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 <generated/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(int, ftrace_cpu_disabled);
91 static inline void ftrace_disable_cpu(void)
94 __this_cpu_inc(per_cpu_var(ftrace_cpu_disabled));
97 static inline void ftrace_enable_cpu(void)
99 __this_cpu_dec(per_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_tr_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 arch_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 arch_spinlock_t ftrace_max_lock =
503 (arch_spinlock_t)__ARCH_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 arch_spin_lock(&ftrace_max_lock);
559 tr->buffer = max_tr.buffer;
562 __update_max_tr(tr, tsk, cpu);
563 arch_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 arch_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 arch_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 arch_spinlock_t trace_cmdline_lock = __ARCH_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 (!arch_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 arch_spin_unlock(&trace_cmdline_lock);
945 void trace_find_cmdline(int pid, char comm[])
950 strcpy(comm, "<idle>");
954 if (WARN_ON_ONCE(pid < 0)) {
955 strcpy(comm, "<XXX>");
959 if (pid > PID_MAX_DEFAULT) {
960 strcpy(comm, "<...>");
965 arch_spin_lock(&trace_cmdline_lock);
966 map = map_pid_to_cmdline[pid];
967 if (map != NO_CMDLINE_MAP)
968 strcpy(comm, saved_cmdlines[map]);
970 strcpy(comm, "<...>");
972 arch_spin_unlock(&trace_cmdline_lock);
976 void tracing_record_cmdline(struct task_struct *tsk)
978 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
982 trace_save_cmdline(tsk);
986 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
989 struct task_struct *tsk = current;
991 entry->preempt_count = pc & 0xff;
992 entry->pid = (tsk) ? tsk->pid : 0;
993 entry->lock_depth = (tsk) ? tsk->lock_depth : 0;
995 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
996 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
998 TRACE_FLAG_IRQS_NOSUPPORT |
1000 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1001 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1002 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1004 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1006 struct ring_buffer_event *
1007 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1010 unsigned long flags, int pc)
1012 struct ring_buffer_event *event;
1014 event = ring_buffer_lock_reserve(buffer, len);
1015 if (event != NULL) {
1016 struct trace_entry *ent = ring_buffer_event_data(event);
1018 tracing_generic_entry_update(ent, flags, pc);
1026 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1027 struct ring_buffer_event *event,
1028 unsigned long flags, int pc,
1031 ring_buffer_unlock_commit(buffer, event);
1033 ftrace_trace_stack(buffer, flags, 6, pc);
1034 ftrace_trace_userstack(buffer, flags, pc);
1040 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1041 struct ring_buffer_event *event,
1042 unsigned long flags, int pc)
1044 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1047 struct ring_buffer_event *
1048 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1049 int type, unsigned long len,
1050 unsigned long flags, int pc)
1052 *current_rb = global_trace.buffer;
1053 return trace_buffer_lock_reserve(*current_rb,
1054 type, len, flags, pc);
1056 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1058 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1059 struct ring_buffer_event *event,
1060 unsigned long flags, int pc)
1062 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1064 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1066 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1067 struct ring_buffer_event *event,
1068 unsigned long flags, int pc)
1070 __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1072 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1074 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1075 struct ring_buffer_event *event)
1077 ring_buffer_discard_commit(buffer, event);
1079 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1082 trace_function(struct trace_array *tr,
1083 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1086 struct ftrace_event_call *call = &event_function;
1087 struct ring_buffer *buffer = tr->buffer;
1088 struct ring_buffer_event *event;
1089 struct ftrace_entry *entry;
1091 /* If we are reading the ring buffer, don't trace */
1092 if (unlikely(__this_cpu_read(per_cpu_var(ftrace_cpu_disabled))))
1095 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1099 entry = ring_buffer_event_data(event);
1101 entry->parent_ip = parent_ip;
1103 if (!filter_check_discard(call, entry, buffer, event))
1104 ring_buffer_unlock_commit(buffer, event);
1108 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1109 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1112 if (likely(!atomic_read(&data->disabled)))
1113 trace_function(tr, ip, parent_ip, flags, pc);
1116 #ifdef CONFIG_STACKTRACE
1117 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1118 unsigned long flags,
1121 struct ftrace_event_call *call = &event_kernel_stack;
1122 struct ring_buffer_event *event;
1123 struct stack_entry *entry;
1124 struct stack_trace trace;
1126 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1127 sizeof(*entry), flags, pc);
1130 entry = ring_buffer_event_data(event);
1131 memset(&entry->caller, 0, sizeof(entry->caller));
1133 trace.nr_entries = 0;
1134 trace.max_entries = FTRACE_STACK_ENTRIES;
1136 trace.entries = entry->caller;
1138 save_stack_trace(&trace);
1139 if (!filter_check_discard(call, entry, buffer, event))
1140 ring_buffer_unlock_commit(buffer, event);
1143 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1146 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1149 __ftrace_trace_stack(buffer, flags, skip, pc);
1152 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1155 __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1159 * trace_dump_stack - record a stack back trace in the trace buffer
1161 void trace_dump_stack(void)
1163 unsigned long flags;
1165 if (tracing_disabled || tracing_selftest_running)
1168 local_save_flags(flags);
1170 /* skipping 3 traces, seems to get us at the caller of this function */
1171 __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count());
1175 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1177 struct ftrace_event_call *call = &event_user_stack;
1178 struct ring_buffer_event *event;
1179 struct userstack_entry *entry;
1180 struct stack_trace trace;
1182 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1185 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1186 sizeof(*entry), flags, pc);
1189 entry = ring_buffer_event_data(event);
1191 entry->tgid = current->tgid;
1192 memset(&entry->caller, 0, sizeof(entry->caller));
1194 trace.nr_entries = 0;
1195 trace.max_entries = FTRACE_STACK_ENTRIES;
1197 trace.entries = entry->caller;
1199 save_stack_trace_user(&trace);
1200 if (!filter_check_discard(call, entry, buffer, event))
1201 ring_buffer_unlock_commit(buffer, event);
1205 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1207 ftrace_trace_userstack(tr, flags, preempt_count());
1211 #endif /* CONFIG_STACKTRACE */
1214 ftrace_trace_special(void *__tr,
1215 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1218 struct ftrace_event_call *call = &event_special;
1219 struct ring_buffer_event *event;
1220 struct trace_array *tr = __tr;
1221 struct ring_buffer *buffer = tr->buffer;
1222 struct special_entry *entry;
1224 event = trace_buffer_lock_reserve(buffer, TRACE_SPECIAL,
1225 sizeof(*entry), 0, pc);
1228 entry = ring_buffer_event_data(event);
1233 if (!filter_check_discard(call, entry, buffer, event))
1234 trace_buffer_unlock_commit(buffer, event, 0, pc);
1238 __trace_special(void *__tr, void *__data,
1239 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1241 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1245 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1247 struct trace_array *tr = &global_trace;
1248 struct trace_array_cpu *data;
1249 unsigned long flags;
1253 if (tracing_disabled)
1256 pc = preempt_count();
1257 local_irq_save(flags);
1258 cpu = raw_smp_processor_id();
1259 data = tr->data[cpu];
1261 if (likely(atomic_inc_return(&data->disabled) == 1))
1262 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1264 atomic_dec(&data->disabled);
1265 local_irq_restore(flags);
1269 * trace_vbprintk - write binary msg to tracing buffer
1272 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1274 static arch_spinlock_t trace_buf_lock =
1275 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1276 static u32 trace_buf[TRACE_BUF_SIZE];
1278 struct ftrace_event_call *call = &event_bprint;
1279 struct ring_buffer_event *event;
1280 struct ring_buffer *buffer;
1281 struct trace_array *tr = &global_trace;
1282 struct trace_array_cpu *data;
1283 struct bprint_entry *entry;
1284 unsigned long flags;
1287 int cpu, len = 0, size, pc;
1289 if (unlikely(tracing_selftest_running || tracing_disabled))
1292 /* Don't pollute graph traces with trace_vprintk internals */
1293 pause_graph_tracing();
1295 pc = preempt_count();
1296 resched = ftrace_preempt_disable();
1297 cpu = raw_smp_processor_id();
1298 data = tr->data[cpu];
1300 disable = atomic_inc_return(&data->disabled);
1301 if (unlikely(disable != 1))
1304 /* Lockdep uses trace_printk for lock tracing */
1305 local_irq_save(flags);
1306 arch_spin_lock(&trace_buf_lock);
1307 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1309 if (len > TRACE_BUF_SIZE || len < 0)
1312 size = sizeof(*entry) + sizeof(u32) * len;
1313 buffer = tr->buffer;
1314 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1318 entry = ring_buffer_event_data(event);
1322 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1323 if (!filter_check_discard(call, entry, buffer, event))
1324 ring_buffer_unlock_commit(buffer, event);
1327 arch_spin_unlock(&trace_buf_lock);
1328 local_irq_restore(flags);
1331 atomic_dec_return(&data->disabled);
1332 ftrace_preempt_enable(resched);
1333 unpause_graph_tracing();
1337 EXPORT_SYMBOL_GPL(trace_vbprintk);
1339 int trace_array_printk(struct trace_array *tr,
1340 unsigned long ip, const char *fmt, ...)
