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 BOOTUP_TRACER_SIZE 100
129 static char bootup_tracer_buf[BOOTUP_TRACER_SIZE] __initdata;
130 static char *default_bootup_tracer;
132 static int __init set_ftrace(char *str)
134 strncpy(bootup_tracer_buf, str, BOOTUP_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_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 * max_tracer_type_len is used to simplify the allocating of
246 * buffers to read userspace tracer names. We keep track of
247 * the longest tracer name registered.
249 static int max_tracer_type_len;
252 * trace_types_lock is used to protect the trace_types list.
253 * This lock is also used to keep user access serialized.
254 * Accesses from userspace will grab this lock while userspace
255 * activities happen inside the kernel.
257 static DEFINE_MUTEX(trace_types_lock);
259 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
260 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
262 /* trace_flags holds trace_options default values */
263 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
264 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
265 TRACE_ITER_GRAPH_TIME;
267 static int trace_stop_count;
268 static DEFINE_SPINLOCK(tracing_start_lock);
271 * trace_wake_up - wake up tasks waiting for trace input
273 * Simply wakes up any task that is blocked on the trace_wait
274 * queue. These is used with trace_poll for tasks polling the trace.
276 void trace_wake_up(void)
279 * The runqueue_is_locked() can fail, but this is the best we
282 if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
283 wake_up(&trace_wait);
286 static int __init set_buf_size(char *str)
288 unsigned long buf_size;
292 buf_size = memparse(str, &str);
293 /* nr_entries can not be zero */
296 trace_buf_size = buf_size;
299 __setup("trace_buf_size=", set_buf_size);
301 unsigned long nsecs_to_usecs(unsigned long nsecs)
306 /* These must match the bit postions in trace_iterator_flags */
307 static const char *trace_options[] = {
336 { trace_clock_local, "local" },
337 { trace_clock_global, "global" },
343 * trace_parser_get_init - gets the buffer for trace parser
345 int trace_parser_get_init(struct trace_parser *parser, int size)
347 memset(parser, 0, sizeof(*parser));
349 parser->buffer = kmalloc(size, GFP_KERNEL);
358 * trace_parser_put - frees the buffer for trace parser
360 void trace_parser_put(struct trace_parser *parser)
362 kfree(parser->buffer);
366 * trace_get_user - reads the user input string separated by space
367 * (matched by isspace(ch))
369 * For each string found the 'struct trace_parser' is updated,
370 * and the function returns.
372 * Returns number of bytes read.
374 * See kernel/trace/trace.h for 'struct trace_parser' details.
376 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
377 size_t cnt, loff_t *ppos)
384 trace_parser_clear(parser);
386 ret = get_user(ch, ubuf++);
394 * The parser is not finished with the last write,
395 * continue reading the user input without skipping spaces.
398 /* skip white space */
399 while (cnt && isspace(ch)) {
400 ret = get_user(ch, ubuf++);
407 /* only spaces were written */
417 /* read the non-space input */
418 while (cnt && !isspace(ch)) {
419 if (parser->idx < parser->size)
420 parser->buffer[parser->idx++] = ch;
425 ret = get_user(ch, ubuf++);
432 /* We either got finished input or we have to wait for another call. */
434 parser->buffer[parser->idx] = 0;
435 parser->cont = false;
438 parser->buffer[parser->idx++] = ch;
448 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
456 if (s->len <= s->readpos)
459 len = s->len - s->readpos;
462 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
472 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
477 if (s->len <= s->readpos)
480 len = s->len - s->readpos;
483 ret = memcpy(buf, s->buffer + s->readpos, cnt);
492 * ftrace_max_lock is used to protect the swapping of buffers
493 * when taking a max snapshot. The buffers themselves are
494 * protected by per_cpu spinlocks. But the action of the swap
495 * needs its own lock.
497 * This is defined as a raw_spinlock_t in order to help
498 * with performance when lockdep debugging is enabled.
500 * It is also used in other places outside the update_max_tr
501 * so it needs to be defined outside of the
502 * CONFIG_TRACER_MAX_TRACE.
504 static raw_spinlock_t ftrace_max_lock =
505 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
507 #ifdef CONFIG_TRACER_MAX_TRACE
508 unsigned long __read_mostly tracing_max_latency;
509 unsigned long __read_mostly tracing_thresh;
512 * Copy the new maximum trace into the separate maximum-trace
513 * structure. (this way the maximum trace is permanently saved,
514 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
517 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
519 struct trace_array_cpu *data = tr->data[cpu];
520 struct trace_array_cpu *max_data = tr->data[cpu];
523 max_tr.time_start = data->preempt_timestamp;
525 max_data = max_tr.data[cpu];
526 max_data->saved_latency = tracing_max_latency;
527 max_data->critical_start = data->critical_start;
528 max_data->critical_end = data->critical_end;
530 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
531 max_data->pid = tsk->pid;
532 max_data->uid = task_uid(tsk);
533 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
534 max_data->policy = tsk->policy;
535 max_data->rt_priority = tsk->rt_priority;
537 /* record this tasks comm */
538 tracing_record_cmdline(tsk);
542 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
544 * @tsk: the task with the latency
545 * @cpu: The cpu that initiated the trace.
547 * Flip the buffers between the @tr and the max_tr and record information
548 * about which task was the cause of this latency.
551 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
553 struct ring_buffer *buf = tr->buffer;
555 if (trace_stop_count)
558 WARN_ON_ONCE(!irqs_disabled());
559 __raw_spin_lock(&ftrace_max_lock);
561 tr->buffer = max_tr.buffer;
564 __update_max_tr(tr, tsk, cpu);
565 __raw_spin_unlock(&ftrace_max_lock);
569 * update_max_tr_single - only copy one trace over, and reset the rest
571 * @tsk - task with the latency
572 * @cpu - the cpu of the buffer to copy.
574 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
577 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
581 if (trace_stop_count)
584 WARN_ON_ONCE(!irqs_disabled());
585 __raw_spin_lock(&ftrace_max_lock);
587 ftrace_disable_cpu();
589 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
593 * We failed to swap the buffer due to a commit taking
594 * place on this CPU. We fail to record, but we reset
595 * the max trace buffer (no one writes directly to it)
596 * and flag that it failed.
598 trace_array_printk(&max_tr, _THIS_IP_,
599 "Failed to swap buffers due to commit in progress\n");
604 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
606 __update_max_tr(tr, tsk, cpu);
607 __raw_spin_unlock(&ftrace_max_lock);
609 #endif /* CONFIG_TRACER_MAX_TRACE */
612 * register_tracer - register a tracer with the ftrace system.
613 * @type - the plugin for the tracer
615 * Register a new plugin tracer.
617 int register_tracer(struct tracer *type)
618 __releases(kernel_lock)
619 __acquires(kernel_lock)
626 pr_info("Tracer must have a name\n");
631 * When this gets called we hold the BKL which means that
632 * preemption is disabled. Various trace selftests however
633 * need to disable and enable preemption for successful tests.
634 * So we drop the BKL here and grab it after the tests again.
637 mutex_lock(&trace_types_lock);
639 tracing_selftest_running = true;
641 for (t = trace_types; t; t = t->next) {
642 if (strcmp(type->name, t->name) == 0) {
644 pr_info("Trace %s already registered\n",
652 type->set_flag = &dummy_set_flag;
654 type->flags = &dummy_tracer_flags;
656 if (!type->flags->opts)
657 type->flags->opts = dummy_tracer_opt;
658 if (!type->wait_pipe)
659 type->wait_pipe = default_wait_pipe;
662 #ifdef CONFIG_FTRACE_STARTUP_TEST
663 if (type->selftest && !tracing_selftest_disabled) {
664 struct tracer *saved_tracer = current_trace;
665 struct trace_array *tr = &global_trace;
668 * Run a selftest on this tracer.
669 * Here we reset the trace buffer, and set the current
670 * tracer to be this tracer. The tracer can then run some
671 * internal tracing to verify that everything is in order.
672 * If we fail, we do not register this tracer.
674 tracing_reset_online_cpus(tr);
676 current_trace = type;
677 /* the test is responsible for initializing and enabling */
678 pr_info("Testing tracer %s: ", type->name);
679 ret = type->selftest(type, tr);
680 /* the test is responsible for resetting too */
681 current_trace = saved_tracer;
683 printk(KERN_CONT "FAILED!\n");
686 /* Only reset on passing, to avoid touching corrupted buffers */
687 tracing_reset_online_cpus(tr);
689 printk(KERN_CONT "PASSED\n");
693 type->next = trace_types;
695 len = strlen(type->name);
696 if (len > max_tracer_type_len)
697 max_tracer_type_len = len;
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, BOOTUP_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)
730 mutex_lock(&trace_types_lock);
731 for (t = &trace_types; *t; t = &(*t)->next) {
735 pr_info("Trace %s not registered\n", type->name);
741 if (type == current_trace && tracer_enabled) {
744 if (current_trace->stop)
745 current_trace->stop(&global_trace);
746 current_trace = &nop_trace;
749 if (strlen(type->name) != max_tracer_type_len)
752 max_tracer_type_len = 0;
753 for (t = &trace_types; *t; t = &(*t)->next) {
754 len = strlen((*t)->name);
755 if (len > max_tracer_type_len)
756 max_tracer_type_len = len;
759 mutex_unlock(&trace_types_lock);
762 static void __tracing_reset(struct trace_array *tr, int cpu)
764 ftrace_disable_cpu();
765 ring_buffer_reset_cpu(tr->buffer, cpu);
769 void tracing_reset(struct trace_array *tr, int cpu)
771 struct ring_buffer *buffer = tr->buffer;
773 ring_buffer_record_disable(buffer);
775 /* Make sure all commits have finished */
777 __tracing_reset(tr, cpu);
779 ring_buffer_record_enable(buffer);
782 void tracing_reset_online_cpus(struct trace_array *tr)
784 struct ring_buffer *buffer = tr->buffer;
787 ring_buffer_record_disable(buffer);
789 /* Make sure all commits have finished */
792 tr->time_start = ftrace_now(tr->cpu);
794 for_each_online_cpu(cpu)
795 __tracing_reset(tr, cpu);
797 ring_buffer_record_enable(buffer);
800 void tracing_reset_current(int cpu)
802 tracing_reset(&global_trace, cpu);
805 void tracing_reset_current_online_cpus(void)
807 tracing_reset_online_cpus(&global_trace);
810 #define SAVED_CMDLINES 128
811 #define NO_CMDLINE_MAP UINT_MAX
812 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
813 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
814 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
815 static int cmdline_idx;
816 static raw_spinlock_t trace_cmdline_lock = __RAW_SPIN_LOCK_UNLOCKED;
818 /* temporary disable recording */
819 static atomic_t trace_record_cmdline_disabled __read_mostly;
821 static void trace_init_cmdlines(void)
823 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
824 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
829 * ftrace_off_permanent - disable all ftrace code permanently
831 * This should only be called when a serious anomally has
832 * been detected. This will turn off the function tracing,
833 * ring buffers, and other tracing utilites. It takes no
834 * locks and can be called from any context.
