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 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 raw_spinlock_t trace_buf_lock =
1255 (raw_spinlock_t)__RAW_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 __raw_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 __raw_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 raw_spinlock_t trace_buf_lock = __RAW_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 __raw_spin_lock(&trace_buf_lock);
1365 strncpy(trace_buf, fmt, TRACE_BUF_SIZE);
1366 len = strlen(trace_buf);
1368 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1370 size = sizeof(*entry) + len + 1;
1371 buffer = tr->buffer;
1372 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1376 entry = ring_buffer_event_data(event);
1379 memcpy(&entry->buf, trace_buf, len);
1380 entry->buf[len] = '\0';
1381 if (!filter_check_discard(call, entry, buffer, event))
1382 ring_buffer_unlock_commit(buffer, event);
1385 __raw_spin_unlock(&trace_buf_lock);
1386 raw_local_irq_restore(irq_flags);
1387 unpause_graph_tracing();
1389 atomic_dec_return(&data->disabled);
1390 preempt_enable_notrace();
1395 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1397 return trace_array_vprintk(&global_trace, ip, fmt, args);
1399 EXPORT_SYMBOL_GPL(trace_vprintk);
1401 enum trace_file_type {
1402 TRACE_FILE_LAT_FMT = 1,
1403 TRACE_FILE_ANNOTATE = 2,
1406 static void trace_iterator_increment(struct trace_iterator *iter)
1408 /* Don't allow ftrace to trace into the ring buffers */
1409 ftrace_disable_cpu();
1412 if (iter->buffer_iter[iter->cpu])
1413 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1415 ftrace_enable_cpu();
1418 static struct trace_entry *
1419 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1421 struct ring_buffer_event *event;
1422 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1424 /* Don't allow ftrace to trace into the ring buffers */
1425 ftrace_disable_cpu();
1428 event = ring_buffer_iter_peek(buf_iter, ts);
1430 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1432 ftrace_enable_cpu();
1434 return event ? ring_buffer_event_data(event) : NULL;
1437 static struct trace_entry *
1438 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1440 struct ring_buffer *buffer = iter->tr->buffer;
1441 struct trace_entry *ent, *next = NULL;
1442 int cpu_file = iter->cpu_file;
1443 u64 next_ts = 0, ts;
1448 * If we are in a per_cpu trace file, don't bother by iterating over
1449 * all cpu and peek directly.
1451 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1452 if (ring_buffer_empty_cpu(buffer, cpu_file))
1454 ent = peek_next_entry(iter, cpu_file, ent_ts);
1456 *ent_cpu = cpu_file;
1461 for_each_tracing_cpu(cpu) {
1463 if (ring_buffer_empty_cpu(buffer, cpu))
1466 ent = peek_next_entry(iter, cpu, &ts);
1469 * Pick the entry with the smallest timestamp:
1471 if (ent && (!next || ts < next_ts)) {
1479 *ent_cpu = next_cpu;
1487 /* Find the next real entry, without updating the iterator itself */
1488 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1489 int *ent_cpu, u64 *ent_ts)
1491 return __find_next_entry(iter, ent_cpu, ent_ts);
1494 /* Find the next real entry, and increment the iterator to the next entry */
1495 static void *find_next_entry_inc(struct trace_iterator *iter)
1497 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1500 trace_iterator_increment(iter);
1502 return iter->ent ? iter : NULL;
1505 static void trace_consume(struct trace_iterator *iter)
1507 /* Don't allow ftrace to trace into the ring buffers */
1508 ftrace_disable_cpu();
1509 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1510 ftrace_enable_cpu();
1513 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1515 struct trace_iterator *iter = m->private;
1521 /* can't go backwards */
1526 ent = find_next_entry_inc(iter);
1530 while (ent && iter->idx < i)
1531 ent = find_next_entry_inc(iter);
1538 static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1540 struct trace_array *tr = iter->tr;
1541 struct ring_buffer_event *event;
1542 struct ring_buffer_iter *buf_iter;
1543 unsigned long entries = 0;
1546 tr->data[cpu]->skipped_entries = 0;
1548 if (!iter->buffer_iter[cpu])
1551 buf_iter = iter->buffer_iter[cpu];
1552 ring_buffer_iter_reset(buf_iter);
1555 * We could have the case with the max latency tracers
1556 * that a reset never took place on a cpu. This is evident
1557 * by the timestamp being before the start of the buffer.
1559 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1560 if (ts >= iter->tr->time_start)
1563 ring_buffer_read(buf_iter, NULL);
1566 tr->data[cpu]->skipped_entries = entries;
1570 * No necessary locking here. The worst thing which can
1571 * happen is loosing events consumed at the same time
1572 * by a trace_pipe reader.
1573 * Other than that, we don't risk to crash the ring buffer
1574 * because it serializes the readers.
1576 * The current tracer is copied to avoid a global locking
1579 static void *s_start(struct seq_file *m, loff_t *pos)
1581 struct trace_iterator *iter = m->private;
1582 static struct tracer *old_tracer;
1583 int cpu_file = iter->cpu_file;
1588 /* copy the tracer to avoid using a global lock all around */
1589 mutex_lock(&trace_types_lock);
1590 if (unlikely(old_tracer != current_trace && current_trace)) {
1591 old_tracer = current_trace;
1592 *iter->trace = *current_trace;
1594 mutex_unlock(&trace_types_lock);
1596 atomic_inc(&trace_record_cmdline_disabled);
1598 if (*pos != iter->pos) {
1603 ftrace_disable_cpu();
1605 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1606 for_each_tracing_cpu(cpu)
1607 tracing_iter_reset(iter, cpu);
1609 tracing_iter_reset(iter, cpu_file);
1611 ftrace_enable_cpu();
1613 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1618 p = s_next(m, p, &l);
1621 trace_event_read_lock();
1625 static void s_stop(struct seq_file *m, void *p)
1627 atomic_dec(&trace_record_cmdline_disabled);
1628 trace_event_read_unlock();
1631 static void print_lat_help_header(struct seq_file *m)
1633 seq_puts(m, "# _------=> CPU# \n");
1634 seq_puts(m, "# / _-----=> irqs-off \n");
1635 seq_puts(m, "# | / _----=> need-resched \n");
1636 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1637 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1638 seq_puts(m, "# |||| /_--=> lock-depth \n");
1639 seq_puts(m, "# |||||/ delay \n");
1640 seq_puts(m, "# cmd pid |||||| time | caller \n");
1641 seq_puts(m, "# \\ / |||||| \\ | / \n");
1644 static void print_func_help_header(struct seq_file *m)
1646 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1647 seq_puts(m, "# | | | | |\n");
1652 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1654 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1655 struct trace_array *tr = iter->tr;
1656 struct trace_array_cpu *data = tr->data[tr->cpu];
1657 struct tracer *type = current_trace;
1658 unsigned long entries = 0;
1659 unsigned long total = 0;
1660 unsigned long count;
1661 const char *name = "preemption";
1668 for_each_tracing_cpu(cpu) {
1669 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1671 * If this buffer has skipped entries, then we hold all
1672 * entries for the trace and we need to ignore the
1673 * ones before the time stamp.
