2 * ring buffer based function tracer
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
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
14 #include <linux/utsrelease.h>
15 #include <linux/kallsyms.h>
16 #include <linux/seq_file.h>
17 #include <linux/notifier.h>
18 #include <linux/debugfs.h>
19 #include <linux/pagemap.h>
20 #include <linux/hardirq.h>
21 #include <linux/linkage.h>
22 #include <linux/uaccess.h>
23 #include <linux/ftrace.h>
24 #include <linux/module.h>
25 #include <linux/percpu.h>
26 #include <linux/kdebug.h>
27 #include <linux/ctype.h>
28 #include <linux/init.h>
29 #include <linux/poll.h>
30 #include <linux/gfp.h>
32 #include <linux/kprobes.h>
33 #include <linux/writeback.h>
35 #include <linux/stacktrace.h>
36 #include <linux/ring_buffer.h>
37 #include <linux/irqflags.h>
40 #include "trace_output.h"
42 #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
44 unsigned long __read_mostly tracing_max_latency;
45 unsigned long __read_mostly tracing_thresh;
48 * We need to change this state when a selftest is running.
49 * A selftest will lurk into the ring-buffer to count the
50 * entries inserted during the selftest although some concurrent
51 * insertions into the ring-buffer such as ftrace_printk could occurred
52 * at the same time, giving false positive or negative results.
54 static bool __read_mostly tracing_selftest_running;
57 * If a tracer is running, we do not want to run SELFTEST.
59 static bool __read_mostly tracing_selftest_disabled;
61 /* For tracers that don't implement custom flags */
62 static struct tracer_opt dummy_tracer_opt[] = {
66 static struct tracer_flags dummy_tracer_flags = {
68 .opts = dummy_tracer_opt
71 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
77 * Kill all tracing for good (never come back).
78 * It is initialized to 1 but will turn to zero if the initialization
79 * of the tracer is successful. But that is the only place that sets
82 int tracing_disabled = 1;
84 static DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
86 static inline void ftrace_disable_cpu(void)
89 local_inc(&__get_cpu_var(ftrace_cpu_disabled));
92 static inline void ftrace_enable_cpu(void)
94 local_dec(&__get_cpu_var(ftrace_cpu_disabled));
98 static cpumask_var_t __read_mostly tracing_buffer_mask;
100 #define for_each_tracing_cpu(cpu) \
101 for_each_cpu(cpu, tracing_buffer_mask)
104 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
106 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
107 * is set, then ftrace_dump is called. This will output the contents
108 * of the ftrace buffers to the console. This is very useful for
109 * capturing traces that lead to crashes and outputing it to a
112 * It is default off, but you can enable it with either specifying
113 * "ftrace_dump_on_oops" in the kernel command line, or setting
114 * /proc/sys/kernel/ftrace_dump_on_oops to true.
116 int ftrace_dump_on_oops;
118 static int tracing_set_tracer(const char *buf);
120 #define BOOTUP_TRACER_SIZE 100
121 static char bootup_tracer_buf[BOOTUP_TRACER_SIZE] __initdata;
122 static char *default_bootup_tracer;
124 static int __init set_ftrace(char *str)
126 strncpy(bootup_tracer_buf, str, BOOTUP_TRACER_SIZE);
127 default_bootup_tracer = bootup_tracer_buf;
130 __setup("ftrace=", set_ftrace);
132 static int __init set_ftrace_dump_on_oops(char *str)
134 ftrace_dump_on_oops = 1;
137 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
140 ns2usecs(cycle_t nsec)
147 cycle_t ftrace_now(int cpu)
149 u64 ts = ring_buffer_time_stamp(cpu);
150 ring_buffer_normalize_time_stamp(cpu, &ts);
155 * The global_trace is the descriptor that holds the tracing
156 * buffers for the live tracing. For each CPU, it contains
157 * a link list of pages that will store trace entries. The
158 * page descriptor of the pages in the memory is used to hold
159 * the link list by linking the lru item in the page descriptor
160 * to each of the pages in the buffer per CPU.
162 * For each active CPU there is a data field that holds the
163 * pages for the buffer for that CPU. Each CPU has the same number
164 * of pages allocated for its buffer.
166 static struct trace_array global_trace;
168 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
171 * The max_tr is used to snapshot the global_trace when a maximum
172 * latency is reached. Some tracers will use this to store a maximum
173 * trace while it continues examining live traces.
175 * The buffers for the max_tr are set up the same as the global_trace.
176 * When a snapshot is taken, the link list of the max_tr is swapped
177 * with the link list of the global_trace and the buffers are reset for
178 * the global_trace so the tracing can continue.
180 static struct trace_array max_tr;
182 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
184 /* tracer_enabled is used to toggle activation of a tracer */
185 static int tracer_enabled = 1;
188 * tracing_is_enabled - return tracer_enabled status
190 * This function is used by other tracers to know the status
191 * of the tracer_enabled flag. Tracers may use this function
192 * to know if it should enable their features when starting
193 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
195 int tracing_is_enabled(void)
197 return tracer_enabled;
201 * trace_buf_size is the size in bytes that is allocated
202 * for a buffer. Note, the number of bytes is always rounded
205 * This number is purposely set to a low number of 16384.
206 * If the dump on oops happens, it will be much appreciated
207 * to not have to wait for all that output. Anyway this can be
208 * boot time and run time configurable.
210 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
212 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
214 /* trace_types holds a link list of available tracers. */
215 static struct tracer *trace_types __read_mostly;
217 /* current_trace points to the tracer that is currently active */
218 static struct tracer *current_trace __read_mostly;
221 * max_tracer_type_len is used to simplify the allocating of
222 * buffers to read userspace tracer names. We keep track of
223 * the longest tracer name registered.
225 static int max_tracer_type_len;
228 * trace_types_lock is used to protect the trace_types list.
229 * This lock is also used to keep user access serialized.
230 * Accesses from userspace will grab this lock while userspace
231 * activities happen inside the kernel.
233 static DEFINE_MUTEX(trace_types_lock);
235 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
236 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
238 /* trace_flags holds trace_options default values */
239 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
240 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO;
243 * trace_wake_up - wake up tasks waiting for trace input
245 * Simply wakes up any task that is blocked on the trace_wait
246 * queue. These is used with trace_poll for tasks polling the trace.
248 void trace_wake_up(void)
251 * The runqueue_is_locked() can fail, but this is the best we
254 if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
255 wake_up(&trace_wait);
258 static int __init set_buf_size(char *str)
260 unsigned long buf_size;
265 ret = strict_strtoul(str, 0, &buf_size);
266 /* nr_entries can not be zero */
267 if (ret < 0 || buf_size == 0)
269 trace_buf_size = buf_size;
272 __setup("trace_buf_size=", set_buf_size);
274 unsigned long nsecs_to_usecs(unsigned long nsecs)
279 /* These must match the bit postions in trace_iterator_flags */
280 static const char *trace_options[] = {
303 * ftrace_max_lock is used to protect the swapping of buffers
304 * when taking a max snapshot. The buffers themselves are
305 * protected by per_cpu spinlocks. But the action of the swap
306 * needs its own lock.
308 * This is defined as a raw_spinlock_t in order to help
309 * with performance when lockdep debugging is enabled.
311 static raw_spinlock_t ftrace_max_lock =
312 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
315 * Copy the new maximum trace into the separate maximum-trace
316 * structure. (this way the maximum trace is permanently saved,
317 * for later retrieval via /debugfs/tracing/latency_trace)
320 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
322 struct trace_array_cpu *data = tr->data[cpu];
325 max_tr.time_start = data->preempt_timestamp;
327 data = max_tr.data[cpu];
328 data->saved_latency = tracing_max_latency;
330 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
331 data->pid = tsk->pid;
332 data->uid = task_uid(tsk);
333 data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
334 data->policy = tsk->policy;
335 data->rt_priority = tsk->rt_priority;
337 /* record this tasks comm */
338 tracing_record_cmdline(current);
342 trace_seq_reset(struct trace_seq *s)
348 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
353 if (s->len <= s->readpos)
356 len = s->len - s->readpos;
359 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
368 trace_print_seq(struct seq_file *m, struct trace_seq *s)
370 int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
373 seq_puts(m, s->buffer);
379 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
381 * @tsk: the task with the latency
382 * @cpu: The cpu that initiated the trace.
