aboutsummaryrefslogtreecommitdiffstats
path: root/kernel
diff options
context:
space:
mode:
authorIngo Molnar <mingo@elte.hu>2008-07-14 10:11:52 -0400
committerIngo Molnar <mingo@elte.hu>2008-07-14 10:11:52 -0400
commit5806b81ac1c0c52665b91723fd4146a4f86e386b (patch)
tree24ea8763bf308ce1407c1de91dc8de4d2655e1c1 /kernel
parentd14c8a680ccfdeb5e7b9be4d61162c2b373bd1e8 (diff)
parent6712e299b7dc78aa4971b85e803435ee6d49a9dd (diff)
Merge branch 'auto-ftrace-next' into tracing/for-linus
Conflicts: arch/x86/kernel/entry_32.S arch/x86/kernel/process_32.c arch/x86/kernel/process_64.c arch/x86/lib/Makefile include/asm-x86/irqflags.h kernel/Makefile kernel/sched.c Signed-off-by: Ingo Molnar <mingo@elte.hu>
Diffstat (limited to 'kernel')
-rw-r--r--kernel/Makefile14
-rw-r--r--kernel/fork.c2
-rw-r--r--kernel/lockdep.c33
-rw-r--r--kernel/marker.c30
-rw-r--r--kernel/printk.c2
-rw-r--r--kernel/sched.c55
-rw-r--r--kernel/semaphore.c1
-rw-r--r--kernel/spinlock.c2
-rw-r--r--kernel/sysctl.c11
-rw-r--r--kernel/trace/Kconfig135
-rw-r--r--kernel/trace/Makefile24
-rw-r--r--kernel/trace/ftrace.c1727
-rw-r--r--kernel/trace/trace.c3161
-rw-r--r--kernel/trace/trace.h339
-rw-r--r--kernel/trace/trace_functions.c81
-rw-r--r--kernel/trace/trace_irqsoff.c486
-rw-r--r--kernel/trace/trace_mmiotrace.c295
-rw-r--r--kernel/trace/trace_sched_switch.c286
-rw-r--r--kernel/trace/trace_sched_wakeup.c448
-rw-r--r--kernel/trace/trace_selftest.c563
-rw-r--r--kernel/trace/trace_selftest_dynamic.c7
-rw-r--r--kernel/trace/trace_sysprof.c363
22 files changed, 8038 insertions, 27 deletions
diff --git a/kernel/Makefile b/kernel/Makefile
index 6c55301112e0..f6328e16dfdd 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -11,6 +11,18 @@ obj-y = sched.o fork.o exec_domain.o panic.o printk.o profile.o \
11 hrtimer.o rwsem.o nsproxy.o srcu.o semaphore.o \ 11 hrtimer.o rwsem.o nsproxy.o srcu.o semaphore.o \
12 notifier.o ksysfs.o pm_qos_params.o sched_clock.o 12 notifier.o ksysfs.o pm_qos_params.o sched_clock.o
13 13
14CFLAGS_REMOVE_sched.o = -mno-spe
15
16ifdef CONFIG_FTRACE
17# Do not trace debug files and internal ftrace files
18CFLAGS_REMOVE_lockdep.o = -pg
19CFLAGS_REMOVE_lockdep_proc.o = -pg
20CFLAGS_REMOVE_mutex-debug.o = -pg
21CFLAGS_REMOVE_rtmutex-debug.o = -pg
22CFLAGS_REMOVE_cgroup-debug.o = -pg
23CFLAGS_REMOVE_sched_clock.o = -pg
24endif
25
14obj-$(CONFIG_SYSCTL_SYSCALL_CHECK) += sysctl_check.o 26obj-$(CONFIG_SYSCTL_SYSCALL_CHECK) += sysctl_check.o
15obj-$(CONFIG_STACKTRACE) += stacktrace.o 27obj-$(CONFIG_STACKTRACE) += stacktrace.o
16obj-y += time/ 28obj-y += time/
@@ -69,6 +81,8 @@ obj-$(CONFIG_TASK_DELAY_ACCT) += delayacct.o
69obj-$(CONFIG_TASKSTATS) += taskstats.o tsacct.o 81obj-$(CONFIG_TASKSTATS) += taskstats.o tsacct.o
70obj-$(CONFIG_MARKERS) += marker.o 82obj-$(CONFIG_MARKERS) += marker.o
71obj-$(CONFIG_LATENCYTOP) += latencytop.o 83obj-$(CONFIG_LATENCYTOP) += latencytop.o
84obj-$(CONFIG_FTRACE) += trace/
85obj-$(CONFIG_TRACING) += trace/
72obj-$(CONFIG_SMP) += sched_cpupri.o 86obj-$(CONFIG_SMP) += sched_cpupri.o
73 87
74ifneq ($(CONFIG_SCHED_NO_NO_OMIT_FRAME_POINTER),y) 88ifneq ($(CONFIG_SCHED_NO_NO_OMIT_FRAME_POINTER),y)
diff --git a/kernel/fork.c b/kernel/fork.c
index 19908b26cf80..d66d676dc362 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -909,7 +909,7 @@ static struct task_struct *copy_process(unsigned long clone_flags,
909 909
910 rt_mutex_init_task(p); 910 rt_mutex_init_task(p);
911 911
912#ifdef CONFIG_TRACE_IRQFLAGS 912#if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_LOCKDEP)
913 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled); 913 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
914 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled); 914 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
915#endif 915#endif
diff --git a/kernel/lockdep.c b/kernel/lockdep.c
index 81a4e4a3f087..65548eff029e 100644
--- a/kernel/lockdep.c
+++ b/kernel/lockdep.c
@@ -39,6 +39,7 @@
39#include <linux/irqflags.h> 39#include <linux/irqflags.h>
40#include <linux/utsname.h> 40#include <linux/utsname.h>
41#include <linux/hash.h> 41#include <linux/hash.h>
42#include <linux/ftrace.h>
42 43
43#include <asm/sections.h> 44#include <asm/sections.h>
44 45
@@ -81,6 +82,8 @@ static int graph_lock(void)
81 __raw_spin_unlock(&lockdep_lock); 82 __raw_spin_unlock(&lockdep_lock);
82 return 0; 83 return 0;
83 } 84 }
85 /* prevent any recursions within lockdep from causing deadlocks */
86 current->lockdep_recursion++;
84 return 1; 87 return 1;
85} 88}
86 89
@@ -89,6 +92,7 @@ static inline int graph_unlock(void)
89 if (debug_locks && !__raw_spin_is_locked(&lockdep_lock)) 92 if (debug_locks && !__raw_spin_is_locked(&lockdep_lock))
90 return DEBUG_LOCKS_WARN_ON(1); 93 return DEBUG_LOCKS_WARN_ON(1);
91 94
95 current->lockdep_recursion--;
92 __raw_spin_unlock(&lockdep_lock); 96 __raw_spin_unlock(&lockdep_lock);
93 return 0; 97 return 0;
94} 98}
@@ -982,7 +986,7 @@ check_noncircular(struct lock_class *source, unsigned int depth)
982 return 1; 986 return 1;
983} 987}
984 988
985#ifdef CONFIG_TRACE_IRQFLAGS 989#if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING)
986/* 990/*
987 * Forwards and backwards subgraph searching, for the purposes of 991 * Forwards and backwards subgraph searching, for the purposes of
988 * proving that two subgraphs can be connected by a new dependency 992 * proving that two subgraphs can be connected by a new dependency
@@ -1680,7 +1684,7 @@ valid_state(struct task_struct *curr, struct held_lock *this,
1680static int mark_lock(struct task_struct *curr, struct held_lock *this, 1684static int mark_lock(struct task_struct *curr, struct held_lock *this,
1681 enum lock_usage_bit new_bit); 1685 enum lock_usage_bit new_bit);
1682 1686
1683#ifdef CONFIG_TRACE_IRQFLAGS 1687#if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING)
1684 1688
1685/* 1689/*
1686 * print irq inversion bug: 1690 * print irq inversion bug:
@@ -2013,11 +2017,13 @@ void early_boot_irqs_on(void)
2013/* 2017/*
2014 * Hardirqs will be enabled: 2018 * Hardirqs will be enabled:
2015 */ 2019 */
2016void trace_hardirqs_on(void) 2020void trace_hardirqs_on_caller(unsigned long a0)
2017{ 2021{
2018 struct task_struct *curr = current; 2022 struct task_struct *curr = current;
2019 unsigned long ip; 2023 unsigned long ip;
2020 2024
2025 time_hardirqs_on(CALLER_ADDR0, a0);
2026
2021 if (unlikely(!debug_locks || current->lockdep_recursion)) 2027 if (unlikely(!debug_locks || current->lockdep_recursion))
2022 return; 2028 return;
2023 2029
@@ -2055,16 +2061,23 @@ void trace_hardirqs_on(void)
2055 curr->hardirq_enable_event = ++curr->irq_events; 2061 curr->hardirq_enable_event = ++curr->irq_events;
2056 debug_atomic_inc(&hardirqs_on_events); 2062 debug_atomic_inc(&hardirqs_on_events);
2057} 2063}
2064EXPORT_SYMBOL(trace_hardirqs_on_caller);
2058 2065
2066void trace_hardirqs_on(void)
2067{
2068 trace_hardirqs_on_caller(CALLER_ADDR0);
2069}
2059EXPORT_SYMBOL(trace_hardirqs_on); 2070EXPORT_SYMBOL(trace_hardirqs_on);
2060 2071
2061/* 2072/*
2062 * Hardirqs were disabled: 2073 * Hardirqs were disabled:
2063 */ 2074 */
2064void trace_hardirqs_off(void) 2075void trace_hardirqs_off_caller(unsigned long a0)
2065{ 2076{
2066 struct task_struct *curr = current; 2077 struct task_struct *curr = current;
2067 2078
2079 time_hardirqs_off(CALLER_ADDR0, a0);
2080
2068 if (unlikely(!debug_locks || current->lockdep_recursion)) 2081 if (unlikely(!debug_locks || current->lockdep_recursion))
2069 return; 2082 return;
2070 2083
@@ -2082,7 +2095,12 @@ void trace_hardirqs_off(void)
2082 } else 2095 } else
2083 debug_atomic_inc(&redundant_hardirqs_off); 2096 debug_atomic_inc(&redundant_hardirqs_off);
2084} 2097}
2098EXPORT_SYMBOL(trace_hardirqs_off_caller);
2085 2099
2100void trace_hardirqs_off(void)
2101{
2102 trace_hardirqs_off_caller(CALLER_ADDR0);
2103}
2086EXPORT_SYMBOL(trace_hardirqs_off); 2104EXPORT_SYMBOL(trace_hardirqs_off);
2087 2105
2088/* 2106/*
@@ -2246,7 +2264,7 @@ static inline int separate_irq_context(struct task_struct *curr,
2246 * Mark a lock with a usage bit, and validate the state transition: 2264 * Mark a lock with a usage bit, and validate the state transition:
2247 */ 2265 */
2248static int mark_lock(struct task_struct *curr, struct held_lock *this, 2266static int mark_lock(struct task_struct *curr, struct held_lock *this,
2249 enum lock_usage_bit new_bit) 2267 enum lock_usage_bit new_bit)
2250{ 2268{
2251 unsigned int new_mask = 1 << new_bit, ret = 1; 2269 unsigned int new_mask = 1 << new_bit, ret = 1;
2252 2270
@@ -2686,7 +2704,7 @@ static void check_flags(unsigned long flags)
2686 * and also avoid lockdep recursion: 2704 * and also avoid lockdep recursion:
2687 */ 2705 */
2688void lock_acquire(struct lockdep_map *lock, unsigned int subclass, 2706void lock_acquire(struct lockdep_map *lock, unsigned int subclass,
2689 int trylock, int read, int check, unsigned long ip) 2707 int trylock, int read, int check, unsigned long ip)
2690{ 2708{
2691 unsigned long flags; 2709 unsigned long flags;
2692 2710
@@ -2708,7 +2726,8 @@ void lock_acquire(struct lockdep_map *lock, unsigned int subclass,
2708 2726
2709EXPORT_SYMBOL_GPL(lock_acquire); 2727EXPORT_SYMBOL_GPL(lock_acquire);
2710 2728
2711void lock_release(struct lockdep_map *lock, int nested, unsigned long ip) 2729void lock_release(struct lockdep_map *lock, int nested,
2730 unsigned long ip)
2712{ 2731{
2713 unsigned long flags; 2732 unsigned long flags;
2714 2733
diff --git a/kernel/marker.c b/kernel/marker.c
index b5a9fe1d50d5..1abfb923b761 100644
--- a/kernel/marker.c
+++ b/kernel/marker.c
@@ -55,8 +55,8 @@ static DEFINE_MUTEX(markers_mutex);
55struct marker_entry { 55struct marker_entry {
56 struct hlist_node hlist; 56 struct hlist_node hlist;
57 char *format; 57 char *format;
58 void (*call)(const struct marker *mdata, /* Probe wrapper */ 58 /* Probe wrapper */
59 void *call_private, const char *fmt, ...); 59 void (*call)(const struct marker *mdata, void *call_private, ...);
60 struct marker_probe_closure single; 60 struct marker_probe_closure single;
61 struct marker_probe_closure *multi; 61 struct marker_probe_closure *multi;
62 int refcount; /* Number of times armed. 0 if disarmed. */ 62 int refcount; /* Number of times armed. 0 if disarmed. */
@@ -91,15 +91,13 @@ EXPORT_SYMBOL_GPL(__mark_empty_function);
91 * marker_probe_cb Callback that prepares the variable argument list for probes. 91 * marker_probe_cb Callback that prepares the variable argument list for probes.
92 * @mdata: pointer of type struct marker 92 * @mdata: pointer of type struct marker
93 * @call_private: caller site private data 93 * @call_private: caller site private data
94 * @fmt: format string
95 * @...: Variable argument list. 94 * @...: Variable argument list.
96 * 95 *
97 * Since we do not use "typical" pointer based RCU in the 1 argument case, we 96 * Since we do not use "typical" pointer based RCU in the 1 argument case, we
98 * need to put a full smp_rmb() in this branch. This is why we do not use 97 * need to put a full smp_rmb() in this branch. This is why we do not use
99 * rcu_dereference() for the pointer read. 98 * rcu_dereference() for the pointer read.
100 */ 99 */
101void marker_probe_cb(const struct marker *mdata, void *call_private, 100void marker_probe_cb(const struct marker *mdata, void *call_private, ...)
102 const char *fmt, ...)
103{ 101{
104 va_list args; 102 va_list args;
105 char ptype; 103 char ptype;
@@ -120,8 +118,9 @@ void marker_probe_cb(const struct marker *mdata, void *call_private,
120 /* Must read the ptr before private data. They are not data 118 /* Must read the ptr before private data. They are not data
121 * dependant, so we put an explicit smp_rmb() here. */ 119 * dependant, so we put an explicit smp_rmb() here. */
122 smp_rmb(); 120 smp_rmb();
123 va_start(args, fmt); 121 va_start(args, call_private);
124 func(mdata->single.probe_private, call_private, fmt, &args); 122 func(mdata->single.probe_private, call_private, mdata->format,
123 &args);
125 va_end(args); 124 va_end(args);
126 } else { 125 } else {
127 struct marker_probe_closure *multi; 126 struct marker_probe_closure *multi;
@@ -136,9 +135,9 @@ void marker_probe_cb(const struct marker *mdata, void *call_private,
136 smp_read_barrier_depends(); 135 smp_read_barrier_depends();
137 multi = mdata->multi; 136 multi = mdata->multi;
138 for (i = 0; multi[i].func; i++) { 137 for (i = 0; multi[i].func; i++) {
139 va_start(args, fmt); 138 va_start(args, call_private);
140 multi[i].func(multi[i].probe_private, call_private, fmt, 139 multi[i].func(multi[i].probe_private, call_private,
141 &args); 140 mdata->format, &args);
142 va_end(args); 141 va_end(args);
143 } 142 }
144 } 143 }
@@ -150,13 +149,11 @@ EXPORT_SYMBOL_GPL(marker_probe_cb);
150 * marker_probe_cb Callback that does not prepare the variable argument list. 149 * marker_probe_cb Callback that does not prepare the variable argument list.
151 * @mdata: pointer of type struct marker 150 * @mdata: pointer of type struct marker
152 * @call_private: caller site private data 151 * @call_private: caller site private data
153 * @fmt: format string
154 * @...: Variable argument list. 152 * @...: Variable argument list.
155 * 153 *
156 * Should be connected to markers "MARK_NOARGS". 154 * Should be connected to markers "MARK_NOARGS".
157 */ 155 */
158void marker_probe_cb_noarg(const struct marker *mdata, 156void marker_probe_cb_noarg(const struct marker *mdata, void *call_private, ...)
159 void *call_private, const char *fmt, ...)
160{ 157{
161 va_list args; /* not initialized */ 158 va_list args; /* not initialized */
162 char ptype; 159 char ptype;
@@ -172,7 +169,8 @@ void marker_probe_cb_noarg(const struct marker *mdata,
172 /* Must read the ptr before private data. They are not data 169 /* Must read the ptr before private data. They are not data
173 * dependant, so we put an explicit smp_rmb() here. */ 170 * dependant, so we put an explicit smp_rmb() here. */
174 smp_rmb(); 171 smp_rmb();
175 func(mdata->single.probe_private, call_private, fmt, &args); 172 func(mdata->single.probe_private, call_private, mdata->format,
173 &args);
176 } else { 174 } else {
177 struct marker_probe_closure *multi; 175 struct marker_probe_closure *multi;
178 int i; 176 int i;
@@ -186,8 +184,8 @@ void marker_probe_cb_noarg(const struct marker *mdata,
186 smp_read_barrier_depends(); 184 smp_read_barrier_depends();
187 multi = mdata->multi; 185 multi = mdata->multi;
188 for (i = 0; multi[i].func; i++) 186 for (i = 0; multi[i].func; i++)
189 multi[i].func(multi[i].probe_private, call_private, fmt, 187 multi[i].func(multi[i].probe_private, call_private,
190 &args); 188 mdata->format, &args);
191 } 189 }
192 preempt_enable(); 190 preempt_enable();
193} 191}
diff --git a/kernel/printk.c b/kernel/printk.c
index 625d240d7ada..5d81a11321fd 100644
--- a/kernel/printk.c
+++ b/kernel/printk.c
@@ -1046,7 +1046,9 @@ void release_console_sem(void)
1046 _log_end = log_end; 1046 _log_end = log_end;
1047 con_start = log_end; /* Flush */ 1047 con_start = log_end; /* Flush */
1048 spin_unlock(&logbuf_lock); 1048 spin_unlock(&logbuf_lock);
1049 stop_critical_timings(); /* don't trace print latency */
1049 call_console_drivers(_con_start, _log_end); 1050 call_console_drivers(_con_start, _log_end);
1051 start_critical_timings();
1050 local_irq_restore(flags); 1052 local_irq_restore(flags);
1051 } 1053 }
1052 console_locked = 0; 1054 console_locked = 0;
diff --git a/kernel/sched.c b/kernel/sched.c
index 591d5e7f757a..c74b0d23c752 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -70,6 +70,7 @@
70#include <linux/bootmem.h> 70#include <linux/bootmem.h>
71#include <linux/debugfs.h> 71#include <linux/debugfs.h>
72#include <linux/ctype.h> 72#include <linux/ctype.h>
73#include <linux/ftrace.h>
73 74
74#include <asm/tlb.h> 75#include <asm/tlb.h>
75#include <asm/irq_regs.h> 76#include <asm/irq_regs.h>
@@ -645,6 +646,24 @@ static inline void update_rq_clock(struct rq *rq)
645# define const_debug static const 646# define const_debug static const
646#endif 647#endif
647 648
649/**
650 * runqueue_is_locked
651 *
652 * Returns true if the current cpu runqueue is locked.
653 * This interface allows printk to be called with the runqueue lock
654 * held and know whether or not it is OK to wake up the klogd.
655 */
656int runqueue_is_locked(void)
657{
658 int cpu = get_cpu();
659 struct rq *rq = cpu_rq(cpu);
660 int ret;
661
662 ret = spin_is_locked(&rq->lock);
663 put_cpu();
664 return ret;
665}
666
648/* 667/*
649 * Debugging: various feature bits 668 * Debugging: various feature bits
650 */ 669 */
@@ -2318,6 +2337,9 @@ out_activate:
2318 success = 1; 2337 success = 1;
2319 2338
2320out_running: 2339out_running:
2340 trace_mark(kernel_sched_wakeup,
2341 "pid %d state %ld ## rq %p task %p rq->curr %p",
2342 p->pid, p->state, rq, p, rq->curr);
2321 check_preempt_curr(rq, p); 2343 check_preempt_curr(rq, p);
2322 2344
2323 p->state = TASK_RUNNING; 2345 p->state = TASK_RUNNING;
@@ -2450,6 +2472,9 @@ void wake_up_new_task(struct task_struct *p, unsigned long clone_flags)
2450 p->sched_class->task_new(rq, p); 2472 p->sched_class->task_new(rq, p);
2451 inc_nr_running(rq); 2473 inc_nr_running(rq);
2452 } 2474 }
2475 trace_mark(kernel_sched_wakeup_new,
2476 "pid %d state %ld ## rq %p task %p rq->curr %p",
2477 p->pid, p->state, rq, p, rq->curr);
2453 check_preempt_curr(rq, p); 2478 check_preempt_curr(rq, p);
2454#ifdef CONFIG_SMP 2479#ifdef CONFIG_SMP
2455 if (p->sched_class->task_wake_up) 2480 if (p->sched_class->task_wake_up)
@@ -2622,6 +2647,11 @@ context_switch(struct rq *rq, struct task_struct *prev,
2622 struct mm_struct *mm, *oldmm; 2647 struct mm_struct *mm, *oldmm;
2623 2648
2624 prepare_task_switch(rq, prev, next); 2649 prepare_task_switch(rq, prev, next);
2650 trace_mark(kernel_sched_schedule,
2651 "prev_pid %d next_pid %d prev_state %ld "
2652 "## rq %p prev %p next %p",
2653 prev->pid, next->pid, prev->state,
2654 rq, prev, next);
2625 mm = next->mm; 2655 mm = next->mm;
2626 oldmm = prev->active_mm; 2656 oldmm = prev->active_mm;
2627 /* 2657 /*
@@ -4221,26 +4251,44 @@ void scheduler_tick(void)
4221#endif 4251#endif
4222} 4252}
4223 4253
4224#if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) 4254#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
4255 defined(CONFIG_PREEMPT_TRACER))
4256
4257static inline unsigned long get_parent_ip(unsigned long addr)
4258{
4259 if (in_lock_functions(addr)) {
4260 addr = CALLER_ADDR2;
4261 if (in_lock_functions(addr))
4262 addr = CALLER_ADDR3;
4263 }
4264 return addr;
4265}
4225 4266
4226void __kprobes add_preempt_count(int val) 4267void __kprobes add_preempt_count(int val)
4227{ 4268{
4269#ifdef CONFIG_DEBUG_PREEMPT
4228 /* 4270 /*
4229 * Underflow? 4271 * Underflow?
4230 */ 4272 */
4231 if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) 4273 if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
4232 return; 4274 return;
4275#endif
4233 preempt_count() += val; 4276 preempt_count() += val;
4277#ifdef CONFIG_DEBUG_PREEMPT
4234 /* 4278 /*
4235 * Spinlock count overflowing soon? 4279 * Spinlock count overflowing soon?
4236 */ 4280 */
4237 DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= 4281 DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
4238 PREEMPT_MASK - 10); 4282 PREEMPT_MASK - 10);
4283#endif
4284 if (preempt_count() == val)
4285 trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
4239} 4286}
4240EXPORT_SYMBOL(add_preempt_count); 4287EXPORT_SYMBOL(add_preempt_count);
4241 4288
4242void __kprobes sub_preempt_count(int val) 4289void __kprobes sub_preempt_count(int val)
4243{ 4290{
4291#ifdef CONFIG_DEBUG_PREEMPT
4244 /* 4292 /*
4245 * Underflow? 4293 * Underflow?
4246 */ 4294 */
@@ -4252,7 +4300,10 @@ void __kprobes sub_preempt_count(int val)
4252 if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && 4300 if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
4253 !(preempt_count() & PREEMPT_MASK))) 4301 !(preempt_count() & PREEMPT_MASK)))
4254 return; 4302 return;
4303#endif
4255 4304
4305 if (preempt_count() == val)
4306 trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
4256 preempt_count() -= val; 4307 preempt_count() -= val;
4257} 4308}
4258EXPORT_SYMBOL(sub_preempt_count); 4309EXPORT_SYMBOL(sub_preempt_count);
@@ -5566,7 +5617,7 @@ out_unlock:
5566 return retval; 5617 return retval;
5567} 5618}
5568 5619
5569static const char stat_nam[] = "RSDTtZX"; 5620static const char stat_nam[] = TASK_STATE_TO_CHAR_STR;
5570 5621
5571void sched_show_task(struct task_struct *p) 5622void sched_show_task(struct task_struct *p)
5572{ 5623{
diff --git a/kernel/semaphore.c b/kernel/semaphore.c
index 5c2942e768cd..aaaeae8244e7 100644
--- a/kernel/semaphore.c
+++ b/kernel/semaphore.c
@@ -31,6 +31,7 @@
31#include <linux/sched.h> 31#include <linux/sched.h>
32#include <linux/semaphore.h> 32#include <linux/semaphore.h>
33#include <linux/spinlock.h> 33#include <linux/spinlock.h>
34#include <linux/ftrace.h>
34 35
35static noinline void __down(struct semaphore *sem); 36static noinline void __down(struct semaphore *sem);
36static noinline int __down_interruptible(struct semaphore *sem); 37static noinline int __down_interruptible(struct semaphore *sem);
diff --git a/kernel/spinlock.c b/kernel/spinlock.c
index ae28c8245123..a1fb54c93cdd 100644
--- a/kernel/spinlock.c
+++ b/kernel/spinlock.c
@@ -436,7 +436,7 @@ int __lockfunc _spin_trylock_bh(spinlock_t *lock)
436} 436}
437EXPORT_SYMBOL(_spin_trylock_bh); 437EXPORT_SYMBOL(_spin_trylock_bh);
438 438
439int in_lock_functions(unsigned long addr) 439notrace int in_lock_functions(unsigned long addr)
440{ 440{
441 /* Linker adds these: start and end of __lockfunc functions */ 441 /* Linker adds these: start and end of __lockfunc functions */
442 extern char __lock_text_start[], __lock_text_end[]; 442 extern char __lock_text_start[], __lock_text_end[];
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index fe8cdc80ff02..18943985ddee 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -46,6 +46,7 @@
46#include <linux/nfs_fs.h> 46#include <linux/nfs_fs.h>
47#include <linux/acpi.h> 47#include <linux/acpi.h>
48#include <linux/reboot.h> 48#include <linux/reboot.h>
49#include <linux/ftrace.h>
49 50
50#include <asm/uaccess.h> 51#include <asm/uaccess.h>
51#include <asm/processor.h> 52#include <asm/processor.h>
@@ -463,6 +464,16 @@ static struct ctl_table kern_table[] = {
463 .mode = 0644, 464 .mode = 0644,
464 .proc_handler = &proc_dointvec, 465 .proc_handler = &proc_dointvec,
465 }, 466 },
467#ifdef CONFIG_FTRACE
468 {
469 .ctl_name = CTL_UNNUMBERED,
470 .procname = "ftrace_enabled",
471 .data = &ftrace_enabled,
472 .maxlen = sizeof(int),
473 .mode = 0644,
474 .proc_handler = &ftrace_enable_sysctl,
475 },
476#endif
466#ifdef CONFIG_KMOD 477#ifdef CONFIG_KMOD
467 { 478 {
468 .ctl_name = KERN_MODPROBE, 479 .ctl_name = KERN_MODPROBE,
diff --git a/kernel/trace/Kconfig b/kernel/trace/Kconfig
new file mode 100644
index 000000000000..263e9e6bbd60
--- /dev/null
+++ b/kernel/trace/Kconfig
@@ -0,0 +1,135 @@
1#
2# Architectures that offer an FTRACE implementation should select HAVE_FTRACE:
3#
4config HAVE_FTRACE
5 bool
6
7config HAVE_DYNAMIC_FTRACE
8 bool
9
10config TRACER_MAX_TRACE
11 bool
12
13config TRACING
14 bool
15 select DEBUG_FS
16 select STACKTRACE
17
18config FTRACE
19 bool "Kernel Function Tracer"
20 depends on HAVE_FTRACE
21 select FRAME_POINTER
22 select TRACING
23 select CONTEXT_SWITCH_TRACER
24 help
25 Enable the kernel to trace every kernel function. This is done
26 by using a compiler feature to insert a small, 5-byte No-Operation
27 instruction to the beginning of every kernel function, which NOP
28 sequence is then dynamically patched into a tracer call when
29 tracing is enabled by the administrator. If it's runtime disabled
30 (the bootup default), then the overhead of the instructions is very
31 small and not measurable even in micro-benchmarks.
32
33config IRQSOFF_TRACER
34 bool "Interrupts-off Latency Tracer"
35 default n
36 depends on TRACE_IRQFLAGS_SUPPORT
37 depends on GENERIC_TIME
38 depends on HAVE_FTRACE
39 select TRACE_IRQFLAGS
40 select TRACING
41 select TRACER_MAX_TRACE
42 help
43 This option measures the time spent in irqs-off critical
44 sections, with microsecond accuracy.
45
46 The default measurement method is a maximum search, which is
47 disabled by default and can be runtime (re-)started
48 via:
49
50 echo 0 > /debugfs/tracing/tracing_max_latency
51
52 (Note that kernel size and overhead increases with this option
53 enabled. This option and the preempt-off timing option can be
54 used together or separately.)
55
56config PREEMPT_TRACER
57 bool "Preemption-off Latency Tracer"
58 default n
59 depends on GENERIC_TIME
60 depends on PREEMPT
61 depends on HAVE_FTRACE
62 select TRACING
63 select TRACER_MAX_TRACE
64 help
65 This option measures the time spent in preemption off critical
66 sections, with microsecond accuracy.
67
68 The default measurement method is a maximum search, which is
69 disabled by default and can be runtime (re-)started
70 via:
71
72 echo 0 > /debugfs/tracing/tracing_max_latency
73
74 (Note that kernel size and overhead increases with this option
75 enabled. This option and the irqs-off timing option can be
76 used together or separately.)
77
78config SYSPROF_TRACER
79 bool "Sysprof Tracer"
80 depends on X86
81 select TRACING
82 help
83 This tracer provides the trace needed by the 'Sysprof' userspace
84 tool.
85
86config SCHED_TRACER
87 bool "Scheduling Latency Tracer"
88 depends on HAVE_FTRACE
89 select TRACING
90 select CONTEXT_SWITCH_TRACER
91 select TRACER_MAX_TRACE
92 help
93 This tracer tracks the latency of the highest priority task
94 to be scheduled in, starting from the point it has woken up.
95
96config CONTEXT_SWITCH_TRACER
97 bool "Trace process context switches"
98 depends on HAVE_FTRACE
99 select TRACING
100 select MARKERS
101 help
102 This tracer gets called from the context switch and records
103 all switching of tasks.
104
105config DYNAMIC_FTRACE
106 bool "enable/disable ftrace tracepoints dynamically"
107 depends on FTRACE
108 depends on HAVE_DYNAMIC_FTRACE
109 default y
110 help
111 This option will modify all the calls to ftrace dynamically
112 (will patch them out of the binary image and replaces them
113 with a No-Op instruction) as they are called. A table is
114 created to dynamically enable them again.
115
116 This way a CONFIG_FTRACE kernel is slightly larger, but otherwise
117 has native performance as long as no tracing is active.
118
119 The changes to the code are done by a kernel thread that
120 wakes up once a second and checks to see if any ftrace calls
121 were made. If so, it runs stop_machine (stops all CPUS)
122 and modifies the code to jump over the call to ftrace.
123
124config FTRACE_SELFTEST
125 bool
126
127config FTRACE_STARTUP_TEST
128 bool "Perform a startup test on ftrace"
129 depends on TRACING
130 select FTRACE_SELFTEST
131 help
132 This option performs a series of startup tests on ftrace. On bootup
133 a series of tests are made to verify that the tracer is
134 functioning properly. It will do tests on all the configured
135 tracers of ftrace.
diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile
new file mode 100644
index 000000000000..71d17de17288
--- /dev/null
+++ b/kernel/trace/Makefile
@@ -0,0 +1,24 @@
1
2# Do not instrument the tracer itself:
3
4ifdef CONFIG_FTRACE
5ORIG_CFLAGS := $(KBUILD_CFLAGS)
6KBUILD_CFLAGS = $(subst -pg,,$(ORIG_CFLAGS))
7
8# selftest needs instrumentation
9CFLAGS_trace_selftest_dynamic.o = -pg
10obj-y += trace_selftest_dynamic.o
11endif
12
13obj-$(CONFIG_FTRACE) += libftrace.o
14
15obj-$(CONFIG_TRACING) += trace.o
16obj-$(CONFIG_CONTEXT_SWITCH_TRACER) += trace_sched_switch.o
17obj-$(CONFIG_SYSPROF_TRACER) += trace_sysprof.o
18obj-$(CONFIG_FTRACE) += trace_functions.o
19obj-$(CONFIG_IRQSOFF_TRACER) += trace_irqsoff.o
20obj-$(CONFIG_PREEMPT_TRACER) += trace_irqsoff.o
21obj-$(CONFIG_SCHED_TRACER) += trace_sched_wakeup.o
22obj-$(CONFIG_MMIOTRACE) += trace_mmiotrace.o
23
24libftrace-y := ftrace.o
diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
new file mode 100644
index 000000000000..4231a3dc224a
--- /dev/null
+++ b/kernel/trace/ftrace.c
@@ -0,0 +1,1727 @@
1/*
2 * Infrastructure for profiling code inserted by 'gcc -pg'.