1675 if (tr->data[cpu]->skipped_entries) {
1676 count -= tr->data[cpu]->skipped_entries;
1677 /* total is the same as the entries */
1681 ring_buffer_overrun_cpu(tr->buffer, cpu);
1685 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1687 seq_puts(m, "# -----------------------------------"
1688 "---------------------------------\n");
1689 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1690 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1691 nsecs_to_usecs(data->saved_latency),
1695 #if defined(CONFIG_PREEMPT_NONE)
1697 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1699 #elif defined(CONFIG_PREEMPT)
1704 /* These are reserved for later use */
1707 seq_printf(m, " #P:%d)\n", num_online_cpus());
1711 seq_puts(m, "# -----------------\n");
1712 seq_printf(m, "# | task: %.16s-%d "
1713 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1714 data->comm, data->pid, data->uid, data->nice,
1715 data->policy, data->rt_priority);
1716 seq_puts(m, "# -----------------\n");
1718 if (data->critical_start) {
1719 seq_puts(m, "# => started at: ");
1720 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1721 trace_print_seq(m, &iter->seq);
1722 seq_puts(m, "\n# => ended at: ");
1723 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1724 trace_print_seq(m, &iter->seq);
1725 seq_puts(m, "\n#\n");
1731 static void test_cpu_buff_start(struct trace_iterator *iter)
1733 struct trace_seq *s = &iter->seq;
1735 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1738 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1741 if (cpumask_test_cpu(iter->cpu, iter->started))
1744 if (iter->tr->data[iter->cpu]->skipped_entries)
1747 cpumask_set_cpu(iter->cpu, iter->started);
1749 /* Don't print started cpu buffer for the first entry of the trace */
1751 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1755 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1757 struct trace_seq *s = &iter->seq;
1758 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1759 struct trace_entry *entry;
1760 struct trace_event *event;
1764 test_cpu_buff_start(iter);
1766 event = ftrace_find_event(entry->type);
1768 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1769 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1770 if (!trace_print_lat_context(iter))
1773 if (!trace_print_context(iter))
1779 return event->trace(iter, sym_flags);
1781 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1784 return TRACE_TYPE_HANDLED;
1786 return TRACE_TYPE_PARTIAL_LINE;
1789 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1791 struct trace_seq *s = &iter->seq;
1792 struct trace_entry *entry;
1793 struct trace_event *event;
1797 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1798 if (!trace_seq_printf(s, "%d %d %llu ",
1799 entry->pid, iter->cpu, iter->ts))
1803 event = ftrace_find_event(entry->type);
1805 return event->raw(iter, 0);
1807 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1810 return TRACE_TYPE_HANDLED;
1812 return TRACE_TYPE_PARTIAL_LINE;
1815 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1817 struct trace_seq *s = &iter->seq;
1818 unsigned char newline = '\n';
1819 struct trace_entry *entry;
1820 struct trace_event *event;
1824 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1825 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1826 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1827 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1830 event = ftrace_find_event(entry->type);
1832 enum print_line_t ret = event->hex(iter, 0);
1833 if (ret != TRACE_TYPE_HANDLED)
1837 SEQ_PUT_FIELD_RET(s, newline);
1839 return TRACE_TYPE_HANDLED;
1842 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1844 struct trace_seq *s = &iter->seq;
1845 struct trace_entry *entry;
1846 struct trace_event *event;
1850 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1851 SEQ_PUT_FIELD_RET(s, entry->pid);
1852 SEQ_PUT_FIELD_RET(s, iter->cpu);
1853 SEQ_PUT_FIELD_RET(s, iter->ts);
1856 event = ftrace_find_event(entry->type);
1857 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1860 static int trace_empty(struct trace_iterator *iter)
1864 /* If we are looking at one CPU buffer, only check that one */
1865 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1866 cpu = iter->cpu_file;
1867 if (iter->buffer_iter[cpu]) {
1868 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1871 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1877 for_each_tracing_cpu(cpu) {
1878 if (iter->buffer_iter[cpu]) {
1879 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1882 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1890 /* Called with trace_event_read_lock() held. */
1891 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1893 enum print_line_t ret;
1895 if (iter->trace && iter->trace->print_line) {
1896 ret = iter->trace->print_line(iter);
1897 if (ret != TRACE_TYPE_UNHANDLED)
1901 if (iter->ent->type == TRACE_BPRINT &&
1902 trace_flags & TRACE_ITER_PRINTK &&
1903 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1904 return trace_print_bprintk_msg_only(iter);
1906 if (iter->ent->type == TRACE_PRINT &&
1907 trace_flags & TRACE_ITER_PRINTK &&
1908 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1909 return trace_print_printk_msg_only(iter);
1911 if (trace_flags & TRACE_ITER_BIN)
1912 return print_bin_fmt(iter);
1914 if (trace_flags & TRACE_ITER_HEX)
1915 return print_hex_fmt(iter);
1917 if (trace_flags & TRACE_ITER_RAW)
1918 return print_raw_fmt(iter);
1920 return print_trace_fmt(iter);
1923 static int s_show(struct seq_file *m, void *v)
1925 struct trace_iterator *iter = v;
1927 if (iter->ent == NULL) {
1929 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1932 if (iter->trace && iter->trace->print_header)
1933 iter->trace->print_header(m);
1934 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1935 /* print nothing if the buffers are empty */
1936 if (trace_empty(iter))
1938 print_trace_header(m, iter);
1939 if (!(trace_flags & TRACE_ITER_VERBOSE))
1940 print_lat_help_header(m);
1942 if (!(trace_flags & TRACE_ITER_VERBOSE))
1943 print_func_help_header(m);
1946 print_trace_line(iter);
1947 trace_print_seq(m, &iter->seq);
1953 static const struct seq_operations tracer_seq_ops = {
1960 static struct trace_iterator *
1961 __tracing_open(struct inode *inode, struct file *file)
1963 long cpu_file = (long) inode->i_private;
1964 void *fail_ret = ERR_PTR(-ENOMEM);
1965 struct trace_iterator *iter;
1969 if (tracing_disabled)
1970 return ERR_PTR(-ENODEV);
1972 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1974 return ERR_PTR(-ENOMEM);
1977 * We make a copy of the current tracer to avoid concurrent
1978 * changes on it while we are reading.