836 void ftrace_off_permanent(void)
838 tracing_disabled = 1;
840 tracing_off_permanent();
844 * tracing_start - quick start of the tracer
846 * If tracing is enabled but was stopped by tracing_stop,
847 * this will start the tracer back up.
849 void tracing_start(void)
851 struct ring_buffer *buffer;
854 if (tracing_disabled)
857 spin_lock_irqsave(&tracing_start_lock, flags);
858 if (--trace_stop_count) {
859 if (trace_stop_count < 0) {
860 /* Someone screwed up their debugging */
862 trace_stop_count = 0;
868 buffer = global_trace.buffer;
870 ring_buffer_record_enable(buffer);
872 buffer = max_tr.buffer;
874 ring_buffer_record_enable(buffer);
878 spin_unlock_irqrestore(&tracing_start_lock, flags);
882 * tracing_stop - quick stop of the tracer
884 * Light weight way to stop tracing. Use in conjunction with
887 void tracing_stop(void)
889 struct ring_buffer *buffer;
893 spin_lock_irqsave(&tracing_start_lock, flags);
894 if (trace_stop_count++)
897 buffer = global_trace.buffer;
899 ring_buffer_record_disable(buffer);
901 buffer = max_tr.buffer;
903 ring_buffer_record_disable(buffer);
906 spin_unlock_irqrestore(&tracing_start_lock, flags);
909 void trace_stop_cmdline_recording(void);
911 static void trace_save_cmdline(struct task_struct *tsk)
915 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
919 * It's not the end of the world if we don't get
920 * the lock, but we also don't want to spin
921 * nor do we want to disable interrupts,
922 * so if we miss here, then better luck next time.
924 if (!__raw_spin_trylock(&trace_cmdline_lock))
927 idx = map_pid_to_cmdline[tsk->pid];
928 if (idx == NO_CMDLINE_MAP) {
929 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
932 * Check whether the cmdline buffer at idx has a pid
933 * mapped. We are going to overwrite that entry so we
934 * need to clear the map_pid_to_cmdline. Otherwise we
935 * would read the new comm for the old pid.
937 pid = map_cmdline_to_pid[idx];
938 if (pid != NO_CMDLINE_MAP)
939 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
941 map_cmdline_to_pid[idx] = tsk->pid;
942 map_pid_to_cmdline[tsk->pid] = idx;
947 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
949 __raw_spin_unlock(&trace_cmdline_lock);
952 void trace_find_cmdline(int pid, char comm[])
957 strcpy(comm, "<idle>");
961 if (pid > PID_MAX_DEFAULT) {
962 strcpy(comm, "<...>");
967 __raw_spin_lock(&trace_cmdline_lock);
968 map = map_pid_to_cmdline[pid];
969 if (map != NO_CMDLINE_MAP)
970 strcpy(comm, saved_cmdlines[map]);
972 strcpy(comm, "<...>");
974 __raw_spin_unlock(&trace_cmdline_lock);
978 void tracing_record_cmdline(struct task_struct *tsk)
980 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
984 trace_save_cmdline(tsk);
988 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
991 struct task_struct *tsk = current;
993 entry->preempt_count = pc & 0xff;
994 entry->pid = (tsk) ? tsk->pid : 0;
995 entry->lock_depth = (tsk) ? tsk->lock_depth : 0;
997 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
998 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1000 TRACE_FLAG_IRQS_NOSUPPORT |
1002 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1003 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1004 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1006 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1008 struct ring_buffer_event *
1009 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1012 unsigned long flags, int pc)
1014 struct ring_buffer_event *event;
1016 event = ring_buffer_lock_reserve(buffer, len);
1017 if (event != NULL) {
1018 struct trace_entry *ent = ring_buffer_event_data(event);
1020 tracing_generic_entry_update(ent, flags, pc);
1028 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1029 struct ring_buffer_event *event,
1030 unsigned long flags, int pc,
1033 ring_buffer_unlock_commit(buffer, event);
1035 ftrace_trace_stack(buffer, flags, 6, pc);
1036 ftrace_trace_userstack(buffer, flags, pc);
1042 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1043 struct ring_buffer_event *event,
1044 unsigned long flags, int pc)
1046 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1049 struct ring_buffer_event *
1050 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1051 int type, unsigned long len,
1052 unsigned long flags, int pc)
1054 *current_rb = global_trace.buffer;
1055 return trace_buffer_lock_reserve(*current_rb,
1056 type, len, flags, pc);
1058 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1060 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1061 struct ring_buffer_event *event,
1062 unsigned long flags, int pc)
1064 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1066 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1068 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1069 struct ring_buffer_event *event,
1070 unsigned long flags, int pc)
1072 __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1074 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1076 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1077 struct ring_buffer_event *event)
1079 ring_buffer_discard_commit(buffer, event);
1081 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1084 trace_function(struct trace_array *tr,
1085 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1088 struct ftrace_event_call *call = &event_function;
1089 struct ring_buffer *buffer = tr->buffer;
1090 struct ring_buffer_event *event;
1091 struct ftrace_entry *entry;
1093 /* If we are reading the ring buffer, don't trace */
1094 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
1097 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1101 entry = ring_buffer_event_data(event);
1103 entry->parent_ip = parent_ip;
1105 if (!filter_check_discard(call, entry, buffer, event))
1106 ring_buffer_unlock_commit(buffer, event);
1110 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1111 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1114 if (likely(!atomic_read(&data->disabled)))
1115 trace_function(tr, ip, parent_ip, flags, pc);
1118 #ifdef CONFIG_STACKTRACE
1119 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1120 unsigned long flags,
1123 struct ftrace_event_call *call = &event_kernel_stack;
1124 struct ring_buffer_event *event;
1125 struct stack_entry *entry;
1126 struct stack_trace trace;
1128 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1129 sizeof(*entry), flags, pc);
1132 entry = ring_buffer_event_data(event);
1133 memset(&entry->caller, 0, sizeof(entry->caller));
1135 trace.nr_entries = 0;
1136 trace.max_entries = FTRACE_STACK_ENTRIES;
1138 trace.entries = entry->caller;
1140 save_stack_trace(&trace);
1141 if (!filter_check_discard(call, entry, buffer, event))
1142 ring_buffer_unlock_commit(buffer, event);
1145 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1148 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1151 __ftrace_trace_stack(buffer, flags, skip, pc);
1154 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1157 __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1161 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1163 struct ftrace_event_call *call = &event_user_stack;
1164 struct ring_buffer_event *event;
1165 struct userstack_entry *entry;
1166 struct stack_trace trace;
1168 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1171 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1172 sizeof(*entry), flags, pc);
1175 entry = ring_buffer_event_data(event);
1177 entry->tgid = current->tgid;
1178 memset(&entry->caller, 0, sizeof(entry->caller));
1180 trace.nr_entries = 0;
1181 trace.max_entries = FTRACE_STACK_ENTRIES;
1183 trace.entries = entry->caller;
1185 save_stack_trace_user(&trace);
1186 if (!filter_check_discard(call, entry, buffer, event))
1187 ring_buffer_unlock_commit(buffer, event);
1191 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1193 ftrace_trace_userstack(tr, flags, preempt_count());
1197 #endif /* CONFIG_STACKTRACE */
1200 ftrace_trace_special(void *__tr,
1201 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1204 struct ring_buffer_event *event;
1205 struct trace_array *tr = __tr;
1206 struct ring_buffer *buffer = tr->buffer;
1207 struct special_entry *entry;
1209 event = trace_buffer_lock_reserve(buffer, TRACE_SPECIAL,
1210 sizeof(*entry), 0, pc);
1213 entry = ring_buffer_event_data(event);
1217 trace_buffer_unlock_commit(buffer, event, 0, pc);
1221 __trace_special(void *__tr, void *__data,
1222 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1224 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1228 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1230 struct trace_array *tr = &global_trace;
1231 struct trace_array_cpu *data;
1232 unsigned long flags;
1236 if (tracing_disabled)
1239 pc = preempt_count();
1240 local_irq_save(flags);
1241 cpu = raw_smp_processor_id();
1242 data = tr->data[cpu];
1244 if (likely(atomic_inc_return(&data->disabled) == 1))
1245 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1247 atomic_dec(&data->disabled);
1248 local_irq_restore(flags);
1252 * trace_vbprintk - write binary msg to tracing buffer
1255 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1257 static raw_spinlock_t trace_buf_lock =
1258 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
1259 static u32 trace_buf[TRACE_BUF_SIZE];
1261 struct ftrace_event_call *call = &event_bprint;
1262 struct ring_buffer_event *event;
1263 struct ring_buffer *buffer;
1264 struct trace_array *tr = &global_trace;
1265 struct trace_array_cpu *data;
1266 struct bprint_entry *entry;
1267 unsigned long flags;
1270 int cpu, len = 0, size, pc;
1272 if (unlikely(tracing_selftest_running || tracing_disabled))
1275 /* Don't pollute graph traces with trace_vprintk internals */
1276 pause_graph_tracing();
1278 pc = preempt_count();
1279 resched = ftrace_preempt_disable();
1280 cpu = raw_smp_processor_id();
1281 data = tr->data[cpu];
1283 disable = atomic_inc_return(&data->disabled);
1284 if (unlikely(disable != 1))
1287 /* Lockdep uses trace_printk for lock tracing */
1288 local_irq_save(flags);
1289 __raw_spin_lock(&trace_buf_lock);
1290 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1292 if (len > TRACE_BUF_SIZE || len < 0)
1295 size = sizeof(*entry) + sizeof(u32) * len;
1296 buffer = tr->buffer;
1297 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1301 entry = ring_buffer_event_data(event);
1305 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1306 if (!filter_check_discard(call, entry, buffer, event))
1307 ring_buffer_unlock_commit(buffer, event);
1310 __raw_spin_unlock(&trace_buf_lock);
1311 local_irq_restore(flags);
1314 atomic_dec_return(&data->disabled);
1315 ftrace_preempt_enable(resched);
1316 unpause_graph_tracing();
1320 EXPORT_SYMBOL_GPL(trace_vbprintk);
1322 int trace_array_printk(struct trace_array *tr,
1323 unsigned long ip, const char *fmt, ...)
