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[] = {
335 { trace_clock_local, "local" },
336 { trace_clock_global, "global" },
342 * trace_parser_get_init - gets the buffer for trace parser
344 int trace_parser_get_init(struct trace_parser *parser, int size)
346 memset(parser, 0, sizeof(*parser));
348 parser->buffer = kmalloc(size, GFP_KERNEL);
357 * trace_parser_put - frees the buffer for trace parser
359 void trace_parser_put(struct trace_parser *parser)
361 kfree(parser->buffer);
365 * trace_get_user - reads the user input string separated by space
366 * (matched by isspace(ch))
368 * For each string found the 'struct trace_parser' is updated,
369 * and the function returns.
371 * Returns number of bytes read.
373 * See kernel/trace/trace.h for 'struct trace_parser' details.
375 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
376 size_t cnt, loff_t *ppos)
383 trace_parser_clear(parser);
385 ret = get_user(ch, ubuf++);
393 * The parser is not finished with the last write,
394 * continue reading the user input without skipping spaces.
397 /* skip white space */
398 while (cnt && isspace(ch)) {
399 ret = get_user(ch, ubuf++);
406 /* only spaces were written */
416 /* read the non-space input */
417 while (cnt && !isspace(ch)) {
418 if (parser->idx < parser->size - 1)
419 parser->buffer[parser->idx++] = ch;
424 ret = get_user(ch, ubuf++);
431 /* We either got finished input or we have to wait for another call. */
433 parser->buffer[parser->idx] = 0;
434 parser->cont = false;
437 parser->buffer[parser->idx++] = ch;
447 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
455 if (s->len <= s->readpos)
458 len = s->len - s->readpos;
461 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
471 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
476 if (s->len <= s->readpos)
479 len = s->len - s->readpos;
482 ret = memcpy(buf, s->buffer + s->readpos, cnt);
491 * ftrace_max_lock is used to protect the swapping of buffers
492 * when taking a max snapshot. The buffers themselves are
493 * protected by per_cpu spinlocks. But the action of the swap
494 * needs its own lock.
496 * This is defined as a raw_spinlock_t in order to help
497 * with performance when lockdep debugging is enabled.
499 * It is also used in other places outside the update_max_tr
500 * so it needs to be defined outside of the
501 * CONFIG_TRACER_MAX_TRACE.
503 static raw_spinlock_t ftrace_max_lock =
504 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
506 #ifdef CONFIG_TRACER_MAX_TRACE
507 unsigned long __read_mostly tracing_max_latency;
508 unsigned long __read_mostly tracing_thresh;
511 * Copy the new maximum trace into the separate maximum-trace
512 * structure. (this way the maximum trace is permanently saved,
513 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
516 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
518 struct trace_array_cpu *data = tr->data[cpu];
519 struct trace_array_cpu *max_data = tr->data[cpu];
522 max_tr.time_start = data->preempt_timestamp;
524 max_data = max_tr.data[cpu];
525 max_data->saved_latency = tracing_max_latency;
526 max_data->critical_start = data->critical_start;
527 max_data->critical_end = data->critical_end;
529 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
530 max_data->pid = tsk->pid;
531 max_data->uid = task_uid(tsk);
532 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
533 max_data->policy = tsk->policy;
534 max_data->rt_priority = tsk->rt_priority;
536 /* record this tasks comm */
537 tracing_record_cmdline(tsk);
541 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
543 * @tsk: the task with the latency
544 * @cpu: The cpu that initiated the trace.
546 * Flip the buffers between the @tr and the max_tr and record information
547 * about which task was the cause of this latency.
550 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
552 struct ring_buffer *buf = tr->buffer;
554 if (trace_stop_count)
557 WARN_ON_ONCE(!irqs_disabled());
558 __raw_spin_lock(&ftrace_max_lock);
560 tr->buffer = max_tr.buffer;
563 __update_max_tr(tr, tsk, cpu);
564 __raw_spin_unlock(&ftrace_max_lock);
568 * update_max_tr_single - only copy one trace over, and reset the rest
570 * @tsk - task with the latency
571 * @cpu - the cpu of the buffer to copy.
573 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
576 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
580 if (trace_stop_count)
583 WARN_ON_ONCE(!irqs_disabled());
584 __raw_spin_lock(&ftrace_max_lock);
586 ftrace_disable_cpu();
588 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
592 * We failed to swap the buffer due to a commit taking
593 * place on this CPU. We fail to record, but we reset
594 * the max trace buffer (no one writes directly to it)
595 * and flag that it failed.
597 trace_array_printk(&max_tr, _THIS_IP_,
598 "Failed to swap buffers due to commit in progress\n");
603 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
605 __update_max_tr(tr, tsk, cpu);
606 __raw_spin_unlock(&ftrace_max_lock);
608 #endif /* CONFIG_TRACER_MAX_TRACE */
611 * register_tracer - register a tracer with the ftrace system.
612 * @type - the plugin for the tracer
614 * Register a new plugin tracer.
616 int register_tracer(struct tracer *type)
617 __releases(kernel_lock)
618 __acquires(kernel_lock)
624 pr_info("Tracer must have a name\n");
628 if (strlen(type->name) > MAX_TRACER_SIZE) {
629 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
634 * When this gets called we hold the BKL which means that
635 * preemption is disabled. Various trace selftests however
636 * need to disable and enable preemption for successful tests.
637 * So we drop the BKL here and grab it after the tests again.
640 mutex_lock(&trace_types_lock);
642 tracing_selftest_running = true;
644 for (t = trace_types; t; t = t->next) {
645 if (strcmp(type->name, t->name) == 0) {
647 pr_info("Tracer %s already registered\n",
655 type->set_flag = &dummy_set_flag;
657 type->flags = &dummy_tracer_flags;
659 if (!type->flags->opts)
660 type->flags->opts = dummy_tracer_opt;
661 if (!type->wait_pipe)
662 type->wait_pipe = default_wait_pipe;
665 #ifdef CONFIG_FTRACE_STARTUP_TEST
666 if (type->selftest && !tracing_selftest_disabled) {
667 struct tracer *saved_tracer = current_trace;
668 struct trace_array *tr = &global_trace;
671 * Run a selftest on this tracer.
672 * Here we reset the trace buffer, and set the current
673 * tracer to be this tracer. The tracer can then run some
674 * internal tracing to verify that everything is in order.
675 * If we fail, we do not register this tracer.
677 tracing_reset_online_cpus(tr);
679 current_trace = type;
680 /* the test is responsible for initializing and enabling */
681 pr_info("Testing tracer %s: ", type->name);
682 ret = type->selftest(type, tr);
683 /* the test is responsible for resetting too */
684 current_trace = saved_tracer;
686 printk(KERN_CONT "FAILED!\n");
689 /* Only reset on passing, to avoid touching corrupted buffers */
690 tracing_reset_online_cpus(tr);
692 printk(KERN_CONT "PASSED\n");
696 type->next = trace_types;
700 tracing_selftest_running = false;
701 mutex_unlock(&trace_types_lock);
703 if (ret || !default_bootup_tracer)
706 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
709 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
710 /* Do we want this tracer to start on bootup? */
711 tracing_set_tracer(type->name);
712 default_bootup_tracer = NULL;
713 /* disable other selftests, since this will break it. */
714 tracing_selftest_disabled = 1;
715 #ifdef CONFIG_FTRACE_STARTUP_TEST
716 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
725 void unregister_tracer(struct tracer *type)
729 mutex_lock(&trace_types_lock);
730 for (t = &trace_types; *t; t = &(*t)->next) {
734 pr_info("Tracer %s not registered\n", type->name);
740 if (type == current_trace && tracer_enabled) {
743 if (current_trace->stop)
744 current_trace->stop(&global_trace);
745 current_trace = &nop_trace;
748 mutex_unlock(&trace_types_lock);
751 static void __tracing_reset(struct trace_array *tr, int cpu)
753 ftrace_disable_cpu();
754 ring_buffer_reset_cpu(tr->buffer, cpu);
758 void tracing_reset(struct trace_array *tr, int cpu)
760 struct ring_buffer *buffer = tr->buffer;
762 ring_buffer_record_disable(buffer);
764 /* Make sure all commits have finished */
766 __tracing_reset(tr, cpu);
768 ring_buffer_record_enable(buffer);
771 void tracing_reset_online_cpus(struct trace_array *tr)
773 struct ring_buffer *buffer = tr->buffer;
776 ring_buffer_record_disable(buffer);
778 /* Make sure all commits have finished */
781 tr->time_start = ftrace_now(tr->cpu);
783 for_each_online_cpu(cpu)
784 __tracing_reset(tr, cpu);
786 ring_buffer_record_enable(buffer);
789 void tracing_reset_current(int cpu)
791 tracing_reset(&global_trace, cpu);
794 void tracing_reset_current_online_cpus(void)
796 tracing_reset_online_cpus(&global_trace);
799 #define SAVED_CMDLINES 128
800 #define NO_CMDLINE_MAP UINT_MAX
801 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
802 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
803 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
804 static int cmdline_idx;
805 static raw_spinlock_t trace_cmdline_lock = __RAW_SPIN_LOCK_UNLOCKED;
807 /* temporary disable recording */
808 static atomic_t trace_record_cmdline_disabled __read_mostly;
810 static void trace_init_cmdlines(void)
812 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
813 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
817 int is_tracing_stopped(void)
819 return trace_stop_count;
823 * ftrace_off_permanent - disable all ftrace code permanently
825 * This should only be called when a serious anomally has
826 * been detected. This will turn off the function tracing,
827 * ring buffers, and other tracing utilites. It takes no
828 * locks and can be called from any context.
830 void ftrace_off_permanent(void)
832 tracing_disabled = 1;
834 tracing_off_permanent();
838 * tracing_start - quick start of the tracer
840 * If tracing is enabled but was stopped by tracing_stop,
841 * this will start the tracer back up.
