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/ring_buffer.h>
15 #include <linux/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/smp_lock.h>
21 #include <linux/notifier.h>
22 #include <linux/irqflags.h>
23 #include <linux/debugfs.h>
24 #include <linux/pagemap.h>
25 #include <linux/hardirq.h>
26 #include <linux/linkage.h>
27 #include <linux/uaccess.h>
28 #include <linux/kprobes.h>
29 #include <linux/ftrace.h>
30 #include <linux/module.h>
31 #include <linux/percpu.h>
32 #include <linux/splice.h>
33 #include <linux/kdebug.h>
34 #include <linux/string.h>
35 #include <linux/ctype.h>
36 #include <linux/init.h>
37 #include <linux/poll.h>
38 #include <linux/gfp.h>
42 #include "trace_output.h"
44 #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
47 * On boot up, the ring buffer is set to the minimum size, so that
48 * we do not waste memory on systems that are not using tracing.
50 int ring_buffer_expanded;
53 * We need to change this state when a selftest is running.
54 * A selftest will lurk into the ring-buffer to count the
55 * entries inserted during the selftest although some concurrent
56 * insertions into the ring-buffer such as trace_printk could occurred
57 * at the same time, giving false positive or negative results.
59 static bool __read_mostly tracing_selftest_running;
62 * If a tracer is running, we do not want to run SELFTEST.
64 bool __read_mostly tracing_selftest_disabled;
66 /* For tracers that don't implement custom flags */
67 static struct tracer_opt dummy_tracer_opt[] = {
71 static struct tracer_flags dummy_tracer_flags = {
73 .opts = dummy_tracer_opt
76 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
82 * Kill all tracing for good (never come back).
83 * It is initialized to 1 but will turn to zero if the initialization
84 * of the tracer is successful. But that is the only place that sets
87 static int tracing_disabled = 1;
89 DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
91 static inline void ftrace_disable_cpu(void)
94 local_inc(&__get_cpu_var(ftrace_cpu_disabled));
97 static inline void ftrace_enable_cpu(void)
99 local_dec(&__get_cpu_var(ftrace_cpu_disabled));
103 static cpumask_var_t __read_mostly tracing_buffer_mask;
105 /* Define which cpu buffers are currently read in trace_pipe */
106 static cpumask_var_t tracing_reader_cpumask;
108 #define for_each_tracing_cpu(cpu) \
109 for_each_cpu(cpu, tracing_buffer_mask)
112 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
114 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
115 * is set, then ftrace_dump is called. This will output the contents
116 * of the ftrace buffers to the console. This is very useful for
117 * capturing traces that lead to crashes and outputing it to a
120 * It is default off, but you can enable it with either specifying
121 * "ftrace_dump_on_oops" in the kernel command line, or setting
122 * /proc/sys/kernel/ftrace_dump_on_oops to true.
124 int ftrace_dump_on_oops;
126 static int tracing_set_tracer(const char *buf);
128 #define MAX_TRACER_SIZE 100
129 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
130 static char *default_bootup_tracer;
132 static int __init set_cmdline_ftrace(char *str)
134 strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
135 default_bootup_tracer = bootup_tracer_buf;
136 /* We are using ftrace early, expand it */
137 ring_buffer_expanded = 1;
140 __setup("ftrace=", set_cmdline_ftrace);
142 static int __init set_ftrace_dump_on_oops(char *str)
144 ftrace_dump_on_oops = 1;
147 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
149 unsigned long long ns2usecs(cycle_t nsec)
157 * The global_trace is the descriptor that holds the tracing
158 * buffers for the live tracing. For each CPU, it contains
159 * a link list of pages that will store trace entries. The
160 * page descriptor of the pages in the memory is used to hold
161 * the link list by linking the lru item in the page descriptor
162 * to each of the pages in the buffer per CPU.
164 * For each active CPU there is a data field that holds the
165 * pages for the buffer for that CPU. Each CPU has the same number
166 * of pages allocated for its buffer.
168 static struct trace_array global_trace;
170 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
172 int filter_current_check_discard(struct ring_buffer *buffer,
173 struct ftrace_event_call *call, void *rec,
174 struct ring_buffer_event *event)
176 return filter_check_discard(call, rec, buffer, event);
178 EXPORT_SYMBOL_GPL(filter_current_check_discard);
180 cycle_t ftrace_now(int cpu)
184 /* Early boot up does not have a buffer yet */
185 if (!global_trace.buffer)
186 return trace_clock_local();
188 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
189 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
195 * The max_tr is used to snapshot the global_trace when a maximum
196 * latency is reached. Some tracers will use this to store a maximum
197 * trace while it continues examining live traces.
199 * The buffers for the max_tr are set up the same as the global_trace.
200 * When a snapshot is taken, the link list of the max_tr is swapped
201 * with the link list of the global_trace and the buffers are reset for
202 * the global_trace so the tracing can continue.
204 static struct trace_array max_tr;
206 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
208 /* tracer_enabled is used to toggle activation of a tracer */
209 static int tracer_enabled = 1;
212 * tracing_is_enabled - return tracer_enabled status
214 * This function is used by other tracers to know the status
215 * of the tracer_enabled flag. Tracers may use this function
216 * to know if it should enable their features when starting
217 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
219 int tracing_is_enabled(void)
221 return tracer_enabled;
225 * trace_buf_size is the size in bytes that is allocated
226 * for a buffer. Note, the number of bytes is always rounded
229 * This number is purposely set to a low number of 16384.
230 * If the dump on oops happens, it will be much appreciated
231 * to not have to wait for all that output. Anyway this can be
232 * boot time and run time configurable.
234 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
236 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
238 /* trace_types holds a link list of available tracers. */
239 static struct tracer *trace_types __read_mostly;
241 /* current_trace points to the tracer that is currently active */
242 static struct tracer *current_trace __read_mostly;
245 * trace_types_lock is used to protect the trace_types list.
246 * This lock is also used to keep user access serialized.
247 * Accesses from userspace will grab this lock while userspace
248 * activities happen inside the kernel.
250 static DEFINE_MUTEX(trace_types_lock);
252 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
253 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
255 /* trace_flags holds trace_options default values */
256 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
257 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
258 TRACE_ITER_GRAPH_TIME;
260 static int trace_stop_count;
261 static DEFINE_SPINLOCK(tracing_start_lock);
264 * trace_wake_up - wake up tasks waiting for trace input
266 * Simply wakes up any task that is blocked on the trace_wait
267 * queue. These is used with trace_poll for tasks polling the trace.
269 void trace_wake_up(void)
273 if (trace_flags & TRACE_ITER_BLOCK)
276 * The runqueue_is_locked() can fail, but this is the best we
280 if (!runqueue_is_locked(cpu))
281 wake_up(&trace_wait);
285 static int __init set_buf_size(char *str)
287 unsigned long buf_size;
291 buf_size = memparse(str, &str);
292 /* nr_entries can not be zero */
295 trace_buf_size = buf_size;
298 __setup("trace_buf_size=", set_buf_size);
300 unsigned long nsecs_to_usecs(unsigned long nsecs)
305 /* These must match the bit postions in trace_iterator_flags */
306 static const char *trace_options[] = {
334 { trace_clock_local, "local" },
335 { trace_clock_global, "global" },
341 * trace_parser_get_init - gets the buffer for trace parser
343 int trace_parser_get_init(struct trace_parser *parser, int size)
345 memset(parser, 0, sizeof(*parser));
347 parser->buffer = kmalloc(size, GFP_KERNEL);
356 * trace_parser_put - frees the buffer for trace parser
358 void trace_parser_put(struct trace_parser *parser)
360 kfree(parser->buffer);
364 * trace_get_user - reads the user input string separated by space
365 * (matched by isspace(ch))
367 * For each string found the 'struct trace_parser' is updated,
368 * and the function returns.
370 * Returns number of bytes read.
372 * See kernel/trace/trace.h for 'struct trace_parser' details.
374 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
375 size_t cnt, loff_t *ppos)
382 trace_parser_clear(parser);
384 ret = get_user(ch, ubuf++);
392 * The parser is not finished with the last write,
393 * continue reading the user input without skipping spaces.
396 /* skip white space */
397 while (cnt && isspace(ch)) {
398 ret = get_user(ch, ubuf++);
405 /* only spaces were written */
415 /* read the non-space input */
416 while (cnt && !isspace(ch)) {
417 if (parser->idx < parser->size - 1)
418 parser->buffer[parser->idx++] = ch;
423 ret = get_user(ch, ubuf++);
430 /* We either got finished input or we have to wait for another call. */
432 parser->buffer[parser->idx] = 0;
433 parser->cont = false;
436 parser->buffer[parser->idx++] = ch;
446 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
454 if (s->len <= s->readpos)
457 len = s->len - s->readpos;
460 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
470 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
475 if (s->len <= s->readpos)
478 len = s->len - s->readpos;
481 ret = memcpy(buf, s->buffer + s->readpos, cnt);
490 * ftrace_max_lock is used to protect the swapping of buffers
491 * when taking a max snapshot. The buffers themselves are
492 * protected by per_cpu spinlocks. But the action of the swap
493 * needs its own lock.
495 * This is defined as a raw_spinlock_t in order to help
496 * with performance when lockdep debugging is enabled.
498 * It is also used in other places outside the update_max_tr
499 * so it needs to be defined outside of the
500 * CONFIG_TRACER_MAX_TRACE.
502 static raw_spinlock_t ftrace_max_lock =
503 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
505 #ifdef CONFIG_TRACER_MAX_TRACE
506 unsigned long __read_mostly tracing_max_latency;
507 unsigned long __read_mostly tracing_thresh;
510 * Copy the new maximum trace into the separate maximum-trace
511 * structure. (this way the maximum trace is permanently saved,
512 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
515 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
517 struct trace_array_cpu *data = tr->data[cpu];
518 struct trace_array_cpu *max_data = tr->data[cpu];
521 max_tr.time_start = data->preempt_timestamp;
523 max_data = max_tr.data[cpu];
524 max_data->saved_latency = tracing_max_latency;
525 max_data->critical_start = data->critical_start;
526 max_data->critical_end = data->critical_end;
528 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
529 max_data->pid = tsk->pid;
530 max_data->uid = task_uid(tsk);
531 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
532 max_data->policy = tsk->policy;
533 max_data->rt_priority = tsk->rt_priority;
535 /* record this tasks comm */
536 tracing_record_cmdline(tsk);
540 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
542 * @tsk: the task with the latency
543 * @cpu: The cpu that initiated the trace.
545 * Flip the buffers between the @tr and the max_tr and record information
546 * about which task was the cause of this latency.
549 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
551 struct ring_buffer *buf = tr->buffer;
553 if (trace_stop_count)
556 WARN_ON_ONCE(!irqs_disabled());
557 __raw_spin_lock(&ftrace_max_lock);
559 tr->buffer = max_tr.buffer;
562 __update_max_tr(tr, tsk, cpu);
563 __raw_spin_unlock(&ftrace_max_lock);
567 * update_max_tr_single - only copy one trace over, and reset the rest
569 * @tsk - task with the latency
570 * @cpu - the cpu of the buffer to copy.