1980 mutex_lock(&trace_types_lock);
1981 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
1986 *iter->trace = *current_trace;
1988 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
1991 if (current_trace && current_trace->print_max)
1994 iter->tr = &global_trace;
1996 mutex_init(&iter->mutex);
1997 iter->cpu_file = cpu_file;
1999 /* Notify the tracer early; before we stop tracing. */
2000 if (iter->trace && iter->trace->open)
2001 iter->trace->open(iter);
2003 /* Annotate start of buffers if we had overruns */
2004 if (ring_buffer_overruns(iter->tr->buffer))
2005 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2007 /* stop the trace while dumping */
2010 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2011 for_each_tracing_cpu(cpu) {
2013 iter->buffer_iter[cpu] =
2014 ring_buffer_read_start(iter->tr->buffer, cpu);
2015 tracing_iter_reset(iter, cpu);
2018 cpu = iter->cpu_file;
2019 iter->buffer_iter[cpu] =
2020 ring_buffer_read_start(iter->tr->buffer, cpu);
2021 tracing_iter_reset(iter, cpu);
2024 ret = seq_open(file, &tracer_seq_ops);
2026 fail_ret = ERR_PTR(ret);
2030 m = file->private_data;
2033 mutex_unlock(&trace_types_lock);
2038 for_each_tracing_cpu(cpu) {
2039 if (iter->buffer_iter[cpu])
2040 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2042 free_cpumask_var(iter->started);
2045 mutex_unlock(&trace_types_lock);
2052 int tracing_open_generic(struct inode *inode, struct file *filp)
2054 if (tracing_disabled)
2057 filp->private_data = inode->i_private;
2061 static int tracing_release(struct inode *inode, struct file *file)
2063 struct seq_file *m = (struct seq_file *)file->private_data;
2064 struct trace_iterator *iter;
2067 if (!(file->f_mode & FMODE_READ))
2072 mutex_lock(&trace_types_lock);
2073 for_each_tracing_cpu(cpu) {
2074 if (iter->buffer_iter[cpu])
2075 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2078 if (iter->trace && iter->trace->close)
2079 iter->trace->close(iter);
2081 /* reenable tracing if it was previously enabled */
2083 mutex_unlock(&trace_types_lock);
2085 seq_release(inode, file);
2086 mutex_destroy(&iter->mutex);
2087 free_cpumask_var(iter->started);
2093 static int tracing_open(struct inode *inode, struct file *file)
2095 struct trace_iterator *iter;
2098 /* If this file was open for write, then erase contents */
2099 if ((file->f_mode & FMODE_WRITE) &&
2100 (file->f_flags & O_TRUNC)) {
2101 long cpu = (long) inode->i_private;
2103 if (cpu == TRACE_PIPE_ALL_CPU)
2104 tracing_reset_online_cpus(&global_trace);
2106 tracing_reset(&global_trace, cpu);
2109 if (file->f_mode & FMODE_READ) {
2110 iter = __tracing_open(inode, file);
2112 ret = PTR_ERR(iter);
2113 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2114 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2120 t_next(struct seq_file *m, void *v, loff_t *pos)
2122 struct tracer *t = v;
2132 static void *t_start(struct seq_file *m, loff_t *pos)
2137 mutex_lock(&trace_types_lock);
2138 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2144 static void t_stop(struct seq_file *m, void *p)
2146 mutex_unlock(&trace_types_lock);
2149 static int t_show(struct seq_file *m, void *v)
2151 struct tracer *t = v;
2156 seq_printf(m, "%s", t->name);
2165 static const struct seq_operations show_traces_seq_ops = {
2172 static int show_traces_open(struct inode *inode, struct file *file)
2174 if (tracing_disabled)
2177 return seq_open(file, &show_traces_seq_ops);
2181 tracing_write_stub(struct file *filp, const char __user *ubuf,
2182 size_t count, loff_t *ppos)
2187 static const struct file_operations tracing_fops = {
2188 .open = tracing_open,
2190 .write = tracing_write_stub,
2191 .llseek = seq_lseek,
2192 .release = tracing_release,
2195 static const struct file_operations show_traces_fops = {
2196 .open = show_traces_open,
2198 .release = seq_release,
2202 * Only trace on a CPU if the bitmask is set:
2204 static cpumask_var_t tracing_cpumask;
2207 * The tracer itself will not take this lock, but still we want
2208 * to provide a consistent cpumask to user-space:
2210 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2213 * Temporary storage for the character representation of the
2214 * CPU bitmask (and one more byte for the newline):
2216 static char mask_str[NR_CPUS + 1];
2219 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2220 size_t count, loff_t *ppos)
2224 mutex_lock(&tracing_cpumask_update_lock);
2226 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2227 if (count - len < 2) {
2231 len += sprintf(mask_str + len, "\n");
2232 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2235 mutex_unlock(&tracing_cpumask_update_lock);
2241 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2242 size_t count, loff_t *ppos)
2245 cpumask_var_t tracing_cpumask_new;
2247 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2250 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2254 mutex_lock(&tracing_cpumask_update_lock);
2256 local_irq_disable();
2257 __raw_spin_lock(&ftrace_max_lock);
2258 for_each_tracing_cpu(cpu) {
2260 * Increase/decrease the disabled counter if we are
2261 * about to flip a bit in the cpumask:
2263 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2264 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2265 atomic_inc(&global_trace.data[cpu]->disabled);
2267 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2268 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2269 atomic_dec(&global_trace.data[cpu]->disabled);
2272 __raw_spin_unlock(&ftrace_max_lock);
2275 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2277 mutex_unlock(&tracing_cpumask_update_lock);
2278 free_cpumask_var(tracing_cpumask_new);
2283 free_cpumask_var(tracing_cpumask_new);
2288 static const struct file_operations tracing_cpumask_fops = {
2289 .open = tracing_open_generic,
2290 .read = tracing_cpumask_read,
2291 .write = tracing_cpumask_write,
2295 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2296 size_t cnt, loff_t *ppos)
2298 struct tracer_opt *trace_opts;
2306 /* calculate max size */
2307 for (i = 0; trace_options[i]; i++) {
2308 len += strlen(trace_options[i]);
2309 len += 3; /* "no" and newline */
2312 mutex_lock(&trace_types_lock);
2313 tracer_flags = current_trace->flags->val;
2314 trace_opts = current_trace->flags->opts;
2317 * Increase the size with names of options specific
2318 * of the current tracer.
2320 for (i = 0; trace_opts[i].name; i++) {
2321 len += strlen(trace_opts[i].name);
2322 len += 3; /* "no" and newline */
2326 buf = kmalloc(len + 1, GFP_KERNEL);
2328 mutex_unlock(&trace_types_lock);
2332 for (i = 0; trace_options[i]; i++) {
2333 if (trace_flags & (1 << i))
2334 r += sprintf(buf + r, "%s\n", trace_options[i]);
2336 r += sprintf(buf + r, "no%s\n", trace_options[i]);
2339 for (i = 0; trace_opts[i].name; i++) {
2340 if (tracer_flags & trace_opts[i].bit)
2341 r += sprintf(buf + r, "%s\n",
2342 trace_opts[i].name);
2344 r += sprintf(buf + r, "no%s\n",
2345 trace_opts[i].name);
2347 mutex_unlock(&trace_types_lock);
2349 WARN_ON(r >= len + 1);
2351 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2357 /* Try to assign a tracer specific option */
2358 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2360 struct tracer_flags *tracer_flags = trace->flags;
2361 struct tracer_opt *opts = NULL;
2365 for (i = 0; tracer_flags->opts[i].name; i++) {
2366 opts = &tracer_flags->opts[i];
2367 len = strlen(opts->name);
2369 if (strncmp(cmp, opts->name, len) == 0) {
2370 ret = trace->set_flag(tracer_flags->val,
2376 if (!tracer_flags->opts[i].name)
2379 /* Refused to handle */
2384 tracer_flags->val &= ~opts->bit;
2386 tracer_flags->val |= opts->bit;
2391 static void set_tracer_flags(unsigned int mask, int enabled)
2393 /* do nothing if flag is already set */
2394 if (!!(trace_flags & mask) == !!enabled)
2398 trace_flags |= mask;
2400 trace_flags &= ~mask;
2404 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2405 size_t cnt, loff_t *ppos)
2413 if (cnt >= sizeof(buf))
2416 if (copy_from_user(&buf, ubuf, cnt))
2421 if (strncmp(buf, "no", 2) == 0) {
2426 for (i = 0; trace_options[i]; i++) {
2427 int len = strlen(trace_options[i]);
2429 if (strncmp(cmp, trace_options[i], len) == 0) {
2430 set_tracer_flags(1 << i, !neg);
2435 /* If no option could be set, test the specific tracer options */
2436 if (!trace_options[i]) {
2437 mutex_lock(&trace_types_lock);
2438 ret = set_tracer_option(current_trace, cmp, neg);
2439 mutex_unlock(&trace_types_lock);
2449 static const struct file_operations tracing_iter_fops = {
2450 .open = tracing_open_generic,
2451 .read = tracing_trace_options_read,
2452 .write = tracing_trace_options_write,
2455 static const char readme_msg[] =
2456 "tracing mini-HOWTO:\n\n"
2457 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2458 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2459 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2460 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2462 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2463 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2465 "# cat /sys/kernel/debug/tracing/trace_options\n"
2466 "noprint-parent nosym-offset nosym-addr noverbose\n"
2467 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2468 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2469 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2470 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2474 tracing_readme_read(struct file *filp, char __user *ubuf,
2475 size_t cnt, loff_t *ppos)
2477 return simple_read_from_buffer(ubuf, cnt, ppos,
2478 readme_msg, strlen(readme_msg));
2481 static const struct file_operations tracing_readme_fops = {
2482 .open = tracing_open_generic,
2483 .read = tracing_readme_read,
2487 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2488 size_t cnt, loff_t *ppos)
2497 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2501 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2509 for (i = 0; i < SAVED_CMDLINES; i++) {
2512 pid = map_cmdline_to_pid[i];
2513 if (pid == -1 || pid == NO_CMDLINE_MAP)
2516 trace_find_cmdline(pid, buf_comm);
2517 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2522 len = simple_read_from_buffer(ubuf, cnt, ppos,
2531 static const struct file_operations tracing_saved_cmdlines_fops = {
2532 .open = tracing_open_generic,
2533 .read = tracing_saved_cmdlines_read,
2537 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2538 size_t cnt, loff_t *ppos)
2543 r = sprintf(buf, "%u\n", tracer_enabled);
2544 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2548 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2549 size_t cnt, loff_t *ppos)
2551 struct trace_array *tr = filp->private_data;
2556 if (cnt >= sizeof(buf))
2559 if (copy_from_user(&buf, ubuf, cnt))
2564 ret = strict_strtoul(buf, 10, &val);
2570 mutex_lock(&trace_types_lock);
2571 if (tracer_enabled ^ val) {
2574 if (current_trace->start)
2575 current_trace->start(tr);
2580 if (current_trace->stop)
2581 current_trace->stop(tr);
2584 mutex_unlock(&trace_types_lock);
2592 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2593 size_t cnt, loff_t *ppos)
2595 char buf[MAX_TRACER_SIZE+2];
2598 mutex_lock(&trace_types_lock);
2600 r = sprintf(buf, "%s\n", current_trace->name);
2602 r = sprintf(buf, "\n");
2603 mutex_unlock(&trace_types_lock);
2605 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2608 int tracer_init(struct tracer *t, struct trace_array *tr)
2610 tracing_reset_online_cpus(tr);
2614 static int tracing_resize_ring_buffer(unsigned long size)
2619 * If kernel or user changes the size of the ring buffer
2620 * we use the size that was given, and we can forget about
2621 * expanding it later.