843 void tracing_start(void)
845 struct ring_buffer *buffer;
848 if (tracing_disabled)
851 spin_lock_irqsave(&tracing_start_lock, flags);
852 if (--trace_stop_count) {
853 if (trace_stop_count < 0) {
854 /* Someone screwed up their debugging */
856 trace_stop_count = 0;
862 buffer = global_trace.buffer;
864 ring_buffer_record_enable(buffer);
866 buffer = max_tr.buffer;
868 ring_buffer_record_enable(buffer);
872 spin_unlock_irqrestore(&tracing_start_lock, flags);
876 * tracing_stop - quick stop of the tracer
878 * Light weight way to stop tracing. Use in conjunction with
881 void tracing_stop(void)
883 struct ring_buffer *buffer;
887 spin_lock_irqsave(&tracing_start_lock, flags);
888 if (trace_stop_count++)
891 buffer = global_trace.buffer;
893 ring_buffer_record_disable(buffer);
895 buffer = max_tr.buffer;
897 ring_buffer_record_disable(buffer);
900 spin_unlock_irqrestore(&tracing_start_lock, flags);
903 void trace_stop_cmdline_recording(void);
905 static void trace_save_cmdline(struct task_struct *tsk)
909 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
913 * It's not the end of the world if we don't get
914 * the lock, but we also don't want to spin
915 * nor do we want to disable interrupts,
916 * so if we miss here, then better luck next time.
918 if (!__raw_spin_trylock(&trace_cmdline_lock))
921 idx = map_pid_to_cmdline[tsk->pid];
922 if (idx == NO_CMDLINE_MAP) {
923 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
926 * Check whether the cmdline buffer at idx has a pid
927 * mapped. We are going to overwrite that entry so we
928 * need to clear the map_pid_to_cmdline. Otherwise we
929 * would read the new comm for the old pid.
931 pid = map_cmdline_to_pid[idx];
932 if (pid != NO_CMDLINE_MAP)
933 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
935 map_cmdline_to_pid[idx] = tsk->pid;
936 map_pid_to_cmdline[tsk->pid] = idx;
941 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
943 __raw_spin_unlock(&trace_cmdline_lock);
946 void trace_find_cmdline(int pid, char comm[])
951 strcpy(comm, "<idle>");
955 if (pid > PID_MAX_DEFAULT) {
956 strcpy(comm, "<...>");
961 __raw_spin_lock(&trace_cmdline_lock);
962 map = map_pid_to_cmdline[pid];
963 if (map != NO_CMDLINE_MAP)
964 strcpy(comm, saved_cmdlines[map]);
966 strcpy(comm, "<...>");
968 __raw_spin_unlock(&trace_cmdline_lock);
972 void tracing_record_cmdline(struct task_struct *tsk)
974 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
978 trace_save_cmdline(tsk);
982 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
985 struct task_struct *tsk = current;
987 entry->preempt_count = pc & 0xff;
988 entry->pid = (tsk) ? tsk->pid : 0;
989 entry->lock_depth = (tsk) ? tsk->lock_depth : 0;
991 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
992 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
994 TRACE_FLAG_IRQS_NOSUPPORT |
996 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
997 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
998 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1000 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1002 struct ring_buffer_event *
1003 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1006 unsigned long flags, int pc)
1008 struct ring_buffer_event *event;
1010 event = ring_buffer_lock_reserve(buffer, len);
1011 if (event != NULL) {
1012 struct trace_entry *ent = ring_buffer_event_data(event);
1014 tracing_generic_entry_update(ent, flags, pc);
1022 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1023 struct ring_buffer_event *event,
1024 unsigned long flags, int pc,
1027 ring_buffer_unlock_commit(buffer, event);
1029 ftrace_trace_stack(buffer, flags, 6, pc);
1030 ftrace_trace_userstack(buffer, flags, pc);
1036 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1037 struct ring_buffer_event *event,
1038 unsigned long flags, int pc)
1040 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1043 struct ring_buffer_event *
1044 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1045 int type, unsigned long len,
1046 unsigned long flags, int pc)
1048 *current_rb = global_trace.buffer;
1049 return trace_buffer_lock_reserve(*current_rb,
1050 type, len, flags, pc);
1052 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1054 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1055 struct ring_buffer_event *event,
1056 unsigned long flags, int pc)
1058 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1060 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1062 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1063 struct ring_buffer_event *event,
1064 unsigned long flags, int pc)
1066 __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1068 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1070 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1071 struct ring_buffer_event *event)
1073 ring_buffer_discard_commit(buffer, event);
1075 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1078 trace_function(struct trace_array *tr,
1079 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1082 struct ftrace_event_call *call = &event_function;
1083 struct ring_buffer *buffer = tr->buffer;
1084 struct ring_buffer_event *event;
1085 struct ftrace_entry *entry;
1087 /* If we are reading the ring buffer, don't trace */
1088 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
1091 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1095 entry = ring_buffer_event_data(event);
1097 entry->parent_ip = parent_ip;
1099 if (!filter_check_discard(call, entry, buffer, event))
1100 ring_buffer_unlock_commit(buffer, event);
1104 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1105 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1108 if (likely(!atomic_read(&data->disabled)))
1109 trace_function(tr, ip, parent_ip, flags, pc);
1112 #ifdef CONFIG_STACKTRACE
1113 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1114 unsigned long flags,
1117 struct ftrace_event_call *call = &event_kernel_stack;
1118 struct ring_buffer_event *event;
1119 struct stack_entry *entry;
1120 struct stack_trace trace;
1122 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1123 sizeof(*entry), flags, pc);
1126 entry = ring_buffer_event_data(event);
1127 memset(&entry->caller, 0, sizeof(entry->caller));
1129 trace.nr_entries = 0;
1130 trace.max_entries = FTRACE_STACK_ENTRIES;
1132 trace.entries = entry->caller;
1134 save_stack_trace(&trace);
1135 if (!filter_check_discard(call, entry, buffer, event))
1136 ring_buffer_unlock_commit(buffer, event);
1139 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1142 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1145 __ftrace_trace_stack(buffer, flags, skip, pc);
1148 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1151 __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1155 * trace_dump_stack - record a stack back trace in the trace buffer
1157 void trace_dump_stack(void)
1159 unsigned long flags;
1161 if (tracing_disabled || tracing_selftest_running)
1164 local_save_flags(flags);
1166 /* skipping 3 traces, seems to get us at the caller of this function */
1167 __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count());
1171 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1173 struct ftrace_event_call *call = &event_user_stack;
1174 struct ring_buffer_event *event;
1175 struct userstack_entry *entry;
1176 struct stack_trace trace;
1178 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1181 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1182 sizeof(*entry), flags, pc);
1185 entry = ring_buffer_event_data(event);
1187 entry->tgid = current->tgid;
1188 memset(&entry->caller, 0, sizeof(entry->caller));
1190 trace.nr_entries = 0;
1191 trace.max_entries = FTRACE_STACK_ENTRIES;
1193 trace.entries = entry->caller;
1195 save_stack_trace_user(&trace);
1196 if (!filter_check_discard(call, entry, buffer, event))
1197 ring_buffer_unlock_commit(buffer, event);
1201 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1203 ftrace_trace_userstack(tr, flags, preempt_count());
1207 #endif /* CONFIG_STACKTRACE */
1210 ftrace_trace_special(void *__tr,
1211 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1214 struct ftrace_event_call *call = &event_special;
1215 struct ring_buffer_event *event;
1216 struct trace_array *tr = __tr;
1217 struct ring_buffer *buffer = tr->buffer;
1218 struct special_entry *entry;
1220 event = trace_buffer_lock_reserve(buffer, TRACE_SPECIAL,
1221 sizeof(*entry), 0, pc);
1224 entry = ring_buffer_event_data(event);
1229 if (!filter_check_discard(call, entry, buffer, event))
1230 trace_buffer_unlock_commit(buffer, event, 0, pc);
1234 __trace_special(void *__tr, void *__data,
1235 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1237 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1241 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1243 struct trace_array *tr = &global_trace;
1244 struct trace_array_cpu *data;
1245 unsigned long flags;
1249 if (tracing_disabled)
1252 pc = preempt_count();
1253 local_irq_save(flags);
1254 cpu = raw_smp_processor_id();
1255 data = tr->data[cpu];
1257 if (likely(atomic_inc_return(&data->disabled) == 1))
1258 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1260 atomic_dec(&data->disabled);
1261 local_irq_restore(flags);
1265 * trace_vbprintk - write binary msg to tracing buffer
1268 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1270 static raw_spinlock_t trace_buf_lock =
1271 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
1272 static u32 trace_buf[TRACE_BUF_SIZE];
1274 struct ftrace_event_call *call = &event_bprint;
1275 struct ring_buffer_event *event;
1276 struct ring_buffer *buffer;
1277 struct trace_array *tr = &global_trace;
1278 struct trace_array_cpu *data;
1279 struct bprint_entry *entry;
1280 unsigned long flags;
1283 int cpu, len = 0, size, pc;
1285 if (unlikely(tracing_selftest_running || tracing_disabled))
1288 /* Don't pollute graph traces with trace_vprintk internals */
1289 pause_graph_tracing();
1291 pc = preempt_count();
1292 resched = ftrace_preempt_disable();
1293 cpu = raw_smp_processor_id();
1294 data = tr->data[cpu];
1296 disable = atomic_inc_return(&data->disabled);
1297 if (unlikely(disable != 1))
1300 /* Lockdep uses trace_printk for lock tracing */
1301 local_irq_save(flags);
1302 __raw_spin_lock(&trace_buf_lock);
1303 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1305 if (len > TRACE_BUF_SIZE || len < 0)
1308 size = sizeof(*entry) + sizeof(u32) * len;
1309 buffer = tr->buffer;
1310 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1314 entry = ring_buffer_event_data(event);
1318 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1319 if (!filter_check_discard(call, entry, buffer, event))
1320 ring_buffer_unlock_commit(buffer, event);
1323 __raw_spin_unlock(&trace_buf_lock);
1324 local_irq_restore(flags);
1327 atomic_dec_return(&data->disabled);
1328 ftrace_preempt_enable(resched);
1329 unpause_graph_tracing();
1333 EXPORT_SYMBOL_GPL(trace_vbprintk);
1335 int trace_array_printk(struct trace_array *tr,
1336 unsigned long ip, const char *fmt, ...)