1345 if (!(trace_flags & TRACE_ITER_PRINTK))
1349 ret = trace_array_vprintk(tr, ip, fmt, ap);
1354 int trace_array_vprintk(struct trace_array *tr,
1355 unsigned long ip, const char *fmt, va_list args)
1357 static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1358 static char trace_buf[TRACE_BUF_SIZE];
1360 struct ftrace_event_call *call = &event_print;
1361 struct ring_buffer_event *event;
1362 struct ring_buffer *buffer;
1363 struct trace_array_cpu *data;
1364 int cpu, len = 0, size, pc;
1365 struct print_entry *entry;
1366 unsigned long irq_flags;
1369 if (tracing_disabled || tracing_selftest_running)
1372 pc = preempt_count();
1373 preempt_disable_notrace();
1374 cpu = raw_smp_processor_id();
1375 data = tr->data[cpu];
1377 disable = atomic_inc_return(&data->disabled);
1378 if (unlikely(disable != 1))
1381 pause_graph_tracing();
1382 raw_local_irq_save(irq_flags);
1383 arch_spin_lock(&trace_buf_lock);
1384 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1386 size = sizeof(*entry) + len + 1;
1387 buffer = tr->buffer;
1388 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1392 entry = ring_buffer_event_data(event);
1395 memcpy(&entry->buf, trace_buf, len);
1396 entry->buf[len] = '\0';
1397 if (!filter_check_discard(call, entry, buffer, event))
1398 ring_buffer_unlock_commit(buffer, event);
1401 arch_spin_unlock(&trace_buf_lock);
1402 raw_local_irq_restore(irq_flags);
1403 unpause_graph_tracing();
1405 atomic_dec_return(&data->disabled);
1406 preempt_enable_notrace();
1411 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1413 return trace_array_vprintk(&global_trace, ip, fmt, args);
1415 EXPORT_SYMBOL_GPL(trace_vprintk);
1417 enum trace_file_type {
1418 TRACE_FILE_LAT_FMT = 1,
1419 TRACE_FILE_ANNOTATE = 2,
1422 static void trace_iterator_increment(struct trace_iterator *iter)
1424 /* Don't allow ftrace to trace into the ring buffers */
1425 ftrace_disable_cpu();
1428 if (iter->buffer_iter[iter->cpu])
1429 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1431 ftrace_enable_cpu();
1434 static struct trace_entry *
1435 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1437 struct ring_buffer_event *event;
1438 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1440 /* Don't allow ftrace to trace into the ring buffers */
1441 ftrace_disable_cpu();
1444 event = ring_buffer_iter_peek(buf_iter, ts);
1446 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1448 ftrace_enable_cpu();
1450 return event ? ring_buffer_event_data(event) : NULL;
1453 static struct trace_entry *
1454 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1456 struct ring_buffer *buffer = iter->tr->buffer;
1457 struct trace_entry *ent, *next = NULL;
1458 int cpu_file = iter->cpu_file;
1459 u64 next_ts = 0, ts;
1464 * If we are in a per_cpu trace file, don't bother by iterating over
1465 * all cpu and peek directly.
1467 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1468 if (ring_buffer_empty_cpu(buffer, cpu_file))
1470 ent = peek_next_entry(iter, cpu_file, ent_ts);
1472 *ent_cpu = cpu_file;
1477 for_each_tracing_cpu(cpu) {
1479 if (ring_buffer_empty_cpu(buffer, cpu))
1482 ent = peek_next_entry(iter, cpu, &ts);
1485 * Pick the entry with the smallest timestamp:
1487 if (ent && (!next || ts < next_ts)) {
1495 *ent_cpu = next_cpu;
1503 /* Find the next real entry, without updating the iterator itself */
1504 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1505 int *ent_cpu, u64 *ent_ts)
1507 return __find_next_entry(iter, ent_cpu, ent_ts);
1510 /* Find the next real entry, and increment the iterator to the next entry */
1511 static void *find_next_entry_inc(struct trace_iterator *iter)
1513 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1516 trace_iterator_increment(iter);
1518 return iter->ent ? iter : NULL;
1521 static void trace_consume(struct trace_iterator *iter)
1523 /* Don't allow ftrace to trace into the ring buffers */
1524 ftrace_disable_cpu();
1525 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1526 ftrace_enable_cpu();
1529 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1531 struct trace_iterator *iter = m->private;
1535 WARN_ON_ONCE(iter->leftover);
1539 /* can't go backwards */
1544 ent = find_next_entry_inc(iter);
1548 while (ent && iter->idx < i)
1549 ent = find_next_entry_inc(iter);
1556 static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1558 struct trace_array *tr = iter->tr;
1559 struct ring_buffer_event *event;
1560 struct ring_buffer_iter *buf_iter;
1561 unsigned long entries = 0;
1564 tr->data[cpu]->skipped_entries = 0;
1566 if (!iter->buffer_iter[cpu])
1569 buf_iter = iter->buffer_iter[cpu];
1570 ring_buffer_iter_reset(buf_iter);
1573 * We could have the case with the max latency tracers
1574 * that a reset never took place on a cpu. This is evident
1575 * by the timestamp being before the start of the buffer.
1577 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1578 if (ts >= iter->tr->time_start)
1581 ring_buffer_read(buf_iter, NULL);
1584 tr->data[cpu]->skipped_entries = entries;
1588 * No necessary locking here. The worst thing which can
1589 * happen is loosing events consumed at the same time
1590 * by a trace_pipe reader.
1591 * Other than that, we don't risk to crash the ring buffer
1592 * because it serializes the readers.
1594 * The current tracer is copied to avoid a global locking
1597 static void *s_start(struct seq_file *m, loff_t *pos)
1599 struct trace_iterator *iter = m->private;
1600 static struct tracer *old_tracer;
1601 int cpu_file = iter->cpu_file;
1606 /* copy the tracer to avoid using a global lock all around */
1607 mutex_lock(&trace_types_lock);
1608 if (unlikely(old_tracer != current_trace && current_trace)) {
1609 old_tracer = current_trace;
1610 *iter->trace = *current_trace;
1612 mutex_unlock(&trace_types_lock);
1614 atomic_inc(&trace_record_cmdline_disabled);
1616 if (*pos != iter->pos) {
1621 ftrace_disable_cpu();
1623 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1624 for_each_tracing_cpu(cpu)
1625 tracing_iter_reset(iter, cpu);
1627 tracing_iter_reset(iter, cpu_file);
1629 ftrace_enable_cpu();
1631 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1636 * If we overflowed the seq_file before, then we want
1637 * to just reuse the trace_seq buffer again.
1643 p = s_next(m, p, &l);
1647 trace_event_read_lock();
1651 static void s_stop(struct seq_file *m, void *p)
1653 atomic_dec(&trace_record_cmdline_disabled);
1654 trace_event_read_unlock();
1657 static void print_lat_help_header(struct seq_file *m)
1659 seq_puts(m, "# _------=> CPU# \n");
1660 seq_puts(m, "# / _-----=> irqs-off \n");
1661 seq_puts(m, "# | / _----=> need-resched \n");
1662 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1663 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1664 seq_puts(m, "# |||| /_--=> lock-depth \n");
1665 seq_puts(m, "# |||||/ delay \n");
1666 seq_puts(m, "# cmd pid |||||| time | caller \n");
1667 seq_puts(m, "# \\ / |||||| \\ | / \n");
1670 static void print_func_help_header(struct seq_file *m)
1672 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1673 seq_puts(m, "# | | | | |\n");
1678 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1680 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1681 struct trace_array *tr = iter->tr;
1682 struct trace_array_cpu *data = tr->data[tr->cpu];
1683 struct tracer *type = current_trace;
1684 unsigned long entries = 0;
1685 unsigned long total = 0;
1686 unsigned long count;
1687 const char *name = "preemption";
1694 for_each_tracing_cpu(cpu) {
1695 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1697 * If this buffer has skipped entries, then we hold all
1698 * entries for the trace and we need to ignore the
1699 * ones before the time stamp.