384 * Flip the buffers between the @tr and the max_tr and record information
385 * about which task was the cause of this latency.
388 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
390 struct ring_buffer *buf = tr->buffer;
392 WARN_ON_ONCE(!irqs_disabled());
393 __raw_spin_lock(&ftrace_max_lock);
395 tr->buffer = max_tr.buffer;
398 ftrace_disable_cpu();
399 ring_buffer_reset(tr->buffer);
402 __update_max_tr(tr, tsk, cpu);
403 __raw_spin_unlock(&ftrace_max_lock);
407 * update_max_tr_single - only copy one trace over, and reset the rest
409 * @tsk - task with the latency
410 * @cpu - the cpu of the buffer to copy.
412 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
415 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
419 WARN_ON_ONCE(!irqs_disabled());
420 __raw_spin_lock(&ftrace_max_lock);
422 ftrace_disable_cpu();
424 ring_buffer_reset(max_tr.buffer);
425 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
429 WARN_ON_ONCE(ret && ret != -EAGAIN);
431 __update_max_tr(tr, tsk, cpu);
432 __raw_spin_unlock(&ftrace_max_lock);
436 * register_tracer - register a tracer with the ftrace system.
437 * @type - the plugin for the tracer
439 * Register a new plugin tracer.
441 int register_tracer(struct tracer *type)
448 pr_info("Tracer must have a name\n");
453 * When this gets called we hold the BKL which means that
454 * preemption is disabled. Various trace selftests however
455 * need to disable and enable preemption for successful tests.
456 * So we drop the BKL here and grab it after the tests again.
459 mutex_lock(&trace_types_lock);
461 tracing_selftest_running = true;
463 for (t = trace_types; t; t = t->next) {
464 if (strcmp(type->name, t->name) == 0) {
466 pr_info("Trace %s already registered\n",
474 type->set_flag = &dummy_set_flag;
476 type->flags = &dummy_tracer_flags;
478 if (!type->flags->opts)
479 type->flags->opts = dummy_tracer_opt;
481 #ifdef CONFIG_FTRACE_STARTUP_TEST
482 if (type->selftest && !tracing_selftest_disabled) {
483 struct tracer *saved_tracer = current_trace;
484 struct trace_array *tr = &global_trace;
488 * Run a selftest on this tracer.
489 * Here we reset the trace buffer, and set the current
490 * tracer to be this tracer. The tracer can then run some
491 * internal tracing to verify that everything is in order.
492 * If we fail, we do not register this tracer.
494 for_each_tracing_cpu(i)
495 tracing_reset(tr, i);
497 current_trace = type;
498 /* the test is responsible for initializing and enabling */
499 pr_info("Testing tracer %s: ", type->name);
500 ret = type->selftest(type, tr);
501 /* the test is responsible for resetting too */
502 current_trace = saved_tracer;
504 printk(KERN_CONT "FAILED!\n");
507 /* Only reset on passing, to avoid touching corrupted buffers */
508 for_each_tracing_cpu(i)
509 tracing_reset(tr, i);
511 printk(KERN_CONT "PASSED\n");
515 type->next = trace_types;
517 len = strlen(type->name);
518 if (len > max_tracer_type_len)
519 max_tracer_type_len = len;
522 tracing_selftest_running = false;
523 mutex_unlock(&trace_types_lock);
525 if (!ret && default_bootup_tracer) {
526 if (!strncmp(default_bootup_tracer, type->name,
527 BOOTUP_TRACER_SIZE)) {
528 printk(KERN_INFO "Starting tracer '%s'\n",
530 /* Do we want this tracer to start on bootup? */
531 tracing_set_tracer(type->name);
532 default_bootup_tracer = NULL;
533 /* disable other selftests, since this will break it. */
534 tracing_selftest_disabled = 1;
535 #ifdef CONFIG_FTRACE_STARTUP_TEST
536 printk(KERN_INFO "Disabling FTRACE selftests due"
537 " to running tracer '%s'\n", type->name);
546 void unregister_tracer(struct tracer *type)
551 mutex_lock(&trace_types_lock);
552 for (t = &trace_types; *t; t = &(*t)->next) {
556 pr_info("Trace %s not registered\n", type->name);
561 if (strlen(type->name) != max_tracer_type_len)
564 max_tracer_type_len = 0;
565 for (t = &trace_types; *t; t = &(*t)->next) {
566 len = strlen((*t)->name);
567 if (len > max_tracer_type_len)
568 max_tracer_type_len = len;
571 mutex_unlock(&trace_types_lock);
574 void tracing_reset(struct trace_array *tr, int cpu)
576 ftrace_disable_cpu();
577 ring_buffer_reset_cpu(tr->buffer, cpu);
581 void tracing_reset_online_cpus(struct trace_array *tr)
585 tr->time_start = ftrace_now(tr->cpu);
587 for_each_online_cpu(cpu)
588 tracing_reset(tr, cpu);
591 #define SAVED_CMDLINES 128
592 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
593 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
594 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
595 static int cmdline_idx;
596 static DEFINE_SPINLOCK(trace_cmdline_lock);
598 /* temporary disable recording */
599 atomic_t trace_record_cmdline_disabled __read_mostly;
601 static void trace_init_cmdlines(void)
603 memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
604 memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
608 static int trace_stop_count;
609 static DEFINE_SPINLOCK(tracing_start_lock);
612 * ftrace_off_permanent - disable all ftrace code permanently
614 * This should only be called when a serious anomally has
615 * been detected. This will turn off the function tracing,
616 * ring buffers, and other tracing utilites. It takes no
617 * locks and can be called from any context.
619 void ftrace_off_permanent(void)
621 tracing_disabled = 1;
623 tracing_off_permanent();
627 * tracing_start - quick start of the tracer
629 * If tracing is enabled but was stopped by tracing_stop,
630 * this will start the tracer back up.
632 void tracing_start(void)
634 struct ring_buffer *buffer;
637 if (tracing_disabled)
640 spin_lock_irqsave(&tracing_start_lock, flags);
641 if (--trace_stop_count) {
642 if (trace_stop_count < 0) {
643 /* Someone screwed up their debugging */
645 trace_stop_count = 0;
651 buffer = global_trace.buffer;
653 ring_buffer_record_enable(buffer);
655 buffer = max_tr.buffer;
657 ring_buffer_record_enable(buffer);
661 spin_unlock_irqrestore(&tracing_start_lock, flags);
665 * tracing_stop - quick stop of the tracer
667 * Light weight way to stop tracing. Use in conjunction with
670 void tracing_stop(void)
672 struct ring_buffer *buffer;
676 spin_lock_irqsave(&tracing_start_lock, flags);
677 if (trace_stop_count++)
680 buffer = global_trace.buffer;
682 ring_buffer_record_disable(buffer);
684 buffer = max_tr.buffer;
686 ring_buffer_record_disable(buffer);
689 spin_unlock_irqrestore(&tracing_start_lock, flags);
692 void trace_stop_cmdline_recording(void);
694 static void trace_save_cmdline(struct task_struct *tsk)
699 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
703 * It's not the end of the world if we don't get
704 * the lock, but we also don't want to spin
705 * nor do we want to disable interrupts,
706 * so if we miss here, then better luck next time.
