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(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[] = {
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 arch_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 arch_spinlock_t ftrace_max_lock =
504 (arch_spinlock_t)__ARCH_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 arch_spin_lock(&ftrace_max_lock);
560 tr->buffer = max_tr.buffer;
563 __update_max_tr(tr, tsk, cpu);
564 arch_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 arch_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 arch_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 arch_spinlock_t trace_cmdline_lock = __ARCH_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 (!arch_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 arch_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 arch_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 arch_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(__this_cpu_read(per_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 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1157 struct ftrace_event_call *call = &event_user_stack;
1158 struct ring_buffer_event *event;
1159 struct userstack_entry *entry;
1160 struct stack_trace trace;
1162 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1165 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1166 sizeof(*entry), flags, pc);
1169 entry = ring_buffer_event_data(event);
1171 entry->tgid = current->tgid;
1172 memset(&entry->caller, 0, sizeof(entry->caller));
1174 trace.nr_entries = 0;
1175 trace.max_entries = FTRACE_STACK_ENTRIES;
1177 trace.entries = entry->caller;
1179 save_stack_trace_user(&trace);
1180 if (!filter_check_discard(call, entry, buffer, event))
1181 ring_buffer_unlock_commit(buffer, event);
1185 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1187 ftrace_trace_userstack(tr, flags, preempt_count());
1191 #endif /* CONFIG_STACKTRACE */
1194 ftrace_trace_special(void *__tr,
1195 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1198 struct ftrace_event_call *call = &event_special;
1199 struct ring_buffer_event *event;
1200 struct trace_array *tr = __tr;
1201 struct ring_buffer *buffer = tr->buffer;
1202 struct special_entry *entry;
1204 event = trace_buffer_lock_reserve(buffer, TRACE_SPECIAL,
1205 sizeof(*entry), 0, pc);
1208 entry = ring_buffer_event_data(event);
1213 if (!filter_check_discard(call, entry, buffer, event))
1214 trace_buffer_unlock_commit(buffer, event, 0, pc);
1218 __trace_special(void *__tr, void *__data,
1219 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1221 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1225 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1227 struct trace_array *tr = &global_trace;
1228 struct trace_array_cpu *data;
1229 unsigned long flags;
1233 if (tracing_disabled)
1236 pc = preempt_count();
1237 local_irq_save(flags);
1238 cpu = raw_smp_processor_id();
1239 data = tr->data[cpu];
1241 if (likely(atomic_inc_return(&data->disabled) == 1))
1242 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1244 atomic_dec(&data->disabled);
1245 local_irq_restore(flags);
1249 * trace_vbprintk - write binary msg to tracing buffer
1252 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1254 static arch_spinlock_t trace_buf_lock =
1255 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1256 static u32 trace_buf[TRACE_BUF_SIZE];
1258 struct ftrace_event_call *call = &event_bprint;
1259 struct ring_buffer_event *event;
1260 struct ring_buffer *buffer;
1261 struct trace_array *tr = &global_trace;
1262 struct trace_array_cpu *data;
1263 struct bprint_entry *entry;
1264 unsigned long flags;
1267 int cpu, len = 0, size, pc;
1269 if (unlikely(tracing_selftest_running || tracing_disabled))
1272 /* Don't pollute graph traces with trace_vprintk internals */
1273 pause_graph_tracing();
1275 pc = preempt_count();
1276 resched = ftrace_preempt_disable();
1277 cpu = raw_smp_processor_id();
1278 data = tr->data[cpu];
1280 disable = atomic_inc_return(&data->disabled);
1281 if (unlikely(disable != 1))
1284 /* Lockdep uses trace_printk for lock tracing */
1285 local_irq_save(flags);
1286 arch_spin_lock(&trace_buf_lock);
1287 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1289 if (len > TRACE_BUF_SIZE || len < 0)
1292 size = sizeof(*entry) + sizeof(u32) * len;
1293 buffer = tr->buffer;
1294 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1298 entry = ring_buffer_event_data(event);
1302 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1303 if (!filter_check_discard(call, entry, buffer, event))
1304 ring_buffer_unlock_commit(buffer, event);
1307 arch_spin_unlock(&trace_buf_lock);
1308 local_irq_restore(flags);
1311 atomic_dec_return(&data->disabled);
1312 ftrace_preempt_enable(resched);
1313 unpause_graph_tracing();
1317 EXPORT_SYMBOL_GPL(trace_vbprintk);
1319 int trace_array_printk(struct trace_array *tr,
1320 unsigned long ip, const char *fmt, ...)
1325 if (!(trace_flags & TRACE_ITER_PRINTK))
1329 ret = trace_array_vprintk(tr, ip, fmt, ap);
1334 int trace_array_vprintk(struct trace_array *tr,
1335 unsigned long ip, const char *fmt, va_list args)
1337 static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1338 static char trace_buf[TRACE_BUF_SIZE];
1340 struct ftrace_event_call *call = &event_print;
1341 struct ring_buffer_event *event;
1342 struct ring_buffer *buffer;
1343 struct trace_array_cpu *data;
1344 int cpu, len = 0, size, pc;
1345 struct print_entry *entry;
1346 unsigned long irq_flags;
1349 if (tracing_disabled || tracing_selftest_running)
1352 pc = preempt_count();
1353 preempt_disable_notrace();
1354 cpu = raw_smp_processor_id();
1355 data = tr->data[cpu];
1357 disable = atomic_inc_return(&data->disabled);
1358 if (unlikely(disable != 1))
1361 pause_graph_tracing();
1362 raw_local_irq_save(irq_flags);
1363 arch_spin_lock(&trace_buf_lock);
1364 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1366 size = sizeof(*entry) + len + 1;
1367 buffer = tr->buffer;
1368 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1372 entry = ring_buffer_event_data(event);
1375 memcpy(&entry->buf, trace_buf, len);
1376 entry->buf[len] = '\0';
1377 if (!filter_check_discard(call, entry, buffer, event))
1378 ring_buffer_unlock_commit(buffer, event);
1381 arch_spin_unlock(&trace_buf_lock);
1382 raw_local_irq_restore(irq_flags);
1383 unpause_graph_tracing();
1385 atomic_dec_return(&data->disabled);
1386 preempt_enable_notrace();
1391 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1393 return trace_array_vprintk(&global_trace, ip, fmt, args);
1395 EXPORT_SYMBOL_GPL(trace_vprintk);
1397 enum trace_file_type {
1398 TRACE_FILE_LAT_FMT = 1,
1399 TRACE_FILE_ANNOTATE = 2,
1402 static void trace_iterator_increment(struct trace_iterator *iter)
1404 /* Don't allow ftrace to trace into the ring buffers */
1405 ftrace_disable_cpu();
1408 if (iter->buffer_iter[iter->cpu])
1409 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1411 ftrace_enable_cpu();
1414 static struct trace_entry *
1415 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1417 struct ring_buffer_event *event;
1418 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1420 /* Don't allow ftrace to trace into the ring buffers */
1421 ftrace_disable_cpu();
1424 event = ring_buffer_iter_peek(buf_iter, ts);
1426 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1428 ftrace_enable_cpu();
1430 return event ? ring_buffer_event_data(event) : NULL;
1433 static struct trace_entry *
1434 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1436 struct ring_buffer *buffer = iter->tr->buffer;
1437 struct trace_entry *ent, *next = NULL;
1438 int cpu_file = iter->cpu_file;
1439 u64 next_ts = 0, ts;
1444 * If we are in a per_cpu trace file, don't bother by iterating over
1445 * all cpu and peek directly.