1701 if (tr->data[cpu]->skipped_entries) {
1702 count -= tr->data[cpu]->skipped_entries;
1703 /* total is the same as the entries */
1707 ring_buffer_overrun_cpu(tr->buffer, cpu);
1711 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1713 seq_puts(m, "# -----------------------------------"
1714 "---------------------------------\n");
1715 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1716 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1717 nsecs_to_usecs(data->saved_latency),
1721 #if defined(CONFIG_PREEMPT_NONE)
1723 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1725 #elif defined(CONFIG_PREEMPT)
1730 /* These are reserved for later use */
1733 seq_printf(m, " #P:%d)\n", num_online_cpus());
1737 seq_puts(m, "# -----------------\n");
1738 seq_printf(m, "# | task: %.16s-%d "
1739 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1740 data->comm, data->pid, data->uid, data->nice,
1741 data->policy, data->rt_priority);
1742 seq_puts(m, "# -----------------\n");
1744 if (data->critical_start) {
1745 seq_puts(m, "# => started at: ");
1746 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1747 trace_print_seq(m, &iter->seq);
1748 seq_puts(m, "\n# => ended at: ");
1749 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1750 trace_print_seq(m, &iter->seq);
1751 seq_puts(m, "\n#\n");
1757 static void test_cpu_buff_start(struct trace_iterator *iter)
1759 struct trace_seq *s = &iter->seq;
1761 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1764 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1767 if (cpumask_test_cpu(iter->cpu, iter->started))
1770 if (iter->tr->data[iter->cpu]->skipped_entries)
1773 cpumask_set_cpu(iter->cpu, iter->started);
1775 /* Don't print started cpu buffer for the first entry of the trace */
1777 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1781 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1783 struct trace_seq *s = &iter->seq;
1784 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1785 struct trace_entry *entry;
1786 struct trace_event *event;
1790 test_cpu_buff_start(iter);
1792 event = ftrace_find_event(entry->type);
1794 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1795 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1796 if (!trace_print_lat_context(iter))
1799 if (!trace_print_context(iter))
1805 return event->trace(iter, sym_flags);
1807 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1810 return TRACE_TYPE_HANDLED;
1812 return TRACE_TYPE_PARTIAL_LINE;
1815 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1817 struct trace_seq *s = &iter->seq;
1818 struct trace_entry *entry;
1819 struct trace_event *event;
1823 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1824 if (!trace_seq_printf(s, "%d %d %llu ",
1825 entry->pid, iter->cpu, iter->ts))
1829 event = ftrace_find_event(entry->type);
1831 return event->raw(iter, 0);
1833 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1836 return TRACE_TYPE_HANDLED;
1838 return TRACE_TYPE_PARTIAL_LINE;
1841 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1843 struct trace_seq *s = &iter->seq;
1844 unsigned char newline = '\n';
1845 struct trace_entry *entry;
1846 struct trace_event *event;
1850 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1851 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1852 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1853 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1856 event = ftrace_find_event(entry->type);
1858 enum print_line_t ret = event->hex(iter, 0);
1859 if (ret != TRACE_TYPE_HANDLED)
1863 SEQ_PUT_FIELD_RET(s, newline);
1865 return TRACE_TYPE_HANDLED;
1868 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1870 struct trace_seq *s = &iter->seq;
1871 struct trace_entry *entry;
1872 struct trace_event *event;
1876 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1877 SEQ_PUT_FIELD_RET(s, entry->pid);
1878 SEQ_PUT_FIELD_RET(s, iter->cpu);
1879 SEQ_PUT_FIELD_RET(s, iter->ts);
1882 event = ftrace_find_event(entry->type);
1883 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1886 static int trace_empty(struct trace_iterator *iter)
1890 /* If we are looking at one CPU buffer, only check that one */
1891 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1892 cpu = iter->cpu_file;
1893 if (iter->buffer_iter[cpu]) {
1894 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1897 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1903 for_each_tracing_cpu(cpu) {
1904 if (iter->buffer_iter[cpu]) {
1905 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1908 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1916 /* Called with trace_event_read_lock() held. */
1917 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1919 enum print_line_t ret;
1921 if (iter->trace && iter->trace->print_line) {
1922 ret = iter->trace->print_line(iter);
1923 if (ret != TRACE_TYPE_UNHANDLED)
1927 if (iter->ent->type == TRACE_BPRINT &&
1928 trace_flags & TRACE_ITER_PRINTK &&
1929 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1930 return trace_print_bprintk_msg_only(iter);
1932 if (iter->ent->type == TRACE_PRINT &&
1933 trace_flags & TRACE_ITER_PRINTK &&
1934 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1935 return trace_print_printk_msg_only(iter);
1937 if (trace_flags & TRACE_ITER_BIN)
1938 return print_bin_fmt(iter);
1940 if (trace_flags & TRACE_ITER_HEX)
1941 return print_hex_fmt(iter);
1943 if (trace_flags & TRACE_ITER_RAW)
1944 return print_raw_fmt(iter);
1946 return print_trace_fmt(iter);
1949 static int s_show(struct seq_file *m, void *v)
1951 struct trace_iterator *iter = v;
1954 if (iter->ent == NULL) {
1956 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1959 if (iter->trace && iter->trace->print_header)
1960 iter->trace->print_header(m);
1961 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1962 /* print nothing if the buffers are empty */
1963 if (trace_empty(iter))
1965 print_trace_header(m, iter);
1966 if (!(trace_flags & TRACE_ITER_VERBOSE))
1967 print_lat_help_header(m);
1969 if (!(trace_flags & TRACE_ITER_VERBOSE))
1970 print_func_help_header(m);
1972 } else if (iter->leftover) {
1974 * If we filled the seq_file buffer earlier, we
1975 * want to just show it now.
1977 ret = trace_print_seq(m, &iter->seq);
1979 /* ret should this time be zero, but you never know */
1980 iter->leftover = ret;
1983 print_trace_line(iter);
1984 ret = trace_print_seq(m, &iter->seq);
1986 * If we overflow the seq_file buffer, then it will
1987 * ask us for this data again at start up.
1989 * ret is 0 if seq_file write succeeded.
1992 iter->leftover = ret;
1998 static const struct seq_operations tracer_seq_ops = {
2005 static struct trace_iterator *
2006 __tracing_open(struct inode *inode, struct file *file)
2008 long cpu_file = (long) inode->i_private;
2009 void *fail_ret = ERR_PTR(-ENOMEM);
2010 struct trace_iterator *iter;
2014 if (tracing_disabled)
2015 return ERR_PTR(-ENODEV);
2017 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2019 return ERR_PTR(-ENOMEM);
2022 * We make a copy of the current tracer to avoid concurrent
2023 * changes on it while we are reading.