1341 if (!(trace_flags & TRACE_ITER_PRINTK))
1345 ret = trace_array_vprintk(tr, ip, fmt, ap);
1350 int trace_array_vprintk(struct trace_array *tr,
1351 unsigned long ip, const char *fmt, va_list args)
1353 static raw_spinlock_t trace_buf_lock = __RAW_SPIN_LOCK_UNLOCKED;
1354 static char trace_buf[TRACE_BUF_SIZE];
1356 struct ftrace_event_call *call = &event_print;
1357 struct ring_buffer_event *event;
1358 struct ring_buffer *buffer;
1359 struct trace_array_cpu *data;
1360 int cpu, len = 0, size, pc;
1361 struct print_entry *entry;
1362 unsigned long irq_flags;
1365 if (tracing_disabled || tracing_selftest_running)
1368 pc = preempt_count();
1369 preempt_disable_notrace();
1370 cpu = raw_smp_processor_id();
1371 data = tr->data[cpu];
1373 disable = atomic_inc_return(&data->disabled);
1374 if (unlikely(disable != 1))
1377 pause_graph_tracing();
1378 raw_local_irq_save(irq_flags);
1379 __raw_spin_lock(&trace_buf_lock);
1380 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1382 size = sizeof(*entry) + len + 1;
1383 buffer = tr->buffer;
1384 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1388 entry = ring_buffer_event_data(event);
1391 memcpy(&entry->buf, trace_buf, len);
1392 entry->buf[len] = '\0';
1393 if (!filter_check_discard(call, entry, buffer, event))
1394 ring_buffer_unlock_commit(buffer, event);
1397 __raw_spin_unlock(&trace_buf_lock);
1398 raw_local_irq_restore(irq_flags);
1399 unpause_graph_tracing();
1401 atomic_dec_return(&data->disabled);
1402 preempt_enable_notrace();
1407 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1409 return trace_array_vprintk(&global_trace, ip, fmt, args);
1411 EXPORT_SYMBOL_GPL(trace_vprintk);
1413 enum trace_file_type {
1414 TRACE_FILE_LAT_FMT = 1,
1415 TRACE_FILE_ANNOTATE = 2,
1418 static void trace_iterator_increment(struct trace_iterator *iter)
1420 /* Don't allow ftrace to trace into the ring buffers */
1421 ftrace_disable_cpu();
1424 if (iter->buffer_iter[iter->cpu])
1425 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1427 ftrace_enable_cpu();
1430 static struct trace_entry *
1431 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1433 struct ring_buffer_event *event;
1434 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1436 /* Don't allow ftrace to trace into the ring buffers */
1437 ftrace_disable_cpu();
1440 event = ring_buffer_iter_peek(buf_iter, ts);
1442 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1444 ftrace_enable_cpu();
1446 return event ? ring_buffer_event_data(event) : NULL;
1449 static struct trace_entry *
1450 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1452 struct ring_buffer *buffer = iter->tr->buffer;
1453 struct trace_entry *ent, *next = NULL;
1454 int cpu_file = iter->cpu_file;
1455 u64 next_ts = 0, ts;
1460 * If we are in a per_cpu trace file, don't bother by iterating over
1461 * all cpu and peek directly.
1463 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1464 if (ring_buffer_empty_cpu(buffer, cpu_file))
1466 ent = peek_next_entry(iter, cpu_file, ent_ts);
1468 *ent_cpu = cpu_file;
1473 for_each_tracing_cpu(cpu) {
1475 if (ring_buffer_empty_cpu(buffer, cpu))
1478 ent = peek_next_entry(iter, cpu, &ts);
1481 * Pick the entry with the smallest timestamp:
1483 if (ent && (!next || ts < next_ts)) {
1491 *ent_cpu = next_cpu;
1499 /* Find the next real entry, without updating the iterator itself */
1500 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1501 int *ent_cpu, u64 *ent_ts)
1503 return __find_next_entry(iter, ent_cpu, ent_ts);
1506 /* Find the next real entry, and increment the iterator to the next entry */
1507 static void *find_next_entry_inc(struct trace_iterator *iter)
1509 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1512 trace_iterator_increment(iter);
1514 return iter->ent ? iter : NULL;
1517 static void trace_consume(struct trace_iterator *iter)
1519 /* Don't allow ftrace to trace into the ring buffers */
1520 ftrace_disable_cpu();
1521 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1522 ftrace_enable_cpu();
1525 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1527 struct trace_iterator *iter = m->private;
1531 WARN_ON_ONCE(iter->leftover);
1535 /* can't go backwards */
1540 ent = find_next_entry_inc(iter);
1544 while (ent && iter->idx < i)
1545 ent = find_next_entry_inc(iter);
1552 static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1554 struct trace_array *tr = iter->tr;
1555 struct ring_buffer_event *event;
1556 struct ring_buffer_iter *buf_iter;
1557 unsigned long entries = 0;
1560 tr->data[cpu]->skipped_entries = 0;
1562 if (!iter->buffer_iter[cpu])
1565 buf_iter = iter->buffer_iter[cpu];
1566 ring_buffer_iter_reset(buf_iter);
1569 * We could have the case with the max latency tracers
1570 * that a reset never took place on a cpu. This is evident
1571 * by the timestamp being before the start of the buffer.
1573 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1574 if (ts >= iter->tr->time_start)
1577 ring_buffer_read(buf_iter, NULL);
1580 tr->data[cpu]->skipped_entries = entries;
1584 * No necessary locking here. The worst thing which can
1585 * happen is loosing events consumed at the same time
1586 * by a trace_pipe reader.
1587 * Other than that, we don't risk to crash the ring buffer
1588 * because it serializes the readers.
1590 * The current tracer is copied to avoid a global locking
1593 static void *s_start(struct seq_file *m, loff_t *pos)
1595 struct trace_iterator *iter = m->private;
1596 static struct tracer *old_tracer;
1597 int cpu_file = iter->cpu_file;
1602 /* copy the tracer to avoid using a global lock all around */
1603 mutex_lock(&trace_types_lock);
1604 if (unlikely(old_tracer != current_trace && current_trace)) {
1605 old_tracer = current_trace;
1606 *iter->trace = *current_trace;
1608 mutex_unlock(&trace_types_lock);
1610 atomic_inc(&trace_record_cmdline_disabled);
1612 if (*pos != iter->pos) {
1617 ftrace_disable_cpu();
1619 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1620 for_each_tracing_cpu(cpu)
1621 tracing_iter_reset(iter, cpu);
1623 tracing_iter_reset(iter, cpu_file);
1625 ftrace_enable_cpu();
1627 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1632 * If we overflowed the seq_file before, then we want
1633 * to just reuse the trace_seq buffer again.
1639 p = s_next(m, p, &l);
1643 trace_event_read_lock();
1647 static void s_stop(struct seq_file *m, void *p)
1649 atomic_dec(&trace_record_cmdline_disabled);
1650 trace_event_read_unlock();
1653 static void print_lat_help_header(struct seq_file *m)
1655 seq_puts(m, "# _------=> CPU# \n");
1656 seq_puts(m, "# / _-----=> irqs-off \n");
1657 seq_puts(m, "# | / _----=> need-resched \n");
1658 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1659 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1660 seq_puts(m, "# |||| /_--=> lock-depth \n");
1661 seq_puts(m, "# |||||/ delay \n");
1662 seq_puts(m, "# cmd pid |||||| time | caller \n");
1663 seq_puts(m, "# \\ / |||||| \\ | / \n");
1666 static void print_func_help_header(struct seq_file *m)
1668 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1669 seq_puts(m, "# | | | | |\n");
1674 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1676 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1677 struct trace_array *tr = iter->tr;
1678 struct trace_array_cpu *data = tr->data[tr->cpu];
1679 struct tracer *type = current_trace;
1680 unsigned long entries = 0;
1681 unsigned long total = 0;
1682 unsigned long count;
1683 const char *name = "preemption";
1690 for_each_tracing_cpu(cpu) {
1691 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1693 * If this buffer has skipped entries, then we hold all
1694 * entries for the trace and we need to ignore the
1695 * ones before the time stamp.