1447 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1448 if (ring_buffer_empty_cpu(buffer, cpu_file))
1450 ent = peek_next_entry(iter, cpu_file, ent_ts);
1452 *ent_cpu = cpu_file;
1457 for_each_tracing_cpu(cpu) {
1459 if (ring_buffer_empty_cpu(buffer, cpu))
1462 ent = peek_next_entry(iter, cpu, &ts);
1465 * Pick the entry with the smallest timestamp:
1467 if (ent && (!next || ts < next_ts)) {
1475 *ent_cpu = next_cpu;
1483 /* Find the next real entry, without updating the iterator itself */
1484 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1485 int *ent_cpu, u64 *ent_ts)
1487 return __find_next_entry(iter, ent_cpu, ent_ts);
1490 /* Find the next real entry, and increment the iterator to the next entry */
1491 static void *find_next_entry_inc(struct trace_iterator *iter)
1493 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1496 trace_iterator_increment(iter);
1498 return iter->ent ? iter : NULL;
1501 static void trace_consume(struct trace_iterator *iter)
1503 /* Don't allow ftrace to trace into the ring buffers */
1504 ftrace_disable_cpu();
1505 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1506 ftrace_enable_cpu();
1509 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1511 struct trace_iterator *iter = m->private;
1515 WARN_ON_ONCE(iter->leftover);
1519 /* can't go backwards */
1524 ent = find_next_entry_inc(iter);
1528 while (ent && iter->idx < i)
1529 ent = find_next_entry_inc(iter);
1536 static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1538 struct trace_array *tr = iter->tr;
1539 struct ring_buffer_event *event;
1540 struct ring_buffer_iter *buf_iter;
1541 unsigned long entries = 0;
1544 tr->data[cpu]->skipped_entries = 0;
1546 if (!iter->buffer_iter[cpu])
1549 buf_iter = iter->buffer_iter[cpu];
1550 ring_buffer_iter_reset(buf_iter);
1553 * We could have the case with the max latency tracers
1554 * that a reset never took place on a cpu. This is evident
1555 * by the timestamp being before the start of the buffer.
1557 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1558 if (ts >= iter->tr->time_start)
1561 ring_buffer_read(buf_iter, NULL);
1564 tr->data[cpu]->skipped_entries = entries;
1568 * No necessary locking here. The worst thing which can
1569 * happen is loosing events consumed at the same time
1570 * by a trace_pipe reader.
1571 * Other than that, we don't risk to crash the ring buffer
1572 * because it serializes the readers.
1574 * The current tracer is copied to avoid a global locking
1577 static void *s_start(struct seq_file *m, loff_t *pos)
1579 struct trace_iterator *iter = m->private;
1580 static struct tracer *old_tracer;
1581 int cpu_file = iter->cpu_file;
1586 /* copy the tracer to avoid using a global lock all around */
1587 mutex_lock(&trace_types_lock);
1588 if (unlikely(old_tracer != current_trace && current_trace)) {
1589 old_tracer = current_trace;
1590 *iter->trace = *current_trace;
1592 mutex_unlock(&trace_types_lock);
1594 atomic_inc(&trace_record_cmdline_disabled);
1596 if (*pos != iter->pos) {
1601 ftrace_disable_cpu();
1603 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1604 for_each_tracing_cpu(cpu)
1605 tracing_iter_reset(iter, cpu);
1607 tracing_iter_reset(iter, cpu_file);
1609 ftrace_enable_cpu();
1611 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1616 * If we overflowed the seq_file before, then we want
1617 * to just reuse the trace_seq buffer again.
1623 p = s_next(m, p, &l);
1627 trace_event_read_lock();
1631 static void s_stop(struct seq_file *m, void *p)
1633 atomic_dec(&trace_record_cmdline_disabled);
1634 trace_event_read_unlock();
1637 static void print_lat_help_header(struct seq_file *m)
1639 seq_puts(m, "# _------=> CPU# \n");
1640 seq_puts(m, "# / _-----=> irqs-off \n");
1641 seq_puts(m, "# | / _----=> need-resched \n");
1642 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1643 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1644 seq_puts(m, "# |||| /_--=> lock-depth \n");
1645 seq_puts(m, "# |||||/ delay \n");
1646 seq_puts(m, "# cmd pid |||||| time | caller \n");
1647 seq_puts(m, "# \\ / |||||| \\ | / \n");
1650 static void print_func_help_header(struct seq_file *m)
1652 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1653 seq_puts(m, "# | | | | |\n");
1658 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1660 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1661 struct trace_array *tr = iter->tr;
1662 struct trace_array_cpu *data = tr->data[tr->cpu];
1663 struct tracer *type = current_trace;
1664 unsigned long entries = 0;
1665 unsigned long total = 0;
1666 unsigned long count;
1667 const char *name = "preemption";
1674 for_each_tracing_cpu(cpu) {
1675 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1677 * If this buffer has skipped entries, then we hold all
1678 * entries for the trace and we need to ignore the
1679 * ones before the time stamp.
1681 if (tr->data[cpu]->skipped_entries) {
1682 count -= tr->data[cpu]->skipped_entries;
1683 /* total is the same as the entries */
1687 ring_buffer_overrun_cpu(tr->buffer, cpu);
1691 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1693 seq_puts(m, "# -----------------------------------"
1694 "---------------------------------\n");
1695 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1696 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1697 nsecs_to_usecs(data->saved_latency),
1701 #if defined(CONFIG_PREEMPT_NONE)
1703 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1705 #elif defined(CONFIG_PREEMPT)
1710 /* These are reserved for later use */
1713 seq_printf(m, " #P:%d)\n", num_online_cpus());
1717 seq_puts(m, "# -----------------\n");
1718 seq_printf(m, "# | task: %.16s-%d "
1719 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1720 data->comm, data->pid, data->uid, data->nice,
1721 data->policy, data->rt_priority);
1722 seq_puts(m, "# -----------------\n");
1724 if (data->critical_start) {
1725 seq_puts(m, "# => started at: ");
1726 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1727 trace_print_seq(m, &iter->seq);
1728 seq_puts(m, "\n# => ended at: ");
1729 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1730 trace_print_seq(m, &iter->seq);
1731 seq_puts(m, "\n#\n");
1737 static void test_cpu_buff_start(struct trace_iterator *iter)
1739 struct trace_seq *s = &iter->seq;
1741 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1744 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1747 if (cpumask_test_cpu(iter->cpu, iter->started))
1750 if (iter->tr->data[iter->cpu]->skipped_entries)
1753 cpumask_set_cpu(iter->cpu, iter->started);
1755 /* Don't print started cpu buffer for the first entry of the trace */
1757 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1761 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1763 struct trace_seq *s = &iter->seq;
1764 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1765 struct trace_entry *entry;
1766 struct trace_event *event;
1770 test_cpu_buff_start(iter);
1772 event = ftrace_find_event(entry->type);
1774 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1775 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1776 if (!trace_print_lat_context(iter))
1779 if (!trace_print_context(iter))
1785 return event->trace(iter, sym_flags);
1787 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1790 return TRACE_TYPE_HANDLED;
1792 return TRACE_TYPE_PARTIAL_LINE;
1795 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1797 struct trace_seq *s = &iter->seq;
1798 struct trace_entry *entry;
1799 struct trace_event *event;
1803 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1804 if (!trace_seq_printf(s, "%d %d %llu ",
1805 entry->pid, iter->cpu, iter->ts))
1809 event = ftrace_find_event(entry->type);
1811 return event->raw(iter, 0);
1813 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1816 return TRACE_TYPE_HANDLED;
1818 return TRACE_TYPE_PARTIAL_LINE;