708 if (!spin_trylock(&trace_cmdline_lock))
711 idx = map_pid_to_cmdline[tsk->pid];
712 if (idx >= SAVED_CMDLINES) {
713 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
715 map = map_cmdline_to_pid[idx];
716 if (map <= PID_MAX_DEFAULT)
717 map_pid_to_cmdline[map] = (unsigned)-1;
719 map_pid_to_cmdline[tsk->pid] = idx;
724 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
726 spin_unlock(&trace_cmdline_lock);
729 char *trace_find_cmdline(int pid)
731 char *cmdline = "<...>";
737 if (pid > PID_MAX_DEFAULT)
740 map = map_pid_to_cmdline[pid];
741 if (map >= SAVED_CMDLINES)
744 cmdline = saved_cmdlines[map];
750 void tracing_record_cmdline(struct task_struct *tsk)
752 if (atomic_read(&trace_record_cmdline_disabled))
755 trace_save_cmdline(tsk);
759 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
762 struct task_struct *tsk = current;
764 entry->preempt_count = pc & 0xff;
765 entry->pid = (tsk) ? tsk->pid : 0;
766 entry->tgid = (tsk) ? tsk->tgid : 0;
768 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
769 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
771 TRACE_FLAG_IRQS_NOSUPPORT |
773 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
774 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
775 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
779 trace_function(struct trace_array *tr,
780 unsigned long ip, unsigned long parent_ip, unsigned long flags,
783 struct ring_buffer_event *event;
784 struct ftrace_entry *entry;
785 unsigned long irq_flags;
787 /* If we are reading the ring buffer, don't trace */
788 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
791 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
795 entry = ring_buffer_event_data(event);
796 tracing_generic_entry_update(&entry->ent, flags, pc);
797 entry->ent.type = TRACE_FN;
799 entry->parent_ip = parent_ip;
800 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
803 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
804 static void __trace_graph_entry(struct trace_array *tr,
805 struct ftrace_graph_ent *trace,
809 struct ring_buffer_event *event;
810 struct ftrace_graph_ent_entry *entry;
811 unsigned long irq_flags;
813 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
816 event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry),
820 entry = ring_buffer_event_data(event);
821 tracing_generic_entry_update(&entry->ent, flags, pc);
822 entry->ent.type = TRACE_GRAPH_ENT;
823 entry->graph_ent = *trace;
824 ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
827 static void __trace_graph_return(struct trace_array *tr,
828 struct ftrace_graph_ret *trace,
832 struct ring_buffer_event *event;
833 struct ftrace_graph_ret_entry *entry;
834 unsigned long irq_flags;
836 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
839 event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry),
843 entry = ring_buffer_event_data(event);
844 tracing_generic_entry_update(&entry->ent, flags, pc);
845 entry->ent.type = TRACE_GRAPH_RET;
847 ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
852 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
853 unsigned long ip, unsigned long parent_ip, unsigned long flags,
856 if (likely(!atomic_read(&data->disabled)))
857 trace_function(tr, ip, parent_ip, flags, pc);
860 static void __ftrace_trace_stack(struct trace_array *tr,
864 #ifdef CONFIG_STACKTRACE
865 struct ring_buffer_event *event;
866 struct stack_entry *entry;
867 struct stack_trace trace;
868 unsigned long irq_flags;
870 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
874 entry = ring_buffer_event_data(event);
875 tracing_generic_entry_update(&entry->ent, flags, pc);
876 entry->ent.type = TRACE_STACK;
878 memset(&entry->caller, 0, sizeof(entry->caller));
880 trace.nr_entries = 0;
881 trace.max_entries = FTRACE_STACK_ENTRIES;
883 trace.entries = entry->caller;
885 save_stack_trace(&trace);
886 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
890 static void ftrace_trace_stack(struct trace_array *tr,
894 if (!(trace_flags & TRACE_ITER_STACKTRACE))
897 __ftrace_trace_stack(tr, flags, skip, pc);
900 void __trace_stack(struct trace_array *tr,
904 __ftrace_trace_stack(tr, flags, skip, pc);
907 static void ftrace_trace_userstack(struct trace_array *tr,
908 unsigned long flags, int pc)
910 #ifdef CONFIG_STACKTRACE
911 struct ring_buffer_event *event;
912 struct userstack_entry *entry;
913 struct stack_trace trace;
914 unsigned long irq_flags;
916 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
919 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
923 entry = ring_buffer_event_data(event);
924 tracing_generic_entry_update(&entry->ent, flags, pc);
925 entry->ent.type = TRACE_USER_STACK;
927 memset(&entry->caller, 0, sizeof(entry->caller));
929 trace.nr_entries = 0;
930 trace.max_entries = FTRACE_STACK_ENTRIES;
932 trace.entries = entry->caller;
934 save_stack_trace_user(&trace);
935 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
939 void __trace_userstack(struct trace_array *tr, unsigned long flags)
941 ftrace_trace_userstack(tr, flags, preempt_count());
945 ftrace_trace_special(void *__tr,
946 unsigned long arg1, unsigned long arg2, unsigned long arg3,
949 struct ring_buffer_event *event;
950 struct trace_array *tr = __tr;
951 struct special_entry *entry;
952 unsigned long irq_flags;
954 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
958 entry = ring_buffer_event_data(event);
959 tracing_generic_entry_update(&entry->ent, 0, pc);
960 entry->ent.type = TRACE_SPECIAL;
964 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
965 ftrace_trace_stack(tr, irq_flags, 4, pc);
966 ftrace_trace_userstack(tr, irq_flags, pc);
972 __trace_special(void *__tr, void *__data,
973 unsigned long arg1, unsigned long arg2, unsigned long arg3)
975 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
979 tracing_sched_switch_trace(struct trace_array *tr,
980 struct task_struct *prev,
981 struct task_struct *next,
982 unsigned long flags, int pc)
984 struct ring_buffer_event *event;
985 struct ctx_switch_entry *entry;
986 unsigned long irq_flags;
988 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
992 entry = ring_buffer_event_data(event);
993 tracing_generic_entry_update(&entry->ent, flags, pc);
994 entry->ent.type = TRACE_CTX;
995 entry->prev_pid = prev->pid;
996 entry->prev_prio = prev->prio;
997 entry->prev_state = prev->state;
998 entry->next_pid = next->pid;
999 entry->next_prio = next->prio;
1000 entry->next_state = next->state;
1001 entry->next_cpu = task_cpu(next);
1002 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1003 ftrace_trace_stack(tr, flags, 5, pc);
1004 ftrace_trace_userstack(tr, flags, pc);
1008 tracing_sched_wakeup_trace(struct trace_array *tr,
1009 struct task_struct *wakee,
1010 struct task_struct *curr,
1011 unsigned long flags, int pc)
1013 struct ring_buffer_event *event;
1014 struct ctx_switch_entry *entry;
1015 unsigned long irq_flags;
1017 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1021 entry = ring_buffer_event_data(event);
1022 tracing_generic_entry_update(&entry->ent, flags, pc);
1023 entry->ent.type = TRACE_WAKE;
1024 entry->prev_pid = curr->pid;
1025 entry->prev_prio = curr->prio;
1026 entry->prev_state = curr->state;
1027 entry->next_pid = wakee->pid;
1028 entry->next_prio = wakee->prio;
1029 entry->next_state = wakee->state;
1030 entry->next_cpu = task_cpu(wakee);
1031 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1032 ftrace_trace_stack(tr, flags, 6, pc);
1033 ftrace_trace_userstack(tr, flags, pc);
1039 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1041 struct trace_array *tr = &global_trace;
1042 struct trace_array_cpu *data;
1043 unsigned long flags;
1047 if (tracing_disabled)
1050 pc = preempt_count();
1051 local_irq_save(flags);
1052 cpu = raw_smp_processor_id();
1053 data = tr->data[cpu];
1055 if (likely(atomic_inc_return(&data->disabled) == 1))
1056 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1058 atomic_dec(&data->disabled);
1059 local_irq_restore(flags);
1062 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1063 int trace_graph_entry(struct ftrace_graph_ent *trace)
1065 struct trace_array *tr = &global_trace;
1066 struct trace_array_cpu *data;
1067 unsigned long flags;
1072 if (!ftrace_trace_task(current))
1075 if (!ftrace_graph_addr(trace->func))
1078 local_irq_save(flags);
1079 cpu = raw_smp_processor_id();
1080 data = tr->data[cpu];
1081 disabled = atomic_inc_return(&data->disabled);
1082 if (likely(disabled == 1)) {
1083 pc = preempt_count();
