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"
45 * On boot up, the ring buffer is set to the minimum size, so that
46 * we do not waste memory on systems that are not using tracing.
48 int ring_buffer_expanded;
51 * We need to change this state when a selftest is running.
52 * A selftest will lurk into the ring-buffer to count the
53 * entries inserted during the selftest although some concurrent
54 * insertions into the ring-buffer such as trace_printk could occurred
55 * at the same time, giving false positive or negative results.
57 static bool __read_mostly tracing_selftest_running;
60 * If a tracer is running, we do not want to run SELFTEST.
62 bool __read_mostly tracing_selftest_disabled;
64 /* For tracers that don't implement custom flags */
65 static struct tracer_opt dummy_tracer_opt[] = {
69 static struct tracer_flags dummy_tracer_flags = {
71 .opts = dummy_tracer_opt
74 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
80 * Kill all tracing for good (never come back).
81 * It is initialized to 1 but will turn to zero if the initialization
82 * of the tracer is successful. But that is the only place that sets
85 static int tracing_disabled = 1;
87 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
89 static inline void ftrace_disable_cpu(void)
92 __this_cpu_inc(ftrace_cpu_disabled);
95 static inline void ftrace_enable_cpu(void)
97 __this_cpu_dec(ftrace_cpu_disabled);
101 cpumask_var_t __read_mostly tracing_buffer_mask;
104 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
106 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
107 * is set, then ftrace_dump is called. This will output the contents
108 * of the ftrace buffers to the console. This is very useful for
109 * capturing traces that lead to crashes and outputing it to a
112 * It is default off, but you can enable it with either specifying
113 * "ftrace_dump_on_oops" in the kernel command line, or setting
114 * /proc/sys/kernel/ftrace_dump_on_oops
115 * Set 1 if you want to dump buffers of all CPUs
116 * Set 2 if you want to dump the buffer of the CPU that triggered oops
119 enum ftrace_dump_mode ftrace_dump_on_oops;
121 static int tracing_set_tracer(const char *buf);
123 #define MAX_TRACER_SIZE 100
124 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
125 static char *default_bootup_tracer;
127 static int __init set_cmdline_ftrace(char *str)
129 strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
130 default_bootup_tracer = bootup_tracer_buf;
131 /* We are using ftrace early, expand it */
132 ring_buffer_expanded = 1;
135 __setup("ftrace=", set_cmdline_ftrace);
137 static int __init set_ftrace_dump_on_oops(char *str)
139 if (*str++ != '=' || !*str) {
140 ftrace_dump_on_oops = DUMP_ALL;
144 if (!strcmp("orig_cpu", str)) {
145 ftrace_dump_on_oops = DUMP_ORIG;
151 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
153 unsigned long long ns2usecs(cycle_t nsec)
161 * The global_trace is the descriptor that holds the tracing
162 * buffers for the live tracing. For each CPU, it contains
163 * a link list of pages that will store trace entries. The
164 * page descriptor of the pages in the memory is used to hold
165 * the link list by linking the lru item in the page descriptor
166 * to each of the pages in the buffer per CPU.
168 * For each active CPU there is a data field that holds the
169 * pages for the buffer for that CPU. Each CPU has the same number
170 * of pages allocated for its buffer.
172 static struct trace_array global_trace;
174 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
176 int filter_current_check_discard(struct ring_buffer *buffer,
177 struct ftrace_event_call *call, void *rec,
178 struct ring_buffer_event *event)
180 return filter_check_discard(call, rec, buffer, event);
182 EXPORT_SYMBOL_GPL(filter_current_check_discard);
184 cycle_t ftrace_now(int cpu)
188 /* Early boot up does not have a buffer yet */
189 if (!global_trace.buffer)
190 return trace_clock_local();
192 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
193 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
199 * The max_tr is used to snapshot the global_trace when a maximum
200 * latency is reached. Some tracers will use this to store a maximum
201 * trace while it continues examining live traces.
203 * The buffers for the max_tr are set up the same as the global_trace.
204 * When a snapshot is taken, the link list of the max_tr is swapped
205 * with the link list of the global_trace and the buffers are reset for
206 * the global_trace so the tracing can continue.
208 static struct trace_array max_tr;
210 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
212 /* tracer_enabled is used to toggle activation of a tracer */
213 static int tracer_enabled = 1;
216 * tracing_is_enabled - return tracer_enabled status
218 * This function is used by other tracers to know the status
219 * of the tracer_enabled flag. Tracers may use this function
220 * to know if it should enable their features when starting
221 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
223 int tracing_is_enabled(void)
225 return tracer_enabled;
229 * trace_buf_size is the size in bytes that is allocated
230 * for a buffer. Note, the number of bytes is always rounded
233 * This number is purposely set to a low number of 16384.
234 * If the dump on oops happens, it will be much appreciated
235 * to not have to wait for all that output. Anyway this can be
236 * boot time and run time configurable.
238 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
240 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
242 /* trace_types holds a link list of available tracers. */
243 static struct tracer *trace_types __read_mostly;
245 /* current_trace points to the tracer that is currently active */
246 static struct tracer *current_trace __read_mostly;
249 * trace_types_lock is used to protect the trace_types list.
251 static DEFINE_MUTEX(trace_types_lock);
254 * serialize the access of the ring buffer
256 * ring buffer serializes readers, but it is low level protection.
257 * The validity of the events (which returns by ring_buffer_peek() ..etc)
258 * are not protected by ring buffer.
260 * The content of events may become garbage if we allow other process consumes
261 * these events concurrently:
262 * A) the page of the consumed events may become a normal page
263 * (not reader page) in ring buffer, and this page will be rewrited
264 * by events producer.
265 * B) The page of the consumed events may become a page for splice_read,
266 * and this page will be returned to system.
268 * These primitives allow multi process access to different cpu ring buffer
271 * These primitives don't distinguish read-only and read-consume access.
272 * Multi read-only access are also serialized.
276 static DECLARE_RWSEM(all_cpu_access_lock);
277 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
279 static inline void trace_access_lock(int cpu)
281 if (cpu == TRACE_PIPE_ALL_CPU) {
282 /* gain it for accessing the whole ring buffer. */
283 down_write(&all_cpu_access_lock);
285 /* gain it for accessing a cpu ring buffer. */
287 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
288 down_read(&all_cpu_access_lock);
290 /* Secondly block other access to this @cpu ring buffer. */
291 mutex_lock(&per_cpu(cpu_access_lock, cpu));
295 static inline void trace_access_unlock(int cpu)
297 if (cpu == TRACE_PIPE_ALL_CPU) {
298 up_write(&all_cpu_access_lock);
300 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
301 up_read(&all_cpu_access_lock);
305 static inline void trace_access_lock_init(void)
309 for_each_possible_cpu(cpu)
310 mutex_init(&per_cpu(cpu_access_lock, cpu));
315 static DEFINE_MUTEX(access_lock);
317 static inline void trace_access_lock(int cpu)
320 mutex_lock(&access_lock);
323 static inline void trace_access_unlock(int cpu)
326 mutex_unlock(&access_lock);
329 static inline void trace_access_lock_init(void)
335 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
336 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
338 /* trace_flags holds trace_options default values */
339 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
340 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
341 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE;
343 static int trace_stop_count;
344 static DEFINE_SPINLOCK(tracing_start_lock);
347 * trace_wake_up - wake up tasks waiting for trace input
349 * Simply wakes up any task that is blocked on the trace_wait
350 * queue. These is used with trace_poll for tasks polling the trace.
352 void trace_wake_up(void)
356 if (trace_flags & TRACE_ITER_BLOCK)
359 * The runqueue_is_locked() can fail, but this is the best we
363 if (!runqueue_is_locked(cpu))
364 wake_up(&trace_wait);
368 static int __init set_buf_size(char *str)
370 unsigned long buf_size;
374 buf_size = memparse(str, &str);
375 /* nr_entries can not be zero */
378 trace_buf_size = buf_size;
381 __setup("trace_buf_size=", set_buf_size);
383 static int __init set_tracing_thresh(char *str)
385 unsigned long threshhold;
390 ret = strict_strtoul(str, 0, &threshhold);
393 tracing_thresh = threshhold * 1000;
396 __setup("tracing_thresh=", set_tracing_thresh);
398 unsigned long nsecs_to_usecs(unsigned long nsecs)
403 /* These must match the bit postions in trace_iterator_flags */
404 static const char *trace_options[] = {
434 { trace_clock_local, "local" },
435 { trace_clock_global, "global" },
441 * trace_parser_get_init - gets the buffer for trace parser
443 int trace_parser_get_init(struct trace_parser *parser, int size)
445 memset(parser, 0, sizeof(*parser));
447 parser->buffer = kmalloc(size, GFP_KERNEL);
456 * trace_parser_put - frees the buffer for trace parser
458 void trace_parser_put(struct trace_parser *parser)
460 kfree(parser->buffer);
464 * trace_get_user - reads the user input string separated by space
465 * (matched by isspace(ch))
467 * For each string found the 'struct trace_parser' is updated,
468 * and the function returns.