572 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
575 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
579 if (trace_stop_count)
582 WARN_ON_ONCE(!irqs_disabled());
583 __raw_spin_lock(&ftrace_max_lock);
585 ftrace_disable_cpu();
587 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
591 * We failed to swap the buffer due to a commit taking
592 * place on this CPU. We fail to record, but we reset
593 * the max trace buffer (no one writes directly to it)
594 * and flag that it failed.
596 trace_array_printk(&max_tr, _THIS_IP_,
597 "Failed to swap buffers due to commit in progress\n");
602 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
604 __update_max_tr(tr, tsk, cpu);
605 __raw_spin_unlock(&ftrace_max_lock);
607 #endif /* CONFIG_TRACER_MAX_TRACE */
610 * register_tracer - register a tracer with the ftrace system.
611 * @type - the plugin for the tracer
613 * Register a new plugin tracer.
615 int register_tracer(struct tracer *type)
616 __releases(kernel_lock)
617 __acquires(kernel_lock)
623 pr_info("Tracer must have a name\n");
627 if (strlen(type->name) > MAX_TRACER_SIZE) {
628 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
633 * When this gets called we hold the BKL which means that
634 * preemption is disabled. Various trace selftests however
635 * need to disable and enable preemption for successful tests.
636 * So we drop the BKL here and grab it after the tests again.
639 mutex_lock(&trace_types_lock);
641 tracing_selftest_running = true;
643 for (t = trace_types; t; t = t->next) {
644 if (strcmp(type->name, t->name) == 0) {
646 pr_info("Tracer %s already registered\n",
654 type->set_flag = &dummy_set_flag;
656 type->flags = &dummy_tracer_flags;
658 if (!type->flags->opts)
659 type->flags->opts = dummy_tracer_opt;
660 if (!type->wait_pipe)
661 type->wait_pipe = default_wait_pipe;
664 #ifdef CONFIG_FTRACE_STARTUP_TEST
665 if (type->selftest && !tracing_selftest_disabled) {
666 struct tracer *saved_tracer = current_trace;
667 struct trace_array *tr = &global_trace;
670 * Run a selftest on this tracer.
671 * Here we reset the trace buffer, and set the current
672 * tracer to be this tracer. The tracer can then run some
673 * internal tracing to verify that everything is in order.
674 * If we fail, we do not register this tracer.
676 tracing_reset_online_cpus(tr);
678 current_trace = type;
679 /* the test is responsible for initializing and enabling */
680 pr_info("Testing tracer %s: ", type->name);
681 ret = type->selftest(type, tr);
682 /* the test is responsible for resetting too */
683 current_trace = saved_tracer;
685 printk(KERN_CONT "FAILED!\n");
688 /* Only reset on passing, to avoid touching corrupted buffers */
689 tracing_reset_online_cpus(tr);
691 printk(KERN_CONT "PASSED\n");
695 type->next = trace_types;
699 tracing_selftest_running = false;
700 mutex_unlock(&trace_types_lock);
702 if (ret || !default_bootup_tracer)
705 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
708 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
709 /* Do we want this tracer to start on bootup? */
710 tracing_set_tracer(type->name);
711 default_bootup_tracer = NULL;
712 /* disable other selftests, since this will break it. */
713 tracing_selftest_disabled = 1;
714 #ifdef CONFIG_FTRACE_STARTUP_TEST
715 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
724 void unregister_tracer(struct tracer *type)
728 mutex_lock(&trace_types_lock);
729 for (t = &trace_types; *t; t = &(*t)->next) {
733 pr_info("Tracer %s not registered\n", type->name);
739 if (type == current_trace && tracer_enabled) {
742 if (current_trace->stop)
743 current_trace->stop(&global_trace);
744 current_trace = &nop_trace;
747 mutex_unlock(&trace_types_lock);
750 static void __tracing_reset(struct trace_array *tr, int cpu)
752 ftrace_disable_cpu();
753 ring_buffer_reset_cpu(tr->buffer, cpu);
757 void tracing_reset(struct trace_array *tr, int cpu)
759 struct ring_buffer *buffer = tr->buffer;
761 ring_buffer_record_disable(buffer);
763 /* Make sure all commits have finished */
765 __tracing_reset(tr, cpu);
767 ring_buffer_record_enable(buffer);
770 void tracing_reset_online_cpus(struct trace_array *tr)
772 struct ring_buffer *buffer = tr->buffer;
775 ring_buffer_record_disable(buffer);
777 /* Make sure all commits have finished */
780 tr->time_start = ftrace_now(tr->cpu);
782 for_each_online_cpu(cpu)
783 __tracing_reset(tr, cpu);
785 ring_buffer_record_enable(buffer);
788 void tracing_reset_current(int cpu)
790 tracing_reset(&global_trace, cpu);
793 void tracing_reset_current_online_cpus(void)
795 tracing_reset_online_cpus(&global_trace);
798 #define SAVED_CMDLINES 128
799 #define NO_CMDLINE_MAP UINT_MAX
800 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
801 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
802 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
803 static int cmdline_idx;
804 static raw_spinlock_t trace_cmdline_lock = __RAW_SPIN_LOCK_UNLOCKED;
806 /* temporary disable recording */
807 static atomic_t trace_record_cmdline_disabled __read_mostly;
809 static void trace_init_cmdlines(void)
811 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
812 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
816 int is_tracing_stopped(void)
818 return trace_stop_count;
822 * ftrace_off_permanent - disable all ftrace code permanently
824 * This should only be called when a serious anomally has
825 * been detected. This will turn off the function tracing,
826 * ring buffers, and other tracing utilites. It takes no
827 * locks and can be called from any context.
829 void ftrace_off_permanent(void)
831 tracing_disabled = 1;
833 tracing_off_permanent();
837 * tracing_start - quick start of the tracer
839 * If tracing is enabled but was stopped by tracing_stop,
840 * this will start the tracer back up.
842 void tracing_start(void)
844 struct ring_buffer *buffer;
847 if (tracing_disabled)
850 spin_lock_irqsave(&tracing_start_lock, flags);
851 if (--trace_stop_count) {
852 if (trace_stop_count < 0) {
853 /* Someone screwed up their debugging */
855 trace_stop_count = 0;
861 buffer = global_trace.buffer;
863 ring_buffer_record_enable(buffer);
865 buffer = max_tr.buffer;
867 ring_buffer_record_enable(buffer);
871 spin_unlock_irqrestore(&tracing_start_lock, flags);
875 * tracing_stop - quick stop of the tracer
877 * Light weight way to stop tracing. Use in conjunction with
880 void tracing_stop(void)
882 struct ring_buffer *buffer;
886 spin_lock_irqsave(&tracing_start_lock, flags);
887 if (trace_stop_count++)
890 buffer = global_trace.buffer;
892 ring_buffer_record_disable(buffer);
894 buffer = max_tr.buffer;
896 ring_buffer_record_disable(buffer);
899 spin_unlock_irqrestore(&tracing_start_lock, flags);
902 void trace_stop_cmdline_recording(void);
904 static void trace_save_cmdline(struct task_struct *tsk)
908 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
912 * It's not the end of the world if we don't get
913 * the lock, but we also don't want to spin
914 * nor do we want to disable interrupts,
915 * so if we miss here, then better luck next time.
917 if (!__raw_spin_trylock(&trace_cmdline_lock))
920 idx = map_pid_to_cmdline[tsk->pid];
921 if (idx == NO_CMDLINE_MAP) {
922 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
925 * Check whether the cmdline buffer at idx has a pid
926 * mapped. We are going to overwrite that entry so we
927 * need to clear the map_pid_to_cmdline. Otherwise we
928 * would read the new comm for the old pid.
930 pid = map_cmdline_to_pid[idx];
931 if (pid != NO_CMDLINE_MAP)
932 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
934 map_cmdline_to_pid[idx] = tsk->pid;
935 map_pid_to_cmdline[tsk->pid] = idx;
940 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
942 __raw_spin_unlock(&trace_cmdline_lock);
945 void trace_find_cmdline(int pid, char comm[])
950 strcpy(comm, "<idle>");
954 if (pid > PID_MAX_DEFAULT) {
955 strcpy(comm, "<...>");
960 __raw_spin_lock(&trace_cmdline_lock);
961 map = map_pid_to_cmdline[pid];
962 if (map != NO_CMDLINE_MAP)
963 strcpy(comm, saved_cmdlines[map]);
965 strcpy(comm, "<...>");
967 __raw_spin_unlock(&trace_cmdline_lock);
971 void tracing_record_cmdline(struct task_struct *tsk)
973 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
977 trace_save_cmdline(tsk);
981 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
984 struct task_struct *tsk = current;
986 entry->preempt_count = pc & 0xff;
987 entry->pid = (tsk) ? tsk->pid : 0;
988 entry->lock_depth = (tsk) ? tsk->lock_depth : 0;
990 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
991 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
993 TRACE_FLAG_IRQS_NOSUPPORT |
995 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
996 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
997 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
999 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1001 struct ring_buffer_event *
1002 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1005 unsigned long flags, int pc)
1007 struct ring_buffer_event *event;
1009 event = ring_buffer_lock_reserve(buffer, len);
1010 if (event != NULL) {
1011 struct trace_entry *ent = ring_buffer_event_data(event);
1013 tracing_generic_entry_update(ent, flags, pc);
1021 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1022 struct ring_buffer_event *event,
1023 unsigned long flags, int pc,
1026 ring_buffer_unlock_commit(buffer, event);
1028 ftrace_trace_stack(buffer, flags, 6, pc);
1029 ftrace_trace_userstack(buffer, flags, pc);
1035 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1036 struct ring_buffer_event *event,
1037 unsigned long flags, int pc)
1039 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1042 struct ring_buffer_event *
1043 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1044 int type, unsigned long len,
1045 unsigned long flags, int pc)
1047 *current_rb = global_trace.buffer;
1048 return trace_buffer_lock_reserve(*current_rb,
1049 type, len, flags, pc);
1051 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1053 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1054 struct ring_buffer_event *event,
1055 unsigned long flags, int pc)
1057 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1059 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1061 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1062 struct ring_buffer_event *event,
1063 unsigned long flags, int pc)
1065 __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1067 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1069 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1070 struct ring_buffer_event *event)
1072 ring_buffer_discard_commit(buffer, event);
1074 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1077 trace_function(struct trace_array *tr,
1078 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1081 struct ftrace_event_call *call = &event_function;
1082 struct ring_buffer *buffer = tr->buffer;
1083 struct ring_buffer_event *event;
1084 struct ftrace_entry *entry;
1086 /* If we are reading the ring buffer, don't trace */
1087 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
1090 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1094 entry = ring_buffer_event_data(event);
1096 entry->parent_ip = parent_ip;
1098 if (!filter_check_discard(call, entry, buffer, event))
1099 ring_buffer_unlock_commit(buffer, event);