3 *
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2004-2008 Ingo Molnar <mingo@redhat.com>
6 *
7 * Originally ported from the -rt patch by:
8 * Copyright (C) 2007 Arnaldo Carvalho de Melo <acme@redhat.com>
9 *
10 * Based on code in the latency_tracer, that is:
11 *
12 * Copyright (C) 2004-2006 Ingo Molnar
13 * Copyright (C) 2004 William Lee Irwin III
14 */
15
16#include <linux/stop_machine.h>
17#include <linux/clocksource.h>
18#include <linux/kallsyms.h>
19#include <linux/seq_file.h>
20#include <linux/debugfs.h>
21#include <linux/hardirq.h>
22#include <linux/kthread.h>
23#include <linux/uaccess.h>
24#include <linux/kprobes.h>
25#include <linux/ftrace.h>
26#include <linux/sysctl.h>
27#include <linux/ctype.h>
28#include <linux/hash.h>
29#include <linux/list.h>
30
31#include <asm/ftrace.h>
32
33#include "trace.h"
34
35/* ftrace_enabled is a method to turn ftrace on or off */
36int ftrace_enabled __read_mostly;
37static int last_ftrace_enabled;
38
39/*
40 * ftrace_disabled is set when an anomaly is discovered.
41 * ftrace_disabled is much stronger than ftrace_enabled.
42 */
43static int ftrace_disabled __read_mostly;
44
45static DEFINE_SPINLOCK(ftrace_lock);
46static DEFINE_MUTEX(ftrace_sysctl_lock);
47
48static struct ftrace_ops ftrace_list_end __read_mostly =
49{
50 .func = ftrace_stub,
51};
52
53static struct ftrace_ops *ftrace_list __read_mostly = &ftrace_list_end;
54ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
55
56static void ftrace_list_func(unsigned long ip, unsigned long parent_ip)
57{
58 struct ftrace_ops *op = ftrace_list;
59
60 /* in case someone actually ports this to alpha! */
61 read_barrier_depends();
62
63 while (op != &ftrace_list_end) {
64 /* silly alpha */
65 read_barrier_depends();
66 op->func(ip, parent_ip);
67 op = op->next;
68 };
69}
70
71/**
72 * clear_ftrace_function - reset the ftrace function
73 *
74 * This NULLs the ftrace function and in essence stops
75 * tracing. There may be lag
76 */
77void clear_ftrace_function(void)
78{
79 ftrace_trace_function = ftrace_stub;
80}
81
82static int __register_ftrace_function(struct ftrace_ops *ops)
83{
84 /* Should never be called by interrupts */
85 spin_lock(&ftrace_lock);
86
87 ops->next = ftrace_list;
88 /*
89 * We are entering ops into the ftrace_list but another
90 * CPU might be walking that list. We need to make sure
91 * the ops->next pointer is valid before another CPU sees
92 * the ops pointer included into the ftrace_list.
93 */
94 smp_wmb();
95 ftrace_list = ops;
96
97 if (ftrace_enabled) {
98 /*
99 * For one func, simply call it directly.
100 * For more than one func, call the chain.
101 */
102 if (ops->next == &ftrace_list_end)
103 ftrace_trace_function = ops->func;
104 else
105 ftrace_trace_function = ftrace_list_func;
106 }
107
108 spin_unlock(&ftrace_lock);
109
110 return 0;
111}
112
113static int __unregister_ftrace_function(struct ftrace_ops *ops)
114{
115 struct ftrace_ops **p;
116 int ret = 0;
117
118 spin_lock(&ftrace_lock);
119
120 /*
121 * If we are removing the last function, then simply point
122 * to the ftrace_stub.
123 */
124 if (ftrace_list == ops && ops->next == &ftrace_list_end) {
125 ftrace_trace_function = ftrace_stub;
126 ftrace_list = &ftrace_list_end;
127 goto out;
128 }
129
130 for (p = &ftrace_list; *p != &ftrace_list_end; p = &(*p)->next)
131 if (*p == ops)
132 break;
133
134 if (*p != ops) {
135 ret = -1;
136 goto out;
137 }
138
139 *p = (*p)->next;
140
141 if (ftrace_enabled) {
142 /* If we only have one func left, then call that directly */
143 if (ftrace_list == &ftrace_list_end ||
144 ftrace_list->next == &ftrace_list_end)
145 ftrace_trace_function = ftrace_list->func;
146 }
147
148 out:
149 spin_unlock(&ftrace_lock);
150
151 return ret;
152}
153
154#ifdef CONFIG_DYNAMIC_FTRACE
155
156static struct task_struct *ftraced_task;
157
158enum {
159 FTRACE_ENABLE_CALLS = (1 << 0),
160 FTRACE_DISABLE_CALLS = (1 << 1),
161 FTRACE_UPDATE_TRACE_FUNC = (1 << 2),
162 FTRACE_ENABLE_MCOUNT = (1 << 3),
163 FTRACE_DISABLE_MCOUNT = (1 << 4),
164};
165
166static int ftrace_filtered;
167static int tracing_on;
168static int frozen_record_count;
169
170static struct hlist_head ftrace_hash[FTRACE_HASHSIZE];
171
172static DEFINE_PER_CPU(int, ftrace_shutdown_disable_cpu);
173
174static DEFINE_SPINLOCK(ftrace_shutdown_lock);
175static DEFINE_MUTEX(ftraced_lock);
176static DEFINE_MUTEX(ftrace_regex_lock);
177
178struct ftrace_page {
179 struct ftrace_page *next;
180 unsigned long index;
181 struct dyn_ftrace records[];
182};
183
184#define ENTRIES_PER_PAGE \
185 ((PAGE_SIZE - sizeof(struct ftrace_page)) / sizeof(struct dyn_ftrace))
186
187/* estimate from running different kernels */
188#define NR_TO_INIT 10000
189
190static struct ftrace_page *ftrace_pages_start;
191static struct ftrace_page *ftrace_pages;
192
193static int ftraced_trigger;
194static int ftraced_suspend;
195static int ftraced_stop;
196
197static int ftrace_record_suspend;
198
199static struct dyn_ftrace *ftrace_free_records;
200
201
202#ifdef CONFIG_KPROBES
203static inline void freeze_record(struct dyn_ftrace *rec)
204{
205 if (!(rec->flags & FTRACE_FL_FROZEN)) {
206 rec->flags |= FTRACE_FL_FROZEN;
207 frozen_record_count++;
208 }
209}
210
211static inline void unfreeze_record(struct dyn_ftrace *rec)
212{
213 if (rec->flags & FTRACE_FL_FROZEN) {
214 rec->flags &= ~FTRACE_FL_FROZEN;
215 frozen_record_count--;
216 }
217}
218
219static inline int record_frozen(struct dyn_ftrace *rec)
220{
221 return rec->flags & FTRACE_FL_FROZEN;
222}
223#else
224# define freeze_record(rec) ({ 0; })
225# define unfreeze_record(rec) ({ 0; })
226# define record_frozen(rec) ({ 0; })
227#endif /* CONFIG_KPROBES */
228
229int skip_trace(unsigned long ip)
230{
231 unsigned long fl;
232 struct dyn_ftrace *rec;
233 struct hlist_node *t;
234 struct hlist_head *head;
235
236 if (frozen_record_count == 0)
237 return 0;
238
239 head = &ftrace_hash[hash_long(ip, FTRACE_HASHBITS)];
240 hlist_for_each_entry_rcu(rec, t, head, node) {
241 if (rec->ip == ip) {
242 if (record_frozen(rec)) {
243 if (rec->flags & FTRACE_FL_FAILED)
244 return 1;
245
246 if (!(rec->flags & FTRACE_FL_CONVERTED))
247 return 1;
248
249 if (!tracing_on || !ftrace_enabled)
250 return 1;
251
252 if (ftrace_filtered) {
253 fl = rec->flags & (FTRACE_FL_FILTER |
254 FTRACE_FL_NOTRACE);
255 if (!fl || (fl & FTRACE_FL_NOTRACE))
256 return 1;
257 }
258 }
259 break;
260 }
261 }
262
263 return 0;
264}
265
266static inline int
267ftrace_ip_in_hash(unsigned long ip, unsigned long key)
268{
269 struct dyn_ftrace *p;
270 struct hlist_node *t;
271 int found = 0;
272
273 hlist_for_each_entry_rcu(p, t, &ftrace_hash[key], node) {
274 if (p->ip == ip) {
275 found = 1;
276 break;
277 }
278 }
279
280 return found;
281}
282
283static inline void
284ftrace_add_hash(struct dyn_ftrace *node, unsigned long key)
285{
286 hlist_add_head_rcu(&node->node, &ftrace_hash[key]);
287}
288
289/* called from kstop_machine */
290static inline void ftrace_del_hash(struct dyn_ftrace *node)
291{
292 hlist_del(&node->node);
293}
294
295static void ftrace_free_rec(struct dyn_ftrace *rec)
296{
297 /* no locking, only called from kstop_machine */
298
299 rec->ip = (unsigned long)ftrace_free_records;
300 ftrace_free_records = rec;
301 rec->flags |= FTRACE_FL_FREE;
302}
303
304static struct dyn_ftrace *ftrace_alloc_dyn_node(unsigned long ip)
305{
306 struct dyn_ftrace *rec;
307
308 /* First check for freed records */
309 if (ftrace_free_records) {
310 rec = ftrace_free_records;
311
312 if (unlikely(!(rec->flags & FTRACE_FL_FREE))) {
313 WARN_ON_ONCE(1);
314 ftrace_free_records = NULL;
315 ftrace_disabled = 1;
316 ftrace_enabled = 0;
317 return NULL;
318 }
319
320 ftrace_free_records = (void *)rec->ip;
321 memset(rec, 0, sizeof(*rec));
322 return rec;
323 }
324
325 if (ftrace_pages->index == ENTRIES_PER_PAGE) {
326 if (!ftrace_pages->next)
327 return NULL;
328 ftrace_pages = ftrace_pages->next;
329 }
330
331 return &ftrace_pages->records[ftrace_pages->index++];
332}
333
334static void
335ftrace_record_ip(unsigned long ip)
336{
337 struct dyn_ftrace *node;
338 unsigned long flags;
339 unsigned long key;
340 int resched;
341 int atomic;
342 int cpu;
343
344 if (!ftrace_enabled || ftrace_disabled)
345 return;
346
347 resched = need_resched();
348 preempt_disable_notrace();
349
350 /*
351 * We simply need to protect against recursion.
352 * Use the the raw version of smp_processor_id and not
353 * __get_cpu_var which can call debug hooks that can
354 * cause a recursive crash here.
355 */
356 cpu = raw_smp_processor_id();
357 per_cpu(ftrace_shutdown_disable_cpu, cpu)++;
358 if (per_cpu(ftrace_shutdown_disable_cpu, cpu) != 1)
359 goto out;
360
361 if (unlikely(ftrace_record_suspend))
362 goto out;
363
364 key = hash_long(ip, FTRACE_HASHBITS);
365
366 WARN_ON_ONCE(key >= FTRACE_HASHSIZE);
367
368 if (ftrace_ip_in_hash(ip, key))
369 goto out;
370
371 atomic = irqs_disabled();
372
373 spin_lock_irqsave(&ftrace_shutdown_lock, flags);
374
375 /* This ip may have hit the hash before the lock */
376 if (ftrace_ip_in_hash(ip, key))
377 goto out_unlock;
378
379 node = ftrace_alloc_dyn_node(ip);
380 if (!node)
381 goto out_unlock;
382
383 node->ip = ip;
384
385 ftrace_add_hash(node, key);
386
387 ftraced_trigger = 1;
388
389 out_unlock:
390 spin_unlock_irqrestore(&ftrace_shutdown_lock, flags);
391 out:
392 per_cpu(ftrace_shutdown_disable_cpu, cpu)--;
393
394 /* prevent recursion with scheduler */
395 if (resched)
396 preempt_enable_no_resched_notrace();
397 else
398 preempt_enable_notrace();
399}
400
401#define FTRACE_ADDR ((long)(ftrace_caller))
402
403static int
404__ftrace_replace_code(struct dyn_ftrace *rec,
405 unsigned char *old, unsigned char *new, int enable)
406{
407 unsigned long ip, fl;
408
409 ip = rec->ip;
410
411 if (ftrace_filtered && enable) {
412 /*
413 * If filtering is on:
414 *
415 * If this record is set to be filtered and
416 * is enabled then do nothing.
417 *
418 * If this record is set to be filtered and
419 * it is not enabled, enable it.
420 *
421 * If this record is not set to be filtered
422 * and it is not enabled do nothing.
423 *
424 * If this record is set not to trace then
425 * do nothing.
426 *
427 * If this record is set not to trace and
428 * it is enabled then disable it.
429 *
430 * If this record is not set to be filtered and
431 * it is enabled, disable it.
432 */
433
434 fl = rec->flags & (FTRACE_FL_FILTER | FTRACE_FL_NOTRACE |
435 FTRACE_FL_ENABLED);
436
437 if ((fl == (FTRACE_FL_FILTER | FTRACE_FL_ENABLED)) ||
438 (fl == (FTRACE_FL_FILTER | FTRACE_FL_NOTRACE)) ||
439 !fl || (fl == FTRACE_FL_NOTRACE))
440 return 0;
441
442 /*
443 * If it is enabled disable it,
444 * otherwise enable it!
445 */
446 if (fl & FTRACE_FL_ENABLED) {
447 /* swap new and old */
448 new = old;
449 old = ftrace_call_replace(ip, FTRACE_ADDR);
450 rec->flags &= ~FTRACE_FL_ENABLED;
451 } else {
452 new = ftrace_call_replace(ip, FTRACE_ADDR);
453 rec->flags |= FTRACE_FL_ENABLED;
454 }
455 } else {
456
457 if (enable) {
458 /*
459 * If this record is set not to trace and is
460 * not enabled, do nothing.
461 */
462 fl = rec->flags & (FTRACE_FL_NOTRACE | FTRACE_FL_ENABLED);
463 if (fl == FTRACE_FL_NOTRACE)
464 return 0;
465
466 new = ftrace_call_replace(ip, FTRACE_ADDR);
467 } else
468 old = ftrace_call_replace(ip, FTRACE_ADDR);
469
470 if (enable) {
471 if (rec->flags & FTRACE_FL_ENABLED)
472 return 0;
473 rec->flags |= FTRACE_FL_ENABLED;
474 } else {
475 if (!(rec->flags & FTRACE_FL_ENABLED))
476 return 0;
477 rec->flags &= ~FTRACE_FL_ENABLED;
478 }
479 }
480
481 return ftrace_modify_code(ip, old, new);
482}
483
484static void ftrace_replace_code(int enable)
485{
486 int i, failed;
487 unsigned char *new = NULL, *old = NULL;
488 struct dyn_ftrace *rec;
489 struct ftrace_page *pg;
490
491 if (enable)
492 old = ftrace_nop_replace();
493 else
494 new = ftrace_nop_replace();
495
496 for (pg = ftrace_pages_start; pg; pg = pg->next) {
497 for (i = 0; i < pg->index; i++) {
498 rec = &pg->records[i];
499
500 /* don't modify code that has already faulted */
501 if (rec->flags & FTRACE_FL_FAILED)
502 continue;
503
504 /* ignore updates to this record's mcount site */
505 if (get_kprobe((void *)rec->ip)) {
506 freeze_record(rec);
507 continue;
508 } else {
509 unfreeze_record(rec);
510 }
511
512 failed = __ftrace_replace_code(rec, old, new, enable);
513 if (failed && (rec->flags & FTRACE_FL_CONVERTED)) {
514 rec->flags |= FTRACE_FL_FAILED;
515 if ((system_state == SYSTEM_BOOTING) ||
516 !core_kernel_text(rec->ip)) {
517 ftrace_del_hash(rec);
518 ftrace_free_rec(rec);
519 }
520 }
521 }
522 }
523}
524
525static void ftrace_shutdown_replenish(void)
526{
527 if (ftrace_pages->next)
528 return;
529
530 /* allocate another page */
531 ftrace_pages->next = (void *)get_zeroed_page(GFP_KERNEL);
532}
533
534static int
535ftrace_code_disable(struct dyn_ftrace *rec)
536{
537 unsigned long ip;
538 unsigned char *nop, *call;
539 int failed;
540
541 ip = rec->ip;
542
543 nop = ftrace_nop_replace();
544 call = ftrace_call_replace(ip, MCOUNT_ADDR);
545
546 failed = ftrace_modify_code(ip, call, nop);
547 if (failed) {
548 rec->flags |= FTRACE_FL_FAILED;
549 return 0;
550 }
551 return 1;
552}
553
554static int __ftrace_update_code(void *ignore);
555
556static int __ftrace_modify_code(void *data)
557{
558 unsigned long addr;
559 int *command = data;
560
561 if (*command & FTRACE_ENABLE_CALLS) {
562 /*
563 * Update any recorded ips now that we have the
564 * machine stopped
565 */
566 __ftrace_update_code(NULL);
567 ftrace_replace_code(1);
568 tracing_on = 1;
569 } else if (*command & FTRACE_DISABLE_CALLS) {
570 ftrace_replace_code(0);
571 tracing_on = 0;
572 }
573
574 if (*command & FTRACE_UPDATE_TRACE_FUNC)
575 ftrace_update_ftrace_func(ftrace_trace_function);
576
577 if (*command & FTRACE_ENABLE_MCOUNT) {
578 addr = (unsigned long)ftrace_record_ip;
579 ftrace_mcount_set(&addr);
580 } else if (*command & FTRACE_DISABLE_MCOUNT) {
581 addr = (unsigned long)ftrace_stub;
582 ftrace_mcount_set(&addr);
583 }
584
585 return 0;
586}
587
588static void ftrace_run_update_code(int command)
589{
590 stop_machine_run(__ftrace_modify_code, &command, NR_CPUS);
591}
592
593void ftrace_disable_daemon(void)
594{
595 /* Stop the daemon from calling kstop_machine */
596 mutex_lock(&ftraced_lock);
597 ftraced_stop = 1;
598 mutex_unlock(&ftraced_lock);
599
600 ftrace_force_update();
601}
602
603void ftrace_enable_daemon(void)
604{
605 mutex_lock(&ftraced_lock);
606 ftraced_stop = 0;
607 mutex_unlock(&ftraced_lock);
608
609 ftrace_force_update();
610}
611
612static ftrace_func_t saved_ftrace_func;
613
614static void ftrace_startup(void)
615{
616 int command = 0;
617
618 if (unlikely(ftrace_disabled))
619 return;
620
621 mutex_lock(&ftraced_lock);
622 ftraced_suspend++;
623 if (ftraced_suspend == 1)
624 command |= FTRACE_ENABLE_CALLS;
625
626 if (saved_ftrace_func != ftrace_trace_function) {
627 saved_ftrace_func = ftrace_trace_function;
628 command |= FTRACE_UPDATE_TRACE_FUNC;
629 }
630
631 if (!command || !ftrace_enabled)
632 goto out;
633
634 ftrace_run_update_code(command);
635 out:
636 mutex_unlock(&ftraced_lock);
637}
638
639static void ftrace_shutdown(void)
640{
641 int command = 0;
642
643 if (unlikely(ftrace_disabled))
644 return;
645
646 mutex_lock(&ftraced_lock);
647 ftraced_suspend--;
648 if (!ftraced_suspend)
649 command |= FTRACE_DISABLE_CALLS;
650
651 if (saved_ftrace_func != ftrace_trace_function) {
652 saved_ftrace_func = ftrace_trace_function;
653 command |= FTRACE_UPDATE_TRACE_FUNC;
654 }
655
656 if (!command || !ftrace_enabled)
657 goto out;
658
659 ftrace_run_update_code(command);
660 out:
661 mutex_unlock(&ftraced_lock);
662}
663
664static void ftrace_startup_sysctl(void)
665{
666 int command = FTRACE_ENABLE_MCOUNT;
667
668 if (unlikely(ftrace_disabled))
669 return;
670
671 mutex_lock(&ftraced_lock);
672 /* Force update next time */
673 saved_ftrace_func = NULL;
674 /* ftraced_suspend is true if we want ftrace running */
675 if (ftraced_suspend)
676 command |= FTRACE_ENABLE_CALLS;
677
678 ftrace_run_update_code(command);
679 mutex_unlock(&ftraced_lock);
680}
681
682static void ftrace_shutdown_sysctl(void)
683{
684 int command = FTRACE_DISABLE_MCOUNT;
685
686 if (unlikely(ftrace_disabled))
687 return;
688
689 mutex_lock(&ftraced_lock);
690 /* ftraced_suspend is true if ftrace is running */
691 if (ftraced_suspend)
692 command |= FTRACE_DISABLE_CALLS;
693
694 ftrace_run_update_code(command);
695 mutex_unlock(&ftraced_lock);
696}
697
698static cycle_t ftrace_update_time;
699static unsigned long ftrace_update_cnt;
700unsigned long ftrace_update_tot_cnt;
701
702static int __ftrace_update_code(void *ignore)
703{
704 int i, save_ftrace_enabled;
705 cycle_t start, stop;
706 struct dyn_ftrace *p;
707 struct hlist_node *t, *n;
708 struct hlist_head *head, temp_list;
709
710 /* Don't be recording funcs now */
711 ftrace_record_suspend++;
712 save_ftrace_enabled = ftrace_enabled;
713 ftrace_enabled = 0;
714
715 start = ftrace_now(raw_smp_processor_id());
716 ftrace_update_cnt = 0;
717
718 /* No locks needed, the machine is stopped! */
719 for (i = 0; i < FTRACE_HASHSIZE; i++) {
720 INIT_HLIST_HEAD(&temp_list);
721 head = &ftrace_hash[i];
722
723 /* all CPUS are stopped, we are safe to modify code */
724 hlist_for_each_entry_safe(p, t, n, head, node) {
725 /* Skip over failed records which have not been
726 * freed. */
727 if (p->flags & FTRACE_FL_FAILED)
728 continue;
729
730 /* Unconverted records are always at the head of the
731 * hash bucket. Once we encounter a converted record,
732 * simply skip over to the next bucket. Saves ftraced
733 * some processor cycles (ftrace does its bid for
734 * global warming :-p ). */
735 if (p->flags & (FTRACE_FL_CONVERTED))
736 break;
737
738 /* Ignore updates to this record's mcount site.
739 * Reintroduce this record at the head of this
740 * bucket to attempt to "convert" it again if
741 * the kprobe on it is unregistered before the
742 * next run. */
743 if (get_kprobe((void *)p->ip)) {
744 ftrace_del_hash(p);
745 INIT_HLIST_NODE(&p->node);
746 hlist_add_head(&p->node, &temp_list);
747 freeze_record(p);
748 continue;
749 } else {
750 unfreeze_record(p);
751 }
752
753 /* convert record (i.e, patch mcount-call with NOP) */
754 if (ftrace_code_disable(p)) {
755 p->flags |= FTRACE_FL_CONVERTED;
756 ftrace_update_cnt++;
757 } else {
758 if ((system_state == SYSTEM_BOOTING) ||
759 !core_kernel_text(p->ip)) {
760 ftrace_del_hash(p);
761 ftrace_free_rec(p);
762 }
763 }
764 }
765
766 hlist_for_each_entry_safe(p, t, n, &temp_list, node) {
767 hlist_del(&p->node);
768 INIT_HLIST_NODE(&p->node);
769 hlist_add_head(&p->node, head);
770 }
771 }
772
773 stop = ftrace_now(raw_smp_processor_id());
774 ftrace_update_time = stop - start;
775 ftrace_update_tot_cnt += ftrace_update_cnt;
776 ftraced_trigger = 0;
777
778 ftrace_enabled = save_ftrace_enabled;
779 ftrace_record_suspend--;
780
781 return 0;
782}
783
784static int ftrace_update_code(void)
785{
786 if (unlikely(ftrace_disabled) ||
787 !ftrace_enabled || !ftraced_trigger)
788 return 0;
789
790 stop_machine_run(__ftrace_update_code, NULL, NR_CPUS);
791
792 return 1;
793}
794
795static int ftraced(void *ignore)
796{
797 unsigned long usecs;
798
799 while (!kthread_should_stop()) {
800
801 set_current_state(TASK_INTERRUPTIBLE);
802
803 /* check once a second */
804 schedule_timeout(HZ);
805
806 if (unlikely(ftrace_disabled))
807 continue;
808
809 mutex_lock(&ftrace_sysctl_lock);
810 mutex_lock(&ftraced_lock);
811 if (!ftraced_suspend && !ftraced_stop &&
812 ftrace_update_code()) {
813 usecs = nsecs_to_usecs(ftrace_update_time);
814 if (ftrace_update_tot_cnt > 100000) {
815 ftrace_update_tot_cnt = 0;
816 pr_info("hm, dftrace overflow: %lu change%s"
817 " (%lu total) in %lu usec%s\n",
818 ftrace_update_cnt,
819 ftrace_update_cnt != 1 ? "s" : "",
820 ftrace_update_tot_cnt,
821 usecs, usecs != 1 ? "s" : "");
822 ftrace_disabled = 1;
823 WARN_ON_ONCE(1);
824 }
825 }
826 mutex_unlock(&ftraced_lock);
827 mutex_unlock(&ftrace_sysctl_lock);
828
829 ftrace_shutdown_replenish();
830 }
831 __set_current_state(TASK_RUNNING);
832 return 0;
833}
834
835static int __init ftrace_dyn_table_alloc(void)
836{
837 struct ftrace_page *pg;
838 int cnt;
839 int i;
840
841 /* allocate a few pages */
842 ftrace_pages_start = (void *)get_zeroed_page(GFP_KERNEL);
843 if (!ftrace_pages_start)
844 return -1;
845
846 /*
847 * Allocate a few more pages.
848 *
849 * TODO: have some parser search vmlinux before
850 * final linking to find all calls to ftrace.
851 * Then we can:
852 * a) know how many pages to allocate.
853 * and/or
854 * b) set up the table then.
855 *
856 * The dynamic code is still necessary for
857 * modules.