1821 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1823 struct trace_seq *s = &iter->seq;
1824 unsigned char newline = '\n';
1825 struct trace_entry *entry;
1826 struct trace_event *event;
1830 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1831 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1832 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1833 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1836 event = ftrace_find_event(entry->type);
1838 enum print_line_t ret = event->hex(iter, 0);
1839 if (ret != TRACE_TYPE_HANDLED)
1843 SEQ_PUT_FIELD_RET(s, newline);
1845 return TRACE_TYPE_HANDLED;
1848 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1850 struct trace_seq *s = &iter->seq;
1851 struct trace_entry *entry;
1852 struct trace_event *event;
1856 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1857 SEQ_PUT_FIELD_RET(s, entry->pid);
1858 SEQ_PUT_FIELD_RET(s, iter->cpu);
1859 SEQ_PUT_FIELD_RET(s, iter->ts);
1862 event = ftrace_find_event(entry->type);
1863 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1866 static int trace_empty(struct trace_iterator *iter)
1870 /* If we are looking at one CPU buffer, only check that one */
1871 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1872 cpu = iter->cpu_file;
1873 if (iter->buffer_iter[cpu]) {
1874 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1877 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1883 for_each_tracing_cpu(cpu) {
1884 if (iter->buffer_iter[cpu]) {
1885 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1888 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1896 /* Called with trace_event_read_lock() held. */
1897 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1899 enum print_line_t ret;
1901 if (iter->trace && iter->trace->print_line) {
1902 ret = iter->trace->print_line(iter);
1903 if (ret != TRACE_TYPE_UNHANDLED)
1907 if (iter->ent->type == TRACE_BPRINT &&
1908 trace_flags & TRACE_ITER_PRINTK &&
1909 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1910 return trace_print_bprintk_msg_only(iter);
1912 if (iter->ent->type == TRACE_PRINT &&
1913 trace_flags & TRACE_ITER_PRINTK &&
1914 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1915 return trace_print_printk_msg_only(iter);
1917 if (trace_flags & TRACE_ITER_BIN)
1918 return print_bin_fmt(iter);
1920 if (trace_flags & TRACE_ITER_HEX)
1921 return print_hex_fmt(iter);
1923 if (trace_flags & TRACE_ITER_RAW)
1924 return print_raw_fmt(iter);
1926 return print_trace_fmt(iter);
1929 static int s_show(struct seq_file *m, void *v)
1931 struct trace_iterator *iter = v;
1934 if (iter->ent == NULL) {
1936 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1939 if (iter->trace && iter->trace->print_header)
1940 iter->trace->print_header(m);
1941 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1942 /* print nothing if the buffers are empty */
1943 if (trace_empty(iter))
1945 print_trace_header(m, iter);
1946 if (!(trace_flags & TRACE_ITER_VERBOSE))
1947 print_lat_help_header(m);
1949 if (!(trace_flags & TRACE_ITER_VERBOSE))
1950 print_func_help_header(m);
1952 } else if (iter->leftover) {
1954 * If we filled the seq_file buffer earlier, we
1955 * want to just show it now.
1957 ret = trace_print_seq(m, &iter->seq);
1959 /* ret should this time be zero, but you never know */
1960 iter->leftover = ret;
1963 print_trace_line(iter);
1964 ret = trace_print_seq(m, &iter->seq);
1966 * If we overflow the seq_file buffer, then it will
1967 * ask us for this data again at start up.
1969 * ret is 0 if seq_file write succeeded.
1972 iter->leftover = ret;
1978 static const struct seq_operations tracer_seq_ops = {
1985 static struct trace_iterator *
1986 __tracing_open(struct inode *inode, struct file *file)
1988 long cpu_file = (long) inode->i_private;
1989 void *fail_ret = ERR_PTR(-ENOMEM);
1990 struct trace_iterator *iter;
1994 if (tracing_disabled)
1995 return ERR_PTR(-ENODEV);
1997 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1999 return ERR_PTR(-ENOMEM);
2002 * We make a copy of the current tracer to avoid concurrent
2003 * changes on it while we are reading.
2005 mutex_lock(&trace_types_lock);
2006 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2011 *iter->trace = *current_trace;
2013 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2016 if (current_trace && current_trace->print_max)
2019 iter->tr = &global_trace;
2021 mutex_init(&iter->mutex);
2022 iter->cpu_file = cpu_file;
2024 /* Notify the tracer early; before we stop tracing. */
2025 if (iter->trace && iter->trace->open)
2026 iter->trace->open(iter);
2028 /* Annotate start of buffers if we had overruns */
2029 if (ring_buffer_overruns(iter->tr->buffer))
2030 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2032 /* stop the trace while dumping */
2035 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2036 for_each_tracing_cpu(cpu) {
2038 iter->buffer_iter[cpu] =
2039 ring_buffer_read_start(iter->tr->buffer, cpu);
2040 tracing_iter_reset(iter, cpu);
2043 cpu = iter->cpu_file;
2044 iter->buffer_iter[cpu] =
2045 ring_buffer_read_start(iter->tr->buffer, cpu);
2046 tracing_iter_reset(iter, cpu);
2049 ret = seq_open(file, &tracer_seq_ops);
2051 fail_ret = ERR_PTR(ret);
2055 m = file->private_data;
2058 mutex_unlock(&trace_types_lock);
2063 for_each_tracing_cpu(cpu) {
2064 if (iter->buffer_iter[cpu])
2065 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2067 free_cpumask_var(iter->started);
2070 mutex_unlock(&trace_types_lock);
2077 int tracing_open_generic(struct inode *inode, struct file *filp)
2079 if (tracing_disabled)
2082 filp->private_data = inode->i_private;
2086 static int tracing_release(struct inode *inode, struct file *file)
2088 struct seq_file *m = (struct seq_file *)file->private_data;
2089 struct trace_iterator *iter;
2092 if (!(file->f_mode & FMODE_READ))
2097 mutex_lock(&trace_types_lock);
2098 for_each_tracing_cpu(cpu) {
2099 if (iter->buffer_iter[cpu])
2100 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2103 if (iter->trace && iter->trace->close)
2104 iter->trace->close(iter);
2106 /* reenable tracing if it was previously enabled */
2108 mutex_unlock(&trace_types_lock);
2110 seq_release(inode, file);
2111 mutex_destroy(&iter->mutex);
2112 free_cpumask_var(iter->started);
2118 static int tracing_open(struct inode *inode, struct file *file)
2120 struct trace_iterator *iter;
2123 /* If this file was open for write, then erase contents */
2124 if ((file->f_mode & FMODE_WRITE) &&
2125 (file->f_flags & O_TRUNC)) {
2126 long cpu = (long) inode->i_private;
2128 if (cpu == TRACE_PIPE_ALL_CPU)
2129 tracing_reset_online_cpus(&global_trace);
2131 tracing_reset(&global_trace, cpu);
2134 if (file->f_mode & FMODE_READ) {
2135 iter = __tracing_open(inode, file);
2137 ret = PTR_ERR(iter);
2138 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2139 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2145 t_next(struct seq_file *m, void *v, loff_t *pos)
2147 struct tracer *t = v;
2157 static void *t_start(struct seq_file *m, loff_t *pos)
2162 mutex_lock(&trace_types_lock);
2163 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2169 static void t_stop(struct seq_file *m, void *p)
2171 mutex_unlock(&trace_types_lock);
2174 static int t_show(struct seq_file *m, void *v)
2176 struct tracer *t = v;
2181 seq_printf(m, "%s", t->name);
2190 static const struct seq_operations show_traces_seq_ops = {
2197 static int show_traces_open(struct inode *inode, struct file *file)
2199 if (tracing_disabled)
2202 return seq_open(file, &show_traces_seq_ops);
2206 tracing_write_stub(struct file *filp, const char __user *ubuf,
2207 size_t count, loff_t *ppos)
2212 static const struct file_operations tracing_fops = {
2213 .open = tracing_open,
2215 .write = tracing_write_stub,
2216 .llseek = seq_lseek,
2217 .release = tracing_release,
2220 static const struct file_operations show_traces_fops = {