1103 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1104 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1107 if (likely(!atomic_read(&data->disabled)))
1108 trace_function(tr, ip, parent_ip, flags, pc);
1111 #ifdef CONFIG_STACKTRACE
1112 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1113 unsigned long flags,
1116 struct ftrace_event_call *call = &event_kernel_stack;
1117 struct ring_buffer_event *event;
1118 struct stack_entry *entry;
1119 struct stack_trace trace;
1121 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1122 sizeof(*entry), flags, pc);
1125 entry = ring_buffer_event_data(event);
1126 memset(&entry->caller, 0, sizeof(entry->caller));
1128 trace.nr_entries = 0;
1129 trace.max_entries = FTRACE_STACK_ENTRIES;
1131 trace.entries = entry->caller;
1133 save_stack_trace(&trace);
1134 if (!filter_check_discard(call, entry, buffer, event))
1135 ring_buffer_unlock_commit(buffer, event);
1138 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1141 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1144 __ftrace_trace_stack(buffer, flags, skip, pc);
1147 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1150 __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1154 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1156 struct ftrace_event_call *call = &event_user_stack;
1157 struct ring_buffer_event *event;
1158 struct userstack_entry *entry;
1159 struct stack_trace trace;
1161 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1164 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1165 sizeof(*entry), flags, pc);
1168 entry = ring_buffer_event_data(event);
1170 entry->tgid = current->tgid;
1171 memset(&entry->caller, 0, sizeof(entry->caller));
1173 trace.nr_entries = 0;
1174 trace.max_entries = FTRACE_STACK_ENTRIES;
1176 trace.entries = entry->caller;
1178 save_stack_trace_user(&trace);
1179 if (!filter_check_discard(call, entry, buffer, event))
1180 ring_buffer_unlock_commit(buffer, event);
1184 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1186 ftrace_trace_userstack(tr, flags, preempt_count());
1190 #endif /* CONFIG_STACKTRACE */
1193 ftrace_trace_special(void *__tr,
1194 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1197 struct ftrace_event_call *call = &event_special;
1198 struct ring_buffer_event *event;
1199 struct trace_array *tr = __tr;
1200 struct ring_buffer *buffer = tr->buffer;
1201 struct special_entry *entry;
1203 event = trace_buffer_lock_reserve(buffer, TRACE_SPECIAL,
1204 sizeof(*entry), 0, pc);
1207 entry = ring_buffer_event_data(event);
1212 if (!filter_check_discard(call, entry, buffer, event))
1213 trace_buffer_unlock_commit(buffer, event, 0, pc);
1217 __trace_special(void *__tr, void *__data,
1218 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1220 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1224 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1226 struct trace_array *tr = &global_trace;
1227 struct trace_array_cpu *data;
1228 unsigned long flags;
1232 if (tracing_disabled)
1235 pc = preempt_count();
1236 local_irq_save(flags);
1237 cpu = raw_smp_processor_id();
1238 data = tr->data[cpu];
1240 if (likely(atomic_inc_return(&data->disabled) == 1))
1241 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1243 atomic_dec(&data->disabled);
1244 local_irq_restore(flags);
1248 * trace_vbprintk - write binary msg to tracing buffer
1251 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1253 static raw_spinlock_t trace_buf_lock =
1254 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
1255 static u32 trace_buf[TRACE_BUF_SIZE];
1257 struct ftrace_event_call *call = &event_bprint;
1258 struct ring_buffer_event *event;
1259 struct ring_buffer *buffer;
1260 struct trace_array *tr = &global_trace;
1261 struct trace_array_cpu *data;
1262 struct bprint_entry *entry;
1263 unsigned long flags;
1266 int cpu, len = 0, size, pc;
1268 if (unlikely(tracing_selftest_running || tracing_disabled))
1271 /* Don't pollute graph traces with trace_vprintk internals */
1272 pause_graph_tracing();
1274 pc = preempt_count();
1275 resched = ftrace_preempt_disable();
1276 cpu = raw_smp_processor_id();
1277 data = tr->data[cpu];
1279 disable = atomic_inc_return(&data->disabled);
1280 if (unlikely(disable != 1))
1283 /* Lockdep uses trace_printk for lock tracing */
1284 local_irq_save(flags);
1285 __raw_spin_lock(&trace_buf_lock);
1286 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1288 if (len > TRACE_BUF_SIZE || len < 0)
1291 size = sizeof(*entry) + sizeof(u32) * len;
1292 buffer = tr->buffer;
1293 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1297 entry = ring_buffer_event_data(event);
1301 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1302 if (!filter_check_discard(call, entry, buffer, event))
1303 ring_buffer_unlock_commit(buffer, event);
1306 __raw_spin_unlock(&trace_buf_lock);
1307 local_irq_restore(flags);
1310 atomic_dec_return(&data->disabled);
1311 ftrace_preempt_enable(resched);
1312 unpause_graph_tracing();
1316 EXPORT_SYMBOL_GPL(trace_vbprintk);
1318 int trace_array_printk(struct trace_array *tr,
1319 unsigned long ip, const char *fmt, ...)
1324 if (!(trace_flags & TRACE_ITER_PRINTK))
1328 ret = trace_array_vprintk(tr, ip, fmt, ap);
1333 int trace_array_vprintk(struct trace_array *tr,
1334 unsigned long ip, const char *fmt, va_list args)
1336 static raw_spinlock_t trace_buf_lock = __RAW_SPIN_LOCK_UNLOCKED;
1337 static char trace_buf[TRACE_BUF_SIZE];
1339 struct ftrace_event_call *call = &event_print;
1340 struct ring_buffer_event *event;
1341 struct ring_buffer *buffer;
1342 struct trace_array_cpu *data;
1343 int cpu, len = 0, size, pc;
1344 struct print_entry *entry;
1345 unsigned long irq_flags;
1348 if (tracing_disabled || tracing_selftest_running)
1351 pc = preempt_count();
1352 preempt_disable_notrace();
1353 cpu = raw_smp_processor_id();
1354 data = tr->data[cpu];
1356 disable = atomic_inc_return(&data->disabled);
1357 if (unlikely(disable != 1))
1360 pause_graph_tracing();
1361 raw_local_irq_save(irq_flags);
1362 __raw_spin_lock(&trace_buf_lock);
1363 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1365 size = sizeof(*entry) + len + 1;
1366 buffer = tr->buffer;
1367 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1371 entry = ring_buffer_event_data(event);
1374 memcpy(&entry->buf, trace_buf, len);
1375 entry->buf[len] = '\0';
1376 if (!filter_check_discard(call, entry, buffer, event))
1377 ring_buffer_unlock_commit(buffer, event);
1380 __raw_spin_unlock(&trace_buf_lock);
1381 raw_local_irq_restore(irq_flags);
1382 unpause_graph_tracing();
1384 atomic_dec_return(&data->disabled);
1385 preempt_enable_notrace();
1390 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1392 return trace_array_vprintk(&global_trace, ip, fmt, args);
1394 EXPORT_SYMBOL_GPL(trace_vprintk);
1396 enum trace_file_type {
1397 TRACE_FILE_LAT_FMT = 1,
1398 TRACE_FILE_ANNOTATE = 2,
1401 static void trace_iterator_increment(struct trace_iterator *iter)
1403 /* Don't allow ftrace to trace into the ring buffers */
1404 ftrace_disable_cpu();
1407 if (iter->buffer_iter[iter->cpu])
1408 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1410 ftrace_enable_cpu();
1413 static struct trace_entry *
1414 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1416 struct ring_buffer_event *event;
1417 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1419 /* Don't allow ftrace to trace into the ring buffers */
1420 ftrace_disable_cpu();
1423 event = ring_buffer_iter_peek(buf_iter, ts);
1425 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1427 ftrace_enable_cpu();
1429 return event ? ring_buffer_event_data(event) : NULL;
1432 static struct trace_entry *
1433 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1435 struct ring_buffer *buffer = iter->tr->buffer;
1436 struct trace_entry *ent, *next = NULL;
1437 int cpu_file = iter->cpu_file;
1438 u64 next_ts = 0, ts;
1443 * If we are in a per_cpu trace file, don't bother by iterating over
1444 * all cpu and peek directly.
1446 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1447 if (ring_buffer_empty_cpu(buffer, cpu_file))
1449 ent = peek_next_entry(iter, cpu_file, ent_ts);
1451 *ent_cpu = cpu_file;
1456 for_each_tracing_cpu(cpu) {
1458 if (ring_buffer_empty_cpu(buffer, cpu))
1461 ent = peek_next_entry(iter, cpu, &ts);
1464 * Pick the entry with the smallest timestamp:
1466 if (ent && (!next || ts < next_ts)) {
1474 *ent_cpu = next_cpu;
1482 /* Find the next real entry, without updating the iterator itself */
1483 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1484 int *ent_cpu, u64 *ent_ts)
1486 return __find_next_entry(iter, ent_cpu, ent_ts);
1489 /* Find the next real entry, and increment the iterator to the next entry */
1490 static void *find_next_entry_inc(struct trace_iterator *iter)
1492 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1495 trace_iterator_increment(iter);
1497 return iter->ent ? iter : NULL;
1500 static void trace_consume(struct trace_iterator *iter)
1502 /* Don't allow ftrace to trace into the ring buffers */
1503 ftrace_disable_cpu();
1504 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1505 ftrace_enable_cpu();
1508 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1510 struct trace_iterator *iter = m->private;
1514 WARN_ON_ONCE(iter->leftover);
1518 /* can't go backwards */
1523 ent = find_next_entry_inc(iter);
1527 while (ent && iter->idx < i)
1528 ent = find_next_entry_inc(iter);
1535 static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1537 struct trace_array *tr = iter->tr;
1538 struct ring_buffer_event *event;
1539 struct ring_buffer_iter *buf_iter;
1540 unsigned long entries = 0;
1543 tr->data[cpu]->skipped_entries = 0;
1545 if (!iter->buffer_iter[cpu])
1548 buf_iter = iter->buffer_iter[cpu];
1549 ring_buffer_iter_reset(buf_iter);
1552 * We could have the case with the max latency tracers
1553 * that a reset never took place on a cpu. This is evident
1554 * by the timestamp being before the start of the buffer.
1556 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1557 if (ts >= iter->tr->time_start)
1560 ring_buffer_read(buf_iter, NULL);
1563 tr->data[cpu]->skipped_entries = entries;
1567 * No necessary locking here. The worst thing which can
1568 * happen is loosing events consumed at the same time
1569 * by a trace_pipe reader.
1570 * Other than that, we don't risk to crash the ring buffer
1571 * because it serializes the readers.
1573 * The current tracer is copied to avoid a global locking
1576 static void *s_start(struct seq_file *m, loff_t *pos)
1578 struct trace_iterator *iter = m->private;
1579 static struct tracer *old_tracer;
1580 int cpu_file = iter->cpu_file;
1585 /* copy the tracer to avoid using a global lock all around */
1586 mutex_lock(&trace_types_lock);
1587 if (unlikely(old_tracer != current_trace && current_trace)) {
1588 old_tracer = current_trace;
1589 *iter->trace = *current_trace;
1591 mutex_unlock(&trace_types_lock);
1593 atomic_inc(&trace_record_cmdline_disabled);
1595 if (*pos != iter->pos) {
1600 ftrace_disable_cpu();
1602 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1603 for_each_tracing_cpu(cpu)
1604 tracing_iter_reset(iter, cpu);
1606 tracing_iter_reset(iter, cpu_file);
1608 ftrace_enable_cpu();
1610 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1615 * If we overflowed the seq_file before, then we want
1616 * to just reuse the trace_seq buffer again.