1697 if (tr->data[cpu]->skipped_entries) {
1698 count -= tr->data[cpu]->skipped_entries;
1699 /* total is the same as the entries */
1703 ring_buffer_overrun_cpu(tr->buffer, cpu);
1707 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1709 seq_puts(m, "# -----------------------------------"
1710 "---------------------------------\n");
1711 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1712 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1713 nsecs_to_usecs(data->saved_latency),
1717 #if defined(CONFIG_PREEMPT_NONE)
1719 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1721 #elif defined(CONFIG_PREEMPT)
1726 /* These are reserved for later use */
1729 seq_printf(m, " #P:%d)\n", num_online_cpus());
1733 seq_puts(m, "# -----------------\n");
1734 seq_printf(m, "# | task: %.16s-%d "
1735 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1736 data->comm, data->pid, data->uid, data->nice,
1737 data->policy, data->rt_priority);
1738 seq_puts(m, "# -----------------\n");
1740 if (data->critical_start) {
1741 seq_puts(m, "# => started at: ");
1742 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1743 trace_print_seq(m, &iter->seq);
1744 seq_puts(m, "\n# => ended at: ");
1745 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1746 trace_print_seq(m, &iter->seq);
1747 seq_puts(m, "\n#\n");
1753 static void test_cpu_buff_start(struct trace_iterator *iter)
1755 struct trace_seq *s = &iter->seq;
1757 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1760 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1763 if (cpumask_test_cpu(iter->cpu, iter->started))
1766 if (iter->tr->data[iter->cpu]->skipped_entries)
1769 cpumask_set_cpu(iter->cpu, iter->started);
1771 /* Don't print started cpu buffer for the first entry of the trace */
1773 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1777 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1779 struct trace_seq *s = &iter->seq;
1780 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1781 struct trace_entry *entry;
1782 struct trace_event *event;
1786 test_cpu_buff_start(iter);
1788 event = ftrace_find_event(entry->type);
1790 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1791 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1792 if (!trace_print_lat_context(iter))
1795 if (!trace_print_context(iter))
1801 return event->trace(iter, sym_flags);
1803 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1806 return TRACE_TYPE_HANDLED;
1808 return TRACE_TYPE_PARTIAL_LINE;
1811 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1813 struct trace_seq *s = &iter->seq;
1814 struct trace_entry *entry;
1815 struct trace_event *event;
1819 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1820 if (!trace_seq_printf(s, "%d %d %llu ",
1821 entry->pid, iter->cpu, iter->ts))
1825 event = ftrace_find_event(entry->type);
1827 return event->raw(iter, 0);
1829 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1832 return TRACE_TYPE_HANDLED;
1834 return TRACE_TYPE_PARTIAL_LINE;
1837 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1839 struct trace_seq *s = &iter->seq;
1840 unsigned char newline = '\n';
1841 struct trace_entry *entry;
1842 struct trace_event *event;
1846 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1847 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1848 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1849 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1852 event = ftrace_find_event(entry->type);
1854 enum print_line_t ret = event->hex(iter, 0);
1855 if (ret != TRACE_TYPE_HANDLED)
1859 SEQ_PUT_FIELD_RET(s, newline);
1861 return TRACE_TYPE_HANDLED;
1864 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1866 struct trace_seq *s = &iter->seq;
1867 struct trace_entry *entry;
1868 struct trace_event *event;
1872 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1873 SEQ_PUT_FIELD_RET(s, entry->pid);
1874 SEQ_PUT_FIELD_RET(s, iter->cpu);
1875 SEQ_PUT_FIELD_RET(s, iter->ts);
1878 event = ftrace_find_event(entry->type);
1879 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1882 static int trace_empty(struct trace_iterator *iter)
1886 /* If we are looking at one CPU buffer, only check that one */
1887 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1888 cpu = iter->cpu_file;
1889 if (iter->buffer_iter[cpu]) {
1890 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1893 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1899 for_each_tracing_cpu(cpu) {
1900 if (iter->buffer_iter[cpu]) {
1901 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1904 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1912 /* Called with trace_event_read_lock() held. */
1913 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1915 enum print_line_t ret;
1917 if (iter->trace && iter->trace->print_line) {
1918 ret = iter->trace->print_line(iter);
1919 if (ret != TRACE_TYPE_UNHANDLED)
1923 if (iter->ent->type == TRACE_BPRINT &&
1924 trace_flags & TRACE_ITER_PRINTK &&
1925 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1926 return trace_print_bprintk_msg_only(iter);
1928 if (iter->ent->type == TRACE_PRINT &&
1929 trace_flags & TRACE_ITER_PRINTK &&
1930 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1931 return trace_print_printk_msg_only(iter);
1933 if (trace_flags & TRACE_ITER_BIN)
1934 return print_bin_fmt(iter);
1936 if (trace_flags & TRACE_ITER_HEX)
1937 return print_hex_fmt(iter);
1939 if (trace_flags & TRACE_ITER_RAW)
1940 return print_raw_fmt(iter);
1942 return print_trace_fmt(iter);
1945 static int s_show(struct seq_file *m, void *v)
1947 struct trace_iterator *iter = v;
1950 if (iter->ent == NULL) {
1952 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1955 if (iter->trace && iter->trace->print_header)
1956 iter->trace->print_header(m);
1957 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1958 /* print nothing if the buffers are empty */
1959 if (trace_empty(iter))
1961 print_trace_header(m, iter);
1962 if (!(trace_flags & TRACE_ITER_VERBOSE))
1963 print_lat_help_header(m);
1965 if (!(trace_flags & TRACE_ITER_VERBOSE))
1966 print_func_help_header(m);
1968 } else if (iter->leftover) {
1970 * If we filled the seq_file buffer earlier, we
1971 * want to just show it now.
1973 ret = trace_print_seq(m, &iter->seq);
1975 /* ret should this time be zero, but you never know */
1976 iter->leftover = ret;
1979 print_trace_line(iter);
1980 ret = trace_print_seq(m, &iter->seq);
1982 * If we overflow the seq_file buffer, then it will
1983 * ask us for this data again at start up.
1985 * ret is 0 if seq_file write succeeded.
1988 iter->leftover = ret;
1994 static const struct seq_operations tracer_seq_ops = {
2001 static struct trace_iterator *
2002 __tracing_open(struct inode *inode, struct file *file)
2004 long cpu_file = (long) inode->i_private;
2005 void *fail_ret = ERR_PTR(-ENOMEM);
2006 struct trace_iterator *iter;
2010 if (tracing_disabled)
2011 return ERR_PTR(-ENODEV);
2013 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2015 return ERR_PTR(-ENOMEM);
2018 * We make a copy of the current tracer to avoid concurrent
2019 * changes on it while we are reading.
2021 mutex_lock(&trace_types_lock);
2022 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2027 *iter->trace = *current_trace;
2029 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2032 if (current_trace && current_trace->print_max)
2035 iter->tr = &global_trace;
2037 mutex_init(&iter->mutex);
2038 iter->cpu_file = cpu_file;
2040 /* Notify the tracer early; before we stop tracing. */
2041 if (iter->trace && iter->trace->open)
2042 iter->trace->open(iter);
2044 /* Annotate start of buffers if we had overruns */
2045 if (ring_buffer_overruns(iter->tr->buffer))
2046 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2048 /* stop the trace while dumping */
2051 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2052 for_each_tracing_cpu(cpu) {
2054 iter->buffer_iter[cpu] =
2055 ring_buffer_read_start(iter->tr->buffer, cpu);
2056 tracing_iter_reset(iter, cpu);
2059 cpu = iter->cpu_file;
2060 iter->buffer_iter[cpu] =
2061 ring_buffer_read_start(iter->tr->buffer, cpu);
2062 tracing_iter_reset(iter, cpu);
2065 ret = seq_open(file, &tracer_seq_ops);
2067 fail_ret = ERR_PTR(ret);
2071 m = file->private_data;
2074 mutex_unlock(&trace_types_lock);
2079 for_each_tracing_cpu(cpu) {
2080 if (iter->buffer_iter[cpu])
2081 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2083 free_cpumask_var(iter->started);
2086 mutex_unlock(&trace_types_lock);
2093 int tracing_open_generic(struct inode *inode, struct file *filp)
2095 if (tracing_disabled)
2098 filp->private_data = inode->i_private;
2102 static int tracing_release(struct inode *inode, struct file *file)
2104 struct seq_file *m = (struct seq_file *)file->private_data;
2105 struct trace_iterator *iter;
2108 if (!(file->f_mode & FMODE_READ))
2113 mutex_lock(&trace_types_lock);
2114 for_each_tracing_cpu(cpu) {
2115 if (iter->buffer_iter[cpu])
2116 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2119 if (iter->trace && iter->trace->close)
2120 iter->trace->close(iter);
2122 /* reenable tracing if it was previously enabled */
2124 mutex_unlock(&trace_types_lock);
2126 seq_release(inode, file);
2127 mutex_destroy(&iter->mutex);
2128 free_cpumask_var(iter->started);
2134 static int tracing_open(struct inode *inode, struct file *file)
2136 struct trace_iterator *iter;
2139 /* If this file was open for write, then erase contents */
2140 if ((file->f_mode & FMODE_WRITE) &&
2141 (file->f_flags & O_TRUNC)) {
2142 long cpu = (long) inode->i_private;
2144 if (cpu == TRACE_PIPE_ALL_CPU)
2145 tracing_reset_online_cpus(&global_trace);
2147 tracing_reset(&global_trace, cpu);
2150 if (file->f_mode & FMODE_READ) {
2151 iter = __tracing_open(inode, file);
2153 ret = PTR_ERR(iter);
2154 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2155 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2161 t_next(struct seq_file *m, void *v, loff_t *pos)
2163 struct tracer *t = v;
2173 static void *t_start(struct seq_file *m, loff_t *pos)
2178 mutex_lock(&trace_types_lock);
2179 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2185 static void t_stop(struct seq_file *m, void *p)
2187 mutex_unlock(&trace_types_lock);
2190 static int t_show(struct seq_file *m, void *v)
2192 struct tracer *t = v;
2197 seq_printf(m, "%s", t->name);
2206 static const struct seq_operations show_traces_seq_ops = {
2213 static int show_traces_open(struct inode *inode, struct file *file)
2215 if (tracing_disabled)
2218 return seq_open(file, &show_traces_seq_ops);
2222 tracing_write_stub(struct file *filp, const char __user *ubuf,
2223 size_t count, loff_t *ppos)
2228 static const struct file_operations tracing_fops = {
2229 .open = tracing_open,
2231 .write = tracing_write_stub,
2232 .llseek = seq_lseek,
2233 .release = tracing_release,
2236 static const struct file_operations show_traces_fops = {
2237 .open = show_traces_open,
2239 .release = seq_release,
2243 * Only trace on a CPU if the bitmask is set:
2245 static cpumask_var_t tracing_cpumask;
2248 * The tracer itself will not take this lock, but still we want
2249 * to provide a consistent cpumask to user-space:
2251 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2254 * Temporary storage for the character representation of the
2255 * CPU bitmask (and one more byte for the newline):
2257 static char mask_str[NR_CPUS + 1];
2260 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2261 size_t count, loff_t *ppos)
2265 mutex_lock(&tracing_cpumask_update_lock);
2267 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2268 if (count - len < 2) {
2272 len += sprintf(mask_str + len, "\n");
2273 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2276 mutex_unlock(&tracing_cpumask_update_lock);
2282 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2283 size_t count, loff_t *ppos)
2286 cpumask_var_t tracing_cpumask_new;
2288 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2291 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2295 mutex_lock(&tracing_cpumask_update_lock);
2297 local_irq_disable();
2298 __raw_spin_lock(&ftrace_max_lock);
2299 for_each_tracing_cpu(cpu) {
2301 * Increase/decrease the disabled counter if we are
2302 * about to flip a bit in the cpumask:
2304 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2305 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2306 atomic_inc(&global_trace.data[cpu]->disabled);
2308 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2309 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2310 atomic_dec(&global_trace.data[cpu]->disabled);
2313 __raw_spin_unlock(&ftrace_max_lock);
2316 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2318 mutex_unlock(&tracing_cpumask_update_lock);
2319 free_cpumask_var(tracing_cpumask_new);
2324 free_cpumask_var(tracing_cpumask_new);
2329 static const struct file_operations tracing_cpumask_fops = {
2330 .open = tracing_open_generic,
2331 .read = tracing_cpumask_read,
2332 .write = tracing_cpumask_write,
2336 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2337 size_t cnt, loff_t *ppos)
2339 struct tracer_opt *trace_opts;
2347 /* calculate max size */
2348 for (i = 0; trace_options[i]; i++) {
2349 len += strlen(trace_options[i]);
2350 len += 3; /* "no" and newline */
2353 mutex_lock(&trace_types_lock);
2354 tracer_flags = current_trace->flags->val;
2355 trace_opts = current_trace->flags->opts;
2358 * Increase the size with names of options specific
2359 * of the current tracer.