2221 .open = show_traces_open,
2223 .release = seq_release,
2227 * Only trace on a CPU if the bitmask is set:
2229 static cpumask_var_t tracing_cpumask;
2232 * The tracer itself will not take this lock, but still we want
2233 * to provide a consistent cpumask to user-space:
2235 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2238 * Temporary storage for the character representation of the
2239 * CPU bitmask (and one more byte for the newline):
2241 static char mask_str[NR_CPUS + 1];
2244 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2245 size_t count, loff_t *ppos)
2249 mutex_lock(&tracing_cpumask_update_lock);
2251 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2252 if (count - len < 2) {
2256 len += sprintf(mask_str + len, "\n");
2257 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2260 mutex_unlock(&tracing_cpumask_update_lock);
2266 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2267 size_t count, loff_t *ppos)
2270 cpumask_var_t tracing_cpumask_new;
2272 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2275 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2279 mutex_lock(&tracing_cpumask_update_lock);
2281 local_irq_disable();
2282 arch_spin_lock(&ftrace_max_lock);
2283 for_each_tracing_cpu(cpu) {
2285 * Increase/decrease the disabled counter if we are
2286 * about to flip a bit in the cpumask:
2288 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2289 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2290 atomic_inc(&global_trace.data[cpu]->disabled);
2292 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2293 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2294 atomic_dec(&global_trace.data[cpu]->disabled);
2297 arch_spin_unlock(&ftrace_max_lock);
2300 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2302 mutex_unlock(&tracing_cpumask_update_lock);
2303 free_cpumask_var(tracing_cpumask_new);
2308 free_cpumask_var(tracing_cpumask_new);
2313 static const struct file_operations tracing_cpumask_fops = {
2314 .open = tracing_open_generic,
2315 .read = tracing_cpumask_read,
2316 .write = tracing_cpumask_write,
2320 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2321 size_t cnt, loff_t *ppos)
2323 struct tracer_opt *trace_opts;
2331 /* calculate max size */
2332 for (i = 0; trace_options[i]; i++) {
2333 len += strlen(trace_options[i]);
2334 len += 3; /* "no" and newline */
2337 mutex_lock(&trace_types_lock);
2338 tracer_flags = current_trace->flags->val;
2339 trace_opts = current_trace->flags->opts;
2342 * Increase the size with names of options specific
2343 * of the current tracer.
2345 for (i = 0; trace_opts[i].name; i++) {
2346 len += strlen(trace_opts[i].name);
2347 len += 3; /* "no" and newline */
2351 buf = kmalloc(len + 1, GFP_KERNEL);
2353 mutex_unlock(&trace_types_lock);
2357 for (i = 0; trace_options[i]; i++) {
2358 if (trace_flags & (1 << i))
2359 r += sprintf(buf + r, "%s\n", trace_options[i]);
2361 r += sprintf(buf + r, "no%s\n", trace_options[i]);
2364 for (i = 0; trace_opts[i].name; i++) {
2365 if (tracer_flags & trace_opts[i].bit)
2366 r += sprintf(buf + r, "%s\n",
2367 trace_opts[i].name);
2369 r += sprintf(buf + r, "no%s\n",
2370 trace_opts[i].name);
2372 mutex_unlock(&trace_types_lock);
2374 WARN_ON(r >= len + 1);
2376 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2382 /* Try to assign a tracer specific option */
2383 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2385 struct tracer_flags *tracer_flags = trace->flags;
2386 struct tracer_opt *opts = NULL;
2390 for (i = 0; tracer_flags->opts[i].name; i++) {
2391 opts = &tracer_flags->opts[i];
2392 len = strlen(opts->name);
2394 if (strncmp(cmp, opts->name, len) == 0) {
2395 ret = trace->set_flag(tracer_flags->val,
2401 if (!tracer_flags->opts[i].name)
2404 /* Refused to handle */
2409 tracer_flags->val &= ~opts->bit;
2411 tracer_flags->val |= opts->bit;
2416 static void set_tracer_flags(unsigned int mask, int enabled)
2418 /* do nothing if flag is already set */
2419 if (!!(trace_flags & mask) == !!enabled)
2423 trace_flags |= mask;
2425 trace_flags &= ~mask;
2429 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2430 size_t cnt, loff_t *ppos)
2438 if (cnt >= sizeof(buf))
2441 if (copy_from_user(&buf, ubuf, cnt))
2446 if (strncmp(buf, "no", 2) == 0) {
2451 for (i = 0; trace_options[i]; i++) {
2452 int len = strlen(trace_options[i]);
2454 if (strncmp(cmp, trace_options[i], len) == 0) {
2455 set_tracer_flags(1 << i, !neg);
2460 /* If no option could be set, test the specific tracer options */
2461 if (!trace_options[i]) {
2462 mutex_lock(&trace_types_lock);
2463 ret = set_tracer_option(current_trace, cmp, neg);
2464 mutex_unlock(&trace_types_lock);
2474 static const struct file_operations tracing_iter_fops = {
2475 .open = tracing_open_generic,
2476 .read = tracing_trace_options_read,
2477 .write = tracing_trace_options_write,
2480 static const char readme_msg[] =
2481 "tracing mini-HOWTO:\n\n"
2482 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2483 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2484 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2485 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2487 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2488 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2490 "# cat /sys/kernel/debug/tracing/trace_options\n"
2491 "noprint-parent nosym-offset nosym-addr noverbose\n"
2492 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2493 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2494 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2495 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2499 tracing_readme_read(struct file *filp, char __user *ubuf,
2500 size_t cnt, loff_t *ppos)
2502 return simple_read_from_buffer(ubuf, cnt, ppos,
2503 readme_msg, strlen(readme_msg));
2506 static const struct file_operations tracing_readme_fops = {
2507 .open = tracing_open_generic,
2508 .read = tracing_readme_read,
2512 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2513 size_t cnt, loff_t *ppos)
2522 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2526 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2534 for (i = 0; i < SAVED_CMDLINES; i++) {
2537 pid = map_cmdline_to_pid[i];
2538 if (pid == -1 || pid == NO_CMDLINE_MAP)
2541 trace_find_cmdline(pid, buf_comm);
2542 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2547 len = simple_read_from_buffer(ubuf, cnt, ppos,
2556 static const struct file_operations tracing_saved_cmdlines_fops = {
2557 .open = tracing_open_generic,
2558 .read = tracing_saved_cmdlines_read,
2562 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2563 size_t cnt, loff_t *ppos)
2568 r = sprintf(buf, "%u\n", tracer_enabled);
2569 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2573 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2574 size_t cnt, loff_t *ppos)
2576 struct trace_array *tr = filp->private_data;
2581 if (cnt >= sizeof(buf))
2584 if (copy_from_user(&buf, ubuf, cnt))
2589 ret = strict_strtoul(buf, 10, &val);
2595 mutex_lock(&trace_types_lock);
2596 if (tracer_enabled ^ val) {
2599 if (current_trace->start)
2600 current_trace->start(tr);
2605 if (current_trace->stop)
2606 current_trace->stop(tr);
2609 mutex_unlock(&trace_types_lock);
2617 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2618 size_t cnt, loff_t *ppos)
2620 char buf[MAX_TRACER_SIZE+2];
2623 mutex_lock(&trace_types_lock);
2625 r = sprintf(buf, "%s\n", current_trace->name);
2627 r = sprintf(buf, "\n");
2628 mutex_unlock(&trace_types_lock);
2630 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2633 int tracer_init(struct tracer *t, struct trace_array *tr)
2635 tracing_reset_online_cpus(tr);
2639 static int tracing_resize_ring_buffer(unsigned long size)
2644 * If kernel or user changes the size of the ring buffer
2645 * we use the size that was given, and we can forget about
2646 * expanding it later.