1622 p = s_next(m, p, &l);
1626 trace_event_read_lock();
1630 static void s_stop(struct seq_file *m, void *p)
1632 atomic_dec(&trace_record_cmdline_disabled);
1633 trace_event_read_unlock();
1636 static void print_lat_help_header(struct seq_file *m)
1638 seq_puts(m, "# _------=> CPU# \n");
1639 seq_puts(m, "# / _-----=> irqs-off \n");
1640 seq_puts(m, "# | / _----=> need-resched \n");
1641 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1642 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1643 seq_puts(m, "# |||| /_--=> lock-depth \n");
1644 seq_puts(m, "# |||||/ delay \n");
1645 seq_puts(m, "# cmd pid |||||| time | caller \n");
1646 seq_puts(m, "# \\ / |||||| \\ | / \n");
1649 static void print_func_help_header(struct seq_file *m)
1651 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1652 seq_puts(m, "# | | | | |\n");
1657 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1659 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1660 struct trace_array *tr = iter->tr;
1661 struct trace_array_cpu *data = tr->data[tr->cpu];
1662 struct tracer *type = current_trace;
1663 unsigned long entries = 0;
1664 unsigned long total = 0;
1665 unsigned long count;
1666 const char *name = "preemption";
1673 for_each_tracing_cpu(cpu) {
1674 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1676 * If this buffer has skipped entries, then we hold all
1677 * entries for the trace and we need to ignore the
1678 * ones before the time stamp.
1680 if (tr->data[cpu]->skipped_entries) {
1681 count -= tr->data[cpu]->skipped_entries;
1682 /* total is the same as the entries */
1686 ring_buffer_overrun_cpu(tr->buffer, cpu);
1690 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1692 seq_puts(m, "# -----------------------------------"
1693 "---------------------------------\n");
1694 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1695 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1696 nsecs_to_usecs(data->saved_latency),
1700 #if defined(CONFIG_PREEMPT_NONE)
1702 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1704 #elif defined(CONFIG_PREEMPT)
1709 /* These are reserved for later use */
1712 seq_printf(m, " #P:%d)\n", num_online_cpus());
1716 seq_puts(m, "# -----------------\n");
1717 seq_printf(m, "# | task: %.16s-%d "
1718 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1719 data->comm, data->pid, data->uid, data->nice,
1720 data->policy, data->rt_priority);
1721 seq_puts(m, "# -----------------\n");
1723 if (data->critical_start) {
1724 seq_puts(m, "# => started at: ");
1725 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1726 trace_print_seq(m, &iter->seq);
1727 seq_puts(m, "\n# => ended at: ");
1728 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1729 trace_print_seq(m, &iter->seq);
1730 seq_puts(m, "\n#\n");
1736 static void test_cpu_buff_start(struct trace_iterator *iter)
1738 struct trace_seq *s = &iter->seq;
1740 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1743 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1746 if (cpumask_test_cpu(iter->cpu, iter->started))
1749 if (iter->tr->data[iter->cpu]->skipped_entries)
1752 cpumask_set_cpu(iter->cpu, iter->started);
1754 /* Don't print started cpu buffer for the first entry of the trace */
1756 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1760 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1762 struct trace_seq *s = &iter->seq;
1763 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1764 struct trace_entry *entry;
1765 struct trace_event *event;
1769 test_cpu_buff_start(iter);
1771 event = ftrace_find_event(entry->type);
1773 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1774 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1775 if (!trace_print_lat_context(iter))
1778 if (!trace_print_context(iter))
1784 return event->trace(iter, sym_flags);
1786 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1789 return TRACE_TYPE_HANDLED;
1791 return TRACE_TYPE_PARTIAL_LINE;
1794 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1796 struct trace_seq *s = &iter->seq;
1797 struct trace_entry *entry;
1798 struct trace_event *event;
1802 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1803 if (!trace_seq_printf(s, "%d %d %llu ",
1804 entry->pid, iter->cpu, iter->ts))
1808 event = ftrace_find_event(entry->type);
1810 return event->raw(iter, 0);
1812 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1815 return TRACE_TYPE_HANDLED;
1817 return TRACE_TYPE_PARTIAL_LINE;
1820 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1822 struct trace_seq *s = &iter->seq;
1823 unsigned char newline = '\n';
1824 struct trace_entry *entry;
1825 struct trace_event *event;
1829 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1830 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1831 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1832 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1835 event = ftrace_find_event(entry->type);
1837 enum print_line_t ret = event->hex(iter, 0);
1838 if (ret != TRACE_TYPE_HANDLED)
1842 SEQ_PUT_FIELD_RET(s, newline);
1844 return TRACE_TYPE_HANDLED;
1847 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1849 struct trace_seq *s = &iter->seq;
1850 struct trace_entry *entry;
1851 struct trace_event *event;
1855 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1856 SEQ_PUT_FIELD_RET(s, entry->pid);
1857 SEQ_PUT_FIELD_RET(s, iter->cpu);
1858 SEQ_PUT_FIELD_RET(s, iter->ts);
1861 event = ftrace_find_event(entry->type);
1862 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1865 static int trace_empty(struct trace_iterator *iter)
1869 /* If we are looking at one CPU buffer, only check that one */
1870 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1871 cpu = iter->cpu_file;
1872 if (iter->buffer_iter[cpu]) {
1873 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1876 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1882 for_each_tracing_cpu(cpu) {
1883 if (iter->buffer_iter[cpu]) {
1884 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1887 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1895 /* Called with trace_event_read_lock() held. */
1896 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1898 enum print_line_t ret;
1900 if (iter->trace && iter->trace->print_line) {
1901 ret = iter->trace->print_line(iter);
1902 if (ret != TRACE_TYPE_UNHANDLED)
1906 if (iter->ent->type == TRACE_BPRINT &&
1907 trace_flags & TRACE_ITER_PRINTK &&
1908 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1909 return trace_print_bprintk_msg_only(iter);
1911 if (iter->ent->type == TRACE_PRINT &&
1912 trace_flags & TRACE_ITER_PRINTK &&
1913 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1914 return trace_print_printk_msg_only(iter);
1916 if (trace_flags & TRACE_ITER_BIN)
1917 return print_bin_fmt(iter);
1919 if (trace_flags & TRACE_ITER_HEX)
1920 return print_hex_fmt(iter);
1922 if (trace_flags & TRACE_ITER_RAW)
1923 return print_raw_fmt(iter);
1925 return print_trace_fmt(iter);
1928 static int s_show(struct seq_file *m, void *v)
1930 struct trace_iterator *iter = v;
1933 if (iter->ent == NULL) {
1935 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1938 if (iter->trace && iter->trace->print_header)
1939 iter->trace->print_header(m);
1940 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1941 /* print nothing if the buffers are empty */
1942 if (trace_empty(iter))
1944 print_trace_header(m, iter);
1945 if (!(trace_flags & TRACE_ITER_VERBOSE))
1946 print_lat_help_header(m);
1948 if (!(trace_flags & TRACE_ITER_VERBOSE))
1949 print_func_help_header(m);
1951 } else if (iter->leftover) {
1953 * If we filled the seq_file buffer earlier, we
1954 * want to just show it now.
1956 ret = trace_print_seq(m, &iter->seq);
1958 /* ret should this time be zero, but you never know */
1959 iter->leftover = ret;
1962 print_trace_line(iter);
1963 ret = trace_print_seq(m, &iter->seq);
1965 * If we overflow the seq_file buffer, then it will
1966 * ask us for this data again at start up.
1968 * ret is 0 if seq_file write succeeded.
1971 iter->leftover = ret;
1977 static const struct seq_operations tracer_seq_ops = {
1984 static struct trace_iterator *
1985 __tracing_open(struct inode *inode, struct file *file)
1987 long cpu_file = (long) inode->i_private;
1988 void *fail_ret = ERR_PTR(-ENOMEM);
1989 struct trace_iterator *iter;
1993 if (tracing_disabled)
1994 return ERR_PTR(-ENODEV);
1996 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1998 return ERR_PTR(-ENOMEM);
2001 * We make a copy of the current tracer to avoid concurrent
2002 * changes on it while we are reading.