858 */
859
860 pg = ftrace_pages = ftrace_pages_start;
861
862 cnt = NR_TO_INIT / ENTRIES_PER_PAGE;
863
864 for (i = 0; i < cnt; i++) {
865 pg->next = (void *)get_zeroed_page(GFP_KERNEL);
866
867 /* If we fail, we'll try later anyway */
868 if (!pg->next)
869 break;
870
871 pg = pg->next;
872 }
873
874 return 0;
875}
876
877enum {
878 FTRACE_ITER_FILTER = (1 << 0),
879 FTRACE_ITER_CONT = (1 << 1),
880 FTRACE_ITER_NOTRACE = (1 << 2),
881 FTRACE_ITER_FAILURES = (1 << 3),
882};
883
884#define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
885
886struct ftrace_iterator {
887 loff_t pos;
888 struct ftrace_page *pg;
889 unsigned idx;
890 unsigned flags;
891 unsigned char buffer[FTRACE_BUFF_MAX+1];
892 unsigned buffer_idx;
893 unsigned filtered;
894};
895
896static void *
897t_next(struct seq_file *m, void *v, loff_t *pos)
898{
899 struct ftrace_iterator *iter = m->private;
900 struct dyn_ftrace *rec = NULL;
901
902 (*pos)++;
903
904 retry:
905 if (iter->idx >= iter->pg->index) {
906 if (iter->pg->next) {
907 iter->pg = iter->pg->next;
908 iter->idx = 0;
909 goto retry;
910 }
911 } else {
912 rec = &iter->pg->records[iter->idx++];
913 if ((!(iter->flags & FTRACE_ITER_FAILURES) &&
914 (rec->flags & FTRACE_FL_FAILED)) ||
915
916 ((iter->flags & FTRACE_ITER_FAILURES) &&
917 (!(rec->flags & FTRACE_FL_FAILED) ||
918 (rec->flags & FTRACE_FL_FREE))) ||
919
920 ((iter->flags & FTRACE_ITER_FILTER) &&
921 !(rec->flags & FTRACE_FL_FILTER)) ||
922
923 ((iter->flags & FTRACE_ITER_NOTRACE) &&
924 !(rec->flags & FTRACE_FL_NOTRACE))) {
925 rec = NULL;
926 goto retry;
927 }
928 }
929
930 iter->pos = *pos;
931
932 return rec;
933}
934
935static void *t_start(struct seq_file *m, loff_t *pos)
936{
937 struct ftrace_iterator *iter = m->private;
938 void *p = NULL;
939 loff_t l = -1;
940
941 if (*pos != iter->pos) {
942 for (p = t_next(m, p, &l); p && l < *pos; p = t_next(m, p, &l))
943 ;
944 } else {
945 l = *pos;
946 p = t_next(m, p, &l);
947 }
948
949 return p;
950}
951
952static void t_stop(struct seq_file *m, void *p)
953{
954}
955
956static int t_show(struct seq_file *m, void *v)
957{
958 struct dyn_ftrace *rec = v;
959 char str[KSYM_SYMBOL_LEN];
960
961 if (!rec)
962 return 0;
963
964 kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
965
966 seq_printf(m, "%s\n", str);
967
968 return 0;
969}
970
971static struct seq_operations show_ftrace_seq_ops = {
972 .start = t_start,
973 .next = t_next,
974 .stop = t_stop,
975 .show = t_show,
976};
977
978static int
979ftrace_avail_open(struct inode *inode, struct file *file)
980{
981 struct ftrace_iterator *iter;
982 int ret;
983
984 if (unlikely(ftrace_disabled))
985 return -ENODEV;
986
987 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
988 if (!iter)
989 return -ENOMEM;
990
991 iter->pg = ftrace_pages_start;
992 iter->pos = -1;
993
994 ret = seq_open(file, &show_ftrace_seq_ops);
995 if (!ret) {
996 struct seq_file *m = file->private_data;
997
998 m->private = iter;
999 } else {
1000 kfree(iter);
1001 }
1002
1003 return ret;
1004}
1005
1006int ftrace_avail_release(struct inode *inode, struct file *file)
1007{
1008 struct seq_file *m = (struct seq_file *)file->private_data;
1009 struct ftrace_iterator *iter = m->private;
1010
1011 seq_release(inode, file);
1012 kfree(iter);
1013
1014 return 0;
1015}
1016
1017static int
1018ftrace_failures_open(struct inode *inode, struct file *file)
1019{
1020 int ret;
1021 struct seq_file *m;
1022 struct ftrace_iterator *iter;
1023
1024 ret = ftrace_avail_open(inode, file);
1025 if (!ret) {
1026 m = (struct seq_file *)file->private_data;
1027 iter = (struct ftrace_iterator *)m->private;
1028 iter->flags = FTRACE_ITER_FAILURES;
1029 }
1030
1031 return ret;
1032}
1033
1034
1035static void ftrace_filter_reset(int enable)
1036{
1037 struct ftrace_page *pg;
1038 struct dyn_ftrace *rec;
1039 unsigned long type = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1040 unsigned i;
1041
1042 /* keep kstop machine from running */
1043 preempt_disable();
1044 if (enable)
1045 ftrace_filtered = 0;
1046 pg = ftrace_pages_start;
1047 while (pg) {
1048 for (i = 0; i < pg->index; i++) {
1049 rec = &pg->records[i];
1050 if (rec->flags & FTRACE_FL_FAILED)
1051 continue;
1052 rec->flags &= ~type;
1053 }
1054 pg = pg->next;
1055 }
1056 preempt_enable();
1057}
1058
1059static int
1060ftrace_regex_open(struct inode *inode, struct file *file, int enable)
1061{
1062 struct ftrace_iterator *iter;
1063 int ret = 0;
1064
1065 if (unlikely(ftrace_disabled))
1066 return -ENODEV;
1067
1068 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1069 if (!iter)
1070 return -ENOMEM;
1071
1072 mutex_lock(&ftrace_regex_lock);
1073 if ((file->f_mode & FMODE_WRITE) &&
1074 !(file->f_flags & O_APPEND))
1075 ftrace_filter_reset(enable);
1076
1077 if (file->f_mode & FMODE_READ) {
1078 iter->pg = ftrace_pages_start;
1079 iter->pos = -1;
1080 iter->flags = enable ? FTRACE_ITER_FILTER :
1081 FTRACE_ITER_NOTRACE;
1082
1083 ret = seq_open(file, &show_ftrace_seq_ops);
1084 if (!ret) {
1085 struct seq_file *m = file->private_data;
1086 m->private = iter;
1087 } else
1088 kfree(iter);
1089 } else
1090 file->private_data = iter;
1091 mutex_unlock(&ftrace_regex_lock);
1092
1093 return ret;
1094}
1095
1096static int
1097ftrace_filter_open(struct inode *inode, struct file *file)
1098{
1099 return ftrace_regex_open(inode, file, 1);
1100}
1101
1102static int
1103ftrace_notrace_open(struct inode *inode, struct file *file)
1104{
1105 return ftrace_regex_open(inode, file, 0);
1106}
1107
1108static ssize_t
1109ftrace_regex_read(struct file *file, char __user *ubuf,
1110 size_t cnt, loff_t *ppos)
1111{
1112 if (file->f_mode & FMODE_READ)
1113 return seq_read(file, ubuf, cnt, ppos);
1114 else
1115 return -EPERM;
1116}
1117
1118static loff_t
1119ftrace_regex_lseek(struct file *file, loff_t offset, int origin)
1120{
1121 loff_t ret;
1122
1123 if (file->f_mode & FMODE_READ)
1124 ret = seq_lseek(file, offset, origin);
1125 else
1126 file->f_pos = ret = 1;
1127
1128 return ret;
1129}
1130
1131enum {
1132 MATCH_FULL,
1133 MATCH_FRONT_ONLY,
1134 MATCH_MIDDLE_ONLY,
1135 MATCH_END_ONLY,
1136};
1137
1138static void
1139ftrace_match(unsigned char *buff, int len, int enable)
1140{
1141 char str[KSYM_SYMBOL_LEN];
1142 char *search = NULL;
1143 struct ftrace_page *pg;
1144 struct dyn_ftrace *rec;
1145 int type = MATCH_FULL;
1146 unsigned long flag = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1147 unsigned i, match = 0, search_len = 0;
1148
1149 for (i = 0; i < len; i++) {
1150 if (buff[i] == '*') {
1151 if (!i) {
1152 search = buff + i + 1;
1153 type = MATCH_END_ONLY;
1154 search_len = len - (i + 1);
1155 } else {
1156 if (type == MATCH_END_ONLY) {
1157 type = MATCH_MIDDLE_ONLY;
1158 } else {
1159 match = i;
1160 type = MATCH_FRONT_ONLY;
1161 }
1162 buff[i] = 0;
1163 break;
1164 }
1165 }
1166 }
1167
1168 /* keep kstop machine from running */
1169 preempt_disable();
1170 if (enable)
1171 ftrace_filtered = 1;
1172 pg = ftrace_pages_start;
1173 while (pg) {
1174 for (i = 0; i < pg->index; i++) {
1175 int matched = 0;
1176 char *ptr;
1177
1178 rec = &pg->records[i];
1179 if (rec->flags & FTRACE_FL_FAILED)
1180 continue;
1181 kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
1182 switch (type) {
1183 case MATCH_FULL:
1184 if (strcmp(str, buff) == 0)
1185 matched = 1;
1186 break;
1187 case MATCH_FRONT_ONLY:
1188 if (memcmp(str, buff, match) == 0)
1189 matched = 1;
1190 break;
1191 case MATCH_MIDDLE_ONLY:
1192 if (strstr(str, search))
1193 matched = 1;
1194 break;
1195 case MATCH_END_ONLY:
1196 ptr = strstr(str, search);
1197 if (ptr && (ptr[search_len] == 0))
1198 matched = 1;
1199 break;
1200 }
1201 if (matched)
1202 rec->flags |= flag;
1203 }
1204 pg = pg->next;
1205 }
1206 preempt_enable();
1207}
1208
1209static ssize_t
1210ftrace_regex_write(struct file *file, const char __user *ubuf,
1211 size_t cnt, loff_t *ppos, int enable)
1212{
1213 struct ftrace_iterator *iter;
1214 char ch;
1215 size_t read = 0;
1216 ssize_t ret;
1217
1218 if (!cnt || cnt < 0)
1219 return 0;
1220
1221 mutex_lock(&ftrace_regex_lock);
1222
1223 if (file->f_mode & FMODE_READ) {
1224 struct seq_file *m = file->private_data;
1225 iter = m->private;
1226 } else
1227 iter = file->private_data;
1228
1229 if (!*ppos) {
1230 iter->flags &= ~FTRACE_ITER_CONT;
1231 iter->buffer_idx = 0;
1232 }
1233
1234 ret = get_user(ch, ubuf++);
1235 if (ret)
1236 goto out;
1237 read++;
1238 cnt--;
1239
1240 if (!(iter->flags & ~FTRACE_ITER_CONT)) {
1241 /* skip white space */
1242 while (cnt && isspace(ch)) {
1243 ret = get_user(ch, ubuf++);
1244 if (ret)
1245 goto out;
1246 read++;
1247 cnt--;
1248 }
1249
1250 if (isspace(ch)) {
1251 file->f_pos += read;
1252 ret = read;
1253 goto out;
1254 }
1255
1256 iter->buffer_idx = 0;
1257 }
1258
1259 while (cnt && !isspace(ch)) {
1260 if (iter->buffer_idx < FTRACE_BUFF_MAX)
1261 iter->buffer[iter->buffer_idx++] = ch;
1262 else {
1263 ret = -EINVAL;
1264 goto out;
1265 }
1266 ret = get_user(ch, ubuf++);
1267 if (ret)
1268 goto out;
1269 read++;
1270 cnt--;
1271 }
1272
1273 if (isspace(ch)) {
1274 iter->filtered++;
1275 iter->buffer[iter->buffer_idx] = 0;
1276 ftrace_match(iter->buffer, iter->buffer_idx, enable);
1277 iter->buffer_idx = 0;
1278 } else
1279 iter->flags |= FTRACE_ITER_CONT;
1280
1281
1282 file->f_pos += read;
1283
1284 ret = read;
1285 out:
1286 mutex_unlock(&ftrace_regex_lock);
1287
1288 return ret;
1289}
1290
1291static ssize_t
1292ftrace_filter_write(struct file *file, const char __user *ubuf,
1293 size_t cnt, loff_t *ppos)
1294{
1295 return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
1296}
1297
1298static ssize_t
1299ftrace_notrace_write(struct file *file, const char __user *ubuf,
1300 size_t cnt, loff_t *ppos)
1301{
1302 return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
1303}
1304
1305static void
1306ftrace_set_regex(unsigned char *buf, int len, int reset, int enable)
1307{
1308 if (unlikely(ftrace_disabled))
1309 return;
1310
1311 mutex_lock(&ftrace_regex_lock);
1312 if (reset)
1313 ftrace_filter_reset(enable);
1314 if (buf)
1315 ftrace_match(buf, len, enable);
1316 mutex_unlock(&ftrace_regex_lock);
1317}
1318
1319/**
1320 * ftrace_set_filter - set a function to filter on in ftrace
1321 * @buf - the string that holds the function filter text.
1322 * @len - the length of the string.
1323 * @reset - non zero to reset all filters before applying this filter.
1324 *
1325 * Filters denote which functions should be enabled when tracing is enabled.
1326 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
1327 */
1328void ftrace_set_filter(unsigned char *buf, int len, int reset)
1329{
1330 ftrace_set_regex(buf, len, reset, 1);
1331}
1332
1333/**
1334 * ftrace_set_notrace - set a function to not trace in ftrace
1335 * @buf - the string that holds the function notrace text.
1336 * @len - the length of the string.
1337 * @reset - non zero to reset all filters before applying this filter.
1338 *
1339 * Notrace Filters denote which functions should not be enabled when tracing
1340 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
1341 * for tracing.
1342 */
1343void ftrace_set_notrace(unsigned char *buf, int len, int reset)
1344{
1345 ftrace_set_regex(buf, len, reset, 0);
1346}
1347
1348static int
1349ftrace_regex_release(struct inode *inode, struct file *file, int enable)
1350{
1351 struct seq_file *m = (struct seq_file *)file->private_data;
1352 struct ftrace_iterator *iter;
1353
1354 mutex_lock(&ftrace_regex_lock);
1355 if (file->f_mode & FMODE_READ) {
1356 iter = m->private;
1357
1358 seq_release(inode, file);
1359 } else
1360 iter = file->private_data;
1361
1362 if (iter->buffer_idx) {
1363 iter->filtered++;
1364 iter->buffer[iter->buffer_idx] = 0;
1365 ftrace_match(iter->buffer, iter->buffer_idx, enable);
1366 }
1367
1368 mutex_lock(&ftrace_sysctl_lock);
1369 mutex_lock(&ftraced_lock);
1370 if (iter->filtered && ftraced_suspend && ftrace_enabled)
1371 ftrace_run_update_code(FTRACE_ENABLE_CALLS);
1372 mutex_unlock(&ftraced_lock);
1373 mutex_unlock(&ftrace_sysctl_lock);
1374
1375 kfree(iter);
1376 mutex_unlock(&ftrace_regex_lock);
1377 return 0;
1378}
1379
1380static int
1381ftrace_filter_release(struct inode *inode, struct file *file)
1382{
1383 return ftrace_regex_release(inode, file, 1);
1384}
1385
1386static int
1387ftrace_notrace_release(struct inode *inode, struct file *file)
1388{
1389 return ftrace_regex_release(inode, file, 0);
1390}
1391
1392static ssize_t
1393ftraced_read(struct file *filp, char __user *ubuf,
1394 size_t cnt, loff_t *ppos)
1395{
1396 /* don't worry about races */
1397 char *buf = ftraced_stop ? "disabled\n" : "enabled\n";
1398 int r = strlen(buf);
1399
1400 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1401}
1402
1403static ssize_t
1404ftraced_write(struct file *filp, const char __user *ubuf,
1405 size_t cnt, loff_t *ppos)
1406{
1407 char buf[64];
1408 long val;
1409 int ret;
1410
1411 if (cnt >= sizeof(buf))
1412 return -EINVAL;
1413
1414 if (copy_from_user(&buf, ubuf, cnt))
1415 return -EFAULT;
1416
1417 if (strncmp(buf, "enable", 6) == 0)
1418 val = 1;
1419 else if (strncmp(buf, "disable", 7) == 0)
1420 val = 0;
1421 else {
1422 buf[cnt] = 0;
1423
1424 ret = strict_strtoul(buf, 10, &val);
1425 if (ret < 0)
1426 return ret;
1427
1428 val = !!val;
1429 }
1430
1431 if (val)
1432 ftrace_enable_daemon();
1433 else
1434 ftrace_disable_daemon();
1435
1436 filp->f_pos += cnt;
1437
1438 return cnt;
1439}
1440
1441static struct file_operations ftrace_avail_fops = {
1442 .open = ftrace_avail_open,
1443 .read = seq_read,
1444 .llseek = seq_lseek,
1445 .release = ftrace_avail_release,
1446};
1447
1448static struct file_operations ftrace_failures_fops = {
1449 .open = ftrace_failures_open,
1450 .read = seq_read,
1451 .llseek = seq_lseek,
1452 .release = ftrace_avail_release,
1453};
1454
1455static struct file_operations ftrace_filter_fops = {
1456 .open = ftrace_filter_open,
1457 .read = ftrace_regex_read,
1458 .write = ftrace_filter_write,
1459 .llseek = ftrace_regex_lseek,
1460 .release = ftrace_filter_release,
1461};
1462
1463static struct file_operations ftrace_notrace_fops = {
1464 .open = ftrace_notrace_open,
1465 .read = ftrace_regex_read,
1466 .write = ftrace_notrace_write,
1467 .llseek = ftrace_regex_lseek,
1468 .release = ftrace_notrace_release,
1469};
1470
1471static struct file_operations ftraced_fops = {
1472 .open = tracing_open_generic,
1473 .read = ftraced_read,
1474 .write = ftraced_write,
1475};
1476
1477/**
1478 * ftrace_force_update - force an update to all recording ftrace functions
1479 */
1480int ftrace_force_update(void)
1481{
1482 int ret = 0;
1483
1484 if (unlikely(ftrace_disabled))
1485 return -ENODEV;
1486
1487 mutex_lock(&ftrace_sysctl_lock);
1488 mutex_lock(&ftraced_lock);
1489
1490 /*
1491 * If ftraced_trigger is not set, then there is nothing
1492 * to update.
1493 */
1494 if (ftraced_trigger && !ftrace_update_code())
1495 ret = -EBUSY;
1496
1497 mutex_unlock(&ftraced_lock);
1498 mutex_unlock(&ftrace_sysctl_lock);
1499
1500 return ret;
1501}
1502
1503static void ftrace_force_shutdown(void)
1504{
1505 struct task_struct *task;
1506 int command = FTRACE_DISABLE_CALLS | FTRACE_UPDATE_TRACE_FUNC;
1507
1508 mutex_lock(&ftraced_lock);
1509 task = ftraced_task;
1510 ftraced_task = NULL;
1511 ftraced_suspend = -1;
1512 ftrace_run_update_code(command);
1513 mutex_unlock(&ftraced_lock);
1514
1515 if (task)
1516 kthread_stop(task);
1517}
1518
1519static __init int ftrace_init_debugfs(void)
1520{
1521 struct dentry *d_tracer;
1522 struct dentry *entry;
1523
1524 d_tracer = tracing_init_dentry();
1525
1526 entry = debugfs_create_file("available_filter_functions", 0444,
1527 d_tracer, NULL, &ftrace_avail_fops);
1528 if (!entry)
1529 pr_warning("Could not create debugfs "
1530 "'available_filter_functions' entry\n");
1531
1532 entry = debugfs_create_file("failures", 0444,
1533 d_tracer, NULL, &ftrace_failures_fops);
1534 if (!entry)
1535 pr_warning("Could not create debugfs 'failures' entry\n");
1536
1537 entry = debugfs_create_file("set_ftrace_filter", 0644, d_tracer,
1538 NULL, &ftrace_filter_fops);
1539 if (!entry)
1540 pr_warning("Could not create debugfs "
1541 "'set_ftrace_filter' entry\n");
1542
1543 entry = debugfs_create_file("set_ftrace_notrace", 0644, d_tracer,
1544 NULL, &ftrace_notrace_fops);
1545 if (!entry)
1546 pr_warning("Could not create debugfs "
1547 "'set_ftrace_notrace' entry\n");
1548
1549 entry = debugfs_create_file("ftraced_enabled", 0644, d_tracer,
1550 NULL, &ftraced_fops);
1551 if (!entry)
1552 pr_warning("Could not create debugfs "
1553 "'ftraced_enabled' entry\n");
1554 return 0;
1555}
1556
1557fs_initcall(ftrace_init_debugfs);
1558
1559static int __init ftrace_dynamic_init(void)
1560{
1561 struct task_struct *p;
1562 unsigned long addr;
1563 int ret;
1564
1565 addr = (unsigned long)ftrace_record_ip;
1566
1567 stop_machine_run(ftrace_dyn_arch_init, &addr, NR_CPUS);
1568
1569 /* ftrace_dyn_arch_init places the return code in addr */
1570 if (addr) {
1571 ret = (int)addr;
1572 goto failed;
1573 }
1574
1575 ret = ftrace_dyn_table_alloc();
1576 if (ret)
1577 goto failed;
1578
1579 p = kthread_run(ftraced, NULL, "ftraced");
1580 if (IS_ERR(p)) {
1581 ret = -1;
1582 goto failed;
1583 }
1584
1585 last_ftrace_enabled = ftrace_enabled = 1;
1586 ftraced_task = p;
1587
1588 return 0;
1589
1590 failed:
1591 ftrace_disabled = 1;
1592 return ret;
1593}
1594
1595core_initcall(ftrace_dynamic_init);
1596#else
1597# define ftrace_startup() do { } while (0)
1598# define ftrace_shutdown() do { } while (0)
1599# define ftrace_startup_sysctl() do { } while (0)
1600# define ftrace_shutdown_sysctl() do { } while (0)
1601# define ftrace_force_shutdown() do { } while (0)
1602#endif /* CONFIG_DYNAMIC_FTRACE */
1603
1604/**
1605 * ftrace_kill_atomic - kill ftrace from critical sections
1606 *
1607 * This function should be used by panic code. It stops ftrace
1608 * but in a not so nice way. If you need to simply kill ftrace
1609 * from a non-atomic section, use ftrace_kill.
1610 */
1611void ftrace_kill_atomic(void)
1612{
1613 ftrace_disabled = 1;
1614 ftrace_enabled = 0;
1615#ifdef CONFIG_DYNAMIC_FTRACE
1616 ftraced_suspend = -1;
1617#endif
1618 clear_ftrace_function();
1619}
1620
1621/**
1622 * ftrace_kill - totally shutdown ftrace
1623 *
1624 * This is a safety measure. If something was detected that seems
1625 * wrong, calling this function will keep ftrace from doing
1626 * any more modifications, and updates.
1627 * used when something went wrong.
1628 */
1629void ftrace_kill(void)
1630{
1631 mutex_lock(&ftrace_sysctl_lock);
1632 ftrace_disabled = 1;
1633 ftrace_enabled = 0;
1634
1635 clear_ftrace_function();
1636 mutex_unlock(&ftrace_sysctl_lock);
1637
1638 /* Try to totally disable ftrace */
1639 ftrace_force_shutdown();
1640}
1641
1642/**
1643 * register_ftrace_function - register a function for profiling
1644 * @ops - ops structure that holds the function for profiling.
1645 *
1646 * Register a function to be called by all functions in the
1647 * kernel.
1648 *
1649 * Note: @ops->func and all the functions it calls must be labeled
1650 * with "notrace", otherwise it will go into a
1651 * recursive loop.
1652 */
1653int register_ftrace_function(struct ftrace_ops *ops)
1654{
1655 int ret;
1656
1657 if (unlikely(ftrace_disabled))
1658 return -1;
1659
1660 mutex_lock(&ftrace_sysctl_lock);
1661 ret = __register_ftrace_function(ops);
1662 ftrace_startup();
1663 mutex_unlock(&ftrace_sysctl_lock);
1664
1665 return ret;
1666}
1667
1668/**
1669 * unregister_ftrace_function - unresgister a function for profiling.
1670 * @ops - ops structure that holds the function to unregister
1671 *
1672 * Unregister a function that was added to be called by ftrace profiling.
1673 */
1674int unregister_ftrace_function(struct ftrace_ops *ops)
1675{
1676 int ret;
1677
1678 mutex_lock(&ftrace_sysctl_lock);
1679 ret = __unregister_ftrace_function(ops);
1680 ftrace_shutdown();
1681 mutex_unlock(&ftrace_sysctl_lock);
1682
1683 return ret;
1684}
1685
1686int
1687ftrace_enable_sysctl(struct ctl_table *table, int write,
1688 struct file *file, void __user *buffer, size_t *lenp,
1689 loff_t *ppos)
1690{
1691 int ret;
1692
1693 if (unlikely(ftrace_disabled))
1694 return -ENODEV;
1695
1696 mutex_lock(&ftrace_sysctl_lock);
1697
1698 ret = proc_dointvec(table, write, file, buffer, lenp, ppos);
1699
1700 if (ret || !write || (last_ftrace_enabled == ftrace_enabled))
1701 goto out;
1702
1703 last_ftrace_enabled = ftrace_enabled;
1704
1705 if (ftrace_enabled) {
1706
1707 ftrace_startup_sysctl();
1708
1709 /* we are starting ftrace again */
1710 if (ftrace_list != &ftrace_list_end) {
1711 if (ftrace_list->next == &ftrace_list_end)
1712 ftrace_trace_function = ftrace_list->func;
1713 else
1714 ftrace_trace_function = ftrace_list_func;
1715 }
1716
1717 } else {
1718 /* stopping ftrace calls (just send to ftrace_stub) */
1719 ftrace_trace_function = ftrace_stub;
1720
1721 ftrace_shutdown_sysctl();
1722 }
1723
1724 out:
1725 mutex_unlock(&ftrace_sysctl_lock);
1726 return ret;
1727}
diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c
new file mode 100644
index 000000000000..868e121c8e38
--- /dev/null
+++ b/kernel/trace/trace.c
@@ -0,0 +1,3161 @@
1/*
2 * ring buffer based function tracer
3 *
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
6 *
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
9 *
10 * Based on code from the latency_tracer, that is:
11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 William Lee Irwin III
13 */
14#include <linux/utsrelease.h>
15#include <linux/kallsyms.h>
16#include <linux/seq_file.h>
17#include <linux/debugfs.h>
18#include <linux/pagemap.h>
19#include <linux/hardirq.h>
20#include <linux/linkage.h>
21#include <linux/uaccess.h>
22#include <linux/ftrace.h>
23#include <linux/module.h>
24#include <linux/percpu.h>
25#include <linux/ctype.h>
26#include <linux/init.h>
27#include <linux/poll.h>
28#include <linux/gfp.h>
29#include <linux/fs.h>
30#include <linux/kprobes.h>
31#include <linux/writeback.h>
32
33#include <linux/stacktrace.h>
34
35#include "trace.h"
36
37unsigned long __read_mostly tracing_max_latency = (cycle_t)ULONG_MAX;
38unsigned long __read_mostly tracing_thresh;
39
40static unsigned long __read_mostly tracing_nr_buffers;
41static cpumask_t __read_mostly tracing_buffer_mask;
42
43#define for_each_tracing_cpu(cpu) \
44 for_each_cpu_mask(cpu, tracing_buffer_mask)
45
46static int trace_alloc_page(void);
47static int trace_free_page(void);
48
49static int tracing_disabled = 1;
50
51static unsigned long tracing_pages_allocated;
52
53long
54ns2usecs(cycle_t nsec)
55{
56 nsec += 500;
57 do_div(nsec, 1000);
58 return nsec;
59}
60
61cycle_t ftrace_now(int cpu)
62{
63 return cpu_clock(cpu);
64}
65
66/*
67 * The global_trace is the descriptor that holds the tracing
68 * buffers for the live tracing. For each CPU, it contains
69 * a link list of pages that will store trace entries. The
70 * page descriptor of the pages in the memory is used to hold
71 * the link list by linking the lru item in the page descriptor
72 * to each of the pages in the buffer per CPU.
73 *
74 * For each active CPU there is a data field that holds the
75 * pages for the buffer for that CPU. Each CPU has the same number
76 * of pages allocated for its buffer.
77 */
78static struct trace_array global_trace;
79
80static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
81
82/*
83 * The max_tr is used to snapshot the global_trace when a maximum
84 * latency is reached. Some tracers will use this to store a maximum
85 * trace while it continues examining live traces.
86 *
87 * The buffers for the max_tr are set up the same as the global_trace.
88 * When a snapshot is taken, the link list of the max_tr is swapped
89 * with the link list of the global_trace and the buffers are reset for
90 * the global_trace so the tracing can continue.
91 */
92static struct trace_array max_tr;
93
94static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
95
96/* tracer_enabled is used to toggle activation of a tracer */
97static int tracer_enabled = 1;
98
99/* function tracing enabled */
100int ftrace_function_enabled;
101
102/*
103 * trace_nr_entries is the number of entries that is allocated
104 * for a buffer. Note, the number of entries is always rounded
105 * to ENTRIES_PER_PAGE.
106 */
107static unsigned long trace_nr_entries = 65536UL;
108
109/* trace_types holds a link list of available tracers. */
110static struct tracer *trace_types __read_mostly;
111
112/* current_trace points to the tracer that is currently active */
113static struct tracer *current_trace __read_mostly;
114
115/*
116 * max_tracer_type_len is used to simplify the allocating of
117 * buffers to read userspace tracer names. We keep track of
118 * the longest tracer name registered.
119 */
120static int max_tracer_type_len;
121
122/*
123 * trace_types_lock is used to protect the trace_types list.
124 * This lock is also used to keep user access serialized.
125 * Accesses from userspace will grab this lock while userspace
126 * activities happen inside the kernel.
127 */
128static DEFINE_MUTEX(trace_types_lock);
129
130/* trace_wait is a waitqueue for tasks blocked on trace_poll */
131static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
132
133/* trace_flags holds iter_ctrl options */
134unsigned long trace_flags = TRACE_ITER_PRINT_PARENT;
135
136static notrace void no_trace_init(struct trace_array *tr)
137{
138 int cpu;
139
140 ftrace_function_enabled = 0;
141 if(tr->ctrl)
142 for_each_online_cpu(cpu)
143 tracing_reset(tr->data[cpu]);
144 tracer_enabled = 0;
145}
146
147/* dummy trace to disable tracing */
148static struct tracer no_tracer __read_mostly = {
149 .name = "none",
150 .init = no_trace_init
151};
152
153
154/**
155 * trace_wake_up - wake up tasks waiting for trace input
156 *
157 * Simply wakes up any task that is blocked on the trace_wait
158 * queue. These is used with trace_poll for tasks polling the trace.
159 */
160void trace_wake_up(void)
161{
162 /*
163 * The runqueue_is_locked() can fail, but this is the best we
164 * have for now:
165 */
166 if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
167 wake_up(&trace_wait);
168}
169
170#define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
171
172static int __init set_nr_entries(char *str)
173{
174 unsigned long nr_entries;
175 int ret;
176
177 if (!str)
178 return 0;
179 ret = strict_strtoul(str, 0, &nr_entries);
180 /* nr_entries can not be zero */
181 if (ret < 0 || nr_entries == 0)
182 return 0;
183 trace_nr_entries = nr_entries;
184 return 1;
185}
186__setup("trace_entries=", set_nr_entries);
187
188unsigned long nsecs_to_usecs(unsigned long nsecs)
189{
190 return nsecs / 1000;
191}
192
193/*
194 * trace_flag_type is an enumeration that holds different
195 * states when a trace occurs. These are:
196 * IRQS_OFF - interrupts were disabled
197 * NEED_RESCED - reschedule is requested
198 * HARDIRQ - inside an interrupt handler
199 * SOFTIRQ - inside a softirq handler
200 */
201enum trace_flag_type {
202 TRACE_FLAG_IRQS_OFF = 0x01,
203 TRACE_FLAG_NEED_RESCHED = 0x02,
204 TRACE_FLAG_HARDIRQ = 0x04,
205 TRACE_FLAG_SOFTIRQ = 0x08,
206};
207
208/*
209 * TRACE_ITER_SYM_MASK masks the options in trace_flags that
210 * control the output of kernel symbols.
211 */
212#define TRACE_ITER_SYM_MASK \
213 (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
214
215/* These must match the bit postions in trace_iterator_flags */
216static const char *trace_options[] = {
217 "print-parent",
218 "sym-offset",
219 "sym-addr",
220 "verbose",
221 "raw",
222 "hex",
223 "bin",
224 "block",
225 "stacktrace",
226 "sched-tree",
227 NULL
228};
229
230/*
231 * ftrace_max_lock is used to protect the swapping of buffers
232 * when taking a max snapshot. The buffers themselves are
233 * protected by per_cpu spinlocks. But the action of the swap
234 * needs its own lock.
235 *
236 * This is defined as a raw_spinlock_t in order to help
237 * with performance when lockdep debugging is enabled.
238 */
239static raw_spinlock_t ftrace_max_lock =
240 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
241
242/*
243 * Copy the new maximum trace into the separate maximum-trace
244 * structure. (this way the maximum trace is permanently saved,
245 * for later retrieval via /debugfs/tracing/latency_trace)
246 */
247static void
248__update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
249{
250 struct trace_array_cpu *data = tr->data[cpu];
251
252 max_tr.cpu = cpu;
253 max_tr.time_start = data->preempt_timestamp;
254
255 data = max_tr.data[cpu];
256 data->saved_latency = tracing_max_latency;
257
258 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
259 data->pid = tsk->pid;
260 data->uid = tsk->uid;
261 data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
262 data->policy = tsk->policy;
263 data->rt_priority = tsk->rt_priority;
264
265 /* record this tasks comm */
266 tracing_record_cmdline(current);
267}
268
269#define CHECK_COND(cond) \
270 if (unlikely(cond)) { \
271 tracing_disabled = 1; \
272 WARN_ON(1); \
273 return -1; \
274 }
275
276/**
277 * check_pages - integrity check of trace buffers
278 *
279 * As a safty measure we check to make sure the data pages have not
280 * been corrupted.
281 */
282int check_pages(struct trace_array_cpu *data)
283{
284 struct page *page, *tmp;
285
286 CHECK_COND(data->trace_pages.next->prev != &data->trace_pages);
287 CHECK_COND(data->trace_pages.prev->next != &data->trace_pages);
288
289 list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
290 CHECK_COND(page->lru.next->prev != &page->lru);
291 CHECK_COND(page->lru.prev->next != &page->lru);
292 }
293
294 return 0;
295}
296
297/**
298 * head_page - page address of the first page in per_cpu buffer.
299 *
300 * head_page returns the page address of the first page in
301 * a per_cpu buffer. This also preforms various consistency
302 * checks to make sure the buffer has not been corrupted.
303 */
304void *head_page(struct trace_array_cpu *data)
305{
306 struct page *page;
307
308 if (list_empty(&data->trace_pages))
309 return NULL;
310
311 page = list_entry(data->trace_pages.next, struct page, lru);
312 BUG_ON(&page->lru == &data->trace_pages);
313
314 return page_address(page);
315}
316
317/**
318 * trace_seq_printf - sequence printing of trace information
319 * @s: trace sequence descriptor
320 * @fmt: printf format string
321 *
322 * The tracer may use either sequence operations or its own
323 * copy to user routines. To simplify formating of a trace
324 * trace_seq_printf is used to store strings into a special
325 * buffer (@s). Then the output may be either used by
326 * the sequencer or pulled into another buffer.
327 */
328int
329trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
330{
331 int len = (PAGE_SIZE - 1) - s->len;
332 va_list ap;
333 int ret;
334
335 if (!len)
336 return 0;
337
338 va_start(ap, fmt);
339 ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
340 va_end(ap);
341
342 /* If we can't write it all, don't bother writing anything */
343 if (ret >= len)
344 return 0;
345
346 s->len += ret;
347
348 return len;
349}
350
351/**
352 * trace_seq_puts - trace sequence printing of simple string
353 * @s: trace sequence descriptor
354 * @str: simple string to record
355 *
356 * The tracer may use either the sequence operations or its own
357 * copy to user routines. This function records a simple string
358 * into a special buffer (@s) for later retrieval by a sequencer
359 * or other mechanism.
360 */
361static int
362trace_seq_puts(struct trace_seq *s, const char *str)
363{
364 int len = strlen(str);
365
366 if (len > ((PAGE_SIZE - 1) - s->len))
367 return 0;
368
369 memcpy(s->buffer + s->len, str, len);
370 s->len += len;
371
372 return len;
373}
374
375static int
376trace_seq_putc(struct trace_seq *s, unsigned char c)
377{
378 if (s->len >= (PAGE_SIZE - 1))
379 return 0;
380
381 s->buffer[s->len++] = c;
382
383 return 1;
384}
385
386static int
387trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
388{
389 if (len > ((PAGE_SIZE - 1) - s->len))
390 return 0;
391
392 memcpy(s->buffer + s->len, mem, len);
393 s->len += len;
394
395 return len;
396}
397
398#define HEX_CHARS 17
399static const char hex2asc[] = "0123456789abcdef";
400
401static int
402trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
403{
404 unsigned char hex[HEX_CHARS];
405 unsigned char *data = mem;
406 unsigned char byte;
407 int i, j;
408
409 BUG_ON(len >= HEX_CHARS);
410
411#ifdef __BIG_ENDIAN
412 for (i = 0, j = 0; i < len; i++) {
413#else
414 for (i = len-1, j = 0; i >= 0; i--) {
415#endif
416 byte = data[i];
417
418 hex[j++] = hex2asc[byte & 0x0f];
419 hex[j++] = hex2asc[byte >> 4];
420 }
421 hex[j++] = ' ';
422
423 return trace_seq_putmem(s, hex, j);
424}
425
426static void
427trace_seq_reset(struct trace_seq *s)
428{
429 s->len = 0;
430 s->readpos = 0;
431}
432
433ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
434{
435 int len;
436 int ret;
437
438 if (s->len <= s->readpos)
439 return -EBUSY;
440
441 len = s->len - s->readpos;
442 if (cnt > len)
443 cnt = len;
444 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
445 if (ret)
446 return -EFAULT;
447
448 s->readpos += len;
449 return cnt;
450}
451
452static void
453trace_print_seq(struct seq_file *m, struct trace_seq *s)
454{
455 int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
456
457 s->buffer[len] = 0;
458 seq_puts(m, s->buffer);
459
460 trace_seq_reset(s);
461}
462
463/*
464 * flip the trace buffers between two trace descriptors.
465 * This usually is the buffers between the global_trace and
466 * the max_tr to record a snapshot of a current trace.
467 *
468 * The ftrace_max_lock must be held.
469 */
470static void
471flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
472{
473 struct list_head flip_pages;
474
475 INIT_LIST_HEAD(&flip_pages);
476
477 memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
478 sizeof(struct trace_array_cpu) -
479 offsetof(struct trace_array_cpu, trace_head_idx));
480
481 check_pages(tr1);
482 check_pages(tr2);
483 list_splice_init(&tr1->trace_pages, &flip_pages);
484 list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
485 list_splice_init(&flip_pages, &tr2->trace_pages);
486 BUG_ON(!list_empty(&flip_pages));
487 check_pages(tr1);
488 check_pages(tr2);
489}
490
491/**
492 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
493 * @tr: tracer
494 * @tsk: the task with the latency
495 * @cpu: The cpu that initiated the trace.