1084 __trace_graph_entry(tr, trace, flags, pc);
1086 /* Only do the atomic if it is not already set */
1087 if (!test_tsk_trace_graph(current))
1088 set_tsk_trace_graph(current);
1089 atomic_dec(&data->disabled);
1090 local_irq_restore(flags);
1095 void trace_graph_return(struct ftrace_graph_ret *trace)
1097 struct trace_array *tr = &global_trace;
1098 struct trace_array_cpu *data;
1099 unsigned long flags;
1104 local_irq_save(flags);
1105 cpu = raw_smp_processor_id();
1106 data = tr->data[cpu];
1107 disabled = atomic_inc_return(&data->disabled);
1108 if (likely(disabled == 1)) {
1109 pc = preempt_count();
1110 __trace_graph_return(tr, trace, flags, pc);
1113 clear_tsk_trace_graph(current);
1114 atomic_dec(&data->disabled);
1115 local_irq_restore(flags);
1117 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1119 enum trace_file_type {
1120 TRACE_FILE_LAT_FMT = 1,
1121 TRACE_FILE_ANNOTATE = 2,
1124 static void trace_iterator_increment(struct trace_iterator *iter)
1126 /* Don't allow ftrace to trace into the ring buffers */
1127 ftrace_disable_cpu();
1130 if (iter->buffer_iter[iter->cpu])
1131 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1133 ftrace_enable_cpu();
1136 static struct trace_entry *
1137 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1139 struct ring_buffer_event *event;
1140 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1142 /* Don't allow ftrace to trace into the ring buffers */
1143 ftrace_disable_cpu();
1146 event = ring_buffer_iter_peek(buf_iter, ts);
1148 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1150 ftrace_enable_cpu();
1152 return event ? ring_buffer_event_data(event) : NULL;
1155 static struct trace_entry *
1156 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1158 struct ring_buffer *buffer = iter->tr->buffer;
1159 struct trace_entry *ent, *next = NULL;
1160 u64 next_ts = 0, ts;
1164 for_each_tracing_cpu(cpu) {
1166 if (ring_buffer_empty_cpu(buffer, cpu))
1169 ent = peek_next_entry(iter, cpu, &ts);
1172 * Pick the entry with the smallest timestamp:
1174 if (ent && (!next || ts < next_ts)) {
1182 *ent_cpu = next_cpu;
1190 /* Find the next real entry, without updating the iterator itself */
1191 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1192 int *ent_cpu, u64 *ent_ts)
1194 return __find_next_entry(iter, ent_cpu, ent_ts);
1197 /* Find the next real entry, and increment the iterator to the next entry */
1198 static void *find_next_entry_inc(struct trace_iterator *iter)
1200 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1203 trace_iterator_increment(iter);
1205 return iter->ent ? iter : NULL;
1208 static void trace_consume(struct trace_iterator *iter)
1210 /* Don't allow ftrace to trace into the ring buffers */
1211 ftrace_disable_cpu();
1212 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1213 ftrace_enable_cpu();
1216 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1218 struct trace_iterator *iter = m->private;
1224 /* can't go backwards */
1229 ent = find_next_entry_inc(iter);
1233 while (ent && iter->idx < i)
1234 ent = find_next_entry_inc(iter);
1241 static void *s_start(struct seq_file *m, loff_t *pos)
1243 struct trace_iterator *iter = m->private;
1248 mutex_lock(&trace_types_lock);
1250 if (!current_trace || current_trace != iter->trace) {
1251 mutex_unlock(&trace_types_lock);
1255 atomic_inc(&trace_record_cmdline_disabled);
1257 if (*pos != iter->pos) {
1262 ftrace_disable_cpu();
1264 for_each_tracing_cpu(cpu) {
1265 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1268 ftrace_enable_cpu();
1270 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1275 p = s_next(m, p, &l);
1281 static void s_stop(struct seq_file *m, void *p)
1283 atomic_dec(&trace_record_cmdline_disabled);
1284 mutex_unlock(&trace_types_lock);
1287 static void print_lat_help_header(struct seq_file *m)
1289 seq_puts(m, "# _------=> CPU# \n");
1290 seq_puts(m, "# / _-----=> irqs-off \n");
1291 seq_puts(m, "# | / _----=> need-resched \n");
1292 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1293 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1294 seq_puts(m, "# |||| / \n");
1295 seq_puts(m, "# ||||| delay \n");
1296 seq_puts(m, "# cmd pid ||||| time | caller \n");
1297 seq_puts(m, "# \\ / ||||| \\ | / \n");
1300 static void print_func_help_header(struct seq_file *m)
1302 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1303 seq_puts(m, "# | | | | |\n");
1308 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1310 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1311 struct trace_array *tr = iter->tr;
1312 struct trace_array_cpu *data = tr->data[tr->cpu];
1313 struct tracer *type = current_trace;
1314 unsigned long total;
1315 unsigned long entries;
1316 const char *name = "preemption";
1321 entries = ring_buffer_entries(iter->tr->buffer);
1323 ring_buffer_overruns(iter->tr->buffer);
1325 seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1327 seq_puts(m, "-----------------------------------"
1328 "---------------------------------\n");
1329 seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1330 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1331 nsecs_to_usecs(data->saved_latency),
1335 #if defined(CONFIG_PREEMPT_NONE)
1337 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1339 #elif defined(CONFIG_PREEMPT)
1344 /* These are reserved for later use */
1347 seq_printf(m, " #P:%d)\n", num_online_cpus());
1351 seq_puts(m, " -----------------\n");
1352 seq_printf(m, " | task: %.16s-%d "
1353 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1354 data->comm, data->pid, data->uid, data->nice,
1355 data->policy, data->rt_priority);
1356 seq_puts(m, " -----------------\n");
1358 if (data->critical_start) {
1359 seq_puts(m, " => started at: ");
1360 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1361 trace_print_seq(m, &iter->seq);
1362 seq_puts(m, "\n => ended at: ");
1363 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1364 trace_print_seq(m, &iter->seq);
1371 static void test_cpu_buff_start(struct trace_iterator *iter)
1373 struct trace_seq *s = &iter->seq;
1375 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1378 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1381 if (cpumask_test_cpu(iter->cpu, iter->started))
1384 cpumask_set_cpu(iter->cpu, iter->started);
1385 trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1388 static enum print_line_t print_lat_fmt(struct trace_iterator *iter)
1390 struct trace_seq *s = &iter->seq;
1391 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1392 struct trace_event *event;
1393 struct trace_entry *entry = iter->ent;
1395 test_cpu_buff_start(iter);
1397 event = ftrace_find_event(entry->type);
1399 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1400 if (!trace_print_lat_context(iter))
1405 return event->latency_trace(iter, sym_flags);
1407 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1410 return TRACE_TYPE_HANDLED;
1412 return TRACE_TYPE_PARTIAL_LINE;
1415 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1417 struct trace_seq *s = &iter->seq;
1418 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1419 struct trace_entry *entry;
1420 struct trace_event *event;
1424 test_cpu_buff_start(iter);
1426 event = ftrace_find_event(entry->type);
1428 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1429 if (!trace_print_context(iter))
1434 return event->trace(iter, sym_flags);
1436 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1439 return TRACE_TYPE_HANDLED;
1441 return TRACE_TYPE_PARTIAL_LINE;
1444 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1446 struct trace_seq *s = &iter->seq;
1447 struct trace_entry *entry;
1448 struct trace_event *event;
1452 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1453 if (!trace_seq_printf(s, "%d %d %llu ",
1454 entry->pid, iter->cpu, iter->ts))
1458 event = ftrace_find_event(entry->type);
1460 return event->raw(iter, 0);
1462 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1465 return TRACE_TYPE_HANDLED;
1467 return TRACE_TYPE_PARTIAL_LINE;
1470 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1472 struct trace_seq *s = &iter->seq;
1473 unsigned char newline = '\n';
1474 struct trace_entry *entry;
1475 struct trace_event *event;
1479 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1480 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1481 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1482 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1485 event = ftrace_find_event(entry->type);