2004 mutex_lock(&trace_types_lock);
2005 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2010 *iter->trace = *current_trace;
2012 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2015 if (current_trace && current_trace->print_max)
2018 iter->tr = &global_trace;
2020 mutex_init(&iter->mutex);
2021 iter->cpu_file = cpu_file;
2023 /* Notify the tracer early; before we stop tracing. */
2024 if (iter->trace && iter->trace->open)
2025 iter->trace->open(iter);
2027 /* Annotate start of buffers if we had overruns */
2028 if (ring_buffer_overruns(iter->tr->buffer))
2029 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2031 /* stop the trace while dumping */
2034 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2035 for_each_tracing_cpu(cpu) {
2037 iter->buffer_iter[cpu] =
2038 ring_buffer_read_start(iter->tr->buffer, cpu);
2039 tracing_iter_reset(iter, cpu);
2042 cpu = iter->cpu_file;
2043 iter->buffer_iter[cpu] =
2044 ring_buffer_read_start(iter->tr->buffer, cpu);
2045 tracing_iter_reset(iter, cpu);
2048 ret = seq_open(file, &tracer_seq_ops);
2050 fail_ret = ERR_PTR(ret);
2054 m = file->private_data;
2057 mutex_unlock(&trace_types_lock);
2062 for_each_tracing_cpu(cpu) {
2063 if (iter->buffer_iter[cpu])
2064 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2066 free_cpumask_var(iter->started);
2069 mutex_unlock(&trace_types_lock);
2076 int tracing_open_generic(struct inode *inode, struct file *filp)
2078 if (tracing_disabled)
2081 filp->private_data = inode->i_private;
2085 static int tracing_release(struct inode *inode, struct file *file)
2087 struct seq_file *m = (struct seq_file *)file->private_data;
2088 struct trace_iterator *iter;
2091 if (!(file->f_mode & FMODE_READ))
2096 mutex_lock(&trace_types_lock);
2097 for_each_tracing_cpu(cpu) {
2098 if (iter->buffer_iter[cpu])
2099 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2102 if (iter->trace && iter->trace->close)
2103 iter->trace->close(iter);
2105 /* reenable tracing if it was previously enabled */
2107 mutex_unlock(&trace_types_lock);
2109 seq_release(inode, file);
2110 mutex_destroy(&iter->mutex);
2111 free_cpumask_var(iter->started);
2117 static int tracing_open(struct inode *inode, struct file *file)
2119 struct trace_iterator *iter;
2122 /* If this file was open for write, then erase contents */
2123 if ((file->f_mode & FMODE_WRITE) &&
2124 (file->f_flags & O_TRUNC)) {
2125 long cpu = (long) inode->i_private;
2127 if (cpu == TRACE_PIPE_ALL_CPU)
2128 tracing_reset_online_cpus(&global_trace);
2130 tracing_reset(&global_trace, cpu);
2133 if (file->f_mode & FMODE_READ) {
2134 iter = __tracing_open(inode, file);
2136 ret = PTR_ERR(iter);
2137 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2138 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2144 t_next(struct seq_file *m, void *v, loff_t *pos)
2146 struct tracer *t = v;
2156 static void *t_start(struct seq_file *m, loff_t *pos)
2161 mutex_lock(&trace_types_lock);
2162 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2168 static void t_stop(struct seq_file *m, void *p)
2170 mutex_unlock(&trace_types_lock);
2173 static int t_show(struct seq_file *m, void *v)
2175 struct tracer *t = v;
2180 seq_printf(m, "%s", t->name);
2189 static const struct seq_operations show_traces_seq_ops = {
2196 static int show_traces_open(struct inode *inode, struct file *file)
2198 if (tracing_disabled)
2201 return seq_open(file, &show_traces_seq_ops);
2205 tracing_write_stub(struct file *filp, const char __user *ubuf,
2206 size_t count, loff_t *ppos)
2211 static const struct file_operations tracing_fops = {
2212 .open = tracing_open,
2214 .write = tracing_write_stub,
2215 .llseek = seq_lseek,
2216 .release = tracing_release,
2219 static const struct file_operations show_traces_fops = {
2220 .open = show_traces_open,
2222 .release = seq_release,
2226 * Only trace on a CPU if the bitmask is set:
2228 static cpumask_var_t tracing_cpumask;
2231 * The tracer itself will not take this lock, but still we want
2232 * to provide a consistent cpumask to user-space:
2234 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2237 * Temporary storage for the character representation of the
2238 * CPU bitmask (and one more byte for the newline):
2240 static char mask_str[NR_CPUS + 1];
2243 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2244 size_t count, loff_t *ppos)
2248 mutex_lock(&tracing_cpumask_update_lock);
2250 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2251 if (count - len < 2) {
2255 len += sprintf(mask_str + len, "\n");
2256 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2259 mutex_unlock(&tracing_cpumask_update_lock);
2265 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2266 size_t count, loff_t *ppos)
2269 cpumask_var_t tracing_cpumask_new;
2271 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2274 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2278 mutex_lock(&tracing_cpumask_update_lock);
2280 local_irq_disable();
2281 __raw_spin_lock(&ftrace_max_lock);
2282 for_each_tracing_cpu(cpu) {
2284 * Increase/decrease the disabled counter if we are
2285 * about to flip a bit in the cpumask:
2287 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2288 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2289 atomic_inc(&global_trace.data[cpu]->disabled);
2291 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2292 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2293 atomic_dec(&global_trace.data[cpu]->disabled);
2296 __raw_spin_unlock(&ftrace_max_lock);
2299 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2301 mutex_unlock(&tracing_cpumask_update_lock);
2302 free_cpumask_var(tracing_cpumask_new);
2307 free_cpumask_var(tracing_cpumask_new);
2312 static const struct file_operations tracing_cpumask_fops = {
2313 .open = tracing_open_generic,
2314 .read = tracing_cpumask_read,
2315 .write = tracing_cpumask_write,
2318 static int tracing_trace_options_show(struct seq_file *m, void *v)
2320 struct tracer_opt *trace_opts;
2324 mutex_lock(&trace_types_lock);
2325 tracer_flags = current_trace->flags->val;
2326 trace_opts = current_trace->flags->opts;
2328 for (i = 0; trace_options[i]; i++) {
2329 if (trace_flags & (1 << i))
2330 seq_printf(m, "%s\n", trace_options[i]);
2332 seq_printf(m, "no%s\n", trace_options[i]);
2335 for (i = 0; trace_opts[i].name; i++) {
2336 if (tracer_flags & trace_opts[i].bit)
2337 seq_printf(m, "%s\n", trace_opts[i].name);
2339 seq_printf(m, "no%s\n", trace_opts[i].name);
2341 mutex_unlock(&trace_types_lock);
2346 /* Try to assign a tracer specific option */
2347 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2349 struct tracer_flags *tracer_flags = trace->flags;
2350 struct tracer_opt *opts = NULL;
2354 for (i = 0; tracer_flags->opts[i].name; i++) {
2355 opts = &tracer_flags->opts[i];
2356 len = strlen(opts->name);
2358 if (strncmp(cmp, opts->name, len) == 0) {
2359 ret = trace->set_flag(tracer_flags->val,
2365 if (!tracer_flags->opts[i].name)
2368 /* Refused to handle */
2373 tracer_flags->val &= ~opts->bit;
2375 tracer_flags->val |= opts->bit;
2380 static void set_tracer_flags(unsigned int mask, int enabled)
2382 /* do nothing if flag is already set */
2383 if (!!(trace_flags & mask) == !!enabled)
2387 trace_flags |= mask;
2389 trace_flags &= ~mask;
2393 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2394 size_t cnt, loff_t *ppos)
2402 if (cnt >= sizeof(buf))
2405 if (copy_from_user(&buf, ubuf, cnt))
2410 if (strncmp(buf, "no", 2) == 0) {
2415 for (i = 0; trace_options[i]; i++) {
2416 int len = strlen(trace_options[i]);
2418 if (strncmp(cmp, trace_options[i], len) == 0) {
2419 set_tracer_flags(1 << i, !neg);
2424 /* If no option could be set, test the specific tracer options */
2425 if (!trace_options[i]) {
2426 mutex_lock(&trace_types_lock);
2427 ret = set_tracer_option(current_trace, cmp, neg);
2428 mutex_unlock(&trace_types_lock);
2438 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2440 if (tracing_disabled)
2442 return single_open(file, tracing_trace_options_show, NULL);
2445 static const struct file_operations tracing_iter_fops = {
2446 .open = tracing_trace_options_open,
2448 .llseek = seq_lseek,
2449 .release = single_release,
2450 .write = tracing_trace_options_write,
2453 static const char readme_msg[] =
2454 "tracing mini-HOWTO:\n\n"
2455 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2456 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2457 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2458 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2460 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2461 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2463 "# cat /sys/kernel/debug/tracing/trace_options\n"
2464 "noprint-parent nosym-offset nosym-addr noverbose\n"
2465 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2466 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2467 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2468 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2472 tracing_readme_read(struct file *filp, char __user *ubuf,
2473 size_t cnt, loff_t *ppos)
2475 return simple_read_from_buffer(ubuf, cnt, ppos,
2476 readme_msg, strlen(readme_msg));
2479 static const struct file_operations tracing_readme_fops = {
2480 .open = tracing_open_generic,
2481 .read = tracing_readme_read,
2485 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2486 size_t cnt, loff_t *ppos)
2495 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2499 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2507 for (i = 0; i < SAVED_CMDLINES; i++) {
2510 pid = map_cmdline_to_pid[i];
2511 if (pid == -1 || pid == NO_CMDLINE_MAP)
2514 trace_find_cmdline(pid, buf_comm);
2515 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2520 len = simple_read_from_buffer(ubuf, cnt, ppos,
2529 static const struct file_operations tracing_saved_cmdlines_fops = {
2530 .open = tracing_open_generic,
2531 .read = tracing_saved_cmdlines_read,
2535 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2536 size_t cnt, loff_t *ppos)
2541 r = sprintf(buf, "%u\n", tracer_enabled);
2542 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2546 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2547 size_t cnt, loff_t *ppos)
2549 struct trace_array *tr = filp->private_data;
2554 if (cnt >= sizeof(buf))
2557 if (copy_from_user(&buf, ubuf, cnt))
2562 ret = strict_strtoul(buf, 10, &val);
2568 mutex_lock(&trace_types_lock);
2569 if (tracer_enabled ^ val) {
2572 if (current_trace->start)
2573 current_trace->start(tr);
2578 if (current_trace->stop)
2579 current_trace->stop(tr);
2582 mutex_unlock(&trace_types_lock);
2590 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2591 size_t cnt, loff_t *ppos)
2593 char buf[MAX_TRACER_SIZE+2];
2596 mutex_lock(&trace_types_lock);
2598 r = sprintf(buf, "%s\n", current_trace->name);
2600 r = sprintf(buf, "\n");
2601 mutex_unlock(&trace_types_lock);
2603 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2606 int tracer_init(struct tracer *t, struct trace_array *tr)
2608 tracing_reset_online_cpus(tr);
2612 static int tracing_resize_ring_buffer(unsigned long size)
2617 * If kernel or user changes the size of the ring buffer
2618 * we use the size that was given, and we can forget about
2619 * expanding it later.
2621 ring_buffer_expanded = 1;
2623 ret = ring_buffer_resize(global_trace.buffer, size);
2627 ret = ring_buffer_resize(max_tr.buffer, size);
2631 r = ring_buffer_resize(global_trace.buffer,
2632 global_trace.entries);
2635 * AARGH! We are left with different
2636 * size max buffer!!!!
2637 * The max buffer is our "snapshot" buffer.
2638 * When a tracer needs a snapshot (one of the
2639 * latency tracers), it swaps the max buffer
2640 * with the saved snap shot. We succeeded to
2641 * update the size of the main buffer, but failed to
2642 * update the size of the max buffer. But when we tried
2643 * to reset the main buffer to the original size, we
2644 * failed there too. This is very unlikely to
2645 * happen, but if it does, warn and kill all
2649 tracing_disabled = 1;
2654 global_trace.entries = size;
2660 * tracing_update_buffers - used by tracing facility to expand ring buffers
2662 * To save on memory when the tracing is never used on a system with it
2663 * configured in. The ring buffers are set to a minimum size. But once
2664 * a user starts to use the tracing facility, then they need to grow
2665 * to their default size.
2667 * This function is to be called when a tracer is about to be used.
2669 int tracing_update_buffers(void)
2673 mutex_lock(&trace_types_lock);
2674 if (!ring_buffer_expanded)
2675 ret = tracing_resize_ring_buffer(trace_buf_size);
2676 mutex_unlock(&trace_types_lock);
2681 struct trace_option_dentry;
2683 static struct trace_option_dentry *
2684 create_trace_option_files(struct tracer *tracer);
2687 destroy_trace_option_files(struct trace_option_dentry *topts);
2689 static int tracing_set_tracer(const char *buf)
2691 static struct trace_option_dentry *topts;
2692 struct trace_array *tr = &global_trace;
2696 mutex_lock(&trace_types_lock);
2698 if (!ring_buffer_expanded) {
2699 ret = tracing_resize_ring_buffer(trace_buf_size);
2705 for (t = trace_types; t; t = t->next) {
2706 if (strcmp(t->name, buf) == 0)
2713 if (t == current_trace)
2716 trace_branch_disable();
2717 if (current_trace && current_trace->reset)
2718 current_trace->reset(tr);
2720 destroy_trace_option_files(topts);
2724 topts = create_trace_option_files(current_trace);
2727 ret = tracer_init(t, tr);
2732 trace_branch_enable(tr);
2734 mutex_unlock(&trace_types_lock);
2740 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2741 size_t cnt, loff_t *ppos)
2743 char buf[MAX_TRACER_SIZE+1];
2750 if (cnt > MAX_TRACER_SIZE)
2751 cnt = MAX_TRACER_SIZE;
2753 if (copy_from_user(&buf, ubuf, cnt))
2758 /* strip ending whitespace. */
2759 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2762 err = tracing_set_tracer(buf);
2772 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2773 size_t cnt, loff_t *ppos)
2775 unsigned long *ptr = filp->private_data;
2779 r = snprintf(buf, sizeof(buf), "%ld\n",
2780 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2781 if (r > sizeof(buf))
2783 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2787 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2788 size_t cnt, loff_t *ppos)
2790 unsigned long *ptr = filp->private_data;
2795 if (cnt >= sizeof(buf))
2798 if (copy_from_user(&buf, ubuf, cnt))
2803 ret = strict_strtoul(buf, 10, &val);
2812 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2814 long cpu_file = (long) inode->i_private;
2815 struct trace_iterator *iter;
2818 if (tracing_disabled)
2821 mutex_lock(&trace_types_lock);
2823 /* We only allow one reader per cpu */
2824 if (cpu_file == TRACE_PIPE_ALL_CPU) {
2825 if (!cpumask_empty(tracing_reader_cpumask)) {
2829 cpumask_setall(tracing_reader_cpumask);
2831 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2832 cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2839 /* create a buffer to store the information to pass to userspace */
2840 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2847 * We make a copy of the current tracer to avoid concurrent
2848 * changes on it while we are reading.
