags(flags);
+       ftrace(tr, data, ip, parent_ip, flags);
+       check_critical_timing(tr, data, parent_ip, cpu);
+       data->critical_start = 0;
+       atomic_dec(&data->disabled);
+}
+
+void notrace start_critical_timings(void)
+{
+       unsigned long flags;
+
+       local_save_flags(flags);
+
+       if (irqs_disabled_flags(flags))
+               start_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
+}
+
+void notrace stop_critical_timings(void)
+{
+       unsigned long flags;
+
+       local_save_flags(flags);
+
+       if (irqs_disabled_flags(flags))
+               stop_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
+}
+
+#ifdef CONFIG_PROVE_LOCKING
+void notrace time_hardirqs_on(unsigned long a0, unsigned long a1)
+{
+       unsigned long flags;
+
+       local_save_flags(flags);
+
+       if (irqs_disabled_flags(flags))
+               stop_critical_timing(a0, a1);
+}
+
+void notrace time_hardirqs_off(unsigned long a0, unsigned long a1)
+{
+       unsigned long flags;
+
+       local_save_flags(flags);
+
+       if (irqs_disabled_flags(flags))
+               start_critical_timing(a0, a1);
+}
+
+#else /* !CONFIG_PROVE_LOCKING */
+
+/*
+ * Stubs:
+ */
+
+void early_boot_irqs_off(void)
+{
+}
+
+void early_boot_irqs_on(void)
+{
+}
+
+void trace_softirqs_on(unsigned long ip)
+{
+}
+
+void trace_softirqs_off(unsigned long ip)
+{
+}
+
+inline void print_irqtrace_events(struct task_struct *curr)
+{
+}
+
+/*
+ * We are only interested in hardirq on/off events:
+ */
+void notrace trace_hardirqs_on(void)
+{
+       unsigned long flags;
+
+       local_save_flags(flags);
+
+       if (irqs_disabled_flags(flags))
+               stop_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
+}
+EXPORT_SYMBOL(trace_hardirqs_on);
+
+void notrace trace_hardirqs_off(void)
+{
+       unsigned long flags;
+
+       local_save_flags(flags);
+
+       if (irqs_disabled_flags(flags))
+               start_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
+}
+EXPORT_SYMBOL(trace_hardirqs_off);
+
+void notrace trace_hardirqs_on_caller(unsigned long caller_addr)
+{
+       unsigned long flags;
+
+       local_save_flags(flags);
+
+       if (irqs_disabled_flags(flags))
+               stop_critical_timing(CALLER_ADDR0, caller_addr);
+}
+EXPORT_SYMBOL(trace_hardirqs_on_caller);
+
+void notrace trace_hardirqs_off_caller(unsigned long caller_addr)
+{
+       unsigned long flags;
+
+       local_save_flags(flags);
+
+       if (irqs_disabled_flags(flags))
+               start_critical_timing(CALLER_ADDR0, caller_addr);
+}
+EXPORT_SYMBOL(trace_hardirqs_off_caller);
+
+#endif /* CONFIG_PROVE_LOCKING */
+
+static void start_irqsoff_tracer(struct trace_array *tr)
+{
+       tracer_enabled = 1;
+       register_ftrace_function(&trace_ops);
+}
+
+static void stop_irqsoff_tracer(struct trace_array *tr)
+{
+       unregister_ftrace_function(&trace_ops);
+       tracer_enabled = 0;
+}
+
+static void irqsoff_tracer_init(struct trace_array *tr)
+{
+       irqsoff_trace = tr;
+       /* make sure that the tracer is visibel */
+       smp_wmb();
+
+       if (tr->ctrl)
+               start_irqsoff_tracer(tr);
+}
+
+static void irqsoff_tracer_reset(struct trace_array *tr)
+{
+       if (tr->ctrl)
+               stop_irqsoff_tracer(tr);
+}
+
+static void irqsoff_tracer_ctrl_update(struct trace_array *tr)
+{
+       if (tr->ctrl)
+               start_irqsoff_tracer(tr);
+       else
+               stop_irqsoff_tracer(tr);
+}
+
+static void notrace irqsoff_tracer_open(struct trace_iterator *iter)
+{
+       /* stop the trace while dumping */
+       if (iter->tr->ctrl)
+               stop_irqsoff_tracer(iter->tr);
+}
+
+static void notrace irqsoff_tracer_close(struct trace_iterator *iter)
+{
+       if (iter->tr->ctrl)
+               start_irqsoff_tracer(iter->tr);
+}
+
+static struct tracer irqsoff_tracer __read_mostly =
+{
+       .name           = "irqsoff",
+       .init           = irqsoff_tracer_init,
+       .reset          = irqsoff_tracer_reset,
+       .open           = irqsoff_tracer_open,
+       .close          = irqsoff_tracer_close,
+       .ctrl_update    = irqsoff_tracer_ctrl_update,
+       .print_max      = 1,
+};
+
+__init static int init_irqsoff_tracer(void)
+{
+       register_tracer(&irqsoff_tracer);
+
+       return 0;
+}
+device_initcall(init_irqsoff_tracer);