1328 if (!(trace_flags & TRACE_ITER_PRINTK))
1332 ret = trace_array_vprintk(tr, ip, fmt, ap);
1337 int trace_array_vprintk(struct trace_array *tr,
1338 unsigned long ip, const char *fmt, va_list args)
1340 static raw_spinlock_t trace_buf_lock = __RAW_SPIN_LOCK_UNLOCKED;
1341 static char trace_buf[TRACE_BUF_SIZE];
1343 struct ftrace_event_call *call = &event_print;
1344 struct ring_buffer_event *event;
1345 struct ring_buffer *buffer;
1346 struct trace_array_cpu *data;
1347 int cpu, len = 0, size, pc;
1348 struct print_entry *entry;
1349 unsigned long irq_flags;
1352 if (tracing_disabled || tracing_selftest_running)
1355 pc = preempt_count();
1356 preempt_disable_notrace();
1357 cpu = raw_smp_processor_id();
1358 data = tr->data[cpu];
1360 disable = atomic_inc_return(&data->disabled);
1361 if (unlikely(disable != 1))
1364 pause_graph_tracing();
1365 raw_local_irq_save(irq_flags);
1366 __raw_spin_lock(&trace_buf_lock);
1367 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1369 len = min(len, TRACE_BUF_SIZE-1);
1372 size = sizeof(*entry) + len + 1;
1373 buffer = tr->buffer;
1374 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1378 entry = ring_buffer_event_data(event);
1381 memcpy(&entry->buf, trace_buf, len);
1382 entry->buf[len] = 0;
1383 if (!filter_check_discard(call, entry, buffer, event))
1384 ring_buffer_unlock_commit(buffer, event);
1387 __raw_spin_unlock(&trace_buf_lock);
1388 raw_local_irq_restore(irq_flags);
1389 unpause_graph_tracing();
1391 atomic_dec_return(&data->disabled);
1392 preempt_enable_notrace();
1397 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1399 return trace_array_printk(&global_trace, ip, fmt, args);
1401 EXPORT_SYMBOL_GPL(trace_vprintk);
1403 enum trace_file_type {
1404 TRACE_FILE_LAT_FMT = 1,
1405 TRACE_FILE_ANNOTATE = 2,
1408 static void trace_iterator_increment(struct trace_iterator *iter)
1410 /* Don't allow ftrace to trace into the ring buffers */
1411 ftrace_disable_cpu();
1414 if (iter->buffer_iter[iter->cpu])
1415 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1417 ftrace_enable_cpu();
1420 static struct trace_entry *
1421 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1423 struct ring_buffer_event *event;
1424 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1426 /* Don't allow ftrace to trace into the ring buffers */
1427 ftrace_disable_cpu();
1430 event = ring_buffer_iter_peek(buf_iter, ts);
1432 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1434 ftrace_enable_cpu();
1436 return event ? ring_buffer_event_data(event) : NULL;
1439 static struct trace_entry *
1440 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1442 struct ring_buffer *buffer = iter->tr->buffer;
1443 struct trace_entry *ent, *next = NULL;
1444 int cpu_file = iter->cpu_file;
1445 u64 next_ts = 0, ts;
1450 * If we are in a per_cpu trace file, don't bother by iterating over
1451 * all cpu and peek directly.
1453 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1454 if (ring_buffer_empty_cpu(buffer, cpu_file))
1456 ent = peek_next_entry(iter, cpu_file, ent_ts);
1458 *ent_cpu = cpu_file;
1463 for_each_tracing_cpu(cpu) {
1465 if (ring_buffer_empty_cpu(buffer, cpu))
1468 ent = peek_next_entry(iter, cpu, &ts);
1471 * Pick the entry with the smallest timestamp:
1473 if (ent && (!next || ts < next_ts)) {
1481 *ent_cpu = next_cpu;
1489 /* Find the next real entry, without updating the iterator itself */
1490 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1491 int *ent_cpu, u64 *ent_ts)
1493 return __find_next_entry(iter, ent_cpu, ent_ts);
1496 /* Find the next real entry, and increment the iterator to the next entry */
1497 static void *find_next_entry_inc(struct trace_iterator *iter)
1499 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1502 trace_iterator_increment(iter);
1504 return iter->ent ? iter : NULL;
1507 static void trace_consume(struct trace_iterator *iter)
1509 /* Don't allow ftrace to trace into the ring buffers */
1510 ftrace_disable_cpu();
1511 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1512 ftrace_enable_cpu();
1515 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1517 struct trace_iterator *iter = m->private;
1523 /* can't go backwards */
1528 ent = find_next_entry_inc(iter);
1532 while (ent && iter->idx < i)
1533 ent = find_next_entry_inc(iter);
1540 static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1542 struct trace_array *tr = iter->tr;
1543 struct ring_buffer_event *event;
1544 struct ring_buffer_iter *buf_iter;
1545 unsigned long entries = 0;
1548 tr->data[cpu]->skipped_entries = 0;
1550 if (!iter->buffer_iter[cpu])
1553 buf_iter = iter->buffer_iter[cpu];
1554 ring_buffer_iter_reset(buf_iter);
1557 * We could have the case with the max latency tracers
1558 * that a reset never took place on a cpu. This is evident
1559 * by the timestamp being before the start of the buffer.
1561 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1562 if (ts >= iter->tr->time_start)
1565 ring_buffer_read(buf_iter, NULL);
1568 tr->data[cpu]->skipped_entries = entries;
1572 * No necessary locking here. The worst thing which can
1573 * happen is loosing events consumed at the same time
1574 * by a trace_pipe reader.
1575 * Other than that, we don't risk to crash the ring buffer
1576 * because it serializes the readers.
1578 * The current tracer is copied to avoid a global locking
1581 static void *s_start(struct seq_file *m, loff_t *pos)
1583 struct trace_iterator *iter = m->private;
1584 static struct tracer *old_tracer;
1585 int cpu_file = iter->cpu_file;
1590 /* copy the tracer to avoid using a global lock all around */
1591 mutex_lock(&trace_types_lock);
1592 if (unlikely(old_tracer != current_trace && current_trace)) {
1593 old_tracer = current_trace;
1594 *iter->trace = *current_trace;
1596 mutex_unlock(&trace_types_lock);
1598 atomic_inc(&trace_record_cmdline_disabled);
1600 if (*pos != iter->pos) {
1605 ftrace_disable_cpu();
1607 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1608 for_each_tracing_cpu(cpu)
1609 tracing_iter_reset(iter, cpu);
1611 tracing_iter_reset(iter, cpu_file);
1613 ftrace_enable_cpu();
1615 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1620 p = s_next(m, p, &l);
1623 trace_event_read_lock();
1627 static void s_stop(struct seq_file *m, void *p)
1629 atomic_dec(&trace_record_cmdline_disabled);
1630 trace_event_read_unlock();
1633 static void print_lat_help_header(struct seq_file *m)
1635 seq_puts(m, "# _------=> CPU# \n");
1636 seq_puts(m, "# / _-----=> irqs-off \n");
1637 seq_puts(m, "# | / _----=> need-resched \n");
1638 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1639 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1640 seq_puts(m, "# |||| /_--=> lock-depth \n");
1641 seq_puts(m, "# |||||/ delay \n");
1642 seq_puts(m, "# cmd pid |||||| time | caller \n");
1643 seq_puts(m, "# \\ / |||||| \\ | / \n");
1646 static void print_func_help_header(struct seq_file *m)
1648 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1649 seq_puts(m, "# | | | | |\n");
1654 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1656 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1657 struct trace_array *tr = iter->tr;
1658 struct trace_array_cpu *data = tr->data[tr->cpu];
1659 struct tracer *type = current_trace;
1660 unsigned long entries = 0;
1661 unsigned long total = 0;
1662 unsigned long count;
1663 const char *name = "preemption";
1670 for_each_tracing_cpu(cpu) {
1671 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1673 * If this buffer has skipped entries, then we hold all
1674 * entries for the trace and we need to ignore the
1675 * ones before the time stamp.