1487 enum print_line_t ret = event->hex(iter, 0);
1488 if (ret != TRACE_TYPE_HANDLED)
1492 SEQ_PUT_FIELD_RET(s, newline);
1494 return TRACE_TYPE_HANDLED;
1497 static enum print_line_t print_printk_msg_only(struct trace_iterator *iter)
1499 struct trace_seq *s = &iter->seq;
1500 struct trace_entry *entry = iter->ent;
1501 struct print_entry *field;
1504 trace_assign_type(field, entry);
1506 ret = trace_seq_printf(s, "%s", field->buf);
1508 return TRACE_TYPE_PARTIAL_LINE;
1510 return TRACE_TYPE_HANDLED;
1513 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1515 struct trace_seq *s = &iter->seq;
1516 struct trace_entry *entry;
1517 struct trace_event *event;
1521 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1522 SEQ_PUT_FIELD_RET(s, entry->pid);
1523 SEQ_PUT_FIELD_RET(s, entry->cpu);
1524 SEQ_PUT_FIELD_RET(s, iter->ts);
1527 event = ftrace_find_event(entry->type);
1528 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1531 static int trace_empty(struct trace_iterator *iter)
1535 for_each_tracing_cpu(cpu) {
1536 if (iter->buffer_iter[cpu]) {
1537 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1540 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1548 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1550 enum print_line_t ret;
1552 if (iter->trace && iter->trace->print_line) {
1553 ret = iter->trace->print_line(iter);
1554 if (ret != TRACE_TYPE_UNHANDLED)
1558 if (iter->ent->type == TRACE_PRINT &&
1559 trace_flags & TRACE_ITER_PRINTK &&
1560 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1561 return print_printk_msg_only(iter);
1563 if (trace_flags & TRACE_ITER_BIN)
1564 return print_bin_fmt(iter);
1566 if (trace_flags & TRACE_ITER_HEX)
1567 return print_hex_fmt(iter);
1569 if (trace_flags & TRACE_ITER_RAW)
1570 return print_raw_fmt(iter);
1572 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1573 return print_lat_fmt(iter);
1575 return print_trace_fmt(iter);
1578 static int s_show(struct seq_file *m, void *v)
1580 struct trace_iterator *iter = v;
1582 if (iter->ent == NULL) {
1584 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1587 if (iter->trace && iter->trace->print_header)
1588 iter->trace->print_header(m);
1589 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1590 /* print nothing if the buffers are empty */
1591 if (trace_empty(iter))
1593 print_trace_header(m, iter);
1594 if (!(trace_flags & TRACE_ITER_VERBOSE))
1595 print_lat_help_header(m);
1597 if (!(trace_flags & TRACE_ITER_VERBOSE))
1598 print_func_help_header(m);
1601 print_trace_line(iter);
1602 trace_print_seq(m, &iter->seq);
1608 static struct seq_operations tracer_seq_ops = {
1615 static struct trace_iterator *
1616 __tracing_open(struct inode *inode, struct file *file, int *ret)
1618 struct trace_iterator *iter;
1622 if (tracing_disabled) {
1627 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1633 mutex_lock(&trace_types_lock);
1634 if (current_trace && current_trace->print_max)
1637 iter->tr = inode->i_private;
1638 iter->trace = current_trace;
1641 /* Notify the tracer early; before we stop tracing. */
1642 if (iter->trace && iter->trace->open)
1643 iter->trace->open(iter);
1645 /* Annotate start of buffers if we had overruns */
1646 if (ring_buffer_overruns(iter->tr->buffer))
1647 iter->iter_flags |= TRACE_FILE_ANNOTATE;
1650 for_each_tracing_cpu(cpu) {
1652 iter->buffer_iter[cpu] =
1653 ring_buffer_read_start(iter->tr->buffer, cpu);
1655 if (!iter->buffer_iter[cpu])
1659 /* TODO stop tracer */
1660 *ret = seq_open(file, &tracer_seq_ops);
1664 m = file->private_data;
1667 /* stop the trace while dumping */
1670 mutex_unlock(&trace_types_lock);
1676 for_each_tracing_cpu(cpu) {
1677 if (iter->buffer_iter[cpu])
1678 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1680 mutex_unlock(&trace_types_lock);
1683 return ERR_PTR(-ENOMEM);
1686 int tracing_open_generic(struct inode *inode, struct file *filp)
1688 if (tracing_disabled)
1691 filp->private_data = inode->i_private;
1695 int tracing_release(struct inode *inode, struct file *file)
1697 struct seq_file *m = (struct seq_file *)file->private_data;
1698 struct trace_iterator *iter = m->private;
1701 mutex_lock(&trace_types_lock);
1702 for_each_tracing_cpu(cpu) {
1703 if (iter->buffer_iter[cpu])
1704 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1707 if (iter->trace && iter->trace->close)
1708 iter->trace->close(iter);
1710 /* reenable tracing if it was previously enabled */
1712 mutex_unlock(&trace_types_lock);
1714 seq_release(inode, file);
1719 static int tracing_open(struct inode *inode, struct file *file)
1723 __tracing_open(inode, file, &ret);
1728 static int tracing_lt_open(struct inode *inode, struct file *file)
1730 struct trace_iterator *iter;
1733 iter = __tracing_open(inode, file, &ret);
1736 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1743 t_next(struct seq_file *m, void *v, loff_t *pos)
1745 struct tracer *t = m->private;
1757 static void *t_start(struct seq_file *m, loff_t *pos)
1759 struct tracer *t = m->private;
1762 mutex_lock(&trace_types_lock);
1763 for (; t && l < *pos; t = t_next(m, t, &l))
1769 static void t_stop(struct seq_file *m, void *p)
1771 mutex_unlock(&trace_types_lock);
1774 static int t_show(struct seq_file *m, void *v)
1776 struct tracer *t = v;
1781 seq_printf(m, "%s", t->name);
1790 static struct seq_operations show_traces_seq_ops = {
1797 static int show_traces_open(struct inode *inode, struct file *file)
1801 if (tracing_disabled)
1804 ret = seq_open(file, &show_traces_seq_ops);
1806 struct seq_file *m = file->private_data;
1807 m->private = trace_types;
1813 static struct file_operations tracing_fops = {
1814 .open = tracing_open,
1816 .llseek = seq_lseek,
1817 .release = tracing_release,
1820 static struct file_operations tracing_lt_fops = {
1821 .open = tracing_lt_open,
1823 .llseek = seq_lseek,
1824 .release = tracing_release,
1827 static struct file_operations show_traces_fops = {
1828 .open = show_traces_open,
1830 .release = seq_release,
1834 * Only trace on a CPU if the bitmask is set:
1836 static cpumask_var_t tracing_cpumask;
1839 * The tracer itself will not take this lock, but still we want
1840 * to provide a consistent cpumask to user-space:
1842 static DEFINE_MUTEX(tracing_cpumask_update_lock);
1845 * Temporary storage for the character representation of the
1846 * CPU bitmask (and one more byte for the newline):
1848 static char mask_str[NR_CPUS + 1];
1851 tracing_cpumask_read(struct file *filp, char __user *ubuf,
1852 size_t count, loff_t *ppos)
1856 mutex_lock(&tracing_cpumask_update_lock);
1858 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
1859 if (count - len < 2) {
1863 len += sprintf(mask_str + len, "\n");
1864 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
1867 mutex_unlock(&tracing_cpumask_update_lock);
1873 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
1874 size_t count, loff_t *ppos)
1877 cpumask_var_t tracing_cpumask_new;
1879 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
1882 mutex_lock(&tracing_cpumask_update_lock);
1883 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
1887 local_irq_disable();
1888 __raw_spin_lock(&ftrace_max_lock);
1889 for_each_tracing_cpu(cpu) {
1891 * Increase/decrease the disabled counter if we are
1892 * about to flip a bit in the cpumask:
1894 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
1895 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
1896 atomic_inc(&global_trace.data[cpu]->disabled);
1898 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
1899 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
1900 atomic_dec(&global_trace.data[cpu]->disabled);
1903 __raw_spin_unlock(&ftrace_max_lock);
1906 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
1908 mutex_unlock(&tracing_cpumask_update_lock);
1909 free_cpumask_var(tracing_cpumask_new);
1914 mutex_unlock(&tracing_cpumask_update_lock);
1915 free_cpumask_var(tracing_cpumask);
1920 static struct file_operations tracing_cpumask_fops = {
1921 .open = tracing_open_generic,
1922 .read = tracing_cpumask_read,
1923 .write = tracing_cpumask_write,
1927 tracing_trace_options_read(struct file *filp, char __user *ubuf,
1928 size_t cnt, loff_t *ppos)
1934 u32 tracer_flags = current_trace->flags->val;
1935 struct tracer_opt *trace_opts = current_trace->flags->opts;
1938 /* calulate max size */
1939 for (i = 0; trace_options[i]; i++) {
1940 len += strlen(trace_options[i]);
1941 len += 3; /* "no" and space */
1945 * Increase the size with names of options specific
1946 * of the current tracer.
