2 * ring buffer based function tracer
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 * Based on code from the latency_tracer, that is:
11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 William Lee Irwin III
14 #include <linux/ring_buffer.h>
15 #include <generated/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/notifier.h>
21 #include <linux/irqflags.h>
22 #include <linux/debugfs.h>
23 #include <linux/pagemap.h>
24 #include <linux/hardirq.h>
25 #include <linux/linkage.h>
26 #include <linux/uaccess.h>
27 #include <linux/kprobes.h>
28 #include <linux/ftrace.h>
29 #include <linux/module.h>
30 #include <linux/percpu.h>
31 #include <linux/splice.h>
32 #include <linux/kdebug.h>
33 #include <linux/string.h>
34 #include <linux/rwsem.h>
35 #include <linux/slab.h>
36 #include <linux/ctype.h>
37 #include <linux/init.h>
38 #include <linux/poll.h>
42 #include "trace_output.h"
44 #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
47 * On boot up, the ring buffer is set to the minimum size, so that
48 * we do not waste memory on systems that are not using tracing.
50 int ring_buffer_expanded;
53 * We need to change this state when a selftest is running.
54 * A selftest will lurk into the ring-buffer to count the
55 * entries inserted during the selftest although some concurrent
56 * insertions into the ring-buffer such as trace_printk could occurred
57 * at the same time, giving false positive or negative results.
59 static bool __read_mostly tracing_selftest_running;
62 * If a tracer is running, we do not want to run SELFTEST.
64 bool __read_mostly tracing_selftest_disabled;
66 /* For tracers that don't implement custom flags */
67 static struct tracer_opt dummy_tracer_opt[] = {
71 static struct tracer_flags dummy_tracer_flags = {
73 .opts = dummy_tracer_opt
76 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
82 * Kill all tracing for good (never come back).
83 * It is initialized to 1 but will turn to zero if the initialization
84 * of the tracer is successful. But that is the only place that sets
87 static int tracing_disabled = 1;
89 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
91 static inline void ftrace_disable_cpu(void)
94 __this_cpu_inc(ftrace_cpu_disabled);
97 static inline void ftrace_enable_cpu(void)
99 __this_cpu_dec(ftrace_cpu_disabled);
103 cpumask_var_t __read_mostly tracing_buffer_mask;
106 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
108 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
109 * is set, then ftrace_dump is called. This will output the contents
110 * of the ftrace buffers to the console. This is very useful for
111 * capturing traces that lead to crashes and outputing it to a
114 * It is default off, but you can enable it with either specifying
115 * "ftrace_dump_on_oops" in the kernel command line, or setting
116 * /proc/sys/kernel/ftrace_dump_on_oops
117 * Set 1 if you want to dump buffers of all CPUs
118 * Set 2 if you want to dump the buffer of the CPU that triggered oops
121 enum ftrace_dump_mode ftrace_dump_on_oops;
123 static int tracing_set_tracer(const char *buf);
125 #define MAX_TRACER_SIZE 100
126 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
127 static char *default_bootup_tracer;
129 static int __init set_cmdline_ftrace(char *str)
131 strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
132 default_bootup_tracer = bootup_tracer_buf;
133 /* We are using ftrace early, expand it */
134 ring_buffer_expanded = 1;
137 __setup("ftrace=", set_cmdline_ftrace);
139 static int __init set_ftrace_dump_on_oops(char *str)
141 if (*str++ != '=' || !*str) {
142 ftrace_dump_on_oops = DUMP_ALL;
146 if (!strcmp("orig_cpu", str)) {
147 ftrace_dump_on_oops = DUMP_ORIG;
153 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
155 unsigned long long ns2usecs(cycle_t nsec)
163 * The global_trace is the descriptor that holds the tracing
164 * buffers for the live tracing. For each CPU, it contains
165 * a link list of pages that will store trace entries. The
166 * page descriptor of the pages in the memory is used to hold
167 * the link list by linking the lru item in the page descriptor
168 * to each of the pages in the buffer per CPU.
170 * For each active CPU there is a data field that holds the
171 * pages for the buffer for that CPU. Each CPU has the same number
172 * of pages allocated for its buffer.
174 static struct trace_array global_trace;
176 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
178 int filter_current_check_discard(struct ring_buffer *buffer,
179 struct ftrace_event_call *call, void *rec,
180 struct ring_buffer_event *event)
182 return filter_check_discard(call, rec, buffer, event);
184 EXPORT_SYMBOL_GPL(filter_current_check_discard);
186 cycle_t ftrace_now(int cpu)
190 /* Early boot up does not have a buffer yet */
191 if (!global_trace.buffer)
192 return trace_clock_local();
194 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
195 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
201 * The max_tr is used to snapshot the global_trace when a maximum
202 * latency is reached. Some tracers will use this to store a maximum
203 * trace while it continues examining live traces.
205 * The buffers for the max_tr are set up the same as the global_trace.
206 * When a snapshot is taken, the link list of the max_tr is swapped
207 * with the link list of the global_trace and the buffers are reset for
208 * the global_trace so the tracing can continue.
210 static struct trace_array max_tr;
212 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
214 /* tracer_enabled is used to toggle activation of a tracer */
215 static int tracer_enabled = 1;
218 * tracing_is_enabled - return tracer_enabled status
220 * This function is used by other tracers to know the status
221 * of the tracer_enabled flag. Tracers may use this function
222 * to know if it should enable their features when starting
223 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
225 int tracing_is_enabled(void)
227 return tracer_enabled;
231 * trace_buf_size is the size in bytes that is allocated
232 * for a buffer. Note, the number of bytes is always rounded
235 * This number is purposely set to a low number of 16384.
236 * If the dump on oops happens, it will be much appreciated
237 * to not have to wait for all that output. Anyway this can be
238 * boot time and run time configurable.
240 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
242 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
244 /* trace_types holds a link list of available tracers. */
245 static struct tracer *trace_types __read_mostly;
247 /* current_trace points to the tracer that is currently active */
248 static struct tracer *current_trace __read_mostly;
251 * trace_types_lock is used to protect the trace_types list.
253 static DEFINE_MUTEX(trace_types_lock);
256 * serialize the access of the ring buffer
258 * ring buffer serializes readers, but it is low level protection.
259 * The validity of the events (which returns by ring_buffer_peek() ..etc)
260 * are not protected by ring buffer.
262 * The content of events may become garbage if we allow other process consumes
263 * these events concurrently:
264 * A) the page of the consumed events may become a normal page
265 * (not reader page) in ring buffer, and this page will be rewrited
266 * by events producer.
267 * B) The page of the consumed events may become a page for splice_read,
268 * and this page will be returned to system.
270 * These primitives allow multi process access to different cpu ring buffer
273 * These primitives don't distinguish read-only and read-consume access.
274 * Multi read-only access are also serialized.
278 static DECLARE_RWSEM(all_cpu_access_lock);
279 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
281 static inline void trace_access_lock(int cpu)
283 if (cpu == TRACE_PIPE_ALL_CPU) {
284 /* gain it for accessing the whole ring buffer. */
285 down_write(&all_cpu_access_lock);
287 /* gain it for accessing a cpu ring buffer. */
289 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
290 down_read(&all_cpu_access_lock);
292 /* Secondly block other access to this @cpu ring buffer. */
293 mutex_lock(&per_cpu(cpu_access_lock, cpu));
297 static inline void trace_access_unlock(int cpu)
299 if (cpu == TRACE_PIPE_ALL_CPU) {
300 up_write(&all_cpu_access_lock);
302 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
303 up_read(&all_cpu_access_lock);
307 static inline void trace_access_lock_init(void)
311 for_each_possible_cpu(cpu)
312 mutex_init(&per_cpu(cpu_access_lock, cpu));
317 static DEFINE_MUTEX(access_lock);
319 static inline void trace_access_lock(int cpu)
322 mutex_lock(&access_lock);
325 static inline void trace_access_unlock(int cpu)
328 mutex_unlock(&access_lock);
331 static inline void trace_access_lock_init(void)
337 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
338 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
340 /* trace_flags holds trace_options default values */
341 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
342 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
343 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD;
345 static int trace_stop_count;
346 static DEFINE_SPINLOCK(tracing_start_lock);
349 * trace_wake_up - wake up tasks waiting for trace input
351 * Simply wakes up any task that is blocked on the trace_wait
352 * queue. These is used with trace_poll for tasks polling the trace.
354 void trace_wake_up(void)
358 if (trace_flags & TRACE_ITER_BLOCK)
361 * The runqueue_is_locked() can fail, but this is the best we
365 if (!runqueue_is_locked(cpu))
366 wake_up(&trace_wait);
370 static int __init set_buf_size(char *str)
372 unsigned long buf_size;
376 buf_size = memparse(str, &str);
377 /* nr_entries can not be zero */
380 trace_buf_size = buf_size;
383 __setup("trace_buf_size=", set_buf_size);
385 static int __init set_tracing_thresh(char *str)
387 unsigned long threshhold;
392 ret = strict_strtoul(str, 0, &threshhold);
395 tracing_thresh = threshhold * 1000;
398 __setup("tracing_thresh=", set_tracing_thresh);
400 unsigned long nsecs_to_usecs(unsigned long nsecs)
405 /* These must match the bit postions in trace_iterator_flags */
406 static const char *trace_options[] = {
435 { trace_clock_local, "local" },
436 { trace_clock_global, "global" },
442 * trace_parser_get_init - gets the buffer for trace parser
444 int trace_parser_get_init(struct trace_parser *parser, int size)
446 memset(parser, 0, sizeof(*parser));
448 parser->buffer = kmalloc(size, GFP_KERNEL);
457 * trace_parser_put - frees the buffer for trace parser
459 void trace_parser_put(struct trace_parser *parser)
461 kfree(parser->buffer);
465 * trace_get_user - reads the user input string separated by space
466 * (matched by isspace(ch))
468 * For each string found the 'struct trace_parser' is updated,
469 * and the function returns.
471 * Returns number of bytes read.
473 * See kernel/trace/trace.h for 'struct trace_parser' details.
475 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
476 size_t cnt, loff_t *ppos)
483 trace_parser_clear(parser);
485 ret = get_user(ch, ubuf++);
493 * The parser is not finished with the last write,
494 * continue reading the user input without skipping spaces.
497 /* skip white space */
498 while (cnt && isspace(ch)) {
499 ret = get_user(ch, ubuf++);
506 /* only spaces were written */
516 /* read the non-space input */
517 while (cnt && !isspace(ch)) {
518 if (parser->idx < parser->size - 1)
519 parser->buffer[parser->idx++] = ch;
524 ret = get_user(ch, ubuf++);
531 /* We either got finished input or we have to wait for another call. */
533 parser->buffer[parser->idx] = 0;
534 parser->cont = false;
537 parser->buffer[parser->idx++] = ch;
547 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
555 if (s->len <= s->readpos)
558 len = s->len - s->readpos;
561 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
571 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
576 if (s->len <= s->readpos)
579 len = s->len - s->readpos;
582 ret = memcpy(buf, s->buffer + s->readpos, cnt);
591 * ftrace_max_lock is used to protect the swapping of buffers
592 * when taking a max snapshot. The buffers themselves are
593 * protected by per_cpu spinlocks. But the action of the swap
594 * needs its own lock.
596 * This is defined as a arch_spinlock_t in order to help
597 * with performance when lockdep debugging is enabled.
599 * It is also used in other places outside the update_max_tr
600 * so it needs to be defined outside of the
601 * CONFIG_TRACER_MAX_TRACE.