1677 if (tr->data[cpu]->skipped_entries) {
1678 count -= tr->data[cpu]->skipped_entries;
1679 /* total is the same as the entries */
1683 ring_buffer_overrun_cpu(tr->buffer, cpu);
1687 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1689 seq_puts(m, "# -----------------------------------"
1690 "---------------------------------\n");
1691 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1692 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1693 nsecs_to_usecs(data->saved_latency),
1697 #if defined(CONFIG_PREEMPT_NONE)
1699 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1701 #elif defined(CONFIG_PREEMPT)
1706 /* These are reserved for later use */
1709 seq_printf(m, " #P:%d)\n", num_online_cpus());
1713 seq_puts(m, "# -----------------\n");
1714 seq_printf(m, "# | task: %.16s-%d "
1715 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1716 data->comm, data->pid, data->uid, data->nice,
1717 data->policy, data->rt_priority);
1718 seq_puts(m, "# -----------------\n");
1720 if (data->critical_start) {
1721 seq_puts(m, "# => started at: ");
1722 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1723 trace_print_seq(m, &iter->seq);
1724 seq_puts(m, "\n# => ended at: ");
1725 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1726 trace_print_seq(m, &iter->seq);
1727 seq_puts(m, "\n#\n");
1733 static void test_cpu_buff_start(struct trace_iterator *iter)
1735 struct trace_seq *s = &iter->seq;
1737 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1740 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1743 if (cpumask_test_cpu(iter->cpu, iter->started))
1746 if (iter->tr->data[iter->cpu]->skipped_entries)
1749 cpumask_set_cpu(iter->cpu, iter->started);
1751 /* Don't print started cpu buffer for the first entry of the trace */
1753 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1757 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1759 struct trace_seq *s = &iter->seq;
1760 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1761 struct trace_entry *entry;
1762 struct trace_event *event;
1766 test_cpu_buff_start(iter);
1768 event = ftrace_find_event(entry->type);
1770 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1771 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1772 if (!trace_print_lat_context(iter))
1775 if (!trace_print_context(iter))
1781 return event->trace(iter, sym_flags);
1783 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1786 return TRACE_TYPE_HANDLED;
1788 return TRACE_TYPE_PARTIAL_LINE;
1791 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1793 struct trace_seq *s = &iter->seq;
1794 struct trace_entry *entry;
1795 struct trace_event *event;
1799 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1800 if (!trace_seq_printf(s, "%d %d %llu ",
1801 entry->pid, iter->cpu, iter->ts))
1805 event = ftrace_find_event(entry->type);
1807 return event->raw(iter, 0);
1809 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1812 return TRACE_TYPE_HANDLED;
1814 return TRACE_TYPE_PARTIAL_LINE;
1817 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1819 struct trace_seq *s = &iter->seq;
1820 unsigned char newline = '\n';
1821 struct trace_entry *entry;
1822 struct trace_event *event;
1826 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1827 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1828 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1829 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1832 event = ftrace_find_event(entry->type);
1834 enum print_line_t ret = event->hex(iter, 0);
1835 if (ret != TRACE_TYPE_HANDLED)
1839 SEQ_PUT_FIELD_RET(s, newline);
1841 return TRACE_TYPE_HANDLED;
1844 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1846 struct trace_seq *s = &iter->seq;
1847 struct trace_entry *entry;
1848 struct trace_event *event;
1852 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1853 SEQ_PUT_FIELD_RET(s, entry->pid);
1854 SEQ_PUT_FIELD_RET(s, iter->cpu);
1855 SEQ_PUT_FIELD_RET(s, iter->ts);
1858 event = ftrace_find_event(entry->type);
1859 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1862 static int trace_empty(struct trace_iterator *iter)
1866 /* If we are looking at one CPU buffer, only check that one */
1867 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1868 cpu = iter->cpu_file;
1869 if (iter->buffer_iter[cpu]) {
1870 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1873 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1879 for_each_tracing_cpu(cpu) {
1880 if (iter->buffer_iter[cpu]) {
1881 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1884 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1892 /* Called with trace_event_read_lock() held. */
1893 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1895 enum print_line_t ret;
1897 if (iter->trace && iter->trace->print_line) {
1898 ret = iter->trace->print_line(iter);
1899 if (ret != TRACE_TYPE_UNHANDLED)
1903 if (iter->ent->type == TRACE_BPRINT &&
1904 trace_flags & TRACE_ITER_PRINTK &&
1905 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1906 return trace_print_bprintk_msg_only(iter);
1908 if (iter->ent->type == TRACE_PRINT &&
1909 trace_flags & TRACE_ITER_PRINTK &&
1910 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1911 return trace_print_printk_msg_only(iter);
1913 if (trace_flags & TRACE_ITER_BIN)
1914 return print_bin_fmt(iter);
1916 if (trace_flags & TRACE_ITER_HEX)
1917 return print_hex_fmt(iter);
1919 if (trace_flags & TRACE_ITER_RAW)
1920 return print_raw_fmt(iter);
1922 return print_trace_fmt(iter);
1925 static int s_show(struct seq_file *m, void *v)
1927 struct trace_iterator *iter = v;
1929 if (iter->ent == NULL) {
1931 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1934 if (iter->trace && iter->trace->print_header)
1935 iter->trace->print_header(m);
1936 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1937 /* print nothing if the buffers are empty */
1938 if (trace_empty(iter))
1940 print_trace_header(m, iter);
1941 if (!(trace_flags & TRACE_ITER_VERBOSE))
1942 print_lat_help_header(m);
1944 if (!(trace_flags & TRACE_ITER_VERBOSE))
1945 print_func_help_header(m);
1948 print_trace_line(iter);
1949 trace_print_seq(m, &iter->seq);
1955 static struct seq_operations tracer_seq_ops = {
1962 static struct trace_iterator *
1963 __tracing_open(struct inode *inode, struct file *file)
1965 long cpu_file = (long) inode->i_private;
1966 void *fail_ret = ERR_PTR(-ENOMEM);
1967 struct trace_iterator *iter;
1971 if (tracing_disabled)
1972 return ERR_PTR(-ENODEV);
1974 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1976 return ERR_PTR(-ENOMEM);
1979 * We make a copy of the current tracer to avoid concurrent
1980 * changes on it while we are reading.
1982 mutex_lock(&trace_types_lock);
1983 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
1988 *iter->trace = *current_trace;
1990 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL))
1993 cpumask_clear(iter->started);
1995 if (current_trace && current_trace->print_max)
1998 iter->tr = &global_trace;
2000 mutex_init(&iter->mutex);
2001 iter->cpu_file = cpu_file;
2003 /* Notify the tracer early; before we stop tracing. */
2004 if (iter->trace && iter->trace->open)
2005 iter->trace->open(iter);
2007 /* Annotate start of buffers if we had overruns */
2008 if (ring_buffer_overruns(iter->tr->buffer))
2009 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2011 /* stop the trace while dumping */
2014 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2015 for_each_tracing_cpu(cpu) {
2017 iter->buffer_iter[cpu] =
2018 ring_buffer_read_start(iter->tr->buffer, cpu);
2019 tracing_iter_reset(iter, cpu);
2022 cpu = iter->cpu_file;
2023 iter->buffer_iter[cpu] =
2024 ring_buffer_read_start(iter->tr->buffer, cpu);
2025 tracing_iter_reset(iter, cpu);
2028 ret = seq_open(file, &tracer_seq_ops);
2030 fail_ret = ERR_PTR(ret);
2034 m = file->private_data;
2037 mutex_unlock(&trace_types_lock);
2042 for_each_tracing_cpu(cpu) {
2043 if (iter->buffer_iter[cpu])
2044 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2046 free_cpumask_var(iter->started);
2049 mutex_unlock(&trace_types_lock);
2056 int tracing_open_generic(struct inode *inode, struct file *filp)
2058 if (tracing_disabled)
2061 filp->private_data = inode->i_private;
2065 static int tracing_release(struct inode *inode, struct file *file)
2067 struct seq_file *m = (struct seq_file *)file->private_data;
2068 struct trace_iterator *iter;
2071 if (!(file->f_mode & FMODE_READ))
2076 mutex_lock(&trace_types_lock);
2077 for_each_tracing_cpu(cpu) {
2078 if (iter->buffer_iter[cpu])
2079 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2082 if (iter->trace && iter->trace->close)
2083 iter->trace->close(iter);
2085 /* reenable tracing if it was previously enabled */
2087 mutex_unlock(&trace_types_lock);
2089 seq_release(inode, file);
2090 mutex_destroy(&iter->mutex);
2091 free_cpumask_var(iter->started);
2097 static int tracing_open(struct inode *inode, struct file *file)
2099 struct trace_iterator *iter;
2102 /* If this file was open for write, then erase contents */
2103 if ((file->f_mode & FMODE_WRITE) &&
2104 (file->f_flags & O_TRUNC)) {
2105 long cpu = (long) inode->i_private;
2107 if (cpu == TRACE_PIPE_ALL_CPU)
2108 tracing_reset_online_cpus(&global_trace);
2110 tracing_reset(&global_trace, cpu);
2113 if (file->f_mode & FMODE_READ) {
2114 iter = __tracing_open(inode, file);
2116 ret = PTR_ERR(iter);
2117 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2118 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2124 t_next(struct seq_file *m, void *v, loff_t *pos)
2126 struct tracer *t = v;
2136 static void *t_start(struct seq_file *m, loff_t *pos)
2141 mutex_lock(&trace_types_lock);
2142 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2148 static void t_stop(struct seq_file *m, void *p)
2150 mutex_unlock(&trace_types_lock);
2153 static int t_show(struct seq_file *m, void *v)
2155 struct tracer *t = v;
2160 seq_printf(m, "%s", t->name);
2169 static struct seq_operations show_traces_seq_ops = {
2176 static int show_traces_open(struct inode *inode, struct file *file)
2178 if (tracing_disabled)
2181 return seq_open(file, &show_traces_seq_ops);
2185 tracing_write_stub(struct file *filp, const char __user *ubuf,
2186 size_t count, loff_t *ppos)
2191 static const struct file_operations tracing_fops = {
2192 .open = tracing_open,
2194 .write = tracing_write_stub,
2195 .llseek = seq_lseek,
2196 .release = tracing_release,
2199 static const struct file_operations show_traces_fops = {
2200 .open = show_traces_open,
2202 .release = seq_release,
2206 * Only trace on a CPU if the bitmask is set:
2208 static cpumask_var_t tracing_cpumask;
2211 * The tracer itself will not take this lock, but still we want
2212 * to provide a consistent cpumask to user-space:
2214 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2217 * Temporary storage for the character representation of the
2218 * CPU bitmask (and one more byte for the newline):
2220 static char mask_str[NR_CPUS + 1];
2223 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2224 size_t count, loff_t *ppos)
2228 mutex_lock(&tracing_cpumask_update_lock);
2230 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2231 if (count - len < 2) {
2235 len += sprintf(mask_str + len, "\n");
2236 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2239 mutex_unlock(&tracing_cpumask_update_lock);
2245 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2246 size_t count, loff_t *ppos)
2249 cpumask_var_t tracing_cpumask_new;
2251 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2254 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2258 mutex_lock(&tracing_cpumask_update_lock);
2260 local_irq_disable();
2261 __raw_spin_lock(&ftrace_max_lock);
2262 for_each_tracing_cpu(cpu) {
2264 * Increase/decrease the disabled counter if we are
2265 * about to flip a bit in the cpumask:
2267 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2268 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2269 atomic_inc(&global_trace.data[cpu]->disabled);
2271 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2272 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2273 atomic_dec(&global_trace.data[cpu]->disabled);
2276 __raw_spin_unlock(&ftrace_max_lock);
2279 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2281 mutex_unlock(&tracing_cpumask_update_lock);
2282 free_cpumask_var(tracing_cpumask_new);
2287 free_cpumask_var(tracing_cpumask_new);
2292 static const struct file_operations tracing_cpumask_fops = {
2293 .open = tracing_open_generic,
2294 .read = tracing_cpumask_read,
2295 .write = tracing_cpumask_write,
2299 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2300 size_t cnt, loff_t *ppos)
2302 struct tracer_opt *trace_opts;
2310 /* calculate max size */
2311 for (i = 0; trace_options[i]; i++) {
2312 len += strlen(trace_options[i]);
2313 len += 3; /* "no" and newline */
2316 mutex_lock(&trace_types_lock);
2317 tracer_flags = current_trace->flags->val;
2318 trace_opts = current_trace->flags->opts;
2321 * Increase the size with names of options specific
2322 * of the current tracer.