2025 mutex_lock(&trace_types_lock);
2026 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2031 *iter->trace = *current_trace;
2033 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2036 if (current_trace && current_trace->print_max)
2039 iter->tr = &global_trace;
2041 mutex_init(&iter->mutex);
2042 iter->cpu_file = cpu_file;
2044 /* Notify the tracer early; before we stop tracing. */
2045 if (iter->trace && iter->trace->open)
2046 iter->trace->open(iter);
2048 /* Annotate start of buffers if we had overruns */
2049 if (ring_buffer_overruns(iter->tr->buffer))
2050 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2052 /* stop the trace while dumping */
2055 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2056 for_each_tracing_cpu(cpu) {
2058 iter->buffer_iter[cpu] =
2059 ring_buffer_read_start(iter->tr->buffer, cpu);
2060 tracing_iter_reset(iter, cpu);
2063 cpu = iter->cpu_file;
2064 iter->buffer_iter[cpu] =
2065 ring_buffer_read_start(iter->tr->buffer, cpu);
2066 tracing_iter_reset(iter, cpu);
2069 ret = seq_open(file, &tracer_seq_ops);
2071 fail_ret = ERR_PTR(ret);
2075 m = file->private_data;
2078 mutex_unlock(&trace_types_lock);
2083 for_each_tracing_cpu(cpu) {
2084 if (iter->buffer_iter[cpu])
2085 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2087 free_cpumask_var(iter->started);
2090 mutex_unlock(&trace_types_lock);
2097 int tracing_open_generic(struct inode *inode, struct file *filp)
2099 if (tracing_disabled)
2102 filp->private_data = inode->i_private;
2106 static int tracing_release(struct inode *inode, struct file *file)
2108 struct seq_file *m = (struct seq_file *)file->private_data;
2109 struct trace_iterator *iter;
2112 if (!(file->f_mode & FMODE_READ))
2117 mutex_lock(&trace_types_lock);
2118 for_each_tracing_cpu(cpu) {
2119 if (iter->buffer_iter[cpu])
2120 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2123 if (iter->trace && iter->trace->close)
2124 iter->trace->close(iter);
2126 /* reenable tracing if it was previously enabled */
2128 mutex_unlock(&trace_types_lock);
2130 seq_release(inode, file);
2131 mutex_destroy(&iter->mutex);
2132 free_cpumask_var(iter->started);
2138 static int tracing_open(struct inode *inode, struct file *file)
2140 struct trace_iterator *iter;
2143 /* If this file was open for write, then erase contents */
2144 if ((file->f_mode & FMODE_WRITE) &&
2145 (file->f_flags & O_TRUNC)) {
2146 long cpu = (long) inode->i_private;
2148 if (cpu == TRACE_PIPE_ALL_CPU)
2149 tracing_reset_online_cpus(&global_trace);
2151 tracing_reset(&global_trace, cpu);
2154 if (file->f_mode & FMODE_READ) {
2155 iter = __tracing_open(inode, file);
2157 ret = PTR_ERR(iter);
2158 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2159 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2165 t_next(struct seq_file *m, void *v, loff_t *pos)
2167 struct tracer *t = v;
2177 static void *t_start(struct seq_file *m, loff_t *pos)
2182 mutex_lock(&trace_types_lock);
2183 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2189 static void t_stop(struct seq_file *m, void *p)
2191 mutex_unlock(&trace_types_lock);
2194 static int t_show(struct seq_file *m, void *v)
2196 struct tracer *t = v;
2201 seq_printf(m, "%s", t->name);
2210 static const struct seq_operations show_traces_seq_ops = {
2217 static int show_traces_open(struct inode *inode, struct file *file)
2219 if (tracing_disabled)
2222 return seq_open(file, &show_traces_seq_ops);
2226 tracing_write_stub(struct file *filp, const char __user *ubuf,
2227 size_t count, loff_t *ppos)
2232 static const struct file_operations tracing_fops = {
2233 .open = tracing_open,
2235 .write = tracing_write_stub,
2236 .llseek = seq_lseek,
2237 .release = tracing_release,
2240 static const struct file_operations show_traces_fops = {
2241 .open = show_traces_open,
2243 .release = seq_release,
2247 * Only trace on a CPU if the bitmask is set:
2249 static cpumask_var_t tracing_cpumask;
2252 * The tracer itself will not take this lock, but still we want
2253 * to provide a consistent cpumask to user-space:
2255 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2258 * Temporary storage for the character representation of the
2259 * CPU bitmask (and one more byte for the newline):
2261 static char mask_str[NR_CPUS + 1];
2264 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2265 size_t count, loff_t *ppos)
2269 mutex_lock(&tracing_cpumask_update_lock);
2271 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2272 if (count - len < 2) {
2276 len += sprintf(mask_str + len, "\n");
2277 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2280 mutex_unlock(&tracing_cpumask_update_lock);
2286 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2287 size_t count, loff_t *ppos)
2290 cpumask_var_t tracing_cpumask_new;
2292 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2295 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2299 mutex_lock(&tracing_cpumask_update_lock);
2301 local_irq_disable();
2302 arch_spin_lock(&ftrace_max_lock);
2303 for_each_tracing_cpu(cpu) {
2305 * Increase/decrease the disabled counter if we are
2306 * about to flip a bit in the cpumask:
2308 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2309 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2310 atomic_inc(&global_trace.data[cpu]->disabled);
2312 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2313 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2314 atomic_dec(&global_trace.data[cpu]->disabled);
2317 arch_spin_unlock(&ftrace_max_lock);
2320 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2322 mutex_unlock(&tracing_cpumask_update_lock);
2323 free_cpumask_var(tracing_cpumask_new);
2328 free_cpumask_var(tracing_cpumask_new);
2333 static const struct file_operations tracing_cpumask_fops = {
2334 .open = tracing_open_generic,
2335 .read = tracing_cpumask_read,
2336 .write = tracing_cpumask_write,
2339 static int tracing_trace_options_show(struct seq_file *m, void *v)
2341 struct tracer_opt *trace_opts;
2345 mutex_lock(&trace_types_lock);
2346 tracer_flags = current_trace->flags->val;
2347 trace_opts = current_trace->flags->opts;
2349 for (i = 0; trace_options[i]; i++) {
2350 if (trace_flags & (1 << i))
2351 seq_printf(m, "%s\n", trace_options[i]);
2353 seq_printf(m, "no%s\n", trace_options[i]);
2356 for (i = 0; trace_opts[i].name; i++) {
2357 if (tracer_flags & trace_opts[i].bit)
2358 seq_printf(m, "%s\n", trace_opts[i].name);
2360 seq_printf(m, "no%s\n", trace_opts[i].name);
2362 mutex_unlock(&trace_types_lock);
2367 static int __set_tracer_option(struct tracer *trace,
2368 struct tracer_flags *tracer_flags,
2369 struct tracer_opt *opts, int neg)
2373 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2378 tracer_flags->val &= ~opts->bit;
2380 tracer_flags->val |= opts->bit;
2384 /* Try to assign a tracer specific option */
2385 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2387 struct tracer_flags *tracer_flags = trace->flags;
2388 struct tracer_opt *opts = NULL;
2391 for (i = 0; tracer_flags->opts[i].name; i++) {
2392 opts = &tracer_flags->opts[i];
2394 if (strcmp(cmp, opts->name) == 0)
2395 return __set_tracer_option(trace, trace->flags,
2402 static void set_tracer_flags(unsigned int mask, int enabled)
2404 /* do nothing if flag is already set */
2405 if (!!(trace_flags & mask) == !!enabled)
2409 trace_flags |= mask;
2411 trace_flags &= ~mask;
2415 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2416 size_t cnt, loff_t *ppos)
2424 if (cnt >= sizeof(buf))
2427 if (copy_from_user(&buf, ubuf, cnt))
2431 cmp = strstrip(buf);
2433 if (strncmp(cmp, "no", 2) == 0) {
2438 for (i = 0; trace_options[i]; i++) {
2439 if (strcmp(cmp, trace_options[i]) == 0) {
2440 set_tracer_flags(1 << i, !neg);
2445 /* If no option could be set, test the specific tracer options */
2446 if (!trace_options[i]) {
2447 mutex_lock(&trace_types_lock);
2448 ret = set_tracer_option(current_trace, cmp, neg);
2449 mutex_unlock(&trace_types_lock);
2459 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2461 if (tracing_disabled)
2463 return single_open(file, tracing_trace_options_show, NULL);
2466 static const struct file_operations tracing_iter_fops = {
2467 .open = tracing_trace_options_open,
2469 .llseek = seq_lseek,
2470 .release = single_release,
2471 .write = tracing_trace_options_write,
2474 static const char readme_msg[] =
2475 "tracing mini-HOWTO:\n\n"
2476 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2477 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2478 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2479 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2481 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2482 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2484 "# cat /sys/kernel/debug/tracing/trace_options\n"
2485 "noprint-parent nosym-offset nosym-addr noverbose\n"
2486 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2487 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2488 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2489 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2493 tracing_readme_read(struct file *filp, char __user *ubuf,
2494 size_t cnt, loff_t *ppos)
2496 return simple_read_from_buffer(ubuf, cnt, ppos,
2497 readme_msg, strlen(readme_msg));
2500 static const struct file_operations tracing_readme_fops = {
2501 .open = tracing_open_generic,
2502 .read = tracing_readme_read,
2506 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2507 size_t cnt, loff_t *ppos)
2516 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2520 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2528 for (i = 0; i < SAVED_CMDLINES; i++) {
2531 pid = map_cmdline_to_pid[i];
2532 if (pid == -1 || pid == NO_CMDLINE_MAP)
2535 trace_find_cmdline(pid, buf_comm);
2536 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2541 len = simple_read_from_buffer(ubuf, cnt, ppos,
2550 static const struct file_operations tracing_saved_cmdlines_fops = {
2551 .open = tracing_open_generic,
2552 .read = tracing_saved_cmdlines_read,
2556 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2557 size_t cnt, loff_t *ppos)
2562 r = sprintf(buf, "%u\n", tracer_enabled);
2563 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2567 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2568 size_t cnt, loff_t *ppos)
2570 struct trace_array *tr = filp->private_data;
2575 if (cnt >= sizeof(buf))
2578 if (copy_from_user(&buf, ubuf, cnt))
2583 ret = strict_strtoul(buf, 10, &val);
2589 mutex_lock(&trace_types_lock);
2590 if (tracer_enabled ^ val) {
2593 if (current_trace->start)
2594 current_trace->start(tr);
2599 if (current_trace->stop)
2600 current_trace->stop(tr);
2603 mutex_unlock(&trace_types_lock);
2611 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2612 size_t cnt, loff_t *ppos)
2614 char buf[MAX_TRACER_SIZE+2];
2617 mutex_lock(&trace_types_lock);
2619 r = sprintf(buf, "%s\n", current_trace->name);
2621 r = sprintf(buf, "\n");
2622 mutex_unlock(&trace_types_lock);
2624 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2627 int tracer_init(struct tracer *t, struct trace_array *tr)
2629 tracing_reset_online_cpus(tr);
2633 static int tracing_resize_ring_buffer(unsigned long size)
2638 * If kernel or user changes the size of the ring buffer
2639 * we use the size that was given, and we can forget about
2640 * expanding it later.