2648 ring_buffer_expanded = 1;
2650 ret = ring_buffer_resize(global_trace.buffer, size);
2654 ret = ring_buffer_resize(max_tr.buffer, size);
2658 r = ring_buffer_resize(global_trace.buffer,
2659 global_trace.entries);
2662 * AARGH! We are left with different
2663 * size max buffer!!!!
2664 * The max buffer is our "snapshot" buffer.
2665 * When a tracer needs a snapshot (one of the
2666 * latency tracers), it swaps the max buffer
2667 * with the saved snap shot. We succeeded to
2668 * update the size of the main buffer, but failed to
2669 * update the size of the max buffer. But when we tried
2670 * to reset the main buffer to the original size, we
2671 * failed there too. This is very unlikely to
2672 * happen, but if it does, warn and kill all
2676 tracing_disabled = 1;
2681 global_trace.entries = size;
2687 * tracing_update_buffers - used by tracing facility to expand ring buffers
2689 * To save on memory when the tracing is never used on a system with it
2690 * configured in. The ring buffers are set to a minimum size. But once
2691 * a user starts to use the tracing facility, then they need to grow
2692 * to their default size.
2694 * This function is to be called when a tracer is about to be used.
2696 int tracing_update_buffers(void)
2700 mutex_lock(&trace_types_lock);
2701 if (!ring_buffer_expanded)
2702 ret = tracing_resize_ring_buffer(trace_buf_size);
2703 mutex_unlock(&trace_types_lock);
2708 struct trace_option_dentry;
2710 static struct trace_option_dentry *
2711 create_trace_option_files(struct tracer *tracer);
2714 destroy_trace_option_files(struct trace_option_dentry *topts);
2716 static int tracing_set_tracer(const char *buf)
2718 static struct trace_option_dentry *topts;
2719 struct trace_array *tr = &global_trace;
2723 mutex_lock(&trace_types_lock);
2725 if (!ring_buffer_expanded) {
2726 ret = tracing_resize_ring_buffer(trace_buf_size);
2732 for (t = trace_types; t; t = t->next) {
2733 if (strcmp(t->name, buf) == 0)
2740 if (t == current_trace)
2743 trace_branch_disable();
2744 if (current_trace && current_trace->reset)
2745 current_trace->reset(tr);
2747 destroy_trace_option_files(topts);
2751 topts = create_trace_option_files(current_trace);
2754 ret = tracer_init(t, tr);
2759 trace_branch_enable(tr);
2761 mutex_unlock(&trace_types_lock);
2767 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2768 size_t cnt, loff_t *ppos)
2770 char buf[MAX_TRACER_SIZE+1];
2777 if (cnt > MAX_TRACER_SIZE)
2778 cnt = MAX_TRACER_SIZE;
2780 if (copy_from_user(&buf, ubuf, cnt))
2785 /* strip ending whitespace. */
2786 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2789 err = tracing_set_tracer(buf);
2799 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2800 size_t cnt, loff_t *ppos)
2802 unsigned long *ptr = filp->private_data;
2806 r = snprintf(buf, sizeof(buf), "%ld\n",
2807 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2808 if (r > sizeof(buf))
2810 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2814 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2815 size_t cnt, loff_t *ppos)
2817 unsigned long *ptr = filp->private_data;
2822 if (cnt >= sizeof(buf))
2825 if (copy_from_user(&buf, ubuf, cnt))
2830 ret = strict_strtoul(buf, 10, &val);
2839 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2841 long cpu_file = (long) inode->i_private;
2842 struct trace_iterator *iter;
2845 if (tracing_disabled)
2848 mutex_lock(&trace_types_lock);
2850 /* We only allow one reader per cpu */
2851 if (cpu_file == TRACE_PIPE_ALL_CPU) {
2852 if (!cpumask_empty(tracing_reader_cpumask)) {
2856 cpumask_setall(tracing_reader_cpumask);
2858 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2859 cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2866 /* create a buffer to store the information to pass to userspace */
2867 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2874 * We make a copy of the current tracer to avoid concurrent
2875 * changes on it while we are reading.
2877 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2883 *iter->trace = *current_trace;
2885 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2890 /* trace pipe does not show start of buffer */
2891 cpumask_setall(iter->started);
2893 if (trace_flags & TRACE_ITER_LATENCY_FMT)
2894 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2896 iter->cpu_file = cpu_file;
2897 iter->tr = &global_trace;
2898 mutex_init(&iter->mutex);
2899 filp->private_data = iter;
2901 if (iter->trace->pipe_open)
2902 iter->trace->pipe_open(iter);
2905 mutex_unlock(&trace_types_lock);
2911 mutex_unlock(&trace_types_lock);
2915 static int tracing_release_pipe(struct inode *inode, struct file *file)
2917 struct trace_iterator *iter = file->private_data;
2919 mutex_lock(&trace_types_lock);
2921 if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2922 cpumask_clear(tracing_reader_cpumask);
2924 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2927 if (iter->trace->pipe_close)
2928 iter->trace->pipe_close(iter);
2930 mutex_unlock(&trace_types_lock);
2932 free_cpumask_var(iter->started);
2933 mutex_destroy(&iter->mutex);
2941 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2943 struct trace_iterator *iter = filp->private_data;
2945 if (trace_flags & TRACE_ITER_BLOCK) {
2947 * Always select as readable when in blocking mode
2949 return POLLIN | POLLRDNORM;
2951 if (!trace_empty(iter))
2952 return POLLIN | POLLRDNORM;
2953 poll_wait(filp, &trace_wait, poll_table);
2954 if (!trace_empty(iter))
2955 return POLLIN | POLLRDNORM;
2962 void default_wait_pipe(struct trace_iterator *iter)
2966 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2968 if (trace_empty(iter))
2971 finish_wait(&trace_wait, &wait);
2975 * This is a make-shift waitqueue.
2976 * A tracer might use this callback on some rare cases:
2978 * 1) the current tracer might hold the runqueue lock when it wakes up
2979 * a reader, hence a deadlock (sched, function, and function graph tracers)
2980 * 2) the function tracers, trace all functions, we don't want
2981 * the overhead of calling wake_up and friends
2982 * (and tracing them too)
2984 * Anyway, this is really very primitive wakeup.