2361 for (i = 0; trace_opts[i].name; i++) {
2362 len += strlen(trace_opts[i].name);
2363 len += 3; /* "no" and newline */
2367 buf = kmalloc(len + 1, GFP_KERNEL);
2369 mutex_unlock(&trace_types_lock);
2373 for (i = 0; trace_options[i]; i++) {
2374 if (trace_flags & (1 << i))
2375 r += sprintf(buf + r, "%s\n", trace_options[i]);
2377 r += sprintf(buf + r, "no%s\n", trace_options[i]);
2380 for (i = 0; trace_opts[i].name; i++) {
2381 if (tracer_flags & trace_opts[i].bit)
2382 r += sprintf(buf + r, "%s\n",
2383 trace_opts[i].name);
2385 r += sprintf(buf + r, "no%s\n",
2386 trace_opts[i].name);
2388 mutex_unlock(&trace_types_lock);
2390 WARN_ON(r >= len + 1);
2392 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2398 /* Try to assign a tracer specific option */
2399 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2401 struct tracer_flags *tracer_flags = trace->flags;
2402 struct tracer_opt *opts = NULL;
2406 for (i = 0; tracer_flags->opts[i].name; i++) {
2407 opts = &tracer_flags->opts[i];
2408 len = strlen(opts->name);
2410 if (strncmp(cmp, opts->name, len) == 0) {
2411 ret = trace->set_flag(tracer_flags->val,
2417 if (!tracer_flags->opts[i].name)
2420 /* Refused to handle */
2425 tracer_flags->val &= ~opts->bit;
2427 tracer_flags->val |= opts->bit;
2432 static void set_tracer_flags(unsigned int mask, int enabled)
2434 /* do nothing if flag is already set */
2435 if (!!(trace_flags & mask) == !!enabled)
2439 trace_flags |= mask;
2441 trace_flags &= ~mask;
2445 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2446 size_t cnt, loff_t *ppos)
2454 if (cnt >= sizeof(buf))
2457 if (copy_from_user(&buf, ubuf, cnt))
2462 if (strncmp(buf, "no", 2) == 0) {
2467 for (i = 0; trace_options[i]; i++) {
2468 int len = strlen(trace_options[i]);
2470 if (strncmp(cmp, trace_options[i], len) == 0) {
2471 set_tracer_flags(1 << i, !neg);
2476 /* If no option could be set, test the specific tracer options */
2477 if (!trace_options[i]) {
2478 mutex_lock(&trace_types_lock);
2479 ret = set_tracer_option(current_trace, cmp, neg);
2480 mutex_unlock(&trace_types_lock);
2490 static const struct file_operations tracing_iter_fops = {
2491 .open = tracing_open_generic,
2492 .read = tracing_trace_options_read,
2493 .write = tracing_trace_options_write,
2496 static const char readme_msg[] =
2497 "tracing mini-HOWTO:\n\n"
2498 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2499 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2500 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2501 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2503 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2504 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2506 "# cat /sys/kernel/debug/tracing/trace_options\n"
2507 "noprint-parent nosym-offset nosym-addr noverbose\n"
2508 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2509 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2510 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2511 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2515 tracing_readme_read(struct file *filp, char __user *ubuf,
2516 size_t cnt, loff_t *ppos)
2518 return simple_read_from_buffer(ubuf, cnt, ppos,
2519 readme_msg, strlen(readme_msg));
2522 static const struct file_operations tracing_readme_fops = {
2523 .open = tracing_open_generic,
2524 .read = tracing_readme_read,
2528 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2529 size_t cnt, loff_t *ppos)
2538 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2542 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2550 for (i = 0; i < SAVED_CMDLINES; i++) {
2553 pid = map_cmdline_to_pid[i];
2554 if (pid == -1 || pid == NO_CMDLINE_MAP)
2557 trace_find_cmdline(pid, buf_comm);
2558 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2563 len = simple_read_from_buffer(ubuf, cnt, ppos,
2572 static const struct file_operations tracing_saved_cmdlines_fops = {
2573 .open = tracing_open_generic,
2574 .read = tracing_saved_cmdlines_read,
2578 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2579 size_t cnt, loff_t *ppos)
2584 r = sprintf(buf, "%u\n", tracer_enabled);
2585 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2589 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2590 size_t cnt, loff_t *ppos)
2592 struct trace_array *tr = filp->private_data;
2597 if (cnt >= sizeof(buf))
2600 if (copy_from_user(&buf, ubuf, cnt))
2605 ret = strict_strtoul(buf, 10, &val);
2611 mutex_lock(&trace_types_lock);
2612 if (tracer_enabled ^ val) {
2615 if (current_trace->start)
2616 current_trace->start(tr);
2621 if (current_trace->stop)
2622 current_trace->stop(tr);
2625 mutex_unlock(&trace_types_lock);
2633 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2634 size_t cnt, loff_t *ppos)
2636 char buf[MAX_TRACER_SIZE+2];
2639 mutex_lock(&trace_types_lock);
2641 r = sprintf(buf, "%s\n", current_trace->name);
2643 r = sprintf(buf, "\n");
2644 mutex_unlock(&trace_types_lock);
2646 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2649 int tracer_init(struct tracer *t, struct trace_array *tr)
2651 tracing_reset_online_cpus(tr);
2655 static int tracing_resize_ring_buffer(unsigned long size)
2660 * If kernel or user changes the size of the ring buffer
2661 * we use the size that was given, and we can forget about
2662 * expanding it later.
2664 ring_buffer_expanded = 1;
2666 ret = ring_buffer_resize(global_trace.buffer, size);
2670 ret = ring_buffer_resize(max_tr.buffer, size);
2674 r = ring_buffer_resize(global_trace.buffer,
2675 global_trace.entries);
2678 * AARGH! We are left with different
2679 * size max buffer!!!!
2680 * The max buffer is our "snapshot" buffer.
2681 * When a tracer needs a snapshot (one of the
2682 * latency tracers), it swaps the max buffer
2683 * with the saved snap shot. We succeeded to
2684 * update the size of the main buffer, but failed to
2685 * update the size of the max buffer. But when we tried
2686 * to reset the main buffer to the original size, we
2687 * failed there too. This is very unlikely to
2688 * happen, but if it does, warn and kill all
2692 tracing_disabled = 1;
2697 global_trace.entries = size;
2703 * tracing_update_buffers - used by tracing facility to expand ring buffers
2705 * To save on memory when the tracing is never used on a system with it
2706 * configured in. The ring buffers are set to a minimum size. But once
2707 * a user starts to use the tracing facility, then they need to grow
2708 * to their default size.
2710 * This function is to be called when a tracer is about to be used.
2712 int tracing_update_buffers(void)
2716 mutex_lock(&trace_types_lock);
2717 if (!ring_buffer_expanded)
2718 ret = tracing_resize_ring_buffer(trace_buf_size);
2719 mutex_unlock(&trace_types_lock);
2724 struct trace_option_dentry;
2726 static struct trace_option_dentry *
2727 create_trace_option_files(struct tracer *tracer);
2730 destroy_trace_option_files(struct trace_option_dentry *topts);
2732 static int tracing_set_tracer(const char *buf)
2734 static struct trace_option_dentry *topts;
2735 struct trace_array *tr = &global_trace;
2739 mutex_lock(&trace_types_lock);
2741 if (!ring_buffer_expanded) {
2742 ret = tracing_resize_ring_buffer(trace_buf_size);
2748 for (t = trace_types; t; t = t->next) {
2749 if (strcmp(t->name, buf) == 0)
2756 if (t == current_trace)
2759 trace_branch_disable();
2760 if (current_trace && current_trace->reset)
2761 current_trace->reset(tr);
2763 destroy_trace_option_files(topts);
2767 topts = create_trace_option_files(current_trace);
2770 ret = tracer_init(t, tr);
2775 trace_branch_enable(tr);
2777 mutex_unlock(&trace_types_lock);
2783 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2784 size_t cnt, loff_t *ppos)
2786 char buf[MAX_TRACER_SIZE+1];
2793 if (cnt > MAX_TRACER_SIZE)
2794 cnt = MAX_TRACER_SIZE;
2796 if (copy_from_user(&buf, ubuf, cnt))
2801 /* strip ending whitespace. */
2802 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2805 err = tracing_set_tracer(buf);
2815 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2816 size_t cnt, loff_t *ppos)
2818 unsigned long *ptr = filp->private_data;
2822 r = snprintf(buf, sizeof(buf), "%ld\n",
2823 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2824 if (r > sizeof(buf))
2826 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2830 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2831 size_t cnt, loff_t *ppos)
2833 unsigned long *ptr = filp->private_data;
2838 if (cnt >= sizeof(buf))
2841 if (copy_from_user(&buf, ubuf, cnt))
2846 ret = strict_strtoul(buf, 10, &val);
2855 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2857 long cpu_file = (long) inode->i_private;
2858 struct trace_iterator *iter;
2861 if (tracing_disabled)
2864 mutex_lock(&trace_types_lock);
2866 /* We only allow one reader per cpu */
2867 if (cpu_file == TRACE_PIPE_ALL_CPU) {
2868 if (!cpumask_empty(tracing_reader_cpumask)) {
2872 cpumask_setall(tracing_reader_cpumask);
2874 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2875 cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2882 /* create a buffer to store the information to pass to userspace */
2883 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2890 * We make a copy of the current tracer to avoid concurrent
2891 * changes on it while we are reading.
2893 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2899 *iter->trace = *current_trace;
2901 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2906 /* trace pipe does not show start of buffer */
2907 cpumask_setall(iter->started);
2909 if (trace_flags & TRACE_ITER_LATENCY_FMT)
2910 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2912 iter->cpu_file = cpu_file;
2913 iter->tr = &global_trace;
2914 mutex_init(&iter->mutex);
2915 filp->private_data = iter;
2917 if (iter->trace->pipe_open)
2918 iter->trace->pipe_open(iter);
2921 mutex_unlock(&trace_types_lock);
2927 mutex_unlock(&trace_types_lock);
2931 static int tracing_release_pipe(struct inode *inode, struct file *file)
2933 struct trace_iterator *iter = file->private_data;
2935 mutex_lock(&trace_types_lock);
2937 if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2938 cpumask_clear(tracing_reader_cpumask);
2940 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2943 if (iter->trace->pipe_close)
2944 iter->trace->pipe_close(iter);
2946 mutex_unlock(&trace_types_lock);
2948 free_cpumask_var(iter->started);
2949 mutex_destroy(&iter->mutex);
2957 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2959 struct trace_iterator *iter = filp->private_data;
2961 if (trace_flags & TRACE_ITER_BLOCK) {
2963 * Always select as readable when in blocking mode
2965 return POLLIN | POLLRDNORM;
2967 if (!trace_empty(iter))
2968 return POLLIN | POLLRDNORM;
2969 poll_wait(filp, &trace_wait, poll_table);
2970 if (!trace_empty(iter))
2971 return POLLIN | POLLRDNORM;
2978 void default_wait_pipe(struct trace_iterator *iter)
2982 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2984 if (trace_empty(iter))
2987 finish_wait(&trace_wait, &wait);
2991 * This is a make-shift waitqueue.