2623 ring_buffer_expanded = 1;
2625 ret = ring_buffer_resize(global_trace.buffer, size);
2629 ret = ring_buffer_resize(max_tr.buffer, size);
2633 r = ring_buffer_resize(global_trace.buffer,
2634 global_trace.entries);
2637 * AARGH! We are left with different
2638 * size max buffer!!!!
2639 * The max buffer is our "snapshot" buffer.
2640 * When a tracer needs a snapshot (one of the
2641 * latency tracers), it swaps the max buffer
2642 * with the saved snap shot. We succeeded to
2643 * update the size of the main buffer, but failed to
2644 * update the size of the max buffer. But when we tried
2645 * to reset the main buffer to the original size, we
2646 * failed there too. This is very unlikely to
2647 * happen, but if it does, warn and kill all
2651 tracing_disabled = 1;
2656 global_trace.entries = size;
2662 * tracing_update_buffers - used by tracing facility to expand ring buffers
2664 * To save on memory when the tracing is never used on a system with it
2665 * configured in. The ring buffers are set to a minimum size. But once
2666 * a user starts to use the tracing facility, then they need to grow
2667 * to their default size.
2669 * This function is to be called when a tracer is about to be used.
2671 int tracing_update_buffers(void)
2675 mutex_lock(&trace_types_lock);
2676 if (!ring_buffer_expanded)
2677 ret = tracing_resize_ring_buffer(trace_buf_size);
2678 mutex_unlock(&trace_types_lock);
2683 struct trace_option_dentry;
2685 static struct trace_option_dentry *
2686 create_trace_option_files(struct tracer *tracer);
2689 destroy_trace_option_files(struct trace_option_dentry *topts);
2691 static int tracing_set_tracer(const char *buf)
2693 static struct trace_option_dentry *topts;
2694 struct trace_array *tr = &global_trace;
2698 mutex_lock(&trace_types_lock);
2700 if (!ring_buffer_expanded) {
2701 ret = tracing_resize_ring_buffer(trace_buf_size);
2707 for (t = trace_types; t; t = t->next) {
2708 if (strcmp(t->name, buf) == 0)
2715 if (t == current_trace)
2718 trace_branch_disable();
2719 if (current_trace && current_trace->reset)
2720 current_trace->reset(tr);
2722 destroy_trace_option_files(topts);
2726 topts = create_trace_option_files(current_trace);
2729 ret = tracer_init(t, tr);
2734 trace_branch_enable(tr);
2736 mutex_unlock(&trace_types_lock);
2742 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2743 size_t cnt, loff_t *ppos)
2745 char buf[MAX_TRACER_SIZE+1];
2752 if (cnt > MAX_TRACER_SIZE)
2753 cnt = MAX_TRACER_SIZE;
2755 if (copy_from_user(&buf, ubuf, cnt))
2760 /* strip ending whitespace. */
2761 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2764 err = tracing_set_tracer(buf);
2774 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2775 size_t cnt, loff_t *ppos)
2777 unsigned long *ptr = filp->private_data;
2781 r = snprintf(buf, sizeof(buf), "%ld\n",
2782 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2783 if (r > sizeof(buf))
2785 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2789 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2790 size_t cnt, loff_t *ppos)
2792 unsigned long *ptr = filp->private_data;
2797 if (cnt >= sizeof(buf))
2800 if (copy_from_user(&buf, ubuf, cnt))
2805 ret = strict_strtoul(buf, 10, &val);
2814 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2816 long cpu_file = (long) inode->i_private;
2817 struct trace_iterator *iter;
2820 if (tracing_disabled)
2823 mutex_lock(&trace_types_lock);
2825 /* We only allow one reader per cpu */
2826 if (cpu_file == TRACE_PIPE_ALL_CPU) {
2827 if (!cpumask_empty(tracing_reader_cpumask)) {
2831 cpumask_setall(tracing_reader_cpumask);
2833 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2834 cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2841 /* create a buffer to store the information to pass to userspace */
2842 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2849 * We make a copy of the current tracer to avoid concurrent
2850 * changes on it while we are reading.
2852 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2858 *iter->trace = *current_trace;
2860 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2865 /* trace pipe does not show start of buffer */
2866 cpumask_setall(iter->started);
2868 if (trace_flags & TRACE_ITER_LATENCY_FMT)
2869 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2871 iter->cpu_file = cpu_file;
2872 iter->tr = &global_trace;
2873 mutex_init(&iter->mutex);
2874 filp->private_data = iter;
2876 if (iter->trace->pipe_open)
2877 iter->trace->pipe_open(iter);
2880 mutex_unlock(&trace_types_lock);
2886 mutex_unlock(&trace_types_lock);
2890 static int tracing_release_pipe(struct inode *inode, struct file *file)
2892 struct trace_iterator *iter = file->private_data;
2894 mutex_lock(&trace_types_lock);
2896 if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2897 cpumask_clear(tracing_reader_cpumask);
2899 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2901 mutex_unlock(&trace_types_lock);
2903 free_cpumask_var(iter->started);
2904 mutex_destroy(&iter->mutex);
2912 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2914 struct trace_iterator *iter = filp->private_data;
2916 if (trace_flags & TRACE_ITER_BLOCK) {
2918 * Always select as readable when in blocking mode
2920 return POLLIN | POLLRDNORM;
2922 if (!trace_empty(iter))
2923 return POLLIN | POLLRDNORM;
2924 poll_wait(filp, &trace_wait, poll_table);
2925 if (!trace_empty(iter))
2926 return POLLIN | POLLRDNORM;
2933 void default_wait_pipe(struct trace_iterator *iter)
2937 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2939 if (trace_empty(iter))
2942 finish_wait(&trace_wait, &wait);
2946 * This is a make-shift waitqueue.