496 *
497 * Flip the buffers between the @tr and the max_tr and record information
498 * about which task was the cause of this latency.
499 */
500void
501update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
502{
503 struct trace_array_cpu *data;
504 int i;
505
506 WARN_ON_ONCE(!irqs_disabled());
507 __raw_spin_lock(&ftrace_max_lock);
508 /* clear out all the previous traces */
509 for_each_tracing_cpu(i) {
510 data = tr->data[i];
511 flip_trace(max_tr.data[i], data);
512 tracing_reset(data);
513 }
514
515 __update_max_tr(tr, tsk, cpu);
516 __raw_spin_unlock(&ftrace_max_lock);
517}
518
519/**
520 * update_max_tr_single - only copy one trace over, and reset the rest
521 * @tr - tracer
522 * @tsk - task with the latency
523 * @cpu - the cpu of the buffer to copy.
524 *
525 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
526 */
527void
528update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
529{
530 struct trace_array_cpu *data = tr->data[cpu];
531 int i;
532
533 WARN_ON_ONCE(!irqs_disabled());
534 __raw_spin_lock(&ftrace_max_lock);
535 for_each_tracing_cpu(i)
536 tracing_reset(max_tr.data[i]);
537
538 flip_trace(max_tr.data[cpu], data);
539 tracing_reset(data);
540
541 __update_max_tr(tr, tsk, cpu);
542 __raw_spin_unlock(&ftrace_max_lock);
543}
544
545/**
546 * register_tracer - register a tracer with the ftrace system.
547 * @type - the plugin for the tracer
548 *
549 * Register a new plugin tracer.
550 */
551int register_tracer(struct tracer *type)
552{
553 struct tracer *t;
554 int len;
555 int ret = 0;
556
557 if (!type->name) {
558 pr_info("Tracer must have a name\n");
559 return -1;
560 }
561
562 mutex_lock(&trace_types_lock);
563 for (t = trace_types; t; t = t->next) {
564 if (strcmp(type->name, t->name) == 0) {
565 /* already found */
566 pr_info("Trace %s already registered\n",
567 type->name);
568 ret = -1;
569 goto out;
570 }
571 }
572
573#ifdef CONFIG_FTRACE_STARTUP_TEST
574 if (type->selftest) {
575 struct tracer *saved_tracer = current_trace;
576 struct trace_array_cpu *data;
577 struct trace_array *tr = &global_trace;
578 int saved_ctrl = tr->ctrl;
579 int i;
580 /*
581 * Run a selftest on this tracer.
582 * Here we reset the trace buffer, and set the current
583 * tracer to be this tracer. The tracer can then run some
584 * internal tracing to verify that everything is in order.
585 * If we fail, we do not register this tracer.
586 */
587 for_each_tracing_cpu(i) {
588 data = tr->data[i];
589 if (!head_page(data))
590 continue;
591 tracing_reset(data);
592 }
593 current_trace = type;
594 tr->ctrl = 0;
595 /* the test is responsible for initializing and enabling */
596 pr_info("Testing tracer %s: ", type->name);
597 ret = type->selftest(type, tr);
598 /* the test is responsible for resetting too */
599 current_trace = saved_tracer;
600 tr->ctrl = saved_ctrl;
601 if (ret) {
602 printk(KERN_CONT "FAILED!\n");
603 goto out;
604 }
605 /* Only reset on passing, to avoid touching corrupted buffers */
606 for_each_tracing_cpu(i) {
607 data = tr->data[i];
608 if (!head_page(data))
609 continue;
610 tracing_reset(data);
611 }
612 printk(KERN_CONT "PASSED\n");
613 }
614#endif
615
616 type->next = trace_types;
617 trace_types = type;
618 len = strlen(type->name);
619 if (len > max_tracer_type_len)
620 max_tracer_type_len = len;
621
622 out:
623 mutex_unlock(&trace_types_lock);
624
625 return ret;
626}
627
628void unregister_tracer(struct tracer *type)
629{
630 struct tracer **t;
631 int len;
632
633 mutex_lock(&trace_types_lock);
634 for (t = &trace_types; *t; t = &(*t)->next) {
635 if (*t == type)
636 goto found;
637 }
638 pr_info("Trace %s not registered\n", type->name);
639 goto out;
640
641 found:
642 *t = (*t)->next;
643 if (strlen(type->name) != max_tracer_type_len)
644 goto out;
645
646 max_tracer_type_len = 0;
647 for (t = &trace_types; *t; t = &(*t)->next) {
648 len = strlen((*t)->name);
649 if (len > max_tracer_type_len)
650 max_tracer_type_len = len;
651 }
652 out:
653 mutex_unlock(&trace_types_lock);
654}
655
656void tracing_reset(struct trace_array_cpu *data)
657{
658 data->trace_idx = 0;
659 data->overrun = 0;
660 data->trace_head = data->trace_tail = head_page(data);
661 data->trace_head_idx = 0;
662 data->trace_tail_idx = 0;
663}
664
665#define SAVED_CMDLINES 128
666static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
667static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
668static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
669static int cmdline_idx;
670static DEFINE_SPINLOCK(trace_cmdline_lock);
671
672/* temporary disable recording */
673atomic_t trace_record_cmdline_disabled __read_mostly;
674
675static void trace_init_cmdlines(void)
676{
677 memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
678 memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
679 cmdline_idx = 0;
680}
681
682void trace_stop_cmdline_recording(void);
683
684static void trace_save_cmdline(struct task_struct *tsk)
685{
686 unsigned map;
687 unsigned idx;
688
689 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
690 return;
691
692 /*
693 * It's not the end of the world if we don't get
694 * the lock, but we also don't want to spin
695 * nor do we want to disable interrupts,
696 * so if we miss here, then better luck next time.
697 */
698 if (!spin_trylock(&trace_cmdline_lock))
699 return;
700
701 idx = map_pid_to_cmdline[tsk->pid];
702 if (idx >= SAVED_CMDLINES) {
703 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
704
705 map = map_cmdline_to_pid[idx];
706 if (map <= PID_MAX_DEFAULT)
707 map_pid_to_cmdline[map] = (unsigned)-1;
708
709 map_pid_to_cmdline[tsk->pid] = idx;
710
711 cmdline_idx = idx;
712 }
713
714 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
715
716 spin_unlock(&trace_cmdline_lock);
717}
718
719static char *trace_find_cmdline(int pid)
720{
721 char *cmdline = "<...>";
722 unsigned map;
723
724 if (!pid)
725 return "<idle>";
726
727 if (pid > PID_MAX_DEFAULT)
728 goto out;
729
730 map = map_pid_to_cmdline[pid];
731 if (map >= SAVED_CMDLINES)
732 goto out;
733
734 cmdline = saved_cmdlines[map];
735
736 out:
737 return cmdline;
738}
739
740void tracing_record_cmdline(struct task_struct *tsk)
741{
742 if (atomic_read(&trace_record_cmdline_disabled))
743 return;
744
745 trace_save_cmdline(tsk);
746}
747
748static inline struct list_head *
749trace_next_list(struct trace_array_cpu *data, struct list_head *next)
750{
751 /*
752 * Roundrobin - but skip the head (which is not a real page):
753 */
754 next = next->next;
755 if (unlikely(next == &data->trace_pages))
756 next = next->next;
757 BUG_ON(next == &data->trace_pages);
758
759 return next;
760}
761
762static inline void *
763trace_next_page(struct trace_array_cpu *data, void *addr)
764{
765 struct list_head *next;
766 struct page *page;
767
768 page = virt_to_page(addr);
769
770 next = trace_next_list(data, &page->lru);
771 page = list_entry(next, struct page, lru);
772
773 return page_address(page);
774}
775
776static inline struct trace_entry *
777tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
778{
779 unsigned long idx, idx_next;
780 struct trace_entry *entry;
781
782 data->trace_idx++;
783 idx = data->trace_head_idx;
784 idx_next = idx + 1;
785
786 BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
787
788 entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
789
790 if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
791 data->trace_head = trace_next_page(data, data->trace_head);
792 idx_next = 0;
793 }
794
795 if (data->trace_head == data->trace_tail &&
796 idx_next == data->trace_tail_idx) {
797 /* overrun */
798 data->overrun++;
799 data->trace_tail_idx++;
800 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
801 data->trace_tail =
802 trace_next_page(data, data->trace_tail);
803 data->trace_tail_idx = 0;
804 }
805 }
806
807 data->trace_head_idx = idx_next;
808
809 return entry;
810}
811
812static inline void
813tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
814{
815 struct task_struct *tsk = current;
816 unsigned long pc;
817
818 pc = preempt_count();
819
820 entry->preempt_count = pc & 0xff;
821 entry->pid = (tsk) ? tsk->pid : 0;
822 entry->t = ftrace_now(raw_smp_processor_id());
823 entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
824 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
825 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
826 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
827}
828
829void
830trace_function(struct trace_array *tr, struct trace_array_cpu *data,
831 unsigned long ip, unsigned long parent_ip, unsigned long flags)
832{
833 struct trace_entry *entry;
834 unsigned long irq_flags;
835
836 raw_local_irq_save(irq_flags);
837 __raw_spin_lock(&data->lock);
838 entry = tracing_get_trace_entry(tr, data);
839 tracing_generic_entry_update(entry, flags);
840 entry->type = TRACE_FN;
841 entry->fn.ip = ip;
842 entry->fn.parent_ip = parent_ip;
843 __raw_spin_unlock(&data->lock);
844 raw_local_irq_restore(irq_flags);
845}
846
847void
848ftrace(struct trace_array *tr, struct trace_array_cpu *data,
849 unsigned long ip, unsigned long parent_ip, unsigned long flags)
850{
851 if (likely(!atomic_read(&data->disabled)))
852 trace_function(tr, data, ip, parent_ip, flags);
853}
854
855#ifdef CONFIG_MMIOTRACE
856void __trace_mmiotrace_rw(struct trace_array *tr, struct trace_array_cpu *data,
857 struct mmiotrace_rw *rw)
858{
859 struct trace_entry *entry;
860 unsigned long irq_flags;
861
862 raw_local_irq_save(irq_flags);
863 __raw_spin_lock(&data->lock);
864
865 entry = tracing_get_trace_entry(tr, data);
866 tracing_generic_entry_update(entry, 0);
867 entry->type = TRACE_MMIO_RW;
868 entry->mmiorw = *rw;
869
870 __raw_spin_unlock(&data->lock);
871 raw_local_irq_restore(irq_flags);
872
873 trace_wake_up();
874}
875
876void __trace_mmiotrace_map(struct trace_array *tr, struct trace_array_cpu *data,
877 struct mmiotrace_map *map)
878{
879 struct trace_entry *entry;
880 unsigned long irq_flags;
881
882 raw_local_irq_save(irq_flags);
883 __raw_spin_lock(&data->lock);
884
885 entry = tracing_get_trace_entry(tr, data);
886 tracing_generic_entry_update(entry, 0);
887 entry->type = TRACE_MMIO_MAP;
888 entry->mmiomap = *map;
889
890 __raw_spin_unlock(&data->lock);
891 raw_local_irq_restore(irq_flags);
892
893 trace_wake_up();
894}
895#endif
896
897void __trace_stack(struct trace_array *tr,
898 struct trace_array_cpu *data,
899 unsigned long flags,
900 int skip)
901{
902 struct trace_entry *entry;
903 struct stack_trace trace;
904
905 if (!(trace_flags & TRACE_ITER_STACKTRACE))
906 return;
907
908 entry = tracing_get_trace_entry(tr, data);
909 tracing_generic_entry_update(entry, flags);
910 entry->type = TRACE_STACK;
911
912 memset(&entry->stack, 0, sizeof(entry->stack));
913
914 trace.nr_entries = 0;
915 trace.max_entries = FTRACE_STACK_ENTRIES;
916 trace.skip = skip;
917 trace.entries = entry->stack.caller;
918
919 save_stack_trace(&trace);
920}
921
922void
923__trace_special(void *__tr, void *__data,
924 unsigned long arg1, unsigned long arg2, unsigned long arg3)
925{
926 struct trace_array_cpu *data = __data;
927 struct trace_array *tr = __tr;
928 struct trace_entry *entry;
929 unsigned long irq_flags;
930
931 raw_local_irq_save(irq_flags);
932 __raw_spin_lock(&data->lock);
933 entry = tracing_get_trace_entry(tr, data);
934 tracing_generic_entry_update(entry, 0);
935 entry->type = TRACE_SPECIAL;
936 entry->special.arg1 = arg1;
937 entry->special.arg2 = arg2;
938 entry->special.arg3 = arg3;
939 __trace_stack(tr, data, irq_flags, 4);
940 __raw_spin_unlock(&data->lock);
941 raw_local_irq_restore(irq_flags);
942
943 trace_wake_up();
944}
945
946void
947tracing_sched_switch_trace(struct trace_array *tr,
948 struct trace_array_cpu *data,
949 struct task_struct *prev,
950 struct task_struct *next,
951 unsigned long flags)
952{
953 struct trace_entry *entry;
954 unsigned long irq_flags;
955
956 raw_local_irq_save(irq_flags);
957 __raw_spin_lock(&data->lock);
958 entry = tracing_get_trace_entry(tr, data);
959 tracing_generic_entry_update(entry, flags);
960 entry->type = TRACE_CTX;
961 entry->ctx.prev_pid = prev->pid;
962 entry->ctx.prev_prio = prev->prio;
963 entry->ctx.prev_state = prev->state;
964 entry->ctx.next_pid = next->pid;
965 entry->ctx.next_prio = next->prio;
966 entry->ctx.next_state = next->state;
967 __trace_stack(tr, data, flags, 5);
968 __raw_spin_unlock(&data->lock);
969 raw_local_irq_restore(irq_flags);
970}
971
972void
973tracing_sched_wakeup_trace(struct trace_array *tr,
974 struct trace_array_cpu *data,
975 struct task_struct *wakee,
976 struct task_struct *curr,
977 unsigned long flags)
978{
979 struct trace_entry *entry;
980 unsigned long irq_flags;
981
982 raw_local_irq_save(irq_flags);
983 __raw_spin_lock(&data->lock);
984 entry = tracing_get_trace_entry(tr, data);
985 tracing_generic_entry_update(entry, flags);
986 entry->type = TRACE_WAKE;
987 entry->ctx.prev_pid = curr->pid;
988 entry->ctx.prev_prio = curr->prio;
989 entry->ctx.prev_state = curr->state;
990 entry->ctx.next_pid = wakee->pid;
991 entry->ctx.next_prio = wakee->prio;
992 entry->ctx.next_state = wakee->state;
993 __trace_stack(tr, data, flags, 6);
994 __raw_spin_unlock(&data->lock);
995 raw_local_irq_restore(irq_flags);
996
997 trace_wake_up();
998}
999
1000void
1001ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1002{
1003 struct trace_array *tr = &global_trace;
1004 struct trace_array_cpu *data;
1005 unsigned long flags;
1006 long disabled;
1007 int cpu;
1008
1009 if (tracing_disabled || current_trace == &no_tracer || !tr->ctrl)
1010 return;
1011
1012 local_irq_save(flags);
1013 cpu = raw_smp_processor_id();
1014 data = tr->data[cpu];
1015 disabled = atomic_inc_return(&data->disabled);
1016
1017 if (likely(disabled == 1))
1018 __trace_special(tr, data, arg1, arg2, arg3);
1019
1020 atomic_dec(&data->disabled);
1021 local_irq_restore(flags);
1022}
1023
1024#ifdef CONFIG_FTRACE
1025static void
1026function_trace_call(unsigned long ip, unsigned long parent_ip)
1027{
1028 struct trace_array *tr = &global_trace;
1029 struct trace_array_cpu *data;
1030 unsigned long flags;
1031 long disabled;
1032 int cpu;
1033
1034 if (unlikely(!ftrace_function_enabled))
1035 return;
1036
1037 if (skip_trace(ip))
1038 return;
1039
1040 local_irq_save(flags);
1041 cpu = raw_smp_processor_id();
1042 data = tr->data[cpu];
1043 disabled = atomic_inc_return(&data->disabled);
1044
1045 if (likely(disabled == 1))
1046 trace_function(tr, data, ip, parent_ip, flags);
1047
1048 atomic_dec(&data->disabled);
1049 local_irq_restore(flags);
1050}
1051
1052static struct ftrace_ops trace_ops __read_mostly =
1053{
1054 .func = function_trace_call,
1055};
1056
1057void tracing_start_function_trace(void)
1058{
1059 ftrace_function_enabled = 0;
1060 register_ftrace_function(&trace_ops);
1061 if (tracer_enabled)
1062 ftrace_function_enabled = 1;
1063}
1064
1065void tracing_stop_function_trace(void)
1066{
1067 ftrace_function_enabled = 0;
1068 unregister_ftrace_function(&trace_ops);
1069}
1070#endif
1071
1072enum trace_file_type {
1073 TRACE_FILE_LAT_FMT = 1,
1074};
1075
1076static struct trace_entry *
1077trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
1078 struct trace_iterator *iter, int cpu)
1079{
1080 struct page *page;
1081 struct trace_entry *array;
1082
1083 if (iter->next_idx[cpu] >= tr->entries ||
1084 iter->next_idx[cpu] >= data->trace_idx ||
1085 (data->trace_head == data->trace_tail &&
1086 data->trace_head_idx == data->trace_tail_idx))
1087 return NULL;
1088
1089 if (!iter->next_page[cpu]) {
1090 /* Initialize the iterator for this cpu trace buffer */
1091 WARN_ON(!data->trace_tail);
1092 page = virt_to_page(data->trace_tail);
1093 iter->next_page[cpu] = &page->lru;
1094 iter->next_page_idx[cpu] = data->trace_tail_idx;
1095 }
1096
1097 page = list_entry(iter->next_page[cpu], struct page, lru);
1098 BUG_ON(&data->trace_pages == &page->lru);
1099
1100 array = page_address(page);
1101
1102 WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
1103 return &array[iter->next_page_idx[cpu]];
1104}
1105
1106static struct trace_entry *
1107find_next_entry(struct trace_iterator *iter, int *ent_cpu)
1108{
1109 struct trace_array *tr = iter->tr;
1110 struct trace_entry *ent, *next = NULL;
1111 int next_cpu = -1;
1112 int cpu;
1113
1114 for_each_tracing_cpu(cpu) {
1115 if (!head_page(tr->data[cpu]))
1116 continue;
1117 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
1118 /*
1119 * Pick the entry with the smallest timestamp:
1120 */
1121 if (ent && (!next || ent->t < next->t)) {
1122 next = ent;
1123 next_cpu = cpu;
1124 }
1125 }
1126
1127 if (ent_cpu)
1128 *ent_cpu = next_cpu;
1129
1130 return next;
1131}
1132
1133static void trace_iterator_increment(struct trace_iterator *iter)
1134{
1135 iter->idx++;
1136 iter->next_idx[iter->cpu]++;
1137 iter->next_page_idx[iter->cpu]++;
1138
1139 if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
1140 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
1141
1142 iter->next_page_idx[iter->cpu] = 0;
1143 iter->next_page[iter->cpu] =
1144 trace_next_list(data, iter->next_page[iter->cpu]);
1145 }
1146}
1147
1148static void trace_consume(struct trace_iterator *iter)
1149{
1150 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
1151
1152 data->trace_tail_idx++;
1153 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
1154 data->trace_tail = trace_next_page(data, data->trace_tail);
1155 data->trace_tail_idx = 0;
1156 }
1157
1158 /* Check if we empty it, then reset the index */
1159 if (data->trace_head == data->trace_tail &&
1160 data->trace_head_idx == data->trace_tail_idx)
1161 data->trace_idx = 0;
1162}
1163
1164static void *find_next_entry_inc(struct trace_iterator *iter)
1165{
1166 struct trace_entry *next;
1167 int next_cpu = -1;
1168
1169 next = find_next_entry(iter, &next_cpu);
1170
1171 iter->prev_ent = iter->ent;
1172 iter->prev_cpu = iter->cpu;
1173
1174 iter->ent = next;
1175 iter->cpu = next_cpu;
1176
1177 if (next)
1178 trace_iterator_increment(iter);
1179
1180 return next ? iter : NULL;
1181}
1182
1183static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1184{
1185 struct trace_iterator *iter = m->private;
1186 void *last_ent = iter->ent;
1187 int i = (int)*pos;
1188 void *ent;
1189
1190 (*pos)++;
1191
1192 /* can't go backwards */
1193 if (iter->idx > i)
1194 return NULL;
1195
1196 if (iter->idx < 0)
1197 ent = find_next_entry_inc(iter);
1198 else
1199 ent = iter;
1200
1201 while (ent && iter->idx < i)
1202 ent = find_next_entry_inc(iter);
1203
1204 iter->pos = *pos;
1205
1206 if (last_ent && !ent)
1207 seq_puts(m, "\n\nvim:ft=help\n");
1208
1209 return ent;
1210}
1211
1212static void *s_start(struct seq_file *m, loff_t *pos)
1213{
1214 struct trace_iterator *iter = m->private;
1215 void *p = NULL;
1216 loff_t l = 0;
1217 int i;
1218
1219 mutex_lock(&trace_types_lock);
1220
1221 if (!current_trace || current_trace != iter->trace) {
1222 mutex_unlock(&trace_types_lock);
1223 return NULL;
1224 }
1225
1226 atomic_inc(&trace_record_cmdline_disabled);
1227
1228 /* let the tracer grab locks here if needed */
1229 if (current_trace->start)
1230 current_trace->start(iter);
1231
1232 if (*pos != iter->pos) {
1233 iter->ent = NULL;
1234 iter->cpu = 0;
1235 iter->idx = -1;
1236 iter->prev_ent = NULL;
1237 iter->prev_cpu = -1;
1238
1239 for_each_tracing_cpu(i) {
1240 iter->next_idx[i] = 0;
1241 iter->next_page[i] = NULL;
1242 }
1243
1244 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1245 ;
1246
1247 } else {
1248 l = *pos - 1;
1249 p = s_next(m, p, &l);
1250 }
1251
1252 return p;
1253}
1254
1255static void s_stop(struct seq_file *m, void *p)
1256{
1257 struct trace_iterator *iter = m->private;
1258
1259 atomic_dec(&trace_record_cmdline_disabled);
1260
1261 /* let the tracer release locks here if needed */
1262 if (current_trace && current_trace == iter->trace && iter->trace->stop)
1263 iter->trace->stop(iter);
1264
1265 mutex_unlock(&trace_types_lock);
1266}
1267
1268#define KRETPROBE_MSG "[unknown/kretprobe'd]"
1269
1270#ifdef CONFIG_KRETPROBES
1271static inline int kretprobed(unsigned long addr)
1272{
1273 return addr == (unsigned long)kretprobe_trampoline;
1274}
1275#else
1276static inline int kretprobed(unsigned long addr)
1277{
1278 return 0;
1279}
1280#endif /* CONFIG_KRETPROBES */
1281
1282static int
1283seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1284{
1285#ifdef CONFIG_KALLSYMS
1286 char str[KSYM_SYMBOL_LEN];
1287
1288 kallsyms_lookup(address, NULL, NULL, NULL, str);
1289
1290 return trace_seq_printf(s, fmt, str);
1291#endif
1292 return 1;
1293}
1294
1295static int
1296seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1297 unsigned long address)
1298{
1299#ifdef CONFIG_KALLSYMS
1300 char str[KSYM_SYMBOL_LEN];
1301
1302 sprint_symbol(str, address);
1303 return trace_seq_printf(s, fmt, str);
1304#endif
1305 return 1;
1306}
1307
1308#ifndef CONFIG_64BIT
1309# define IP_FMT "%08lx"
1310#else
1311# define IP_FMT "%016lx"
1312#endif
1313
1314static int
1315seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1316{
1317 int ret;
1318
1319 if (!ip)
1320 return trace_seq_printf(s, "0");
1321
1322 if (sym_flags & TRACE_ITER_SYM_OFFSET)
1323 ret = seq_print_sym_offset(s, "%s", ip);
1324 else
1325 ret = seq_print_sym_short(s, "%s", ip);
1326
1327 if (!ret)
1328 return 0;
1329
1330 if (sym_flags & TRACE_ITER_SYM_ADDR)
1331 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1332 return ret;
1333}
1334
1335static void print_lat_help_header(struct seq_file *m)
1336{
1337 seq_puts(m, "# _------=> CPU# \n");
1338 seq_puts(m, "# / _-----=> irqs-off \n");
1339 seq_puts(m, "# | / _----=> need-resched \n");
1340 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1341 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1342 seq_puts(m, "# |||| / \n");
1343 seq_puts(m, "# ||||| delay \n");
1344 seq_puts(m, "# cmd pid ||||| time | caller \n");
1345 seq_puts(m, "# \\ / ||||| \\ | / \n");
1346}
1347
1348static void print_func_help_header(struct seq_file *m)
1349{
1350 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1351 seq_puts(m, "# | | | | |\n");
1352}
1353
1354
1355static void
1356print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1357{
1358 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1359 struct trace_array *tr = iter->tr;
1360 struct trace_array_cpu *data = tr->data[tr->cpu];
1361 struct tracer *type = current_trace;
1362 unsigned long total = 0;
1363 unsigned long entries = 0;
1364 int cpu;
1365 const char *name = "preemption";
1366
1367 if (type)
1368 name = type->name;
1369
1370 for_each_tracing_cpu(cpu) {
1371 if (head_page(tr->data[cpu])) {
1372 total += tr->data[cpu]->trace_idx;
1373 if (tr->data[cpu]->trace_idx > tr->entries)
1374 entries += tr->entries;
1375 else
1376 entries += tr->data[cpu]->trace_idx;
1377 }
1378 }
1379
1380 seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1381 name, UTS_RELEASE);
1382 seq_puts(m, "-----------------------------------"
1383 "---------------------------------\n");
1384 seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1385 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1386 nsecs_to_usecs(data->saved_latency),
1387 entries,
1388 total,
1389 tr->cpu,
1390#if defined(CONFIG_PREEMPT_NONE)
1391 "server",
1392#elif defined(CONFIG_PREEMPT_VOLUNTARY)
1393 "desktop",
1394#elif defined(CONFIG_PREEMPT)
1395 "preempt",
1396#else
1397 "unknown",
1398#endif
1399 /* These are reserved for later use */
1400 0, 0, 0, 0);
1401#ifdef CONFIG_SMP
1402 seq_printf(m, " #P:%d)\n", num_online_cpus());
1403#else
1404 seq_puts(m, ")\n");
1405#endif
1406 seq_puts(m, " -----------------\n");
1407 seq_printf(m, " | task: %.16s-%d "
1408 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1409 data->comm, data->pid, data->uid, data->nice,
1410 data->policy, data->rt_priority);
1411 seq_puts(m, " -----------------\n");
1412
1413 if (data->critical_start) {
1414 seq_puts(m, " => started at: ");
1415 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1416 trace_print_seq(m, &iter->seq);
1417 seq_puts(m, "\n => ended at: ");
1418 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1419 trace_print_seq(m, &iter->seq);
1420 seq_puts(m, "\n");
1421 }
1422
1423 seq_puts(m, "\n");
1424}
1425
1426static void
1427lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1428{
1429 int hardirq, softirq;
1430 char *comm;
1431
1432 comm = trace_find_cmdline(entry->pid);
1433
1434 trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1435 trace_seq_printf(s, "%d", cpu);
1436 trace_seq_printf(s, "%c%c",
1437 (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1438 ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1439
1440 hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1441 softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1442 if (hardirq && softirq) {
1443 trace_seq_putc(s, 'H');
1444 } else {
1445 if (hardirq) {
1446 trace_seq_putc(s, 'h');
1447 } else {
1448 if (softirq)
1449 trace_seq_putc(s, 's');
1450 else
1451 trace_seq_putc(s, '.');
1452 }
1453 }
1454
1455 if (entry->preempt_count)
1456 trace_seq_printf(s, "%x", entry->preempt_count);
1457 else
1458 trace_seq_puts(s, ".");
1459}
1460
1461unsigned long preempt_mark_thresh = 100;
1462
1463static void
1464lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1465 unsigned long rel_usecs)
1466{
1467 trace_seq_printf(s, " %4lldus", abs_usecs);
1468 if (rel_usecs > preempt_mark_thresh)
1469 trace_seq_puts(s, "!: ");
1470 else if (rel_usecs > 1)
1471 trace_seq_puts(s, "+: ");
1472 else
1473 trace_seq_puts(s, " : ");
1474}
1475
1476static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1477
1478static int
1479print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1480{
1481 struct trace_seq *s = &iter->seq;
1482 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1483 struct trace_entry *next_entry = find_next_entry(iter, NULL);
1484 unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1485 struct trace_entry *entry = iter->ent;
1486 unsigned long abs_usecs;
1487 unsigned long rel_usecs;
1488 char *comm;
1489 int S, T;
1490 int i;
1491 unsigned state;
1492
1493 if (!next_entry)
1494 next_entry = entry;
1495 rel_usecs = ns2usecs(next_entry->t - entry->t);
1496 abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1497
1498 if (verbose) {
1499 comm = trace_find_cmdline(entry->pid);
1500 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1501 " %ld.%03ldms (+%ld.%03ldms): ",
1502 comm,
1503 entry->pid, cpu, entry->flags,
1504 entry->preempt_count, trace_idx,
1505 ns2usecs(entry->t),
1506 abs_usecs/1000,
1507 abs_usecs % 1000, rel_usecs/1000,
1508 rel_usecs % 1000);
1509 } else {
1510 lat_print_generic(s, entry, cpu);
1511 lat_print_timestamp(s, abs_usecs, rel_usecs);
1512 }
1513 switch (entry->type) {
1514 case TRACE_FN:
1515 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1516 trace_seq_puts(s, " (");
1517 if (kretprobed(entry->fn.parent_ip))
1518 trace_seq_puts(s, KRETPROBE_MSG);
1519 else
1520 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1521 trace_seq_puts(s, ")\n");
1522 break;
1523 case TRACE_CTX:
1524 case TRACE_WAKE:
1525 T = entry->ctx.next_state < sizeof(state_to_char) ?
1526 state_to_char[entry->ctx.next_state] : 'X';
1527
1528 state = entry->ctx.prev_state ? __ffs(entry->ctx.prev_state) + 1 : 0;
1529 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1530 comm = trace_find_cmdline(entry->ctx.next_pid);
1531 trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c %s\n",
1532 entry->ctx.prev_pid,
1533 entry->ctx.prev_prio,
1534 S, entry->type == TRACE_CTX ? "==>" : " +",
1535 entry->ctx.next_pid,
1536 entry->ctx.next_prio,
1537 T, comm);
1538 break;
1539 case TRACE_SPECIAL:
1540 trace_seq_printf(s, "# %ld %ld %ld\n",
1541 entry->special.arg1,
1542 entry->special.arg2,
1543 entry->special.arg3);
1544 break;
1545 case TRACE_STACK:
1546 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1547 if (i)
1548 trace_seq_puts(s, " <= ");
1549 seq_print_ip_sym(s, entry->stack.caller[i], sym_flags);
1550 }
1551 trace_seq_puts(s, "\n");
1552 break;
1553 default:
1554 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1555 }
1556 return 1;
1557}
1558
1559static int print_trace_fmt(struct trace_iterator *iter)
1560{
1561 struct trace_seq *s = &iter->seq;
1562 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1563 struct trace_entry *entry;
1564 unsigned long usec_rem;
1565 unsigned long long t;
1566 unsigned long secs;
1567 char *comm;
1568 int ret;
1569 int S, T;
1570 int i;
1571
1572 entry = iter->ent;
1573
1574 comm = trace_find_cmdline(iter->ent->pid);
1575
1576 t = ns2usecs(entry->t);
1577 usec_rem = do_div(t, 1000000ULL);
1578 secs = (unsigned long)t;
1579
1580 ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1581 if (!ret)
1582 return 0;
1583 ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1584 if (!ret)
1585 return 0;
1586 ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1587 if (!ret)
1588 return 0;
1589
1590 switch (entry->type) {
1591 case TRACE_FN:
1592 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1593 if (!ret)
1594 return 0;
1595 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1596 entry->fn.parent_ip) {
1597 ret = trace_seq_printf(s, " <-");
1598 if (!ret)
1599 return 0;
1600 if (kretprobed(entry->fn.parent_ip))
1601 ret = trace_seq_puts(s, KRETPROBE_MSG);
1602 else
1603 ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1604 sym_flags);
1605 if (!ret)
1606 return 0;
1607 }
1608 ret = trace_seq_printf(s, "\n");
1609 if (!ret)
1610 return 0;
1611 break;
1612 case TRACE_CTX:
1613 case TRACE_WAKE:
1614 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1615 state_to_char[entry->ctx.prev_state] : 'X';