2850 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2856 *iter->trace = *current_trace;
2858 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2863 /* trace pipe does not show start of buffer */
2864 cpumask_setall(iter->started);
2866 if (trace_flags & TRACE_ITER_LATENCY_FMT)
2867 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2869 iter->cpu_file = cpu_file;
2870 iter->tr = &global_trace;
2871 mutex_init(&iter->mutex);
2872 filp->private_data = iter;
2874 if (iter->trace->pipe_open)
2875 iter->trace->pipe_open(iter);
2878 mutex_unlock(&trace_types_lock);
2884 mutex_unlock(&trace_types_lock);
2888 static int tracing_release_pipe(struct inode *inode, struct file *file)
2890 struct trace_iterator *iter = file->private_data;
2892 mutex_lock(&trace_types_lock);
2894 if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2895 cpumask_clear(tracing_reader_cpumask);
2897 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2900 if (iter->trace->pipe_close)
2901 iter->trace->pipe_close(iter);
2903 mutex_unlock(&trace_types_lock);
2905 free_cpumask_var(iter->started);
2906 mutex_destroy(&iter->mutex);
2914 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2916 struct trace_iterator *iter = filp->private_data;
2918 if (trace_flags & TRACE_ITER_BLOCK) {
2920 * Always select as readable when in blocking mode
2922 return POLLIN | POLLRDNORM;
2924 if (!trace_empty(iter))
2925 return POLLIN | POLLRDNORM;
2926 poll_wait(filp, &trace_wait, poll_table);
2927 if (!trace_empty(iter))
2928 return POLLIN | POLLRDNORM;
2935 void default_wait_pipe(struct trace_iterator *iter)
2939 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2941 if (trace_empty(iter))
2944 finish_wait(&trace_wait, &wait);
2948 * This is a make-shift waitqueue.
2949 * A tracer might use this callback on some rare cases:
2951 * 1) the current tracer might hold the runqueue lock when it wakes up
2952 * a reader, hence a deadlock (sched, function, and function graph tracers)
2953 * 2) the function tracers, trace all functions, we don't want
2954 * the overhead of calling wake_up and friends
2955 * (and tracing them too)
2957 * Anyway, this is really very primitive wakeup.
2959 void poll_wait_pipe(struct trace_iterator *iter)
2961 set_current_state(TASK_INTERRUPTIBLE);
2962 /* sleep for 100 msecs, and try again. */
2963 schedule_timeout(HZ / 10);
2966 /* Must be called with trace_types_lock mutex held. */
2967 static int tracing_wait_pipe(struct file *filp)
2969 struct trace_iterator *iter = filp->private_data;
2971 while (trace_empty(iter)) {
2973 if ((filp->f_flags & O_NONBLOCK)) {
2977 mutex_unlock(&iter->mutex);
2979 iter->trace->wait_pipe(iter);
2981 mutex_lock(&iter->mutex);
2983 if (signal_pending(current))
2987 * We block until we read something and tracing is disabled.
2988 * We still block if tracing is disabled, but we have never
2989 * read anything. This allows a user to cat this file, and
2990 * then enable tracing. But after we have read something,
2991 * we give an EOF when tracing is again disabled.
2993 * iter->pos will be 0 if we haven't read anything.
2995 if (!tracer_enabled && iter->pos)
3006 tracing_read_pipe(struct file *filp, char __user *ubuf,
3007 size_t cnt, loff_t *ppos)
3009 struct trace_iterator *iter = filp->private_data;
3010 static struct tracer *old_tracer;
3013 /* return any leftover data */
3014 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3018 trace_seq_init(&iter->seq);
3020 /* copy the tracer to avoid using a global lock all around */
3021 mutex_lock(&trace_types_lock);
3022 if (unlikely(old_tracer != current_trace && current_trace)) {
3023 old_tracer = current_trace;
3024 *iter->trace = *current_trace;
3026 mutex_unlock(&trace_types_lock);
3029 * Avoid more than one consumer on a single file descriptor
3030 * This is just a matter of traces coherency, the ring buffer itself
3033 mutex_lock(&iter->mutex);
3034 if (iter->trace->read) {
3035 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3041 sret = tracing_wait_pipe(filp);
3045 /* stop when tracing is finished */
3046 if (trace_empty(iter)) {
3051 if (cnt >= PAGE_SIZE)
3052 cnt = PAGE_SIZE - 1;
3054 /* reset all but tr, trace, and overruns */
3055 memset(&iter->seq, 0,
3056 sizeof(struct trace_iterator) -
3057 offsetof(struct trace_iterator, seq));
3060 trace_event_read_lock();
3061 while (find_next_entry_inc(iter) != NULL) {
3062 enum print_line_t ret;
3063 int len = iter->seq.len;
3065 ret = print_trace_line(iter);
3066 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3067 /* don't print partial lines */
3068 iter->seq.len = len;
3071 if (ret != TRACE_TYPE_NO_CONSUME)
3072 trace_consume(iter);
3074 if (iter->seq.len >= cnt)
3077 trace_event_read_unlock();
3079 /* Now copy what we have to the user */
3080 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3081 if (iter->seq.readpos >= iter->seq.len)
3082 trace_seq_init(&iter->seq);
3085 * If there was nothing to send to user, inspite of consuming trace
3086 * entries, go back to wait for more entries.
3092 mutex_unlock(&iter->mutex);
3097 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3098 struct pipe_buffer *buf)
3100 __free_page(buf->page);
3103 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3106 __free_page(spd->pages[idx]);
3109 static struct pipe_buf_operations tracing_pipe_buf_ops = {
3111 .map = generic_pipe_buf_map,
3112 .unmap = generic_pipe_buf_unmap,
3113 .confirm = generic_pipe_buf_confirm,
3114 .release = tracing_pipe_buf_release,
3115 .steal = generic_pipe_buf_steal,
3116 .get = generic_pipe_buf_get,
3120 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3125 /* Seq buffer is page-sized, exactly what we need. */
3127 count = iter->seq.len;
3128 ret = print_trace_line(iter);
3129 count = iter->seq.len - count;
3132 iter->seq.len -= count;
3135 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3136 iter->seq.len -= count;
3140 if (ret != TRACE_TYPE_NO_CONSUME)
3141 trace_consume(iter);
3143 if (!find_next_entry_inc(iter)) {
3153 static ssize_t tracing_splice_read_pipe(struct file *filp,
3155 struct pipe_inode_info *pipe,
3159 struct page *pages[PIPE_BUFFERS];
3160 struct partial_page partial[PIPE_BUFFERS];
3161 struct trace_iterator *iter = filp->private_data;
3162 struct splice_pipe_desc spd = {
3165 .nr_pages = 0, /* This gets updated below. */
3167 .ops = &tracing_pipe_buf_ops,
3168 .spd_release = tracing_spd_release_pipe,
3170 static struct tracer *old_tracer;
3175 /* copy the tracer to avoid using a global lock all around */
3176 mutex_lock(&trace_types_lock);
3177 if (unlikely(old_tracer != current_trace && current_trace)) {
3178 old_tracer = current_trace;
3179 *iter->trace = *current_trace;
3181 mutex_unlock(&trace_types_lock);
3183 mutex_lock(&iter->mutex);
3185 if (iter->trace->splice_read) {
3186 ret = iter->trace->splice_read(iter, filp,
3187 ppos, pipe, len, flags);
3192 ret = tracing_wait_pipe(filp);
3196 if (!iter->ent && !find_next_entry_inc(iter)) {
3201 trace_event_read_lock();
3203 /* Fill as many pages as possible. */
3204 for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3205 pages[i] = alloc_page(GFP_KERNEL);
3209 rem = tracing_fill_pipe_page(rem, iter);
3211 /* Copy the data into the page, so we can start over. */
3212 ret = trace_seq_to_buffer(&iter->seq,
3213 page_address(pages[i]),
3216 __free_page(pages[i]);
3219 partial[i].offset = 0;
3220 partial[i].len = iter->seq.len;
3222 trace_seq_init(&iter->seq);
3225 trace_event_read_unlock();
3226 mutex_unlock(&iter->mutex);
3230 return splice_to_pipe(pipe, &spd);
3233 mutex_unlock(&iter->mutex);
3239 tracing_entries_read(struct file *filp, char __user *ubuf,
3240 size_t cnt, loff_t *ppos)
3242 struct trace_array *tr = filp->private_data;
3246 mutex_lock(&trace_types_lock);
3247 if (!ring_buffer_expanded)
3248 r = sprintf(buf, "%lu (expanded: %lu)\n",
3250 trace_buf_size >> 10);
3252 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3253 mutex_unlock(&trace_types_lock);
3255 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3259 tracing_entries_write(struct file *filp, const char __user *ubuf,
3260 size_t cnt, loff_t *ppos)
3266 if (cnt >= sizeof(buf))
3269 if (copy_from_user(&buf, ubuf, cnt))
3274 ret = strict_strtoul(buf, 10, &val);
3278 /* must have at least 1 entry */
3282 mutex_lock(&trace_types_lock);
3286 /* disable all cpu buffers */
3287 for_each_tracing_cpu(cpu) {
3288 if (global_trace.data[cpu])
3289 atomic_inc(&global_trace.data[cpu]->disabled);
3290 if (max_tr.data[cpu])
3291 atomic_inc(&max_tr.data[cpu]->disabled);
3294 /* value is in KB */
3297 if (val != global_trace.entries) {
3298 ret = tracing_resize_ring_buffer(val);
3307 /* If check pages failed, return ENOMEM */
3308 if (tracing_disabled)
3311 for_each_tracing_cpu(cpu) {
3312 if (global_trace.data[cpu])
3313 atomic_dec(&global_trace.data[cpu]->disabled);
3314 if (max_tr.data[cpu])
3315 atomic_dec(&max_tr.data[cpu]->disabled);
3319 max_tr.entries = global_trace.entries;
3320 mutex_unlock(&trace_types_lock);
3325 static int mark_printk(const char *fmt, ...)