2947 * A tracer might use this callback on some rare cases:
2949 * 1) the current tracer might hold the runqueue lock when it wakes up
2950 * a reader, hence a deadlock (sched, function, and function graph tracers)
2951 * 2) the function tracers, trace all functions, we don't want
2952 * the overhead of calling wake_up and friends
2953 * (and tracing them too)
2955 * Anyway, this is really very primitive wakeup.
2957 void poll_wait_pipe(struct trace_iterator *iter)
2959 set_current_state(TASK_INTERRUPTIBLE);
2960 /* sleep for 100 msecs, and try again. */
2961 schedule_timeout(HZ / 10);
2964 /* Must be called with trace_types_lock mutex held. */
2965 static int tracing_wait_pipe(struct file *filp)
2967 struct trace_iterator *iter = filp->private_data;
2969 while (trace_empty(iter)) {
2971 if ((filp->f_flags & O_NONBLOCK)) {
2975 mutex_unlock(&iter->mutex);
2977 iter->trace->wait_pipe(iter);
2979 mutex_lock(&iter->mutex);
2981 if (signal_pending(current))
2985 * We block until we read something and tracing is disabled.
2986 * We still block if tracing is disabled, but we have never
2987 * read anything. This allows a user to cat this file, and
2988 * then enable tracing. But after we have read something,
2989 * we give an EOF when tracing is again disabled.
2991 * iter->pos will be 0 if we haven't read anything.
2993 if (!tracer_enabled && iter->pos)
3004 tracing_read_pipe(struct file *filp, char __user *ubuf,
3005 size_t cnt, loff_t *ppos)
3007 struct trace_iterator *iter = filp->private_data;
3008 static struct tracer *old_tracer;
3011 /* return any leftover data */
3012 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3016 trace_seq_init(&iter->seq);
3018 /* copy the tracer to avoid using a global lock all around */
3019 mutex_lock(&trace_types_lock);
3020 if (unlikely(old_tracer != current_trace && current_trace)) {
3021 old_tracer = current_trace;
3022 *iter->trace = *current_trace;
3024 mutex_unlock(&trace_types_lock);
3027 * Avoid more than one consumer on a single file descriptor
3028 * This is just a matter of traces coherency, the ring buffer itself
3031 mutex_lock(&iter->mutex);
3032 if (iter->trace->read) {
3033 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3039 sret = tracing_wait_pipe(filp);
3043 /* stop when tracing is finished */
3044 if (trace_empty(iter)) {
3049 if (cnt >= PAGE_SIZE)
3050 cnt = PAGE_SIZE - 1;
3052 /* reset all but tr, trace, and overruns */
3053 memset(&iter->seq, 0,
3054 sizeof(struct trace_iterator) -
3055 offsetof(struct trace_iterator, seq));
3058 trace_event_read_lock();
3059 while (find_next_entry_inc(iter) != NULL) {
3060 enum print_line_t ret;
3061 int len = iter->seq.len;
3063 ret = print_trace_line(iter);
3064 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3065 /* don't print partial lines */
3066 iter->seq.len = len;
3069 if (ret != TRACE_TYPE_NO_CONSUME)
3070 trace_consume(iter);
3072 if (iter->seq.len >= cnt)
3075 trace_event_read_unlock();
3077 /* Now copy what we have to the user */
3078 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3079 if (iter->seq.readpos >= iter->seq.len)
3080 trace_seq_init(&iter->seq);
3083 * If there was nothing to send to user, inspite of consuming trace
3084 * entries, go back to wait for more entries.
3090 mutex_unlock(&iter->mutex);
3095 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3096 struct pipe_buffer *buf)
3098 __free_page(buf->page);
3101 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3104 __free_page(spd->pages[idx]);
3107 static struct pipe_buf_operations tracing_pipe_buf_ops = {
3109 .map = generic_pipe_buf_map,
3110 .unmap = generic_pipe_buf_unmap,
3111 .confirm = generic_pipe_buf_confirm,
3112 .release = tracing_pipe_buf_release,
3113 .steal = generic_pipe_buf_steal,
3114 .get = generic_pipe_buf_get,
3118 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3123 /* Seq buffer is page-sized, exactly what we need. */
3125 count = iter->seq.len;
3126 ret = print_trace_line(iter);
3127 count = iter->seq.len - count;
3130 iter->seq.len -= count;
3133 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3134 iter->seq.len -= count;
3138 if (ret != TRACE_TYPE_NO_CONSUME)
3139 trace_consume(iter);
3141 if (!find_next_entry_inc(iter)) {
3151 static ssize_t tracing_splice_read_pipe(struct file *filp,
3153 struct pipe_inode_info *pipe,
3157 struct page *pages[PIPE_BUFFERS];
3158 struct partial_page partial[PIPE_BUFFERS];
3159 struct trace_iterator *iter = filp->private_data;
3160 struct splice_pipe_desc spd = {
3163 .nr_pages = 0, /* This gets updated below. */
3165 .ops = &tracing_pipe_buf_ops,
3166 .spd_release = tracing_spd_release_pipe,
3168 static struct tracer *old_tracer;
3173 /* copy the tracer to avoid using a global lock all around */
3174 mutex_lock(&trace_types_lock);
3175 if (unlikely(old_tracer != current_trace && current_trace)) {
3176 old_tracer = current_trace;
3177 *iter->trace = *current_trace;
3179 mutex_unlock(&trace_types_lock);
3181 mutex_lock(&iter->mutex);
3183 if (iter->trace->splice_read) {
3184 ret = iter->trace->splice_read(iter, filp,
3185 ppos, pipe, len, flags);
3190 ret = tracing_wait_pipe(filp);
3194 if (!iter->ent && !find_next_entry_inc(iter)) {
3199 trace_event_read_lock();
3201 /* Fill as many pages as possible. */
3202 for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3203 pages[i] = alloc_page(GFP_KERNEL);
3207 rem = tracing_fill_pipe_page(rem, iter);
3209 /* Copy the data into the page, so we can start over. */
3210 ret = trace_seq_to_buffer(&iter->seq,
3211 page_address(pages[i]),
3214 __free_page(pages[i]);
3217 partial[i].offset = 0;
3218 partial[i].len = iter->seq.len;
3220 trace_seq_init(&iter->seq);
3223 trace_event_read_unlock();
3224 mutex_unlock(&iter->mutex);
3228 return splice_to_pipe(pipe, &spd);
3231 mutex_unlock(&iter->mutex);
3237 tracing_entries_read(struct file *filp, char __user *ubuf,
3238 size_t cnt, loff_t *ppos)
3240 struct trace_array *tr = filp->private_data;
3244 mutex_lock(&trace_types_lock);
3245 if (!ring_buffer_expanded)