603 static arch_spinlock_t ftrace_max_lock =
604 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
606 unsigned long __read_mostly tracing_thresh;
608 #ifdef CONFIG_TRACER_MAX_TRACE
609 unsigned long __read_mostly tracing_max_latency;
612 * Copy the new maximum trace into the separate maximum-trace
613 * structure. (this way the maximum trace is permanently saved,
614 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
617 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
619 struct trace_array_cpu *data = tr->data[cpu];
620 struct trace_array_cpu *max_data;
623 max_tr.time_start = data->preempt_timestamp;
625 max_data = max_tr.data[cpu];
626 max_data->saved_latency = tracing_max_latency;
627 max_data->critical_start = data->critical_start;
628 max_data->critical_end = data->critical_end;
630 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
631 max_data->pid = tsk->pid;
632 max_data->uid = task_uid(tsk);
633 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
634 max_data->policy = tsk->policy;
635 max_data->rt_priority = tsk->rt_priority;
637 /* record this tasks comm */
638 tracing_record_cmdline(tsk);
642 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
644 * @tsk: the task with the latency
645 * @cpu: The cpu that initiated the trace.
647 * Flip the buffers between the @tr and the max_tr and record information
648 * about which task was the cause of this latency.
651 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
653 struct ring_buffer *buf = tr->buffer;
655 if (trace_stop_count)
658 WARN_ON_ONCE(!irqs_disabled());
659 if (!current_trace->use_max_tr) {
663 arch_spin_lock(&ftrace_max_lock);
665 tr->buffer = max_tr.buffer;
668 __update_max_tr(tr, tsk, cpu);
669 arch_spin_unlock(&ftrace_max_lock);
673 * update_max_tr_single - only copy one trace over, and reset the rest
675 * @tsk - task with the latency
676 * @cpu - the cpu of the buffer to copy.
678 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
681 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
685 if (trace_stop_count)
688 WARN_ON_ONCE(!irqs_disabled());
689 if (!current_trace->use_max_tr) {
694 arch_spin_lock(&ftrace_max_lock);
696 ftrace_disable_cpu();
698 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
702 * We failed to swap the buffer due to a commit taking
703 * place on this CPU. We fail to record, but we reset
704 * the max trace buffer (no one writes directly to it)
705 * and flag that it failed.
707 trace_array_printk(&max_tr, _THIS_IP_,
708 "Failed to swap buffers due to commit in progress\n");
713 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
715 __update_max_tr(tr, tsk, cpu);
716 arch_spin_unlock(&ftrace_max_lock);
718 #endif /* CONFIG_TRACER_MAX_TRACE */
721 * register_tracer - register a tracer with the ftrace system.
722 * @type - the plugin for the tracer
724 * Register a new plugin tracer.
726 int register_tracer(struct tracer *type)
727 __releases(kernel_lock)
728 __acquires(kernel_lock)
734 pr_info("Tracer must have a name\n");
738 if (strlen(type->name) >= MAX_TRACER_SIZE) {
739 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
743 mutex_lock(&trace_types_lock);
745 tracing_selftest_running = true;
747 for (t = trace_types; t; t = t->next) {
748 if (strcmp(type->name, t->name) == 0) {
750 pr_info("Tracer %s already registered\n",
758 type->set_flag = &dummy_set_flag;
760 type->flags = &dummy_tracer_flags;
762 if (!type->flags->opts)
763 type->flags->opts = dummy_tracer_opt;
764 if (!type->wait_pipe)
765 type->wait_pipe = default_wait_pipe;
768 #ifdef CONFIG_FTRACE_STARTUP_TEST
769 if (type->selftest && !tracing_selftest_disabled) {
770 struct tracer *saved_tracer = current_trace;
771 struct trace_array *tr = &global_trace;
774 * Run a selftest on this tracer.
775 * Here we reset the trace buffer, and set the current
776 * tracer to be this tracer. The tracer can then run some
777 * internal tracing to verify that everything is in order.
778 * If we fail, we do not register this tracer.
780 tracing_reset_online_cpus(tr);
782 current_trace = type;
783 /* the test is responsible for initializing and enabling */
784 pr_info("Testing tracer %s: ", type->name);
785 ret = type->selftest(type, tr);
786 /* the test is responsible for resetting too */
787 current_trace = saved_tracer;
789 printk(KERN_CONT "FAILED!\n");
792 /* Only reset on passing, to avoid touching corrupted buffers */
793 tracing_reset_online_cpus(tr);
795 printk(KERN_CONT "PASSED\n");
799 type->next = trace_types;
803 tracing_selftest_running = false;
804 mutex_unlock(&trace_types_lock);
806 if (ret || !default_bootup_tracer)
809 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
812 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
813 /* Do we want this tracer to start on bootup? */
814 tracing_set_tracer(type->name);
815 default_bootup_tracer = NULL;
816 /* disable other selftests, since this will break it. */
817 tracing_selftest_disabled = 1;
818 #ifdef CONFIG_FTRACE_STARTUP_TEST
819 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
827 void unregister_tracer(struct tracer *type)
831 mutex_lock(&trace_types_lock);
832 for (t = &trace_types; *t; t = &(*t)->next) {
836 pr_info("Tracer %s not registered\n", type->name);
842 if (type == current_trace && tracer_enabled) {
845 if (current_trace->stop)
846 current_trace->stop(&global_trace);
847 current_trace = &nop_trace;
850 mutex_unlock(&trace_types_lock);
853 static void __tracing_reset(struct ring_buffer *buffer, int cpu)
855 ftrace_disable_cpu();
856 ring_buffer_reset_cpu(buffer, cpu);
860 void tracing_reset(struct trace_array *tr, int cpu)
862 struct ring_buffer *buffer = tr->buffer;
864 ring_buffer_record_disable(buffer);
866 /* Make sure all commits have finished */
868 __tracing_reset(buffer, cpu);
870 ring_buffer_record_enable(buffer);
873 void tracing_reset_online_cpus(struct trace_array *tr)
875 struct ring_buffer *buffer = tr->buffer;
878 ring_buffer_record_disable(buffer);
880 /* Make sure all commits have finished */
883 tr->time_start = ftrace_now(tr->cpu);
885 for_each_online_cpu(cpu)
886 __tracing_reset(buffer, cpu);
888 ring_buffer_record_enable(buffer);
891 void tracing_reset_current(int cpu)
893 tracing_reset(&global_trace, cpu);
896 void tracing_reset_current_online_cpus(void)
898 tracing_reset_online_cpus(&global_trace);
901 #define SAVED_CMDLINES 128
902 #define NO_CMDLINE_MAP UINT_MAX
903 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
904 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
905 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
906 static int cmdline_idx;
907 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
909 /* temporary disable recording */
910 static atomic_t trace_record_cmdline_disabled __read_mostly;
912 static void trace_init_cmdlines(void)
914 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
915 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
919 int is_tracing_stopped(void)
921 return trace_stop_count;
925 * ftrace_off_permanent - disable all ftrace code permanently
927 * This should only be called when a serious anomally has
928 * been detected. This will turn off the function tracing,
929 * ring buffers, and other tracing utilites. It takes no
930 * locks and can be called from any context.
932 void ftrace_off_permanent(void)
934 tracing_disabled = 1;
936 tracing_off_permanent();
940 * tracing_start - quick start of the tracer
942 * If tracing is enabled but was stopped by tracing_stop,
943 * this will start the tracer back up.
945 void tracing_start(void)
947 struct ring_buffer *buffer;
950 if (tracing_disabled)
953 spin_lock_irqsave(&tracing_start_lock, flags);
954 if (--trace_stop_count) {
955 if (trace_stop_count < 0) {
956 /* Someone screwed up their debugging */
958 trace_stop_count = 0;
963 /* Prevent the buffers from switching */
964 arch_spin_lock(&ftrace_max_lock);
966 buffer = global_trace.buffer;
968 ring_buffer_record_enable(buffer);
970 buffer = max_tr.buffer;
972 ring_buffer_record_enable(buffer);
974 arch_spin_unlock(&ftrace_max_lock);
978 spin_unlock_irqrestore(&tracing_start_lock, flags);
982 * tracing_stop - quick stop of the tracer
984 * Light weight way to stop tracing. Use in conjunction with
987 void tracing_stop(void)
989 struct ring_buffer *buffer;
993 spin_lock_irqsave(&tracing_start_lock, flags);
994 if (trace_stop_count++)
997 /* Prevent the buffers from switching */
998 arch_spin_lock(&ftrace_max_lock);
1000 buffer = global_trace.buffer;
1002 ring_buffer_record_disable(buffer);
1004 buffer = max_tr.buffer;
1006 ring_buffer_record_disable(buffer);
1008 arch_spin_unlock(&ftrace_max_lock);
1011 spin_unlock_irqrestore(&tracing_start_lock, flags);
1014 void trace_stop_cmdline_recording(void);
1016 static void trace_save_cmdline(struct task_struct *tsk)
1020 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1024 * It's not the end of the world if we don't get
1025 * the lock, but we also don't want to spin
1026 * nor do we want to disable interrupts,
1027 * so if we miss here, then better luck next time.
1029 if (!arch_spin_trylock(&trace_cmdline_lock))
1032 idx = map_pid_to_cmdline[tsk->pid];
1033 if (idx == NO_CMDLINE_MAP) {
1034 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1037 * Check whether the cmdline buffer at idx has a pid
1038 * mapped. We are going to overwrite that entry so we
1039 * need to clear the map_pid_to_cmdline. Otherwise we
1040 * would read the new comm for the old pid.