2992 * A tracer might use this callback on some rare cases:
2994 * 1) the current tracer might hold the runqueue lock when it wakes up
2995 * a reader, hence a deadlock (sched, function, and function graph tracers)
2996 * 2) the function tracers, trace all functions, we don't want
2997 * the overhead of calling wake_up and friends
2998 * (and tracing them too)
3000 * Anyway, this is really very primitive wakeup.
3002 void poll_wait_pipe(struct trace_iterator *iter)
3004 set_current_state(TASK_INTERRUPTIBLE);
3005 /* sleep for 100 msecs, and try again. */
3006 schedule_timeout(HZ / 10);
3009 /* Must be called with trace_types_lock mutex held. */
3010 static int tracing_wait_pipe(struct file *filp)
3012 struct trace_iterator *iter = filp->private_data;
3014 while (trace_empty(iter)) {
3016 if ((filp->f_flags & O_NONBLOCK)) {
3020 mutex_unlock(&iter->mutex);
3022 iter->trace->wait_pipe(iter);
3024 mutex_lock(&iter->mutex);
3026 if (signal_pending(current))
3030 * We block until we read something and tracing is disabled.
3031 * We still block if tracing is disabled, but we have never
3032 * read anything. This allows a user to cat this file, and
3033 * then enable tracing. But after we have read something,
3034 * we give an EOF when tracing is again disabled.
3036 * iter->pos will be 0 if we haven't read anything.
3038 if (!tracer_enabled && iter->pos)
3049 tracing_read_pipe(struct file *filp, char __user *ubuf,
3050 size_t cnt, loff_t *ppos)
3052 struct trace_iterator *iter = filp->private_data;
3053 static struct tracer *old_tracer;
3056 /* return any leftover data */
3057 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3061 trace_seq_init(&iter->seq);
3063 /* copy the tracer to avoid using a global lock all around */
3064 mutex_lock(&trace_types_lock);
3065 if (unlikely(old_tracer != current_trace && current_trace)) {
3066 old_tracer = current_trace;
3067 *iter->trace = *current_trace;
3069 mutex_unlock(&trace_types_lock);
3072 * Avoid more than one consumer on a single file descriptor
3073 * This is just a matter of traces coherency, the ring buffer itself
3076 mutex_lock(&iter->mutex);
3077 if (iter->trace->read) {
3078 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3084 sret = tracing_wait_pipe(filp);
3088 /* stop when tracing is finished */
3089 if (trace_empty(iter)) {
3094 if (cnt >= PAGE_SIZE)
3095 cnt = PAGE_SIZE - 1;
3097 /* reset all but tr, trace, and overruns */
3098 memset(&iter->seq, 0,
3099 sizeof(struct trace_iterator) -
3100 offsetof(struct trace_iterator, seq));
3103 trace_event_read_lock();
3104 while (find_next_entry_inc(iter) != NULL) {
3105 enum print_line_t ret;
3106 int len = iter->seq.len;
3108 ret = print_trace_line(iter);
3109 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3110 /* don't print partial lines */
3111 iter->seq.len = len;
3114 if (ret != TRACE_TYPE_NO_CONSUME)
3115 trace_consume(iter);
3117 if (iter->seq.len >= cnt)
3120 trace_event_read_unlock();
3122 /* Now copy what we have to the user */
3123 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3124 if (iter->seq.readpos >= iter->seq.len)
3125 trace_seq_init(&iter->seq);
3128 * If there was nothing to send to user, inspite of consuming trace
3129 * entries, go back to wait for more entries.
3135 mutex_unlock(&iter->mutex);
3140 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3141 struct pipe_buffer *buf)
3143 __free_page(buf->page);
3146 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3149 __free_page(spd->pages[idx]);
3152 static struct pipe_buf_operations tracing_pipe_buf_ops = {
3154 .map = generic_pipe_buf_map,
3155 .unmap = generic_pipe_buf_unmap,
3156 .confirm = generic_pipe_buf_confirm,
3157 .release = tracing_pipe_buf_release,
3158 .steal = generic_pipe_buf_steal,
3159 .get = generic_pipe_buf_get,
3163 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3168 /* Seq buffer is page-sized, exactly what we need. */
3170 count = iter->seq.len;
3171 ret = print_trace_line(iter);
3172 count = iter->seq.len - count;
3175 iter->seq.len -= count;
3178 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3179 iter->seq.len -= count;
3183 if (ret != TRACE_TYPE_NO_CONSUME)
3184 trace_consume(iter);
3186 if (!find_next_entry_inc(iter)) {
3196 static ssize_t tracing_splice_read_pipe(struct file *filp,
3198 struct pipe_inode_info *pipe,
3202 struct page *pages[PIPE_BUFFERS];
3203 struct partial_page partial[PIPE_BUFFERS];
3204 struct trace_iterator *iter = filp->private_data;
3205 struct splice_pipe_desc spd = {
3208 .nr_pages = 0, /* This gets updated below. */
3210 .ops = &tracing_pipe_buf_ops,
3211 .spd_release = tracing_spd_release_pipe,
3213 static struct tracer *old_tracer;
3218 /* copy the tracer to avoid using a global lock all around */
3219 mutex_lock(&trace_types_lock);
3220 if (unlikely(old_tracer != current_trace && current_trace)) {
3221 old_tracer = current_trace;
3222 *iter->trace = *current_trace;
3224 mutex_unlock(&trace_types_lock);
3226 mutex_lock(&iter->mutex);
3228 if (iter->trace->splice_read) {
3229 ret = iter->trace->splice_read(iter, filp,
3230 ppos, pipe, len, flags);
3235 ret = tracing_wait_pipe(filp);
3239 if (!iter->ent && !find_next_entry_inc(iter)) {
3244 trace_event_read_lock();
3246 /* Fill as many pages as possible. */
3247 for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3248 pages[i] = alloc_page(GFP_KERNEL);
3252 rem = tracing_fill_pipe_page(rem, iter);
3254 /* Copy the data into the page, so we can start over. */
3255 ret = trace_seq_to_buffer(&iter->seq,
3256 page_address(pages[i]),
3259 __free_page(pages[i]);
3262 partial[i].offset = 0;
3263 partial[i].len = iter->seq.len;
3265 trace_seq_init(&iter->seq);
3268 trace_event_read_unlock();
3269 mutex_unlock(&iter->mutex);
3273 return splice_to_pipe(pipe, &spd);
3276 mutex_unlock(&iter->mutex);
3282 tracing_entries_read(struct file *filp, char __user *ubuf,
3283 size_t cnt, loff_t *ppos)
3285 struct trace_array *tr = filp->private_data;
3289 mutex_lock(&trace_types_lock);
3290 if (!ring_buffer_expanded)
3291 r = sprintf(buf, "%lu (expanded: %lu)\n",
3293 trace_buf_size >> 10);
3295 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3296 mutex_unlock(&trace_types_lock);
3298 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3302 tracing_entries_write(struct file *filp, const char __user *ubuf,
3303 size_t cnt, loff_t *ppos)
3309 if (cnt >= sizeof(buf))
3312 if (copy_from_user(&buf, ubuf, cnt))
3317 ret = strict_strtoul(buf, 10, &val);
3321 /* must have at least 1 entry */
3325 mutex_lock(&trace_types_lock);
3329 /* disable all cpu buffers */
3330 for_each_tracing_cpu(cpu) {
3331 if (global_trace.data[cpu])
3332 atomic_inc(&global_trace.data[cpu]->disabled);
3333 if (max_tr.data[cpu])
3334 atomic_inc(&max_tr.data[cpu]->disabled);
3337 /* value is in KB */
3340 if (val != global_trace.entries) {
3341 ret = tracing_resize_ring_buffer(val);
3350 /* If check pages failed, return ENOMEM */
3351 if (tracing_disabled)
3354 for_each_tracing_cpu(cpu) {
3355 if (global_trace.data[cpu])
3356 atomic_dec(&global_trace.data[cpu]->disabled);
3357 if (max_tr.data[cpu])
3358 atomic_dec(&max_tr.data[cpu]->disabled);
3362 max_tr.entries = global_trace.entries;
3363 mutex_unlock(&trace_types_lock);
3368 static int mark_printk(const char *fmt, ...)