2986 void poll_wait_pipe(struct trace_iterator *iter)
2988 set_current_state(TASK_INTERRUPTIBLE);
2989 /* sleep for 100 msecs, and try again. */
2990 schedule_timeout(HZ / 10);
2993 /* Must be called with trace_types_lock mutex held. */
2994 static int tracing_wait_pipe(struct file *filp)
2996 struct trace_iterator *iter = filp->private_data;
2998 while (trace_empty(iter)) {
3000 if ((filp->f_flags & O_NONBLOCK)) {
3004 mutex_unlock(&iter->mutex);
3006 iter->trace->wait_pipe(iter);
3008 mutex_lock(&iter->mutex);
3010 if (signal_pending(current))
3014 * We block until we read something and tracing is disabled.
3015 * We still block if tracing is disabled, but we have never
3016 * read anything. This allows a user to cat this file, and
3017 * then enable tracing. But after we have read something,
3018 * we give an EOF when tracing is again disabled.
3020 * iter->pos will be 0 if we haven't read anything.
3022 if (!tracer_enabled && iter->pos)
3033 tracing_read_pipe(struct file *filp, char __user *ubuf,
3034 size_t cnt, loff_t *ppos)
3036 struct trace_iterator *iter = filp->private_data;
3037 static struct tracer *old_tracer;
3040 /* return any leftover data */
3041 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3045 trace_seq_init(&iter->seq);
3047 /* copy the tracer to avoid using a global lock all around */
3048 mutex_lock(&trace_types_lock);
3049 if (unlikely(old_tracer != current_trace && current_trace)) {
3050 old_tracer = current_trace;
3051 *iter->trace = *current_trace;
3053 mutex_unlock(&trace_types_lock);
3056 * Avoid more than one consumer on a single file descriptor
3057 * This is just a matter of traces coherency, the ring buffer itself
3060 mutex_lock(&iter->mutex);
3061 if (iter->trace->read) {
3062 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3068 sret = tracing_wait_pipe(filp);
3072 /* stop when tracing is finished */
3073 if (trace_empty(iter)) {
3078 if (cnt >= PAGE_SIZE)
3079 cnt = PAGE_SIZE - 1;
3081 /* reset all but tr, trace, and overruns */
3082 memset(&iter->seq, 0,
3083 sizeof(struct trace_iterator) -
3084 offsetof(struct trace_iterator, seq));
3087 trace_event_read_lock();
3088 while (find_next_entry_inc(iter) != NULL) {
3089 enum print_line_t ret;
3090 int len = iter->seq.len;
3092 ret = print_trace_line(iter);
3093 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3094 /* don't print partial lines */
3095 iter->seq.len = len;
3098 if (ret != TRACE_TYPE_NO_CONSUME)
3099 trace_consume(iter);
3101 if (iter->seq.len >= cnt)
3104 trace_event_read_unlock();
3106 /* Now copy what we have to the user */
3107 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3108 if (iter->seq.readpos >= iter->seq.len)
3109 trace_seq_init(&iter->seq);
3112 * If there was nothing to send to user, inspite of consuming trace
3113 * entries, go back to wait for more entries.
3119 mutex_unlock(&iter->mutex);
3124 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3125 struct pipe_buffer *buf)
3127 __free_page(buf->page);
3130 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3133 __free_page(spd->pages[idx]);
3136 static struct pipe_buf_operations tracing_pipe_buf_ops = {
3138 .map = generic_pipe_buf_map,
3139 .unmap = generic_pipe_buf_unmap,
3140 .confirm = generic_pipe_buf_confirm,
3141 .release = tracing_pipe_buf_release,
3142 .steal = generic_pipe_buf_steal,
3143 .get = generic_pipe_buf_get,
3147 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3152 /* Seq buffer is page-sized, exactly what we need. */
3154 count = iter->seq.len;
3155 ret = print_trace_line(iter);
3156 count = iter->seq.len - count;
3159 iter->seq.len -= count;
3162 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3163 iter->seq.len -= count;
3167 if (ret != TRACE_TYPE_NO_CONSUME)
3168 trace_consume(iter);
3170 if (!find_next_entry_inc(iter)) {
3180 static ssize_t tracing_splice_read_pipe(struct file *filp,
3182 struct pipe_inode_info *pipe,
3186 struct page *pages[PIPE_BUFFERS];
3187 struct partial_page partial[PIPE_BUFFERS];
3188 struct trace_iterator *iter = filp->private_data;
3189 struct splice_pipe_desc spd = {
3192 .nr_pages = 0, /* This gets updated below. */
3194 .ops = &tracing_pipe_buf_ops,
3195 .spd_release = tracing_spd_release_pipe,
3197 static struct tracer *old_tracer;
3202 /* copy the tracer to avoid using a global lock all around */
3203 mutex_lock(&trace_types_lock);
3204 if (unlikely(old_tracer != current_trace && current_trace)) {
3205 old_tracer = current_trace;
3206 *iter->trace = *current_trace;
3208 mutex_unlock(&trace_types_lock);
3210 mutex_lock(&iter->mutex);
3212 if (iter->trace->splice_read) {
3213 ret = iter->trace->splice_read(iter, filp,
3214 ppos, pipe, len, flags);
3219 ret = tracing_wait_pipe(filp);
3223 if (!iter->ent && !find_next_entry_inc(iter)) {
3228 trace_event_read_lock();
3230 /* Fill as many pages as possible. */
3231 for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3232 pages[i] = alloc_page(GFP_KERNEL);
3236 rem = tracing_fill_pipe_page(rem, iter);
3238 /* Copy the data into the page, so we can start over. */
3239 ret = trace_seq_to_buffer(&iter->seq,
3240 page_address(pages[i]),
3243 __free_page(pages[i]);
3246 partial[i].offset = 0;
3247 partial[i].len = iter->seq.len;
3249 trace_seq_init(&iter->seq);
3252 trace_event_read_unlock();
3253 mutex_unlock(&iter->mutex);
3257 return splice_to_pipe(pipe, &spd);
3260 mutex_unlock(&iter->mutex);
3266 tracing_entries_read(struct file *filp, char __user *ubuf,
3267 size_t cnt, loff_t *ppos)
3269 struct trace_array *tr = filp->private_data;
3273 mutex_lock(&trace_types_lock);
3274 if (!ring_buffer_expanded)
3275 r = sprintf(buf, "%lu (expanded: %lu)\n",
3277 trace_buf_size >> 10);
3279 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3280 mutex_unlock(&trace_types_lock);
3282 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3286 tracing_entries_write(struct file *filp, const char __user *ubuf,
3287 size_t cnt, loff_t *ppos)
3293 if (cnt >= sizeof(buf))
3296 if (copy_from_user(&buf, ubuf, cnt))
3301 ret = strict_strtoul(buf, 10, &val);
3305 /* must have at least 1 entry */
3309 mutex_lock(&trace_types_lock);
3313 /* disable all cpu buffers */
3314 for_each_tracing_cpu(cpu) {
3315 if (global_trace.data[cpu])
3316 atomic_inc(&global_trace.data[cpu]->disabled);
3317 if (max_tr.data[cpu])
3318 atomic_inc(&max_tr.data[cpu]->disabled);
3321 /* value is in KB */
3324 if (val != global_trace.entries) {
3325 ret = tracing_resize_ring_buffer(val);
3334 /* If check pages failed, return ENOMEM */
3335 if (tracing_disabled)
3338 for_each_tracing_cpu(cpu) {
3339 if (global_trace.data[cpu])
3340 atomic_dec(&global_trace.data[cpu]->disabled);
3341 if (max_tr.data[cpu])
3342 atomic_dec(&max_tr.data[cpu]->disabled);
3346 max_tr.entries = global_trace.entries;
3347 mutex_unlock(&trace_types_lock);
3352 static int mark_printk(const char *fmt, ...)