470 * Returns number of bytes read.
472 * See kernel/trace/trace.h for 'struct trace_parser' details.
474 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
475 size_t cnt, loff_t *ppos)
482 trace_parser_clear(parser);
484 ret = get_user(ch, ubuf++);
492 * The parser is not finished with the last write,
493 * continue reading the user input without skipping spaces.
496 /* skip white space */
497 while (cnt && isspace(ch)) {
498 ret = get_user(ch, ubuf++);
505 /* only spaces were written */
515 /* read the non-space input */
516 while (cnt && !isspace(ch)) {
517 if (parser->idx < parser->size - 1)
518 parser->buffer[parser->idx++] = ch;
523 ret = get_user(ch, ubuf++);
530 /* We either got finished input or we have to wait for another call. */
532 parser->buffer[parser->idx] = 0;
533 parser->cont = false;
536 parser->buffer[parser->idx++] = ch;
546 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
554 if (s->len <= s->readpos)
557 len = s->len - s->readpos;
560 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
570 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
575 if (s->len <= s->readpos)
578 len = s->len - s->readpos;
581 ret = memcpy(buf, s->buffer + s->readpos, cnt);
590 * ftrace_max_lock is used to protect the swapping of buffers
591 * when taking a max snapshot. The buffers themselves are
592 * protected by per_cpu spinlocks. But the action of the swap
593 * needs its own lock.
595 * This is defined as a arch_spinlock_t in order to help
596 * with performance when lockdep debugging is enabled.
598 * It is also used in other places outside the update_max_tr
599 * so it needs to be defined outside of the
600 * CONFIG_TRACER_MAX_TRACE.
602 static arch_spinlock_t ftrace_max_lock =
603 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
605 unsigned long __read_mostly tracing_thresh;
607 #ifdef CONFIG_TRACER_MAX_TRACE
608 unsigned long __read_mostly tracing_max_latency;
611 * Copy the new maximum trace into the separate maximum-trace
612 * structure. (this way the maximum trace is permanently saved,
613 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
616 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
618 struct trace_array_cpu *data = tr->data[cpu];
619 struct trace_array_cpu *max_data;
622 max_tr.time_start = data->preempt_timestamp;
624 max_data = max_tr.data[cpu];
625 max_data->saved_latency = tracing_max_latency;
626 max_data->critical_start = data->critical_start;
627 max_data->critical_end = data->critical_end;
629 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
630 max_data->pid = tsk->pid;
631 max_data->uid = task_uid(tsk);
632 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
633 max_data->policy = tsk->policy;
634 max_data->rt_priority = tsk->rt_priority;
636 /* record this tasks comm */
637 tracing_record_cmdline(tsk);
641 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
643 * @tsk: the task with the latency
644 * @cpu: The cpu that initiated the trace.
646 * Flip the buffers between the @tr and the max_tr and record information
647 * about which task was the cause of this latency.
650 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
652 struct ring_buffer *buf = tr->buffer;
654 if (trace_stop_count)
657 WARN_ON_ONCE(!irqs_disabled());
658 if (!current_trace->use_max_tr) {
662 arch_spin_lock(&ftrace_max_lock);
664 tr->buffer = max_tr.buffer;
667 __update_max_tr(tr, tsk, cpu);
668 arch_spin_unlock(&ftrace_max_lock);
672 * update_max_tr_single - only copy one trace over, and reset the rest
674 * @tsk - task with the latency
675 * @cpu - the cpu of the buffer to copy.
677 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
680 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
684 if (trace_stop_count)
687 WARN_ON_ONCE(!irqs_disabled());
688 if (!current_trace->use_max_tr) {
693 arch_spin_lock(&ftrace_max_lock);
695 ftrace_disable_cpu();
697 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
701 * We failed to swap the buffer due to a commit taking
702 * place on this CPU. We fail to record, but we reset
703 * the max trace buffer (no one writes directly to it)
704 * and flag that it failed.
706 trace_array_printk(&max_tr, _THIS_IP_,
707 "Failed to swap buffers due to commit in progress\n");
712 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
714 __update_max_tr(tr, tsk, cpu);
715 arch_spin_unlock(&ftrace_max_lock);
717 #endif /* CONFIG_TRACER_MAX_TRACE */
720 * register_tracer - register a tracer with the ftrace system.
721 * @type - the plugin for the tracer
723 * Register a new plugin tracer.
725 int register_tracer(struct tracer *type)
726 __releases(kernel_lock)
727 __acquires(kernel_lock)
733 pr_info("Tracer must have a name\n");
737 if (strlen(type->name) >= MAX_TRACER_SIZE) {
738 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
742 mutex_lock(&trace_types_lock);
744 tracing_selftest_running = true;
746 for (t = trace_types; t; t = t->next) {
747 if (strcmp(type->name, t->name) == 0) {
749 pr_info("Tracer %s already registered\n",
757 type->set_flag = &dummy_set_flag;
759 type->flags = &dummy_tracer_flags;
761 if (!type->flags->opts)
762 type->flags->opts = dummy_tracer_opt;
763 if (!type->wait_pipe)
764 type->wait_pipe = default_wait_pipe;
767 #ifdef CONFIG_FTRACE_STARTUP_TEST
768 if (type->selftest && !tracing_selftest_disabled) {
769 struct tracer *saved_tracer = current_trace;
770 struct trace_array *tr = &global_trace;
773 * Run a selftest on this tracer.
774 * Here we reset the trace buffer, and set the current
775 * tracer to be this tracer. The tracer can then run some
776 * internal tracing to verify that everything is in order.
777 * If we fail, we do not register this tracer.
779 tracing_reset_online_cpus(tr);
781 current_trace = type;
782 /* the test is responsible for initializing and enabling */
783 pr_info("Testing tracer %s: ", type->name);
784 ret = type->selftest(type, tr);
785 /* the test is responsible for resetting too */
786 current_trace = saved_tracer;
788 printk(KERN_CONT "FAILED!\n");
791 /* Only reset on passing, to avoid touching corrupted buffers */
792 tracing_reset_online_cpus(tr);
794 printk(KERN_CONT "PASSED\n");
798 type->next = trace_types;
802 tracing_selftest_running = false;
803 mutex_unlock(&trace_types_lock);
805 if (ret || !default_bootup_tracer)
808 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
811 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
812 /* Do we want this tracer to start on bootup? */
813 tracing_set_tracer(type->name);
814 default_bootup_tracer = NULL;
815 /* disable other selftests, since this will break it. */
816 tracing_selftest_disabled = 1;
817 #ifdef CONFIG_FTRACE_STARTUP_TEST
818 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
826 void unregister_tracer(struct tracer *type)
830 mutex_lock(&trace_types_lock);
831 for (t = &trace_types; *t; t = &(*t)->next) {
835 pr_info("Tracer %s not registered\n", type->name);
841 if (type == current_trace && tracer_enabled) {
844 if (current_trace->stop)
845 current_trace->stop(&global_trace);
846 current_trace = &nop_trace;
849 mutex_unlock(&trace_types_lock);
852 static void __tracing_reset(struct ring_buffer *buffer, int cpu)
854 ftrace_disable_cpu();
855 ring_buffer_reset_cpu(buffer, cpu);
859 void tracing_reset(struct trace_array *tr, int cpu)
861 struct ring_buffer *buffer = tr->buffer;
863 ring_buffer_record_disable(buffer);
865 /* Make sure all commits have finished */
867 __tracing_reset(buffer, cpu);
869 ring_buffer_record_enable(buffer);
872 void tracing_reset_online_cpus(struct trace_array *tr)
874 struct ring_buffer *buffer = tr->buffer;
877 ring_buffer_record_disable(buffer);
879 /* Make sure all commits have finished */
882 tr->time_start = ftrace_now(tr->cpu);
884 for_each_online_cpu(cpu)
885 __tracing_reset(buffer, cpu);
887 ring_buffer_record_enable(buffer);
890 void tracing_reset_current(int cpu)
892 tracing_reset(&global_trace, cpu);
895 void tracing_reset_current_online_cpus(void)
897 tracing_reset_online_cpus(&global_trace);
900 #define SAVED_CMDLINES 128
901 #define NO_CMDLINE_MAP UINT_MAX
902 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
903 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
904 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
905 static int cmdline_idx;
906 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
908 /* temporary disable recording */
909 static atomic_t trace_record_cmdline_disabled __read_mostly;
911 static void trace_init_cmdlines(void)
913 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
914 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
918 int is_tracing_stopped(void)
920 return trace_stop_count;
924 * ftrace_off_permanent - disable all ftrace code permanently
926 * This should only be called when a serious anomally has
927 * been detected. This will turn off the function tracing,
928 * ring buffers, and other tracing utilites. It takes no
929 * locks and can be called from any context.