3372 va_start(args, fmt);
3373 ret = trace_vprintk(0, fmt, args);
3379 tracing_mark_write(struct file *filp, const char __user *ubuf,
3380 size_t cnt, loff_t *fpos)
3384 if (tracing_disabled)
3387 if (cnt > TRACE_BUF_SIZE)
3388 cnt = TRACE_BUF_SIZE;
3390 buf = kmalloc(cnt + 2, GFP_KERNEL);
3394 if (copy_from_user(buf, ubuf, cnt)) {
3398 if (buf[cnt-1] != '\n') {
3404 cnt = mark_printk("%s", buf);
3411 static ssize_t tracing_clock_read(struct file *filp, char __user *ubuf,
3412 size_t cnt, loff_t *ppos)
3418 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3419 bufiter += snprintf(buf + bufiter, sizeof(buf) - bufiter,
3420 "%s%s%s%s", i ? " " : "",
3421 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3422 i == trace_clock_id ? "]" : "");
3423 bufiter += snprintf(buf + bufiter, sizeof(buf) - bufiter, "\n");
3425 return simple_read_from_buffer(ubuf, cnt, ppos, buf, bufiter);
3428 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3429 size_t cnt, loff_t *fpos)
3432 const char *clockstr;
3435 if (cnt >= sizeof(buf))
3438 if (copy_from_user(&buf, ubuf, cnt))
3443 clockstr = strstrip(buf);
3445 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3446 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3449 if (i == ARRAY_SIZE(trace_clocks))
3454 mutex_lock(&trace_types_lock);
3456 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3458 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3460 mutex_unlock(&trace_types_lock);
3467 static const struct file_operations tracing_max_lat_fops = {
3468 .open = tracing_open_generic,
3469 .read = tracing_max_lat_read,
3470 .write = tracing_max_lat_write,
3473 static const struct file_operations tracing_ctrl_fops = {
3474 .open = tracing_open_generic,
3475 .read = tracing_ctrl_read,
3476 .write = tracing_ctrl_write,
3479 static const struct file_operations set_tracer_fops = {
3480 .open = tracing_open_generic,
3481 .read = tracing_set_trace_read,
3482 .write = tracing_set_trace_write,
3485 static const struct file_operations tracing_pipe_fops = {
3486 .open = tracing_open_pipe,
3487 .poll = tracing_poll_pipe,
3488 .read = tracing_read_pipe,
3489 .splice_read = tracing_splice_read_pipe,
3490 .release = tracing_release_pipe,
3493 static const struct file_operations tracing_entries_fops = {
3494 .open = tracing_open_generic,
3495 .read = tracing_entries_read,
3496 .write = tracing_entries_write,
3499 static const struct file_operations tracing_mark_fops = {
3500 .open = tracing_open_generic,
3501 .write = tracing_mark_write,
3504 static const struct file_operations trace_clock_fops = {
3505 .open = tracing_open_generic,
3506 .read = tracing_clock_read,
3507 .write = tracing_clock_write,
3510 struct ftrace_buffer_info {
3511 struct trace_array *tr;
3517 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3519 int cpu = (int)(long)inode->i_private;
3520 struct ftrace_buffer_info *info;
3522 if (tracing_disabled)
3525 info = kzalloc(sizeof(*info), GFP_KERNEL);
3529 info->tr = &global_trace;
3532 /* Force reading ring buffer for first read */
3533 info->read = (unsigned int)-1;
3535 filp->private_data = info;
3537 return nonseekable_open(inode, filp);
3541 tracing_buffers_read(struct file *filp, char __user *ubuf,
3542 size_t count, loff_t *ppos)
3544 struct ftrace_buffer_info *info = filp->private_data;
3553 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3557 /* Do we have previous read data to read? */
3558 if (info->read < PAGE_SIZE)
3563 ret = ring_buffer_read_page(info->tr->buffer,
3570 pos = ring_buffer_page_len(info->spare);
3572 if (pos < PAGE_SIZE)
3573 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3576 size = PAGE_SIZE - info->read;
3580 ret = copy_to_user(ubuf, info->spare + info->read, size);
3591 static int tracing_buffers_release(struct inode *inode, struct file *file)
3593 struct ftrace_buffer_info *info = file->private_data;
3596 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3603 struct ring_buffer *buffer;
3608 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3609 struct pipe_buffer *buf)
3611 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3616 ring_buffer_free_read_page(ref->buffer, ref->page);
3621 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3622 struct pipe_buffer *buf)
3627 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3628 struct pipe_buffer *buf)
3630 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3635 /* Pipe buffer operations for a buffer. */
3636 static struct pipe_buf_operations buffer_pipe_buf_ops = {
3638 .map = generic_pipe_buf_map,
3639 .unmap = generic_pipe_buf_unmap,
3640 .confirm = generic_pipe_buf_confirm,
3641 .release = buffer_pipe_buf_release,
3642 .steal = buffer_pipe_buf_steal,
3643 .get = buffer_pipe_buf_get,
3647 * Callback from splice_to_pipe(), if we need to release some pages
3648 * at the end of the spd in case we error'ed out in filling the pipe.
3650 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3652 struct buffer_ref *ref =
3653 (struct buffer_ref *)spd->partial[i].private;
3658 ring_buffer_free_read_page(ref->buffer, ref->page);
3660 spd->partial[i].private = 0;
3664 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3665 struct pipe_inode_info *pipe, size_t len,
3668 struct ftrace_buffer_info *info = file->private_data;
3669 struct partial_page partial[PIPE_BUFFERS];
3670 struct page *pages[PIPE_BUFFERS];
3671 struct splice_pipe_desc spd = {
3675 .ops = &buffer_pipe_buf_ops,
3676 .spd_release = buffer_spd_release,
3678 struct buffer_ref *ref;
3679 int entries, size, i;
3682 if (*ppos & (PAGE_SIZE - 1)) {
3683 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3687 if (len & (PAGE_SIZE - 1)) {
3688 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3689 if (len < PAGE_SIZE)
3694 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3696 for (i = 0; i < PIPE_BUFFERS && len && entries; i++, len -= PAGE_SIZE) {
3700 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3705 ref->buffer = info->tr->buffer;
3706 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3712 r = ring_buffer_read_page(ref->buffer, &ref->page,
3715 ring_buffer_free_read_page(ref->buffer,
3722 * zero out any left over data, this is going to
3725 size = ring_buffer_page_len(ref->page);
3726 if (size < PAGE_SIZE)
3727 memset(ref->page + size, 0, PAGE_SIZE - size);
3729 page = virt_to_page(ref->page);
3731 spd.pages[i] = page;
3732 spd.partial[i].len = PAGE_SIZE;
3733 spd.partial[i].offset = 0;
3734 spd.partial[i].private = (unsigned long)ref;
3738 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3743 /* did we read anything? */
3744 if (!spd.nr_pages) {
3745 if (flags & SPLICE_F_NONBLOCK)
3753 ret = splice_to_pipe(pipe, &spd);
3758 static const struct file_operations tracing_buffers_fops = {
3759 .open = tracing_buffers_open,
3760 .read = tracing_buffers_read,
3761 .release = tracing_buffers_release,
3762 .splice_read = tracing_buffers_splice_read,
3763 .llseek = no_llseek,
3767 tracing_stats_read(struct file *filp, char __user *ubuf,
3768 size_t count, loff_t *ppos)
3770 unsigned long cpu = (unsigned long)filp->private_data;
3771 struct trace_array *tr = &global_trace;
3772 struct trace_seq *s;
3775 s = kmalloc(sizeof(*s), GFP_KERNEL);
3781 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3782 trace_seq_printf(s, "entries: %ld\n", cnt);
3784 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3785 trace_seq_printf(s, "overrun: %ld\n", cnt);
3787 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3788 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3790 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3797 static const struct file_operations tracing_stats_fops = {
3798 .open = tracing_open_generic,
3799 .read = tracing_stats_read,
3802 #ifdef CONFIG_DYNAMIC_FTRACE
3804 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3810 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3811 size_t cnt, loff_t *ppos)
3813 static char ftrace_dyn_info_buffer[1024];
3814 static DEFINE_MUTEX(dyn_info_mutex);
3815 unsigned long *p = filp->private_data;
3816 char *buf = ftrace_dyn_info_buffer;
3817 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3820 mutex_lock(&dyn_info_mutex);
3821 r = sprintf(buf, "%ld ", *p);
3823 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3826 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3828 mutex_unlock(&dyn_info_mutex);
3833 static const struct file_operations tracing_dyn_info_fops = {
3834 .open = tracing_open_generic,
3835 .read = tracing_read_dyn_info,
3839 static struct dentry *d_tracer;
3841 struct dentry *tracing_init_dentry(void)
3848 if (!debugfs_initialized())
3851 d_tracer = debugfs_create_dir("tracing", NULL);
3853 if (!d_tracer && !once) {
3855 pr_warning("Could not create debugfs directory 'tracing'\n");
3862 static struct dentry *d_percpu;
3864 struct dentry *tracing_dentry_percpu(void)
3867 struct dentry *d_tracer;
3872 d_tracer = tracing_init_dentry();
3877 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3879 if (!d_percpu && !once) {
3881 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3888 static void tracing_init_debugfs_percpu(long cpu)
3890 struct dentry *d_percpu = tracing_dentry_percpu();
3891 struct dentry *d_cpu;
3892 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3895 if (cpu > 999 || cpu < 0)
3898 sprintf(cpu_dir, "cpu%ld", cpu);
3899 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3901 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3905 /* per cpu trace_pipe */
3906 trace_create_file("trace_pipe", 0444, d_cpu,
3907 (void *) cpu, &tracing_pipe_fops);
3910 trace_create_file("trace", 0644, d_cpu,
3911 (void *) cpu, &tracing_fops);
3913 trace_create_file("trace_pipe_raw", 0444, d_cpu,
3914 (void *) cpu, &tracing_buffers_fops);
3916 trace_create_file("stats", 0444, d_cpu,
3917 (void *) cpu, &tracing_stats_fops);
3920 #ifdef CONFIG_FTRACE_SELFTEST
3921 /* Let selftest have access to static functions in this file */
3922 #include "trace_selftest.c"
3925 struct trace_option_dentry {
3926 struct tracer_opt *opt;
3927 struct tracer_flags *flags;
3928 struct dentry *entry;
3932 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3935 struct trace_option_dentry *topt = filp->private_data;
3938 if (topt->flags->val & topt->opt->bit)
3943 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3947 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3950 struct trace_option_dentry *topt = filp->private_data;
3955 if (cnt >= sizeof(buf))
3958 if (copy_from_user(&buf, ubuf, cnt))
3963 ret = strict_strtoul(buf, 10, &val);
3970 /* do nothing if already cleared */
3971 if (!(topt->flags->val & topt->opt->bit))
3974 mutex_lock(&trace_types_lock);
3975 if (current_trace->set_flag)
3976 ret = current_trace->set_flag(topt->flags->val,