3356 va_start(args, fmt);
3357 ret = trace_vprintk(0, fmt, args);
3363 tracing_mark_write(struct file *filp, const char __user *ubuf,
3364 size_t cnt, loff_t *fpos)
3368 if (tracing_disabled)
3371 if (cnt > TRACE_BUF_SIZE)
3372 cnt = TRACE_BUF_SIZE;
3374 buf = kmalloc(cnt + 2, GFP_KERNEL);
3378 if (copy_from_user(buf, ubuf, cnt)) {
3382 if (buf[cnt-1] != '\n') {
3388 cnt = mark_printk("%s", buf);
3395 static ssize_t tracing_clock_read(struct file *filp, char __user *ubuf,
3396 size_t cnt, loff_t *ppos)
3402 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3403 bufiter += snprintf(buf + bufiter, sizeof(buf) - bufiter,
3404 "%s%s%s%s", i ? " " : "",
3405 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3406 i == trace_clock_id ? "]" : "");
3407 bufiter += snprintf(buf + bufiter, sizeof(buf) - bufiter, "\n");
3409 return simple_read_from_buffer(ubuf, cnt, ppos, buf, bufiter);
3412 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3413 size_t cnt, loff_t *fpos)
3416 const char *clockstr;
3419 if (cnt >= sizeof(buf))
3422 if (copy_from_user(&buf, ubuf, cnt))
3427 clockstr = strstrip(buf);
3429 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3430 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3433 if (i == ARRAY_SIZE(trace_clocks))
3438 mutex_lock(&trace_types_lock);
3440 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3442 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3444 mutex_unlock(&trace_types_lock);
3451 static const struct file_operations tracing_max_lat_fops = {
3452 .open = tracing_open_generic,
3453 .read = tracing_max_lat_read,
3454 .write = tracing_max_lat_write,
3457 static const struct file_operations tracing_ctrl_fops = {
3458 .open = tracing_open_generic,
3459 .read = tracing_ctrl_read,
3460 .write = tracing_ctrl_write,
3463 static const struct file_operations set_tracer_fops = {
3464 .open = tracing_open_generic,
3465 .read = tracing_set_trace_read,
3466 .write = tracing_set_trace_write,
3469 static const struct file_operations tracing_pipe_fops = {
3470 .open = tracing_open_pipe,
3471 .poll = tracing_poll_pipe,
3472 .read = tracing_read_pipe,
3473 .splice_read = tracing_splice_read_pipe,
3474 .release = tracing_release_pipe,
3477 static const struct file_operations tracing_entries_fops = {
3478 .open = tracing_open_generic,
3479 .read = tracing_entries_read,
3480 .write = tracing_entries_write,
3483 static const struct file_operations tracing_mark_fops = {
3484 .open = tracing_open_generic,
3485 .write = tracing_mark_write,
3488 static const struct file_operations trace_clock_fops = {
3489 .open = tracing_open_generic,
3490 .read = tracing_clock_read,
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 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);
3954 /* do nothing if already cleared */
3955 if (!(topt->flags->val & topt->opt->bit))
3958 mutex_lock(&trace_types_lock);
3959 if (current_trace->set_flag)
3960 ret = current_trace->set_flag(topt->flags->val,
3962 mutex_unlock(&trace_types_lock);
3965 topt->flags->val &= ~topt->opt->bit;
3968 /* do nothing if already set */
3969 if (topt->flags->val & topt->opt->bit)
3972 mutex_lock(&trace_types_lock);
3973 if (current_trace->set_flag)
3974 ret = current_trace->set_flag(topt->flags->val,
3976 mutex_unlock(&trace_types_lock);
3979 topt->flags->val |= topt->opt->bit;
3992 static const struct file_operations trace_options_fops = {
3993 .open = tracing_open_generic,
3994 .read = trace_options_read,
3995 .write = trace_options_write,
3999 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4002 long index = (long)filp->private_data;
4005 if (trace_flags & (1 << index))
4010 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4014 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4017 long index = (long)filp->private_data;
4022 if (cnt >= sizeof(buf))
4025 if (copy_from_user(&buf, ubuf, cnt))
4030 ret = strict_strtoul(buf, 10, &val);
4034 if (val != 0 && val != 1)
4036 set_tracer_flags(1 << index, val);
4043 static const struct file_operations trace_options_core_fops = {
4044 .open = tracing_open_generic,
4045 .read = trace_options_core_read,
4046 .write = trace_options_core_write,
4049 struct dentry *trace_create_file(const char *name,
4051 struct dentry *parent,
4053 const struct file_operations *fops)
4057 ret = debugfs_create_file(name, mode, parent, data, fops);
4059 pr_warning("Could not create debugfs '%s' entry\n", name);
4065 static struct dentry *trace_options_init_dentry(void)
4067 struct dentry *d_tracer;
4068 static struct dentry *t_options;
4073 d_tracer = tracing_init_dentry();
4077 t_options = debugfs_create_dir("options", d_tracer);
4079 pr_warning("Could not create debugfs directory 'options'\n");
4087 create_trace_option_file(struct trace_option_dentry *topt,
4088 struct tracer_flags *flags,
4089 struct tracer_opt *opt)
4091 struct dentry *t_options;
4093 t_options = trace_options_init_dentry();
4097 topt->flags = flags;
4100 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4101 &trace_options_fops);
4105 static struct trace_option_dentry *
4106 create_trace_option_files(struct tracer *tracer)
4108 struct trace_option_dentry *topts;
4109 struct tracer_flags *flags;
4110 struct tracer_opt *opts;
4116 flags = tracer->flags;
4118 if (!flags || !flags->opts)
4123 for (cnt = 0; opts[cnt].name; cnt++)
4126 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4130 for (cnt = 0; opts[cnt].name; cnt++)
4131 create_trace_option_file(&topts[cnt], flags,
4138 destroy_trace_option_files(struct trace_option_dentry *topts)
4145 for (cnt = 0; topts[cnt].opt; cnt++) {
4146 if (topts[cnt].entry)
4147 debugfs_remove(topts[cnt].entry);
4153 static struct dentry *
4154 create_trace_option_core_file(const char *option, long index)
4156 struct dentry *t_options;
4158 t_options = trace_options_init_dentry();
4162 return trace_create_file(option, 0644, t_options, (void *)index,
4163 &trace_options_core_fops);
4166 static __init void create_trace_options_dir(void)
4168 struct dentry *t_options;
4171 t_options = trace_options_init_dentry();
4175 for (i = 0; trace_options[i]; i++)
4176 create_trace_option_core_file(trace_options[i], i);
4179 static __init int tracer_init_debugfs(void)
4181 struct dentry *d_tracer;
4184 d_tracer = tracing_init_dentry();
4186 trace_create_file("tracing_enabled", 0644, d_tracer,
4187 &global_trace, &tracing_ctrl_fops);
4189 trace_create_file("trace_options", 0644, d_tracer,
4190 NULL, &tracing_iter_fops);
4192 trace_create_file("tracing_cpumask", 0644, d_tracer,
4193 NULL, &tracing_cpumask_fops);
4195 trace_create_file("trace", 0644, d_tracer,
4196 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4198 trace_create_file("available_tracers", 0444, d_tracer,
4199 &global_trace, &show_traces_fops);
4201 trace_create_file("current_tracer", 0644, d_tracer,
4202 &global_trace, &set_tracer_fops);
4204 #ifdef CONFIG_TRACER_MAX_TRACE
4205 trace_create_file("tracing_max_latency", 0644, d_tracer,
4206 &tracing_max_latency, &tracing_max_lat_fops);
4208 trace_create_file("tracing_thresh", 0644, d_tracer,
4209 &tracing_thresh, &tracing_max_lat_fops);
4212 trace_create_file("README", 0444, d_tracer,
4213 NULL, &tracing_readme_fops);
4215 trace_create_file("trace_pipe", 0444, d_tracer,
4216 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4218 trace_create_file("buffer_size_kb", 0644, d_tracer,
4219 &global_trace, &tracing_entries_fops);
4221 trace_create_file("trace_marker", 0220, d_tracer,
4222 NULL, &tracing_mark_fops);
4224 trace_create_file("saved_cmdlines", 0444, d_tracer,
4225 NULL, &tracing_saved_cmdlines_fops);
4227 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4230 #ifdef CONFIG_DYNAMIC_FTRACE
4231 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4232 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4234 #ifdef CONFIG_SYSPROF_TRACER
4235 init_tracer_sysprof_debugfs(d_tracer);
4238 create_trace_options_dir();
4240 for_each_tracing_cpu(cpu)
4241 tracing_init_debugfs_percpu(cpu);
4246 static int trace_panic_handler(struct notifier_block *this,
4247 unsigned long event, void *unused)
4249 if (ftrace_dump_on_oops)
4254 static struct notifier_block trace_panic_notifier = {
4255 .notifier_call = trace_panic_handler,
4257 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4260 static int trace_die_handler(struct notifier_block *self,
4266 if (ftrace_dump_on_oops)
4275 static struct notifier_block trace_die_notifier = {
4276 .notifier_call = trace_die_handler,
4281 * printk is set to max of 1024, we really don't need it that big.