2642 ring_buffer_expanded = 1;
2644 ret = ring_buffer_resize(global_trace.buffer, size);
2648 ret = ring_buffer_resize(max_tr.buffer, size);
2652 r = ring_buffer_resize(global_trace.buffer,
2653 global_trace.entries);
2656 * AARGH! We are left with different
2657 * size max buffer!!!!
2658 * The max buffer is our "snapshot" buffer.
2659 * When a tracer needs a snapshot (one of the
2660 * latency tracers), it swaps the max buffer
2661 * with the saved snap shot. We succeeded to
2662 * update the size of the main buffer, but failed to
2663 * update the size of the max buffer. But when we tried
2664 * to reset the main buffer to the original size, we
2665 * failed there too. This is very unlikely to
2666 * happen, but if it does, warn and kill all
2670 tracing_disabled = 1;
2675 global_trace.entries = size;
2681 * tracing_update_buffers - used by tracing facility to expand ring buffers
2683 * To save on memory when the tracing is never used on a system with it
2684 * configured in. The ring buffers are set to a minimum size. But once
2685 * a user starts to use the tracing facility, then they need to grow
2686 * to their default size.
2688 * This function is to be called when a tracer is about to be used.
2690 int tracing_update_buffers(void)
2694 mutex_lock(&trace_types_lock);
2695 if (!ring_buffer_expanded)
2696 ret = tracing_resize_ring_buffer(trace_buf_size);
2697 mutex_unlock(&trace_types_lock);
2702 struct trace_option_dentry;
2704 static struct trace_option_dentry *
2705 create_trace_option_files(struct tracer *tracer);
2708 destroy_trace_option_files(struct trace_option_dentry *topts);
2710 static int tracing_set_tracer(const char *buf)
2712 static struct trace_option_dentry *topts;
2713 struct trace_array *tr = &global_trace;
2717 mutex_lock(&trace_types_lock);
2719 if (!ring_buffer_expanded) {
2720 ret = tracing_resize_ring_buffer(trace_buf_size);
2726 for (t = trace_types; t; t = t->next) {
2727 if (strcmp(t->name, buf) == 0)
2734 if (t == current_trace)
2737 trace_branch_disable();
2738 if (current_trace && current_trace->reset)
2739 current_trace->reset(tr);
2741 destroy_trace_option_files(topts);
2745 topts = create_trace_option_files(current_trace);
2748 ret = tracer_init(t, tr);
2753 trace_branch_enable(tr);
2755 mutex_unlock(&trace_types_lock);
2761 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2762 size_t cnt, loff_t *ppos)
2764 char buf[MAX_TRACER_SIZE+1];
2771 if (cnt > MAX_TRACER_SIZE)
2772 cnt = MAX_TRACER_SIZE;
2774 if (copy_from_user(&buf, ubuf, cnt))
2779 /* strip ending whitespace. */
2780 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2783 err = tracing_set_tracer(buf);
2793 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2794 size_t cnt, loff_t *ppos)
2796 unsigned long *ptr = filp->private_data;
2800 r = snprintf(buf, sizeof(buf), "%ld\n",
2801 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2802 if (r > sizeof(buf))
2804 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2808 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2809 size_t cnt, loff_t *ppos)
2811 unsigned long *ptr = filp->private_data;
2816 if (cnt >= sizeof(buf))
2819 if (copy_from_user(&buf, ubuf, cnt))
2824 ret = strict_strtoul(buf, 10, &val);
2833 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2835 long cpu_file = (long) inode->i_private;
2836 struct trace_iterator *iter;
2839 if (tracing_disabled)
2842 mutex_lock(&trace_types_lock);
2844 /* We only allow one reader per cpu */
2845 if (cpu_file == TRACE_PIPE_ALL_CPU) {
2846 if (!cpumask_empty(tracing_reader_cpumask)) {
2850 cpumask_setall(tracing_reader_cpumask);
2852 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2853 cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2860 /* create a buffer to store the information to pass to userspace */
2861 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2868 * We make a copy of the current tracer to avoid concurrent
2869 * changes on it while we are reading.
2871 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2877 *iter->trace = *current_trace;
2879 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2884 /* trace pipe does not show start of buffer */
2885 cpumask_setall(iter->started);
2887 if (trace_flags & TRACE_ITER_LATENCY_FMT)
2888 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2890 iter->cpu_file = cpu_file;
2891 iter->tr = &global_trace;
2892 mutex_init(&iter->mutex);
2893 filp->private_data = iter;
2895 if (iter->trace->pipe_open)
2896 iter->trace->pipe_open(iter);
2899 mutex_unlock(&trace_types_lock);
2905 mutex_unlock(&trace_types_lock);
2909 static int tracing_release_pipe(struct inode *inode, struct file *file)
2911 struct trace_iterator *iter = file->private_data;
2913 mutex_lock(&trace_types_lock);
2915 if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2916 cpumask_clear(tracing_reader_cpumask);
2918 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2921 if (iter->trace->pipe_close)
2922 iter->trace->pipe_close(iter);
2924 mutex_unlock(&trace_types_lock);
2926 free_cpumask_var(iter->started);
2927 mutex_destroy(&iter->mutex);
2935 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2937 struct trace_iterator *iter = filp->private_data;
2939 if (trace_flags & TRACE_ITER_BLOCK) {
2941 * Always select as readable when in blocking mode
2943 return POLLIN | POLLRDNORM;
2945 if (!trace_empty(iter))
2946 return POLLIN | POLLRDNORM;
2947 poll_wait(filp, &trace_wait, poll_table);
2948 if (!trace_empty(iter))
2949 return POLLIN | POLLRDNORM;
2956 void default_wait_pipe(struct trace_iterator *iter)
2960 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2962 if (trace_empty(iter))
2965 finish_wait(&trace_wait, &wait);
2969 * This is a make-shift waitqueue.
2970 * A tracer might use this callback on some rare cases:
2972 * 1) the current tracer might hold the runqueue lock when it wakes up
2973 * a reader, hence a deadlock (sched, function, and function graph tracers)
2974 * 2) the function tracers, trace all functions, we don't want
2975 * the overhead of calling wake_up and friends
2976 * (and tracing them too)
2978 * Anyway, this is really very primitive wakeup.
2980 void poll_wait_pipe(struct trace_iterator *iter)
2982 set_current_state(TASK_INTERRUPTIBLE);
2983 /* sleep for 100 msecs, and try again. */
2984 schedule_timeout(HZ / 10);
2987 /* Must be called with trace_types_lock mutex held. */
2988 static int tracing_wait_pipe(struct file *filp)
2990 struct trace_iterator *iter = filp->private_data;
2992 while (trace_empty(iter)) {
2994 if ((filp->f_flags & O_NONBLOCK)) {
2998 mutex_unlock(&iter->mutex);
3000 iter->trace->wait_pipe(iter);
3002 mutex_lock(&iter->mutex);
3004 if (signal_pending(current))
3008 * We block until we read something and tracing is disabled.
3009 * We still block if tracing is disabled, but we have never
3010 * read anything. This allows a user to cat this file, and
3011 * then enable tracing. But after we have read something,
3012 * we give an EOF when tracing is again disabled.
3014 * iter->pos will be 0 if we haven't read anything.
3016 if (!tracer_enabled && iter->pos)
3027 tracing_read_pipe(struct file *filp, char __user *ubuf,
3028 size_t cnt, loff_t *ppos)
3030 struct trace_iterator *iter = filp->private_data;
3031 static struct tracer *old_tracer;
3034 /* return any leftover data */
3035 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3039 trace_seq_init(&iter->seq);
3041 /* copy the tracer to avoid using a global lock all around */
3042 mutex_lock(&trace_types_lock);
3043 if (unlikely(old_tracer != current_trace && current_trace)) {
3044 old_tracer = current_trace;
3045 *iter->trace = *current_trace;
3047 mutex_unlock(&trace_types_lock);
3050 * Avoid more than one consumer on a single file descriptor
3051 * This is just a matter of traces coherency, the ring buffer itself
3054 mutex_lock(&iter->mutex);
3055 if (iter->trace->read) {
3056 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3062 sret = tracing_wait_pipe(filp);
3066 /* stop when tracing is finished */
3067 if (trace_empty(iter)) {
3072 if (cnt >= PAGE_SIZE)
3073 cnt = PAGE_SIZE - 1;
3075 /* reset all but tr, trace, and overruns */
3076 memset(&iter->seq, 0,
3077 sizeof(struct trace_iterator) -
3078 offsetof(struct trace_iterator, seq));
3081 trace_event_read_lock();
3082 while (find_next_entry_inc(iter) != NULL) {
3083 enum print_line_t ret;
3084 int len = iter->seq.len;
3086 ret = print_trace_line(iter);
3087 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3088 /* don't print partial lines */
3089 iter->seq.len = len;
3092 if (ret != TRACE_TYPE_NO_CONSUME)
3093 trace_consume(iter);
3095 if (iter->seq.len >= cnt)
3098 trace_event_read_unlock();
3100 /* Now copy what we have to the user */
3101 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3102 if (iter->seq.readpos >= iter->seq.len)
3103 trace_seq_init(&iter->seq);
3106 * If there was nothing to send to user, inspite of consuming trace
3107 * entries, go back to wait for more entries.