1042 pid = map_cmdline_to_pid[idx];
1043 if (pid != NO_CMDLINE_MAP)
1044 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1046 map_cmdline_to_pid[idx] = tsk->pid;
1047 map_pid_to_cmdline[tsk->pid] = idx;
1052 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1054 arch_spin_unlock(&trace_cmdline_lock);
1057 void trace_find_cmdline(int pid, char comm[])
1062 strcpy(comm, "<idle>");
1066 if (WARN_ON_ONCE(pid < 0)) {
1067 strcpy(comm, "<XXX>");
1071 if (pid > PID_MAX_DEFAULT) {
1072 strcpy(comm, "<...>");
1077 arch_spin_lock(&trace_cmdline_lock);
1078 map = map_pid_to_cmdline[pid];
1079 if (map != NO_CMDLINE_MAP)
1080 strcpy(comm, saved_cmdlines[map]);
1082 strcpy(comm, "<...>");
1084 arch_spin_unlock(&trace_cmdline_lock);
1088 void tracing_record_cmdline(struct task_struct *tsk)
1090 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1094 trace_save_cmdline(tsk);
1098 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1101 struct task_struct *tsk = current;
1103 entry->preempt_count = pc & 0xff;
1104 entry->pid = (tsk) ? tsk->pid : 0;
1105 entry->lock_depth = (tsk) ? tsk->lock_depth : 0;
1107 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1108 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1110 TRACE_FLAG_IRQS_NOSUPPORT |
1112 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1113 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1114 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1116 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1118 struct ring_buffer_event *
1119 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1122 unsigned long flags, int pc)
1124 struct ring_buffer_event *event;
1126 event = ring_buffer_lock_reserve(buffer, len);
1127 if (event != NULL) {
1128 struct trace_entry *ent = ring_buffer_event_data(event);
1130 tracing_generic_entry_update(ent, flags, pc);
1138 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1139 struct ring_buffer_event *event,
1140 unsigned long flags, int pc,
1143 ring_buffer_unlock_commit(buffer, event);
1145 ftrace_trace_stack(buffer, flags, 6, pc);
1146 ftrace_trace_userstack(buffer, flags, pc);
1152 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1153 struct ring_buffer_event *event,
1154 unsigned long flags, int pc)
1156 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1159 struct ring_buffer_event *
1160 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1161 int type, unsigned long len,
1162 unsigned long flags, int pc)
1164 *current_rb = global_trace.buffer;
1165 return trace_buffer_lock_reserve(*current_rb,
1166 type, len, flags, pc);
1168 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1170 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1171 struct ring_buffer_event *event,
1172 unsigned long flags, int pc)
1174 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1176 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1178 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1179 struct ring_buffer_event *event,
1180 unsigned long flags, int pc)
1182 __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1184 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1186 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1187 struct ring_buffer_event *event)
1189 ring_buffer_discard_commit(buffer, event);
1191 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1194 trace_function(struct trace_array *tr,
1195 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1198 struct ftrace_event_call *call = &event_function;
1199 struct ring_buffer *buffer = tr->buffer;
1200 struct ring_buffer_event *event;
1201 struct ftrace_entry *entry;
1203 /* If we are reading the ring buffer, don't trace */
1204 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1207 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1211 entry = ring_buffer_event_data(event);
1213 entry->parent_ip = parent_ip;
1215 if (!filter_check_discard(call, entry, buffer, event))
1216 ring_buffer_unlock_commit(buffer, event);
1220 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1221 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1224 if (likely(!atomic_read(&data->disabled)))
1225 trace_function(tr, ip, parent_ip, flags, pc);
1228 #ifdef CONFIG_STACKTRACE
1229 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1230 unsigned long flags,
1233 struct ftrace_event_call *call = &event_kernel_stack;
1234 struct ring_buffer_event *event;
1235 struct stack_entry *entry;
1236 struct stack_trace trace;
1238 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1239 sizeof(*entry), flags, pc);
1242 entry = ring_buffer_event_data(event);
1243 memset(&entry->caller, 0, sizeof(entry->caller));
1245 trace.nr_entries = 0;
1246 trace.max_entries = FTRACE_STACK_ENTRIES;
1248 trace.entries = entry->caller;
1250 save_stack_trace(&trace);
1251 if (!filter_check_discard(call, entry, buffer, event))
1252 ring_buffer_unlock_commit(buffer, event);
1255 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1258 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1261 __ftrace_trace_stack(buffer, flags, skip, pc);
1264 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1267 __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1271 * trace_dump_stack - record a stack back trace in the trace buffer
1273 void trace_dump_stack(void)
1275 unsigned long flags;
1277 if (tracing_disabled || tracing_selftest_running)
1280 local_save_flags(flags);
1282 /* skipping 3 traces, seems to get us at the caller of this function */
1283 __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count());
1286 static DEFINE_PER_CPU(int, user_stack_count);
1289 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1291 struct ftrace_event_call *call = &event_user_stack;
1292 struct ring_buffer_event *event;
1293 struct userstack_entry *entry;
1294 struct stack_trace trace;
1296 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1300 * NMIs can not handle page faults, even with fix ups.
1301 * The save user stack can (and often does) fault.
1303 if (unlikely(in_nmi()))
1307 * prevent recursion, since the user stack tracing may
1308 * trigger other kernel events.
1311 if (__this_cpu_read(user_stack_count))
1314 __this_cpu_inc(user_stack_count);
1318 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1319 sizeof(*entry), flags, pc);
1322 entry = ring_buffer_event_data(event);
1324 entry->tgid = current->tgid;
1325 memset(&entry->caller, 0, sizeof(entry->caller));
1327 trace.nr_entries = 0;
1328 trace.max_entries = FTRACE_STACK_ENTRIES;
1330 trace.entries = entry->caller;
1332 save_stack_trace_user(&trace);
1333 if (!filter_check_discard(call, entry, buffer, event))
1334 ring_buffer_unlock_commit(buffer, event);
1336 __this_cpu_dec(user_stack_count);
1343 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1345 ftrace_trace_userstack(tr, flags, preempt_count());
1349 #endif /* CONFIG_STACKTRACE */
1352 * trace_vbprintk - write binary msg to tracing buffer
1355 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1357 static arch_spinlock_t trace_buf_lock =
1358 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1359 static u32 trace_buf[TRACE_BUF_SIZE];
1361 struct ftrace_event_call *call = &event_bprint;
1362 struct ring_buffer_event *event;
1363 struct ring_buffer *buffer;
1364 struct trace_array *tr = &global_trace;
1365 struct trace_array_cpu *data;
1366 struct bprint_entry *entry;
1367 unsigned long flags;
1369 int cpu, len = 0, size, pc;
1371 if (unlikely(tracing_selftest_running || tracing_disabled))
1374 /* Don't pollute graph traces with trace_vprintk internals */
1375 pause_graph_tracing();
1377 pc = preempt_count();
1378 preempt_disable_notrace();
1379 cpu = raw_smp_processor_id();
1380 data = tr->data[cpu];
1382 disable = atomic_inc_return(&data->disabled);
1383 if (unlikely(disable != 1))
1386 /* Lockdep uses trace_printk for lock tracing */
1387 local_irq_save(flags);
1388 arch_spin_lock(&trace_buf_lock);
1389 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1391 if (len > TRACE_BUF_SIZE || len < 0)
1394 size = sizeof(*entry) + sizeof(u32) * len;
1395 buffer = tr->buffer;
1396 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1400 entry = ring_buffer_event_data(event);
1404 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1405 if (!filter_check_discard(call, entry, buffer, event)) {
1406 ring_buffer_unlock_commit(buffer, event);
1407 ftrace_trace_stack(buffer, flags, 6, pc);
1411 arch_spin_unlock(&trace_buf_lock);
1412 local_irq_restore(flags);
1415 atomic_dec_return(&data->disabled);
1416 preempt_enable_notrace();
1417 unpause_graph_tracing();
1421 EXPORT_SYMBOL_GPL(trace_vbprintk);
1423 int trace_array_printk(struct trace_array *tr,
1424 unsigned long ip, const char *fmt, ...)
1429 if (!(trace_flags & TRACE_ITER_PRINTK))
1433 ret = trace_array_vprintk(tr, ip, fmt, ap);
1438 int trace_array_vprintk(struct trace_array *tr,
1439 unsigned long ip, const char *fmt, va_list args)
1441 static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1442 static char trace_buf[TRACE_BUF_SIZE];
1444 struct ftrace_event_call *call = &event_print;
1445 struct ring_buffer_event *event;
1446 struct ring_buffer *buffer;
1447 struct trace_array_cpu *data;
1448 int cpu, len = 0, size, pc;
1449 struct print_entry *entry;
1450 unsigned long irq_flags;
1453 if (tracing_disabled || tracing_selftest_running)
1456 pc = preempt_count();
1457 preempt_disable_notrace();
1458 cpu = raw_smp_processor_id();
1459 data = tr->data[cpu];
1461 disable = atomic_inc_return(&data->disabled);
1462 if (unlikely(disable != 1))
1465 pause_graph_tracing();
1466 raw_local_irq_save(irq_flags);
1467 arch_spin_lock(&trace_buf_lock);
1468 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1470 size = sizeof(*entry) + len + 1;
1471 buffer = tr->buffer;
1472 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1476 entry = ring_buffer_event_data(event);
1479 memcpy(&entry->buf, trace_buf, len);
1480 entry->buf[len] = '\0';
1481 if (!filter_check_discard(call, entry, buffer, event)) {
1482 ring_buffer_unlock_commit(buffer, event);
1483 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1487 arch_spin_unlock(&trace_buf_lock);
1488 raw_local_irq_restore(irq_flags);
1489 unpause_graph_tracing();
1491 atomic_dec_return(&data->disabled);
1492 preempt_enable_notrace();
1497 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1499 return trace_array_vprintk(&global_trace, ip, fmt, args);
1501 EXPORT_SYMBOL_GPL(trace_vprintk);
1503 static void trace_iterator_increment(struct trace_iterator *iter)
1505 /* Don't allow ftrace to trace into the ring buffers */
1506 ftrace_disable_cpu();
1509 if (iter->buffer_iter[iter->cpu])
1510 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1512 ftrace_enable_cpu();
1515 static struct trace_entry *
1516 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1517 unsigned long *lost_events)
1519 struct ring_buffer_event *event;
1520 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1522 /* Don't allow ftrace to trace into the ring buffers */
1523 ftrace_disable_cpu();
1526 event = ring_buffer_iter_peek(buf_iter, ts);
1528 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1531 ftrace_enable_cpu();
1533 return event ? ring_buffer_event_data(event) : NULL;
1536 static struct trace_entry *
1537 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1538 unsigned long *missing_events, u64 *ent_ts)
1540 struct ring_buffer *buffer = iter->tr->buffer;
1541 struct trace_entry *ent, *next = NULL;
1542 unsigned long lost_events = 0, next_lost = 0;
1543 int cpu_file = iter->cpu_file;
1544 u64 next_ts = 0, ts;
1549 * If we are in a per_cpu trace file, don't bother by iterating over
1550 * all cpu and peek directly.
1552 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1553 if (ring_buffer_empty_cpu(buffer, cpu_file))
1555 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1557 *ent_cpu = cpu_file;
1562 for_each_tracing_cpu(cpu) {
1564 if (ring_buffer_empty_cpu(buffer, cpu))
1567 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1570 * Pick the entry with the smallest timestamp:
1572 if (ent && (!next || ts < next_ts)) {
1576 next_lost = lost_events;
1581 *ent_cpu = next_cpu;
1587 *missing_events = next_lost;
1592 /* Find the next real entry, without updating the iterator itself */
1593 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1594 int *ent_cpu, u64 *ent_ts)
1596 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1599 /* Find the next real entry, and increment the iterator to the next entry */
1600 void *trace_find_next_entry_inc(struct trace_iterator *iter)
1602 iter->ent = __find_next_entry(iter, &iter->cpu,
1603 &iter->lost_events, &iter->ts);
1606 trace_iterator_increment(iter);
1608 return iter->ent ? iter : NULL;
1611 static void trace_consume(struct trace_iterator *iter)
1613 /* Don't allow ftrace to trace into the ring buffers */
1614 ftrace_disable_cpu();
1615 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1616 &iter->lost_events);
1617 ftrace_enable_cpu();
1620 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1622 struct trace_iterator *iter = m->private;
1626 WARN_ON_ONCE(iter->leftover);
1630 /* can't go backwards */
1635 ent = trace_find_next_entry_inc(iter);
1639 while (ent && iter->idx < i)
1640 ent = trace_find_next_entry_inc(iter);
1647 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1649 struct trace_array *tr = iter->tr;
1650 struct ring_buffer_event *event;
1651 struct ring_buffer_iter *buf_iter;
1652 unsigned long entries = 0;
1655 tr->data[cpu]->skipped_entries = 0;
1657 if (!iter->buffer_iter[cpu])
1660 buf_iter = iter->buffer_iter[cpu];
1661 ring_buffer_iter_reset(buf_iter);
1664 * We could have the case with the max latency tracers
1665 * that a reset never took place on a cpu. This is evident
1666 * by the timestamp being before the start of the buffer.