2324 for (i = 0; trace_opts[i].name; i++) {
2325 len += strlen(trace_opts[i].name);
2326 len += 3; /* "no" and newline */
2330 buf = kmalloc(len + 1, GFP_KERNEL);
2332 mutex_unlock(&trace_types_lock);
2336 for (i = 0; trace_options[i]; i++) {
2337 if (trace_flags & (1 << i))
2338 r += sprintf(buf + r, "%s\n", trace_options[i]);
2340 r += sprintf(buf + r, "no%s\n", trace_options[i]);
2343 for (i = 0; trace_opts[i].name; i++) {
2344 if (tracer_flags & trace_opts[i].bit)
2345 r += sprintf(buf + r, "%s\n",
2346 trace_opts[i].name);
2348 r += sprintf(buf + r, "no%s\n",
2349 trace_opts[i].name);
2351 mutex_unlock(&trace_types_lock);
2353 WARN_ON(r >= len + 1);
2355 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2361 /* Try to assign a tracer specific option */
2362 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2364 struct tracer_flags *tracer_flags = trace->flags;
2365 struct tracer_opt *opts = NULL;
2369 for (i = 0; tracer_flags->opts[i].name; i++) {
2370 opts = &tracer_flags->opts[i];
2371 len = strlen(opts->name);
2373 if (strncmp(cmp, opts->name, len) == 0) {
2374 ret = trace->set_flag(tracer_flags->val,
2380 if (!tracer_flags->opts[i].name)
2383 /* Refused to handle */
2388 tracer_flags->val &= ~opts->bit;
2390 tracer_flags->val |= opts->bit;
2395 static void set_tracer_flags(unsigned int mask, int enabled)
2397 /* do nothing if flag is already set */
2398 if (!!(trace_flags & mask) == !!enabled)
2402 trace_flags |= mask;
2404 trace_flags &= ~mask;
2408 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2409 size_t cnt, loff_t *ppos)
2417 if (cnt >= sizeof(buf))
2420 if (copy_from_user(&buf, ubuf, cnt))
2425 if (strncmp(buf, "no", 2) == 0) {
2430 for (i = 0; trace_options[i]; i++) {
2431 int len = strlen(trace_options[i]);
2433 if (strncmp(cmp, trace_options[i], len) == 0) {
2434 set_tracer_flags(1 << i, !neg);
2439 /* If no option could be set, test the specific tracer options */
2440 if (!trace_options[i]) {
2441 mutex_lock(&trace_types_lock);
2442 ret = set_tracer_option(current_trace, cmp, neg);
2443 mutex_unlock(&trace_types_lock);
2453 static const struct file_operations tracing_iter_fops = {
2454 .open = tracing_open_generic,
2455 .read = tracing_trace_options_read,
2456 .write = tracing_trace_options_write,
2459 static const char readme_msg[] =
2460 "tracing mini-HOWTO:\n\n"
2461 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2462 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2463 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2464 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2466 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2467 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2469 "# cat /sys/kernel/debug/tracing/trace_options\n"
2470 "noprint-parent nosym-offset nosym-addr noverbose\n"
2471 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2472 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2473 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2474 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2478 tracing_readme_read(struct file *filp, char __user *ubuf,
2479 size_t cnt, loff_t *ppos)
2481 return simple_read_from_buffer(ubuf, cnt, ppos,
2482 readme_msg, strlen(readme_msg));
2485 static const struct file_operations tracing_readme_fops = {
2486 .open = tracing_open_generic,
2487 .read = tracing_readme_read,
2491 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2492 size_t cnt, loff_t *ppos)
2501 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2505 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2513 for (i = 0; i < SAVED_CMDLINES; i++) {
2516 pid = map_cmdline_to_pid[i];
2517 if (pid == -1 || pid == NO_CMDLINE_MAP)
2520 trace_find_cmdline(pid, buf_comm);
2521 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2526 len = simple_read_from_buffer(ubuf, cnt, ppos,
2535 static const struct file_operations tracing_saved_cmdlines_fops = {
2536 .open = tracing_open_generic,
2537 .read = tracing_saved_cmdlines_read,
2541 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2542 size_t cnt, loff_t *ppos)
2547 r = sprintf(buf, "%u\n", tracer_enabled);
2548 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2552 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2553 size_t cnt, loff_t *ppos)
2555 struct trace_array *tr = filp->private_data;
2560 if (cnt >= sizeof(buf))
2563 if (copy_from_user(&buf, ubuf, cnt))
2568 ret = strict_strtoul(buf, 10, &val);
2574 mutex_lock(&trace_types_lock);
2575 if (tracer_enabled ^ val) {
2578 if (current_trace->start)
2579 current_trace->start(tr);
2584 if (current_trace->stop)
2585 current_trace->stop(tr);
2588 mutex_unlock(&trace_types_lock);
2596 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2597 size_t cnt, loff_t *ppos)
2599 char buf[max_tracer_type_len+2];
2602 mutex_lock(&trace_types_lock);
2604 r = sprintf(buf, "%s\n", current_trace->name);
2606 r = sprintf(buf, "\n");
2607 mutex_unlock(&trace_types_lock);
2609 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2612 int tracer_init(struct tracer *t, struct trace_array *tr)
2614 tracing_reset_online_cpus(tr);
2618 static int tracing_resize_ring_buffer(unsigned long size)
2623 * If kernel or user changes the size of the ring buffer
2624 * we use the size that was given, and we can forget about
2625 * expanding it later.
2627 ring_buffer_expanded = 1;
2629 ret = ring_buffer_resize(global_trace.buffer, size);
2633 ret = ring_buffer_resize(max_tr.buffer, size);
2637 r = ring_buffer_resize(global_trace.buffer,
2638 global_trace.entries);
2641 * AARGH! We are left with different
2642 * size max buffer!!!!
2643 * The max buffer is our "snapshot" buffer.
2644 * When a tracer needs a snapshot (one of the
2645 * latency tracers), it swaps the max buffer
2646 * with the saved snap shot. We succeeded to
2647 * update the size of the main buffer, but failed to
2648 * update the size of the max buffer. But when we tried
2649 * to reset the main buffer to the original size, we
2650 * failed there too. This is very unlikely to
2651 * happen, but if it does, warn and kill all
2655 tracing_disabled = 1;
2660 global_trace.entries = size;
2666 * tracing_update_buffers - used by tracing facility to expand ring buffers
2668 * To save on memory when the tracing is never used on a system with it
2669 * configured in. The ring buffers are set to a minimum size. But once
2670 * a user starts to use the tracing facility, then they need to grow
2671 * to their default size.
2673 * This function is to be called when a tracer is about to be used.
2675 int tracing_update_buffers(void)
2679 mutex_lock(&trace_types_lock);
2680 if (!ring_buffer_expanded)
2681 ret = tracing_resize_ring_buffer(trace_buf_size);
2682 mutex_unlock(&trace_types_lock);
2687 struct trace_option_dentry;
2689 static struct trace_option_dentry *
2690 create_trace_option_files(struct tracer *tracer);
2693 destroy_trace_option_files(struct trace_option_dentry *topts);
2695 static int tracing_set_tracer(const char *buf)
2697 static struct trace_option_dentry *topts;
2698 struct trace_array *tr = &global_trace;
2702 mutex_lock(&trace_types_lock);
2704 if (!ring_buffer_expanded) {
2705 ret = tracing_resize_ring_buffer(trace_buf_size);
2711 for (t = trace_types; t; t = t->next) {
2712 if (strcmp(t->name, buf) == 0)
2719 if (t == current_trace)
2722 trace_branch_disable();
2723 if (current_trace && current_trace->reset)
2724 current_trace->reset(tr);
2726 destroy_trace_option_files(topts);
2730 topts = create_trace_option_files(current_trace);
2733 ret = tracer_init(t, tr);
2738 trace_branch_enable(tr);
2740 mutex_unlock(&trace_types_lock);
2746 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2747 size_t cnt, loff_t *ppos)
2749 char buf[max_tracer_type_len+1];
2756 if (cnt > max_tracer_type_len)
2757 cnt = max_tracer_type_len;
2759 if (copy_from_user(&buf, ubuf, cnt))
2764 /* strip ending whitespace. */
2765 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2768 err = tracing_set_tracer(buf);
2778 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2779 size_t cnt, loff_t *ppos)
2781 unsigned long *ptr = filp->private_data;
2785 r = snprintf(buf, sizeof(buf), "%ld\n",
2786 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2787 if (r > sizeof(buf))
2789 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2793 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2794 size_t cnt, loff_t *ppos)
2796 unsigned long *ptr = filp->private_data;
2801 if (cnt >= sizeof(buf))
2804 if (copy_from_user(&buf, ubuf, cnt))
2809 ret = strict_strtoul(buf, 10, &val);
2818 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2820 long cpu_file = (long) inode->i_private;
2821 struct trace_iterator *iter;
2824 if (tracing_disabled)
2827 mutex_lock(&trace_types_lock);
2829 /* We only allow one reader per cpu */
2830 if (cpu_file == TRACE_PIPE_ALL_CPU) {
2831 if (!cpumask_empty(tracing_reader_cpumask)) {
2835 cpumask_setall(tracing_reader_cpumask);
2837 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2838 cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2845 /* create a buffer to store the information to pass to userspace */
2846 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2853 * We make a copy of the current tracer to avoid concurrent
2854 * changes on it while we are reading.
2856 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2862 *iter->trace = *current_trace;
2864 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2869 /* trace pipe does not show start of buffer */
2870 cpumask_setall(iter->started);
2872 if (trace_flags & TRACE_ITER_LATENCY_FMT)
2873 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2875 iter->cpu_file = cpu_file;
2876 iter->tr = &global_trace;
2877 mutex_init(&iter->mutex);
2878 filp->private_data = iter;
2880 if (iter->trace->pipe_open)
2881 iter->trace->pipe_open(iter);
2884 mutex_unlock(&trace_types_lock);
2890 mutex_unlock(&trace_types_lock);
2894 static int tracing_release_pipe(struct inode *inode, struct file *file)
2896 struct trace_iterator *iter = file->private_data;
2898 mutex_lock(&trace_types_lock);
2900 if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2901 cpumask_clear(tracing_reader_cpumask);
2903 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2905 mutex_unlock(&trace_types_lock);
2907 free_cpumask_var(iter->started);
2908 mutex_destroy(&iter->mutex);
2916 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2918 struct trace_iterator *iter = filp->private_data;
2920 if (trace_flags & TRACE_ITER_BLOCK) {
2922 * Always select as readable when in blocking mode
2924 return POLLIN | POLLRDNORM;
2926 if (!trace_empty(iter))
2927 return POLLIN | POLLRDNORM;
2928 poll_wait(filp, &trace_wait, poll_table);
2929 if (!trace_empty(iter))
2930 return POLLIN | POLLRDNORM;
2937 void default_wait_pipe(struct trace_iterator *iter)
2941 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2943 if (trace_empty(iter))
2946 finish_wait(&trace_wait, &wait);
2950 * This is a make-shift waitqueue.