3246 r = sprintf(buf, "%lu (expanded: %lu)\n",
3248 trace_buf_size >> 10);
3250 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3251 mutex_unlock(&trace_types_lock);
3253 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3257 tracing_entries_write(struct file *filp, const char __user *ubuf,
3258 size_t cnt, loff_t *ppos)
3264 if (cnt >= sizeof(buf))
3267 if (copy_from_user(&buf, ubuf, cnt))
3272 ret = strict_strtoul(buf, 10, &val);
3276 /* must have at least 1 entry */
3280 mutex_lock(&trace_types_lock);
3284 /* disable all cpu buffers */
3285 for_each_tracing_cpu(cpu) {
3286 if (global_trace.data[cpu])
3287 atomic_inc(&global_trace.data[cpu]->disabled);
3288 if (max_tr.data[cpu])
3289 atomic_inc(&max_tr.data[cpu]->disabled);
3292 /* value is in KB */
3295 if (val != global_trace.entries) {
3296 ret = tracing_resize_ring_buffer(val);
3305 /* If check pages failed, return ENOMEM */
3306 if (tracing_disabled)
3309 for_each_tracing_cpu(cpu) {
3310 if (global_trace.data[cpu])
3311 atomic_dec(&global_trace.data[cpu]->disabled);
3312 if (max_tr.data[cpu])
3313 atomic_dec(&max_tr.data[cpu]->disabled);
3317 max_tr.entries = global_trace.entries;
3318 mutex_unlock(&trace_types_lock);
3324 tracing_mark_write(struct file *filp, const char __user *ubuf,
3325 size_t cnt, loff_t *fpos)
3329 if (tracing_disabled)
3332 if (cnt > TRACE_BUF_SIZE)
3333 cnt = TRACE_BUF_SIZE;
3335 buf = kmalloc(cnt + 2, GFP_KERNEL);
3339 if (copy_from_user(buf, ubuf, cnt)) {
3343 if (buf[cnt-1] != '\n') {
3349 cnt = trace_vprintk(0, buf, NULL);
3356 static ssize_t tracing_clock_read(struct file *filp, char __user *ubuf,
3357 size_t cnt, loff_t *ppos)
3363 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3364 bufiter += snprintf(buf + bufiter, sizeof(buf) - bufiter,
3365 "%s%s%s%s", i ? " " : "",
3366 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3367 i == trace_clock_id ? "]" : "");
3368 bufiter += snprintf(buf + bufiter, sizeof(buf) - bufiter, "\n");
3370 return simple_read_from_buffer(ubuf, cnt, ppos, buf, bufiter);
3373 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3374 size_t cnt, loff_t *fpos)
3377 const char *clockstr;
3380 if (cnt >= sizeof(buf))
3383 if (copy_from_user(&buf, ubuf, cnt))
3388 clockstr = strstrip(buf);
3390 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3391 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3394 if (i == ARRAY_SIZE(trace_clocks))
3399 mutex_lock(&trace_types_lock);
3401 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3403 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3405 mutex_unlock(&trace_types_lock);
3412 static const struct file_operations tracing_max_lat_fops = {
3413 .open = tracing_open_generic,
3414 .read = tracing_max_lat_read,
3415 .write = tracing_max_lat_write,
3418 static const struct file_operations tracing_ctrl_fops = {
3419 .open = tracing_open_generic,
3420 .read = tracing_ctrl_read,
3421 .write = tracing_ctrl_write,
3424 static const struct file_operations set_tracer_fops = {
3425 .open = tracing_open_generic,
3426 .read = tracing_set_trace_read,
3427 .write = tracing_set_trace_write,
3430 static const struct file_operations tracing_pipe_fops = {
3431 .open = tracing_open_pipe,
3432 .poll = tracing_poll_pipe,
3433 .read = tracing_read_pipe,
3434 .splice_read = tracing_splice_read_pipe,
3435 .release = tracing_release_pipe,
3438 static const struct file_operations tracing_entries_fops = {
3439 .open = tracing_open_generic,
3440 .read = tracing_entries_read,
3441 .write = tracing_entries_write,
3444 static const struct file_operations tracing_mark_fops = {
3445 .open = tracing_open_generic,
3446 .write = tracing_mark_write,
3449 static const struct file_operations trace_clock_fops = {
3450 .open = tracing_open_generic,
3451 .read = tracing_clock_read,
3452 .write = tracing_clock_write,
3455 struct ftrace_buffer_info {
3456 struct trace_array *tr;
3462 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3464 int cpu = (int)(long)inode->i_private;
3465 struct ftrace_buffer_info *info;
3467 if (tracing_disabled)
3470 info = kzalloc(sizeof(*info), GFP_KERNEL);
3474 info->tr = &global_trace;
3477 /* Force reading ring buffer for first read */
3478 info->read = (unsigned int)-1;
3480 filp->private_data = info;
3482 return nonseekable_open(inode, filp);
3486 tracing_buffers_read(struct file *filp, char __user *ubuf,
3487 size_t count, loff_t *ppos)
3489 struct ftrace_buffer_info *info = filp->private_data;
3498 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3502 /* Do we have previous read data to read? */
3503 if (info->read < PAGE_SIZE)
3508 ret = ring_buffer_read_page(info->tr->buffer,
3515 pos = ring_buffer_page_len(info->spare);
3517 if (pos < PAGE_SIZE)
3518 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3521 size = PAGE_SIZE - info->read;
3525 ret = copy_to_user(ubuf, info->spare + info->read, size);
3536 static int tracing_buffers_release(struct inode *inode, struct file *file)
3538 struct ftrace_buffer_info *info = file->private_data;
3541 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3548 struct ring_buffer *buffer;
3553 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3554 struct pipe_buffer *buf)
3556 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3561 ring_buffer_free_read_page(ref->buffer, ref->page);
3566 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3567 struct pipe_buffer *buf)
3572 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3573 struct pipe_buffer *buf)
3575 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3580 /* Pipe buffer operations for a buffer. */
3581 static struct pipe_buf_operations buffer_pipe_buf_ops = {
3583 .map = generic_pipe_buf_map,
3584 .unmap = generic_pipe_buf_unmap,
3585 .confirm = generic_pipe_buf_confirm,
3586 .release = buffer_pipe_buf_release,
3587 .steal = buffer_pipe_buf_steal,
3588 .get = buffer_pipe_buf_get,
3592 * Callback from splice_to_pipe(), if we need to release some pages
3593 * at the end of the spd in case we error'ed out in filling the pipe.