1948 for (i = 0; trace_opts[i].name; i++) {
1949 len += strlen(trace_opts[i].name);
1950 len += 3; /* "no" and space */
1953 /* +2 for \n and \0 */
1954 buf = kmalloc(len + 2, GFP_KERNEL);
1958 for (i = 0; trace_options[i]; i++) {
1959 if (trace_flags & (1 << i))
1960 r += sprintf(buf + r, "%s ", trace_options[i]);
1962 r += sprintf(buf + r, "no%s ", trace_options[i]);
1965 for (i = 0; trace_opts[i].name; i++) {
1966 if (tracer_flags & trace_opts[i].bit)
1967 r += sprintf(buf + r, "%s ",
1968 trace_opts[i].name);
1970 r += sprintf(buf + r, "no%s ",
1971 trace_opts[i].name);
1974 r += sprintf(buf + r, "\n");
1975 WARN_ON(r >= len + 2);
1977 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1984 /* Try to assign a tracer specific option */
1985 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
1987 struct tracer_flags *trace_flags = trace->flags;
1988 struct tracer_opt *opts = NULL;
1992 for (i = 0; trace_flags->opts[i].name; i++) {
1993 opts = &trace_flags->opts[i];
1994 len = strlen(opts->name);
1996 if (strncmp(cmp, opts->name, len) == 0) {
1997 ret = trace->set_flag(trace_flags->val,
2003 if (!trace_flags->opts[i].name)
2006 /* Refused to handle */
2011 trace_flags->val &= ~opts->bit;
2013 trace_flags->val |= opts->bit;
2019 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2020 size_t cnt, loff_t *ppos)
2028 if (cnt >= sizeof(buf))
2031 if (copy_from_user(&buf, ubuf, cnt))
2036 if (strncmp(buf, "no", 2) == 0) {
2041 for (i = 0; trace_options[i]; i++) {
2042 int len = strlen(trace_options[i]);
2044 if (strncmp(cmp, trace_options[i], len) == 0) {
2046 trace_flags &= ~(1 << i);
2048 trace_flags |= (1 << i);
2053 /* If no option could be set, test the specific tracer options */
2054 if (!trace_options[i]) {
2055 ret = set_tracer_option(current_trace, cmp, neg);
2065 static struct file_operations tracing_iter_fops = {
2066 .open = tracing_open_generic,
2067 .read = tracing_trace_options_read,
2068 .write = tracing_trace_options_write,
2071 static const char readme_msg[] =
2072 "tracing mini-HOWTO:\n\n"
2074 "# mount -t debugfs nodev /debug\n\n"
2075 "# cat /debug/tracing/available_tracers\n"
2076 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2077 "# cat /debug/tracing/current_tracer\n"
2079 "# echo sched_switch > /debug/tracing/current_tracer\n"
2080 "# cat /debug/tracing/current_tracer\n"
2082 "# cat /debug/tracing/trace_options\n"
2083 "noprint-parent nosym-offset nosym-addr noverbose\n"
2084 "# echo print-parent > /debug/tracing/trace_options\n"
2085 "# echo 1 > /debug/tracing/tracing_enabled\n"
2086 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2087 "echo 0 > /debug/tracing/tracing_enabled\n"
2091 tracing_readme_read(struct file *filp, char __user *ubuf,
2092 size_t cnt, loff_t *ppos)
2094 return simple_read_from_buffer(ubuf, cnt, ppos,
2095 readme_msg, strlen(readme_msg));
2098 static struct file_operations tracing_readme_fops = {
2099 .open = tracing_open_generic,
2100 .read = tracing_readme_read,
2104 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2105 size_t cnt, loff_t *ppos)
2110 r = sprintf(buf, "%u\n", tracer_enabled);
2111 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2115 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2116 size_t cnt, loff_t *ppos)
2118 struct trace_array *tr = filp->private_data;
2123 if (cnt >= sizeof(buf))
2126 if (copy_from_user(&buf, ubuf, cnt))
2131 ret = strict_strtoul(buf, 10, &val);
2137 mutex_lock(&trace_types_lock);
2138 if (tracer_enabled ^ val) {
2141 if (current_trace->start)
2142 current_trace->start(tr);
2147 if (current_trace->stop)
2148 current_trace->stop(tr);
2151 mutex_unlock(&trace_types_lock);
2159 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2160 size_t cnt, loff_t *ppos)
2162 char buf[max_tracer_type_len+2];
2165 mutex_lock(&trace_types_lock);
2167 r = sprintf(buf, "%s\n", current_trace->name);
2169 r = sprintf(buf, "\n");
2170 mutex_unlock(&trace_types_lock);
2172 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2175 static int tracing_set_tracer(const char *buf)
2177 struct trace_array *tr = &global_trace;
2181 mutex_lock(&trace_types_lock);
2182 for (t = trace_types; t; t = t->next) {
2183 if (strcmp(t->name, buf) == 0)
2190 if (t == current_trace)
2193 trace_branch_disable();
2194 if (current_trace && current_trace->reset)
2195 current_trace->reset(tr);
2204 trace_branch_enable(tr);
2206 mutex_unlock(&trace_types_lock);
2212 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2213 size_t cnt, loff_t *ppos)
2215 char buf[max_tracer_type_len+1];
2222 if (cnt > max_tracer_type_len)
2223 cnt = max_tracer_type_len;
2225 if (copy_from_user(&buf, ubuf, cnt))
2230 /* strip ending whitespace. */
2231 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2234 err = tracing_set_tracer(buf);
2244 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2245 size_t cnt, loff_t *ppos)
2247 unsigned long *ptr = filp->private_data;
2251 r = snprintf(buf, sizeof(buf), "%ld\n",
2252 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2253 if (r > sizeof(buf))
2255 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2259 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2260 size_t cnt, loff_t *ppos)
2262 long *ptr = filp->private_data;
2267 if (cnt >= sizeof(buf))
2270 if (copy_from_user(&buf, ubuf, cnt))
2275 ret = strict_strtoul(buf, 10, &val);
2284 static atomic_t tracing_reader;
2286 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2288 struct trace_iterator *iter;
2290 if (tracing_disabled)
2293 /* We only allow for reader of the pipe */
2294 if (atomic_inc_return(&tracing_reader) != 1) {
2295 atomic_dec(&tracing_reader);
2299 /* create a buffer to store the information to pass to userspace */
2300 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2304 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2309 mutex_lock(&trace_types_lock);
2311 /* trace pipe does not show start of buffer */
2312 cpumask_setall(iter->started);
2314 iter->tr = &global_trace;
2315 iter->trace = current_trace;
2316 filp->private_data = iter;
2318 if (iter->trace->pipe_open)
2319 iter->trace->pipe_open(iter);
2320 mutex_unlock(&trace_types_lock);
2325 static int tracing_release_pipe(struct inode *inode, struct file *file)
2327 struct trace_iterator *iter = file->private_data;
2329 free_cpumask_var(iter->started);
2331 atomic_dec(&tracing_reader);
2337 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2339 struct trace_iterator *iter = filp->private_data;
2341 if (trace_flags & TRACE_ITER_BLOCK) {
2343 * Always select as readable when in blocking mode
2345 return POLLIN | POLLRDNORM;
2347 if (!trace_empty(iter))
2348 return POLLIN | POLLRDNORM;
2349 poll_wait(filp, &trace_wait, poll_table);
2350 if (!trace_empty(iter))
2351 return POLLIN | POLLRDNORM;
2361 tracing_read_pipe(struct file *filp, char __user *ubuf,
2362 size_t cnt, loff_t *ppos)
2364 struct trace_iterator *iter = filp->private_data;
2367 /* return any leftover data */
2368 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2372 trace_seq_reset(&iter->seq);
2374 mutex_lock(&trace_types_lock);
2375 if (iter->trace->read) {
2376 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2383 while (trace_empty(iter)) {
2385 if ((filp->f_flags & O_NONBLOCK)) {
2391 * This is a make-shift waitqueue. The reason we don't use
2392 * an actual wait queue is because:
2393 * 1) we only ever have one waiter
2394 * 2) the tracing, traces all functions, we don't want
2395 * the overhead of calling wake_up and friends
2396 * (and tracing them too)
2397 * Anyway, this is really very primitive wakeup.