1668 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1669 if (ts >= iter->tr->time_start)
1672 ring_buffer_read(buf_iter, NULL);
1675 tr->data[cpu]->skipped_entries = entries;
1679 * The current tracer is copied to avoid a global locking
1682 static void *s_start(struct seq_file *m, loff_t *pos)
1684 struct trace_iterator *iter = m->private;
1685 static struct tracer *old_tracer;
1686 int cpu_file = iter->cpu_file;
1691 /* copy the tracer to avoid using a global lock all around */
1692 mutex_lock(&trace_types_lock);
1693 if (unlikely(old_tracer != current_trace && current_trace)) {
1694 old_tracer = current_trace;
1695 *iter->trace = *current_trace;
1697 mutex_unlock(&trace_types_lock);
1699 atomic_inc(&trace_record_cmdline_disabled);
1701 if (*pos != iter->pos) {
1706 ftrace_disable_cpu();
1708 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1709 for_each_tracing_cpu(cpu)
1710 tracing_iter_reset(iter, cpu);
1712 tracing_iter_reset(iter, cpu_file);
1714 ftrace_enable_cpu();
1717 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1722 * If we overflowed the seq_file before, then we want
1723 * to just reuse the trace_seq buffer again.
1729 p = s_next(m, p, &l);
1733 trace_event_read_lock();
1734 trace_access_lock(cpu_file);
1738 static void s_stop(struct seq_file *m, void *p)
1740 struct trace_iterator *iter = m->private;
1742 atomic_dec(&trace_record_cmdline_disabled);
1743 trace_access_unlock(iter->cpu_file);
1744 trace_event_read_unlock();
1747 static void print_lat_help_header(struct seq_file *m)
1749 seq_puts(m, "# _------=> CPU# \n");
1750 seq_puts(m, "# / _-----=> irqs-off \n");
1751 seq_puts(m, "# | / _----=> need-resched \n");
1752 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1753 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1754 seq_puts(m, "# |||| /_--=> lock-depth \n");
1755 seq_puts(m, "# |||||/ delay \n");
1756 seq_puts(m, "# cmd pid |||||| time | caller \n");
1757 seq_puts(m, "# \\ / |||||| \\ | / \n");
1760 static void print_func_help_header(struct seq_file *m)
1762 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1763 seq_puts(m, "# | | | | |\n");
1768 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1770 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1771 struct trace_array *tr = iter->tr;
1772 struct trace_array_cpu *data = tr->data[tr->cpu];
1773 struct tracer *type = current_trace;
1774 unsigned long entries = 0;
1775 unsigned long total = 0;
1776 unsigned long count;
1777 const char *name = "preemption";
1784 for_each_tracing_cpu(cpu) {
1785 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1787 * If this buffer has skipped entries, then we hold all
1788 * entries for the trace and we need to ignore the
1789 * ones before the time stamp.
1791 if (tr->data[cpu]->skipped_entries) {
1792 count -= tr->data[cpu]->skipped_entries;
1793 /* total is the same as the entries */
1797 ring_buffer_overrun_cpu(tr->buffer, cpu);
1801 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1803 seq_puts(m, "# -----------------------------------"
1804 "---------------------------------\n");
1805 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1806 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1807 nsecs_to_usecs(data->saved_latency),
1811 #if defined(CONFIG_PREEMPT_NONE)
1813 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1815 #elif defined(CONFIG_PREEMPT)
1820 /* These are reserved for later use */
1823 seq_printf(m, " #P:%d)\n", num_online_cpus());
1827 seq_puts(m, "# -----------------\n");
1828 seq_printf(m, "# | task: %.16s-%d "
1829 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1830 data->comm, data->pid, data->uid, data->nice,
1831 data->policy, data->rt_priority);
1832 seq_puts(m, "# -----------------\n");
1834 if (data->critical_start) {
1835 seq_puts(m, "# => started at: ");
1836 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1837 trace_print_seq(m, &iter->seq);
1838 seq_puts(m, "\n# => ended at: ");
1839 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1840 trace_print_seq(m, &iter->seq);
1841 seq_puts(m, "\n#\n");
1847 static void test_cpu_buff_start(struct trace_iterator *iter)
1849 struct trace_seq *s = &iter->seq;
1851 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1854 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1857 if (cpumask_test_cpu(iter->cpu, iter->started))
1860 if (iter->tr->data[iter->cpu]->skipped_entries)
1863 cpumask_set_cpu(iter->cpu, iter->started);
1865 /* Don't print started cpu buffer for the first entry of the trace */
1867 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1871 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1873 struct trace_seq *s = &iter->seq;
1874 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1875 struct trace_entry *entry;
1876 struct trace_event *event;
1880 test_cpu_buff_start(iter);
1882 event = ftrace_find_event(entry->type);
1884 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1885 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1886 if (!trace_print_lat_context(iter))
1889 if (!trace_print_context(iter))
1895 return event->funcs->trace(iter, sym_flags, event);
1897 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1900 return TRACE_TYPE_HANDLED;
1902 return TRACE_TYPE_PARTIAL_LINE;
1905 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1907 struct trace_seq *s = &iter->seq;
1908 struct trace_entry *entry;
1909 struct trace_event *event;
1913 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1914 if (!trace_seq_printf(s, "%d %d %llu ",
1915 entry->pid, iter->cpu, iter->ts))
1919 event = ftrace_find_event(entry->type);
1921 return event->funcs->raw(iter, 0, event);
1923 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1926 return TRACE_TYPE_HANDLED;
1928 return TRACE_TYPE_PARTIAL_LINE;
1931 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1933 struct trace_seq *s = &iter->seq;
1934 unsigned char newline = '\n';
1935 struct trace_entry *entry;
1936 struct trace_event *event;
1940 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1941 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1942 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1943 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1946 event = ftrace_find_event(entry->type);
1948 enum print_line_t ret = event->funcs->hex(iter, 0, event);
1949 if (ret != TRACE_TYPE_HANDLED)
1953 SEQ_PUT_FIELD_RET(s, newline);
1955 return TRACE_TYPE_HANDLED;
1958 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1960 struct trace_seq *s = &iter->seq;
1961 struct trace_entry *entry;
1962 struct trace_event *event;
1966 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1967 SEQ_PUT_FIELD_RET(s, entry->pid);
1968 SEQ_PUT_FIELD_RET(s, iter->cpu);
1969 SEQ_PUT_FIELD_RET(s, iter->ts);
1972 event = ftrace_find_event(entry->type);
1973 return event ? event->funcs->binary(iter, 0, event) :
1977 int trace_empty(struct trace_iterator *iter)
1981 /* If we are looking at one CPU buffer, only check that one */
1982 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1983 cpu = iter->cpu_file;
1984 if (iter->buffer_iter[cpu]) {
1985 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1988 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1994 for_each_tracing_cpu(cpu) {
1995 if (iter->buffer_iter[cpu]) {
1996 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1999 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2007 /* Called with trace_event_read_lock() held. */
2008 enum print_line_t print_trace_line(struct trace_iterator *iter)
2010 enum print_line_t ret;
2012 if (iter->lost_events)
2013 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2014 iter->cpu, iter->lost_events);
2016 if (iter->trace && iter->trace->print_line) {
2017 ret = iter->trace->print_line(iter);
2018 if (ret != TRACE_TYPE_UNHANDLED)
2022 if (iter->ent->type == TRACE_BPRINT &&
2023 trace_flags & TRACE_ITER_PRINTK &&
2024 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2025 return trace_print_bprintk_msg_only(iter);
2027 if (iter->ent->type == TRACE_PRINT &&
2028 trace_flags & TRACE_ITER_PRINTK &&
2029 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2030 return trace_print_printk_msg_only(iter);
2032 if (trace_flags & TRACE_ITER_BIN)
2033 return print_bin_fmt(iter);
2035 if (trace_flags & TRACE_ITER_HEX)
2036 return print_hex_fmt(iter);
2038 if (trace_flags & TRACE_ITER_RAW)
2039 return print_raw_fmt(iter);
2041 return print_trace_fmt(iter);
2044 void trace_default_header(struct seq_file *m)
2046 struct trace_iterator *iter = m->private;
2048 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2049 /* print nothing if the buffers are empty */
2050 if (trace_empty(iter))
2052 print_trace_header(m, iter);
2053 if (!(trace_flags & TRACE_ITER_VERBOSE))
2054 print_lat_help_header(m);
2056 if (!(trace_flags & TRACE_ITER_VERBOSE))
2057 print_func_help_header(m);
2061 static int s_show(struct seq_file *m, void *v)
2063 struct trace_iterator *iter = v;
2066 if (iter->ent == NULL) {
2068 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2071 if (iter->trace && iter->trace->print_header)
2072 iter->trace->print_header(m);
2074 trace_default_header(m);
2076 } else if (iter->leftover) {
2078 * If we filled the seq_file buffer earlier, we
2079 * want to just show it now.
2081 ret = trace_print_seq(m, &iter->seq);
2083 /* ret should this time be zero, but you never know */
2084 iter->leftover = ret;
2087 print_trace_line(iter);
2088 ret = trace_print_seq(m, &iter->seq);
2090 * If we overflow the seq_file buffer, then it will
2091 * ask us for this data again at start up.
2093 * ret is 0 if seq_file write succeeded.
2096 iter->leftover = ret;
2102 static const struct seq_operations tracer_seq_ops = {
2109 static struct trace_iterator *
2110 __tracing_open(struct inode *inode, struct file *file)
2112 long cpu_file = (long) inode->i_private;
2113 void *fail_ret = ERR_PTR(-ENOMEM);
2114 struct trace_iterator *iter;
2118 if (tracing_disabled)
2119 return ERR_PTR(-ENODEV);
2121 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2123 return ERR_PTR(-ENOMEM);
2126 * We make a copy of the current tracer to avoid concurrent
2127 * changes on it while we are reading.