3595 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3597 struct buffer_ref *ref =
3598 (struct buffer_ref *)spd->partial[i].private;
3603 ring_buffer_free_read_page(ref->buffer, ref->page);
3605 spd->partial[i].private = 0;
3609 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3610 struct pipe_inode_info *pipe, size_t len,
3613 struct ftrace_buffer_info *info = file->private_data;
3614 struct partial_page partial[PIPE_BUFFERS];
3615 struct page *pages[PIPE_BUFFERS];
3616 struct splice_pipe_desc spd = {
3620 .ops = &buffer_pipe_buf_ops,
3621 .spd_release = buffer_spd_release,
3623 struct buffer_ref *ref;
3624 int entries, size, i;
3627 if (*ppos & (PAGE_SIZE - 1)) {
3628 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3632 if (len & (PAGE_SIZE - 1)) {
3633 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3634 if (len < PAGE_SIZE)
3639 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3641 for (i = 0; i < PIPE_BUFFERS && len && entries; i++, len -= PAGE_SIZE) {
3645 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3650 ref->buffer = info->tr->buffer;
3651 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3657 r = ring_buffer_read_page(ref->buffer, &ref->page,
3660 ring_buffer_free_read_page(ref->buffer,
3667 * zero out any left over data, this is going to
3670 size = ring_buffer_page_len(ref->page);
3671 if (size < PAGE_SIZE)
3672 memset(ref->page + size, 0, PAGE_SIZE - size);
3674 page = virt_to_page(ref->page);
3676 spd.pages[i] = page;
3677 spd.partial[i].len = PAGE_SIZE;
3678 spd.partial[i].offset = 0;
3679 spd.partial[i].private = (unsigned long)ref;
3683 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3688 /* did we read anything? */
3689 if (!spd.nr_pages) {
3690 if (flags & SPLICE_F_NONBLOCK)
3698 ret = splice_to_pipe(pipe, &spd);
3703 static const struct file_operations tracing_buffers_fops = {
3704 .open = tracing_buffers_open,
3705 .read = tracing_buffers_read,
3706 .release = tracing_buffers_release,
3707 .splice_read = tracing_buffers_splice_read,
3708 .llseek = no_llseek,
3712 tracing_stats_read(struct file *filp, char __user *ubuf,
3713 size_t count, loff_t *ppos)
3715 unsigned long cpu = (unsigned long)filp->private_data;
3716 struct trace_array *tr = &global_trace;
3717 struct trace_seq *s;
3720 s = kmalloc(sizeof(*s), GFP_KERNEL);
3726 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3727 trace_seq_printf(s, "entries: %ld\n", cnt);
3729 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3730 trace_seq_printf(s, "overrun: %ld\n", cnt);
3732 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3733 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3735 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3742 static const struct file_operations tracing_stats_fops = {
3743 .open = tracing_open_generic,
3744 .read = tracing_stats_read,
3747 #ifdef CONFIG_DYNAMIC_FTRACE
3749 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3755 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3756 size_t cnt, loff_t *ppos)
3758 static char ftrace_dyn_info_buffer[1024];
3759 static DEFINE_MUTEX(dyn_info_mutex);
3760 unsigned long *p = filp->private_data;
3761 char *buf = ftrace_dyn_info_buffer;
3762 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3765 mutex_lock(&dyn_info_mutex);
3766 r = sprintf(buf, "%ld ", *p);
3768 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3771 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3773 mutex_unlock(&dyn_info_mutex);
3778 static const struct file_operations tracing_dyn_info_fops = {
3779 .open = tracing_open_generic,
3780 .read = tracing_read_dyn_info,
3784 static struct dentry *d_tracer;
3786 struct dentry *tracing_init_dentry(void)
3793 if (!debugfs_initialized())
3796 d_tracer = debugfs_create_dir("tracing", NULL);
3798 if (!d_tracer && !once) {
3800 pr_warning("Could not create debugfs directory 'tracing'\n");
3807 static struct dentry *d_percpu;
3809 struct dentry *tracing_dentry_percpu(void)
3812 struct dentry *d_tracer;
3817 d_tracer = tracing_init_dentry();
3822 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3824 if (!d_percpu && !once) {
3826 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3833 static void tracing_init_debugfs_percpu(long cpu)
3835 struct dentry *d_percpu = tracing_dentry_percpu();
3836 struct dentry *d_cpu;
3837 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3840 if (cpu > 999 || cpu < 0)
3843 sprintf(cpu_dir, "cpu%ld", cpu);
3844 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3846 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3850 /* per cpu trace_pipe */
3851 trace_create_file("trace_pipe", 0444, d_cpu,
3852 (void *) cpu, &tracing_pipe_fops);
3855 trace_create_file("trace", 0644, d_cpu,
3856 (void *) cpu, &tracing_fops);
3858 trace_create_file("trace_pipe_raw", 0444, d_cpu,
3859 (void *) cpu, &tracing_buffers_fops);
3861 trace_create_file("stats", 0444, d_cpu,
3862 (void *) cpu, &tracing_stats_fops);
3865 #ifdef CONFIG_FTRACE_SELFTEST
3866 /* Let selftest have access to static functions in this file */
3867 #include "trace_selftest.c"
3870 struct trace_option_dentry {
3871 struct tracer_opt *opt;
3872 struct tracer_flags *flags;
3873 struct dentry *entry;
3877 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3880 struct trace_option_dentry *topt = filp->private_data;
3883 if (topt->flags->val & topt->opt->bit)
3888 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3892 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3895 struct trace_option_dentry *topt = filp->private_data;
3900 if (cnt >= sizeof(buf))
3903 if (copy_from_user(&buf, ubuf, cnt))
3908 ret = strict_strtoul(buf, 10, &val);
3915 /* do nothing if already cleared */
3916 if (!(topt->flags->val & topt->opt->bit))
3919 mutex_lock(&trace_types_lock);
3920 if (current_trace->set_flag)
3921 ret = current_trace->set_flag(topt->flags->val,
3923 mutex_unlock(&trace_types_lock);
3926 topt->flags->val &= ~topt->opt->bit;
3929 /* do nothing if already set */
3930 if (topt->flags->val & topt->opt->bit)
3933 mutex_lock(&trace_types_lock);
3934 if (current_trace->set_flag)
3935 ret = current_trace->set_flag(topt->flags->val,
3937 mutex_unlock(&trace_types_lock);
3940 topt->flags->val |= topt->opt->bit;
3953 static const struct file_operations trace_options_fops = {
3954 .open = tracing_open_generic,
3955 .read = trace_options_read,
3956 .write = trace_options_write,
3960 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
3963 long index = (long)filp->private_data;
3966 if (trace_flags & (1 << index))
3971 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3975 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
3978 long index = (long)filp->private_data;
3983 if (cnt >= sizeof(buf))
3986 if (copy_from_user(&buf, ubuf, cnt))
3991 ret = strict_strtoul(buf, 10, &val);
3995 if (val != 0 && val != 1)
3997 set_tracer_flags(1 << index, val);
4004 static const struct file_operations trace_options_core_fops = {
4005 .open = tracing_open_generic,
4006 .read = trace_options_core_read,
4007 .write = trace_options_core_write,
4010 struct dentry *trace_create_file(const char *name,
4012 struct dentry *parent,
4014 const struct file_operations *fops)
4018 ret = debugfs_create_file(name, mode, parent, data, fops);
4020 pr_warning("Could not create debugfs '%s' entry\n", name);
4026 static struct dentry *trace_options_init_dentry(void)
4028 struct dentry *d_tracer;
4029 static struct dentry *t_options;
4034 d_tracer = tracing_init_dentry();
4038 t_options = debugfs_create_dir("options", d_tracer);
4040 pr_warning("Could not create debugfs directory 'options'\n");
4048 create_trace_option_file(struct trace_option_dentry *topt,
4049 struct tracer_flags *flags,
4050 struct tracer_opt *opt)
4052 struct dentry *t_options;
4054 t_options = trace_options_init_dentry();
4058 topt->flags = flags;
4061 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4062 &trace_options_fops);
4066 static struct trace_option_dentry *
4067 create_trace_option_files(struct tracer *tracer)
4069 struct trace_option_dentry *topts;
4070 struct tracer_flags *flags;
4071 struct tracer_opt *opts;
4077 flags = tracer->flags;
4079 if (!flags || !flags->opts)
4084 for (cnt = 0; opts[cnt].name; cnt++)
4087 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4091 for (cnt = 0; opts[cnt].name; cnt++)