931 void ftrace_off_permanent(void)
933 tracing_disabled = 1;
935 tracing_off_permanent();
939 * tracing_start - quick start of the tracer
941 * If tracing is enabled but was stopped by tracing_stop,
942 * this will start the tracer back up.
944 void tracing_start(void)
946 struct ring_buffer *buffer;
949 if (tracing_disabled)
952 spin_lock_irqsave(&tracing_start_lock, flags);
953 if (--trace_stop_count) {
954 if (trace_stop_count < 0) {
955 /* Someone screwed up their debugging */
957 trace_stop_count = 0;
962 /* Prevent the buffers from switching */
963 arch_spin_lock(&ftrace_max_lock);
965 buffer = global_trace.buffer;
967 ring_buffer_record_enable(buffer);
969 buffer = max_tr.buffer;
971 ring_buffer_record_enable(buffer);
973 arch_spin_unlock(&ftrace_max_lock);
977 spin_unlock_irqrestore(&tracing_start_lock, flags);
981 * tracing_stop - quick stop of the tracer
983 * Light weight way to stop tracing. Use in conjunction with
986 void tracing_stop(void)
988 struct ring_buffer *buffer;
992 spin_lock_irqsave(&tracing_start_lock, flags);
993 if (trace_stop_count++)
996 /* Prevent the buffers from switching */
997 arch_spin_lock(&ftrace_max_lock);
999 buffer = global_trace.buffer;
1001 ring_buffer_record_disable(buffer);
1003 buffer = max_tr.buffer;
1005 ring_buffer_record_disable(buffer);
1007 arch_spin_unlock(&ftrace_max_lock);
1010 spin_unlock_irqrestore(&tracing_start_lock, flags);
1013 void trace_stop_cmdline_recording(void);
1015 static void trace_save_cmdline(struct task_struct *tsk)
1019 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1023 * It's not the end of the world if we don't get
1024 * the lock, but we also don't want to spin
1025 * nor do we want to disable interrupts,
1026 * so if we miss here, then better luck next time.
1028 if (!arch_spin_trylock(&trace_cmdline_lock))
1031 idx = map_pid_to_cmdline[tsk->pid];
1032 if (idx == NO_CMDLINE_MAP) {
1033 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1036 * Check whether the cmdline buffer at idx has a pid
1037 * mapped. We are going to overwrite that entry so we
1038 * need to clear the map_pid_to_cmdline. Otherwise we
1039 * would read the new comm for the old pid.
1041 pid = map_cmdline_to_pid[idx];
1042 if (pid != NO_CMDLINE_MAP)
1043 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1045 map_cmdline_to_pid[idx] = tsk->pid;
1046 map_pid_to_cmdline[tsk->pid] = idx;
1051 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1053 arch_spin_unlock(&trace_cmdline_lock);
1056 void trace_find_cmdline(int pid, char comm[])
1061 strcpy(comm, "<idle>");
1065 if (WARN_ON_ONCE(pid < 0)) {
1066 strcpy(comm, "<XXX>");
1070 if (pid > PID_MAX_DEFAULT) {
1071 strcpy(comm, "<...>");
1076 arch_spin_lock(&trace_cmdline_lock);
1077 map = map_pid_to_cmdline[pid];
1078 if (map != NO_CMDLINE_MAP)
1079 strcpy(comm, saved_cmdlines[map]);
1081 strcpy(comm, "<...>");
1083 arch_spin_unlock(&trace_cmdline_lock);
1087 void tracing_record_cmdline(struct task_struct *tsk)
1089 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1093 trace_save_cmdline(tsk);
1097 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1100 struct task_struct *tsk = current;
1102 entry->preempt_count = pc & 0xff;
1103 entry->pid = (tsk) ? tsk->pid : 0;
1105 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1106 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1108 TRACE_FLAG_IRQS_NOSUPPORT |
1110 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1111 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1112 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1114 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1116 struct ring_buffer_event *
1117 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1120 unsigned long flags, int pc)
1122 struct ring_buffer_event *event;
1124 event = ring_buffer_lock_reserve(buffer, len);
1125 if (event != NULL) {
1126 struct trace_entry *ent = ring_buffer_event_data(event);
1128 tracing_generic_entry_update(ent, flags, pc);
1136 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1137 struct ring_buffer_event *event,
1138 unsigned long flags, int pc,
1141 ring_buffer_unlock_commit(buffer, event);
1143 ftrace_trace_stack(buffer, flags, 6, pc);
1144 ftrace_trace_userstack(buffer, flags, pc);
1150 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1151 struct ring_buffer_event *event,
1152 unsigned long flags, int pc)
1154 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1157 struct ring_buffer_event *
1158 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1159 int type, unsigned long len,
1160 unsigned long flags, int pc)
1162 *current_rb = global_trace.buffer;
1163 return trace_buffer_lock_reserve(*current_rb,
1164 type, len, flags, pc);
1166 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1168 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1169 struct ring_buffer_event *event,
1170 unsigned long flags, int pc)
1172 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1174 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1176 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1177 struct ring_buffer_event *event,
1178 unsigned long flags, int pc)
1180 __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1182 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1184 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1185 struct ring_buffer_event *event)
1187 ring_buffer_discard_commit(buffer, event);
1189 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1192 trace_function(struct trace_array *tr,
1193 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1196 struct ftrace_event_call *call = &event_function;
1197 struct ring_buffer *buffer = tr->buffer;
1198 struct ring_buffer_event *event;
1199 struct ftrace_entry *entry;
1201 /* If we are reading the ring buffer, don't trace */
1202 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1205 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1209 entry = ring_buffer_event_data(event);
1211 entry->parent_ip = parent_ip;
1213 if (!filter_check_discard(call, entry, buffer, event))
1214 ring_buffer_unlock_commit(buffer, event);
1218 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1219 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1222 if (likely(!atomic_read(&data->disabled)))
1223 trace_function(tr, ip, parent_ip, flags, pc);
1226 #ifdef CONFIG_STACKTRACE
1227 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1228 unsigned long flags,
1231 struct ftrace_event_call *call = &event_kernel_stack;
1232 struct ring_buffer_event *event;
1233 struct stack_entry *entry;
1234 struct stack_trace trace;
1236 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1237 sizeof(*entry), flags, pc);
1240 entry = ring_buffer_event_data(event);
1241 memset(&entry->caller, 0, sizeof(entry->caller));
1243 trace.nr_entries = 0;
1244 trace.max_entries = FTRACE_STACK_ENTRIES;
1246 trace.entries = entry->caller;
1248 save_stack_trace(&trace);
1249 if (!filter_check_discard(call, entry, buffer, event))
1250 ring_buffer_unlock_commit(buffer, event);
1253 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1256 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1259 __ftrace_trace_stack(buffer, flags, skip, pc);
1262 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1265 __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1269 * trace_dump_stack - record a stack back trace in the trace buffer
1271 void trace_dump_stack(void)
1273 unsigned long flags;
1275 if (tracing_disabled || tracing_selftest_running)
1278 local_save_flags(flags);
1280 /* skipping 3 traces, seems to get us at the caller of this function */
1281 __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count());
1284 static DEFINE_PER_CPU(int, user_stack_count);
1287 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1289 struct ftrace_event_call *call = &event_user_stack;
1290 struct ring_buffer_event *event;
1291 struct userstack_entry *entry;
1292 struct stack_trace trace;
1294 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1298 * NMIs can not handle page faults, even with fix ups.