4282 * Nothing should be printing 1000 characters anyway.
4284 #define TRACE_MAX_PRINT 1000
4287 * Define here KERN_TRACE so that we have one place to modify
4288 * it if we decide to change what log level the ftrace dump
4291 #define KERN_TRACE KERN_EMERG
4294 trace_printk_seq(struct trace_seq *s)
4296 /* Probably should print a warning here. */
4300 /* should be zero ended, but we are paranoid. */
4301 s->buffer[s->len] = 0;
4303 printk(KERN_TRACE "%s", s->buffer);
4308 static void __ftrace_dump(bool disable_tracing)
4310 static arch_spinlock_t ftrace_dump_lock =
4311 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4312 /* use static because iter can be a bit big for the stack */
4313 static struct trace_iterator iter;
4314 unsigned int old_userobj;
4315 static int dump_ran;
4316 unsigned long flags;
4320 local_irq_save(flags);
4321 arch_spin_lock(&ftrace_dump_lock);
4329 if (disable_tracing)
4332 for_each_tracing_cpu(cpu) {
4333 atomic_inc(&global_trace.data[cpu]->disabled);
4336 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4338 /* don't look at user memory in panic mode */
4339 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4341 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4343 /* Simulate the iterator */
4344 iter.tr = &global_trace;
4345 iter.trace = current_trace;
4346 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4349 * We need to stop all tracing on all CPUS to read the
4350 * the next buffer. This is a bit expensive, but is
4351 * not done often. We fill all what we can read,
4352 * and then release the locks again.
4355 while (!trace_empty(&iter)) {
4358 printk(KERN_TRACE "---------------------------------\n");
4362 /* reset all but tr, trace, and overruns */
4363 memset(&iter.seq, 0,
4364 sizeof(struct trace_iterator) -
4365 offsetof(struct trace_iterator, seq));
4366 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4369 if (find_next_entry_inc(&iter) != NULL) {
4372 ret = print_trace_line(&iter);
4373 if (ret != TRACE_TYPE_NO_CONSUME)
4374 trace_consume(&iter);
4377 trace_printk_seq(&iter.seq);
4381 printk(KERN_TRACE " (ftrace buffer empty)\n");
4383 printk(KERN_TRACE "---------------------------------\n");
4385 /* Re-enable tracing if requested */
4386 if (!disable_tracing) {
4387 trace_flags |= old_userobj;
4389 for_each_tracing_cpu(cpu) {
4390 atomic_dec(&global_trace.data[cpu]->disabled);
4396 arch_spin_unlock(&ftrace_dump_lock);
4397 local_irq_restore(flags);
4400 /* By default: disable tracing after the dump */
4401 void ftrace_dump(void)
4403 __ftrace_dump(true);
4406 __init static int tracer_alloc_buffers(void)
4412 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4415 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4416 goto out_free_buffer_mask;
4418 if (!zalloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
4419 goto out_free_tracing_cpumask;
4421 /* To save memory, keep the ring buffer size to its minimum */
4422 if (ring_buffer_expanded)
4423 ring_buf_size = trace_buf_size;
4427 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4428 cpumask_copy(tracing_cpumask, cpu_all_mask);
4430 /* TODO: make the number of buffers hot pluggable with CPUS */
4431 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4432 TRACE_BUFFER_FLAGS);
4433 if (!global_trace.buffer) {
4434 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4436 goto out_free_cpumask;
4438 global_trace.entries = ring_buffer_size(global_trace.buffer);
4441 #ifdef CONFIG_TRACER_MAX_TRACE
4442 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4443 TRACE_BUFFER_FLAGS);
4444 if (!max_tr.buffer) {
4445 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4447 ring_buffer_free(global_trace.buffer);
4448 goto out_free_cpumask;
4450 max_tr.entries = ring_buffer_size(max_tr.buffer);
4451 WARN_ON(max_tr.entries != global_trace.entries);
4454 /* Allocate the first page for all buffers */
4455 for_each_tracing_cpu(i) {
4456 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4457 max_tr.data[i] = &per_cpu(max_tr_data, i);
4460 trace_init_cmdlines();
4462 register_tracer(&nop_trace);
4463 current_trace = &nop_trace;
4464 #ifdef CONFIG_BOOT_TRACER
4465 register_tracer(&boot_tracer);
4467 /* All seems OK, enable tracing */
4468 tracing_disabled = 0;
4470 atomic_notifier_chain_register(&panic_notifier_list,
4471 &trace_panic_notifier);
4473 register_die_notifier(&trace_die_notifier);
4478 free_cpumask_var(tracing_reader_cpumask);
4479 out_free_tracing_cpumask:
4480 free_cpumask_var(tracing_cpumask);
4481 out_free_buffer_mask:
4482 free_cpumask_var(tracing_buffer_mask);
4487 __init static int clear_boot_tracer(void)
4490 * The default tracer at boot buffer is an init section.
4491 * This function is called in lateinit. If we did not
4492 * find the boot tracer, then clear it out, to prevent
4493 * later registration from accessing the buffer that is
4494 * about to be freed.
4496 if (!default_bootup_tracer)
4499 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4500 default_bootup_tracer);
4501 default_bootup_tracer = NULL;
4506 early_initcall(tracer_alloc_buffers);
4507 fs_initcall(tracer_init_debugfs);
4508 late_initcall(clear_boot_tracer);