2129 mutex_lock(&trace_types_lock);
2130 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2135 *iter->trace = *current_trace;
2137 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2140 if (current_trace && current_trace->print_max)
2143 iter->tr = &global_trace;
2145 mutex_init(&iter->mutex);
2146 iter->cpu_file = cpu_file;
2148 /* Notify the tracer early; before we stop tracing. */
2149 if (iter->trace && iter->trace->open)
2150 iter->trace->open(iter);
2152 /* Annotate start of buffers if we had overruns */
2153 if (ring_buffer_overruns(iter->tr->buffer))
2154 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2156 /* stop the trace while dumping */
2159 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2160 for_each_tracing_cpu(cpu) {
2161 iter->buffer_iter[cpu] =
2162 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2164 ring_buffer_read_prepare_sync();
2165 for_each_tracing_cpu(cpu) {
2166 ring_buffer_read_start(iter->buffer_iter[cpu]);
2167 tracing_iter_reset(iter, cpu);
2170 cpu = iter->cpu_file;
2171 iter->buffer_iter[cpu] =
2172 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2173 ring_buffer_read_prepare_sync();
2174 ring_buffer_read_start(iter->buffer_iter[cpu]);
2175 tracing_iter_reset(iter, cpu);
2178 ret = seq_open(file, &tracer_seq_ops);
2180 fail_ret = ERR_PTR(ret);
2184 m = file->private_data;
2187 mutex_unlock(&trace_types_lock);
2192 for_each_tracing_cpu(cpu) {
2193 if (iter->buffer_iter[cpu])
2194 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2196 free_cpumask_var(iter->started);
2199 mutex_unlock(&trace_types_lock);
2206 int tracing_open_generic(struct inode *inode, struct file *filp)
2208 if (tracing_disabled)
2211 filp->private_data = inode->i_private;
2215 static int tracing_release(struct inode *inode, struct file *file)
2217 struct seq_file *m = file->private_data;
2218 struct trace_iterator *iter;
2221 if (!(file->f_mode & FMODE_READ))
2226 mutex_lock(&trace_types_lock);
2227 for_each_tracing_cpu(cpu) {
2228 if (iter->buffer_iter[cpu])
2229 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2232 if (iter->trace && iter->trace->close)
2233 iter->trace->close(iter);
2235 /* reenable tracing if it was previously enabled */
2237 mutex_unlock(&trace_types_lock);
2239 seq_release(inode, file);
2240 mutex_destroy(&iter->mutex);
2241 free_cpumask_var(iter->started);
2247 static int tracing_open(struct inode *inode, struct file *file)
2249 struct trace_iterator *iter;
2252 /* If this file was open for write, then erase contents */
2253 if ((file->f_mode & FMODE_WRITE) &&
2254 (file->f_flags & O_TRUNC)) {
2255 long cpu = (long) inode->i_private;
2257 if (cpu == TRACE_PIPE_ALL_CPU)
2258 tracing_reset_online_cpus(&global_trace);
2260 tracing_reset(&global_trace, cpu);
2263 if (file->f_mode & FMODE_READ) {
2264 iter = __tracing_open(inode, file);
2266 ret = PTR_ERR(iter);
2267 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2268 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2274 t_next(struct seq_file *m, void *v, loff_t *pos)
2276 struct tracer *t = v;
2286 static void *t_start(struct seq_file *m, loff_t *pos)
2291 mutex_lock(&trace_types_lock);
2292 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2298 static void t_stop(struct seq_file *m, void *p)
2300 mutex_unlock(&trace_types_lock);
2303 static int t_show(struct seq_file *m, void *v)
2305 struct tracer *t = v;
2310 seq_printf(m, "%s", t->name);
2319 static const struct seq_operations show_traces_seq_ops = {
2326 static int show_traces_open(struct inode *inode, struct file *file)
2328 if (tracing_disabled)
2331 return seq_open(file, &show_traces_seq_ops);
2335 tracing_write_stub(struct file *filp, const char __user *ubuf,
2336 size_t count, loff_t *ppos)
2341 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
2343 if (file->f_mode & FMODE_READ)
2344 return seq_lseek(file, offset, origin);
2349 static const struct file_operations tracing_fops = {
2350 .open = tracing_open,
2352 .write = tracing_write_stub,
2353 .llseek = tracing_seek,
2354 .release = tracing_release,
2357 static const struct file_operations show_traces_fops = {
2358 .open = show_traces_open,
2360 .release = seq_release,
2361 .llseek = seq_lseek,
2365 * Only trace on a CPU if the bitmask is set:
2367 static cpumask_var_t tracing_cpumask;
2370 * The tracer itself will not take this lock, but still we want
2371 * to provide a consistent cpumask to user-space:
2373 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2376 * Temporary storage for the character representation of the
2377 * CPU bitmask (and one more byte for the newline):
2379 static char mask_str[NR_CPUS + 1];
2382 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2383 size_t count, loff_t *ppos)
2387 mutex_lock(&tracing_cpumask_update_lock);
2389 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2390 if (count - len < 2) {
2394 len += sprintf(mask_str + len, "\n");
2395 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2398 mutex_unlock(&tracing_cpumask_update_lock);
2404 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2405 size_t count, loff_t *ppos)
2408 cpumask_var_t tracing_cpumask_new;
2410 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2413 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2417 mutex_lock(&tracing_cpumask_update_lock);
2419 local_irq_disable();
2420 arch_spin_lock(&ftrace_max_lock);
2421 for_each_tracing_cpu(cpu) {
2423 * Increase/decrease the disabled counter if we are
2424 * about to flip a bit in the cpumask:
2426 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2427 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2428 atomic_inc(&global_trace.data[cpu]->disabled);
2430 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2431 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2432 atomic_dec(&global_trace.data[cpu]->disabled);
2435 arch_spin_unlock(&ftrace_max_lock);
2438 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2440 mutex_unlock(&tracing_cpumask_update_lock);
2441 free_cpumask_var(tracing_cpumask_new);
2446 free_cpumask_var(tracing_cpumask_new);
2451 static const struct file_operations tracing_cpumask_fops = {
2452 .open = tracing_open_generic,
2453 .read = tracing_cpumask_read,
2454 .write = tracing_cpumask_write,
2455 .llseek = generic_file_llseek,
2458 static int tracing_trace_options_show(struct seq_file *m, void *v)
2460 struct tracer_opt *trace_opts;
2464 mutex_lock(&trace_types_lock);
2465 tracer_flags = current_trace->flags->val;
2466 trace_opts = current_trace->flags->opts;
2468 for (i = 0; trace_options[i]; i++) {
2469 if (trace_flags & (1 << i))
2470 seq_printf(m, "%s\n", trace_options[i]);
2472 seq_printf(m, "no%s\n", trace_options[i]);
2475 for (i = 0; trace_opts[i].name; i++) {
2476 if (tracer_flags & trace_opts[i].bit)
2477 seq_printf(m, "%s\n", trace_opts[i].name);
2479 seq_printf(m, "no%s\n", trace_opts[i].name);
2481 mutex_unlock(&trace_types_lock);
2486 static int __set_tracer_option(struct tracer *trace,
2487 struct tracer_flags *tracer_flags,
2488 struct tracer_opt *opts, int neg)
2492 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2497 tracer_flags->val &= ~opts->bit;
2499 tracer_flags->val |= opts->bit;
2503 /* Try to assign a tracer specific option */
2504 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2506 struct tracer_flags *tracer_flags = trace->flags;
2507 struct tracer_opt *opts = NULL;
2510 for (i = 0; tracer_flags->opts[i].name; i++) {
2511 opts = &tracer_flags->opts[i];
2513 if (strcmp(cmp, opts->name) == 0)
2514 return __set_tracer_option(trace, trace->flags,
2521 static void set_tracer_flags(unsigned int mask, int enabled)
2523 /* do nothing if flag is already set */
2524 if (!!(trace_flags & mask) == !!enabled)
2528 trace_flags |= mask;
2530 trace_flags &= ~mask;
2532 if (mask == TRACE_ITER_RECORD_CMD)
2533 trace_event_enable_cmd_record(enabled);
2537 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2538 size_t cnt, loff_t *ppos)
2546 if (cnt >= sizeof(buf))
2549 if (copy_from_user(&buf, ubuf, cnt))
2553 cmp = strstrip(buf);
2555 if (strncmp(cmp, "no", 2) == 0) {
2560 for (i = 0; trace_options[i]; i++) {
2561 if (strcmp(cmp, trace_options[i]) == 0) {
2562 set_tracer_flags(1 << i, !neg);
2567 /* If no option could be set, test the specific tracer options */
2568 if (!trace_options[i]) {
2569 mutex_lock(&trace_types_lock);
2570 ret = set_tracer_option(current_trace, cmp, neg);
2571 mutex_unlock(&trace_types_lock);
2581 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2583 if (tracing_disabled)
2585 return single_open(file, tracing_trace_options_show, NULL);
2588 static const struct file_operations tracing_iter_fops = {
2589 .open = tracing_trace_options_open,
2591 .llseek = seq_lseek,
2592 .release = single_release,
2593 .write = tracing_trace_options_write,
2596 static const char readme_msg[] =
2597 "tracing mini-HOWTO:\n\n"
2598 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2599 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2600 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2601 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2603 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2604 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2606 "# cat /sys/kernel/debug/tracing/trace_options\n"
2607 "noprint-parent nosym-offset nosym-addr noverbose\n"
2608 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2609 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2610 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2611 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2615 tracing_readme_read(struct file *filp, char __user *ubuf,
2616 size_t cnt, loff_t *ppos)
2618 return simple_read_from_buffer(ubuf, cnt, ppos,
2619 readme_msg, strlen(readme_msg));
2622 static const struct file_operations tracing_readme_fops = {
2623 .open = tracing_open_generic,
2624 .read = tracing_readme_read,
2625 .llseek = generic_file_llseek,
2629 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2630 size_t cnt, loff_t *ppos)
2639 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2643 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2651 for (i = 0; i < SAVED_CMDLINES; i++) {
2654 pid = map_cmdline_to_pid[i];
2655 if (pid == -1 || pid == NO_CMDLINE_MAP)
2658 trace_find_cmdline(pid, buf_comm);
2659 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2664 len = simple_read_from_buffer(ubuf, cnt, ppos,
2673 static const struct file_operations tracing_saved_cmdlines_fops = {
2674 .open = tracing_open_generic,
2675 .read = tracing_saved_cmdlines_read,
2676 .llseek = generic_file_llseek,
2680 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2681 size_t cnt, loff_t *ppos)
2686 r = sprintf(buf, "%u\n", tracer_enabled);
2687 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2691 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2692 size_t cnt, loff_t *ppos)
2694 struct trace_array *tr = filp->private_data;
2699 if (cnt >= sizeof(buf))
2702 if (copy_from_user(&buf, ubuf, cnt))
2707 ret = strict_strtoul(buf, 10, &val);
2713 mutex_lock(&trace_types_lock);
2714 if (tracer_enabled ^ val) {
2717 if (current_trace->start)
2718 current_trace->start(tr);
2723 if (current_trace->stop)
2724 current_trace->stop(tr);
2727 mutex_unlock(&trace_types_lock);
2735 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2736 size_t cnt, loff_t *ppos)
2738 char buf[MAX_TRACER_SIZE+2];
2741 mutex_lock(&trace_types_lock);
2743 r = sprintf(buf, "%s\n", current_trace->name);
2745 r = sprintf(buf, "\n");
2746 mutex_unlock(&trace_types_lock);
2748 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2751 int tracer_init(struct tracer *t, struct trace_array *tr)
2753 tracing_reset_online_cpus(tr);
2757 static int tracing_resize_ring_buffer(unsigned long size)
2762 * If kernel or user changes the size of the ring buffer
2763 * we use the size that was given, and we can forget about
2764 * expanding it later.
2766 ring_buffer_expanded = 1;
2768 ret = ring_buffer_resize(global_trace.buffer, size);
2772 if (!current_trace->use_max_tr)
2775 ret = ring_buffer_resize(max_tr.buffer, size);
2779 r = ring_buffer_resize(global_trace.buffer,
2780 global_trace.entries);
2783 * AARGH! We are left with different
2784 * size max buffer!!!!
2785 * The max buffer is our "snapshot" buffer.
2786 * When a tracer needs a snapshot (one of the
2787 * latency tracers), it swaps the max buffer
2788 * with the saved snap shot. We succeeded to
2789 * update the size of the main buffer, but failed to
2790 * update the size of the max buffer. But when we tried
2791 * to reset the main buffer to the original size, we
2792 * failed there too. This is very unlikely to
2793 * happen, but if it does, warn and kill all
2797 tracing_disabled = 1;
2802 max_tr.entries = size;
2804 global_trace.entries = size;
2811 * tracing_update_buffers - used by tracing facility to expand ring buffers
2813 * To save on memory when the tracing is never used on a system with it
2814 * configured in. The ring buffers are set to a minimum size. But once
2815 * a user starts to use the tracing facility, then they need to grow
2816 * to their default size.
2818 * This function is to be called when a tracer is about to be used.