3329 va_start(args, fmt);
3330 ret = trace_vprintk(0, fmt, args);
3336 tracing_mark_write(struct file *filp, const char __user *ubuf,
3337 size_t cnt, loff_t *fpos)
3341 if (tracing_disabled)
3344 if (cnt > TRACE_BUF_SIZE)
3345 cnt = TRACE_BUF_SIZE;
3347 buf = kmalloc(cnt + 2, GFP_KERNEL);
3351 if (copy_from_user(buf, ubuf, cnt)) {
3355 if (buf[cnt-1] != '\n') {
3361 cnt = mark_printk("%s", buf);
3368 static int tracing_clock_show(struct seq_file *m, void *v)
3372 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3374 "%s%s%s%s", i ? " " : "",
3375 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3376 i == trace_clock_id ? "]" : "");
3382 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3383 size_t cnt, loff_t *fpos)
3386 const char *clockstr;
3389 if (cnt >= sizeof(buf))
3392 if (copy_from_user(&buf, ubuf, cnt))
3397 clockstr = strstrip(buf);
3399 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3400 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3403 if (i == ARRAY_SIZE(trace_clocks))
3408 mutex_lock(&trace_types_lock);
3410 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3412 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3414 mutex_unlock(&trace_types_lock);
3421 static int tracing_clock_open(struct inode *inode, struct file *file)
3423 if (tracing_disabled)
3425 return single_open(file, tracing_clock_show, NULL);
3428 static const struct file_operations tracing_max_lat_fops = {
3429 .open = tracing_open_generic,
3430 .read = tracing_max_lat_read,
3431 .write = tracing_max_lat_write,
3434 static const struct file_operations tracing_ctrl_fops = {
3435 .open = tracing_open_generic,
3436 .read = tracing_ctrl_read,
3437 .write = tracing_ctrl_write,
3440 static const struct file_operations set_tracer_fops = {
3441 .open = tracing_open_generic,
3442 .read = tracing_set_trace_read,
3443 .write = tracing_set_trace_write,
3446 static const struct file_operations tracing_pipe_fops = {
3447 .open = tracing_open_pipe,
3448 .poll = tracing_poll_pipe,
3449 .read = tracing_read_pipe,
3450 .splice_read = tracing_splice_read_pipe,
3451 .release = tracing_release_pipe,
3454 static const struct file_operations tracing_entries_fops = {
3455 .open = tracing_open_generic,
3456 .read = tracing_entries_read,
3457 .write = tracing_entries_write,
3460 static const struct file_operations tracing_mark_fops = {
3461 .open = tracing_open_generic,
3462 .write = tracing_mark_write,
3465 static const struct file_operations trace_clock_fops = {
3466 .open = tracing_clock_open,
3468 .llseek = seq_lseek,
3469 .release = single_release,
3470 .write = tracing_clock_write,
3473 struct ftrace_buffer_info {
3474 struct trace_array *tr;
3480 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3482 int cpu = (int)(long)inode->i_private;
3483 struct ftrace_buffer_info *info;
3485 if (tracing_disabled)
3488 info = kzalloc(sizeof(*info), GFP_KERNEL);
3492 info->tr = &global_trace;
3495 /* Force reading ring buffer for first read */
3496 info->read = (unsigned int)-1;
3498 filp->private_data = info;
3500 return nonseekable_open(inode, filp);
3504 tracing_buffers_read(struct file *filp, char __user *ubuf,
3505 size_t count, loff_t *ppos)
3507 struct ftrace_buffer_info *info = filp->private_data;
3516 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3520 /* Do we have previous read data to read? */
3521 if (info->read < PAGE_SIZE)
3526 ret = ring_buffer_read_page(info->tr->buffer,
3533 pos = ring_buffer_page_len(info->spare);
3535 if (pos < PAGE_SIZE)
3536 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3539 size = PAGE_SIZE - info->read;
3543 ret = copy_to_user(ubuf, info->spare + info->read, size);
3554 static int tracing_buffers_release(struct inode *inode, struct file *file)
3556 struct ftrace_buffer_info *info = file->private_data;
3559 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3566 struct ring_buffer *buffer;
3571 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3572 struct pipe_buffer *buf)
3574 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3579 ring_buffer_free_read_page(ref->buffer, ref->page);
3584 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3585 struct pipe_buffer *buf)
3590 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3591 struct pipe_buffer *buf)
3593 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3598 /* Pipe buffer operations for a buffer. */
3599 static struct pipe_buf_operations buffer_pipe_buf_ops = {
3601 .map = generic_pipe_buf_map,
3602 .unmap = generic_pipe_buf_unmap,
3603 .confirm = generic_pipe_buf_confirm,
3604 .release = buffer_pipe_buf_release,
3605 .steal = buffer_pipe_buf_steal,
3606 .get = buffer_pipe_buf_get,
3610 * Callback from splice_to_pipe(), if we need to release some pages
3611 * at the end of the spd in case we error'ed out in filling the pipe.
3613 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3615 struct buffer_ref *ref =
3616 (struct buffer_ref *)spd->partial[i].private;
3621 ring_buffer_free_read_page(ref->buffer, ref->page);
3623 spd->partial[i].private = 0;
3627 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3628 struct pipe_inode_info *pipe, size_t len,
3631 struct ftrace_buffer_info *info = file->private_data;
3632 struct partial_page partial[PIPE_BUFFERS];
3633 struct page *pages[PIPE_BUFFERS];
3634 struct splice_pipe_desc spd = {
3638 .ops = &buffer_pipe_buf_ops,
3639 .spd_release = buffer_spd_release,
3641 struct buffer_ref *ref;
3642 int entries, size, i;
3645 if (*ppos & (PAGE_SIZE - 1)) {
3646 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3650 if (len & (PAGE_SIZE - 1)) {
3651 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3652 if (len < PAGE_SIZE)
3657 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3659 for (i = 0; i < PIPE_BUFFERS && len && entries; i++, len -= PAGE_SIZE) {
3663 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3668 ref->buffer = info->tr->buffer;
3669 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3675 r = ring_buffer_read_page(ref->buffer, &ref->page,
3678 ring_buffer_free_read_page(ref->buffer,
3685 * zero out any left over data, this is going to
3688 size = ring_buffer_page_len(ref->page);
3689 if (size < PAGE_SIZE)
3690 memset(ref->page + size, 0, PAGE_SIZE - size);
3692 page = virt_to_page(ref->page);
3694 spd.pages[i] = page;
3695 spd.partial[i].len = PAGE_SIZE;
3696 spd.partial[i].offset = 0;
3697 spd.partial[i].private = (unsigned long)ref;
3701 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3706 /* did we read anything? */
3707 if (!spd.nr_pages) {
3708 if (flags & SPLICE_F_NONBLOCK)
3716 ret = splice_to_pipe(pipe, &spd);
3721 static const struct file_operations tracing_buffers_fops = {
3722 .open = tracing_buffers_open,
3723 .read = tracing_buffers_read,
3724 .release = tracing_buffers_release,
3725 .splice_read = tracing_buffers_splice_read,
3726 .llseek = no_llseek,
3730 tracing_stats_read(struct file *filp, char __user *ubuf,
3731 size_t count, loff_t *ppos)
3733 unsigned long cpu = (unsigned long)filp->private_data;
3734 struct trace_array *tr = &global_trace;
3735 struct trace_seq *s;
3738 s = kmalloc(sizeof(*s), GFP_KERNEL);
3744 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3745 trace_seq_printf(s, "entries: %ld\n", cnt);
3747 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3748 trace_seq_printf(s, "overrun: %ld\n", cnt);
3750 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3751 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3753 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3760 static const struct file_operations tracing_stats_fops = {
3761 .open = tracing_open_generic,
3762 .read = tracing_stats_read,
3765 #ifdef CONFIG_DYNAMIC_FTRACE
3767 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3773 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3774 size_t cnt, loff_t *ppos)
3776 static char ftrace_dyn_info_buffer[1024];
3777 static DEFINE_MUTEX(dyn_info_mutex);
3778 unsigned long *p = filp->private_data;
3779 char *buf = ftrace_dyn_info_buffer;
3780 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3783 mutex_lock(&dyn_info_mutex);
3784 r = sprintf(buf, "%ld ", *p);
3786 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3789 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3791 mutex_unlock(&dyn_info_mutex);
3796 static const struct file_operations tracing_dyn_info_fops = {
3797 .open = tracing_open_generic,
3798 .read = tracing_read_dyn_info,
3802 static struct dentry *d_tracer;
3804 struct dentry *tracing_init_dentry(void)
3811 if (!debugfs_initialized())
3814 d_tracer = debugfs_create_dir("tracing", NULL);
3816 if (!d_tracer && !once) {
3818 pr_warning("Could not create debugfs directory 'tracing'\n");
3825 static struct dentry *d_percpu;
3827 struct dentry *tracing_dentry_percpu(void)
3830 struct dentry *d_tracer;
3835 d_tracer = tracing_init_dentry();
3840 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3842 if (!d_percpu && !once) {
3844 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3851 static void tracing_init_debugfs_percpu(long cpu)
3853 struct dentry *d_percpu = tracing_dentry_percpu();
3854 struct dentry *d_cpu;
3855 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3858 if (cpu > 999 || cpu < 0)
3861 sprintf(cpu_dir, "cpu%ld", cpu);
3862 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3864 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3868 /* per cpu trace_pipe */
3869 trace_create_file("trace_pipe", 0444, d_cpu,
3870 (void *) cpu, &tracing_pipe_fops);
3873 trace_create_file("trace", 0644, d_cpu,
3874 (void *) cpu, &tracing_fops);
3876 trace_create_file("trace_pipe_raw", 0444, d_cpu,
3877 (void *) cpu, &tracing_buffers_fops);
3879 trace_create_file("stats", 0444, d_cpu,
3880 (void *) cpu, &tracing_stats_fops);
3883 #ifdef CONFIG_FTRACE_SELFTEST
3884 /* Let selftest have access to static functions in this file */
3885 #include "trace_selftest.c"
3888 struct trace_option_dentry {
3889 struct tracer_opt *opt;
3890 struct tracer_flags *flags;
3891 struct dentry *entry;
3895 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3898 struct trace_option_dentry *topt = filp->private_data;
3901 if (topt->flags->val & topt->opt->bit)
3906 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3910 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3913 struct trace_option_dentry *topt = filp->private_data;
3918 if (cnt >= sizeof(buf))
3921 if (copy_from_user(&buf, ubuf, cnt))
3926 ret = strict_strtoul(buf, 10, &val);
3933 /* do nothing if already cleared */
3934 if (!(topt->flags->val & topt->opt->bit))
3937 mutex_lock(&trace_types_lock);
3938 if (current_trace->set_flag)
3939 ret = current_trace->set_flag(topt->flags->val,
3941 mutex_unlock(&trace_types_lock);