1616 T = entry->ctx.next_state < sizeof(state_to_char) ?
1617 state_to_char[entry->ctx.next_state] : 'X';
1618 ret = trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c\n",
1619 entry->ctx.prev_pid,
1620 entry->ctx.prev_prio,
1621 S,
1622 entry->type == TRACE_CTX ? "==>" : " +",
1623 entry->ctx.next_pid,
1624 entry->ctx.next_prio,
1625 T);
1626 if (!ret)
1627 return 0;
1628 break;
1629 case TRACE_SPECIAL:
1630 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1631 entry->special.arg1,
1632 entry->special.arg2,
1633 entry->special.arg3);
1634 if (!ret)
1635 return 0;
1636 break;
1637 case TRACE_STACK:
1638 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1639 if (i) {
1640 ret = trace_seq_puts(s, " <= ");
1641 if (!ret)
1642 return 0;
1643 }
1644 ret = seq_print_ip_sym(s, entry->stack.caller[i],
1645 sym_flags);
1646 if (!ret)
1647 return 0;
1648 }
1649 ret = trace_seq_puts(s, "\n");
1650 if (!ret)
1651 return 0;
1652 break;
1653 }
1654 return 1;
1655}
1656
1657static int print_raw_fmt(struct trace_iterator *iter)
1658{
1659 struct trace_seq *s = &iter->seq;
1660 struct trace_entry *entry;
1661 int ret;
1662 int S, T;
1663
1664 entry = iter->ent;
1665
1666 ret = trace_seq_printf(s, "%d %d %llu ",
1667 entry->pid, iter->cpu, entry->t);
1668 if (!ret)
1669 return 0;
1670
1671 switch (entry->type) {
1672 case TRACE_FN:
1673 ret = trace_seq_printf(s, "%x %x\n",
1674 entry->fn.ip, entry->fn.parent_ip);
1675 if (!ret)
1676 return 0;
1677 break;
1678 case TRACE_CTX:
1679 case TRACE_WAKE:
1680 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1681 state_to_char[entry->ctx.prev_state] : 'X';
1682 T = entry->ctx.next_state < sizeof(state_to_char) ?
1683 state_to_char[entry->ctx.next_state] : 'X';
1684 if (entry->type == TRACE_WAKE)
1685 S = '+';
1686 ret = trace_seq_printf(s, "%d %d %c %d %d %c\n",
1687 entry->ctx.prev_pid,
1688 entry->ctx.prev_prio,
1689 S,
1690 entry->ctx.next_pid,
1691 entry->ctx.next_prio,
1692 T);
1693 if (!ret)
1694 return 0;
1695 break;
1696 case TRACE_SPECIAL:
1697 case TRACE_STACK:
1698 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1699 entry->special.arg1,
1700 entry->special.arg2,
1701 entry->special.arg3);
1702 if (!ret)
1703 return 0;
1704 break;
1705 }
1706 return 1;
1707}
1708
1709#define SEQ_PUT_FIELD_RET(s, x) \
1710do { \
1711 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
1712 return 0; \
1713} while (0)
1714
1715#define SEQ_PUT_HEX_FIELD_RET(s, x) \
1716do { \
1717 if (!trace_seq_putmem_hex(s, &(x), sizeof(x))) \
1718 return 0; \
1719} while (0)
1720
1721static int print_hex_fmt(struct trace_iterator *iter)
1722{
1723 struct trace_seq *s = &iter->seq;
1724 unsigned char newline = '\n';
1725 struct trace_entry *entry;
1726 int S, T;
1727
1728 entry = iter->ent;
1729
1730 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1731 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1732 SEQ_PUT_HEX_FIELD_RET(s, entry->t);
1733
1734 switch (entry->type) {
1735 case TRACE_FN:
1736 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.ip);
1737 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1738 break;
1739 case TRACE_CTX:
1740 case TRACE_WAKE:
1741 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1742 state_to_char[entry->ctx.prev_state] : 'X';
1743 T = entry->ctx.next_state < sizeof(state_to_char) ?
1744 state_to_char[entry->ctx.next_state] : 'X';
1745 if (entry->type == TRACE_WAKE)
1746 S = '+';
1747 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_pid);
1748 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_prio);
1749 SEQ_PUT_HEX_FIELD_RET(s, S);
1750 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_pid);
1751 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_prio);
1752 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1753 SEQ_PUT_HEX_FIELD_RET(s, T);
1754 break;
1755 case TRACE_SPECIAL:
1756 case TRACE_STACK:
1757 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg1);
1758 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg2);
1759 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg3);
1760 break;
1761 }
1762 SEQ_PUT_FIELD_RET(s, newline);
1763
1764 return 1;
1765}
1766
1767static int print_bin_fmt(struct trace_iterator *iter)
1768{
1769 struct trace_seq *s = &iter->seq;
1770 struct trace_entry *entry;
1771
1772 entry = iter->ent;
1773
1774 SEQ_PUT_FIELD_RET(s, entry->pid);
1775 SEQ_PUT_FIELD_RET(s, entry->cpu);
1776 SEQ_PUT_FIELD_RET(s, entry->t);
1777
1778 switch (entry->type) {
1779 case TRACE_FN:
1780 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1781 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1782 break;
1783 case TRACE_CTX:
1784 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1785 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1786 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1787 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1788 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1789 SEQ_PUT_FIELD_RET(s, entry->ctx.next_state);
1790 break;
1791 case TRACE_SPECIAL:
1792 case TRACE_STACK:
1793 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1794 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1795 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1796 break;
1797 }
1798 return 1;
1799}
1800
1801static int trace_empty(struct trace_iterator *iter)
1802{
1803 struct trace_array_cpu *data;
1804 int cpu;
1805
1806 for_each_tracing_cpu(cpu) {
1807 data = iter->tr->data[cpu];
1808
1809 if (head_page(data) && data->trace_idx &&
1810 (data->trace_tail != data->trace_head ||
1811 data->trace_tail_idx != data->trace_head_idx))
1812 return 0;
1813 }
1814 return 1;
1815}
1816
1817static int print_trace_line(struct trace_iterator *iter)
1818{
1819 if (iter->trace && iter->trace->print_line)
1820 return iter->trace->print_line(iter);
1821
1822 if (trace_flags & TRACE_ITER_BIN)
1823 return print_bin_fmt(iter);
1824
1825 if (trace_flags & TRACE_ITER_HEX)
1826 return print_hex_fmt(iter);
1827
1828 if (trace_flags & TRACE_ITER_RAW)
1829 return print_raw_fmt(iter);
1830
1831 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1832 return print_lat_fmt(iter, iter->idx, iter->cpu);
1833
1834 return print_trace_fmt(iter);
1835}
1836
1837static int s_show(struct seq_file *m, void *v)
1838{
1839 struct trace_iterator *iter = v;
1840
1841 if (iter->ent == NULL) {
1842 if (iter->tr) {
1843 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1844 seq_puts(m, "#\n");
1845 }
1846 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1847 /* print nothing if the buffers are empty */
1848 if (trace_empty(iter))
1849 return 0;
1850 print_trace_header(m, iter);
1851 if (!(trace_flags & TRACE_ITER_VERBOSE))
1852 print_lat_help_header(m);
1853 } else {
1854 if (!(trace_flags & TRACE_ITER_VERBOSE))
1855 print_func_help_header(m);
1856 }
1857 } else {
1858 print_trace_line(iter);
1859 trace_print_seq(m, &iter->seq);
1860 }
1861
1862 return 0;
1863}
1864
1865static struct seq_operations tracer_seq_ops = {
1866 .start = s_start,
1867 .next = s_next,
1868 .stop = s_stop,
1869 .show = s_show,
1870};
1871
1872static struct trace_iterator *
1873__tracing_open(struct inode *inode, struct file *file, int *ret)
1874{
1875 struct trace_iterator *iter;
1876
1877 if (tracing_disabled) {
1878 *ret = -ENODEV;
1879 return NULL;
1880 }
1881
1882 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1883 if (!iter) {
1884 *ret = -ENOMEM;
1885 goto out;
1886 }
1887
1888 mutex_lock(&trace_types_lock);
1889 if (current_trace && current_trace->print_max)
1890 iter->tr = &max_tr;
1891 else
1892 iter->tr = inode->i_private;
1893 iter->trace = current_trace;
1894 iter->pos = -1;
1895
1896 /* TODO stop tracer */
1897 *ret = seq_open(file, &tracer_seq_ops);
1898 if (!*ret) {
1899 struct seq_file *m = file->private_data;
1900 m->private = iter;
1901
1902 /* stop the trace while dumping */
1903 if (iter->tr->ctrl) {
1904 tracer_enabled = 0;
1905 ftrace_function_enabled = 0;
1906 }
1907
1908 if (iter->trace && iter->trace->open)
1909 iter->trace->open(iter);
1910 } else {
1911 kfree(iter);
1912 iter = NULL;
1913 }
1914 mutex_unlock(&trace_types_lock);
1915
1916 out:
1917 return iter;
1918}
1919
1920int tracing_open_generic(struct inode *inode, struct file *filp)
1921{
1922 if (tracing_disabled)
1923 return -ENODEV;
1924
1925 filp->private_data = inode->i_private;
1926 return 0;
1927}
1928
1929int tracing_release(struct inode *inode, struct file *file)
1930{
1931 struct seq_file *m = (struct seq_file *)file->private_data;
1932 struct trace_iterator *iter = m->private;
1933
1934 mutex_lock(&trace_types_lock);
1935 if (iter->trace && iter->trace->close)
1936 iter->trace->close(iter);
1937
1938 /* reenable tracing if it was previously enabled */
1939 if (iter->tr->ctrl) {
1940 tracer_enabled = 1;
1941 /*
1942 * It is safe to enable function tracing even if it
1943 * isn't used
1944 */
1945 ftrace_function_enabled = 1;
1946 }
1947 mutex_unlock(&trace_types_lock);
1948
1949 seq_release(inode, file);
1950 kfree(iter);
1951 return 0;
1952}
1953
1954static int tracing_open(struct inode *inode, struct file *file)
1955{
1956 int ret;
1957
1958 __tracing_open(inode, file, &ret);
1959
1960 return ret;
1961}
1962
1963static int tracing_lt_open(struct inode *inode, struct file *file)
1964{
1965 struct trace_iterator *iter;
1966 int ret;
1967
1968 iter = __tracing_open(inode, file, &ret);
1969
1970 if (!ret)
1971 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1972
1973 return ret;
1974}
1975
1976
1977static void *
1978t_next(struct seq_file *m, void *v, loff_t *pos)
1979{
1980 struct tracer *t = m->private;
1981
1982 (*pos)++;
1983
1984 if (t)
1985 t = t->next;
1986
1987 m->private = t;
1988
1989 return t;
1990}
1991
1992static void *t_start(struct seq_file *m, loff_t *pos)
1993{
1994 struct tracer *t = m->private;
1995 loff_t l = 0;
1996
1997 mutex_lock(&trace_types_lock);
1998 for (; t && l < *pos; t = t_next(m, t, &l))
1999 ;
2000
2001 return t;
2002}
2003
2004static void t_stop(struct seq_file *m, void *p)
2005{
2006 mutex_unlock(&trace_types_lock);
2007}
2008
2009static int t_show(struct seq_file *m, void *v)
2010{
2011 struct tracer *t = v;
2012
2013 if (!t)
2014 return 0;
2015
2016 seq_printf(m, "%s", t->name);
2017 if (t->next)
2018 seq_putc(m, ' ');
2019 else
2020 seq_putc(m, '\n');
2021
2022 return 0;
2023}
2024
2025static struct seq_operations show_traces_seq_ops = {
2026 .start = t_start,
2027 .next = t_next,
2028 .stop = t_stop,
2029 .show = t_show,
2030};
2031
2032static int show_traces_open(struct inode *inode, struct file *file)
2033{
2034 int ret;
2035
2036 if (tracing_disabled)
2037 return -ENODEV;
2038
2039 ret = seq_open(file, &show_traces_seq_ops);
2040 if (!ret) {
2041 struct seq_file *m = file->private_data;
2042 m->private = trace_types;
2043 }
2044
2045 return ret;
2046}
2047
2048static struct file_operations tracing_fops = {
2049 .open = tracing_open,
2050 .read = seq_read,
2051 .llseek = seq_lseek,
2052 .release = tracing_release,
2053};
2054
2055static struct file_operations tracing_lt_fops = {
2056 .open = tracing_lt_open,
2057 .read = seq_read,
2058 .llseek = seq_lseek,
2059 .release = tracing_release,
2060};
2061
2062static struct file_operations show_traces_fops = {
2063 .open = show_traces_open,
2064 .read = seq_read,
2065 .release = seq_release,
2066};
2067
2068/*
2069 * Only trace on a CPU if the bitmask is set:
2070 */
2071static cpumask_t tracing_cpumask = CPU_MASK_ALL;
2072
2073/*
2074 * When tracing/tracing_cpu_mask is modified then this holds
2075 * the new bitmask we are about to install:
2076 */
2077static cpumask_t tracing_cpumask_new;
2078
2079/*
2080 * The tracer itself will not take this lock, but still we want
2081 * to provide a consistent cpumask to user-space:
2082 */
2083static DEFINE_MUTEX(tracing_cpumask_update_lock);
2084
2085/*
2086 * Temporary storage for the character representation of the
2087 * CPU bitmask (and one more byte for the newline):
2088 */
2089static char mask_str[NR_CPUS + 1];
2090
2091static ssize_t
2092tracing_cpumask_read(struct file *filp, char __user *ubuf,
2093 size_t count, loff_t *ppos)
2094{
2095 int len;
2096
2097 mutex_lock(&tracing_cpumask_update_lock);
2098
2099 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2100 if (count - len < 2) {
2101 count = -EINVAL;
2102 goto out_err;
2103 }
2104 len += sprintf(mask_str + len, "\n");
2105 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2106
2107out_err:
2108 mutex_unlock(&tracing_cpumask_update_lock);
2109
2110 return count;
2111}
2112
2113static ssize_t
2114tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2115 size_t count, loff_t *ppos)
2116{
2117 int err, cpu;
2118
2119 mutex_lock(&tracing_cpumask_update_lock);
2120 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2121 if (err)
2122 goto err_unlock;
2123
2124 raw_local_irq_disable();
2125 __raw_spin_lock(&ftrace_max_lock);
2126 for_each_tracing_cpu(cpu) {
2127 /*
2128 * Increase/decrease the disabled counter if we are
2129 * about to flip a bit in the cpumask:
2130 */
2131 if (cpu_isset(cpu, tracing_cpumask) &&
2132 !cpu_isset(cpu, tracing_cpumask_new)) {
2133 atomic_inc(&global_trace.data[cpu]->disabled);
2134 }
2135 if (!cpu_isset(cpu, tracing_cpumask) &&
2136 cpu_isset(cpu, tracing_cpumask_new)) {
2137 atomic_dec(&global_trace.data[cpu]->disabled);
2138 }
2139 }
2140 __raw_spin_unlock(&ftrace_max_lock);
2141 raw_local_irq_enable();
2142
2143 tracing_cpumask = tracing_cpumask_new;
2144
2145 mutex_unlock(&tracing_cpumask_update_lock);
2146
2147 return count;
2148
2149err_unlock:
2150 mutex_unlock(&tracing_cpumask_update_lock);
2151
2152 return err;
2153}
2154
2155static struct file_operations tracing_cpumask_fops = {
2156 .open = tracing_open_generic,
2157 .read = tracing_cpumask_read,
2158 .write = tracing_cpumask_write,
2159};
2160
2161static ssize_t
2162tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
2163 size_t cnt, loff_t *ppos)
2164{
2165 char *buf;
2166 int r = 0;
2167 int len = 0;
2168 int i;
2169
2170 /* calulate max size */
2171 for (i = 0; trace_options[i]; i++) {
2172 len += strlen(trace_options[i]);
2173 len += 3; /* "no" and space */
2174 }
2175
2176 /* +2 for \n and \0 */
2177 buf = kmalloc(len + 2, GFP_KERNEL);
2178 if (!buf)
2179 return -ENOMEM;
2180
2181 for (i = 0; trace_options[i]; i++) {
2182 if (trace_flags & (1 << i))
2183 r += sprintf(buf + r, "%s ", trace_options[i]);
2184 else
2185 r += sprintf(buf + r, "no%s ", trace_options[i]);
2186 }
2187
2188 r += sprintf(buf + r, "\n");
2189 WARN_ON(r >= len + 2);
2190
2191 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2192
2193 kfree(buf);
2194
2195 return r;
2196}
2197
2198static ssize_t
2199tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
2200 size_t cnt, loff_t *ppos)
2201{
2202 char buf[64];
2203 char *cmp = buf;
2204 int neg = 0;
2205 int i;
2206
2207 if (cnt >= sizeof(buf))
2208 return -EINVAL;
2209
2210 if (copy_from_user(&buf, ubuf, cnt))
2211 return -EFAULT;
2212
2213 buf[cnt] = 0;
2214
2215 if (strncmp(buf, "no", 2) == 0) {
2216 neg = 1;
2217 cmp += 2;
2218 }
2219
2220 for (i = 0; trace_options[i]; i++) {
2221 int len = strlen(trace_options[i]);
2222
2223 if (strncmp(cmp, trace_options[i], len) == 0) {
2224 if (neg)
2225 trace_flags &= ~(1 << i);
2226 else
2227 trace_flags |= (1 << i);
2228 break;
2229 }
2230 }
2231 /*
2232 * If no option could be set, return an error:
2233 */
2234 if (!trace_options[i])
2235 return -EINVAL;
2236
2237 filp->f_pos += cnt;
2238
2239 return cnt;
2240}
2241
2242static struct file_operations tracing_iter_fops = {
2243 .open = tracing_open_generic,
2244 .read = tracing_iter_ctrl_read,
2245 .write = tracing_iter_ctrl_write,
2246};
2247
2248static const char readme_msg[] =
2249 "tracing mini-HOWTO:\n\n"
2250 "# mkdir /debug\n"
2251 "# mount -t debugfs nodev /debug\n\n"
2252 "# cat /debug/tracing/available_tracers\n"
2253 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2254 "# cat /debug/tracing/current_tracer\n"
2255 "none\n"
2256 "# echo sched_switch > /debug/tracing/current_tracer\n"
2257 "# cat /debug/tracing/current_tracer\n"
2258 "sched_switch\n"
2259 "# cat /debug/tracing/iter_ctrl\n"
2260 "noprint-parent nosym-offset nosym-addr noverbose\n"
2261 "# echo print-parent > /debug/tracing/iter_ctrl\n"
2262 "# echo 1 > /debug/tracing/tracing_enabled\n"
2263 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2264 "echo 0 > /debug/tracing/tracing_enabled\n"
2265;
2266
2267static ssize_t
2268tracing_readme_read(struct file *filp, char __user *ubuf,
2269 size_t cnt, loff_t *ppos)
2270{
2271 return simple_read_from_buffer(ubuf, cnt, ppos,
2272 readme_msg, strlen(readme_msg));
2273}
2274
2275static struct file_operations tracing_readme_fops = {
2276 .open = tracing_open_generic,
2277 .read = tracing_readme_read,
2278};
2279
2280static ssize_t
2281tracing_ctrl_read(struct file *filp, char __user *ubuf,
2282 size_t cnt, loff_t *ppos)
2283{
2284 struct trace_array *tr = filp->private_data;
2285 char buf[64];
2286 int r;
2287
2288 r = sprintf(buf, "%ld\n", tr->ctrl);
2289 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2290}
2291
2292static ssize_t
2293tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2294 size_t cnt, loff_t *ppos)
2295{
2296 struct trace_array *tr = filp->private_data;
2297 char buf[64];
2298 long val;
2299 int ret;
2300
2301 if (cnt >= sizeof(buf))
2302 return -EINVAL;
2303
2304 if (copy_from_user(&buf, ubuf, cnt))
2305 return -EFAULT;
2306
2307 buf[cnt] = 0;
2308
2309 ret = strict_strtoul(buf, 10, &val);
2310 if (ret < 0)
2311 return ret;
2312
2313 val = !!val;
2314
2315 mutex_lock(&trace_types_lock);
2316 if (tr->ctrl ^ val) {
2317 if (val)
2318 tracer_enabled = 1;
2319 else
2320 tracer_enabled = 0;
2321
2322 tr->ctrl = val;
2323
2324 if (current_trace && current_trace->ctrl_update)
2325 current_trace->ctrl_update(tr);
2326 }
2327 mutex_unlock(&trace_types_lock);
2328
2329 filp->f_pos += cnt;
2330
2331 return cnt;
2332}
2333
2334static ssize_t
2335tracing_set_trace_read(struct file *filp, char __user *ubuf,
2336 size_t cnt, loff_t *ppos)
2337{
2338 char buf[max_tracer_type_len+2];
2339 int r;
2340
2341 mutex_lock(&trace_types_lock);
2342 if (current_trace)
2343 r = sprintf(buf, "%s\n", current_trace->name);
2344 else
2345 r = sprintf(buf, "\n");
2346 mutex_unlock(&trace_types_lock);
2347
2348 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2349}
2350
2351static ssize_t
2352tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2353 size_t cnt, loff_t *ppos)
2354{
2355 struct trace_array *tr = &global_trace;
2356 struct tracer *t;
2357 char buf[max_tracer_type_len+1];
2358 int i;
2359
2360 if (cnt > max_tracer_type_len)
2361 cnt = max_tracer_type_len;
2362
2363 if (copy_from_user(&buf, ubuf, cnt))
2364 return -EFAULT;
2365
2366 buf[cnt] = 0;
2367
2368 /* strip ending whitespace. */
2369 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2370 buf[i] = 0;
2371
2372 mutex_lock(&trace_types_lock);
2373 for (t = trace_types; t; t = t->next) {
2374 if (strcmp(t->name, buf) == 0)
2375 break;
2376 }
2377 if (!t || t == current_trace)
2378 goto out;
2379
2380 if (current_trace && current_trace->reset)
2381 current_trace->reset(tr);
2382
2383 current_trace = t;
2384 if (t->init)
2385 t->init(tr);
2386
2387 out:
2388 mutex_unlock(&trace_types_lock);
2389
2390 filp->f_pos += cnt;
2391
2392 return cnt;
2393}
2394
2395static ssize_t
2396tracing_max_lat_read(struct file *filp, char __user *ubuf,
2397 size_t cnt, loff_t *ppos)
2398{
2399 unsigned long *ptr = filp->private_data;
2400 char buf[64];
2401 int r;
2402
2403 r = snprintf(buf, sizeof(buf), "%ld\n",
2404 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2405 if (r > sizeof(buf))
2406 r = sizeof(buf);
2407 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2408}
2409
2410static ssize_t
2411tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2412 size_t cnt, loff_t *ppos)
2413{
2414 long *ptr = filp->private_data;
2415 char buf[64];
2416 long val;
2417 int ret;
2418
2419 if (cnt >= sizeof(buf))
2420 return -EINVAL;
2421
2422 if (copy_from_user(&buf, ubuf, cnt))
2423 return -EFAULT;
2424
2425 buf[cnt] = 0;
2426
2427 ret = strict_strtoul(buf, 10, &val);
2428 if (ret < 0)
2429 return ret;
2430
2431 *ptr = val * 1000;
2432
2433 return cnt;
2434}
2435
2436static atomic_t tracing_reader;
2437
2438static int tracing_open_pipe(struct inode *inode, struct file *filp)
2439{
2440 struct trace_iterator *iter;
2441
2442 if (tracing_disabled)
2443 return -ENODEV;
2444
2445 /* We only allow for reader of the pipe */
2446 if (atomic_inc_return(&tracing_reader) != 1) {
2447 atomic_dec(&tracing_reader);
2448 return -EBUSY;
2449 }
2450
2451 /* create a buffer to store the information to pass to userspace */
2452 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2453 if (!iter)
2454 return -ENOMEM;
2455
2456 mutex_lock(&trace_types_lock);
2457 iter->tr = &global_trace;
2458 iter->trace = current_trace;
2459 filp->private_data = iter;
2460
2461 if (iter->trace->pipe_open)
2462 iter->trace->pipe_open(iter);
2463 mutex_unlock(&trace_types_lock);
2464
2465 return 0;
2466}
2467
2468static int tracing_release_pipe(struct inode *inode, struct file *file)
2469{
2470 struct trace_iterator *iter = file->private_data;
2471
2472 kfree(iter);
2473 atomic_dec(&tracing_reader);
2474
2475 return 0;
2476}
2477
2478static unsigned int
2479tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2480{
2481 struct trace_iterator *iter = filp->private_data;
2482
2483 if (trace_flags & TRACE_ITER_BLOCK) {
2484 /*
2485 * Always select as readable when in blocking mode
2486 */
2487 return POLLIN | POLLRDNORM;
2488 } else {
2489 if (!trace_empty(iter))
2490 return POLLIN | POLLRDNORM;
2491 poll_wait(filp, &trace_wait, poll_table);
2492 if (!trace_empty(iter))
2493 return POLLIN | POLLRDNORM;
2494
2495 return 0;
2496 }
2497}
2498
2499/*
2500 * Consumer reader.
2501 */
2502static ssize_t
2503tracing_read_pipe(struct file *filp, char __user *ubuf,
2504 size_t cnt, loff_t *ppos)
2505{
2506 struct trace_iterator *iter = filp->private_data;
2507 struct trace_array_cpu *data;
2508 static cpumask_t mask;
2509 unsigned long flags;
2510#ifdef CONFIG_FTRACE
2511 int ftrace_save;
2512#endif
2513 int cpu;
2514 ssize_t sret;
2515
2516 /* return any leftover data */
2517 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2518 if (sret != -EBUSY)
2519 return sret;
2520 sret = 0;
2521
2522 trace_seq_reset(&iter->seq);
2523
2524 mutex_lock(&trace_types_lock);
2525 if (iter->trace->read) {
2526 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2527 if (sret)
2528 goto out;
2529 }
2530
2531 while (trace_empty(iter)) {
2532
2533 if ((filp->f_flags & O_NONBLOCK)) {
2534 sret = -EAGAIN;
2535 goto out;
2536 }
2537
2538 /*
2539 * This is a make-shift waitqueue. The reason we don't use
2540 * an actual wait queue is because:
2541 * 1) we only ever have one waiter
2542 * 2) the tracing, traces all functions, we don't want
2543 * the overhead of calling wake_up and friends
2544 * (and tracing them too)
2545 * Anyway, this is really very primitive wakeup.
2546 */
2547 set_current_state(TASK_INTERRUPTIBLE);
2548 iter->tr->waiter = current;
2549
2550 mutex_unlock(&trace_types_lock);
2551
2552 /* sleep for 100 msecs, and try again. */
2553 schedule_timeout(HZ/10);
2554
2555 mutex_lock(&trace_types_lock);
2556
2557 iter->tr->waiter = NULL;
2558
2559 if (signal_pending(current)) {
2560 sret = -EINTR;
2561 goto out;
2562 }
2563
2564 if (iter->trace != current_trace)
2565 goto out;
2566
2567 /*
2568 * We block until we read something and tracing is disabled.
2569 * We still block if tracing is disabled, but we have never
2570 * read anything. This allows a user to cat this file, and
2571 * then enable tracing. But after we have read something,
2572 * we give an EOF when tracing is again disabled.
2573 *
2574 * iter->pos will be 0 if we haven't read anything.
2575 */
2576 if (!tracer_enabled && iter->pos)
2577 break;
2578
2579 continue;
2580 }
2581
2582 /* stop when tracing is finished */
2583 if (trace_empty(iter))
2584 goto out;
2585
2586 if (cnt >= PAGE_SIZE)
2587 cnt = PAGE_SIZE - 1;
2588
2589 /* reset all but tr, trace, and overruns */
2590 memset(&iter->seq, 0,
2591 sizeof(struct trace_iterator) -
2592 offsetof(struct trace_iterator, seq));
2593 iter->pos = -1;
2594
2595 /*
2596 * We need to stop all tracing on all CPUS to read the
2597 * the next buffer. This is a bit expensive, but is
2598 * not done often. We fill all what we can read,
2599 * and then release the locks again.