2820 int tracing_update_buffers(void)
2824 mutex_lock(&trace_types_lock);
2825 if (!ring_buffer_expanded)
2826 ret = tracing_resize_ring_buffer(trace_buf_size);
2827 mutex_unlock(&trace_types_lock);
2832 struct trace_option_dentry;
2834 static struct trace_option_dentry *
2835 create_trace_option_files(struct tracer *tracer);
2838 destroy_trace_option_files(struct trace_option_dentry *topts);
2840 static int tracing_set_tracer(const char *buf)
2842 static struct trace_option_dentry *topts;
2843 struct trace_array *tr = &global_trace;
2847 mutex_lock(&trace_types_lock);
2849 if (!ring_buffer_expanded) {
2850 ret = tracing_resize_ring_buffer(trace_buf_size);
2856 for (t = trace_types; t; t = t->next) {
2857 if (strcmp(t->name, buf) == 0)
2864 if (t == current_trace)
2867 trace_branch_disable();
2868 if (current_trace && current_trace->reset)
2869 current_trace->reset(tr);
2870 if (current_trace && current_trace->use_max_tr) {
2872 * We don't free the ring buffer. instead, resize it because
2873 * The max_tr ring buffer has some state (e.g. ring->clock) and
2874 * we want preserve it.
2876 ring_buffer_resize(max_tr.buffer, 1);
2879 destroy_trace_option_files(topts);
2883 topts = create_trace_option_files(current_trace);
2884 if (current_trace->use_max_tr) {
2885 ret = ring_buffer_resize(max_tr.buffer, global_trace.entries);
2888 max_tr.entries = global_trace.entries;
2892 ret = tracer_init(t, tr);
2897 trace_branch_enable(tr);
2899 mutex_unlock(&trace_types_lock);
2905 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2906 size_t cnt, loff_t *ppos)
2908 char buf[MAX_TRACER_SIZE+1];
2915 if (cnt > MAX_TRACER_SIZE)
2916 cnt = MAX_TRACER_SIZE;
2918 if (copy_from_user(&buf, ubuf, cnt))
2923 /* strip ending whitespace. */
2924 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2927 err = tracing_set_tracer(buf);
2937 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2938 size_t cnt, loff_t *ppos)
2940 unsigned long *ptr = filp->private_data;
2944 r = snprintf(buf, sizeof(buf), "%ld\n",
2945 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2946 if (r > sizeof(buf))
2948 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2952 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2953 size_t cnt, loff_t *ppos)
2955 unsigned long *ptr = filp->private_data;
2960 if (cnt >= sizeof(buf))
2963 if (copy_from_user(&buf, ubuf, cnt))
2968 ret = strict_strtoul(buf, 10, &val);
2977 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2979 long cpu_file = (long) inode->i_private;
2980 struct trace_iterator *iter;
2983 if (tracing_disabled)
2986 mutex_lock(&trace_types_lock);
2988 /* create a buffer to store the information to pass to userspace */
2989 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2996 * We make a copy of the current tracer to avoid concurrent
2997 * changes on it while we are reading.
2999 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3005 *iter->trace = *current_trace;
3007 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3012 /* trace pipe does not show start of buffer */
3013 cpumask_setall(iter->started);
3015 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3016 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3018 iter->cpu_file = cpu_file;
3019 iter->tr = &global_trace;
3020 mutex_init(&iter->mutex);
3021 filp->private_data = iter;
3023 if (iter->trace->pipe_open)
3024 iter->trace->pipe_open(iter);
3026 nonseekable_open(inode, filp);
3028 mutex_unlock(&trace_types_lock);
3034 mutex_unlock(&trace_types_lock);
3038 static int tracing_release_pipe(struct inode *inode, struct file *file)
3040 struct trace_iterator *iter = file->private_data;
3042 mutex_lock(&trace_types_lock);
3044 if (iter->trace->pipe_close)
3045 iter->trace->pipe_close(iter);
3047 mutex_unlock(&trace_types_lock);
3049 free_cpumask_var(iter->started);
3050 mutex_destroy(&iter->mutex);
3058 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3060 struct trace_iterator *iter = filp->private_data;
3062 if (trace_flags & TRACE_ITER_BLOCK) {
3064 * Always select as readable when in blocking mode
3066 return POLLIN | POLLRDNORM;
3068 if (!trace_empty(iter))
3069 return POLLIN | POLLRDNORM;
3070 poll_wait(filp, &trace_wait, poll_table);
3071 if (!trace_empty(iter))
3072 return POLLIN | POLLRDNORM;
3079 void default_wait_pipe(struct trace_iterator *iter)
3083 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3085 if (trace_empty(iter))
3088 finish_wait(&trace_wait, &wait);
3092 * This is a make-shift waitqueue.
3093 * A tracer might use this callback on some rare cases:
3095 * 1) the current tracer might hold the runqueue lock when it wakes up
3096 * a reader, hence a deadlock (sched, function, and function graph tracers)
3097 * 2) the function tracers, trace all functions, we don't want
3098 * the overhead of calling wake_up and friends
3099 * (and tracing them too)
3101 * Anyway, this is really very primitive wakeup.
3103 void poll_wait_pipe(struct trace_iterator *iter)
3105 set_current_state(TASK_INTERRUPTIBLE);
3106 /* sleep for 100 msecs, and try again. */
3107 schedule_timeout(HZ / 10);
3110 /* Must be called with trace_types_lock mutex held. */
3111 static int tracing_wait_pipe(struct file *filp)
3113 struct trace_iterator *iter = filp->private_data;
3115 while (trace_empty(iter)) {
3117 if ((filp->f_flags & O_NONBLOCK)) {
3121 mutex_unlock(&iter->mutex);
3123 iter->trace->wait_pipe(iter);
3125 mutex_lock(&iter->mutex);
3127 if (signal_pending(current))
3131 * We block until we read something and tracing is disabled.
3132 * We still block if tracing is disabled, but we have never
3133 * read anything. This allows a user to cat this file, and
3134 * then enable tracing. But after we have read something,
3135 * we give an EOF when tracing is again disabled.
3137 * iter->pos will be 0 if we haven't read anything.
3139 if (!tracer_enabled && iter->pos)
3150 tracing_read_pipe(struct file *filp, char __user *ubuf,
3151 size_t cnt, loff_t *ppos)
3153 struct trace_iterator *iter = filp->private_data;
3154 static struct tracer *old_tracer;
3157 /* return any leftover data */
3158 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3162 trace_seq_init(&iter->seq);
3164 /* copy the tracer to avoid using a global lock all around */
3165 mutex_lock(&trace_types_lock);
3166 if (unlikely(old_tracer != current_trace && current_trace)) {
3167 old_tracer = current_trace;
3168 *iter->trace = *current_trace;
3170 mutex_unlock(&trace_types_lock);
3173 * Avoid more than one consumer on a single file descriptor
3174 * This is just a matter of traces coherency, the ring buffer itself
3177 mutex_lock(&iter->mutex);
3178 if (iter->trace->read) {
3179 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3185 sret = tracing_wait_pipe(filp);
3189 /* stop when tracing is finished */
3190 if (trace_empty(iter)) {
3195 if (cnt >= PAGE_SIZE)
3196 cnt = PAGE_SIZE - 1;
3198 /* reset all but tr, trace, and overruns */
3199 memset(&iter->seq, 0,
3200 sizeof(struct trace_iterator) -
3201 offsetof(struct trace_iterator, seq));
3204 trace_event_read_lock();
3205 trace_access_lock(iter->cpu_file);
3206 while (trace_find_next_entry_inc(iter) != NULL) {
3207 enum print_line_t ret;
3208 int len = iter->seq.len;
3210 ret = print_trace_line(iter);
3211 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3212 /* don't print partial lines */
3213 iter->seq.len = len;
3216 if (ret != TRACE_TYPE_NO_CONSUME)
3217 trace_consume(iter);
3219 if (iter->seq.len >= cnt)
3222 trace_access_unlock(iter->cpu_file);
3223 trace_event_read_unlock();
3225 /* Now copy what we have to the user */
3226 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3227 if (iter->seq.readpos >= iter->seq.len)
3228 trace_seq_init(&iter->seq);
3231 * If there was nothing to send to user, inspite of consuming trace
3232 * entries, go back to wait for more entries.
3238 mutex_unlock(&iter->mutex);
3243 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3244 struct pipe_buffer *buf)
3246 __free_page(buf->page);
3249 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3252 __free_page(spd->pages[idx]);
3255 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3257 .map = generic_pipe_buf_map,
3258 .unmap = generic_pipe_buf_unmap,
3259 .confirm = generic_pipe_buf_confirm,
3260 .release = tracing_pipe_buf_release,
3261 .steal = generic_pipe_buf_steal,
3262 .get = generic_pipe_buf_get,
3266 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3271 /* Seq buffer is page-sized, exactly what we need. */
3273 count = iter->seq.len;
3274 ret = print_trace_line(iter);
3275 count = iter->seq.len - count;
3278 iter->seq.len -= count;
3281 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3282 iter->seq.len -= count;
3286 if (ret != TRACE_TYPE_NO_CONSUME)
3287 trace_consume(iter);
3289 if (!trace_find_next_entry_inc(iter)) {
3299 static ssize_t tracing_splice_read_pipe(struct file *filp,
3301 struct pipe_inode_info *pipe,
3305 struct page *pages_def[PIPE_DEF_BUFFERS];
3306 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3307 struct trace_iterator *iter = filp->private_data;
3308 struct splice_pipe_desc spd = {
3310 .partial = partial_def,
3311 .nr_pages = 0, /* This gets updated below. */
3313 .ops = &tracing_pipe_buf_ops,
3314 .spd_release = tracing_spd_release_pipe,
3316 static struct tracer *old_tracer;
3321 if (splice_grow_spd(pipe, &spd))
3324 /* copy the tracer to avoid using a global lock all around */
3325 mutex_lock(&trace_types_lock);
3326 if (unlikely(old_tracer != current_trace && current_trace)) {
3327 old_tracer = current_trace;
3328 *iter->trace = *current_trace;
3330 mutex_unlock(&trace_types_lock);
3332 mutex_lock(&iter->mutex);
3334 if (iter->trace->splice_read) {
3335 ret = iter->trace->splice_read(iter, filp,
3336 ppos, pipe, len, flags);
3341 ret = tracing_wait_pipe(filp);
3345 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
3350 trace_event_read_lock();
3351 trace_access_lock(iter->cpu_file);
3353 /* Fill as many pages as possible. */
3354 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3355 spd.pages[i] = alloc_page(GFP_KERNEL);
3359 rem = tracing_fill_pipe_page(rem, iter);
3361 /* Copy the data into the page, so we can start over. */
3362 ret = trace_seq_to_buffer(&iter->seq,
3363 page_address(spd.pages[i]),
3366 __free_page(spd.pages[i]);
3369 spd.partial[i].offset = 0;
3370 spd.partial[i].len = iter->seq.len;
3372 trace_seq_init(&iter->seq);
3375 trace_access_unlock(iter->cpu_file);
3376 trace_event_read_unlock();
3377 mutex_unlock(&iter->mutex);
3381 ret = splice_to_pipe(pipe, &spd);
3383 splice_shrink_spd(pipe, &spd);
3387 mutex_unlock(&iter->mutex);
3392 tracing_entries_read(struct file *filp, char __user *ubuf,
3393 size_t cnt, loff_t *ppos)
3395 struct trace_array *tr = filp->private_data;
3399 mutex_lock(&trace_types_lock);
3400 if (!ring_buffer_expanded)
3401 r = sprintf(buf, "%lu (expanded: %lu)\n",
3403 trace_buf_size >> 10);
3405 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3406 mutex_unlock(&trace_types_lock);
3408 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3412 tracing_entries_write(struct file *filp, const char __user *ubuf,
3413 size_t cnt, loff_t *ppos)
3419 if (cnt >= sizeof(buf))
3422 if (copy_from_user(&buf, ubuf, cnt))
3427 ret = strict_strtoul(buf, 10, &val);
3431 /* must have at least 1 entry */
3435 mutex_lock(&trace_types_lock);
3439 /* disable all cpu buffers */
3440 for_each_tracing_cpu(cpu) {
3441 if (global_trace.data[cpu])
3442 atomic_inc(&global_trace.data[cpu]->disabled);
3443 if (max_tr.data[cpu])
3444 atomic_inc(&max_tr.data[cpu]->disabled);
3447 /* value is in KB */
3450 if (val != global_trace.entries) {
3451 ret = tracing_resize_ring_buffer(val);
3460 /* If check pages failed, return ENOMEM */
3461 if (tracing_disabled)
3464 for_each_tracing_cpu(cpu) {
3465 if (global_trace.data[cpu])
3466 atomic_dec(&global_trace.data[cpu]->disabled);
3467 if (max_tr.data[cpu])
3468 atomic_dec(&max_tr.data[cpu]->disabled);
3472 mutex_unlock(&trace_types_lock);
3477 static int mark_printk(const char *fmt, ...)