3944 topt->flags->val &= ~topt->opt->bit;
3947 /* do nothing if already set */
3948 if (topt->flags->val & topt->opt->bit)
3951 mutex_lock(&trace_types_lock);
3952 if (current_trace->set_flag)
3953 ret = current_trace->set_flag(topt->flags->val,
3955 mutex_unlock(&trace_types_lock);
3958 topt->flags->val |= topt->opt->bit;
3971 static const struct file_operations trace_options_fops = {
3972 .open = tracing_open_generic,
3973 .read = trace_options_read,
3974 .write = trace_options_write,
3978 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
3981 long index = (long)filp->private_data;
3984 if (trace_flags & (1 << index))
3989 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3993 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
3996 long index = (long)filp->private_data;
4001 if (cnt >= sizeof(buf))
4004 if (copy_from_user(&buf, ubuf, cnt))
4009 ret = strict_strtoul(buf, 10, &val);
4013 if (val != 0 && val != 1)
4015 set_tracer_flags(1 << index, val);
4022 static const struct file_operations trace_options_core_fops = {
4023 .open = tracing_open_generic,
4024 .read = trace_options_core_read,
4025 .write = trace_options_core_write,
4028 struct dentry *trace_create_file(const char *name,
4030 struct dentry *parent,
4032 const struct file_operations *fops)
4036 ret = debugfs_create_file(name, mode, parent, data, fops);
4038 pr_warning("Could not create debugfs '%s' entry\n", name);
4044 static struct dentry *trace_options_init_dentry(void)
4046 struct dentry *d_tracer;
4047 static struct dentry *t_options;
4052 d_tracer = tracing_init_dentry();
4056 t_options = debugfs_create_dir("options", d_tracer);
4058 pr_warning("Could not create debugfs directory 'options'\n");
4066 create_trace_option_file(struct trace_option_dentry *topt,
4067 struct tracer_flags *flags,
4068 struct tracer_opt *opt)
4070 struct dentry *t_options;
4072 t_options = trace_options_init_dentry();
4076 topt->flags = flags;
4079 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4080 &trace_options_fops);
4084 static struct trace_option_dentry *
4085 create_trace_option_files(struct tracer *tracer)
4087 struct trace_option_dentry *topts;
4088 struct tracer_flags *flags;
4089 struct tracer_opt *opts;
4095 flags = tracer->flags;
4097 if (!flags || !flags->opts)
4102 for (cnt = 0; opts[cnt].name; cnt++)
4105 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4109 for (cnt = 0; opts[cnt].name; cnt++)
4110 create_trace_option_file(&topts[cnt], flags,
4117 destroy_trace_option_files(struct trace_option_dentry *topts)
4124 for (cnt = 0; topts[cnt].opt; cnt++) {
4125 if (topts[cnt].entry)
4126 debugfs_remove(topts[cnt].entry);
4132 static struct dentry *
4133 create_trace_option_core_file(const char *option, long index)
4135 struct dentry *t_options;
4137 t_options = trace_options_init_dentry();
4141 return trace_create_file(option, 0644, t_options, (void *)index,
4142 &trace_options_core_fops);
4145 static __init void create_trace_options_dir(void)
4147 struct dentry *t_options;
4150 t_options = trace_options_init_dentry();
4154 for (i = 0; trace_options[i]; i++)
4155 create_trace_option_core_file(trace_options[i], i);
4158 static __init int tracer_init_debugfs(void)
4160 struct dentry *d_tracer;
4163 d_tracer = tracing_init_dentry();
4165 trace_create_file("tracing_enabled", 0644, d_tracer,
4166 &global_trace, &tracing_ctrl_fops);
4168 trace_create_file("trace_options", 0644, d_tracer,
4169 NULL, &tracing_iter_fops);
4171 trace_create_file("tracing_cpumask", 0644, d_tracer,
4172 NULL, &tracing_cpumask_fops);
4174 trace_create_file("trace", 0644, d_tracer,
4175 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4177 trace_create_file("available_tracers", 0444, d_tracer,
4178 &global_trace, &show_traces_fops);
4180 trace_create_file("current_tracer", 0644, d_tracer,
4181 &global_trace, &set_tracer_fops);
4183 #ifdef CONFIG_TRACER_MAX_TRACE
4184 trace_create_file("tracing_max_latency", 0644, d_tracer,
4185 &tracing_max_latency, &tracing_max_lat_fops);
4187 trace_create_file("tracing_thresh", 0644, d_tracer,
4188 &tracing_thresh, &tracing_max_lat_fops);
4191 trace_create_file("README", 0444, d_tracer,
4192 NULL, &tracing_readme_fops);
4194 trace_create_file("trace_pipe", 0444, d_tracer,
4195 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4197 trace_create_file("buffer_size_kb", 0644, d_tracer,
4198 &global_trace, &tracing_entries_fops);
4200 trace_create_file("trace_marker", 0220, d_tracer,
4201 NULL, &tracing_mark_fops);
4203 trace_create_file("saved_cmdlines", 0444, d_tracer,
4204 NULL, &tracing_saved_cmdlines_fops);
4206 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4209 #ifdef CONFIG_DYNAMIC_FTRACE
4210 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4211 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4213 #ifdef CONFIG_SYSPROF_TRACER
4214 init_tracer_sysprof_debugfs(d_tracer);
4217 create_trace_options_dir();
4219 for_each_tracing_cpu(cpu)
4220 tracing_init_debugfs_percpu(cpu);
4225 static int trace_panic_handler(struct notifier_block *this,
4226 unsigned long event, void *unused)
4228 if (ftrace_dump_on_oops)
4233 static struct notifier_block trace_panic_notifier = {
4234 .notifier_call = trace_panic_handler,
4236 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4239 static int trace_die_handler(struct notifier_block *self,
4245 if (ftrace_dump_on_oops)
4254 static struct notifier_block trace_die_notifier = {
4255 .notifier_call = trace_die_handler,
4260 * printk is set to max of 1024, we really don't need it that big.
4261 * Nothing should be printing 1000 characters anyway.
4263 #define TRACE_MAX_PRINT 1000
4266 * Define here KERN_TRACE so that we have one place to modify
4267 * it if we decide to change what log level the ftrace dump
4270 #define KERN_TRACE KERN_EMERG
4273 trace_printk_seq(struct trace_seq *s)
4275 /* Probably should print a warning here. */
4279 /* should be zero ended, but we are paranoid. */
4280 s->buffer[s->len] = 0;
4282 printk(KERN_TRACE "%s", s->buffer);
4287 static void __ftrace_dump(bool disable_tracing)
4289 static raw_spinlock_t ftrace_dump_lock =
4290 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
4291 /* use static because iter can be a bit big for the stack */
4292 static struct trace_iterator iter;
4293 unsigned int old_userobj;
4294 static int dump_ran;
4295 unsigned long flags;
4299 local_irq_save(flags);
4300 __raw_spin_lock(&ftrace_dump_lock);
4308 if (disable_tracing)
4311 for_each_tracing_cpu(cpu) {
4312 atomic_inc(&global_trace.data[cpu]->disabled);
4315 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4317 /* don't look at user memory in panic mode */
4318 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4320 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4322 /* Simulate the iterator */
4323 iter.tr = &global_trace;
4324 iter.trace = current_trace;
4325 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4328 * We need to stop all tracing on all CPUS to read the
4329 * the next buffer. This is a bit expensive, but is
4330 * not done often. We fill all what we can read,
4331 * and then release the locks again.
4334 while (!trace_empty(&iter)) {
4337 printk(KERN_TRACE "---------------------------------\n");
4341 /* reset all but tr, trace, and overruns */
4342 memset(&iter.seq, 0,
4343 sizeof(struct trace_iterator) -
4344 offsetof(struct trace_iterator, seq));
4345 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4348 if (find_next_entry_inc(&iter) != NULL) {
4351 ret = print_trace_line(&iter);
4352 if (ret != TRACE_TYPE_NO_CONSUME)
4353 trace_consume(&iter);
4356 trace_printk_seq(&iter.seq);
4360 printk(KERN_TRACE " (ftrace buffer empty)\n");
4362 printk(KERN_TRACE "---------------------------------\n");
4364 /* Re-enable tracing if requested */
4365 if (!disable_tracing) {
4366 trace_flags |= old_userobj;
4368 for_each_tracing_cpu(cpu) {
4369 atomic_dec(&global_trace.data[cpu]->disabled);
4375 __raw_spin_unlock(&ftrace_dump_lock);
4376 local_irq_restore(flags);
4379 /* By default: disable tracing after the dump */
4380 void ftrace_dump(void)
4382 __ftrace_dump(true);
4385 __init static int tracer_alloc_buffers(void)
4391 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4394 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4395 goto out_free_buffer_mask;
4397 if (!zalloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
4398 goto out_free_tracing_cpumask;
4400 /* To save memory, keep the ring buffer size to its minimum */
4401 if (ring_buffer_expanded)
4402 ring_buf_size = trace_buf_size;
4406 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4407 cpumask_copy(tracing_cpumask, cpu_all_mask);
4409 /* TODO: make the number of buffers hot pluggable with CPUS */
4410 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4411 TRACE_BUFFER_FLAGS);
4412 if (!global_trace.buffer) {
4413 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4415 goto out_free_cpumask;
4417 global_trace.entries = ring_buffer_size(global_trace.buffer);
4420 #ifdef CONFIG_TRACER_MAX_TRACE
4421 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4422 TRACE_BUFFER_FLAGS);
4423 if (!max_tr.buffer) {
4424 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4426 ring_buffer_free(global_trace.buffer);
4427 goto out_free_cpumask;
4429 max_tr.entries = ring_buffer_size(max_tr.buffer);
4430 WARN_ON(max_tr.entries != global_trace.entries);
4433 /* Allocate the first page for all buffers */
4434 for_each_tracing_cpu(i) {
4435 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4436 max_tr.data[i] = &per_cpu(max_data, i);
4439 trace_init_cmdlines();
4441 register_tracer(&nop_trace);
4442 current_trace = &nop_trace;
4443 #ifdef CONFIG_BOOT_TRACER
4444 register_tracer(&boot_tracer);
4446 /* All seems OK, enable tracing */
4447 tracing_disabled = 0;
4449 atomic_notifier_chain_register(&panic_notifier_list,
4450 &trace_panic_notifier);
4452 register_die_notifier(&trace_die_notifier);
4457 free_cpumask_var(tracing_reader_cpumask);
4458 out_free_tracing_cpumask:
4459 free_cpumask_var(tracing_cpumask);
4460 out_free_buffer_mask:
4461 free_cpumask_var(tracing_buffer_mask);
4466 __init static int clear_boot_tracer(void)
4469 * The default tracer at boot buffer is an init section.
4470 * This function is called in lateinit. If we did not
4471 * find the boot tracer, then clear it out, to prevent
4472 * later registration from accessing the buffer that is
4473 * about to be freed.
4475 if (!default_bootup_tracer)
4478 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4479 default_bootup_tracer);
4480 default_bootup_tracer = NULL;
4485 early_initcall(tracer_alloc_buffers);
4486 fs_initcall(tracer_init_debugfs);
4487 late_initcall(clear_boot_tracer);