2600 */
2601
2602 cpus_clear(mask);
2603 local_irq_save(flags);
2604#ifdef CONFIG_FTRACE
2605 ftrace_save = ftrace_enabled;
2606 ftrace_enabled = 0;
2607#endif
2608 smp_wmb();
2609 for_each_tracing_cpu(cpu) {
2610 data = iter->tr->data[cpu];
2611
2612 if (!head_page(data) || !data->trace_idx)
2613 continue;
2614
2615 atomic_inc(&data->disabled);
2616 cpu_set(cpu, mask);
2617 }
2618
2619 for_each_cpu_mask(cpu, mask) {
2620 data = iter->tr->data[cpu];
2621 __raw_spin_lock(&data->lock);
2622
2623 if (data->overrun > iter->last_overrun[cpu])
2624 iter->overrun[cpu] +=
2625 data->overrun - iter->last_overrun[cpu];
2626 iter->last_overrun[cpu] = data->overrun;
2627 }
2628
2629 while (find_next_entry_inc(iter) != NULL) {
2630 int ret;
2631 int len = iter->seq.len;
2632
2633 ret = print_trace_line(iter);
2634 if (!ret) {
2635 /* don't print partial lines */
2636 iter->seq.len = len;
2637 break;
2638 }
2639
2640 trace_consume(iter);
2641
2642 if (iter->seq.len >= cnt)
2643 break;
2644 }
2645
2646 for_each_cpu_mask(cpu, mask) {
2647 data = iter->tr->data[cpu];
2648 __raw_spin_unlock(&data->lock);
2649 }
2650
2651 for_each_cpu_mask(cpu, mask) {
2652 data = iter->tr->data[cpu];
2653 atomic_dec(&data->disabled);
2654 }
2655#ifdef CONFIG_FTRACE
2656 ftrace_enabled = ftrace_save;
2657#endif
2658 local_irq_restore(flags);
2659
2660 /* Now copy what we have to the user */
2661 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2662 if (iter->seq.readpos >= iter->seq.len)
2663 trace_seq_reset(&iter->seq);
2664 if (sret == -EBUSY)
2665 sret = 0;
2666
2667out:
2668 mutex_unlock(&trace_types_lock);
2669
2670 return sret;
2671}
2672
2673static ssize_t
2674tracing_entries_read(struct file *filp, char __user *ubuf,
2675 size_t cnt, loff_t *ppos)
2676{
2677 struct trace_array *tr = filp->private_data;
2678 char buf[64];
2679 int r;
2680
2681 r = sprintf(buf, "%lu\n", tr->entries);
2682 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2683}
2684
2685static ssize_t
2686tracing_entries_write(struct file *filp, const char __user *ubuf,
2687 size_t cnt, loff_t *ppos)
2688{
2689 unsigned long val;
2690 char buf[64];
2691 int i, ret;
2692
2693 if (cnt >= sizeof(buf))
2694 return -EINVAL;
2695
2696 if (copy_from_user(&buf, ubuf, cnt))
2697 return -EFAULT;
2698
2699 buf[cnt] = 0;
2700
2701 ret = strict_strtoul(buf, 10, &val);
2702 if (ret < 0)
2703 return ret;
2704
2705 /* must have at least 1 entry */
2706 if (!val)
2707 return -EINVAL;
2708
2709 mutex_lock(&trace_types_lock);
2710
2711 if (current_trace != &no_tracer) {
2712 cnt = -EBUSY;
2713 pr_info("ftrace: set current_tracer to none"
2714 " before modifying buffer size\n");
2715 goto out;
2716 }
2717
2718 if (val > global_trace.entries) {
2719 long pages_requested;
2720 unsigned long freeable_pages;
2721
2722 /* make sure we have enough memory before mapping */
2723 pages_requested =
2724 (val + (ENTRIES_PER_PAGE-1)) / ENTRIES_PER_PAGE;
2725
2726 /* account for each buffer (and max_tr) */
2727 pages_requested *= tracing_nr_buffers * 2;
2728
2729 /* Check for overflow */
2730 if (pages_requested < 0) {
2731 cnt = -ENOMEM;
2732 goto out;
2733 }
2734
2735 freeable_pages = determine_dirtyable_memory();
2736
2737 /* we only allow to request 1/4 of useable memory */
2738 if (pages_requested >
2739 ((freeable_pages + tracing_pages_allocated) / 4)) {
2740 cnt = -ENOMEM;
2741 goto out;
2742 }
2743
2744 while (global_trace.entries < val) {
2745 if (trace_alloc_page()) {
2746 cnt = -ENOMEM;
2747 goto out;
2748 }
2749 /* double check that we don't go over the known pages */
2750 if (tracing_pages_allocated > pages_requested)
2751 break;
2752 }
2753
2754 } else {
2755 /* include the number of entries in val (inc of page entries) */
2756 while (global_trace.entries > val + (ENTRIES_PER_PAGE - 1))
2757 trace_free_page();
2758 }
2759
2760 /* check integrity */
2761 for_each_tracing_cpu(i)
2762 check_pages(global_trace.data[i]);
2763
2764 filp->f_pos += cnt;
2765
2766 /* If check pages failed, return ENOMEM */
2767 if (tracing_disabled)
2768 cnt = -ENOMEM;
2769 out:
2770 max_tr.entries = global_trace.entries;
2771 mutex_unlock(&trace_types_lock);
2772
2773 return cnt;
2774}
2775
2776static struct file_operations tracing_max_lat_fops = {
2777 .open = tracing_open_generic,
2778 .read = tracing_max_lat_read,
2779 .write = tracing_max_lat_write,
2780};
2781
2782static struct file_operations tracing_ctrl_fops = {
2783 .open = tracing_open_generic,
2784 .read = tracing_ctrl_read,
2785 .write = tracing_ctrl_write,
2786};
2787
2788static struct file_operations set_tracer_fops = {
2789 .open = tracing_open_generic,
2790 .read = tracing_set_trace_read,
2791 .write = tracing_set_trace_write,
2792};
2793
2794static struct file_operations tracing_pipe_fops = {
2795 .open = tracing_open_pipe,
2796 .poll = tracing_poll_pipe,
2797 .read = tracing_read_pipe,
2798 .release = tracing_release_pipe,
2799};
2800
2801static struct file_operations tracing_entries_fops = {
2802 .open = tracing_open_generic,
2803 .read = tracing_entries_read,
2804 .write = tracing_entries_write,
2805};
2806
2807#ifdef CONFIG_DYNAMIC_FTRACE
2808
2809static ssize_t
2810tracing_read_long(struct file *filp, char __user *ubuf,
2811 size_t cnt, loff_t *ppos)
2812{
2813 unsigned long *p = filp->private_data;
2814 char buf[64];
2815 int r;
2816
2817 r = sprintf(buf, "%ld\n", *p);
2818
2819 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2820}
2821
2822static struct file_operations tracing_read_long_fops = {
2823 .open = tracing_open_generic,
2824 .read = tracing_read_long,
2825};
2826#endif
2827
2828static struct dentry *d_tracer;
2829
2830struct dentry *tracing_init_dentry(void)
2831{
2832 static int once;
2833
2834 if (d_tracer)
2835 return d_tracer;
2836
2837 d_tracer = debugfs_create_dir("tracing", NULL);
2838
2839 if (!d_tracer && !once) {
2840 once = 1;
2841 pr_warning("Could not create debugfs directory 'tracing'\n");
2842 return NULL;
2843 }
2844
2845 return d_tracer;
2846}
2847
2848#ifdef CONFIG_FTRACE_SELFTEST
2849/* Let selftest have access to static functions in this file */
2850#include "trace_selftest.c"
2851#endif
2852
2853static __init void tracer_init_debugfs(void)
2854{
2855 struct dentry *d_tracer;
2856 struct dentry *entry;
2857
2858 d_tracer = tracing_init_dentry();
2859
2860 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2861 &global_trace, &tracing_ctrl_fops);
2862 if (!entry)
2863 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2864
2865 entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2866 NULL, &tracing_iter_fops);
2867 if (!entry)
2868 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2869
2870 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
2871 NULL, &tracing_cpumask_fops);
2872 if (!entry)
2873 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2874
2875 entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2876 &global_trace, &tracing_lt_fops);
2877 if (!entry)
2878 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2879
2880 entry = debugfs_create_file("trace", 0444, d_tracer,
2881 &global_trace, &tracing_fops);
2882 if (!entry)
2883 pr_warning("Could not create debugfs 'trace' entry\n");
2884
2885 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2886 &global_trace, &show_traces_fops);
2887 if (!entry)
2888 pr_warning("Could not create debugfs 'trace' entry\n");
2889
2890 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2891 &global_trace, &set_tracer_fops);
2892 if (!entry)
2893 pr_warning("Could not create debugfs 'trace' entry\n");
2894
2895 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2896 &tracing_max_latency,
2897 &tracing_max_lat_fops);
2898 if (!entry)
2899 pr_warning("Could not create debugfs "
2900 "'tracing_max_latency' entry\n");
2901
2902 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2903 &tracing_thresh, &tracing_max_lat_fops);
2904 if (!entry)
2905 pr_warning("Could not create debugfs "
2906 "'tracing_threash' entry\n");
2907 entry = debugfs_create_file("README", 0644, d_tracer,
2908 NULL, &tracing_readme_fops);
2909 if (!entry)
2910 pr_warning("Could not create debugfs 'README' entry\n");
2911
2912 entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2913 NULL, &tracing_pipe_fops);
2914 if (!entry)
2915 pr_warning("Could not create debugfs "
2916 "'tracing_threash' entry\n");
2917
2918 entry = debugfs_create_file("trace_entries", 0644, d_tracer,
2919 &global_trace, &tracing_entries_fops);
2920 if (!entry)
2921 pr_warning("Could not create debugfs "
2922 "'tracing_threash' entry\n");
2923
2924#ifdef CONFIG_DYNAMIC_FTRACE
2925 entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2926 &ftrace_update_tot_cnt,
2927 &tracing_read_long_fops);
2928 if (!entry)
2929 pr_warning("Could not create debugfs "
2930 "'dyn_ftrace_total_info' entry\n");
2931#endif
2932#ifdef CONFIG_SYSPROF_TRACER
2933 init_tracer_sysprof_debugfs(d_tracer);
2934#endif
2935}
2936
2937static int trace_alloc_page(void)
2938{
2939 struct trace_array_cpu *data;
2940 struct page *page, *tmp;
2941 LIST_HEAD(pages);
2942 void *array;
2943 unsigned pages_allocated = 0;
2944 int i;
2945
2946 /* first allocate a page for each CPU */
2947 for_each_tracing_cpu(i) {
2948 array = (void *)__get_free_page(GFP_KERNEL);
2949 if (array == NULL) {
2950 printk(KERN_ERR "tracer: failed to allocate page"
2951 "for trace buffer!\n");
2952 goto free_pages;
2953 }
2954
2955 pages_allocated++;
2956 page = virt_to_page(array);
2957 list_add(&page->lru, &pages);
2958
2959/* Only allocate if we are actually using the max trace */
2960#ifdef CONFIG_TRACER_MAX_TRACE
2961 array = (void *)__get_free_page(GFP_KERNEL);
2962 if (array == NULL) {
2963 printk(KERN_ERR "tracer: failed to allocate page"
2964 "for trace buffer!\n");
2965 goto free_pages;
2966 }
2967 pages_allocated++;
2968 page = virt_to_page(array);
2969 list_add(&page->lru, &pages);
2970#endif
2971 }
2972
2973 /* Now that we successfully allocate a page per CPU, add them */
2974 for_each_tracing_cpu(i) {
2975 data = global_trace.data[i];
2976 page = list_entry(pages.next, struct page, lru);
2977 list_del_init(&page->lru);
2978 list_add_tail(&page->lru, &data->trace_pages);
2979 ClearPageLRU(page);
2980
2981#ifdef CONFIG_TRACER_MAX_TRACE
2982 data = max_tr.data[i];
2983 page = list_entry(pages.next, struct page, lru);
2984 list_del_init(&page->lru);
2985 list_add_tail(&page->lru, &data->trace_pages);
2986 SetPageLRU(page);
2987#endif
2988 }
2989 tracing_pages_allocated += pages_allocated;
2990 global_trace.entries += ENTRIES_PER_PAGE;
2991
2992 return 0;
2993
2994 free_pages:
2995 list_for_each_entry_safe(page, tmp, &pages, lru) {
2996 list_del_init(&page->lru);
2997 __free_page(page);
2998 }
2999 return -ENOMEM;
3000}
3001
3002static int trace_free_page(void)
3003{
3004 struct trace_array_cpu *data;
3005 struct page *page;
3006 struct list_head *p;
3007 int i;
3008 int ret = 0;
3009
3010 /* free one page from each buffer */
3011 for_each_tracing_cpu(i) {
3012 data = global_trace.data[i];
3013 p = data->trace_pages.next;
3014 if (p == &data->trace_pages) {
3015 /* should never happen */
3016 WARN_ON(1);
3017 tracing_disabled = 1;
3018 ret = -1;
3019 break;
3020 }
3021 page = list_entry(p, struct page, lru);
3022 ClearPageLRU(page);
3023 list_del(&page->lru);
3024 tracing_pages_allocated--;
3025 tracing_pages_allocated--;
3026 __free_page(page);
3027
3028 tracing_reset(data);
3029
3030#ifdef CONFIG_TRACER_MAX_TRACE
3031 data = max_tr.data[i];
3032 p = data->trace_pages.next;
3033 if (p == &data->trace_pages) {
3034 /* should never happen */
3035 WARN_ON(1);
3036 tracing_disabled = 1;
3037 ret = -1;
3038 break;
3039 }
3040 page = list_entry(p, struct page, lru);
3041 ClearPageLRU(page);
3042 list_del(&page->lru);
3043 __free_page(page);
3044
3045 tracing_reset(data);
3046#endif
3047 }
3048 global_trace.entries -= ENTRIES_PER_PAGE;
3049
3050 return ret;
3051}
3052
3053__init static int tracer_alloc_buffers(void)
3054{
3055 struct trace_array_cpu *data;
3056 void *array;
3057 struct page *page;
3058 int pages = 0;
3059 int ret = -ENOMEM;
3060 int i;
3061
3062 /* TODO: make the number of buffers hot pluggable with CPUS */
3063 tracing_nr_buffers = num_possible_cpus();
3064 tracing_buffer_mask = cpu_possible_map;
3065
3066 /* Allocate the first page for all buffers */
3067 for_each_tracing_cpu(i) {
3068 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
3069 max_tr.data[i] = &per_cpu(max_data, i);
3070
3071 array = (void *)__get_free_page(GFP_KERNEL);
3072 if (array == NULL) {
3073 printk(KERN_ERR "tracer: failed to allocate page"
3074 "for trace buffer!\n");
3075 goto free_buffers;
3076 }
3077
3078 /* set the array to the list */
3079 INIT_LIST_HEAD(&data->trace_pages);
3080 page = virt_to_page(array);
3081 list_add(&page->lru, &data->trace_pages);
3082 /* use the LRU flag to differentiate the two buffers */
3083 ClearPageLRU(page);
3084
3085 data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
3086 max_tr.data[i]->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
3087
3088/* Only allocate if we are actually using the max trace */
3089#ifdef CONFIG_TRACER_MAX_TRACE
3090 array = (void *)__get_free_page(GFP_KERNEL);
3091 if (array == NULL) {
3092 printk(KERN_ERR "tracer: failed to allocate page"
3093 "for trace buffer!\n");
3094 goto free_buffers;
3095 }
3096
3097 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
3098 page = virt_to_page(array);
3099 list_add(&page->lru, &max_tr.data[i]->trace_pages);
3100 SetPageLRU(page);
3101#endif
3102 }
3103
3104 /*
3105 * Since we allocate by orders of pages, we may be able to
3106 * round up a bit.
3107 */
3108 global_trace.entries = ENTRIES_PER_PAGE;
3109 pages++;
3110
3111 while (global_trace.entries < trace_nr_entries) {
3112 if (trace_alloc_page())
3113 break;
3114 pages++;
3115 }
3116 max_tr.entries = global_trace.entries;
3117
3118 pr_info("tracer: %d pages allocated for %ld entries of %ld bytes\n",
3119 pages, trace_nr_entries, (long)TRACE_ENTRY_SIZE);
3120 pr_info(" actual entries %ld\n", global_trace.entries);
3121
3122 tracer_init_debugfs();
3123
3124 trace_init_cmdlines();
3125
3126 register_tracer(&no_tracer);
3127 current_trace = &no_tracer;
3128
3129 /* All seems OK, enable tracing */
3130 global_trace.ctrl = tracer_enabled;
3131 tracing_disabled = 0;
3132
3133 return 0;
3134
3135 free_buffers:
3136 for (i-- ; i >= 0; i--) {
3137 struct page *page, *tmp;
3138 struct trace_array_cpu *data = global_trace.data[i];
3139
3140 if (data) {
3141 list_for_each_entry_safe(page, tmp,
3142 &data->trace_pages, lru) {
3143 list_del_init(&page->lru);
3144 __free_page(page);
3145 }
3146 }
3147
3148#ifdef CONFIG_TRACER_MAX_TRACE
3149 data = max_tr.data[i];
3150 if (data) {
3151 list_for_each_entry_safe(page, tmp,
3152 &data->trace_pages, lru) {
3153 list_del_init(&page->lru);
3154 __free_page(page);
3155 }
3156 }
3157#endif
3158 }
3159 return ret;
3160}
3161fs_initcall(tracer_alloc_buffers);
diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
new file mode 100644
index 000000000000..f69f86788c2b
--- /dev/null
+++ b/kernel/trace/trace.h
@@ -0,0 +1,339 @@
1#ifndef _LINUX_KERNEL_TRACE_H
2#define _LINUX_KERNEL_TRACE_H
3
4#include <linux/fs.h>
5#include <asm/atomic.h>
6#include <linux/sched.h>
7#include <linux/clocksource.h>
8#include <linux/mmiotrace.h>
9
10enum trace_type {
11 __TRACE_FIRST_TYPE = 0,
12
13 TRACE_FN,
14 TRACE_CTX,
15 TRACE_WAKE,
16 TRACE_STACK,
17 TRACE_SPECIAL,
18 TRACE_MMIO_RW,
19 TRACE_MMIO_MAP,
20
21 __TRACE_LAST_TYPE
22};
23
24/*
25 * Function trace entry - function address and parent function addres:
26 */
27struct ftrace_entry {
28 unsigned long ip;
29 unsigned long parent_ip;
30};
31
32/*
33 * Context switch trace entry - which task (and prio) we switched from/to:
34 */
35struct ctx_switch_entry {
36 unsigned int prev_pid;
37 unsigned char prev_prio;
38 unsigned char prev_state;
39 unsigned int next_pid;
40 unsigned char next_prio;
41 unsigned char next_state;
42};
43
44/*
45 * Special (free-form) trace entry:
46 */
47struct special_entry {
48 unsigned long arg1;
49 unsigned long arg2;
50 unsigned long arg3;
51};
52
53/*
54 * Stack-trace entry:
55 */
56
57#define FTRACE_STACK_ENTRIES 8
58
59struct stack_entry {
60 unsigned long caller[FTRACE_STACK_ENTRIES];
61};
62
63/*
64 * The trace entry - the most basic unit of tracing. This is what
65 * is printed in the end as a single line in the trace output, such as:
66 *
67 * bash-15816 [01] 235.197585: idle_cpu <- irq_enter
68 */
69struct trace_entry {
70 char type;
71 char cpu;
72 char flags;
73 char preempt_count;
74 int pid;
75 cycle_t t;
76 union {
77 struct ftrace_entry fn;
78 struct ctx_switch_entry ctx;
79 struct special_entry special;
80 struct stack_entry stack;
81 struct mmiotrace_rw mmiorw;
82 struct mmiotrace_map mmiomap;
83 };
84};
85
86#define TRACE_ENTRY_SIZE sizeof(struct trace_entry)
87
88/*
89 * The CPU trace array - it consists of thousands of trace entries
90 * plus some other descriptor data: (for example which task started
91 * the trace, etc.)
92 */
93struct trace_array_cpu {
94 struct list_head trace_pages;
95 atomic_t disabled;
96 raw_spinlock_t lock;
97 struct lock_class_key lock_key;
98
99 /* these fields get copied into max-trace: */
100 unsigned trace_head_idx;
101 unsigned trace_tail_idx;
102 void *trace_head; /* producer */
103 void *trace_tail; /* consumer */
104 unsigned long trace_idx;
105 unsigned long overrun;
106 unsigned long saved_latency;
107 unsigned long critical_start;
108 unsigned long critical_end;
109 unsigned long critical_sequence;
110 unsigned long nice;
111 unsigned long policy;
112 unsigned long rt_priority;
113 cycle_t preempt_timestamp;
114 pid_t pid;
115 uid_t uid;
116 char comm[TASK_COMM_LEN];
117};
118
119struct trace_iterator;
120
121/*
122 * The trace array - an array of per-CPU trace arrays. This is the
123 * highest level data structure that individual tracers deal with.
124 * They have on/off state as well:
125 */
126struct trace_array {
127 unsigned long entries;
128 long ctrl;
129 int cpu;
130 cycle_t time_start;
131 struct task_struct *waiter;
132 struct trace_array_cpu *data[NR_CPUS];
133};
134
135/*
136 * A specific tracer, represented by methods that operate on a trace array:
137 */
138struct tracer {
139 const char *name;
140 void (*init)(struct trace_array *tr);
141 void (*reset)(struct trace_array *tr);
142 void (*open)(struct trace_iterator *iter);
143 void (*pipe_open)(struct trace_iterator *iter);
144 void (*close)(struct trace_iterator *iter);
145 void (*start)(struct trace_iterator *iter);
146 void (*stop)(struct trace_iterator *iter);
147 ssize_t (*read)(struct trace_iterator *iter,
148 struct file *filp, char __user *ubuf,
149 size_t cnt, loff_t *ppos);
150 void (*ctrl_update)(struct trace_array *tr);
151#ifdef CONFIG_FTRACE_STARTUP_TEST
152 int (*selftest)(struct tracer *trace,
153 struct trace_array *tr);
154#endif
155 int (*print_line)(struct trace_iterator *iter);
156 struct tracer *next;
157 int print_max;
158};
159
160struct trace_seq {
161 unsigned char buffer[PAGE_SIZE];
162 unsigned int len;
163 unsigned int readpos;
164};
165
166/*
167 * Trace iterator - used by printout routines who present trace
168 * results to users and which routines might sleep, etc:
169 */
170struct trace_iterator {
171 struct trace_array *tr;
172 struct tracer *trace;
173 void *private;
174 long last_overrun[NR_CPUS];
175 long overrun[NR_CPUS];
176
177 /* The below is zeroed out in pipe_read */
178 struct trace_seq seq;
179 struct trace_entry *ent;
180 int cpu;
181
182 struct trace_entry *prev_ent;
183 int prev_cpu;
184
185 unsigned long iter_flags;
186 loff_t pos;
187 unsigned long next_idx[NR_CPUS];
188 struct list_head *next_page[NR_CPUS];
189 unsigned next_page_idx[NR_CPUS];
190 long idx;
191};
192
193void tracing_reset(struct trace_array_cpu *data);
194int tracing_open_generic(struct inode *inode, struct file *filp);
195struct dentry *tracing_init_dentry(void);
196void init_tracer_sysprof_debugfs(struct dentry *d_tracer);
197
198void ftrace(struct trace_array *tr,
199 struct trace_array_cpu *data,
200 unsigned long ip,
201 unsigned long parent_ip,
202 unsigned long flags);
203void tracing_sched_switch_trace(struct trace_array *tr,
204 struct trace_array_cpu *data,
205 struct task_struct *prev,
206 struct task_struct *next,
207 unsigned long flags);
208void tracing_record_cmdline(struct task_struct *tsk);
209
210void tracing_sched_wakeup_trace(struct trace_array *tr,
211 struct trace_array_cpu *data,
212 struct task_struct *wakee,
213 struct task_struct *cur,
214 unsigned long flags);
215void trace_special(struct trace_array *tr,
216 struct trace_array_cpu *data,
217 unsigned long arg1,
218 unsigned long arg2,
219 unsigned long arg3);
220void trace_function(struct trace_array *tr,
221 struct trace_array_cpu *data,
222 unsigned long ip,
223 unsigned long parent_ip,
224 unsigned long flags);
225
226void tracing_start_cmdline_record(void);
227void tracing_stop_cmdline_record(void);
228int register_tracer(struct tracer *type);
229void unregister_tracer(struct tracer *type);
230
231extern unsigned long nsecs_to_usecs(unsigned long nsecs);
232
233extern unsigned long tracing_max_latency;
234extern unsigned long tracing_thresh;
235
236void update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu);
237void update_max_tr_single(struct trace_array *tr,
238 struct task_struct *tsk, int cpu);
239
240extern cycle_t ftrace_now(int cpu);
241
242#ifdef CONFIG_FTRACE
243void tracing_start_function_trace(void);
244void tracing_stop_function_trace(void);
245#else
246# define tracing_start_function_trace() do { } while (0)
247# define tracing_stop_function_trace() do { } while (0)
248#endif
249
250#ifdef CONFIG_CONTEXT_SWITCH_TRACER
251typedef void
252(*tracer_switch_func_t)(void *private,
253 void *__rq,
254 struct task_struct *prev,
255 struct task_struct *next);
256
257struct tracer_switch_ops {
258 tracer_switch_func_t func;
259 void *private;
260 struct tracer_switch_ops *next;
261};
262
263#endif /* CONFIG_CONTEXT_SWITCH_TRACER */
264
265#ifdef CONFIG_DYNAMIC_FTRACE
266extern unsigned long ftrace_update_tot_cnt;
267#define DYN_FTRACE_TEST_NAME trace_selftest_dynamic_test_func
268extern int DYN_FTRACE_TEST_NAME(void);
269#endif
270
271#ifdef CONFIG_MMIOTRACE
272extern void __trace_mmiotrace_rw(struct trace_array *tr,
273 struct trace_array_cpu *data,
274 struct mmiotrace_rw *rw);
275extern void __trace_mmiotrace_map(struct trace_array *tr,
276 struct trace_array_cpu *data,
277 struct mmiotrace_map *map);
278#endif
279
280#ifdef CONFIG_FTRACE_STARTUP_TEST
281#ifdef CONFIG_FTRACE
282extern int trace_selftest_startup_function(struct tracer *trace,
283 struct trace_array *tr);
284#endif
285#ifdef CONFIG_IRQSOFF_TRACER
286extern int trace_selftest_startup_irqsoff(struct tracer *trace,
287 struct trace_array *tr);
288#endif
289#ifdef CONFIG_PREEMPT_TRACER
290extern int trace_selftest_startup_preemptoff(struct tracer *trace,
291 struct trace_array *tr);
292#endif
293#if defined(CONFIG_IRQSOFF_TRACER) && defined(CONFIG_PREEMPT_TRACER)
294extern int trace_selftest_startup_preemptirqsoff(struct tracer *trace,
295 struct trace_array *tr);
296#endif
297#ifdef CONFIG_SCHED_TRACER
298extern int trace_selftest_startup_wakeup(struct tracer *trace,
299 struct trace_array *tr);
300#endif
301#ifdef CONFIG_CONTEXT_SWITCH_TRACER
302extern int trace_selftest_startup_sched_switch(struct tracer *trace,
303 struct trace_array *tr);
304#endif
305#ifdef CONFIG_SYSPROF_TRACER
306extern int trace_selftest_startup_sysprof(struct tracer *trace,
307 struct trace_array *tr);
308#endif
309#endif /* CONFIG_FTRACE_STARTUP_TEST */
310
311extern void *head_page(struct trace_array_cpu *data);
312extern int trace_seq_printf(struct trace_seq *s, const char *fmt, ...);
313extern ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf,
314 size_t cnt);
315extern long ns2usecs(cycle_t nsec);
316
317extern unsigned long trace_flags;
318
319/*
320 * trace_iterator_flags is an enumeration that defines bit
321 * positions into trace_flags that controls the output.
322 *
323 * NOTE: These bits must match the trace_options array in
324 * trace.c.
325 */
326enum trace_iterator_flags {
327 TRACE_ITER_PRINT_PARENT = 0x01,
328 TRACE_ITER_SYM_OFFSET = 0x02,
329 TRACE_ITER_SYM_ADDR = 0x04,
330 TRACE_ITER_VERBOSE = 0x08,
331 TRACE_ITER_RAW = 0x10,
332 TRACE_ITER_HEX = 0x20,
333 TRACE_ITER_BIN = 0x40,
334 TRACE_ITER_BLOCK = 0x80,
335 TRACE_ITER_STACKTRACE = 0x100,
336 TRACE_ITER_SCHED_TREE = 0x200,
337};
338
339#endif /* _LINUX_KERNEL_TRACE_H */
diff --git a/kernel/trace/trace_functions.c b/kernel/trace/trace_functions.c
new file mode 100644
index 000000000000..312144897970
--- /dev/null
+++ b/kernel/trace/trace_functions.c
@@ -0,0 +1,81 @@
1/*
2 * ring buffer based function tracer
3 *
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
6 *
7 * Based on code from the latency_tracer, that is:
8 *
9 * Copyright (C) 2004-2006 Ingo Molnar
10 * Copyright (C) 2004 William Lee Irwin III
11 */
12#include <linux/debugfs.h>
13#include <linux/uaccess.h>
14#include <linux/ftrace.h>
15#include <linux/fs.h>
16
17#include "trace.h"
18
19static void function_reset(struct trace_array *tr)
20{
21 int cpu;
22
23 tr->time_start = ftrace_now(tr->cpu);
24
25 for_each_online_cpu(cpu)
26 tracing_reset(tr->data[cpu]);
27}
28
29static void start_function_trace(struct trace_array *tr)
30{
31 tr->cpu = get_cpu();
32 function_reset(tr);
33 put_cpu();
34
35 tracing_start_cmdline_record();
36 tracing_start_function_trace();
37}
38
39static void stop_function_trace(struct trace_array *tr)
40{
41 tracing_stop_function_trace();
42 tracing_stop_cmdline_record();
43}
44
45static void function_trace_init(struct trace_array *tr)
46{
47 if (tr->ctrl)
48 start_function_trace(tr);
49}
50
51static void function_trace_reset(struct trace_array *tr)
52{
53 if (tr->ctrl)
54 stop_function_trace(tr);
55}
56
57static void function_trace_ctrl_update(struct trace_array *tr)
58{
59 if (tr->ctrl)
60 start_function_trace(tr);
61 else
62 stop_function_trace(tr);
63}
64
65static struct tracer function_trace __read_mostly =
66{
67 .name = "ftrace",
68 .init = function_trace_init,
69 .reset = function_trace_reset,
70 .ctrl_update = function_trace_ctrl_update,
71#ifdef CONFIG_FTRACE_SELFTEST
72 .selftest = trace_selftest_startup_function,
73#endif
74};
75
76static __init int init_function_trace(void)
77{
78 return register_tracer(&function_trace);
79}
80
81device_initcall(init_function_trace);
diff --git a/kernel/trace/trace_irqsoff.c b/kernel/trace/trace_irqsoff.c
new file mode 100644
index 000000000000..421d6fe3650e
--- /dev/null
+++ b/kernel/trace/trace_irqsoff.c
@@ -0,0 +1,486 @@
1/*
2 * trace irqs off criticall timings
3 *
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
6 *
7 * From code in the latency_tracer, that is:
8 *
9 * Copyright (C) 2004-2006 Ingo Molnar
10 * Copyright (C) 2004 William Lee Irwin III
11 */
12#include <linux/kallsyms.h>
13#include <linux/debugfs.h>
14#include <linux/uaccess.h>
15#include <linux/module.h>
16#include <linux/ftrace.h>
17#include <linux/fs.h>
18
19#include "trace.h"
20
21static struct trace_array *irqsoff_trace __read_mostly;
22static int tracer_enabled __read_mostly;
23
24static DEFINE_PER_CPU(int, tracing_cpu);
25
26static DEFINE_SPINLOCK(max_trace_lock);
27
28enum {
29 TRACER_IRQS_OFF = (1 << 1),
30 TRACER_PREEMPT_OFF = (1 << 2),
31};
32
33static int trace_type __read_mostly;
34
35#ifdef CONFIG_PREEMPT_TRACER
36static inline int
37preempt_trace(void)
38{
39 return ((trace_type & TRACER_PREEMPT_OFF) && preempt_count());
40}
41#else
42# define preempt_trace() (0)
43#endif
44
45#ifdef CONFIG_IRQSOFF_TRACER
46static inline int
47irq_trace(void)
48{
49 return ((trace_type & TRACER_IRQS_OFF) &&
50 irqs_disabled());
51}
52#else
53# define irq_trace() (0)
54#endif
55
56/*
57 * Sequence count - we record it when starting a measurement and
58 * skip the latency if the sequence has changed - some other section
59 * did a maximum and could disturb our measurement with serial console
60 * printouts, etc. Truly coinciding maximum latencies should be rare
61 * and what happens together happens separately as well, so this doesnt
62 * decrease the validity of the maximum found:
63 */
64static __cacheline_aligned_in_smp unsigned long max_sequence;
65
66#ifdef CONFIG_FTRACE
67/*
68 * irqsoff uses its own tracer function to keep the overhead down:
69 */
70static void
71irqsoff_tracer_call(unsigned long ip, unsigned long parent_ip)
72{
73 struct trace_array *tr = irqsoff_trace;
74 struct trace_array_cpu *data;
75 unsigned long flags;
76 long disabled;
77 int cpu;
78
79 /*
80 * Does not matter if we preempt. We test the flags
81 * afterward, to see if irqs are disabled or not.
82 * If we preempt and get a false positive, the flags
83 * test will fail.
84 */
85 cpu = raw_smp_processor_id();
86 if (likely(!per_cpu(tracing_cpu, cpu)))
87 return;
88
89 local_save_flags(flags);
90 /* slight chance to get a false positive on tracing_cpu */
91 if (!irqs_disabled_flags(flags))
92 return;
93
94 data = tr->data[cpu];
95 disabled = atomic_inc_return(&data->disabled);
96
97 if (likely(disabled == 1))
98 trace_function(tr, data, ip, parent_ip, flags);
99
100 atomic_dec(&data->disabled);
101}
102
103static struct ftrace_ops trace_ops __read_mostly =
104{
105 .func = irqsoff_tracer_call,
106};
107#endif /* CONFIG_FTRACE */
108
109/*
110 * Should this new latency be reported/recorded?