3481 va_start(args, fmt);
3482 ret = trace_vprintk(0, fmt, args);
3488 tracing_mark_write(struct file *filp, const char __user *ubuf,
3489 size_t cnt, loff_t *fpos)
3494 if (tracing_disabled)
3497 if (cnt > TRACE_BUF_SIZE)
3498 cnt = TRACE_BUF_SIZE;
3500 buf = kmalloc(cnt + 2, GFP_KERNEL);
3504 if (copy_from_user(buf, ubuf, cnt)) {
3508 if (buf[cnt-1] != '\n') {
3514 written = mark_printk("%s", buf);
3518 /* don't tell userspace we wrote more - it might confuse them */
3525 static int tracing_clock_show(struct seq_file *m, void *v)
3529 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3531 "%s%s%s%s", i ? " " : "",
3532 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3533 i == trace_clock_id ? "]" : "");
3539 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3540 size_t cnt, loff_t *fpos)
3543 const char *clockstr;
3546 if (cnt >= sizeof(buf))
3549 if (copy_from_user(&buf, ubuf, cnt))
3554 clockstr = strstrip(buf);
3556 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3557 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3560 if (i == ARRAY_SIZE(trace_clocks))
3565 mutex_lock(&trace_types_lock);
3567 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3569 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3571 mutex_unlock(&trace_types_lock);
3578 static int tracing_clock_open(struct inode *inode, struct file *file)
3580 if (tracing_disabled)
3582 return single_open(file, tracing_clock_show, NULL);
3585 static const struct file_operations tracing_max_lat_fops = {
3586 .open = tracing_open_generic,
3587 .read = tracing_max_lat_read,
3588 .write = tracing_max_lat_write,
3589 .llseek = generic_file_llseek,
3592 static const struct file_operations tracing_ctrl_fops = {
3593 .open = tracing_open_generic,
3594 .read = tracing_ctrl_read,
3595 .write = tracing_ctrl_write,
3596 .llseek = generic_file_llseek,
3599 static const struct file_operations set_tracer_fops = {
3600 .open = tracing_open_generic,
3601 .read = tracing_set_trace_read,
3602 .write = tracing_set_trace_write,
3603 .llseek = generic_file_llseek,
3606 static const struct file_operations tracing_pipe_fops = {
3607 .open = tracing_open_pipe,
3608 .poll = tracing_poll_pipe,
3609 .read = tracing_read_pipe,
3610 .splice_read = tracing_splice_read_pipe,
3611 .release = tracing_release_pipe,
3612 .llseek = no_llseek,
3615 static const struct file_operations tracing_entries_fops = {
3616 .open = tracing_open_generic,
3617 .read = tracing_entries_read,
3618 .write = tracing_entries_write,
3619 .llseek = generic_file_llseek,
3622 static const struct file_operations tracing_mark_fops = {
3623 .open = tracing_open_generic,
3624 .write = tracing_mark_write,
3625 .llseek = generic_file_llseek,
3628 static const struct file_operations trace_clock_fops = {
3629 .open = tracing_clock_open,
3631 .llseek = seq_lseek,
3632 .release = single_release,
3633 .write = tracing_clock_write,
3636 struct ftrace_buffer_info {
3637 struct trace_array *tr;
3643 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3645 int cpu = (int)(long)inode->i_private;
3646 struct ftrace_buffer_info *info;
3648 if (tracing_disabled)
3651 info = kzalloc(sizeof(*info), GFP_KERNEL);
3655 info->tr = &global_trace;
3658 /* Force reading ring buffer for first read */
3659 info->read = (unsigned int)-1;
3661 filp->private_data = info;
3663 return nonseekable_open(inode, filp);
3667 tracing_buffers_read(struct file *filp, char __user *ubuf,
3668 size_t count, loff_t *ppos)
3670 struct ftrace_buffer_info *info = filp->private_data;
3678 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3682 /* Do we have previous read data to read? */
3683 if (info->read < PAGE_SIZE)
3688 trace_access_lock(info->cpu);
3689 ret = ring_buffer_read_page(info->tr->buffer,
3693 trace_access_unlock(info->cpu);
3698 size = PAGE_SIZE - info->read;
3702 ret = copy_to_user(ubuf, info->spare + info->read, size);
3713 static int tracing_buffers_release(struct inode *inode, struct file *file)
3715 struct ftrace_buffer_info *info = file->private_data;
3718 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3725 struct ring_buffer *buffer;
3730 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3731 struct pipe_buffer *buf)
3733 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3738 ring_buffer_free_read_page(ref->buffer, ref->page);
3743 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3744 struct pipe_buffer *buf)
3749 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3750 struct pipe_buffer *buf)
3752 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3757 /* Pipe buffer operations for a buffer. */
3758 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
3760 .map = generic_pipe_buf_map,
3761 .unmap = generic_pipe_buf_unmap,
3762 .confirm = generic_pipe_buf_confirm,
3763 .release = buffer_pipe_buf_release,
3764 .steal = buffer_pipe_buf_steal,
3765 .get = buffer_pipe_buf_get,
3769 * Callback from splice_to_pipe(), if we need to release some pages
3770 * at the end of the spd in case we error'ed out in filling the pipe.
3772 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3774 struct buffer_ref *ref =
3775 (struct buffer_ref *)spd->partial[i].private;
3780 ring_buffer_free_read_page(ref->buffer, ref->page);
3782 spd->partial[i].private = 0;
3786 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3787 struct pipe_inode_info *pipe, size_t len,
3790 struct ftrace_buffer_info *info = file->private_data;
3791 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3792 struct page *pages_def[PIPE_DEF_BUFFERS];
3793 struct splice_pipe_desc spd = {
3795 .partial = partial_def,
3797 .ops = &buffer_pipe_buf_ops,
3798 .spd_release = buffer_spd_release,
3800 struct buffer_ref *ref;
3801 int entries, size, i;
3804 if (splice_grow_spd(pipe, &spd))
3807 if (*ppos & (PAGE_SIZE - 1)) {
3808 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3813 if (len & (PAGE_SIZE - 1)) {
3814 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3815 if (len < PAGE_SIZE) {
3822 trace_access_lock(info->cpu);
3823 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3825 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
3829 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3834 ref->buffer = info->tr->buffer;
3835 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3841 r = ring_buffer_read_page(ref->buffer, &ref->page,
3844 ring_buffer_free_read_page(ref->buffer,
3851 * zero out any left over data, this is going to
3854 size = ring_buffer_page_len(ref->page);
3855 if (size < PAGE_SIZE)
3856 memset(ref->page + size, 0, PAGE_SIZE - size);
3858 page = virt_to_page(ref->page);
3860 spd.pages[i] = page;
3861 spd.partial[i].len = PAGE_SIZE;
3862 spd.partial[i].offset = 0;
3863 spd.partial[i].private = (unsigned long)ref;
3867 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3870 trace_access_unlock(info->cpu);
3873 /* did we read anything? */
3874 if (!spd.nr_pages) {
3875 if (flags & SPLICE_F_NONBLOCK)
3883 ret = splice_to_pipe(pipe, &spd);
3884 splice_shrink_spd(pipe, &spd);
3889 static const struct file_operations tracing_buffers_fops = {
3890 .open = tracing_buffers_open,
3891 .read = tracing_buffers_read,
3892 .release = tracing_buffers_release,
3893 .splice_read = tracing_buffers_splice_read,
3894 .llseek = no_llseek,
3898 tracing_stats_read(struct file *filp, char __user *ubuf,
3899 size_t count, loff_t *ppos)
3901 unsigned long cpu = (unsigned long)filp->private_data;
3902 struct trace_array *tr = &global_trace;
3903 struct trace_seq *s;
3906 s = kmalloc(sizeof(*s), GFP_KERNEL);
3912 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3913 trace_seq_printf(s, "entries: %ld\n", cnt);
3915 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3916 trace_seq_printf(s, "overrun: %ld\n", cnt);
3918 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3919 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3921 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3928 static const struct file_operations tracing_stats_fops = {
3929 .open = tracing_open_generic,
3930 .read = tracing_stats_read,
3931 .llseek = generic_file_llseek,
3934 #ifdef CONFIG_DYNAMIC_FTRACE
3936 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3942 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3943 size_t cnt, loff_t *ppos)
3945 static char ftrace_dyn_info_buffer[1024];
3946 static DEFINE_MUTEX(dyn_info_mutex);
3947 unsigned long *p = filp->private_data;
3948 char *buf = ftrace_dyn_info_buffer;
3949 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3952 mutex_lock(&dyn_info_mutex);
3953 r = sprintf(buf, "%ld ", *p);
3955 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3958 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3960 mutex_unlock(&dyn_info_mutex);
3965 static const struct file_operations tracing_dyn_info_fops = {
3966 .open = tracing_open_generic,
3967 .read = tracing_read_dyn_info,
3968 .llseek = generic_file_llseek,
3972 static struct dentry *d_tracer;
3974 struct dentry *tracing_init_dentry(void)
3981 if (!debugfs_initialized())
3984 d_tracer = debugfs_create_dir("tracing", NULL);
3986 if (!d_tracer && !once) {
3988 pr_warning("Could not create debugfs directory 'tracing'\n");
3995 static struct dentry *d_percpu;
3997 struct dentry *tracing_dentry_percpu(void)
4000 struct dentry *d_tracer;
4005 d_tracer = tracing_init_dentry();
4010 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
4012 if (!d_percpu && !once) {
4014 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4021 static void tracing_init_debugfs_percpu(long cpu)
4023 struct dentry *d_percpu = tracing_dentry_percpu();
4024 struct dentry *d_cpu;
4025 char cpu_dir[30]; /* 30 characters should be more than enough */
4027 snprintf(cpu_dir, 30, "cpu%ld", cpu);
4028 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4030 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4034 /* per cpu trace_pipe */
4035 trace_create_file("trace_pipe", 0444, d_cpu,
4036 (void *) cpu, &tracing_pipe_fops);
4039 trace_create_file("trace", 0644, d_cpu,
4040 (void *) cpu, &tracing_fops);
4042 trace_create_file("trace_pipe_raw", 0444, d_cpu,
4043 (void *) cpu, &tracing_buffers_fops);
4045 trace_create_file("stats", 0444, d_cpu,
4046 (void *) cpu, &tracing_stats_fops);
4049 #ifdef CONFIG_FTRACE_SELFTEST
4050 /* Let selftest have access to static functions in this file */
4051 #include "trace_selftest.c"
4054 struct trace_option_dentry {
4055 struct tracer_opt *opt;
4056 struct tracer_flags *flags;
4057 struct dentry *entry;
4061 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4064 struct trace_option_dentry *topt = filp->private_data;
4067 if (topt->flags->val & topt->opt->bit)
4072 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4076 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4079 struct trace_option_dentry *topt = filp->private_data;
4084 if (cnt >= sizeof(buf))