1299 * The save user stack can (and often does) fault.
1301 if (unlikely(in_nmi()))
1305 * prevent recursion, since the user stack tracing may
1306 * trigger other kernel events.
1309 if (__this_cpu_read(user_stack_count))
1312 __this_cpu_inc(user_stack_count);
1314 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1315 sizeof(*entry), flags, pc);
1317 goto out_drop_count;
1318 entry = ring_buffer_event_data(event);
1320 entry->tgid = current->tgid;
1321 memset(&entry->caller, 0, sizeof(entry->caller));
1323 trace.nr_entries = 0;
1324 trace.max_entries = FTRACE_STACK_ENTRIES;
1326 trace.entries = entry->caller;
1328 save_stack_trace_user(&trace);
1329 if (!filter_check_discard(call, entry, buffer, event))
1330 ring_buffer_unlock_commit(buffer, event);
1333 __this_cpu_dec(user_stack_count);
1339 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1341 ftrace_trace_userstack(tr, flags, preempt_count());
1345 #endif /* CONFIG_STACKTRACE */
1348 * trace_vbprintk - write binary msg to tracing buffer
1351 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1353 static arch_spinlock_t trace_buf_lock =
1354 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1355 static u32 trace_buf[TRACE_BUF_SIZE];
1357 struct ftrace_event_call *call = &event_bprint;
1358 struct ring_buffer_event *event;
1359 struct ring_buffer *buffer;
1360 struct trace_array *tr = &global_trace;
1361 struct trace_array_cpu *data;
1362 struct bprint_entry *entry;
1363 unsigned long flags;
1365 int cpu, len = 0, size, pc;
1367 if (unlikely(tracing_selftest_running || tracing_disabled))
1370 /* Don't pollute graph traces with trace_vprintk internals */
1371 pause_graph_tracing();
1373 pc = preempt_count();
1374 preempt_disable_notrace();
1375 cpu = raw_smp_processor_id();
1376 data = tr->data[cpu];
1378 disable = atomic_inc_return(&data->disabled);
1379 if (unlikely(disable != 1))
1382 /* Lockdep uses trace_printk for lock tracing */
1383 local_irq_save(flags);
1384 arch_spin_lock(&trace_buf_lock);
1385 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1387 if (len > TRACE_BUF_SIZE || len < 0)
1390 size = sizeof(*entry) + sizeof(u32) * len;
1391 buffer = tr->buffer;
1392 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1396 entry = ring_buffer_event_data(event);
1400 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1401 if (!filter_check_discard(call, entry, buffer, event)) {
1402 ring_buffer_unlock_commit(buffer, event);
1403 ftrace_trace_stack(buffer, flags, 6, pc);
1407 arch_spin_unlock(&trace_buf_lock);
1408 local_irq_restore(flags);
1411 atomic_dec_return(&data->disabled);
1412 preempt_enable_notrace();
1413 unpause_graph_tracing();
1417 EXPORT_SYMBOL_GPL(trace_vbprintk);
1419 int trace_array_printk(struct trace_array *tr,
1420 unsigned long ip, const char *fmt, ...)
1425 if (!(trace_flags & TRACE_ITER_PRINTK))
1429 ret = trace_array_vprintk(tr, ip, fmt, ap);
1434 int trace_array_vprintk(struct trace_array *tr,
1435 unsigned long ip, const char *fmt, va_list args)
1437 static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1438 static char trace_buf[TRACE_BUF_SIZE];
1440 struct ftrace_event_call *call = &event_print;
1441 struct ring_buffer_event *event;
1442 struct ring_buffer *buffer;
1443 struct trace_array_cpu *data;
1444 int cpu, len = 0, size, pc;
1445 struct print_entry *entry;
1446 unsigned long irq_flags;
1449 if (tracing_disabled || tracing_selftest_running)
1452 pc = preempt_count();
1453 preempt_disable_notrace();
1454 cpu = raw_smp_processor_id();
1455 data = tr->data[cpu];
1457 disable = atomic_inc_return(&data->disabled);
1458 if (unlikely(disable != 1))
1461 pause_graph_tracing();
1462 raw_local_irq_save(irq_flags);
1463 arch_spin_lock(&trace_buf_lock);
1464 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1466 size = sizeof(*entry) + len + 1;
1467 buffer = tr->buffer;
1468 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1472 entry = ring_buffer_event_data(event);
1475 memcpy(&entry->buf, trace_buf, len);
1476 entry->buf[len] = '\0';
1477 if (!filter_check_discard(call, entry, buffer, event)) {
1478 ring_buffer_unlock_commit(buffer, event);
1479 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1483 arch_spin_unlock(&trace_buf_lock);
1484 raw_local_irq_restore(irq_flags);
1485 unpause_graph_tracing();
1487 atomic_dec_return(&data->disabled);
1488 preempt_enable_notrace();
1493 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1495 return trace_array_vprintk(&global_trace, ip, fmt, args);
1497 EXPORT_SYMBOL_GPL(trace_vprintk);
1499 static void trace_iterator_increment(struct trace_iterator *iter)
1501 /* Don't allow ftrace to trace into the ring buffers */
1502 ftrace_disable_cpu();
1505 if (iter->buffer_iter[iter->cpu])
1506 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1508 ftrace_enable_cpu();
1511 static struct trace_entry *
1512 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1513 unsigned long *lost_events)
1515 struct ring_buffer_event *event;
1516 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1518 /* Don't allow ftrace to trace into the ring buffers */
1519 ftrace_disable_cpu();
1522 event = ring_buffer_iter_peek(buf_iter, ts);
1524 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1527 ftrace_enable_cpu();
1529 return event ? ring_buffer_event_data(event) : NULL;
1532 static struct trace_entry *
1533 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1534 unsigned long *missing_events, u64 *ent_ts)
1536 struct ring_buffer *buffer = iter->tr->buffer;
1537 struct trace_entry *ent, *next = NULL;
1538 unsigned long lost_events = 0, next_lost = 0;
1539 int cpu_file = iter->cpu_file;
1540 u64 next_ts = 0, ts;
1545 * If we are in a per_cpu trace file, don't bother by iterating over
1546 * all cpu and peek directly.
1548 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1549 if (ring_buffer_empty_cpu(buffer, cpu_file))
1551 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1553 *ent_cpu = cpu_file;
1558 for_each_tracing_cpu(cpu) {
1560 if (ring_buffer_empty_cpu(buffer, cpu))
1563 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1566 * Pick the entry with the smallest timestamp:
1568 if (ent && (!next || ts < next_ts)) {
1572 next_lost = lost_events;
1577 *ent_cpu = next_cpu;
1583 *missing_events = next_lost;
1588 /* Find the next real entry, without updating the iterator itself */
1589 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1590 int *ent_cpu, u64 *ent_ts)
1592 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1595 /* Find the next real entry, and increment the iterator to the next entry */
1596 void *trace_find_next_entry_inc(struct trace_iterator *iter)
1598 iter->ent = __find_next_entry(iter, &iter->cpu,
1599 &iter->lost_events, &iter->ts);
1602 trace_iterator_increment(iter);
1604 return iter->ent ? iter : NULL;
1607 static void trace_consume(struct trace_iterator *iter)
1609 /* Don't allow ftrace to trace into the ring buffers */
1610 ftrace_disable_cpu();
1611 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1612 &iter->lost_events);
1613 ftrace_enable_cpu();
1616 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1618 struct trace_iterator *iter = m->private;
1622 WARN_ON_ONCE(iter->leftover);
1626 /* can't go backwards */
1631 ent = trace_find_next_entry_inc(iter);
1635 while (ent && iter->idx < i)
1636 ent = trace_find_next_entry_inc(iter);
1643 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1645 struct trace_array *tr = iter->tr;
1646 struct ring_buffer_event *event;
1647 struct ring_buffer_iter *buf_iter;
1648 unsigned long entries = 0;
1651 tr->data[cpu]->skipped_entries = 0;
1653 if (!iter->buffer_iter[cpu])
1656 buf_iter = iter->buffer_iter[cpu];
1657 ring_buffer_iter_reset(buf_iter);
1660 * We could have the case with the max latency tracers
1661 * that a reset never took place on a cpu. This is evident
1662 * by the timestamp being before the start of the buffer.