3978 mutex_unlock(&trace_types_lock);
3981 topt->flags->val &= ~topt->opt->bit;
3984 /* do nothing if already set */
3985 if (topt->flags->val & topt->opt->bit)
3988 mutex_lock(&trace_types_lock);
3989 if (current_trace->set_flag)
3990 ret = current_trace->set_flag(topt->flags->val,
3992 mutex_unlock(&trace_types_lock);
3995 topt->flags->val |= topt->opt->bit;
4008 static const struct file_operations trace_options_fops = {
4009 .open = tracing_open_generic,
4010 .read = trace_options_read,
4011 .write = trace_options_write,
4015 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4018 long index = (long)filp->private_data;
4021 if (trace_flags & (1 << index))
4026 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4030 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4033 long index = (long)filp->private_data;
4038 if (cnt >= sizeof(buf))
4041 if (copy_from_user(&buf, ubuf, cnt))
4046 ret = strict_strtoul(buf, 10, &val);
4050 if (val != 0 && val != 1)
4052 set_tracer_flags(1 << index, val);
4059 static const struct file_operations trace_options_core_fops = {
4060 .open = tracing_open_generic,
4061 .read = trace_options_core_read,
4062 .write = trace_options_core_write,
4065 struct dentry *trace_create_file(const char *name,
4067 struct dentry *parent,
4069 const struct file_operations *fops)
4073 ret = debugfs_create_file(name, mode, parent, data, fops);
4075 pr_warning("Could not create debugfs '%s' entry\n", name);
4081 static struct dentry *trace_options_init_dentry(void)
4083 struct dentry *d_tracer;
4084 static struct dentry *t_options;
4089 d_tracer = tracing_init_dentry();
4093 t_options = debugfs_create_dir("options", d_tracer);
4095 pr_warning("Could not create debugfs directory 'options'\n");
4103 create_trace_option_file(struct trace_option_dentry *topt,
4104 struct tracer_flags *flags,
4105 struct tracer_opt *opt)
4107 struct dentry *t_options;
4109 t_options = trace_options_init_dentry();
4113 topt->flags = flags;
4116 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4117 &trace_options_fops);
4121 static struct trace_option_dentry *
4122 create_trace_option_files(struct tracer *tracer)
4124 struct trace_option_dentry *topts;
4125 struct tracer_flags *flags;
4126 struct tracer_opt *opts;
4132 flags = tracer->flags;
4134 if (!flags || !flags->opts)
4139 for (cnt = 0; opts[cnt].name; cnt++)
4142 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4146 for (cnt = 0; opts[cnt].name; cnt++)
4147 create_trace_option_file(&topts[cnt], flags,
4154 destroy_trace_option_files(struct trace_option_dentry *topts)
4161 for (cnt = 0; topts[cnt].opt; cnt++) {
4162 if (topts[cnt].entry)
4163 debugfs_remove(topts[cnt].entry);
4169 static struct dentry *
4170 create_trace_option_core_file(const char *option, long index)
4172 struct dentry *t_options;
4174 t_options = trace_options_init_dentry();
4178 return trace_create_file(option, 0644, t_options, (void *)index,
4179 &trace_options_core_fops);
4182 static __init void create_trace_options_dir(void)
4184 struct dentry *t_options;
4187 t_options = trace_options_init_dentry();
4191 for (i = 0; trace_options[i]; i++)
4192 create_trace_option_core_file(trace_options[i], i);
4195 static __init int tracer_init_debugfs(void)
4197 struct dentry *d_tracer;
4200 d_tracer = tracing_init_dentry();
4202 trace_create_file("tracing_enabled", 0644, d_tracer,
4203 &global_trace, &tracing_ctrl_fops);
4205 trace_create_file("trace_options", 0644, d_tracer,
4206 NULL, &tracing_iter_fops);
4208 trace_create_file("tracing_cpumask", 0644, d_tracer,
4209 NULL, &tracing_cpumask_fops);
4211 trace_create_file("trace", 0644, d_tracer,
4212 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4214 trace_create_file("available_tracers", 0444, d_tracer,
4215 &global_trace, &show_traces_fops);
4217 trace_create_file("current_tracer", 0644, d_tracer,
4218 &global_trace, &set_tracer_fops);
4220 #ifdef CONFIG_TRACER_MAX_TRACE
4221 trace_create_file("tracing_max_latency", 0644, d_tracer,
4222 &tracing_max_latency, &tracing_max_lat_fops);
4224 trace_create_file("tracing_thresh", 0644, d_tracer,
4225 &tracing_thresh, &tracing_max_lat_fops);
4228 trace_create_file("README", 0444, d_tracer,
4229 NULL, &tracing_readme_fops);
4231 trace_create_file("trace_pipe", 0444, d_tracer,
4232 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4234 trace_create_file("buffer_size_kb", 0644, d_tracer,
4235 &global_trace, &tracing_entries_fops);
4237 trace_create_file("trace_marker", 0220, d_tracer,
4238 NULL, &tracing_mark_fops);
4240 trace_create_file("saved_cmdlines", 0444, d_tracer,
4241 NULL, &tracing_saved_cmdlines_fops);
4243 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4246 #ifdef CONFIG_DYNAMIC_FTRACE
4247 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4248 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4250 #ifdef CONFIG_SYSPROF_TRACER
4251 init_tracer_sysprof_debugfs(d_tracer);
4254 create_trace_options_dir();
4256 for_each_tracing_cpu(cpu)
4257 tracing_init_debugfs_percpu(cpu);
4262 static int trace_panic_handler(struct notifier_block *this,
4263 unsigned long event, void *unused)
4265 if (ftrace_dump_on_oops)
4270 static struct notifier_block trace_panic_notifier = {
4271 .notifier_call = trace_panic_handler,
4273 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4276 static int trace_die_handler(struct notifier_block *self,
4282 if (ftrace_dump_on_oops)
4291 static struct notifier_block trace_die_notifier = {
4292 .notifier_call = trace_die_handler,
4297 * printk is set to max of 1024, we really don't need it that big.
4298 * Nothing should be printing 1000 characters anyway.
4300 #define TRACE_MAX_PRINT 1000
4303 * Define here KERN_TRACE so that we have one place to modify
4304 * it if we decide to change what log level the ftrace dump
4307 #define KERN_TRACE KERN_EMERG
4310 trace_printk_seq(struct trace_seq *s)
4312 /* Probably should print a warning here. */
4316 /* should be zero ended, but we are paranoid. */
4317 s->buffer[s->len] = 0;
4319 printk(KERN_TRACE "%s", s->buffer);
4324 static void __ftrace_dump(bool disable_tracing)
4326 static raw_spinlock_t ftrace_dump_lock =
4327 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
4328 /* use static because iter can be a bit big for the stack */
4329 static struct trace_iterator iter;
4330 unsigned int old_userobj;
4331 static int dump_ran;
4332 unsigned long flags;
4336 local_irq_save(flags);
4337 __raw_spin_lock(&ftrace_dump_lock);
4345 if (disable_tracing)
4348 for_each_tracing_cpu(cpu) {
4349 atomic_inc(&global_trace.data[cpu]->disabled);
4352 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4354 /* don't look at user memory in panic mode */
4355 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4357 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4359 /* Simulate the iterator */
4360 iter.tr = &global_trace;
4361 iter.trace = current_trace;
4362 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4365 * We need to stop all tracing on all CPUS to read the
4366 * the next buffer. This is a bit expensive, but is
4367 * not done often. We fill all what we can read,
4368 * and then release the locks again.
4371 while (!trace_empty(&iter)) {
4374 printk(KERN_TRACE "---------------------------------\n");
4378 /* reset all but tr, trace, and overruns */
4379 memset(&iter.seq, 0,
4380 sizeof(struct trace_iterator) -
4381 offsetof(struct trace_iterator, seq));
4382 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4385 if (find_next_entry_inc(&iter) != NULL) {
4388 ret = print_trace_line(&iter);
4389 if (ret != TRACE_TYPE_NO_CONSUME)
4390 trace_consume(&iter);
4393 trace_printk_seq(&iter.seq);
4397 printk(KERN_TRACE " (ftrace buffer empty)\n");
4399 printk(KERN_TRACE "---------------------------------\n");
4401 /* Re-enable tracing if requested */
4402 if (!disable_tracing) {
4403 trace_flags |= old_userobj;
4405 for_each_tracing_cpu(cpu) {
4406 atomic_dec(&global_trace.data[cpu]->disabled);
4412 __raw_spin_unlock(&ftrace_dump_lock);
4413 local_irq_restore(flags);
4416 /* By default: disable tracing after the dump */
4417 void ftrace_dump(void)
4419 __ftrace_dump(true);
4422 __init static int tracer_alloc_buffers(void)
4428 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4431 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4432 goto out_free_buffer_mask;
4434 if (!zalloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
4435 goto out_free_tracing_cpumask;
4437 /* To save memory, keep the ring buffer size to its minimum */
4438 if (ring_buffer_expanded)
4439 ring_buf_size = trace_buf_size;
4443 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4444 cpumask_copy(tracing_cpumask, cpu_all_mask);
4446 /* TODO: make the number of buffers hot pluggable with CPUS */
4447 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4448 TRACE_BUFFER_FLAGS);
4449 if (!global_trace.buffer) {
4450 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4452 goto out_free_cpumask;
4454 global_trace.entries = ring_buffer_size(global_trace.buffer);
4457 #ifdef CONFIG_TRACER_MAX_TRACE
4458 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4459 TRACE_BUFFER_FLAGS);
4460 if (!max_tr.buffer) {
4461 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4463 ring_buffer_free(global_trace.buffer);
4464 goto out_free_cpumask;
4466 max_tr.entries = ring_buffer_size(max_tr.buffer);
4467 WARN_ON(max_tr.entries != global_trace.entries);
4470 /* Allocate the first page for all buffers */
4471 for_each_tracing_cpu(i) {
4472 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4473 max_tr.data[i] = &per_cpu(max_data, i);
4476 trace_init_cmdlines();
4478 register_tracer(&nop_trace);
4479 current_trace = &nop_trace;
4480 #ifdef CONFIG_BOOT_TRACER
4481 register_tracer(&boot_tracer);
4483 /* All seems OK, enable tracing */
4484 tracing_disabled = 0;
4486 atomic_notifier_chain_register(&panic_notifier_list,
4487 &trace_panic_notifier);
4489 register_die_notifier(&trace_die_notifier);
4494 free_cpumask_var(tracing_reader_cpumask);
4495 out_free_tracing_cpumask:
4496 free_cpumask_var(tracing_cpumask);
4497 out_free_buffer_mask:
4498 free_cpumask_var(tracing_buffer_mask);
4503 __init static int clear_boot_tracer(void)
4506 * The default tracer at boot buffer is an init section.
4507 * This function is called in lateinit. If we did not
4508 * find the boot tracer, then clear it out, to prevent
4509 * later registration from accessing the buffer that is
4510 * about to be freed.
4512 if (!default_bootup_tracer)
4515 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4516 default_bootup_tracer);
4517 default_bootup_tracer = NULL;
4522 early_initcall(tracer_alloc_buffers);
4523 fs_initcall(tracer_init_debugfs);
4524 late_initcall(clear_boot_tracer);