4087 if (copy_from_user(&buf, ubuf, cnt))
4092 ret = strict_strtoul(buf, 10, &val);
4096 if (val != 0 && val != 1)
4099 if (!!(topt->flags->val & topt->opt->bit) != val) {
4100 mutex_lock(&trace_types_lock);
4101 ret = __set_tracer_option(current_trace, topt->flags,
4103 mutex_unlock(&trace_types_lock);
4114 static const struct file_operations trace_options_fops = {
4115 .open = tracing_open_generic,
4116 .read = trace_options_read,
4117 .write = trace_options_write,
4118 .llseek = generic_file_llseek,
4122 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4125 long index = (long)filp->private_data;
4128 if (trace_flags & (1 << index))
4133 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4137 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4140 long index = (long)filp->private_data;
4145 if (cnt >= sizeof(buf))
4148 if (copy_from_user(&buf, ubuf, cnt))
4153 ret = strict_strtoul(buf, 10, &val);
4157 if (val != 0 && val != 1)
4159 set_tracer_flags(1 << index, val);
4166 static const struct file_operations trace_options_core_fops = {
4167 .open = tracing_open_generic,
4168 .read = trace_options_core_read,
4169 .write = trace_options_core_write,
4170 .llseek = generic_file_llseek,
4173 struct dentry *trace_create_file(const char *name,
4175 struct dentry *parent,
4177 const struct file_operations *fops)
4181 ret = debugfs_create_file(name, mode, parent, data, fops);
4183 pr_warning("Could not create debugfs '%s' entry\n", name);
4189 static struct dentry *trace_options_init_dentry(void)
4191 struct dentry *d_tracer;
4192 static struct dentry *t_options;
4197 d_tracer = tracing_init_dentry();
4201 t_options = debugfs_create_dir("options", d_tracer);
4203 pr_warning("Could not create debugfs directory 'options'\n");
4211 create_trace_option_file(struct trace_option_dentry *topt,
4212 struct tracer_flags *flags,
4213 struct tracer_opt *opt)
4215 struct dentry *t_options;
4217 t_options = trace_options_init_dentry();
4221 topt->flags = flags;
4224 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4225 &trace_options_fops);
4229 static struct trace_option_dentry *
4230 create_trace_option_files(struct tracer *tracer)
4232 struct trace_option_dentry *topts;
4233 struct tracer_flags *flags;
4234 struct tracer_opt *opts;
4240 flags = tracer->flags;
4242 if (!flags || !flags->opts)
4247 for (cnt = 0; opts[cnt].name; cnt++)
4250 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4254 for (cnt = 0; opts[cnt].name; cnt++)
4255 create_trace_option_file(&topts[cnt], flags,
4262 destroy_trace_option_files(struct trace_option_dentry *topts)
4269 for (cnt = 0; topts[cnt].opt; cnt++) {
4270 if (topts[cnt].entry)
4271 debugfs_remove(topts[cnt].entry);
4277 static struct dentry *
4278 create_trace_option_core_file(const char *option, long index)
4280 struct dentry *t_options;
4282 t_options = trace_options_init_dentry();
4286 return trace_create_file(option, 0644, t_options, (void *)index,
4287 &trace_options_core_fops);
4290 static __init void create_trace_options_dir(void)
4292 struct dentry *t_options;
4295 t_options = trace_options_init_dentry();
4299 for (i = 0; trace_options[i]; i++)
4300 create_trace_option_core_file(trace_options[i], i);
4303 static __init int tracer_init_debugfs(void)
4305 struct dentry *d_tracer;
4308 trace_access_lock_init();
4310 d_tracer = tracing_init_dentry();
4312 trace_create_file("tracing_enabled", 0644, d_tracer,
4313 &global_trace, &tracing_ctrl_fops);
4315 trace_create_file("trace_options", 0644, d_tracer,
4316 NULL, &tracing_iter_fops);
4318 trace_create_file("tracing_cpumask", 0644, d_tracer,
4319 NULL, &tracing_cpumask_fops);
4321 trace_create_file("trace", 0644, d_tracer,
4322 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4324 trace_create_file("available_tracers", 0444, d_tracer,
4325 &global_trace, &show_traces_fops);
4327 trace_create_file("current_tracer", 0644, d_tracer,
4328 &global_trace, &set_tracer_fops);
4330 #ifdef CONFIG_TRACER_MAX_TRACE
4331 trace_create_file("tracing_max_latency", 0644, d_tracer,
4332 &tracing_max_latency, &tracing_max_lat_fops);
4335 trace_create_file("tracing_thresh", 0644, d_tracer,
4336 &tracing_thresh, &tracing_max_lat_fops);
4338 trace_create_file("README", 0444, d_tracer,
4339 NULL, &tracing_readme_fops);
4341 trace_create_file("trace_pipe", 0444, d_tracer,
4342 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4344 trace_create_file("buffer_size_kb", 0644, d_tracer,
4345 &global_trace, &tracing_entries_fops);
4347 trace_create_file("trace_marker", 0220, d_tracer,
4348 NULL, &tracing_mark_fops);
4350 trace_create_file("saved_cmdlines", 0444, d_tracer,
4351 NULL, &tracing_saved_cmdlines_fops);
4353 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4356 #ifdef CONFIG_DYNAMIC_FTRACE
4357 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4358 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4361 create_trace_options_dir();
4363 for_each_tracing_cpu(cpu)
4364 tracing_init_debugfs_percpu(cpu);
4369 static int trace_panic_handler(struct notifier_block *this,
4370 unsigned long event, void *unused)
4372 if (ftrace_dump_on_oops)
4373 ftrace_dump(ftrace_dump_on_oops);
4377 static struct notifier_block trace_panic_notifier = {
4378 .notifier_call = trace_panic_handler,
4380 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4383 static int trace_die_handler(struct notifier_block *self,
4389 if (ftrace_dump_on_oops)
4390 ftrace_dump(ftrace_dump_on_oops);
4398 static struct notifier_block trace_die_notifier = {
4399 .notifier_call = trace_die_handler,
4404 * printk is set to max of 1024, we really don't need it that big.
4405 * Nothing should be printing 1000 characters anyway.
4407 #define TRACE_MAX_PRINT 1000
4410 * Define here KERN_TRACE so that we have one place to modify
4411 * it if we decide to change what log level the ftrace dump
4414 #define KERN_TRACE KERN_EMERG
4417 trace_printk_seq(struct trace_seq *s)
4419 /* Probably should print a warning here. */
4423 /* should be zero ended, but we are paranoid. */
4424 s->buffer[s->len] = 0;
4426 printk(KERN_TRACE "%s", s->buffer);
4431 void trace_init_global_iter(struct trace_iterator *iter)
4433 iter->tr = &global_trace;
4434 iter->trace = current_trace;
4435 iter->cpu_file = TRACE_PIPE_ALL_CPU;
4439 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4441 static arch_spinlock_t ftrace_dump_lock =
4442 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4443 /* use static because iter can be a bit big for the stack */
4444 static struct trace_iterator iter;
4445 unsigned int old_userobj;
4446 static int dump_ran;
4447 unsigned long flags;
4451 local_irq_save(flags);
4452 arch_spin_lock(&ftrace_dump_lock);
4460 if (disable_tracing)
4463 trace_init_global_iter(&iter);
4465 for_each_tracing_cpu(cpu) {
4466 atomic_inc(&iter.tr->data[cpu]->disabled);
4469 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4471 /* don't look at user memory in panic mode */
4472 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4474 /* Simulate the iterator */
4475 iter.tr = &global_trace;
4476 iter.trace = current_trace;
4478 switch (oops_dump_mode) {
4480 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4483 iter.cpu_file = raw_smp_processor_id();
4488 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
4489 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4492 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4495 * We need to stop all tracing on all CPUS to read the
4496 * the next buffer. This is a bit expensive, but is
4497 * not done often. We fill all what we can read,
4498 * and then release the locks again.
4501 while (!trace_empty(&iter)) {
4504 printk(KERN_TRACE "---------------------------------\n");
4508 /* reset all but tr, trace, and overruns */
4509 memset(&iter.seq, 0,
4510 sizeof(struct trace_iterator) -
4511 offsetof(struct trace_iterator, seq));
4512 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4515 if (trace_find_next_entry_inc(&iter) != NULL) {
4518 ret = print_trace_line(&iter);
4519 if (ret != TRACE_TYPE_NO_CONSUME)
4520 trace_consume(&iter);
4523 trace_printk_seq(&iter.seq);
4527 printk(KERN_TRACE " (ftrace buffer empty)\n");
4529 printk(KERN_TRACE "---------------------------------\n");
4532 /* Re-enable tracing if requested */
4533 if (!disable_tracing) {
4534 trace_flags |= old_userobj;
4536 for_each_tracing_cpu(cpu) {
4537 atomic_dec(&iter.tr->data[cpu]->disabled);
4543 arch_spin_unlock(&ftrace_dump_lock);
4544 local_irq_restore(flags);
4547 /* By default: disable tracing after the dump */
4548 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
4550 __ftrace_dump(true, oops_dump_mode);
4553 __init static int tracer_alloc_buffers(void)
4559 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4562 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4563 goto out_free_buffer_mask;
4565 /* To save memory, keep the ring buffer size to its minimum */
4566 if (ring_buffer_expanded)
4567 ring_buf_size = trace_buf_size;
4571 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4572 cpumask_copy(tracing_cpumask, cpu_all_mask);
4574 /* TODO: make the number of buffers hot pluggable with CPUS */
4575 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4576 TRACE_BUFFER_FLAGS);
4577 if (!global_trace.buffer) {
4578 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4580 goto out_free_cpumask;
4582 global_trace.entries = ring_buffer_size(global_trace.buffer);
4585 #ifdef CONFIG_TRACER_MAX_TRACE
4586 max_tr.buffer = ring_buffer_alloc(1, TRACE_BUFFER_FLAGS);
4587 if (!max_tr.buffer) {
4588 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4590 ring_buffer_free(global_trace.buffer);
4591 goto out_free_cpumask;
4596 /* Allocate the first page for all buffers */
4597 for_each_tracing_cpu(i) {
4598 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4599 max_tr.data[i] = &per_cpu(max_tr_data, i);
4602 trace_init_cmdlines();
4604 register_tracer(&nop_trace);
4605 current_trace = &nop_trace;
4606 /* All seems OK, enable tracing */
4607 tracing_disabled = 0;
4609 atomic_notifier_chain_register(&panic_notifier_list,
4610 &trace_panic_notifier);
4612 register_die_notifier(&trace_die_notifier);
4617 free_cpumask_var(tracing_cpumask);
4618 out_free_buffer_mask:
4619 free_cpumask_var(tracing_buffer_mask);
4624 __init static int clear_boot_tracer(void)
4627 * The default tracer at boot buffer is an init section.
4628 * This function is called in lateinit. If we did not
4629 * find the boot tracer, then clear it out, to prevent
4630 * later registration from accessing the buffer that is
4631 * about to be freed.
4633 if (!default_bootup_tracer)
4636 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4637 default_bootup_tracer);
4638 default_bootup_tracer = NULL;
4643 early_initcall(tracer_alloc_buffers);
4644 fs_initcall(tracer_init_debugfs);
4645 late_initcall(clear_boot_tracer);