1664 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1665 if (ts >= iter->tr->time_start)
1668 ring_buffer_read(buf_iter, NULL);
1671 tr->data[cpu]->skipped_entries = entries;
1675 * The current tracer is copied to avoid a global locking
1678 static void *s_start(struct seq_file *m, loff_t *pos)
1680 struct trace_iterator *iter = m->private;
1681 static struct tracer *old_tracer;
1682 int cpu_file = iter->cpu_file;
1687 /* copy the tracer to avoid using a global lock all around */
1688 mutex_lock(&trace_types_lock);
1689 if (unlikely(old_tracer != current_trace && current_trace)) {
1690 old_tracer = current_trace;
1691 *iter->trace = *current_trace;
1693 mutex_unlock(&trace_types_lock);
1695 atomic_inc(&trace_record_cmdline_disabled);
1697 if (*pos != iter->pos) {
1702 ftrace_disable_cpu();
1704 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1705 for_each_tracing_cpu(cpu)
1706 tracing_iter_reset(iter, cpu);
1708 tracing_iter_reset(iter, cpu_file);
1710 ftrace_enable_cpu();
1713 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1718 * If we overflowed the seq_file before, then we want
1719 * to just reuse the trace_seq buffer again.
1725 p = s_next(m, p, &l);
1729 trace_event_read_lock();
1730 trace_access_lock(cpu_file);
1734 static void s_stop(struct seq_file *m, void *p)
1736 struct trace_iterator *iter = m->private;
1738 atomic_dec(&trace_record_cmdline_disabled);
1739 trace_access_unlock(iter->cpu_file);
1740 trace_event_read_unlock();
1743 static void print_lat_help_header(struct seq_file *m)
1745 seq_puts(m, "# _------=> CPU# \n");
1746 seq_puts(m, "# / _-----=> irqs-off \n");
1747 seq_puts(m, "# | / _----=> need-resched \n");
1748 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1749 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1750 seq_puts(m, "# |||| / delay \n");
1751 seq_puts(m, "# cmd pid ||||| time | caller \n");
1752 seq_puts(m, "# \\ / ||||| \\ | / \n");
1755 static void print_func_help_header(struct seq_file *m)
1757 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1758 seq_puts(m, "# | | | | |\n");
1763 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1765 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1766 struct trace_array *tr = iter->tr;
1767 struct trace_array_cpu *data = tr->data[tr->cpu];
1768 struct tracer *type = current_trace;
1769 unsigned long entries = 0;
1770 unsigned long total = 0;
1771 unsigned long count;
1772 const char *name = "preemption";
1779 for_each_tracing_cpu(cpu) {
1780 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1782 * If this buffer has skipped entries, then we hold all
1783 * entries for the trace and we need to ignore the
1784 * ones before the time stamp.
1786 if (tr->data[cpu]->skipped_entries) {
1787 count -= tr->data[cpu]->skipped_entries;
1788 /* total is the same as the entries */
1792 ring_buffer_overrun_cpu(tr->buffer, cpu);
1796 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1798 seq_puts(m, "# -----------------------------------"
1799 "---------------------------------\n");
1800 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1801 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1802 nsecs_to_usecs(data->saved_latency),
1806 #if defined(CONFIG_PREEMPT_NONE)
1808 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1810 #elif defined(CONFIG_PREEMPT)
1815 /* These are reserved for later use */
1818 seq_printf(m, " #P:%d)\n", num_online_cpus());
1822 seq_puts(m, "# -----------------\n");
1823 seq_printf(m, "# | task: %.16s-%d "
1824 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1825 data->comm, data->pid, data->uid, data->nice,
1826 data->policy, data->rt_priority);
1827 seq_puts(m, "# -----------------\n");
1829 if (data->critical_start) {
1830 seq_puts(m, "# => started at: ");
1831 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1832 trace_print_seq(m, &iter->seq);
1833 seq_puts(m, "\n# => ended at: ");
1834 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1835 trace_print_seq(m, &iter->seq);
1836 seq_puts(m, "\n#\n");
1842 static void test_cpu_buff_start(struct trace_iterator *iter)
1844 struct trace_seq *s = &iter->seq;
1846 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1849 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1852 if (cpumask_test_cpu(iter->cpu, iter->started))
1855 if (iter->tr->data[iter->cpu]->skipped_entries)
1858 cpumask_set_cpu(iter->cpu, iter->started);
1860 /* Don't print started cpu buffer for the first entry of the trace */
1862 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1866 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1868 struct trace_seq *s = &iter->seq;
1869 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1870 struct trace_entry *entry;
1871 struct trace_event *event;
1875 test_cpu_buff_start(iter);
1877 event = ftrace_find_event(entry->type);
1879 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1880 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1881 if (!trace_print_lat_context(iter))
1884 if (!trace_print_context(iter))
1890 return event->funcs->trace(iter, sym_flags, event);
1892 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1895 return TRACE_TYPE_HANDLED;
1897 return TRACE_TYPE_PARTIAL_LINE;
1900 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1902 struct trace_seq *s = &iter->seq;
1903 struct trace_entry *entry;
1904 struct trace_event *event;
1908 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1909 if (!trace_seq_printf(s, "%d %d %llu ",
1910 entry->pid, iter->cpu, iter->ts))
1914 event = ftrace_find_event(entry->type);
1916 return event->funcs->raw(iter, 0, event);
1918 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1921 return TRACE_TYPE_HANDLED;
1923 return TRACE_TYPE_PARTIAL_LINE;
1926 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1928 struct trace_seq *s = &iter->seq;
1929 unsigned char newline = '\n';
1930 struct trace_entry *entry;
1931 struct trace_event *event;
1935 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1936 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1937 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1938 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1941 event = ftrace_find_event(entry->type);
1943 enum print_line_t ret = event->funcs->hex(iter, 0, event);
1944 if (ret != TRACE_TYPE_HANDLED)
1948 SEQ_PUT_FIELD_RET(s, newline);
1950 return TRACE_TYPE_HANDLED;
1953 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1955 struct trace_seq *s = &iter->seq;
1956 struct trace_entry *entry;
1957 struct trace_event *event;
1961 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1962 SEQ_PUT_FIELD_RET(s, entry->pid);
1963 SEQ_PUT_FIELD_RET(s, iter->cpu);
1964 SEQ_PUT_FIELD_RET(s, iter->ts);
1967 event = ftrace_find_event(entry->type);
1968 return event ? event->funcs->binary(iter, 0, event) :
1972 int trace_empty(struct trace_iterator *iter)
1976 /* If we are looking at one CPU buffer, only check that one */
1977 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1978 cpu = iter->cpu_file;
1979 if (iter->buffer_iter[cpu]) {
1980 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1983 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1989 for_each_tracing_cpu(cpu) {
1990 if (iter->buffer_iter[cpu]) {
1991 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1994 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2002 /* Called with trace_event_read_lock() held. */
2003 enum print_line_t print_trace_line(struct trace_iterator *iter)
2005 enum print_line_t ret;
2007 if (iter->lost_events)
2008 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2009 iter->cpu, iter->lost_events);
2011 if (iter->trace && iter->trace->print_line) {
2012 ret = iter->trace->print_line(iter);
2013 if (ret != TRACE_TYPE_UNHANDLED)
2017 if (iter->ent->type == TRACE_BPRINT &&
2018 trace_flags & TRACE_ITER_PRINTK &&
2019 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2020 return trace_print_bprintk_msg_only(iter);
2022 if (iter->ent->type == TRACE_PRINT &&
2023 trace_flags & TRACE_ITER_PRINTK &&
2024 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2025 return trace_print_printk_msg_only(iter);
2027 if (trace_flags & TRACE_ITER_BIN)
2028 return print_bin_fmt(iter);
2030 if (trace_flags & TRACE_ITER_HEX)
2031 return print_hex_fmt(iter);
2033 if (trace_flags & TRACE_ITER_RAW)
2034 return print_raw_fmt(iter);
2036 return print_trace_fmt(iter);
2039 void trace_default_header(struct seq_file *m)
2041 struct trace_iterator *iter = m->private;
2043 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2044 /* print nothing if the buffers are empty */
2045 if (trace_empty(iter))
2047 print_trace_header(m, iter);
2048 if (!(trace_flags & TRACE_ITER_VERBOSE))
2049 print_lat_help_header(m);
2051 if (!(trace_flags & TRACE_ITER_VERBOSE))
2052 print_func_help_header(m);
2056 static int s_show(struct seq_file *m, void *v)
2058 struct trace_iterator *iter = v;
2061 if (iter->ent == NULL) {
2063 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2066 if (iter->trace && iter->trace->print_header)
2067 iter->trace->print_header(m);
2069 trace_default_header(m);
2071 } else if (iter->leftover) {
2073 * If we filled the seq_file buffer earlier, we
2074 * want to just show it now.