3113 mutex_unlock(&iter->mutex);
3118 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3119 struct pipe_buffer *buf)
3121 __free_page(buf->page);
3124 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3127 __free_page(spd->pages[idx]);
3130 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3132 .map = generic_pipe_buf_map,
3133 .unmap = generic_pipe_buf_unmap,
3134 .confirm = generic_pipe_buf_confirm,
3135 .release = tracing_pipe_buf_release,
3136 .steal = generic_pipe_buf_steal,
3137 .get = generic_pipe_buf_get,
3141 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3146 /* Seq buffer is page-sized, exactly what we need. */
3148 count = iter->seq.len;
3149 ret = print_trace_line(iter);
3150 count = iter->seq.len - count;
3153 iter->seq.len -= count;
3156 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3157 iter->seq.len -= count;
3161 if (ret != TRACE_TYPE_NO_CONSUME)
3162 trace_consume(iter);
3164 if (!find_next_entry_inc(iter)) {
3174 static ssize_t tracing_splice_read_pipe(struct file *filp,
3176 struct pipe_inode_info *pipe,
3180 struct page *pages[PIPE_BUFFERS];
3181 struct partial_page partial[PIPE_BUFFERS];
3182 struct trace_iterator *iter = filp->private_data;
3183 struct splice_pipe_desc spd = {
3186 .nr_pages = 0, /* This gets updated below. */
3188 .ops = &tracing_pipe_buf_ops,
3189 .spd_release = tracing_spd_release_pipe,
3191 static struct tracer *old_tracer;
3196 /* copy the tracer to avoid using a global lock all around */
3197 mutex_lock(&trace_types_lock);
3198 if (unlikely(old_tracer != current_trace && current_trace)) {
3199 old_tracer = current_trace;
3200 *iter->trace = *current_trace;
3202 mutex_unlock(&trace_types_lock);
3204 mutex_lock(&iter->mutex);
3206 if (iter->trace->splice_read) {
3207 ret = iter->trace->splice_read(iter, filp,
3208 ppos, pipe, len, flags);
3213 ret = tracing_wait_pipe(filp);
3217 if (!iter->ent && !find_next_entry_inc(iter)) {
3222 trace_event_read_lock();
3224 /* Fill as many pages as possible. */
3225 for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3226 pages[i] = alloc_page(GFP_KERNEL);
3230 rem = tracing_fill_pipe_page(rem, iter);
3232 /* Copy the data into the page, so we can start over. */
3233 ret = trace_seq_to_buffer(&iter->seq,
3234 page_address(pages[i]),
3237 __free_page(pages[i]);
3240 partial[i].offset = 0;
3241 partial[i].len = iter->seq.len;
3243 trace_seq_init(&iter->seq);
3246 trace_event_read_unlock();
3247 mutex_unlock(&iter->mutex);
3251 return splice_to_pipe(pipe, &spd);
3254 mutex_unlock(&iter->mutex);
3260 tracing_entries_read(struct file *filp, char __user *ubuf,
3261 size_t cnt, loff_t *ppos)
3263 struct trace_array *tr = filp->private_data;
3267 mutex_lock(&trace_types_lock);
3268 if (!ring_buffer_expanded)
3269 r = sprintf(buf, "%lu (expanded: %lu)\n",
3271 trace_buf_size >> 10);
3273 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3274 mutex_unlock(&trace_types_lock);
3276 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3280 tracing_entries_write(struct file *filp, const char __user *ubuf,
3281 size_t cnt, loff_t *ppos)
3287 if (cnt >= sizeof(buf))
3290 if (copy_from_user(&buf, ubuf, cnt))
3295 ret = strict_strtoul(buf, 10, &val);
3299 /* must have at least 1 entry */
3303 mutex_lock(&trace_types_lock);
3307 /* disable all cpu buffers */
3308 for_each_tracing_cpu(cpu) {
3309 if (global_trace.data[cpu])
3310 atomic_inc(&global_trace.data[cpu]->disabled);
3311 if (max_tr.data[cpu])
3312 atomic_inc(&max_tr.data[cpu]->disabled);
3315 /* value is in KB */
3318 if (val != global_trace.entries) {
3319 ret = tracing_resize_ring_buffer(val);
3328 /* If check pages failed, return ENOMEM */
3329 if (tracing_disabled)
3332 for_each_tracing_cpu(cpu) {
3333 if (global_trace.data[cpu])
3334 atomic_dec(&global_trace.data[cpu]->disabled);
3335 if (max_tr.data[cpu])
3336 atomic_dec(&max_tr.data[cpu]->disabled);
3340 max_tr.entries = global_trace.entries;
3341 mutex_unlock(&trace_types_lock);
3346 static int mark_printk(const char *fmt, ...)
3350 va_start(args, fmt);
3351 ret = trace_vprintk(0, fmt, args);
3357 tracing_mark_write(struct file *filp, const char __user *ubuf,
3358 size_t cnt, loff_t *fpos)
3362 if (tracing_disabled)
3365 if (cnt > TRACE_BUF_SIZE)
3366 cnt = TRACE_BUF_SIZE;
3368 buf = kmalloc(cnt + 2, GFP_KERNEL);
3372 if (copy_from_user(buf, ubuf, cnt)) {
3376 if (buf[cnt-1] != '\n') {
3382 cnt = mark_printk("%s", buf);
3389 static int tracing_clock_show(struct seq_file *m, void *v)
3393 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3395 "%s%s%s%s", i ? " " : "",
3396 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3397 i == trace_clock_id ? "]" : "");
3403 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3404 size_t cnt, loff_t *fpos)
3407 const char *clockstr;
3410 if (cnt >= sizeof(buf))
3413 if (copy_from_user(&buf, ubuf, cnt))
3418 clockstr = strstrip(buf);
3420 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3421 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3424 if (i == ARRAY_SIZE(trace_clocks))
3429 mutex_lock(&trace_types_lock);
3431 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3433 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3435 mutex_unlock(&trace_types_lock);
3442 static int tracing_clock_open(struct inode *inode, struct file *file)
3444 if (tracing_disabled)
3446 return single_open(file, tracing_clock_show, NULL);
3449 static const struct file_operations tracing_max_lat_fops = {
3450 .open = tracing_open_generic,
3451 .read = tracing_max_lat_read,
3452 .write = tracing_max_lat_write,
3455 static const struct file_operations tracing_ctrl_fops = {
3456 .open = tracing_open_generic,
3457 .read = tracing_ctrl_read,
3458 .write = tracing_ctrl_write,
3461 static const struct file_operations set_tracer_fops = {
3462 .open = tracing_open_generic,
3463 .read = tracing_set_trace_read,
3464 .write = tracing_set_trace_write,
3467 static const struct file_operations tracing_pipe_fops = {
3468 .open = tracing_open_pipe,
3469 .poll = tracing_poll_pipe,
3470 .read = tracing_read_pipe,
3471 .splice_read = tracing_splice_read_pipe,
3472 .release = tracing_release_pipe,
3475 static const struct file_operations tracing_entries_fops = {
3476 .open = tracing_open_generic,
3477 .read = tracing_entries_read,
3478 .write = tracing_entries_write,
3481 static const struct file_operations tracing_mark_fops = {
3482 .open = tracing_open_generic,
3483 .write = tracing_mark_write,
3486 static const struct file_operations trace_clock_fops = {
3487 .open = tracing_clock_open,
3489 .llseek = seq_lseek,
3490 .release = single_release,
3491 .write = tracing_clock_write,
3494 struct ftrace_buffer_info {
3495 struct trace_array *tr;
3501 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3503 int cpu = (int)(long)inode->i_private;
3504 struct ftrace_buffer_info *info;
3506 if (tracing_disabled)
3509 info = kzalloc(sizeof(*info), GFP_KERNEL);
3513 info->tr = &global_trace;
3516 /* Force reading ring buffer for first read */
3517 info->read = (unsigned int)-1;
3519 filp->private_data = info;
3521 return nonseekable_open(inode, filp);
3525 tracing_buffers_read(struct file *filp, char __user *ubuf,
3526 size_t count, loff_t *ppos)
3528 struct ftrace_buffer_info *info = filp->private_data;
3537 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3541 /* Do we have previous read data to read? */
3542 if (info->read < PAGE_SIZE)
3547 ret = ring_buffer_read_page(info->tr->buffer,
3554 pos = ring_buffer_page_len(info->spare);
3556 if (pos < PAGE_SIZE)
3557 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3560 size = PAGE_SIZE - info->read;
3564 ret = copy_to_user(ubuf, info->spare + info->read, size);
3575 static int tracing_buffers_release(struct inode *inode, struct file *file)
3577 struct ftrace_buffer_info *info = file->private_data;
3580 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3587 struct ring_buffer *buffer;
3592 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3593 struct pipe_buffer *buf)
3595 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3600 ring_buffer_free_read_page(ref->buffer, ref->page);
3605 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3606 struct pipe_buffer *buf)
3611 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3612 struct pipe_buffer *buf)
3614 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3619 /* Pipe buffer operations for a buffer. */
3620 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
3622 .map = generic_pipe_buf_map,
3623 .unmap = generic_pipe_buf_unmap,
3624 .confirm = generic_pipe_buf_confirm,
3625 .release = buffer_pipe_buf_release,
3626 .steal = buffer_pipe_buf_steal,
3627 .get = buffer_pipe_buf_get,
3631 * Callback from splice_to_pipe(), if we need to release some pages
3632 * at the end of the spd in case we error'ed out in filling the pipe.