2951 * A tracer might use this callback on some rare cases:
2953 * 1) the current tracer might hold the runqueue lock when it wakes up
2954 * a reader, hence a deadlock (sched, function, and function graph tracers)
2955 * 2) the function tracers, trace all functions, we don't want
2956 * the overhead of calling wake_up and friends
2957 * (and tracing them too)
2959 * Anyway, this is really very primitive wakeup.
2961 void poll_wait_pipe(struct trace_iterator *iter)
2963 set_current_state(TASK_INTERRUPTIBLE);
2964 /* sleep for 100 msecs, and try again. */
2965 schedule_timeout(HZ / 10);
2968 /* Must be called with trace_types_lock mutex held. */
2969 static int tracing_wait_pipe(struct file *filp)
2971 struct trace_iterator *iter = filp->private_data;
2973 while (trace_empty(iter)) {
2975 if ((filp->f_flags & O_NONBLOCK)) {
2979 mutex_unlock(&iter->mutex);
2981 iter->trace->wait_pipe(iter);
2983 mutex_lock(&iter->mutex);
2985 if (signal_pending(current))
2989 * We block until we read something and tracing is disabled.
2990 * We still block if tracing is disabled, but we have never
2991 * read anything. This allows a user to cat this file, and
2992 * then enable tracing. But after we have read something,
2993 * we give an EOF when tracing is again disabled.
2995 * iter->pos will be 0 if we haven't read anything.
2997 if (!tracer_enabled && iter->pos)
3008 tracing_read_pipe(struct file *filp, char __user *ubuf,
3009 size_t cnt, loff_t *ppos)
3011 struct trace_iterator *iter = filp->private_data;
3012 static struct tracer *old_tracer;
3015 /* return any leftover data */
3016 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3020 trace_seq_init(&iter->seq);
3022 /* copy the tracer to avoid using a global lock all around */
3023 mutex_lock(&trace_types_lock);
3024 if (unlikely(old_tracer != current_trace && current_trace)) {
3025 old_tracer = current_trace;
3026 *iter->trace = *current_trace;
3028 mutex_unlock(&trace_types_lock);
3031 * Avoid more than one consumer on a single file descriptor
3032 * This is just a matter of traces coherency, the ring buffer itself
3035 mutex_lock(&iter->mutex);
3036 if (iter->trace->read) {
3037 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3043 sret = tracing_wait_pipe(filp);
3047 /* stop when tracing is finished */
3048 if (trace_empty(iter)) {
3053 if (cnt >= PAGE_SIZE)
3054 cnt = PAGE_SIZE - 1;
3056 /* reset all but tr, trace, and overruns */
3057 memset(&iter->seq, 0,
3058 sizeof(struct trace_iterator) -
3059 offsetof(struct trace_iterator, seq));
3062 trace_event_read_lock();
3063 while (find_next_entry_inc(iter) != NULL) {
3064 enum print_line_t ret;
3065 int len = iter->seq.len;
3067 ret = print_trace_line(iter);
3068 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3069 /* don't print partial lines */
3070 iter->seq.len = len;
3073 if (ret != TRACE_TYPE_NO_CONSUME)
3074 trace_consume(iter);
3076 if (iter->seq.len >= cnt)
3079 trace_event_read_unlock();
3081 /* Now copy what we have to the user */
3082 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3083 if (iter->seq.readpos >= iter->seq.len)
3084 trace_seq_init(&iter->seq);
3087 * If there was nothing to send to user, inspite of consuming trace
3088 * entries, go back to wait for more entries.
3094 mutex_unlock(&iter->mutex);
3099 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3100 struct pipe_buffer *buf)
3102 __free_page(buf->page);
3105 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3108 __free_page(spd->pages[idx]);
3111 static struct pipe_buf_operations tracing_pipe_buf_ops = {
3113 .map = generic_pipe_buf_map,
3114 .unmap = generic_pipe_buf_unmap,
3115 .confirm = generic_pipe_buf_confirm,
3116 .release = tracing_pipe_buf_release,
3117 .steal = generic_pipe_buf_steal,
3118 .get = generic_pipe_buf_get,
3122 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3127 /* Seq buffer is page-sized, exactly what we need. */
3129 count = iter->seq.len;
3130 ret = print_trace_line(iter);
3131 count = iter->seq.len - count;
3134 iter->seq.len -= count;
3137 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3138 iter->seq.len -= count;
3142 if (ret != TRACE_TYPE_NO_CONSUME)
3143 trace_consume(iter);
3145 if (!find_next_entry_inc(iter)) {
3155 static ssize_t tracing_splice_read_pipe(struct file *filp,
3157 struct pipe_inode_info *pipe,
3161 struct page *pages[PIPE_BUFFERS];
3162 struct partial_page partial[PIPE_BUFFERS];
3163 struct trace_iterator *iter = filp->private_data;
3164 struct splice_pipe_desc spd = {
3167 .nr_pages = 0, /* This gets updated below. */
3169 .ops = &tracing_pipe_buf_ops,
3170 .spd_release = tracing_spd_release_pipe,
3172 static struct tracer *old_tracer;
3177 /* copy the tracer to avoid using a global lock all around */
3178 mutex_lock(&trace_types_lock);
3179 if (unlikely(old_tracer != current_trace && current_trace)) {
3180 old_tracer = current_trace;
3181 *iter->trace = *current_trace;
3183 mutex_unlock(&trace_types_lock);
3185 mutex_lock(&iter->mutex);
3187 if (iter->trace->splice_read) {
3188 ret = iter->trace->splice_read(iter, filp,
3189 ppos, pipe, len, flags);
3194 ret = tracing_wait_pipe(filp);
3198 if (!iter->ent && !find_next_entry_inc(iter)) {
3203 trace_event_read_lock();
3205 /* Fill as many pages as possible. */
3206 for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3207 pages[i] = alloc_page(GFP_KERNEL);
3211 rem = tracing_fill_pipe_page(rem, iter);
3213 /* Copy the data into the page, so we can start over. */
3214 ret = trace_seq_to_buffer(&iter->seq,
3215 page_address(pages[i]),
3218 __free_page(pages[i]);
3221 partial[i].offset = 0;
3222 partial[i].len = iter->seq.len;
3224 trace_seq_init(&iter->seq);
3227 trace_event_read_unlock();
3228 mutex_unlock(&iter->mutex);
3232 return splice_to_pipe(pipe, &spd);
3235 mutex_unlock(&iter->mutex);
3241 tracing_entries_read(struct file *filp, char __user *ubuf,
3242 size_t cnt, loff_t *ppos)
3244 struct trace_array *tr = filp->private_data;
3248 mutex_lock(&trace_types_lock);
3249 if (!ring_buffer_expanded)
3250 r = sprintf(buf, "%lu (expanded: %lu)\n",
3252 trace_buf_size >> 10);
3254 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3255 mutex_unlock(&trace_types_lock);
3257 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3261 tracing_entries_write(struct file *filp, const char __user *ubuf,
3262 size_t cnt, loff_t *ppos)
3268 if (cnt >= sizeof(buf))
3271 if (copy_from_user(&buf, ubuf, cnt))
3276 ret = strict_strtoul(buf, 10, &val);
3280 /* must have at least 1 entry */
3284 mutex_lock(&trace_types_lock);
3288 /* disable all cpu buffers */
3289 for_each_tracing_cpu(cpu) {
3290 if (global_trace.data[cpu])
3291 atomic_inc(&global_trace.data[cpu]->disabled);
3292 if (max_tr.data[cpu])
3293 atomic_inc(&max_tr.data[cpu]->disabled);
3296 /* value is in KB */
3299 if (val != global_trace.entries) {
3300 ret = tracing_resize_ring_buffer(val);
3309 /* If check pages failed, return ENOMEM */
3310 if (tracing_disabled)
3313 for_each_tracing_cpu(cpu) {
3314 if (global_trace.data[cpu])
3315 atomic_dec(&global_trace.data[cpu]->disabled);
3316 if (max_tr.data[cpu])
3317 atomic_dec(&max_tr.data[cpu]->disabled);
3321 max_tr.entries = global_trace.entries;
3322 mutex_unlock(&trace_types_lock);
3327 static int mark_printk(const char *fmt, ...)