2076 ret = trace_print_seq(m, &iter->seq);
2078 /* ret should this time be zero, but you never know */
2079 iter->leftover = ret;
2082 print_trace_line(iter);
2083 ret = trace_print_seq(m, &iter->seq);
2085 * If we overflow the seq_file buffer, then it will
2086 * ask us for this data again at start up.
2088 * ret is 0 if seq_file write succeeded.
2091 iter->leftover = ret;
2097 static const struct seq_operations tracer_seq_ops = {
2104 static struct trace_iterator *
2105 __tracing_open(struct inode *inode, struct file *file)
2107 long cpu_file = (long) inode->i_private;
2108 void *fail_ret = ERR_PTR(-ENOMEM);
2109 struct trace_iterator *iter;
2113 if (tracing_disabled)
2114 return ERR_PTR(-ENODEV);
2116 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2118 return ERR_PTR(-ENOMEM);
2121 * We make a copy of the current tracer to avoid concurrent
2122 * changes on it while we are reading.
2124 mutex_lock(&trace_types_lock);
2125 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2130 *iter->trace = *current_trace;
2132 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2135 if (current_trace && current_trace->print_max)
2138 iter->tr = &global_trace;
2140 mutex_init(&iter->mutex);
2141 iter->cpu_file = cpu_file;
2143 /* Notify the tracer early; before we stop tracing. */
2144 if (iter->trace && iter->trace->open)
2145 iter->trace->open(iter);
2147 /* Annotate start of buffers if we had overruns */
2148 if (ring_buffer_overruns(iter->tr->buffer))
2149 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2151 /* stop the trace while dumping */
2154 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2155 for_each_tracing_cpu(cpu) {
2156 iter->buffer_iter[cpu] =
2157 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2159 ring_buffer_read_prepare_sync();
2160 for_each_tracing_cpu(cpu) {
2161 ring_buffer_read_start(iter->buffer_iter[cpu]);
2162 tracing_iter_reset(iter, cpu);
2165 cpu = iter->cpu_file;
2166 iter->buffer_iter[cpu] =
2167 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2168 ring_buffer_read_prepare_sync();
2169 ring_buffer_read_start(iter->buffer_iter[cpu]);
2170 tracing_iter_reset(iter, cpu);
2173 ret = seq_open(file, &tracer_seq_ops);
2175 fail_ret = ERR_PTR(ret);
2179 m = file->private_data;
2182 mutex_unlock(&trace_types_lock);
2187 for_each_tracing_cpu(cpu) {
2188 if (iter->buffer_iter[cpu])
2189 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2191 free_cpumask_var(iter->started);
2194 mutex_unlock(&trace_types_lock);
2201 int tracing_open_generic(struct inode *inode, struct file *filp)
2203 if (tracing_disabled)
2206 filp->private_data = inode->i_private;
2210 static int tracing_release(struct inode *inode, struct file *file)
2212 struct seq_file *m = file->private_data;
2213 struct trace_iterator *iter;
2216 if (!(file->f_mode & FMODE_READ))
2221 mutex_lock(&trace_types_lock);
2222 for_each_tracing_cpu(cpu) {
2223 if (iter->buffer_iter[cpu])
2224 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2227 if (iter->trace && iter->trace->close)
2228 iter->trace->close(iter);
2230 /* reenable tracing if it was previously enabled */
2232 mutex_unlock(&trace_types_lock);
2234 seq_release(inode, file);
2235 mutex_destroy(&iter->mutex);
2236 free_cpumask_var(iter->started);
2242 static int tracing_open(struct inode *inode, struct file *file)
2244 struct trace_iterator *iter;
2247 /* If this file was open for write, then erase contents */
2248 if ((file->f_mode & FMODE_WRITE) &&
2249 (file->f_flags & O_TRUNC)) {
2250 long cpu = (long) inode->i_private;
2252 if (cpu == TRACE_PIPE_ALL_CPU)
2253 tracing_reset_online_cpus(&global_trace);
2255 tracing_reset(&global_trace, cpu);
2258 if (file->f_mode & FMODE_READ) {
2259 iter = __tracing_open(inode, file);
2261 ret = PTR_ERR(iter);
2262 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2263 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2269 t_next(struct seq_file *m, void *v, loff_t *pos)
2271 struct tracer *t = v;
2281 static void *t_start(struct seq_file *m, loff_t *pos)
2286 mutex_lock(&trace_types_lock);
2287 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2293 static void t_stop(struct seq_file *m, void *p)
2295 mutex_unlock(&trace_types_lock);
2298 static int t_show(struct seq_file *m, void *v)
2300 struct tracer *t = v;
2305 seq_printf(m, "%s", t->name);
2314 static const struct seq_operations show_traces_seq_ops = {
2321 static int show_traces_open(struct inode *inode, struct file *file)
2323 if (tracing_disabled)
2326 return seq_open(file, &show_traces_seq_ops);
2330 tracing_write_stub(struct file *filp, const char __user *ubuf,
2331 size_t count, loff_t *ppos)
2336 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
2338 if (file->f_mode & FMODE_READ)
2339 return seq_lseek(file, offset, origin);
2344 static const struct file_operations tracing_fops = {
2345 .open = tracing_open,
2347 .write = tracing_write_stub,
2348 .llseek = tracing_seek,
2349 .release = tracing_release,
2352 static const struct file_operations show_traces_fops = {
2353 .open = show_traces_open,
2355 .release = seq_release,
2356 .llseek = seq_lseek,
2360 * Only trace on a CPU if the bitmask is set:
2362 static cpumask_var_t tracing_cpumask;
2365 * The tracer itself will not take this lock, but still we want
2366 * to provide a consistent cpumask to user-space:
2368 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2371 * Temporary storage for the character representation of the
2372 * CPU bitmask (and one more byte for the newline):
2374 static char mask_str[NR_CPUS + 1];
2377 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2378 size_t count, loff_t *ppos)
2382 mutex_lock(&tracing_cpumask_update_lock);
2384 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2385 if (count - len < 2) {
2389 len += sprintf(mask_str + len, "\n");
2390 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2393 mutex_unlock(&tracing_cpumask_update_lock);
2399 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2400 size_t count, loff_t *ppos)
2403 cpumask_var_t tracing_cpumask_new;
2405 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2408 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2412 mutex_lock(&tracing_cpumask_update_lock);
2414 local_irq_disable();
2415 arch_spin_lock(&ftrace_max_lock);
2416 for_each_tracing_cpu(cpu) {
2418 * Increase/decrease the disabled counter if we are
2419 * about to flip a bit in the cpumask:
2421 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2422 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2423 atomic_inc(&global_trace.data[cpu]->disabled);
2425 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2426 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2427 atomic_dec(&global_trace.data[cpu]->disabled);
2430 arch_spin_unlock(&ftrace_max_lock);
2433 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2435 mutex_unlock(&tracing_cpumask_update_lock);
2436 free_cpumask_var(tracing_cpumask_new);
2441 free_cpumask_var(tracing_cpumask_new);
2446 static const struct file_operations tracing_cpumask_fops = {
2447 .open = tracing_open_generic,
2448 .read = tracing_cpumask_read,
2449 .write = tracing_cpumask_write,
2450 .llseek = generic_file_llseek,
2453 static int tracing_trace_options_show(struct seq_file *m, void *v)
2455 struct tracer_opt *trace_opts;
2459 mutex_lock(&trace_types_lock);
2460 tracer_flags = current_trace->flags->val;
2461 trace_opts = current_trace->flags->opts;
2463 for (i = 0; trace_options[i]; i++) {
2464 if (trace_flags & (1 << i))
2465 seq_printf(m, "%s\n", trace_options[i]);
2467 seq_printf(m, "no%s\n", trace_options[i]);
2470 for (i = 0; trace_opts[i].name; i++) {
2471 if (tracer_flags & trace_opts[i].bit)
2472 seq_printf(m, "%s\n", trace_opts[i].name);