3634 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3636 struct buffer_ref *ref =
3637 (struct buffer_ref *)spd->partial[i].private;
3642 ring_buffer_free_read_page(ref->buffer, ref->page);
3644 spd->partial[i].private = 0;
3648 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3649 struct pipe_inode_info *pipe, size_t len,
3652 struct ftrace_buffer_info *info = file->private_data;
3653 struct partial_page partial[PIPE_BUFFERS];
3654 struct page *pages[PIPE_BUFFERS];
3655 struct splice_pipe_desc spd = {
3659 .ops = &buffer_pipe_buf_ops,
3660 .spd_release = buffer_spd_release,
3662 struct buffer_ref *ref;
3663 int entries, size, i;
3666 if (*ppos & (PAGE_SIZE - 1)) {
3667 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3671 if (len & (PAGE_SIZE - 1)) {
3672 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3673 if (len < PAGE_SIZE)
3678 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3680 for (i = 0; i < PIPE_BUFFERS && len && entries; i++, len -= PAGE_SIZE) {
3684 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3689 ref->buffer = info->tr->buffer;
3690 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3696 r = ring_buffer_read_page(ref->buffer, &ref->page,
3699 ring_buffer_free_read_page(ref->buffer,
3706 * zero out any left over data, this is going to
3709 size = ring_buffer_page_len(ref->page);
3710 if (size < PAGE_SIZE)
3711 memset(ref->page + size, 0, PAGE_SIZE - size);
3713 page = virt_to_page(ref->page);
3715 spd.pages[i] = page;
3716 spd.partial[i].len = PAGE_SIZE;
3717 spd.partial[i].offset = 0;
3718 spd.partial[i].private = (unsigned long)ref;
3722 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3727 /* did we read anything? */
3728 if (!spd.nr_pages) {
3729 if (flags & SPLICE_F_NONBLOCK)
3737 ret = splice_to_pipe(pipe, &spd);
3742 static const struct file_operations tracing_buffers_fops = {
3743 .open = tracing_buffers_open,
3744 .read = tracing_buffers_read,
3745 .release = tracing_buffers_release,
3746 .splice_read = tracing_buffers_splice_read,
3747 .llseek = no_llseek,
3751 tracing_stats_read(struct file *filp, char __user *ubuf,
3752 size_t count, loff_t *ppos)
3754 unsigned long cpu = (unsigned long)filp->private_data;
3755 struct trace_array *tr = &global_trace;
3756 struct trace_seq *s;
3759 s = kmalloc(sizeof(*s), GFP_KERNEL);
3765 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3766 trace_seq_printf(s, "entries: %ld\n", cnt);
3768 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3769 trace_seq_printf(s, "overrun: %ld\n", cnt);
3771 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3772 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3774 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3781 static const struct file_operations tracing_stats_fops = {
3782 .open = tracing_open_generic,
3783 .read = tracing_stats_read,
3786 #ifdef CONFIG_DYNAMIC_FTRACE
3788 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3794 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3795 size_t cnt, loff_t *ppos)
3797 static char ftrace_dyn_info_buffer[1024];
3798 static DEFINE_MUTEX(dyn_info_mutex);
3799 unsigned long *p = filp->private_data;
3800 char *buf = ftrace_dyn_info_buffer;
3801 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3804 mutex_lock(&dyn_info_mutex);
3805 r = sprintf(buf, "%ld ", *p);
3807 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3810 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3812 mutex_unlock(&dyn_info_mutex);
3817 static const struct file_operations tracing_dyn_info_fops = {
3818 .open = tracing_open_generic,
3819 .read = tracing_read_dyn_info,
3823 static struct dentry *d_tracer;
3825 struct dentry *tracing_init_dentry(void)
3832 if (!debugfs_initialized())
3835 d_tracer = debugfs_create_dir("tracing", NULL);
3837 if (!d_tracer && !once) {
3839 pr_warning("Could not create debugfs directory 'tracing'\n");
3846 static struct dentry *d_percpu;
3848 struct dentry *tracing_dentry_percpu(void)
3851 struct dentry *d_tracer;
3856 d_tracer = tracing_init_dentry();
3861 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3863 if (!d_percpu && !once) {
3865 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3872 static void tracing_init_debugfs_percpu(long cpu)
3874 struct dentry *d_percpu = tracing_dentry_percpu();
3875 struct dentry *d_cpu;
3876 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3879 if (cpu > 999 || cpu < 0)
3882 sprintf(cpu_dir, "cpu%ld", cpu);
3883 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3885 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3889 /* per cpu trace_pipe */
3890 trace_create_file("trace_pipe", 0444, d_cpu,
3891 (void *) cpu, &tracing_pipe_fops);
3894 trace_create_file("trace", 0644, d_cpu,
3895 (void *) cpu, &tracing_fops);
3897 trace_create_file("trace_pipe_raw", 0444, d_cpu,
3898 (void *) cpu, &tracing_buffers_fops);
3900 trace_create_file("stats", 0444, d_cpu,
3901 (void *) cpu, &tracing_stats_fops);
3904 #ifdef CONFIG_FTRACE_SELFTEST
3905 /* Let selftest have access to static functions in this file */
3906 #include "trace_selftest.c"
3909 struct trace_option_dentry {
3910 struct tracer_opt *opt;
3911 struct tracer_flags *flags;
3912 struct dentry *entry;
3916 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3919 struct trace_option_dentry *topt = filp->private_data;
3922 if (topt->flags->val & topt->opt->bit)
3927 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3931 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3934 struct trace_option_dentry *topt = filp->private_data;
3939 if (cnt >= sizeof(buf))
3942 if (copy_from_user(&buf, ubuf, cnt))
3947 ret = strict_strtoul(buf, 10, &val);
3951 if (val != 0 && val != 1)
3954 if (!!(topt->flags->val & topt->opt->bit) != val) {
3955 mutex_lock(&trace_types_lock);
3956 ret = __set_tracer_option(current_trace, topt->flags,
3958 mutex_unlock(&trace_types_lock);
3969 static const struct file_operations trace_options_fops = {
3970 .open = tracing_open_generic,
3971 .read = trace_options_read,
3972 .write = trace_options_write,
3976 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
3979 long index = (long)filp->private_data;
3982 if (trace_flags & (1 << index))
3987 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3991 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
3994 long index = (long)filp->private_data;
3999 if (cnt >= sizeof(buf))
4002 if (copy_from_user(&buf, ubuf, cnt))
4007 ret = strict_strtoul(buf, 10, &val);
4011 if (val != 0 && val != 1)
4013 set_tracer_flags(1 << index, val);
4020 static const struct file_operations trace_options_core_fops = {
4021 .open = tracing_open_generic,
4022 .read = trace_options_core_read,
4023 .write = trace_options_core_write,
4026 struct dentry *trace_create_file(const char *name,
4028 struct dentry *parent,
4030 const struct file_operations *fops)
4034 ret = debugfs_create_file(name, mode, parent, data, fops);
4036 pr_warning("Could not create debugfs '%s' entry\n", name);
4042 static struct dentry *trace_options_init_dentry(void)
4044 struct dentry *d_tracer;
4045 static struct dentry *t_options;
4050 d_tracer = tracing_init_dentry();
4054 t_options = debugfs_create_dir("options", d_tracer);
4056 pr_warning("Could not create debugfs directory 'options'\n");
4064 create_trace_option_file(struct trace_option_dentry *topt,
4065 struct tracer_flags *flags,
4066 struct tracer_opt *opt)
4068 struct dentry *t_options;
4070 t_options = trace_options_init_dentry();
4074 topt->flags = flags;
4077 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4078 &trace_options_fops);
4082 static struct trace_option_dentry *
4083 create_trace_option_files(struct tracer *tracer)
4085 struct trace_option_dentry *topts;
4086 struct tracer_flags *flags;
4087 struct tracer_opt *opts;
4093 flags = tracer->flags;
4095 if (!flags || !flags->opts)
4100 for (cnt = 0; opts[cnt].name; cnt++)
4103 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4107 for (cnt = 0; opts[cnt].name; cnt++)
4108 create_trace_option_file(&topts[cnt], flags,
4115 destroy_trace_option_files(struct trace_option_dentry *topts)
4122 for (cnt = 0; topts[cnt].opt; cnt++) {