2399 set_current_state(TASK_INTERRUPTIBLE);
2400 iter->tr->waiter = current;
2402 mutex_unlock(&trace_types_lock);
2404 /* sleep for 100 msecs, and try again. */
2405 schedule_timeout(HZ/10);
2407 mutex_lock(&trace_types_lock);
2409 iter->tr->waiter = NULL;
2411 if (signal_pending(current)) {
2416 if (iter->trace != current_trace)
2420 * We block until we read something and tracing is disabled.
2421 * We still block if tracing is disabled, but we have never
2422 * read anything. This allows a user to cat this file, and
2423 * then enable tracing. But after we have read something,
2424 * we give an EOF when tracing is again disabled.
2426 * iter->pos will be 0 if we haven't read anything.
2428 if (!tracer_enabled && iter->pos)
2434 /* stop when tracing is finished */
2435 if (trace_empty(iter))
2438 if (cnt >= PAGE_SIZE)
2439 cnt = PAGE_SIZE - 1;
2441 /* reset all but tr, trace, and overruns */
2442 memset(&iter->seq, 0,
2443 sizeof(struct trace_iterator) -
2444 offsetof(struct trace_iterator, seq));
2447 while (find_next_entry_inc(iter) != NULL) {
2448 enum print_line_t ret;
2449 int len = iter->seq.len;
2451 ret = print_trace_line(iter);
2452 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2453 /* don't print partial lines */
2454 iter->seq.len = len;
2458 trace_consume(iter);
2460 if (iter->seq.len >= cnt)
2464 /* Now copy what we have to the user */
2465 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2466 if (iter->seq.readpos >= iter->seq.len)
2467 trace_seq_reset(&iter->seq);
2470 * If there was nothing to send to user, inspite of consuming trace
2471 * entries, go back to wait for more entries.
2477 mutex_unlock(&trace_types_lock);
2483 tracing_entries_read(struct file *filp, char __user *ubuf,
2484 size_t cnt, loff_t *ppos)
2486 struct trace_array *tr = filp->private_data;
2490 r = sprintf(buf, "%lu\n", tr->entries >> 10);
2491 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2495 tracing_entries_write(struct file *filp, const char __user *ubuf,
2496 size_t cnt, loff_t *ppos)
2502 if (cnt >= sizeof(buf))
2505 if (copy_from_user(&buf, ubuf, cnt))
2510 ret = strict_strtoul(buf, 10, &val);
2514 /* must have at least 1 entry */
2518 mutex_lock(&trace_types_lock);
2522 /* disable all cpu buffers */
2523 for_each_tracing_cpu(cpu) {
2524 if (global_trace.data[cpu])
2525 atomic_inc(&global_trace.data[cpu]->disabled);
2526 if (max_tr.data[cpu])
2527 atomic_inc(&max_tr.data[cpu]->disabled);
2530 /* value is in KB */
2533 if (val != global_trace.entries) {
2534 ret = ring_buffer_resize(global_trace.buffer, val);
2540 ret = ring_buffer_resize(max_tr.buffer, val);
2544 r = ring_buffer_resize(global_trace.buffer,
2545 global_trace.entries);
2547 /* AARGH! We are left with different
2548 * size max buffer!!!! */
2550 tracing_disabled = 1;
2555 global_trace.entries = val;
2560 /* If check pages failed, return ENOMEM */
2561 if (tracing_disabled)
2564 for_each_tracing_cpu(cpu) {
2565 if (global_trace.data[cpu])
2566 atomic_dec(&global_trace.data[cpu]->disabled);
2567 if (max_tr.data[cpu])
2568 atomic_dec(&max_tr.data[cpu]->disabled);
2572 max_tr.entries = global_trace.entries;
2573 mutex_unlock(&trace_types_lock);
2578 static int mark_printk(const char *fmt, ...)
2582 va_start(args, fmt);
2583 ret = trace_vprintk(0, -1, fmt, args);
2589 tracing_mark_write(struct file *filp, const char __user *ubuf,
2590 size_t cnt, loff_t *fpos)
2595 if (tracing_disabled)
2598 if (cnt > TRACE_BUF_SIZE)
2599 cnt = TRACE_BUF_SIZE;
2601 buf = kmalloc(cnt + 1, GFP_KERNEL);
2605 if (copy_from_user(buf, ubuf, cnt)) {
2610 /* Cut from the first nil or newline. */
2612 end = strchr(buf, '\n');
2616 cnt = mark_printk("%s\n", buf);
2623 static struct file_operations tracing_max_lat_fops = {
2624 .open = tracing_open_generic,
2625 .read = tracing_max_lat_read,
2626 .write = tracing_max_lat_write,
2629 static struct file_operations tracing_ctrl_fops = {
2630 .open = tracing_open_generic,
2631 .read = tracing_ctrl_read,
2632 .write = tracing_ctrl_write,
2635 static struct file_operations set_tracer_fops = {
2636 .open = tracing_open_generic,
2637 .read = tracing_set_trace_read,
2638 .write = tracing_set_trace_write,
2641 static struct file_operations tracing_pipe_fops = {
2642 .open = tracing_open_pipe,
2643 .poll = tracing_poll_pipe,
2644 .read = tracing_read_pipe,
2645 .release = tracing_release_pipe,
2648 static struct file_operations tracing_entries_fops = {
2649 .open = tracing_open_generic,
2650 .read = tracing_entries_read,
2651 .write = tracing_entries_write,
2654 static struct file_operations tracing_mark_fops = {
2655 .open = tracing_open_generic,
2656 .write = tracing_mark_write,
2659 #ifdef CONFIG_DYNAMIC_FTRACE
2661 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
2667 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
2668 size_t cnt, loff_t *ppos)
2670 static char ftrace_dyn_info_buffer[1024];
2671 static DEFINE_MUTEX(dyn_info_mutex);
2672 unsigned long *p = filp->private_data;
2673 char *buf = ftrace_dyn_info_buffer;
2674 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
2677 mutex_lock(&dyn_info_mutex);
2678 r = sprintf(buf, "%ld ", *p);
2680 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
2683 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2685 mutex_unlock(&dyn_info_mutex);
2690 static struct file_operations tracing_dyn_info_fops = {
2691 .open = tracing_open_generic,
2692 .read = tracing_read_dyn_info,
2696 static struct dentry *d_tracer;
2698 struct dentry *tracing_init_dentry(void)
2705 d_tracer = debugfs_create_dir("tracing", NULL);
2707 if (!d_tracer && !once) {
2709 pr_warning("Could not create debugfs directory 'tracing'\n");
2716 #ifdef CONFIG_FTRACE_SELFTEST
2717 /* Let selftest have access to static functions in this file */
2718 #include "trace_selftest.c"
2721 static __init int tracer_init_debugfs(void)
2723 struct dentry *d_tracer;
2724 struct dentry *entry;
2726 d_tracer = tracing_init_dentry();
2728 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2729 &global_trace, &tracing_ctrl_fops);
2731 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2733 entry = debugfs_create_file("trace_options", 0644, d_tracer,
2734 NULL, &tracing_iter_fops);
2736 pr_warning("Could not create debugfs 'trace_options' entry\n");
2738 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
2739 NULL, &tracing_cpumask_fops);
2741 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2743 entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2744 &global_trace, &tracing_lt_fops);
2746 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2748 entry = debugfs_create_file("trace", 0444, d_tracer,
2749 &global_trace, &tracing_fops);
2751 pr_warning("Could not create debugfs 'trace' entry\n");
2753 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2754 &global_trace, &show_traces_fops);
2756 pr_warning("Could not create debugfs 'available_tracers' entry\n");
2758 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2759 &global_trace, &set_tracer_fops);
2761 pr_warning("Could not create debugfs 'current_tracer' entry\n");
2763 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2764 &tracing_max_latency,
2765 &tracing_max_lat_fops);
2767 pr_warning("Could not create debugfs "
2768 "'tracing_max_latency' entry\n");
2770 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2771 &tracing_thresh, &tracing_max_lat_fops);
2773 pr_warning("Could not create debugfs "
2774 "'tracing_thresh' entry\n");
2775 entry = debugfs_create_file("README", 0644, d_tracer,
2776 NULL, &tracing_readme_fops);
2778 pr_warning("Could not create debugfs 'README' entry\n");
2780 entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2781 NULL, &tracing_pipe_fops);
2783 pr_warning("Could not create debugfs "
2784 "'trace_pipe' entry\n");
2786 entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
2787 &global_trace, &tracing_entries_fops);
2789 pr_warning("Could not create debugfs "
2790 "'buffer_size_kb' entry\n");
2792 entry = debugfs_create_file("trace_marker", 0220, d_tracer,
2793 NULL, &tracing_mark_fops);
2795 pr_warning("Could not create debugfs "
2796 "'trace_marker' entry\n");
2798 #ifdef CONFIG_DYNAMIC_FTRACE
2799 entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2800 &ftrace_update_tot_cnt,
2801 &tracing_dyn_info_fops);
2803 pr_warning("Could not create debugfs "
2804 "'dyn_ftrace_total_info' entry\n");
2806 #ifdef CONFIG_SYSPROF_TRACER
2807 init_tracer_sysprof_debugfs(d_tracer);
2812 int trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args)
2814 static DEFINE_SPINLOCK(trace_buf_lock);
2815 static char trace_buf[TRACE_BUF_SIZE];
2817 struct ring_buffer_event *event;
2818 struct trace_array *tr = &global_trace;
2819 struct trace_array_cpu *data;
2820 int cpu, len = 0, size, pc;
2821 struct print_entry *entry;
2822 unsigned long irq_flags;
2824 if (tracing_disabled || tracing_selftest_running)
2827 pc = preempt_count();
2828 preempt_disable_notrace();
2829 cpu = raw_smp_processor_id();
2830 data = tr->data[cpu];
2832 if (unlikely(atomic_read(&data->disabled)))
2835 pause_graph_tracing();
2836 spin_lock_irqsave(&trace_buf_lock, irq_flags);
2837 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
2839 len = min(len, TRACE_BUF_SIZE-1);
2842 size = sizeof(*entry) + len + 1;
2843 event = ring_buffer_lock_reserve(tr->buffer, size, &irq_flags);
2846 entry = ring_buffer_event_data(event);
2847 tracing_generic_entry_update(&entry->ent, irq_flags, pc);
2848 entry->ent.type = TRACE_PRINT;
2850 entry->depth = depth;
2852 memcpy(&entry->buf, trace_buf, len);
2853 entry->buf[len] = 0;
2854 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
2857 spin_unlock_irqrestore(&trace_buf_lock, irq_flags);
2858 unpause_graph_tracing();
2860 preempt_enable_notrace();
2864 EXPORT_SYMBOL_GPL(trace_vprintk);
2866 int __ftrace_printk(unsigned long ip, const char *fmt, ...)