4092 create_trace_option_file(&topts[cnt], flags,
4099 destroy_trace_option_files(struct trace_option_dentry *topts)
4106 for (cnt = 0; topts[cnt].opt; cnt++) {
4107 if (topts[cnt].entry)
4108 debugfs_remove(topts[cnt].entry);
4114 static struct dentry *
4115 create_trace_option_core_file(const char *option, long index)
4117 struct dentry *t_options;
4119 t_options = trace_options_init_dentry();
4123 return trace_create_file(option, 0644, t_options, (void *)index,
4124 &trace_options_core_fops);
4127 static __init void create_trace_options_dir(void)
4129 struct dentry *t_options;
4132 t_options = trace_options_init_dentry();
4136 for (i = 0; trace_options[i]; i++)
4137 create_trace_option_core_file(trace_options[i], i);
4140 static __init int tracer_init_debugfs(void)
4142 struct dentry *d_tracer;
4145 d_tracer = tracing_init_dentry();
4147 trace_create_file("tracing_enabled", 0644, d_tracer,
4148 &global_trace, &tracing_ctrl_fops);
4150 trace_create_file("trace_options", 0644, d_tracer,
4151 NULL, &tracing_iter_fops);
4153 trace_create_file("tracing_cpumask", 0644, d_tracer,
4154 NULL, &tracing_cpumask_fops);
4156 trace_create_file("trace", 0644, d_tracer,
4157 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4159 trace_create_file("available_tracers", 0444, d_tracer,
4160 &global_trace, &show_traces_fops);
4162 trace_create_file("current_tracer", 0644, d_tracer,
4163 &global_trace, &set_tracer_fops);
4165 #ifdef CONFIG_TRACER_MAX_TRACE
4166 trace_create_file("tracing_max_latency", 0644, d_tracer,
4167 &tracing_max_latency, &tracing_max_lat_fops);
4169 trace_create_file("tracing_thresh", 0644, d_tracer,
4170 &tracing_thresh, &tracing_max_lat_fops);
4173 trace_create_file("README", 0444, d_tracer,
4174 NULL, &tracing_readme_fops);
4176 trace_create_file("trace_pipe", 0444, d_tracer,
4177 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4179 trace_create_file("buffer_size_kb", 0644, d_tracer,
4180 &global_trace, &tracing_entries_fops);
4182 trace_create_file("trace_marker", 0220, d_tracer,
4183 NULL, &tracing_mark_fops);
4185 trace_create_file("saved_cmdlines", 0444, d_tracer,
4186 NULL, &tracing_saved_cmdlines_fops);
4188 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4191 #ifdef CONFIG_DYNAMIC_FTRACE
4192 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4193 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4195 #ifdef CONFIG_SYSPROF_TRACER
4196 init_tracer_sysprof_debugfs(d_tracer);
4199 create_trace_options_dir();
4201 for_each_tracing_cpu(cpu)
4202 tracing_init_debugfs_percpu(cpu);
4207 static int trace_panic_handler(struct notifier_block *this,
4208 unsigned long event, void *unused)
4210 if (ftrace_dump_on_oops)
4215 static struct notifier_block trace_panic_notifier = {
4216 .notifier_call = trace_panic_handler,
4218 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4221 static int trace_die_handler(struct notifier_block *self,
4227 if (ftrace_dump_on_oops)
4236 static struct notifier_block trace_die_notifier = {
4237 .notifier_call = trace_die_handler,
4242 * printk is set to max of 1024, we really don't need it that big.
4243 * Nothing should be printing 1000 characters anyway.
4245 #define TRACE_MAX_PRINT 1000
4248 * Define here KERN_TRACE so that we have one place to modify
4249 * it if we decide to change what log level the ftrace dump
4252 #define KERN_TRACE KERN_EMERG
4255 trace_printk_seq(struct trace_seq *s)
4257 /* Probably should print a warning here. */
4261 /* should be zero ended, but we are paranoid. */
4262 s->buffer[s->len] = 0;
4264 printk(KERN_TRACE "%s", s->buffer);
4269 static void __ftrace_dump(bool disable_tracing)
4271 static raw_spinlock_t ftrace_dump_lock =
4272 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
4273 /* use static because iter can be a bit big for the stack */
4274 static struct trace_iterator iter;
4275 unsigned int old_userobj;
4276 static int dump_ran;
4277 unsigned long flags;
4281 local_irq_save(flags);
4282 __raw_spin_lock(&ftrace_dump_lock);
4290 if (disable_tracing)
4293 for_each_tracing_cpu(cpu) {
4294 atomic_inc(&global_trace.data[cpu]->disabled);
4297 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4299 /* don't look at user memory in panic mode */
4300 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4302 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4304 /* Simulate the iterator */
4305 iter.tr = &global_trace;
4306 iter.trace = current_trace;
4307 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4310 * We need to stop all tracing on all CPUS to read the
4311 * the next buffer. This is a bit expensive, but is
4312 * not done often. We fill all what we can read,
4313 * and then release the locks again.
4316 while (!trace_empty(&iter)) {
4319 printk(KERN_TRACE "---------------------------------\n");
4323 /* reset all but tr, trace, and overruns */
4324 memset(&iter.seq, 0,
4325 sizeof(struct trace_iterator) -
4326 offsetof(struct trace_iterator, seq));
4327 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4330 if (find_next_entry_inc(&iter) != NULL) {
4333 ret = print_trace_line(&iter);
4334 if (ret != TRACE_TYPE_NO_CONSUME)
4335 trace_consume(&iter);
4338 trace_printk_seq(&iter.seq);
4342 printk(KERN_TRACE " (ftrace buffer empty)\n");
4344 printk(KERN_TRACE "---------------------------------\n");
4346 /* Re-enable tracing if requested */
4347 if (!disable_tracing) {
4348 trace_flags |= old_userobj;
4350 for_each_tracing_cpu(cpu) {
4351 atomic_dec(&global_trace.data[cpu]->disabled);
4357 __raw_spin_unlock(&ftrace_dump_lock);
4358 local_irq_restore(flags);
4361 /* By default: disable tracing after the dump */
4362 void ftrace_dump(void)
4364 __ftrace_dump(true);
4367 __init static int tracer_alloc_buffers(void)
4373 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4376 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4377 goto out_free_buffer_mask;
4379 if (!zalloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
4380 goto out_free_tracing_cpumask;
4382 /* To save memory, keep the ring buffer size to its minimum */
4383 if (ring_buffer_expanded)
4384 ring_buf_size = trace_buf_size;
4388 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4389 cpumask_copy(tracing_cpumask, cpu_all_mask);
4391 /* TODO: make the number of buffers hot pluggable with CPUS */
4392 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4393 TRACE_BUFFER_FLAGS);
4394 if (!global_trace.buffer) {
4395 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4397 goto out_free_cpumask;
4399 global_trace.entries = ring_buffer_size(global_trace.buffer);
4402 #ifdef CONFIG_TRACER_MAX_TRACE
4403 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4404 TRACE_BUFFER_FLAGS);
4405 if (!max_tr.buffer) {
4406 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4408 ring_buffer_free(global_trace.buffer);
4409 goto out_free_cpumask;
4411 max_tr.entries = ring_buffer_size(max_tr.buffer);
4412 WARN_ON(max_tr.entries != global_trace.entries);
4415 /* Allocate the first page for all buffers */
4416 for_each_tracing_cpu(i) {
4417 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4418 max_tr.data[i] = &per_cpu(max_data, i);
4421 trace_init_cmdlines();
4423 register_tracer(&nop_trace);
4424 current_trace = &nop_trace;
4425 #ifdef CONFIG_BOOT_TRACER
4426 register_tracer(&boot_tracer);
4428 /* All seems OK, enable tracing */
4429 tracing_disabled = 0;
4431 atomic_notifier_chain_register(&panic_notifier_list,
4432 &trace_panic_notifier);
4434 register_die_notifier(&trace_die_notifier);
4439 free_cpumask_var(tracing_reader_cpumask);
4440 out_free_tracing_cpumask:
4441 free_cpumask_var(tracing_cpumask);
4442 out_free_buffer_mask:
4443 free_cpumask_var(tracing_buffer_mask);
4448 __init static int clear_boot_tracer(void)
4451 * The default tracer at boot buffer is an init section.
4452 * This function is called in lateinit. If we did not
4453 * find the boot tracer, then clear it out, to prevent
4454 * later registration from accessing the buffer that is
4455 * about to be freed.
4457 if (!default_bootup_tracer)
4460 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4461 default_bootup_tracer);
4462 default_bootup_tracer = NULL;
4467 early_initcall(tracer_alloc_buffers);
4468 fs_initcall(tracer_init_debugfs);
4469 late_initcall(clear_boot_tracer);