111 */
112static int report_latency(cycle_t delta)
113{
114 if (tracing_thresh) {
115 if (delta < tracing_thresh)
116 return 0;
117 } else {
118 if (delta <= tracing_max_latency)
119 return 0;
120 }
121 return 1;
122}
123
124static void
125check_critical_timing(struct trace_array *tr,
126 struct trace_array_cpu *data,
127 unsigned long parent_ip,
128 int cpu)
129{
130 unsigned long latency, t0, t1;
131 cycle_t T0, T1, delta;
132 unsigned long flags;
133
134 /*
135 * usecs conversion is slow so we try to delay the conversion
136 * as long as possible:
137 */
138 T0 = data->preempt_timestamp;
139 T1 = ftrace_now(cpu);
140 delta = T1-T0;
141
142 local_save_flags(flags);
143
144 if (!report_latency(delta))
145 goto out;
146
147 spin_lock_irqsave(&max_trace_lock, flags);
148
149 /* check if we are still the max latency */
150 if (!report_latency(delta))
151 goto out_unlock;
152
153 trace_function(tr, data, CALLER_ADDR0, parent_ip, flags);
154
155 latency = nsecs_to_usecs(delta);
156
157 if (data->critical_sequence != max_sequence)
158 goto out_unlock;
159
160 tracing_max_latency = delta;
161 t0 = nsecs_to_usecs(T0);
162 t1 = nsecs_to_usecs(T1);
163
164 data->critical_end = parent_ip;
165
166 update_max_tr_single(tr, current, cpu);
167
168 max_sequence++;
169
170out_unlock:
171 spin_unlock_irqrestore(&max_trace_lock, flags);
172
173out:
174 data->critical_sequence = max_sequence;
175 data->preempt_timestamp = ftrace_now(cpu);
176 tracing_reset(data);
177 trace_function(tr, data, CALLER_ADDR0, parent_ip, flags);
178}
179
180static inline void
181start_critical_timing(unsigned long ip, unsigned long parent_ip)
182{
183 int cpu;
184 struct trace_array *tr = irqsoff_trace;
185 struct trace_array_cpu *data;
186 unsigned long flags;
187
188 if (likely(!tracer_enabled))
189 return;
190
191 cpu = raw_smp_processor_id();
192
193 if (per_cpu(tracing_cpu, cpu))
194 return;
195
196 data = tr->data[cpu];
197
198 if (unlikely(!data) || atomic_read(&data->disabled))
199 return;
200
201 atomic_inc(&data->disabled);
202
203 data->critical_sequence = max_sequence;
204 data->preempt_timestamp = ftrace_now(cpu);
205 data->critical_start = parent_ip ? : ip;
206 tracing_reset(data);
207
208 local_save_flags(flags);
209
210 trace_function(tr, data, ip, parent_ip, flags);
211
212 per_cpu(tracing_cpu, cpu) = 1;
213
214 atomic_dec(&data->disabled);
215}
216
217static inline void
218stop_critical_timing(unsigned long ip, unsigned long parent_ip)
219{
220 int cpu;
221 struct trace_array *tr = irqsoff_trace;
222 struct trace_array_cpu *data;
223 unsigned long flags;
224
225 cpu = raw_smp_processor_id();
226 /* Always clear the tracing cpu on stopping the trace */
227 if (unlikely(per_cpu(tracing_cpu, cpu)))
228 per_cpu(tracing_cpu, cpu) = 0;
229 else
230 return;
231
232 if (!tracer_enabled)
233 return;
234
235 data = tr->data[cpu];
236
237 if (unlikely(!data) || unlikely(!head_page(data)) ||
238 !data->critical_start || atomic_read(&data->disabled))
239 return;
240
241 atomic_inc(&data->disabled);
242
243 local_save_flags(flags);
244 trace_function(tr, data, ip, parent_ip, flags);
245 check_critical_timing(tr, data, parent_ip ? : ip, cpu);
246 data->critical_start = 0;
247 atomic_dec(&data->disabled);
248}
249
250/* start and stop critical timings used to for stoppage (in idle) */
251void start_critical_timings(void)
252{
253 if (preempt_trace() || irq_trace())
254 start_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
255}
256
257void stop_critical_timings(void)
258{
259 if (preempt_trace() || irq_trace())
260 stop_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
261}
262
263#ifdef CONFIG_IRQSOFF_TRACER
264#ifdef CONFIG_PROVE_LOCKING
265void time_hardirqs_on(unsigned long a0, unsigned long a1)
266{
267 if (!preempt_trace() && irq_trace())
268 stop_critical_timing(a0, a1);
269}
270
271void time_hardirqs_off(unsigned long a0, unsigned long a1)
272{
273 if (!preempt_trace() && irq_trace())
274 start_critical_timing(a0, a1);
275}
276
277#else /* !CONFIG_PROVE_LOCKING */
278
279/*
280 * Stubs:
281 */
282
283void early_boot_irqs_off(void)
284{
285}
286
287void early_boot_irqs_on(void)
288{
289}
290
291void trace_softirqs_on(unsigned long ip)
292{
293}
294
295void trace_softirqs_off(unsigned long ip)
296{
297}
298
299inline void print_irqtrace_events(struct task_struct *curr)
300{
301}
302
303/*
304 * We are only interested in hardirq on/off events:
305 */
306void trace_hardirqs_on(void)
307{
308 if (!preempt_trace() && irq_trace())
309 stop_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
310}
311EXPORT_SYMBOL(trace_hardirqs_on);
312
313void trace_hardirqs_off(void)
314{
315 if (!preempt_trace() && irq_trace())
316 start_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
317}
318EXPORT_SYMBOL(trace_hardirqs_off);
319
320void trace_hardirqs_on_caller(unsigned long caller_addr)
321{
322 if (!preempt_trace() && irq_trace())
323 stop_critical_timing(CALLER_ADDR0, caller_addr);
324}
325EXPORT_SYMBOL(trace_hardirqs_on_caller);
326
327void trace_hardirqs_off_caller(unsigned long caller_addr)
328{
329 if (!preempt_trace() && irq_trace())
330 start_critical_timing(CALLER_ADDR0, caller_addr);
331}
332EXPORT_SYMBOL(trace_hardirqs_off_caller);
333
334#endif /* CONFIG_PROVE_LOCKING */
335#endif /* CONFIG_IRQSOFF_TRACER */
336
337#ifdef CONFIG_PREEMPT_TRACER
338void trace_preempt_on(unsigned long a0, unsigned long a1)
339{
340 stop_critical_timing(a0, a1);
341}
342
343void trace_preempt_off(unsigned long a0, unsigned long a1)
344{
345 start_critical_timing(a0, a1);
346}
347#endif /* CONFIG_PREEMPT_TRACER */
348
349static void start_irqsoff_tracer(struct trace_array *tr)
350{
351 register_ftrace_function(&trace_ops);
352 tracer_enabled = 1;
353}
354
355static void stop_irqsoff_tracer(struct trace_array *tr)
356{
357 tracer_enabled = 0;
358 unregister_ftrace_function(&trace_ops);
359}
360
361static void __irqsoff_tracer_init(struct trace_array *tr)
362{
363 irqsoff_trace = tr;
364 /* make sure that the tracer is visible */
365 smp_wmb();
366
367 if (tr->ctrl)
368 start_irqsoff_tracer(tr);
369}
370
371static void irqsoff_tracer_reset(struct trace_array *tr)
372{
373 if (tr->ctrl)
374 stop_irqsoff_tracer(tr);
375}
376
377static void irqsoff_tracer_ctrl_update(struct trace_array *tr)
378{
379 if (tr->ctrl)
380 start_irqsoff_tracer(tr);
381 else
382 stop_irqsoff_tracer(tr);
383}
384
385static void irqsoff_tracer_open(struct trace_iterator *iter)
386{
387 /* stop the trace while dumping */
388 if (iter->tr->ctrl)
389 stop_irqsoff_tracer(iter->tr);
390}
391
392static void irqsoff_tracer_close(struct trace_iterator *iter)
393{
394 if (iter->tr->ctrl)
395 start_irqsoff_tracer(iter->tr);
396}
397
398#ifdef CONFIG_IRQSOFF_TRACER
399static void irqsoff_tracer_init(struct trace_array *tr)
400{
401 trace_type = TRACER_IRQS_OFF;
402
403 __irqsoff_tracer_init(tr);
404}
405static struct tracer irqsoff_tracer __read_mostly =
406{
407 .name = "irqsoff",
408 .init = irqsoff_tracer_init,
409 .reset = irqsoff_tracer_reset,
410 .open = irqsoff_tracer_open,
411 .close = irqsoff_tracer_close,
412 .ctrl_update = irqsoff_tracer_ctrl_update,
413 .print_max = 1,
414#ifdef CONFIG_FTRACE_SELFTEST
415 .selftest = trace_selftest_startup_irqsoff,
416#endif
417};
418# define register_irqsoff(trace) register_tracer(&trace)
419#else
420# define register_irqsoff(trace) do { } while (0)
421#endif
422
423#ifdef CONFIG_PREEMPT_TRACER
424static void preemptoff_tracer_init(struct trace_array *tr)
425{
426 trace_type = TRACER_PREEMPT_OFF;
427
428 __irqsoff_tracer_init(tr);
429}
430
431static struct tracer preemptoff_tracer __read_mostly =
432{
433 .name = "preemptoff",
434 .init = preemptoff_tracer_init,
435 .reset = irqsoff_tracer_reset,
436 .open = irqsoff_tracer_open,
437 .close = irqsoff_tracer_close,
438 .ctrl_update = irqsoff_tracer_ctrl_update,
439 .print_max = 1,
440#ifdef CONFIG_FTRACE_SELFTEST
441 .selftest = trace_selftest_startup_preemptoff,
442#endif
443};
444# define register_preemptoff(trace) register_tracer(&trace)
445#else
446# define register_preemptoff(trace) do { } while (0)
447#endif
448
449#if defined(CONFIG_IRQSOFF_TRACER) && \
450 defined(CONFIG_PREEMPT_TRACER)
451
452static void preemptirqsoff_tracer_init(struct trace_array *tr)
453{
454 trace_type = TRACER_IRQS_OFF | TRACER_PREEMPT_OFF;
455
456 __irqsoff_tracer_init(tr);
457}
458
459static struct tracer preemptirqsoff_tracer __read_mostly =
460{
461 .name = "preemptirqsoff",
462 .init = preemptirqsoff_tracer_init,
463 .reset = irqsoff_tracer_reset,
464 .open = irqsoff_tracer_open,
465 .close = irqsoff_tracer_close,
466 .ctrl_update = irqsoff_tracer_ctrl_update,
467 .print_max = 1,
468#ifdef CONFIG_FTRACE_SELFTEST
469 .selftest = trace_selftest_startup_preemptirqsoff,
470#endif
471};
472
473# define register_preemptirqsoff(trace) register_tracer(&trace)
474#else
475# define register_preemptirqsoff(trace) do { } while (0)
476#endif
477
478__init static int init_irqsoff_tracer(void)
479{
480 register_irqsoff(irqsoff_tracer);
481 register_preemptoff(preemptoff_tracer);
482 register_preemptirqsoff(preemptirqsoff_tracer);
483
484 return 0;
485}
486device_initcall(init_irqsoff_tracer);
diff --git a/kernel/trace/trace_mmiotrace.c b/kernel/trace/trace_mmiotrace.c
new file mode 100644
index 000000000000..b13dc19dcbb4
--- /dev/null
+++ b/kernel/trace/trace_mmiotrace.c
@@ -0,0 +1,295 @@
1/*
2 * Memory mapped I/O tracing
3 *
4 * Copyright (C) 2008 Pekka Paalanen <pq@iki.fi>
5 */
6
7#define DEBUG 1
8
9#include <linux/kernel.h>
10#include <linux/mmiotrace.h>
11#include <linux/pci.h>
12
13#include "trace.h"
14
15struct header_iter {
16 struct pci_dev *dev;
17};
18
19static struct trace_array *mmio_trace_array;
20static bool overrun_detected;
21
22static void mmio_reset_data(struct trace_array *tr)
23{
24 int cpu;
25
26 overrun_detected = false;
27 tr->time_start = ftrace_now(tr->cpu);
28
29 for_each_online_cpu(cpu)
30 tracing_reset(tr->data[cpu]);
31}
32
33static void mmio_trace_init(struct trace_array *tr)
34{
35 pr_debug("in %s\n", __func__);
36 mmio_trace_array = tr;
37 if (tr->ctrl) {
38 mmio_reset_data(tr);
39 enable_mmiotrace();
40 }
41}
42
43static void mmio_trace_reset(struct trace_array *tr)
44{
45 pr_debug("in %s\n", __func__);
46 if (tr->ctrl)
47 disable_mmiotrace();
48 mmio_reset_data(tr);
49 mmio_trace_array = NULL;
50}
51
52static void mmio_trace_ctrl_update(struct trace_array *tr)
53{
54 pr_debug("in %s\n", __func__);
55 if (tr->ctrl) {
56 mmio_reset_data(tr);
57 enable_mmiotrace();
58 } else {
59 disable_mmiotrace();
60 }
61}
62
63static int mmio_print_pcidev(struct trace_seq *s, const struct pci_dev *dev)
64{
65 int ret = 0;
66 int i;
67 resource_size_t start, end;
68 const struct pci_driver *drv = pci_dev_driver(dev);
69
70 /* XXX: incomplete checks for trace_seq_printf() return value */
71 ret += trace_seq_printf(s, "PCIDEV %02x%02x %04x%04x %x",
72 dev->bus->number, dev->devfn,
73 dev->vendor, dev->device, dev->irq);
74 /*
75 * XXX: is pci_resource_to_user() appropriate, since we are
76 * supposed to interpret the __ioremap() phys_addr argument based on
77 * these printed values?
78 */
79 for (i = 0; i < 7; i++) {
80 pci_resource_to_user(dev, i, &dev->resource[i], &start, &end);
81 ret += trace_seq_printf(s, " %llx",
82 (unsigned long long)(start |
83 (dev->resource[i].flags & PCI_REGION_FLAG_MASK)));
84 }
85 for (i = 0; i < 7; i++) {
86 pci_resource_to_user(dev, i, &dev->resource[i], &start, &end);
87 ret += trace_seq_printf(s, " %llx",
88 dev->resource[i].start < dev->resource[i].end ?
89 (unsigned long long)(end - start) + 1 : 0);
90 }
91 if (drv)
92 ret += trace_seq_printf(s, " %s\n", drv->name);
93 else
94 ret += trace_seq_printf(s, " \n");
95 return ret;
96}
97
98static void destroy_header_iter(struct header_iter *hiter)
99{
100 if (!hiter)
101 return;
102 pci_dev_put(hiter->dev);
103 kfree(hiter);
104}
105
106static void mmio_pipe_open(struct trace_iterator *iter)
107{
108 struct header_iter *hiter;
109 struct trace_seq *s = &iter->seq;
110
111 trace_seq_printf(s, "VERSION 20070824\n");
112
113 hiter = kzalloc(sizeof(*hiter), GFP_KERNEL);
114 if (!hiter)
115 return;
116
117 hiter->dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, NULL);
118 iter->private = hiter;
119}
120
121/* XXX: This is not called when the pipe is closed! */
122static void mmio_close(struct trace_iterator *iter)
123{
124 struct header_iter *hiter = iter->private;
125 destroy_header_iter(hiter);
126 iter->private = NULL;
127}
128
129static unsigned long count_overruns(struct trace_iterator *iter)
130{
131 int cpu;
132 unsigned long cnt = 0;
133 for_each_online_cpu(cpu) {
134 cnt += iter->overrun[cpu];
135 iter->overrun[cpu] = 0;
136 }
137 return cnt;
138}
139
140static ssize_t mmio_read(struct trace_iterator *iter, struct file *filp,
141 char __user *ubuf, size_t cnt, loff_t *ppos)
142{
143 ssize_t ret;
144 struct header_iter *hiter = iter->private;
145 struct trace_seq *s = &iter->seq;
146 unsigned long n;
147
148 n = count_overruns(iter);
149 if (n) {
150 /* XXX: This is later than where events were lost. */
151 trace_seq_printf(s, "MARK 0.000000 Lost %lu events.\n", n);
152 if (!overrun_detected)
153 pr_warning("mmiotrace has lost events.\n");
154 overrun_detected = true;
155 goto print_out;
156 }
157
158 if (!hiter)
159 return 0;
160
161 mmio_print_pcidev(s, hiter->dev);
162 hiter->dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, hiter->dev);
163
164 if (!hiter->dev) {
165 destroy_header_iter(hiter);
166 iter->private = NULL;
167 }
168
169print_out:
170 ret = trace_seq_to_user(s, ubuf, cnt);
171 return (ret == -EBUSY) ? 0 : ret;
172}
173
174static int mmio_print_rw(struct trace_iterator *iter)
175{
176 struct trace_entry *entry = iter->ent;
177 struct mmiotrace_rw *rw = &entry->mmiorw;
178 struct trace_seq *s = &iter->seq;
179 unsigned long long t = ns2usecs(entry->t);
180 unsigned long usec_rem = do_div(t, 1000000ULL);
181 unsigned secs = (unsigned long)t;
182 int ret = 1;
183
184 switch (entry->mmiorw.opcode) {
185 case MMIO_READ:
186 ret = trace_seq_printf(s,
187 "R %d %lu.%06lu %d 0x%llx 0x%lx 0x%lx %d\n",
188 rw->width, secs, usec_rem, rw->map_id,
189 (unsigned long long)rw->phys,
190 rw->value, rw->pc, 0);
191 break;
192 case MMIO_WRITE:
193 ret = trace_seq_printf(s,
194 "W %d %lu.%06lu %d 0x%llx 0x%lx 0x%lx %d\n",
195 rw->width, secs, usec_rem, rw->map_id,
196 (unsigned long long)rw->phys,
197 rw->value, rw->pc, 0);
198 break;
199 case MMIO_UNKNOWN_OP:
200 ret = trace_seq_printf(s,
201 "UNKNOWN %lu.%06lu %d 0x%llx %02x,%02x,%02x 0x%lx %d\n",
202 secs, usec_rem, rw->map_id,
203 (unsigned long long)rw->phys,
204 (rw->value >> 16) & 0xff, (rw->value >> 8) & 0xff,
205 (rw->value >> 0) & 0xff, rw->pc, 0);
206 break;
207 default:
208 ret = trace_seq_printf(s, "rw what?\n");
209 break;
210 }
211 if (ret)
212 return 1;
213 return 0;
214}
215
216static int mmio_print_map(struct trace_iterator *iter)
217{
218 struct trace_entry *entry = iter->ent;
219 struct mmiotrace_map *m = &entry->mmiomap;
220 struct trace_seq *s = &iter->seq;
221 unsigned long long t = ns2usecs(entry->t);
222 unsigned long usec_rem = do_div(t, 1000000ULL);
223 unsigned secs = (unsigned long)t;
224 int ret = 1;
225
226 switch (entry->mmiorw.opcode) {
227 case MMIO_PROBE:
228 ret = trace_seq_printf(s,
229 "MAP %lu.%06lu %d 0x%llx 0x%lx 0x%lx 0x%lx %d\n",
230 secs, usec_rem, m->map_id,
231 (unsigned long long)m->phys, m->virt, m->len,
232 0UL, 0);
233 break;
234 case MMIO_UNPROBE:
235 ret = trace_seq_printf(s,
236 "UNMAP %lu.%06lu %d 0x%lx %d\n",
237 secs, usec_rem, m->map_id, 0UL, 0);
238 break;
239 default:
240 ret = trace_seq_printf(s, "map what?\n");
241 break;
242 }
243 if (ret)
244 return 1;
245 return 0;
246}
247
248/* return 0 to abort printing without consuming current entry in pipe mode */
249static int mmio_print_line(struct trace_iterator *iter)
250{
251 switch (iter->ent->type) {
252 case TRACE_MMIO_RW:
253 return mmio_print_rw(iter);
254 case TRACE_MMIO_MAP:
255 return mmio_print_map(iter);
256 default:
257 return 1; /* ignore unknown entries */
258 }
259}
260
261static struct tracer mmio_tracer __read_mostly =
262{
263 .name = "mmiotrace",
264 .init = mmio_trace_init,
265 .reset = mmio_trace_reset,
266 .pipe_open = mmio_pipe_open,
267 .close = mmio_close,
268 .read = mmio_read,
269 .ctrl_update = mmio_trace_ctrl_update,
270 .print_line = mmio_print_line,
271};
272
273__init static int init_mmio_trace(void)
274{
275 return register_tracer(&mmio_tracer);
276}
277device_initcall(init_mmio_trace);
278
279void mmio_trace_rw(struct mmiotrace_rw *rw)
280{
281 struct trace_array *tr = mmio_trace_array;
282 struct trace_array_cpu *data = tr->data[smp_processor_id()];
283 __trace_mmiotrace_rw(tr, data, rw);
284}
285
286void mmio_trace_mapping(struct mmiotrace_map *map)
287{
288 struct trace_array *tr = mmio_trace_array;
289 struct trace_array_cpu *data;
290
291 preempt_disable();
292 data = tr->data[smp_processor_id()];
293 __trace_mmiotrace_map(tr, data, map);
294 preempt_enable();
295}
diff --git a/kernel/trace/trace_sched_switch.c b/kernel/trace/trace_sched_switch.c
new file mode 100644
index 000000000000..cb817a209aa0
--- /dev/null
+++ b/kernel/trace/trace_sched_switch.c
@@ -0,0 +1,286 @@
1/*
2 * trace context switch
3 *
4 * Copyright (C) 2007 Steven Rostedt <srostedt@redhat.com>
5 *
6 */
7#include <linux/module.h>
8#include <linux/fs.h>
9#include <linux/debugfs.h>
10#include <linux/kallsyms.h>
11#include <linux/uaccess.h>
12#include <linux/marker.h>
13#include <linux/ftrace.h>
14
15#include "trace.h"
16
17static struct trace_array *ctx_trace;
18static int __read_mostly tracer_enabled;
19static atomic_t sched_ref;
20
21static void
22sched_switch_func(void *private, void *__rq, struct task_struct *prev,
23 struct task_struct *next)
24{
25 struct trace_array **ptr = private;
26 struct trace_array *tr = *ptr;
27 struct trace_array_cpu *data;
28 unsigned long flags;
29 long disabled;
30 int cpu;
31
32 tracing_record_cmdline(prev);
33 tracing_record_cmdline(next);
34
35 if (!tracer_enabled)
36 return;
37
38 local_irq_save(flags);
39 cpu = raw_smp_processor_id();
40 data = tr->data[cpu];
41 disabled = atomic_inc_return(&data->disabled);
42
43 if (likely(disabled == 1))
44 tracing_sched_switch_trace(tr, data, prev, next, flags);
45
46 atomic_dec(&data->disabled);
47 local_irq_restore(flags);
48}
49
50static notrace void
51sched_switch_callback(void *probe_data, void *call_data,
52 const char *format, va_list *args)
53{
54 struct task_struct *prev;
55 struct task_struct *next;
56 struct rq *__rq;
57
58 if (!atomic_read(&sched_ref))
59 return;
60
61 /* skip prev_pid %d next_pid %d prev_state %ld */
62 (void)va_arg(*args, int);
63 (void)va_arg(*args, int);
64 (void)va_arg(*args, long);
65 __rq = va_arg(*args, typeof(__rq));
66 prev = va_arg(*args, typeof(prev));
67 next = va_arg(*args, typeof(next));
68
69 /*
70 * If tracer_switch_func only points to the local
71 * switch func, it still needs the ptr passed to it.
72 */
73 sched_switch_func(probe_data, __rq, prev, next);
74}
75
76static void
77wakeup_func(void *private, void *__rq, struct task_struct *wakee, struct
78 task_struct *curr)
79{
80 struct trace_array **ptr = private;
81 struct trace_array *tr = *ptr;
82 struct trace_array_cpu *data;
83 unsigned long flags;
84 long disabled;
85 int cpu;
86
87 if (!tracer_enabled)
88 return;
89
90 tracing_record_cmdline(curr);
91
92 local_irq_save(flags);
93 cpu = raw_smp_processor_id();
94 data = tr->data[cpu];
95 disabled = atomic_inc_return(&data->disabled);
96
97 if (likely(disabled == 1))
98 tracing_sched_wakeup_trace(tr, data, wakee, curr, flags);
99
100 atomic_dec(&data->disabled);
101 local_irq_restore(flags);
102}
103
104static notrace void
105wake_up_callback(void *probe_data, void *call_data,
106 const char *format, va_list *args)
107{
108 struct task_struct *curr;
109 struct task_struct *task;
110 struct rq *__rq;
111
112 if (likely(!tracer_enabled))
113 return;
114
115 /* Skip pid %d state %ld */
116 (void)va_arg(*args, int);
117 (void)va_arg(*args, long);
118 /* now get the meat: "rq %p task %p rq->curr %p" */
119 __rq = va_arg(*args, typeof(__rq));
120 task = va_arg(*args, typeof(task));
121 curr = va_arg(*args, typeof(curr));
122
123 tracing_record_cmdline(task);
124 tracing_record_cmdline(curr);
125
126 wakeup_func(probe_data, __rq, task, curr);
127}
128
129static void sched_switch_reset(struct trace_array *tr)
130{
131 int cpu;
132
133 tr->time_start = ftrace_now(tr->cpu);
134
135 for_each_online_cpu(cpu)
136 tracing_reset(tr->data[cpu]);
137}
138
139static int tracing_sched_register(void)
140{
141 int ret;
142
143 ret = marker_probe_register("kernel_sched_wakeup",
144 "pid %d state %ld ## rq %p task %p rq->curr %p",
145 wake_up_callback,
146 &ctx_trace);
147 if (ret) {
148 pr_info("wakeup trace: Couldn't add marker"
149 " probe to kernel_sched_wakeup\n");
150 return ret;
151 }
152
153 ret = marker_probe_register("kernel_sched_wakeup_new",
154 "pid %d state %ld ## rq %p task %p rq->curr %p",
155 wake_up_callback,
156 &ctx_trace);
157 if (ret) {
158 pr_info("wakeup trace: Couldn't add marker"
159 " probe to kernel_sched_wakeup_new\n");
160 goto fail_deprobe;
161 }
162
163 ret = marker_probe_register("kernel_sched_schedule",
164 "prev_pid %d next_pid %d prev_state %ld "
165 "## rq %p prev %p next %p",
166 sched_switch_callback,
167 &ctx_trace);
168 if (ret) {
169 pr_info("sched trace: Couldn't add marker"
170 " probe to kernel_sched_schedule\n");
171 goto fail_deprobe_wake_new;
172 }
173
174 return ret;
175fail_deprobe_wake_new:
176 marker_probe_unregister("kernel_sched_wakeup_new",
177 wake_up_callback,
178 &ctx_trace);
179fail_deprobe:
180 marker_probe_unregister("kernel_sched_wakeup",
181 wake_up_callback,
182 &ctx_trace);
183 return ret;
184}
185
186static void tracing_sched_unregister(void)
187{
188 marker_probe_unregister("kernel_sched_schedule",
189 sched_switch_callback,
190 &ctx_trace);
191 marker_probe_unregister("kernel_sched_wakeup_new",
192 wake_up_callback,
193 &ctx_trace);
194 marker_probe_unregister("kernel_sched_wakeup",
195 wake_up_callback,
196 &ctx_trace);
197}
198
199static void tracing_start_sched_switch(void)
200{
201 long ref;
202
203 ref = atomic_inc_return(&sched_ref);
204 if (ref == 1)
205 tracing_sched_register();
206}
207
208static void tracing_stop_sched_switch(void)
209{
210 long ref;
211
212 ref = atomic_dec_and_test(&sched_ref);
213 if (ref)
214 tracing_sched_unregister();
215}
216
217void tracing_start_cmdline_record(void)
218{
219 tracing_start_sched_switch();
220}
221
222void tracing_stop_cmdline_record(void)
223{
224 tracing_stop_sched_switch();
225}
226
227static void start_sched_trace(struct trace_array *tr)
228{
229 sched_switch_reset(tr);
230 tracing_start_cmdline_record();
231 tracer_enabled = 1;
232}
233
234static void stop_sched_trace(struct trace_array *tr)
235{
236 tracer_enabled = 0;
237 tracing_stop_cmdline_record();
238}
239
240static void sched_switch_trace_init(struct trace_array *tr)
241{
242 ctx_trace = tr;
243
244 if (tr->ctrl)
245 start_sched_trace(tr);
246}
247
248static void sched_switch_trace_reset(struct trace_array *tr)
249{
250 if (tr->ctrl)
251 stop_sched_trace(tr);
252}
253
254static void sched_switch_trace_ctrl_update(struct trace_array *tr)
255{
256 /* When starting a new trace, reset the buffers */
257 if (tr->ctrl)
258 start_sched_trace(tr);
259 else
260 stop_sched_trace(tr);
261}
262
263static struct tracer sched_switch_trace __read_mostly =
264{
265 .name = "sched_switch",
266 .init = sched_switch_trace_init,
267 .reset = sched_switch_trace_reset,
268 .ctrl_update = sched_switch_trace_ctrl_update,
269#ifdef CONFIG_FTRACE_SELFTEST
270 .selftest = trace_selftest_startup_sched_switch,
271#endif
272};
273
274__init static int init_sched_switch_trace(void)
275{
276 int ret = 0;
277
278 if (atomic_read(&sched_ref))
279 ret = tracing_sched_register();
280 if (ret) {
281 pr_info("error registering scheduler trace\n");
282 return ret;
283 }
284 return register_tracer(&sched_switch_trace);
285}
286device_initcall(init_sched_switch_trace);
diff --git a/kernel/trace/trace_sched_wakeup.c b/kernel/trace/trace_sched_wakeup.c
new file mode 100644
index 000000000000..3c8d61df4474
--- /dev/null
+++ b/kernel/trace/trace_sched_wakeup.c
@@ -0,0 +1,448 @@
1/*
2 * trace task wakeup timings
3 *
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
6 *
7 * Based on code from the latency_tracer, that is:
8 *
9 * Copyright (C) 2004-2006 Ingo Molnar
10 * Copyright (C) 2004 William Lee Irwin III
11 */
12#include <linux/module.h>
13#include <linux/fs.h>
14#include <linux/debugfs.h>
15#include <linux/kallsyms.h>
16#include <linux/uaccess.h>
17#include <linux/ftrace.h>
18#include <linux/marker.h>
19
20#include "trace.h"
21
22static struct trace_array *wakeup_trace;
23static int __read_mostly tracer_enabled;
24
25static struct task_struct *wakeup_task;
26static int wakeup_cpu;
27static unsigned wakeup_prio = -1;
28
29static DEFINE_SPINLOCK(wakeup_lock);
30
31static void __wakeup_reset(struct trace_array *tr);
32
33#ifdef CONFIG_FTRACE
34/*
35 * irqsoff uses its own tracer function to keep the overhead down:
36 */
37static void
38wakeup_tracer_call(unsigned long ip, unsigned long parent_ip)
39{
40 struct trace_array *tr = wakeup_trace;
41 struct trace_array_cpu *data;
42 unsigned long flags;
43 long disabled;
44 int resched;
45 int cpu;
46
47 if (likely(!wakeup_task))
48 return;
49
50 resched = need_resched();
51 preempt_disable_notrace();
52
53 cpu = raw_smp_processor_id();
54 data = tr->data[cpu];
55 disabled = atomic_inc_return(&data->disabled);
56 if (unlikely(disabled != 1))
57 goto out;
58
59 spin_lock_irqsave(&wakeup_lock, flags);
60
61 if (unlikely(!wakeup_task))
62 goto unlock;
63
64 /*
65 * The task can't disappear because it needs to
66 * wake up first, and we have the wakeup_lock.
67 */
68 if (task_cpu(wakeup_task) != cpu)
69 goto unlock;
70
71 trace_function(tr, data, ip, parent_ip, flags);
72
73 unlock:
74 spin_unlock_irqrestore(&wakeup_lock, flags);
75
76 out:
77 atomic_dec(&data->disabled);
78
79 /*
80 * To prevent recursion from the scheduler, if the
81 * resched flag was set before we entered, then
82 * don't reschedule.
83 */
84 if (resched)
85 preempt_enable_no_resched_notrace();
86 else
87 preempt_enable_notrace();
88}
89
90static struct ftrace_ops trace_ops __read_mostly =
91{
92 .func = wakeup_tracer_call,
93};
94#endif /* CONFIG_FTRACE */
95
96/*
97 * Should this new latency be reported/recorded?
98 */
99static int report_latency(cycle_t delta)
100{
101 if (tracing_thresh) {
102 if (delta < tracing_thresh)
103 return 0;
104 } else {
105 if (delta <= tracing_max_latency)
106 return 0;
107 }
108 return 1;
109}
110
111static void notrace
112wakeup_sched_switch(void *private, void *rq, struct task_struct *prev,
113 struct task_struct *next)
114{
115 unsigned long latency = 0, t0 = 0, t1 = 0;
116 struct trace_array **ptr = private;
117 struct trace_array *tr = *ptr;
118 struct trace_array_cpu *data;
119 cycle_t T0, T1, delta;
120 unsigned long flags;
121 long disabled;
122 int cpu;
123
124 if (unlikely(!tracer_enabled))
125 return;
126
127 /*
128 * When we start a new trace, we set wakeup_task to NULL
129 * and then set tracer_enabled = 1. We want to make sure
130 * that another CPU does not see the tracer_enabled = 1
131 * and the wakeup_task with an older task, that might
132 * actually be the same as next.
133 */
134 smp_rmb();
135
136 if (next != wakeup_task)
137 return;
138
139 /* The task we are waiting for is waking up */
140 data = tr->data[wakeup_cpu];
141
142 /* disable local data, not wakeup_cpu data */
143 cpu = raw_smp_processor_id();
144 disabled = atomic_inc_return(&tr->data[cpu]->disabled);
145 if (likely(disabled != 1))
146 goto out;
147
148 spin_lock_irqsave(&wakeup_lock, flags);
149
150 /* We could race with grabbing wakeup_lock */
151 if (unlikely(!tracer_enabled || next != wakeup_task))
152 goto out_unlock;
153
154 trace_function(tr, data, CALLER_ADDR1, CALLER_ADDR2, flags);
155
156 /*
157 * usecs conversion is slow so we try to delay the conversion
158 * as long as possible:
159 */
160 T0 = data->preempt_timestamp;
161 T1 = ftrace_now(cpu);
162 delta = T1-T0;
163
164 if (!report_latency(delta))
165 goto out_unlock;
166
167 latency = nsecs_to_usecs(delta);
168
169 tracing_max_latency = delta;
170 t0 = nsecs_to_usecs(T0);
171 t1 = nsecs_to_usecs(T1);
172
173 update_max_tr(tr, wakeup_task, wakeup_cpu);
174
175out_unlock:
176 __wakeup_reset(tr);
177 spin_unlock_irqrestore(&wakeup_lock, flags);
178out:
179 atomic_dec(&tr->data[cpu]->disabled);
180}
181
182static notrace void
183sched_switch_callback(void *probe_data, void *call_data,
184 const char *format, va_list *args)
185{
186 struct task_struct *prev;
187 struct task_struct *next;
188 struct rq *__rq;
189
190 /* skip prev_pid %d next_pid %d prev_state %ld */
191 (void)va_arg(*args, int);
192 (void)va_arg(*args, int);
193 (void)va_arg(*args, long);
194 __rq = va_arg(*args, typeof(__rq));
195 prev = va_arg(*args, typeof(prev));
196 next = va_arg(*args, typeof(next));
197
198 tracing_record_cmdline(prev);
199
200 /*
201 * If tracer_switch_func only points to the local
202 * switch func, it still needs the ptr passed to it.