3331 va_start(args, fmt);
3332 ret = trace_vprintk(0, fmt, args);
3338 tracing_mark_write(struct file *filp, const char __user *ubuf,
3339 size_t cnt, loff_t *fpos)
3344 if (tracing_disabled)
3347 if (cnt > TRACE_BUF_SIZE)
3348 cnt = TRACE_BUF_SIZE;
3350 buf = kmalloc(cnt + 1, GFP_KERNEL);
3354 if (copy_from_user(buf, ubuf, cnt)) {
3359 /* Cut from the first nil or newline. */
3361 end = strchr(buf, '\n');
3365 cnt = mark_printk("%s\n", buf);
3372 static ssize_t tracing_clock_read(struct file *filp, char __user *ubuf,
3373 size_t cnt, loff_t *ppos)
3379 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3380 bufiter += snprintf(buf + bufiter, sizeof(buf) - bufiter,
3381 "%s%s%s%s", i ? " " : "",
3382 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3383 i == trace_clock_id ? "]" : "");
3384 bufiter += snprintf(buf + bufiter, sizeof(buf) - bufiter, "\n");
3386 return simple_read_from_buffer(ubuf, cnt, ppos, buf, bufiter);
3389 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3390 size_t cnt, loff_t *fpos)
3393 const char *clockstr;
3396 if (cnt >= sizeof(buf))
3399 if (copy_from_user(&buf, ubuf, cnt))
3404 clockstr = strstrip(buf);
3406 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3407 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3410 if (i == ARRAY_SIZE(trace_clocks))
3415 mutex_lock(&trace_types_lock);
3417 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3419 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3421 mutex_unlock(&trace_types_lock);
3428 static const struct file_operations tracing_max_lat_fops = {
3429 .open = tracing_open_generic,
3430 .read = tracing_max_lat_read,
3431 .write = tracing_max_lat_write,
3434 static const struct file_operations tracing_ctrl_fops = {
3435 .open = tracing_open_generic,
3436 .read = tracing_ctrl_read,
3437 .write = tracing_ctrl_write,
3440 static const struct file_operations set_tracer_fops = {
3441 .open = tracing_open_generic,
3442 .read = tracing_set_trace_read,
3443 .write = tracing_set_trace_write,
3446 static const struct file_operations tracing_pipe_fops = {
3447 .open = tracing_open_pipe,
3448 .poll = tracing_poll_pipe,
3449 .read = tracing_read_pipe,
3450 .splice_read = tracing_splice_read_pipe,
3451 .release = tracing_release_pipe,
3454 static const struct file_operations tracing_entries_fops = {
3455 .open = tracing_open_generic,
3456 .read = tracing_entries_read,
3457 .write = tracing_entries_write,
3460 static const struct file_operations tracing_mark_fops = {
3461 .open = tracing_open_generic,
3462 .write = tracing_mark_write,
3465 static const struct file_operations trace_clock_fops = {
3466 .open = tracing_open_generic,
3467 .read = tracing_clock_read,
3468 .write = tracing_clock_write,
3471 struct ftrace_buffer_info {
3472 struct trace_array *tr;
3478 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3480 int cpu = (int)(long)inode->i_private;
3481 struct ftrace_buffer_info *info;
3483 if (tracing_disabled)
3486 info = kzalloc(sizeof(*info), GFP_KERNEL);
3490 info->tr = &global_trace;
3493 /* Force reading ring buffer for first read */
3494 info->read = (unsigned int)-1;
3496 filp->private_data = info;
3498 return nonseekable_open(inode, filp);
3502 tracing_buffers_read(struct file *filp, char __user *ubuf,
3503 size_t count, loff_t *ppos)
3505 struct ftrace_buffer_info *info = filp->private_data;
3514 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3518 /* Do we have previous read data to read? */
3519 if (info->read < PAGE_SIZE)
3524 ret = ring_buffer_read_page(info->tr->buffer,
3531 pos = ring_buffer_page_len(info->spare);
3533 if (pos < PAGE_SIZE)
3534 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3537 size = PAGE_SIZE - info->read;
3541 ret = copy_to_user(ubuf, info->spare + info->read, size);
3552 static int tracing_buffers_release(struct inode *inode, struct file *file)
3554 struct ftrace_buffer_info *info = file->private_data;
3557 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3564 struct ring_buffer *buffer;
3569 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3570 struct pipe_buffer *buf)
3572 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3577 ring_buffer_free_read_page(ref->buffer, ref->page);
3582 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3583 struct pipe_buffer *buf)
3588 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3589 struct pipe_buffer *buf)
3591 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3596 /* Pipe buffer operations for a buffer. */
3597 static struct pipe_buf_operations buffer_pipe_buf_ops = {
3599 .map = generic_pipe_buf_map,
3600 .unmap = generic_pipe_buf_unmap,
3601 .confirm = generic_pipe_buf_confirm,
3602 .release = buffer_pipe_buf_release,
3603 .steal = buffer_pipe_buf_steal,
3604 .get = buffer_pipe_buf_get,
3608 * Callback from splice_to_pipe(), if we need to release some pages
3609 * at the end of the spd in case we error'ed out in filling the pipe.
3611 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3613 struct buffer_ref *ref =
3614 (struct buffer_ref *)spd->partial[i].private;
3619 ring_buffer_free_read_page(ref->buffer, ref->page);
3621 spd->partial[i].private = 0;
3625 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3626 struct pipe_inode_info *pipe, size_t len,
3629 struct ftrace_buffer_info *info = file->private_data;
3630 struct partial_page partial[PIPE_BUFFERS];
3631 struct page *pages[PIPE_BUFFERS];
3632 struct splice_pipe_desc spd = {
3636 .ops = &buffer_pipe_buf_ops,
3637 .spd_release = buffer_spd_release,
3639 struct buffer_ref *ref;
3640 int entries, size, i;
3643 if (*ppos & (PAGE_SIZE - 1)) {
3644 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3648 if (len & (PAGE_SIZE - 1)) {
3649 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3650 if (len < PAGE_SIZE)
3655 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3657 for (i = 0; i < PIPE_BUFFERS && len && entries; i++, len -= PAGE_SIZE) {
3661 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3666 ref->buffer = info->tr->buffer;
3667 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3673 r = ring_buffer_read_page(ref->buffer, &ref->page,
3676 ring_buffer_free_read_page(ref->buffer,
3683 * zero out any left over data, this is going to
3686 size = ring_buffer_page_len(ref->page);
3687 if (size < PAGE_SIZE)
3688 memset(ref->page + size, 0, PAGE_SIZE - size);
3690 page = virt_to_page(ref->page);
3692 spd.pages[i] = page;
3693 spd.partial[i].len = PAGE_SIZE;
3694 spd.partial[i].offset = 0;
3695 spd.partial[i].private = (unsigned long)ref;
3699 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3704 /* did we read anything? */
3705 if (!spd.nr_pages) {
3706 if (flags & SPLICE_F_NONBLOCK)
3714 ret = splice_to_pipe(pipe, &spd);
3719 static const struct file_operations tracing_buffers_fops = {
3720 .open = tracing_buffers_open,
3721 .read = tracing_buffers_read,
3722 .release = tracing_buffers_release,
3723 .splice_read = tracing_buffers_splice_read,
3724 .llseek = no_llseek,
3728 tracing_stats_read(struct file *filp, char __user *ubuf,
3729 size_t count, loff_t *ppos)
3731 unsigned long cpu = (unsigned long)filp->private_data;
3732 struct trace_array *tr = &global_trace;
3733 struct trace_seq *s;
3736 s = kmalloc(sizeof(*s), GFP_KERNEL);
3742 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3743 trace_seq_printf(s, "entries: %ld\n", cnt);
3745 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3746 trace_seq_printf(s, "overrun: %ld\n", cnt);
3748 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3749 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3751 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3758 static const struct file_operations tracing_stats_fops = {
3759 .open = tracing_open_generic,
3760 .read = tracing_stats_read,
3763 #ifdef CONFIG_DYNAMIC_FTRACE
3765 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3771 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3772 size_t cnt, loff_t *ppos)
3774 static char ftrace_dyn_info_buffer[1024];
3775 static DEFINE_MUTEX(dyn_info_mutex);
3776 unsigned long *p = filp->private_data;
3777 char *buf = ftrace_dyn_info_buffer;
3778 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3781 mutex_lock(&dyn_info_mutex);
3782 r = sprintf(buf, "%ld ", *p);
3784 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3787 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3789 mutex_unlock(&dyn_info_mutex);
3794 static const struct file_operations tracing_dyn_info_fops = {
3795 .open = tracing_open_generic,
3796 .read = tracing_read_dyn_info,
3800 static struct dentry *d_tracer;
3802 struct dentry *tracing_init_dentry(void)
3809 if (!debugfs_initialized())
3812 d_tracer = debugfs_create_dir("tracing", NULL);
3814 if (!d_tracer && !once) {
3816 pr_warning("Could not create debugfs directory 'tracing'\n");
3823 static struct dentry *d_percpu;
3825 struct dentry *tracing_dentry_percpu(void)
3828 struct dentry *d_tracer;
3833 d_tracer = tracing_init_dentry();
3838 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3840 if (!d_percpu && !once) {
3842 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3849 static void tracing_init_debugfs_percpu(long cpu)
3851 struct dentry *d_percpu = tracing_dentry_percpu();
3852 struct dentry *d_cpu;
3853 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3856 if (cpu > 999 || cpu < 0)
3859 sprintf(cpu_dir, "cpu%ld", cpu);
3860 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3862 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3866 /* per cpu trace_pipe */
3867 trace_create_file("trace_pipe", 0444, d_cpu,
3868 (void *) cpu, &tracing_pipe_fops);
3871 trace_create_file("trace", 0644, d_cpu,
3872 (void *) cpu, &tracing_fops);
3874 trace_create_file("trace_pipe_raw", 0444, d_cpu,
3875 (void *) cpu, &tracing_buffers_fops);
3877 trace_create_file("stats", 0444, d_cpu,
3878 (void *) cpu, &tracing_stats_fops);
3881 #ifdef CONFIG_FTRACE_SELFTEST
3882 /* Let selftest have access to static functions in this file */
3883 #include "trace_selftest.c"
3886 struct trace_option_dentry {
3887 struct tracer_opt *opt;
3888 struct tracer_flags *flags;
3889 struct dentry *entry;
3893 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3896 struct trace_option_dentry *topt = filp->private_data;
3899 if (topt->flags->val & topt->opt->bit)
3904 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3908 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3911 struct trace_option_dentry *topt = filp->private_data;
3916 if (cnt >= sizeof(buf))
3919 if (copy_from_user(&buf, ubuf, cnt))
3924 ret = strict_strtoul(buf, 10, &val);
3931 /* do nothing if already cleared */
3932 if (!(topt->flags->val & topt->opt->bit))
3935 mutex_lock(&trace_types_lock);
3936 if (current_trace->set_flag)
3937 ret = current_trace->set_flag(topt->flags->val,