2474 seq_printf(m, "no%s\n", trace_opts[i].name);
2476 mutex_unlock(&trace_types_lock);
2481 static int __set_tracer_option(struct tracer *trace,
2482 struct tracer_flags *tracer_flags,
2483 struct tracer_opt *opts, int neg)
2487 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2492 tracer_flags->val &= ~opts->bit;
2494 tracer_flags->val |= opts->bit;
2498 /* Try to assign a tracer specific option */
2499 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2501 struct tracer_flags *tracer_flags = trace->flags;
2502 struct tracer_opt *opts = NULL;
2505 for (i = 0; tracer_flags->opts[i].name; i++) {
2506 opts = &tracer_flags->opts[i];
2508 if (strcmp(cmp, opts->name) == 0)
2509 return __set_tracer_option(trace, trace->flags,
2516 static void set_tracer_flags(unsigned int mask, int enabled)
2518 /* do nothing if flag is already set */
2519 if (!!(trace_flags & mask) == !!enabled)
2523 trace_flags |= mask;
2525 trace_flags &= ~mask;
2527 if (mask == TRACE_ITER_RECORD_CMD)
2528 trace_event_enable_cmd_record(enabled);
2530 if (mask == TRACE_ITER_OVERWRITE)
2531 ring_buffer_change_overwrite(global_trace.buffer, enabled);
2535 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2536 size_t cnt, loff_t *ppos)
2544 if (cnt >= sizeof(buf))
2547 if (copy_from_user(&buf, ubuf, cnt))
2551 cmp = strstrip(buf);
2553 if (strncmp(cmp, "no", 2) == 0) {
2558 for (i = 0; trace_options[i]; i++) {
2559 if (strcmp(cmp, trace_options[i]) == 0) {
2560 set_tracer_flags(1 << i, !neg);
2565 /* If no option could be set, test the specific tracer options */
2566 if (!trace_options[i]) {
2567 mutex_lock(&trace_types_lock);
2568 ret = set_tracer_option(current_trace, cmp, neg);
2569 mutex_unlock(&trace_types_lock);
2579 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2581 if (tracing_disabled)
2583 return single_open(file, tracing_trace_options_show, NULL);
2586 static const struct file_operations tracing_iter_fops = {
2587 .open = tracing_trace_options_open,
2589 .llseek = seq_lseek,
2590 .release = single_release,
2591 .write = tracing_trace_options_write,
2594 static const char readme_msg[] =
2595 "tracing mini-HOWTO:\n\n"
2596 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2597 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2598 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2599 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2601 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2602 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2604 "# cat /sys/kernel/debug/tracing/trace_options\n"
2605 "noprint-parent nosym-offset nosym-addr noverbose\n"
2606 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2607 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2608 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2609 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2613 tracing_readme_read(struct file *filp, char __user *ubuf,
2614 size_t cnt, loff_t *ppos)
2616 return simple_read_from_buffer(ubuf, cnt, ppos,
2617 readme_msg, strlen(readme_msg));
2620 static const struct file_operations tracing_readme_fops = {
2621 .open = tracing_open_generic,
2622 .read = tracing_readme_read,
2623 .llseek = generic_file_llseek,
2627 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2628 size_t cnt, loff_t *ppos)
2637 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2641 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2649 for (i = 0; i < SAVED_CMDLINES; i++) {
2652 pid = map_cmdline_to_pid[i];
2653 if (pid == -1 || pid == NO_CMDLINE_MAP)
2656 trace_find_cmdline(pid, buf_comm);
2657 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2662 len = simple_read_from_buffer(ubuf, cnt, ppos,
2671 static const struct file_operations tracing_saved_cmdlines_fops = {
2672 .open = tracing_open_generic,
2673 .read = tracing_saved_cmdlines_read,
2674 .llseek = generic_file_llseek,
2678 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2679 size_t cnt, loff_t *ppos)
2684 r = sprintf(buf, "%u\n", tracer_enabled);
2685 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2689 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2690 size_t cnt, loff_t *ppos)
2692 struct trace_array *tr = filp->private_data;
2697 if (cnt >= sizeof(buf))
2700 if (copy_from_user(&buf, ubuf, cnt))
2705 ret = strict_strtoul(buf, 10, &val);
2711 mutex_lock(&trace_types_lock);
2712 if (tracer_enabled ^ val) {
2714 /* Only need to warn if this is used to change the state */
2715 WARN_ONCE(1, "tracing_enabled is deprecated. Use tracing_on");
2719 if (current_trace->start)
2720 current_trace->start(tr);
2725 if (current_trace->stop)
2726 current_trace->stop(tr);
2729 mutex_unlock(&trace_types_lock);
2737 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2738 size_t cnt, loff_t *ppos)
2740 char buf[MAX_TRACER_SIZE+2];
2743 mutex_lock(&trace_types_lock);
2745 r = sprintf(buf, "%s\n", current_trace->name);
2747 r = sprintf(buf, "\n");
2748 mutex_unlock(&trace_types_lock);
2750 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2753 int tracer_init(struct tracer *t, struct trace_array *tr)
2755 tracing_reset_online_cpus(tr);
2759 static int tracing_resize_ring_buffer(unsigned long size)
2764 * If kernel or user changes the size of the ring buffer
2765 * we use the size that was given, and we can forget about
2766 * expanding it later.
2768 ring_buffer_expanded = 1;
2770 ret = ring_buffer_resize(global_trace.buffer, size);
2774 if (!current_trace->use_max_tr)
2777 ret = ring_buffer_resize(max_tr.buffer, size);
2781 r = ring_buffer_resize(global_trace.buffer,
2782 global_trace.entries);
2785 * AARGH! We are left with different
2786 * size max buffer!!!!
2787 * The max buffer is our "snapshot" buffer.
2788 * When a tracer needs a snapshot (one of the
2789 * latency tracers), it swaps the max buffer
2790 * with the saved snap shot. We succeeded to
2791 * update the size of the main buffer, but failed to
2792 * update the size of the max buffer. But when we tried
2793 * to reset the main buffer to the original size, we
2794 * failed there too. This is very unlikely to
2795 * happen, but if it does, warn and kill all
2799 tracing_disabled = 1;
2804 max_tr.entries = size;
2806 global_trace.entries = size;
2813 * tracing_update_buffers - used by tracing facility to expand ring buffers
2815 * To save on memory when the tracing is never used on a system with it
2816 * configured in. The ring buffers are set to a minimum size. But once
2817 * a user starts to use the tracing facility, then they need to grow
2818 * to their default size.
2820 * This function is to be called when a tracer is about to be used.
2822 int tracing_update_buffers(void)
2826 mutex_lock(&trace_types_lock);
2827 if (!ring_buffer_expanded)
2828 ret = tracing_resize_ring_buffer(trace_buf_size);
2829 mutex_unlock(&trace_types_lock);
2834 struct trace_option_dentry;
2836 static struct trace_option_dentry *
2837 create_trace_option_files(struct tracer *tracer);
2840 destroy_trace_option_files(struct trace_option_dentry *topts);
2842 static int tracing_set_tracer(const char *buf)
2844 static struct trace_option_dentry *topts;
2845 struct trace_array *tr = &global_trace;
2849 mutex_lock(&trace_types_lock);
2851 if (!ring_buffer_expanded) {
2852 ret = tracing_resize_ring_buffer(trace_buf_size);
2858 for (t = trace_types; t; t = t->next) {
2859 if (strcmp(t->name, buf) == 0)
2866 if (t == current_trace)
2869 trace_branch_disable();
2870 if (current_trace && current_trace->reset)
2871 current_trace->reset(tr);
2872 if (current_trace && current_trace->use_max_tr) {
2874 * We don't free the ring buffer. instead, resize it because
2875 * The max_tr ring buffer has some state (e.g. ring->clock) and
2876 * we want preserve it.