4123 if (topts[cnt].entry)
4124 debugfs_remove(topts[cnt].entry);
4130 static struct dentry *
4131 create_trace_option_core_file(const char *option, long index)
4133 struct dentry *t_options;
4135 t_options = trace_options_init_dentry();
4139 return trace_create_file(option, 0644, t_options, (void *)index,
4140 &trace_options_core_fops);
4143 static __init void create_trace_options_dir(void)
4145 struct dentry *t_options;
4148 t_options = trace_options_init_dentry();
4152 for (i = 0; trace_options[i]; i++)
4153 create_trace_option_core_file(trace_options[i], i);
4156 static __init int tracer_init_debugfs(void)
4158 struct dentry *d_tracer;
4161 d_tracer = tracing_init_dentry();
4163 trace_create_file("tracing_enabled", 0644, d_tracer,
4164 &global_trace, &tracing_ctrl_fops);
4166 trace_create_file("trace_options", 0644, d_tracer,
4167 NULL, &tracing_iter_fops);
4169 trace_create_file("tracing_cpumask", 0644, d_tracer,
4170 NULL, &tracing_cpumask_fops);
4172 trace_create_file("trace", 0644, d_tracer,
4173 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4175 trace_create_file("available_tracers", 0444, d_tracer,
4176 &global_trace, &show_traces_fops);
4178 trace_create_file("current_tracer", 0644, d_tracer,
4179 &global_trace, &set_tracer_fops);
4181 #ifdef CONFIG_TRACER_MAX_TRACE
4182 trace_create_file("tracing_max_latency", 0644, d_tracer,
4183 &tracing_max_latency, &tracing_max_lat_fops);
4185 trace_create_file("tracing_thresh", 0644, d_tracer,
4186 &tracing_thresh, &tracing_max_lat_fops);
4189 trace_create_file("README", 0444, d_tracer,
4190 NULL, &tracing_readme_fops);
4192 trace_create_file("trace_pipe", 0444, d_tracer,
4193 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4195 trace_create_file("buffer_size_kb", 0644, d_tracer,
4196 &global_trace, &tracing_entries_fops);
4198 trace_create_file("trace_marker", 0220, d_tracer,
4199 NULL, &tracing_mark_fops);
4201 trace_create_file("saved_cmdlines", 0444, d_tracer,
4202 NULL, &tracing_saved_cmdlines_fops);
4204 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4207 #ifdef CONFIG_DYNAMIC_FTRACE
4208 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4209 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4211 #ifdef CONFIG_SYSPROF_TRACER
4212 init_tracer_sysprof_debugfs(d_tracer);
4215 create_trace_options_dir();
4217 for_each_tracing_cpu(cpu)
4218 tracing_init_debugfs_percpu(cpu);
4223 static int trace_panic_handler(struct notifier_block *this,
4224 unsigned long event, void *unused)
4226 if (ftrace_dump_on_oops)
4231 static struct notifier_block trace_panic_notifier = {
4232 .notifier_call = trace_panic_handler,
4234 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4237 static int trace_die_handler(struct notifier_block *self,
4243 if (ftrace_dump_on_oops)
4252 static struct notifier_block trace_die_notifier = {
4253 .notifier_call = trace_die_handler,
4258 * printk is set to max of 1024, we really don't need it that big.
4259 * Nothing should be printing 1000 characters anyway.
4261 #define TRACE_MAX_PRINT 1000
4264 * Define here KERN_TRACE so that we have one place to modify
4265 * it if we decide to change what log level the ftrace dump
4268 #define KERN_TRACE KERN_EMERG
4271 trace_printk_seq(struct trace_seq *s)
4273 /* Probably should print a warning here. */
4277 /* should be zero ended, but we are paranoid. */
4278 s->buffer[s->len] = 0;
4280 printk(KERN_TRACE "%s", s->buffer);
4285 static void __ftrace_dump(bool disable_tracing)
4287 static arch_spinlock_t ftrace_dump_lock =
4288 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4289 /* use static because iter can be a bit big for the stack */
4290 static struct trace_iterator iter;
4291 unsigned int old_userobj;
4292 static int dump_ran;
4293 unsigned long flags;
4297 local_irq_save(flags);
4298 arch_spin_lock(&ftrace_dump_lock);
4306 if (disable_tracing)
4309 for_each_tracing_cpu(cpu) {
4310 atomic_inc(&global_trace.data[cpu]->disabled);
4313 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4315 /* don't look at user memory in panic mode */
4316 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4318 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4320 /* Simulate the iterator */
4321 iter.tr = &global_trace;
4322 iter.trace = current_trace;
4323 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4326 * We need to stop all tracing on all CPUS to read the
4327 * the next buffer. This is a bit expensive, but is
4328 * not done often. We fill all what we can read,
4329 * and then release the locks again.
4332 while (!trace_empty(&iter)) {
4335 printk(KERN_TRACE "---------------------------------\n");
4339 /* reset all but tr, trace, and overruns */
4340 memset(&iter.seq, 0,
4341 sizeof(struct trace_iterator) -
4342 offsetof(struct trace_iterator, seq));
4343 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4346 if (find_next_entry_inc(&iter) != NULL) {
4349 ret = print_trace_line(&iter);
4350 if (ret != TRACE_TYPE_NO_CONSUME)
4351 trace_consume(&iter);
4354 trace_printk_seq(&iter.seq);
4358 printk(KERN_TRACE " (ftrace buffer empty)\n");
4360 printk(KERN_TRACE "---------------------------------\n");
4362 /* Re-enable tracing if requested */
4363 if (!disable_tracing) {
4364 trace_flags |= old_userobj;
4366 for_each_tracing_cpu(cpu) {
4367 atomic_dec(&global_trace.data[cpu]->disabled);
4373 arch_spin_unlock(&ftrace_dump_lock);
4374 local_irq_restore(flags);
4377 /* By default: disable tracing after the dump */
4378 void ftrace_dump(void)
4380 __ftrace_dump(true);
4383 __init static int tracer_alloc_buffers(void)
4389 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4392 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4393 goto out_free_buffer_mask;
4395 if (!zalloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
4396 goto out_free_tracing_cpumask;
4398 /* To save memory, keep the ring buffer size to its minimum */
4399 if (ring_buffer_expanded)
4400 ring_buf_size = trace_buf_size;
4404 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4405 cpumask_copy(tracing_cpumask, cpu_all_mask);
4407 /* TODO: make the number of buffers hot pluggable with CPUS */
4408 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4409 TRACE_BUFFER_FLAGS);
4410 if (!global_trace.buffer) {
4411 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4413 goto out_free_cpumask;
4415 global_trace.entries = ring_buffer_size(global_trace.buffer);
4418 #ifdef CONFIG_TRACER_MAX_TRACE
4419 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4420 TRACE_BUFFER_FLAGS);
4421 if (!max_tr.buffer) {
4422 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4424 ring_buffer_free(global_trace.buffer);
4425 goto out_free_cpumask;
4427 max_tr.entries = ring_buffer_size(max_tr.buffer);
4428 WARN_ON(max_tr.entries != global_trace.entries);
4431 /* Allocate the first page for all buffers */
4432 for_each_tracing_cpu(i) {
4433 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4434 max_tr.data[i] = &per_cpu(max_tr_data, i);
4437 trace_init_cmdlines();
4439 register_tracer(&nop_trace);
4440 current_trace = &nop_trace;
4441 #ifdef CONFIG_BOOT_TRACER
4442 register_tracer(&boot_tracer);
4444 /* All seems OK, enable tracing */
4445 tracing_disabled = 0;
4447 atomic_notifier_chain_register(&panic_notifier_list,
4448 &trace_panic_notifier);
4450 register_die_notifier(&trace_die_notifier);
4455 free_cpumask_var(tracing_reader_cpumask);
4456 out_free_tracing_cpumask:
4457 free_cpumask_var(tracing_cpumask);
4458 out_free_buffer_mask:
4459 free_cpumask_var(tracing_buffer_mask);
4464 __init static int clear_boot_tracer(void)
4467 * The default tracer at boot buffer is an init section.
4468 * This function is called in lateinit. If we did not
4469 * find the boot tracer, then clear it out, to prevent
4470 * later registration from accessing the buffer that is
4471 * about to be freed.
4473 if (!default_bootup_tracer)
4476 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4477 default_bootup_tracer);
4478 default_bootup_tracer = NULL;
4483 early_initcall(tracer_alloc_buffers);
4484 fs_initcall(tracer_init_debugfs);
4485 late_initcall(clear_boot_tracer);