2871 if (!(trace_flags & TRACE_ITER_PRINTK))
2875 ret = trace_vprintk(ip, task_curr_ret_stack(current), fmt, ap);
2879 EXPORT_SYMBOL_GPL(__ftrace_printk);
2881 int __ftrace_vprintk(unsigned long ip, const char *fmt, va_list ap)
2883 if (!(trace_flags & TRACE_ITER_PRINTK))
2886 return trace_vprintk(ip, task_curr_ret_stack(current), fmt, ap);
2888 EXPORT_SYMBOL_GPL(__ftrace_vprintk);
2890 static int trace_panic_handler(struct notifier_block *this,
2891 unsigned long event, void *unused)
2893 if (ftrace_dump_on_oops)
2898 static struct notifier_block trace_panic_notifier = {
2899 .notifier_call = trace_panic_handler,
2901 .priority = 150 /* priority: INT_MAX >= x >= 0 */
2904 static int trace_die_handler(struct notifier_block *self,
2910 if (ftrace_dump_on_oops)
2919 static struct notifier_block trace_die_notifier = {
2920 .notifier_call = trace_die_handler,
2925 * printk is set to max of 1024, we really don't need it that big.
2926 * Nothing should be printing 1000 characters anyway.
2928 #define TRACE_MAX_PRINT 1000
2931 * Define here KERN_TRACE so that we have one place to modify
2932 * it if we decide to change what log level the ftrace dump
2935 #define KERN_TRACE KERN_EMERG
2938 trace_printk_seq(struct trace_seq *s)
2940 /* Probably should print a warning here. */
2944 /* should be zero ended, but we are paranoid. */
2945 s->buffer[s->len] = 0;
2947 printk(KERN_TRACE "%s", s->buffer);
2952 void ftrace_dump(void)
2954 static DEFINE_SPINLOCK(ftrace_dump_lock);
2955 /* use static because iter can be a bit big for the stack */
2956 static struct trace_iterator iter;
2957 static int dump_ran;
2958 unsigned long flags;
2962 spin_lock_irqsave(&ftrace_dump_lock, flags);
2968 /* No turning back! */
2972 for_each_tracing_cpu(cpu) {
2973 atomic_inc(&global_trace.data[cpu]->disabled);
2976 /* don't look at user memory in panic mode */
2977 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
2979 printk(KERN_TRACE "Dumping ftrace buffer:\n");
2981 iter.tr = &global_trace;
2982 iter.trace = current_trace;
2985 * We need to stop all tracing on all CPUS to read the
2986 * the next buffer. This is a bit expensive, but is
2987 * not done often. We fill all what we can read,
2988 * and then release the locks again.
2991 while (!trace_empty(&iter)) {
2994 printk(KERN_TRACE "---------------------------------\n");
2998 /* reset all but tr, trace, and overruns */
2999 memset(&iter.seq, 0,
3000 sizeof(struct trace_iterator) -
3001 offsetof(struct trace_iterator, seq));
3002 iter.iter_flags |= TRACE_FILE_LAT_FMT;
3005 if (find_next_entry_inc(&iter) != NULL) {
3006 print_trace_line(&iter);
3007 trace_consume(&iter);
3010 trace_printk_seq(&iter.seq);
3014 printk(KERN_TRACE " (ftrace buffer empty)\n");
3016 printk(KERN_TRACE "---------------------------------\n");
3019 spin_unlock_irqrestore(&ftrace_dump_lock, flags);
3022 __init static int tracer_alloc_buffers(void)
3024 struct trace_array_cpu *data;
3028 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
3031 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
3032 goto out_free_buffer_mask;
3034 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
3035 cpumask_copy(tracing_cpumask, cpu_all_mask);
3037 /* TODO: make the number of buffers hot pluggable with CPUS */
3038 global_trace.buffer = ring_buffer_alloc(trace_buf_size,
3039 TRACE_BUFFER_FLAGS);
3040 if (!global_trace.buffer) {
3041 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
3043 goto out_free_cpumask;
3045 global_trace.entries = ring_buffer_size(global_trace.buffer);
3048 #ifdef CONFIG_TRACER_MAX_TRACE
3049 max_tr.buffer = ring_buffer_alloc(trace_buf_size,
3050 TRACE_BUFFER_FLAGS);
3051 if (!max_tr.buffer) {
3052 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
3054 ring_buffer_free(global_trace.buffer);
3055 goto out_free_cpumask;
3057 max_tr.entries = ring_buffer_size(max_tr.buffer);
3058 WARN_ON(max_tr.entries != global_trace.entries);
3061 /* Allocate the first page for all buffers */
3062 for_each_tracing_cpu(i) {
3063 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
3064 max_tr.data[i] = &per_cpu(max_data, i);
3067 trace_init_cmdlines();
3069 register_tracer(&nop_trace);
3070 current_trace = &nop_trace;
3071 #ifdef CONFIG_BOOT_TRACER
3072 register_tracer(&boot_tracer);
3074 /* All seems OK, enable tracing */
3075 tracing_disabled = 0;
3077 atomic_notifier_chain_register(&panic_notifier_list,
3078 &trace_panic_notifier);
3080 register_die_notifier(&trace_die_notifier);
3084 free_cpumask_var(tracing_cpumask);
3085 out_free_buffer_mask:
3086 free_cpumask_var(tracing_buffer_mask);
3091 __init static int clear_boot_tracer(void)
3094 * The default tracer at boot buffer is an init section.
3095 * This function is called in lateinit. If we did not
3096 * find the boot tracer, then clear it out, to prevent
3097 * later registration from accessing the buffer that is
3098 * about to be freed.
3100 if (!default_bootup_tracer)
3103 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
3104 default_bootup_tracer);
3105 default_bootup_tracer = NULL;
3110 early_initcall(tracer_alloc_buffers);
3111 fs_initcall(tracer_init_debugfs);
3112 late_initcall(clear_boot_tracer);