203 */
204 wakeup_sched_switch(probe_data, __rq, prev, next);
205}
206
207static void __wakeup_reset(struct trace_array *tr)
208{
209 struct trace_array_cpu *data;
210 int cpu;
211
212 assert_spin_locked(&wakeup_lock);
213
214 for_each_possible_cpu(cpu) {
215 data = tr->data[cpu];
216 tracing_reset(data);
217 }
218
219 wakeup_cpu = -1;
220 wakeup_prio = -1;
221
222 if (wakeup_task)
223 put_task_struct(wakeup_task);
224
225 wakeup_task = NULL;
226}
227
228static void wakeup_reset(struct trace_array *tr)
229{
230 unsigned long flags;
231
232 spin_lock_irqsave(&wakeup_lock, flags);
233 __wakeup_reset(tr);
234 spin_unlock_irqrestore(&wakeup_lock, flags);
235}
236
237static void
238wakeup_check_start(struct trace_array *tr, struct task_struct *p,
239 struct task_struct *curr)
240{
241 int cpu = smp_processor_id();
242 unsigned long flags;
243 long disabled;
244
245 if (likely(!rt_task(p)) ||
246 p->prio >= wakeup_prio ||
247 p->prio >= curr->prio)
248 return;
249
250 disabled = atomic_inc_return(&tr->data[cpu]->disabled);
251 if (unlikely(disabled != 1))
252 goto out;
253
254 /* interrupts should be off from try_to_wake_up */
255 spin_lock(&wakeup_lock);
256
257 /* check for races. */
258 if (!tracer_enabled || p->prio >= wakeup_prio)
259 goto out_locked;
260
261 /* reset the trace */
262 __wakeup_reset(tr);
263
264 wakeup_cpu = task_cpu(p);
265 wakeup_prio = p->prio;
266
267 wakeup_task = p;
268 get_task_struct(wakeup_task);
269
270 local_save_flags(flags);
271
272 tr->data[wakeup_cpu]->preempt_timestamp = ftrace_now(cpu);
273 trace_function(tr, tr->data[wakeup_cpu],
274 CALLER_ADDR1, CALLER_ADDR2, flags);
275
276out_locked:
277 spin_unlock(&wakeup_lock);
278out:
279 atomic_dec(&tr->data[cpu]->disabled);
280}
281
282static notrace void
283wake_up_callback(void *probe_data, void *call_data,
284 const char *format, va_list *args)
285{
286 struct trace_array **ptr = probe_data;
287 struct trace_array *tr = *ptr;
288 struct task_struct *curr;
289 struct task_struct *task;
290 struct rq *__rq;
291
292 if (likely(!tracer_enabled))
293 return;
294
295 /* Skip pid %d state %ld */
296 (void)va_arg(*args, int);
297 (void)va_arg(*args, long);
298 /* now get the meat: "rq %p task %p rq->curr %p" */
299 __rq = va_arg(*args, typeof(__rq));
300 task = va_arg(*args, typeof(task));
301 curr = va_arg(*args, typeof(curr));
302
303 tracing_record_cmdline(task);
304 tracing_record_cmdline(curr);
305
306 wakeup_check_start(tr, task, curr);
307}
308
309static void start_wakeup_tracer(struct trace_array *tr)
310{
311 int ret;
312
313 ret = marker_probe_register("kernel_sched_wakeup",
314 "pid %d state %ld ## rq %p task %p rq->curr %p",
315 wake_up_callback,
316 &wakeup_trace);
317 if (ret) {
318 pr_info("wakeup trace: Couldn't add marker"
319 " probe to kernel_sched_wakeup\n");
320 return;
321 }
322
323 ret = marker_probe_register("kernel_sched_wakeup_new",
324 "pid %d state %ld ## rq %p task %p rq->curr %p",
325 wake_up_callback,
326 &wakeup_trace);
327 if (ret) {
328 pr_info("wakeup trace: Couldn't add marker"
329 " probe to kernel_sched_wakeup_new\n");
330 goto fail_deprobe;
331 }
332
333 ret = marker_probe_register("kernel_sched_schedule",
334 "prev_pid %d next_pid %d prev_state %ld "
335 "## rq %p prev %p next %p",
336 sched_switch_callback,
337 &wakeup_trace);
338 if (ret) {
339 pr_info("sched trace: Couldn't add marker"
340 " probe to kernel_sched_schedule\n");
341 goto fail_deprobe_wake_new;
342 }
343
344 wakeup_reset(tr);
345
346 /*
347 * Don't let the tracer_enabled = 1 show up before
348 * the wakeup_task is reset. This may be overkill since
349 * wakeup_reset does a spin_unlock after setting the
350 * wakeup_task to NULL, but I want to be safe.
351 * This is a slow path anyway.
352 */
353 smp_wmb();
354
355 register_ftrace_function(&trace_ops);
356
357 tracer_enabled = 1;
358
359 return;
360fail_deprobe_wake_new:
361 marker_probe_unregister("kernel_sched_wakeup_new",
362 wake_up_callback,
363 &wakeup_trace);
364fail_deprobe:
365 marker_probe_unregister("kernel_sched_wakeup",
366 wake_up_callback,
367 &wakeup_trace);
368}
369
370static void stop_wakeup_tracer(struct trace_array *tr)
371{
372 tracer_enabled = 0;
373 unregister_ftrace_function(&trace_ops);
374 marker_probe_unregister("kernel_sched_schedule",
375 sched_switch_callback,
376 &wakeup_trace);
377 marker_probe_unregister("kernel_sched_wakeup_new",
378 wake_up_callback,
379 &wakeup_trace);
380 marker_probe_unregister("kernel_sched_wakeup",
381 wake_up_callback,
382 &wakeup_trace);
383}
384
385static void wakeup_tracer_init(struct trace_array *tr)
386{
387 wakeup_trace = tr;
388
389 if (tr->ctrl)
390 start_wakeup_tracer(tr);
391}
392
393static void wakeup_tracer_reset(struct trace_array *tr)
394{
395 if (tr->ctrl) {
396 stop_wakeup_tracer(tr);
397 /* make sure we put back any tasks we are tracing */
398 wakeup_reset(tr);
399 }
400}
401
402static void wakeup_tracer_ctrl_update(struct trace_array *tr)
403{
404 if (tr->ctrl)
405 start_wakeup_tracer(tr);
406 else
407 stop_wakeup_tracer(tr);
408}
409
410static void wakeup_tracer_open(struct trace_iterator *iter)
411{
412 /* stop the trace while dumping */
413 if (iter->tr->ctrl)
414 stop_wakeup_tracer(iter->tr);
415}
416
417static void wakeup_tracer_close(struct trace_iterator *iter)
418{
419 /* forget about any processes we were recording */
420 if (iter->tr->ctrl)
421 start_wakeup_tracer(iter->tr);
422}
423
424static struct tracer wakeup_tracer __read_mostly =
425{
426 .name = "wakeup",
427 .init = wakeup_tracer_init,
428 .reset = wakeup_tracer_reset,
429 .open = wakeup_tracer_open,
430 .close = wakeup_tracer_close,
431 .ctrl_update = wakeup_tracer_ctrl_update,
432 .print_max = 1,
433#ifdef CONFIG_FTRACE_SELFTEST
434 .selftest = trace_selftest_startup_wakeup,
435#endif
436};
437
438__init static int init_wakeup_tracer(void)
439{
440 int ret;
441
442 ret = register_tracer(&wakeup_tracer);
443 if (ret)
444 return ret;
445
446 return 0;
447}
448device_initcall(init_wakeup_tracer);
diff --git a/kernel/trace/trace_selftest.c b/kernel/trace/trace_selftest.c
new file mode 100644
index 000000000000..0911b7e073bf
--- /dev/null
+++ b/kernel/trace/trace_selftest.c
@@ -0,0 +1,563 @@
1/* Include in trace.c */
2
3#include <linux/kthread.h>
4#include <linux/delay.h>
5
6static inline int trace_valid_entry(struct trace_entry *entry)
7{
8 switch (entry->type) {
9 case TRACE_FN:
10 case TRACE_CTX:
11 case TRACE_WAKE:
12 case TRACE_STACK:
13 case TRACE_SPECIAL:
14 return 1;
15 }
16 return 0;
17}
18
19static int
20trace_test_buffer_cpu(struct trace_array *tr, struct trace_array_cpu *data)
21{
22 struct trace_entry *entries;
23 struct page *page;
24 int idx = 0;
25 int i;
26
27 BUG_ON(list_empty(&data->trace_pages));
28 page = list_entry(data->trace_pages.next, struct page, lru);
29 entries = page_address(page);
30
31 check_pages(data);
32 if (head_page(data) != entries)
33 goto failed;
34
35 /*
36 * The starting trace buffer always has valid elements,
37 * if any element exists.
38 */
39 entries = head_page(data);
40
41 for (i = 0; i < tr->entries; i++) {
42
43 if (i < data->trace_idx && !trace_valid_entry(&entries[idx])) {
44 printk(KERN_CONT ".. invalid entry %d ",
45 entries[idx].type);
46 goto failed;
47 }
48
49 idx++;
50 if (idx >= ENTRIES_PER_PAGE) {
51 page = virt_to_page(entries);
52 if (page->lru.next == &data->trace_pages) {
53 if (i != tr->entries - 1) {
54 printk(KERN_CONT ".. entries buffer mismatch");
55 goto failed;
56 }
57 } else {
58 page = list_entry(page->lru.next, struct page, lru);
59 entries = page_address(page);
60 }
61 idx = 0;
62 }
63 }
64
65 page = virt_to_page(entries);
66 if (page->lru.next != &data->trace_pages) {
67 printk(KERN_CONT ".. too many entries");
68 goto failed;
69 }
70
71 return 0;
72
73 failed:
74 /* disable tracing */
75 tracing_disabled = 1;
76 printk(KERN_CONT ".. corrupted trace buffer .. ");
77 return -1;
78}
79
80/*
81 * Test the trace buffer to see if all the elements
82 * are still sane.
83 */
84static int trace_test_buffer(struct trace_array *tr, unsigned long *count)
85{
86 unsigned long flags, cnt = 0;
87 int cpu, ret = 0;
88
89 /* Don't allow flipping of max traces now */
90 raw_local_irq_save(flags);
91 __raw_spin_lock(&ftrace_max_lock);
92 for_each_possible_cpu(cpu) {
93 if (!head_page(tr->data[cpu]))
94 continue;
95
96 cnt += tr->data[cpu]->trace_idx;
97
98 ret = trace_test_buffer_cpu(tr, tr->data[cpu]);
99 if (ret)
100 break;
101 }
102 __raw_spin_unlock(&ftrace_max_lock);
103 raw_local_irq_restore(flags);
104
105 if (count)
106 *count = cnt;
107
108 return ret;
109}
110
111#ifdef CONFIG_FTRACE
112
113#ifdef CONFIG_DYNAMIC_FTRACE
114
115#define __STR(x) #x
116#define STR(x) __STR(x)
117
118/* Test dynamic code modification and ftrace filters */
119int trace_selftest_startup_dynamic_tracing(struct tracer *trace,
120 struct trace_array *tr,
121 int (*func)(void))
122{
123 unsigned long count;
124 int ret;
125 int save_ftrace_enabled = ftrace_enabled;
126 int save_tracer_enabled = tracer_enabled;
127 char *func_name;
128
129 /* The ftrace test PASSED */
130 printk(KERN_CONT "PASSED\n");
131 pr_info("Testing dynamic ftrace: ");
132
133 /* enable tracing, and record the filter function */
134 ftrace_enabled = 1;
135 tracer_enabled = 1;
136
137 /* passed in by parameter to fool gcc from optimizing */
138 func();
139
140 /* update the records */
141 ret = ftrace_force_update();
142 if (ret) {
143 printk(KERN_CONT ".. ftraced failed .. ");
144 return ret;
145 }
146
147 /*
148 * Some archs *cough*PowerPC*cough* add charachters to the
149 * start of the function names. We simply put a '*' to
150 * accomodate them.
151 */
152 func_name = "*" STR(DYN_FTRACE_TEST_NAME);
153
154 /* filter only on our function */
155 ftrace_set_filter(func_name, strlen(func_name), 1);
156
157 /* enable tracing */
158 tr->ctrl = 1;
159 trace->init(tr);
160 /* Sleep for a 1/10 of a second */
161 msleep(100);
162
163 /* we should have nothing in the buffer */
164 ret = trace_test_buffer(tr, &count);
165 if (ret)
166 goto out;
167
168 if (count) {
169 ret = -1;
170 printk(KERN_CONT ".. filter did not filter .. ");
171 goto out;
172 }
173
174 /* call our function again */
175 func();
176
177 /* sleep again */
178 msleep(100);
179
180 /* stop the tracing. */
181 tr->ctrl = 0;
182 trace->ctrl_update(tr);
183 ftrace_enabled = 0;
184
185 /* check the trace buffer */
186 ret = trace_test_buffer(tr, &count);
187 trace->reset(tr);
188
189 /* we should only have one item */
190 if (!ret && count != 1) {
191 printk(KERN_CONT ".. filter failed count=%ld ..", count);
192 ret = -1;
193 goto out;
194 }
195 out:
196 ftrace_enabled = save_ftrace_enabled;
197 tracer_enabled = save_tracer_enabled;
198
199 /* Enable tracing on all functions again */
200 ftrace_set_filter(NULL, 0, 1);
201
202 return ret;
203}
204#else
205# define trace_selftest_startup_dynamic_tracing(trace, tr, func) ({ 0; })
206#endif /* CONFIG_DYNAMIC_FTRACE */
207/*
208 * Simple verification test of ftrace function tracer.
209 * Enable ftrace, sleep 1/10 second, and then read the trace
210 * buffer to see if all is in order.
211 */
212int
213trace_selftest_startup_function(struct tracer *trace, struct trace_array *tr)
214{
215 unsigned long count;
216 int ret;
217 int save_ftrace_enabled = ftrace_enabled;
218 int save_tracer_enabled = tracer_enabled;
219
220 /* make sure msleep has been recorded */
221 msleep(1);
222
223 /* force the recorded functions to be traced */
224 ret = ftrace_force_update();
225 if (ret) {
226 printk(KERN_CONT ".. ftraced failed .. ");
227 return ret;
228 }
229
230 /* start the tracing */
231 ftrace_enabled = 1;
232 tracer_enabled = 1;
233
234 tr->ctrl = 1;
235 trace->init(tr);
236 /* Sleep for a 1/10 of a second */
237 msleep(100);
238 /* stop the tracing. */
239 tr->ctrl = 0;
240 trace->ctrl_update(tr);
241 ftrace_enabled = 0;
242
243 /* check the trace buffer */
244 ret = trace_test_buffer(tr, &count);
245 trace->reset(tr);
246
247 if (!ret && !count) {
248 printk(KERN_CONT ".. no entries found ..");
249 ret = -1;
250 goto out;
251 }
252
253 ret = trace_selftest_startup_dynamic_tracing(trace, tr,
254 DYN_FTRACE_TEST_NAME);
255
256 out:
257 ftrace_enabled = save_ftrace_enabled;
258 tracer_enabled = save_tracer_enabled;
259
260 /* kill ftrace totally if we failed */
261 if (ret)
262 ftrace_kill();
263
264 return ret;
265}
266#endif /* CONFIG_FTRACE */
267
268#ifdef CONFIG_IRQSOFF_TRACER
269int
270trace_selftest_startup_irqsoff(struct tracer *trace, struct trace_array *tr)
271{
272 unsigned long save_max = tracing_max_latency;
273 unsigned long count;
274 int ret;
275
276 /* start the tracing */
277 tr->ctrl = 1;
278 trace->init(tr);
279 /* reset the max latency */
280 tracing_max_latency = 0;
281 /* disable interrupts for a bit */
282 local_irq_disable();
283 udelay(100);
284 local_irq_enable();
285 /* stop the tracing. */
286 tr->ctrl = 0;
287 trace->ctrl_update(tr);
288 /* check both trace buffers */
289 ret = trace_test_buffer(tr, NULL);
290 if (!ret)
291 ret = trace_test_buffer(&max_tr, &count);
292 trace->reset(tr);
293
294 if (!ret && !count) {
295 printk(KERN_CONT ".. no entries found ..");
296 ret = -1;
297 }
298
299 tracing_max_latency = save_max;
300
301 return ret;
302}
303#endif /* CONFIG_IRQSOFF_TRACER */
304
305#ifdef CONFIG_PREEMPT_TRACER
306int
307trace_selftest_startup_preemptoff(struct tracer *trace, struct trace_array *tr)
308{
309 unsigned long save_max = tracing_max_latency;
310 unsigned long count;
311 int ret;
312
313 /* start the tracing */
314 tr->ctrl = 1;
315 trace->init(tr);
316 /* reset the max latency */
317 tracing_max_latency = 0;
318 /* disable preemption for a bit */
319 preempt_disable();
320 udelay(100);
321 preempt_enable();
322 /* stop the tracing. */
323 tr->ctrl = 0;
324 trace->ctrl_update(tr);
325 /* check both trace buffers */
326 ret = trace_test_buffer(tr, NULL);
327 if (!ret)
328 ret = trace_test_buffer(&max_tr, &count);
329 trace->reset(tr);
330
331 if (!ret && !count) {
332 printk(KERN_CONT ".. no entries found ..");
333 ret = -1;
334 }
335
336 tracing_max_latency = save_max;
337
338 return ret;
339}
340#endif /* CONFIG_PREEMPT_TRACER */
341
342#if defined(CONFIG_IRQSOFF_TRACER) && defined(CONFIG_PREEMPT_TRACER)
343int
344trace_selftest_startup_preemptirqsoff(struct tracer *trace, struct trace_array *tr)
345{
346 unsigned long save_max = tracing_max_latency;
347 unsigned long count;
348 int ret;
349
350 /* start the tracing */
351 tr->ctrl = 1;
352 trace->init(tr);
353
354 /* reset the max latency */
355 tracing_max_latency = 0;
356
357 /* disable preemption and interrupts for a bit */
358 preempt_disable();
359 local_irq_disable();
360 udelay(100);
361 preempt_enable();
362 /* reverse the order of preempt vs irqs */
363 local_irq_enable();
364
365 /* stop the tracing. */
366 tr->ctrl = 0;
367 trace->ctrl_update(tr);
368 /* check both trace buffers */
369 ret = trace_test_buffer(tr, NULL);
370 if (ret)
371 goto out;
372
373 ret = trace_test_buffer(&max_tr, &count);
374 if (ret)
375 goto out;
376
377 if (!ret && !count) {
378 printk(KERN_CONT ".. no entries found ..");
379 ret = -1;
380 goto out;
381 }
382
383 /* do the test by disabling interrupts first this time */
384 tracing_max_latency = 0;
385 tr->ctrl = 1;
386 trace->ctrl_update(tr);
387 preempt_disable();
388 local_irq_disable();
389 udelay(100);
390 preempt_enable();
391 /* reverse the order of preempt vs irqs */
392 local_irq_enable();
393
394 /* stop the tracing. */
395 tr->ctrl = 0;
396 trace->ctrl_update(tr);
397 /* check both trace buffers */
398 ret = trace_test_buffer(tr, NULL);
399 if (ret)
400 goto out;
401
402 ret = trace_test_buffer(&max_tr, &count);
403
404 if (!ret && !count) {
405 printk(KERN_CONT ".. no entries found ..");
406 ret = -1;
407 goto out;
408 }
409
410 out:
411 trace->reset(tr);
412 tracing_max_latency = save_max;
413
414 return ret;
415}
416#endif /* CONFIG_IRQSOFF_TRACER && CONFIG_PREEMPT_TRACER */
417
418#ifdef CONFIG_SCHED_TRACER
419static int trace_wakeup_test_thread(void *data)
420{
421 /* Make this a RT thread, doesn't need to be too high */
422 struct sched_param param = { .sched_priority = 5 };
423 struct completion *x = data;
424
425 sched_setscheduler(current, SCHED_FIFO, &param);
426
427 /* Make it know we have a new prio */
428 complete(x);
429
430 /* now go to sleep and let the test wake us up */
431 set_current_state(TASK_INTERRUPTIBLE);
432 schedule();
433
434 /* we are awake, now wait to disappear */
435 while (!kthread_should_stop()) {
436 /*
437 * This is an RT task, do short sleeps to let
438 * others run.
439 */
440 msleep(100);
441 }
442
443 return 0;
444}
445
446int
447trace_selftest_startup_wakeup(struct tracer *trace, struct trace_array *tr)
448{
449 unsigned long save_max = tracing_max_latency;
450 struct task_struct *p;
451 struct completion isrt;
452 unsigned long count;
453 int ret;
454
455 init_completion(&isrt);
456
457 /* create a high prio thread */
458 p = kthread_run(trace_wakeup_test_thread, &isrt, "ftrace-test");
459 if (IS_ERR(p)) {
460 printk(KERN_CONT "Failed to create ftrace wakeup test thread ");
461 return -1;
462 }
463
464 /* make sure the thread is running at an RT prio */
465 wait_for_completion(&isrt);
466
467 /* start the tracing */
468 tr->ctrl = 1;
469 trace->init(tr);
470 /* reset the max latency */
471 tracing_max_latency = 0;
472
473 /* sleep to let the RT thread sleep too */
474 msleep(100);
475
476 /*
477 * Yes this is slightly racy. It is possible that for some
478 * strange reason that the RT thread we created, did not
479 * call schedule for 100ms after doing the completion,
480 * and we do a wakeup on a task that already is awake.
481 * But that is extremely unlikely, and the worst thing that
482 * happens in such a case, is that we disable tracing.
483 * Honestly, if this race does happen something is horrible
484 * wrong with the system.
485 */
486
487 wake_up_process(p);
488
489 /* stop the tracing. */
490 tr->ctrl = 0;
491 trace->ctrl_update(tr);
492 /* check both trace buffers */
493 ret = trace_test_buffer(tr, NULL);
494 if (!ret)
495 ret = trace_test_buffer(&max_tr, &count);
496
497
498 trace->reset(tr);
499
500 tracing_max_latency = save_max;
501
502 /* kill the thread */
503 kthread_stop(p);
504
505 if (!ret && !count) {
506 printk(KERN_CONT ".. no entries found ..");
507 ret = -1;
508 }
509
510 return ret;
511}
512#endif /* CONFIG_SCHED_TRACER */
513
514#ifdef CONFIG_CONTEXT_SWITCH_TRACER
515int
516trace_selftest_startup_sched_switch(struct tracer *trace, struct trace_array *tr)
517{
518 unsigned long count;
519 int ret;
520
521 /* start the tracing */
522 tr->ctrl = 1;
523 trace->init(tr);
524 /* Sleep for a 1/10 of a second */
525 msleep(100);
526 /* stop the tracing. */
527 tr->ctrl = 0;
528 trace->ctrl_update(tr);
529 /* check the trace buffer */
530 ret = trace_test_buffer(tr, &count);
531 trace->reset(tr);
532
533 if (!ret && !count) {
534 printk(KERN_CONT ".. no entries found ..");
535 ret = -1;
536 }
537
538 return ret;
539}
540#endif /* CONFIG_CONTEXT_SWITCH_TRACER */
541
542#ifdef CONFIG_SYSPROF_TRACER
543int
544trace_selftest_startup_sysprof(struct tracer *trace, struct trace_array *tr)
545{
546 unsigned long count;
547 int ret;
548
549 /* start the tracing */
550 tr->ctrl = 1;
551 trace->init(tr);
552 /* Sleep for a 1/10 of a second */
553 msleep(100);
554 /* stop the tracing. */
555 tr->ctrl = 0;
556 trace->ctrl_update(tr);
557 /* check the trace buffer */
558 ret = trace_test_buffer(tr, &count);
559 trace->reset(tr);
560
561 return ret;
562}
563#endif /* CONFIG_SYSPROF_TRACER */
diff --git a/kernel/trace/trace_selftest_dynamic.c b/kernel/trace/trace_selftest_dynamic.c
new file mode 100644
index 000000000000..54dd77cce5bf
--- /dev/null
+++ b/kernel/trace/trace_selftest_dynamic.c
@@ -0,0 +1,7 @@
1#include "trace.h"
2
3int DYN_FTRACE_TEST_NAME(void)
4{
5 /* used to call mcount */
6 return 0;
7}
diff --git a/kernel/trace/trace_sysprof.c b/kernel/trace/trace_sysprof.c
new file mode 100644
index 000000000000..2301e1e7c606
--- /dev/null
+++ b/kernel/trace/trace_sysprof.c
@@ -0,0 +1,363 @@
1/*
2 * trace stack traces
3 *
4 * Copyright (C) 2004-2008, Soeren Sandmann
5 * Copyright (C) 2007 Steven Rostedt <srostedt@redhat.com>
6 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 */
8#include <linux/kallsyms.h>
9#include <linux/debugfs.h>
10#include <linux/hrtimer.h>
11#include <linux/uaccess.h>
12#include <linux/ftrace.h>
13#include <linux/module.h>
14#include <linux/irq.h>
15#include <linux/fs.h>
16
17#include <asm/stacktrace.h>
18
19#include "trace.h"
20
21static struct trace_array *sysprof_trace;
22static int __read_mostly tracer_enabled;
23
24/*
25 * 1 msec sample interval by default:
26 */
27static unsigned long sample_period = 1000000;
28static const unsigned int sample_max_depth = 512;
29
30static DEFINE_MUTEX(sample_timer_lock);
31/*
32 * Per CPU hrtimers that do the profiling:
33 */
34static DEFINE_PER_CPU(struct hrtimer, stack_trace_hrtimer);
35
36struct stack_frame {
37 const void __user *next_fp;
38 unsigned long return_address;
39};
40
41static int copy_stack_frame(const void __user *fp, struct stack_frame *frame)
42{
43 int ret;
44
45 if (!access_ok(VERIFY_READ, fp, sizeof(*frame)))
46 return 0;
47
48 ret = 1;
49 pagefault_disable();
50 if (__copy_from_user_inatomic(frame, fp, sizeof(*frame)))
51 ret = 0;
52 pagefault_enable();
53
54 return ret;
55}
56
57struct backtrace_info {
58 struct trace_array_cpu *data;
59 struct trace_array *tr;
60 int pos;
61};
62
63static void
64backtrace_warning_symbol(void *data, char *msg, unsigned long symbol)
65{
66 /* Ignore warnings */
67}
68
69static void backtrace_warning(void *data, char *msg)
70{
71 /* Ignore warnings */
72}
73
74static int backtrace_stack(void *data, char *name)
75{
76 /* Don't bother with IRQ stacks for now */
77 return -1;
78}
79
80static void backtrace_address(void *data, unsigned long addr, int reliable)
81{
82 struct backtrace_info *info = data;
83
84 if (info->pos < sample_max_depth && reliable) {
85 __trace_special(info->tr, info->data, 1, addr, 0);
86
87 info->pos++;
88 }
89}
90
91const static struct stacktrace_ops backtrace_ops = {
92 .warning = backtrace_warning,
93 .warning_symbol = backtrace_warning_symbol,
94 .stack = backtrace_stack,
95 .address = backtrace_address,
96};
97
98static int
99trace_kernel(struct pt_regs *regs, struct trace_array *tr,
100 struct trace_array_cpu *data)
101{
102 struct backtrace_info info;
103 unsigned long bp;
104 char *stack;
105
106 info.tr = tr;
107 info.data = data;
108 info.pos = 1;
109
110 __trace_special(info.tr, info.data, 1, regs->ip, 0);
111
112 stack = ((char *)regs + sizeof(struct pt_regs));
113#ifdef CONFIG_FRAME_POINTER
114 bp = regs->bp;
115#else
116 bp = 0;
117#endif
118
119 dump_trace(NULL, regs, (void *)stack, bp, &backtrace_ops, &info);
120
121 return info.pos;
122}
123
124static void timer_notify(struct pt_regs *regs, int cpu)
125{
126 struct trace_array_cpu *data;
127 struct stack_frame frame;
128 struct trace_array *tr;
129 const void __user *fp;
130 int is_user;
131 int i;
132
133 if (!regs)
134 return;
135
136 tr = sysprof_trace;
137 data = tr->data[cpu];
138 is_user = user_mode(regs);
139
140 if (!current || current->pid == 0)
141 return;
142
143 if (is_user && current->state != TASK_RUNNING)
144 return;
145
146 __trace_special(tr, data, 0, 0, current->pid);
147
148 if (!is_user)
149 i = trace_kernel(regs, tr, data);
150 else
151 i = 0;
152
153 /*
154 * Trace user stack if we are not a kernel thread
155 */
156 if (current->mm && i < sample_max_depth) {
157 regs = (struct pt_regs *)current->thread.sp0 - 1;
158
159 fp = (void __user *)regs->bp;
160
161 __trace_special(tr, data, 2, regs->ip, 0);
162
163 while (i < sample_max_depth) {
164 frame.next_fp = 0;
165 frame.return_address = 0;
166 if (!copy_stack_frame(fp, &frame))
167 break;
168 if ((unsigned long)fp < regs->sp)
169 break;
170
171 __trace_special(tr, data, 2, frame.return_address,
172 (unsigned long)fp);
173 fp = frame.next_fp;
174
175 i++;
176 }
177
178 }
179
180 /*
181 * Special trace entry if we overflow the max depth:
182 */
183 if (i == sample_max_depth)
184 __trace_special(tr, data, -1, -1, -1);
185
186 __trace_special(tr, data, 3, current->pid, i);
187}
188
189static enum hrtimer_restart stack_trace_timer_fn(struct hrtimer *hrtimer)
190{
191 /* trace here */
192 timer_notify(get_irq_regs(), smp_processor_id());
193
194 hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period));
195
196 return HRTIMER_RESTART;
197}
198
199static void start_stack_timer(int cpu)
200{
201 struct hrtimer *hrtimer = &per_cpu(stack_trace_hrtimer, cpu);
202
203 hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
204 hrtimer->function = stack_trace_timer_fn;
205 hrtimer->cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ;
206
207 hrtimer_start(hrtimer, ns_to_ktime(sample_period), HRTIMER_MODE_REL);
208}
209
210static void start_stack_timers(void)
211{
212 cpumask_t saved_mask = current->cpus_allowed;
213 int cpu;
214
215 for_each_online_cpu(cpu) {
216 set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu));
217 start_stack_timer(cpu);
218 }
219 set_cpus_allowed_ptr(current, &saved_mask);
220}
221
222static void stop_stack_timer(int cpu)
223{
224 struct hrtimer *hrtimer = &per_cpu(stack_trace_hrtimer, cpu);
225
226 hrtimer_cancel(hrtimer);
227}
228
229static void stop_stack_timers(void)
230{
231 int cpu;
232
233 for_each_online_cpu(cpu)
234 stop_stack_timer(cpu);
235}
236
237static void stack_reset(struct trace_array *tr)
238{
239 int cpu;
240
241 tr->time_start = ftrace_now(tr->cpu);
242
243 for_each_online_cpu(cpu)
244 tracing_reset(tr->data[cpu]);
245}
246
247static void start_stack_trace(struct trace_array *tr)
248{
249 mutex_lock(&sample_timer_lock);
250 stack_reset(tr);
251 start_stack_timers();
252 tracer_enabled = 1;
253 mutex_unlock(&sample_timer_lock);
254}
255
256static void stop_stack_trace(struct trace_array *tr)
257{
258 mutex_lock(&sample_timer_lock);
259 stop_stack_timers();
260 tracer_enabled = 0;
261 mutex_unlock(&sample_timer_lock);
262}
263
264static void stack_trace_init(struct trace_array *tr)
265{
266 sysprof_trace = tr;
267
268 if (tr->ctrl)
269 start_stack_trace(tr);
270}
271
272static void stack_trace_reset(struct trace_array *tr)
273{
274 if (tr->ctrl)
275 stop_stack_trace(tr);
276}
277
278static void stack_trace_ctrl_update(struct trace_array *tr)
279{
280 /* When starting a new trace, reset the buffers */
281 if (tr->ctrl)
282 start_stack_trace(tr);
283 else
284 stop_stack_trace(tr);
285}
286
287static struct tracer stack_trace __read_mostly =
288{
289 .name = "sysprof",
290 .init = stack_trace_init,
291 .reset = stack_trace_reset,
292 .ctrl_update = stack_trace_ctrl_update,
293#ifdef CONFIG_FTRACE_SELFTEST
294 .selftest = trace_selftest_startup_sysprof,
295#endif
296};
297
298__init static int init_stack_trace(void)
299{
300 return register_tracer(&stack_trace);
301}
302device_initcall(init_stack_trace);
303
304#define MAX_LONG_DIGITS 22
305
306static ssize_t
307sysprof_sample_read(struct file *filp, char __user *ubuf,
308 size_t cnt, loff_t *ppos)
309{
310 char buf[MAX_LONG_DIGITS];
311 int r;
312
313 r = sprintf(buf, "%ld\n", nsecs_to_usecs(sample_period));
314
315 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
316}
317
318static ssize_t
319sysprof_sample_write(struct file *filp, const char __user *ubuf,
320 size_t cnt, loff_t *ppos)
321{
322 char buf[MAX_LONG_DIGITS];
323 unsigned long val;
324
325 if (cnt > MAX_LONG_DIGITS-1)
326 cnt = MAX_LONG_DIGITS-1;
327
328 if (copy_from_user(&buf, ubuf, cnt))
329 return -EFAULT;
330
331 buf[cnt] = 0;
332
333 val = simple_strtoul(buf, NULL, 10);
334 /*
335 * Enforce a minimum sample period of 100 usecs:
336 */
337 if (val < 100)
338 val = 100;
339
340 mutex_lock(&sample_timer_lock);
341 stop_stack_timers();
342 sample_period = val * 1000;
343 start_stack_timers();
344 mutex_unlock(&sample_timer_lock);
345
346 return cnt;
347}
348
349static struct file_operations sysprof_sample_fops = {
350 .read = sysprof_sample_read,
351 .write = sysprof_sample_write,
352};
353
354void init_tracer_sysprof_debugfs(struct dentry *d_tracer)
355{
356 struct dentry *entry;
357
358 entry = debugfs_create_file("sysprof_sample_period", 0644,
359 d_tracer, NULL, &sysprof_sample_fops);
360 if (entry)
361 return;
362 pr_warning("Could not create debugfs 'dyn_ftrace_total_info' entry\n");
363}