3939 mutex_unlock(&trace_types_lock);
3942 topt->flags->val &= ~topt->opt->bit;
3945 /* do nothing if already set */
3946 if (topt->flags->val & topt->opt->bit)
3949 mutex_lock(&trace_types_lock);
3950 if (current_trace->set_flag)
3951 ret = current_trace->set_flag(topt->flags->val,
3953 mutex_unlock(&trace_types_lock);
3956 topt->flags->val |= topt->opt->bit;
3969 static const struct file_operations trace_options_fops = {
3970 .open = tracing_open_generic,
3971 .read = trace_options_read,
3972 .write = trace_options_write,
3976 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
3979 long index = (long)filp->private_data;
3982 if (trace_flags & (1 << index))
3987 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3991 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
3994 long index = (long)filp->private_data;
3999 if (cnt >= sizeof(buf))
4002 if (copy_from_user(&buf, ubuf, cnt))
4007 ret = strict_strtoul(buf, 10, &val);
4011 if (val != 0 && val != 1)
4013 set_tracer_flags(1 << index, val);
4020 static const struct file_operations trace_options_core_fops = {
4021 .open = tracing_open_generic,
4022 .read = trace_options_core_read,
4023 .write = trace_options_core_write,
4026 struct dentry *trace_create_file(const char *name,
4028 struct dentry *parent,
4030 const struct file_operations *fops)
4034 ret = debugfs_create_file(name, mode, parent, data, fops);
4036 pr_warning("Could not create debugfs '%s' entry\n", name);
4042 static struct dentry *trace_options_init_dentry(void)
4044 struct dentry *d_tracer;
4045 static struct dentry *t_options;
4050 d_tracer = tracing_init_dentry();
4054 t_options = debugfs_create_dir("options", d_tracer);
4056 pr_warning("Could not create debugfs directory 'options'\n");
4064 create_trace_option_file(struct trace_option_dentry *topt,
4065 struct tracer_flags *flags,
4066 struct tracer_opt *opt)
4068 struct dentry *t_options;
4070 t_options = trace_options_init_dentry();
4074 topt->flags = flags;
4077 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4078 &trace_options_fops);
4082 static struct trace_option_dentry *
4083 create_trace_option_files(struct tracer *tracer)
4085 struct trace_option_dentry *topts;
4086 struct tracer_flags *flags;
4087 struct tracer_opt *opts;
4093 flags = tracer->flags;
4095 if (!flags || !flags->opts)
4100 for (cnt = 0; opts[cnt].name; cnt++)
4103 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4107 for (cnt = 0; opts[cnt].name; cnt++)
4108 create_trace_option_file(&topts[cnt], flags,
4115 destroy_trace_option_files(struct trace_option_dentry *topts)
4122 for (cnt = 0; topts[cnt].opt; cnt++) {
4123 if (topts[cnt].entry)
4124 debugfs_remove(topts[cnt].entry);
4130 static struct dentry *
4131 create_trace_option_core_file(const char *option, long index)
4133 struct dentry *t_options;
4135 t_options = trace_options_init_dentry();
4139 return trace_create_file(option, 0644, t_options, (void *)index,
4140 &trace_options_core_fops);
4143 static __init void create_trace_options_dir(void)
4145 struct dentry *t_options;
4148 t_options = trace_options_init_dentry();
4152 for (i = 0; trace_options[i]; i++)
4153 create_trace_option_core_file(trace_options[i], i);
4156 static __init int tracer_init_debugfs(void)
4158 struct dentry *d_tracer;
4161 d_tracer = tracing_init_dentry();
4163 trace_create_file("tracing_enabled", 0644, d_tracer,
4164 &global_trace, &tracing_ctrl_fops);
4166 trace_create_file("trace_options", 0644, d_tracer,
4167 NULL, &tracing_iter_fops);
4169 trace_create_file("tracing_cpumask", 0644, d_tracer,
4170 NULL, &tracing_cpumask_fops);
4172 trace_create_file("trace", 0644, d_tracer,
4173 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4175 trace_create_file("available_tracers", 0444, d_tracer,
4176 &global_trace, &show_traces_fops);
4178 trace_create_file("current_tracer", 0644, d_tracer,
4179 &global_trace, &set_tracer_fops);
4181 #ifdef CONFIG_TRACER_MAX_TRACE
4182 trace_create_file("tracing_max_latency", 0644, d_tracer,
4183 &tracing_max_latency, &tracing_max_lat_fops);
4185 trace_create_file("tracing_thresh", 0644, d_tracer,
4186 &tracing_thresh, &tracing_max_lat_fops);
4189 trace_create_file("README", 0444, d_tracer,
4190 NULL, &tracing_readme_fops);
4192 trace_create_file("trace_pipe", 0444, d_tracer,
4193 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4195 trace_create_file("buffer_size_kb", 0644, d_tracer,
4196 &global_trace, &tracing_entries_fops);
4198 trace_create_file("trace_marker", 0220, d_tracer,
4199 NULL, &tracing_mark_fops);
4201 trace_create_file("saved_cmdlines", 0444, d_tracer,
4202 NULL, &tracing_saved_cmdlines_fops);
4204 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4207 #ifdef CONFIG_DYNAMIC_FTRACE
4208 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4209 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4211 #ifdef CONFIG_SYSPROF_TRACER
4212 init_tracer_sysprof_debugfs(d_tracer);
4215 create_trace_options_dir();
4217 for_each_tracing_cpu(cpu)
4218 tracing_init_debugfs_percpu(cpu);
4223 static int trace_panic_handler(struct notifier_block *this,
4224 unsigned long event, void *unused)
4226 if (ftrace_dump_on_oops)
4231 static struct notifier_block trace_panic_notifier = {
4232 .notifier_call = trace_panic_handler,
4234 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4237 static int trace_die_handler(struct notifier_block *self,
4243 if (ftrace_dump_on_oops)
4252 static struct notifier_block trace_die_notifier = {
4253 .notifier_call = trace_die_handler,
4258 * printk is set to max of 1024, we really don't need it that big.
4259 * Nothing should be printing 1000 characters anyway.
4261 #define TRACE_MAX_PRINT 1000
4264 * Define here KERN_TRACE so that we have one place to modify
4265 * it if we decide to change what log level the ftrace dump
4268 #define KERN_TRACE KERN_EMERG
4271 trace_printk_seq(struct trace_seq *s)
4273 /* Probably should print a warning here. */
4277 /* should be zero ended, but we are paranoid. */
4278 s->buffer[s->len] = 0;
4280 printk(KERN_TRACE "%s", s->buffer);
4285 static void __ftrace_dump(bool disable_tracing)
4287 static raw_spinlock_t ftrace_dump_lock =
4288 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
4289 /* use static because iter can be a bit big for the stack */
4290 static struct trace_iterator iter;
4291 unsigned int old_userobj;
4292 static int dump_ran;
4293 unsigned long flags;
4297 local_irq_save(flags);
4298 __raw_spin_lock(&ftrace_dump_lock);
4306 if (disable_tracing)
4309 for_each_tracing_cpu(cpu) {
4310 atomic_inc(&global_trace.data[cpu]->disabled);
4313 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4315 /* don't look at user memory in panic mode */
4316 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4318 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4320 /* Simulate the iterator */
4321 iter.tr = &global_trace;
4322 iter.trace = current_trace;
4323 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4326 * We need to stop all tracing on all CPUS to read the
4327 * the next buffer. This is a bit expensive, but is
4328 * not done often. We fill all what we can read,
4329 * and then release the locks again.
4332 while (!trace_empty(&iter)) {
4335 printk(KERN_TRACE "---------------------------------\n");
4339 /* reset all but tr, trace, and overruns */
4340 memset(&iter.seq, 0,
4341 sizeof(struct trace_iterator) -
4342 offsetof(struct trace_iterator, seq));
4343 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4346 if (find_next_entry_inc(&iter) != NULL) {
4349 ret = print_trace_line(&iter);
4350 if (ret != TRACE_TYPE_NO_CONSUME)
4351 trace_consume(&iter);
4354 trace_printk_seq(&iter.seq);
4358 printk(KERN_TRACE " (ftrace buffer empty)\n");
4360 printk(KERN_TRACE "---------------------------------\n");
4362 /* Re-enable tracing if requested */
4363 if (!disable_tracing) {
4364 trace_flags |= old_userobj;
4366 for_each_tracing_cpu(cpu) {
4367 atomic_dec(&global_trace.data[cpu]->disabled);
4373 __raw_spin_unlock(&ftrace_dump_lock);
4374 local_irq_restore(flags);
4377 /* By default: disable tracing after the dump */
4378 void ftrace_dump(void)
4380 __ftrace_dump(true);
4383 __init static int tracer_alloc_buffers(void)
4389 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4392 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4393 goto out_free_buffer_mask;
4395 if (!alloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
4396 goto out_free_tracing_cpumask;
4398 /* To save memory, keep the ring buffer size to its minimum */
4399 if (ring_buffer_expanded)
4400 ring_buf_size = trace_buf_size;
4404 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4405 cpumask_copy(tracing_cpumask, cpu_all_mask);
4406 cpumask_clear(tracing_reader_cpumask);
4408 /* TODO: make the number of buffers hot pluggable with CPUS */
4409 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4410 TRACE_BUFFER_FLAGS);
4411 if (!global_trace.buffer) {
4412 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4414 goto out_free_cpumask;
4416 global_trace.entries = ring_buffer_size(global_trace.buffer);
4419 #ifdef CONFIG_TRACER_MAX_TRACE
4420 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4421 TRACE_BUFFER_FLAGS);
4422 if (!max_tr.buffer) {
4423 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4425 ring_buffer_free(global_trace.buffer);
4426 goto out_free_cpumask;
4428 max_tr.entries = ring_buffer_size(max_tr.buffer);
4429 WARN_ON(max_tr.entries != global_trace.entries);
4432 /* Allocate the first page for all buffers */
4433 for_each_tracing_cpu(i) {
4434 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4435 max_tr.data[i] = &per_cpu(max_data, i);
4438 trace_init_cmdlines();
4440 register_tracer(&nop_trace);
4441 current_trace = &nop_trace;
4442 #ifdef CONFIG_BOOT_TRACER
4443 register_tracer(&boot_tracer);
4445 /* All seems OK, enable tracing */
4446 tracing_disabled = 0;
4448 atomic_notifier_chain_register(&panic_notifier_list,
4449 &trace_panic_notifier);
4451 register_die_notifier(&trace_die_notifier);
4456 free_cpumask_var(tracing_reader_cpumask);
4457 out_free_tracing_cpumask:
4458 free_cpumask_var(tracing_cpumask);
4459 out_free_buffer_mask:
4460 free_cpumask_var(tracing_buffer_mask);
4465 __init static int clear_boot_tracer(void)
4468 * The default tracer at boot buffer is an init section.
4469 * This function is called in lateinit. If we did not
4470 * find the boot tracer, then clear it out, to prevent
4471 * later registration from accessing the buffer that is
4472 * about to be freed.
4474 if (!default_bootup_tracer)
4477 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4478 default_bootup_tracer);
4479 default_bootup_tracer = NULL;
4484 early_initcall(tracer_alloc_buffers);
4485 fs_initcall(tracer_init_debugfs);
4486 late_initcall(clear_boot_tracer);