2878 ring_buffer_resize(max_tr.buffer, 1);
2881 destroy_trace_option_files(topts);
2885 topts = create_trace_option_files(current_trace);
2886 if (current_trace->use_max_tr) {
2887 ret = ring_buffer_resize(max_tr.buffer, global_trace.entries);
2890 max_tr.entries = global_trace.entries;
2894 ret = tracer_init(t, tr);
2899 trace_branch_enable(tr);
2901 mutex_unlock(&trace_types_lock);
2907 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2908 size_t cnt, loff_t *ppos)
2910 char buf[MAX_TRACER_SIZE+1];
2917 if (cnt > MAX_TRACER_SIZE)
2918 cnt = MAX_TRACER_SIZE;
2920 if (copy_from_user(&buf, ubuf, cnt))
2925 /* strip ending whitespace. */
2926 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2929 err = tracing_set_tracer(buf);
2939 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2940 size_t cnt, loff_t *ppos)
2942 unsigned long *ptr = filp->private_data;
2946 r = snprintf(buf, sizeof(buf), "%ld\n",
2947 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2948 if (r > sizeof(buf))
2950 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2954 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2955 size_t cnt, loff_t *ppos)
2957 unsigned long *ptr = filp->private_data;
2962 if (cnt >= sizeof(buf))
2965 if (copy_from_user(&buf, ubuf, cnt))
2970 ret = strict_strtoul(buf, 10, &val);
2979 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2981 long cpu_file = (long) inode->i_private;
2982 struct trace_iterator *iter;
2985 if (tracing_disabled)
2988 mutex_lock(&trace_types_lock);
2990 /* create a buffer to store the information to pass to userspace */
2991 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2998 * We make a copy of the current tracer to avoid concurrent
2999 * changes on it while we are reading.
3001 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3007 *iter->trace = *current_trace;
3009 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3014 /* trace pipe does not show start of buffer */
3015 cpumask_setall(iter->started);
3017 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3018 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3020 iter->cpu_file = cpu_file;
3021 iter->tr = &global_trace;
3022 mutex_init(&iter->mutex);
3023 filp->private_data = iter;
3025 if (iter->trace->pipe_open)
3026 iter->trace->pipe_open(iter);
3028 nonseekable_open(inode, filp);
3030 mutex_unlock(&trace_types_lock);
3036 mutex_unlock(&trace_types_lock);
3040 static int tracing_release_pipe(struct inode *inode, struct file *file)
3042 struct trace_iterator *iter = file->private_data;
3044 mutex_lock(&trace_types_lock);
3046 if (iter->trace->pipe_close)
3047 iter->trace->pipe_close(iter);
3049 mutex_unlock(&trace_types_lock);
3051 free_cpumask_var(iter->started);
3052 mutex_destroy(&iter->mutex);
3060 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3062 struct trace_iterator *iter = filp->private_data;
3064 if (trace_flags & TRACE_ITER_BLOCK) {
3066 * Always select as readable when in blocking mode
3068 return POLLIN | POLLRDNORM;
3070 if (!trace_empty(iter))
3071 return POLLIN | POLLRDNORM;
3072 poll_wait(filp, &trace_wait, poll_table);
3073 if (!trace_empty(iter))
3074 return POLLIN | POLLRDNORM;
3081 void default_wait_pipe(struct trace_iterator *iter)
3085 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3087 if (trace_empty(iter))
3090 finish_wait(&trace_wait, &wait);
3094 * This is a make-shift waitqueue.
3095 * A tracer might use this callback on some rare cases:
3097 * 1) the current tracer might hold the runqueue lock when it wakes up
3098 * a reader, hence a deadlock (sched, function, and function graph tracers)
3099 * 2) the function tracers, trace all functions, we don't want
3100 * the overhead of calling wake_up and friends
3101 * (and tracing them too)
3103 * Anyway, this is really very primitive wakeup.
3105 void poll_wait_pipe(struct trace_iterator *iter)
3107 set_current_state(TASK_INTERRUPTIBLE);
3108 /* sleep for 100 msecs, and try again. */
3109 schedule_timeout(HZ / 10);
3112 /* Must be called with trace_types_lock mutex held. */
3113 static int tracing_wait_pipe(struct file *filp)
3115 struct trace_iterator *iter = filp->private_data;
3117 while (trace_empty(iter)) {
3119 if ((filp->f_flags & O_NONBLOCK)) {
3123 mutex_unlock(&iter->mutex);
3125 iter->trace->wait_pipe(iter);
3127 mutex_lock(&iter->mutex);
3129 if (signal_pending(current))
3133 * We block until we read something and tracing is disabled.
3134 * We still block if tracing is disabled, but we have never
3135 * read anything. This allows a user to cat this file, and
3136 * then enable tracing. But after we have read something,
3137 * we give an EOF when tracing is again disabled.
3139 * iter->pos will be 0 if we haven't read anything.
3141 if (!tracer_enabled && iter->pos)
3152 tracing_read_pipe(struct file *filp, char __user *ubuf,
3153 size_t cnt, loff_t *ppos)
3155 struct trace_iterator *iter = filp->private_data;
3156 static struct tracer *old_tracer;
3159 /* return any leftover data */
3160 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3164 trace_seq_init(&iter->seq);
3166 /* copy the tracer to avoid using a global lock all around */
3167 mutex_lock(&trace_types_lock);
3168 if (unlikely(old_tracer != current_trace && current_trace)) {
3169 old_tracer = current_trace;
3170 *iter->trace = *current_trace;
3172 mutex_unlock(&trace_types_lock);
3175 * Avoid more than one consumer on a single file descriptor
3176 * This is just a matter of traces coherency, the ring buffer itself
3179 mutex_lock(&iter->mutex);
3180 if (iter->trace->read) {
3181 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3187 sret = tracing_wait_pipe(filp);
3191 /* stop when tracing is finished */
3192 if (trace_empty(iter)) {
3197 if (cnt >= PAGE_SIZE)
3198 cnt = PAGE_SIZE - 1;
3200 /* reset all but tr, trace, and overruns */
3201 memset(&iter->seq, 0,
3202 sizeof(struct trace_iterator) -
3203 offsetof(struct trace_iterator, seq));
3206 trace_event_read_lock();
3207 trace_access_lock(iter->cpu_file);
3208 while (trace_find_next_entry_inc(iter) != NULL) {
3209 enum print_line_t ret;
3210 int len = iter->seq.len;
3212 ret = print_trace_line(iter);
3213 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3214 /* don't print partial lines */
3215 iter->seq.len = len;
3218 if (ret != TRACE_TYPE_NO_CONSUME)
3219 trace_consume(iter);
3221 if (iter->seq.len >= cnt)
3224 trace_access_unlock(iter->cpu_file);
3225 trace_event_read_unlock();
3227 /* Now copy what we have to the user */
3228 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3229 if (iter->seq.readpos >= iter->seq.len)
3230 trace_seq_init(&iter->seq);
3233 * If there was nothing to send to user, inspite of consuming trace
3234 * entries, go back to wait for more entries.
3240 mutex_unlock(&iter->mutex);
3245 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3246 struct pipe_buffer *buf)
3248 __free_page(buf->page);
3251 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3254 __free_page(spd->pages[idx]);
3257 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3259 .map = generic_pipe_buf_map,
3260 .unmap = generic_pipe_buf_unmap,
3261 .confirm = generic_pipe_buf_confirm,
3262 .release = tracing_pipe_buf_release,
3263 .steal = generic_pipe_buf_steal,
3264 .get = generic_pipe_buf_get,
3268 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3273 /* Seq buffer is page-sized, exactly what we need. */
3275 count = iter->seq.len;
3276 ret = print_trace_line(iter);
3277 count = iter->seq.len - count;
3280 iter->seq.len -= count;
3283 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3284 iter->seq.len -= count;
3288 if (ret != TRACE_TYPE_NO_CONSUME)
3289 trace_consume(iter);
3291 if (!trace_find_next_entry_inc(iter)) {
3301 static ssize_t tracing_splice_read_pipe(struct file *filp,
3303 struct pipe_inode_info *pipe,
3307 struct page *pages_def[PIPE_DEF_BUFFERS];
3308 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3309 struct trace_iterator *iter = filp->private_data;
3310 struct splice_pipe_desc spd = {
3312 .partial = partial_def,
3313 .nr_pages = 0, /* This gets updated below. */
3315 .ops = &tracing_pipe_buf_ops,
3316 .spd_release = tracing_spd_release_pipe,
3318 static struct tracer *old_tracer;
3323 if (splice_grow_spd(pipe, &spd))
3326 /* copy the tracer to avoid using a global lock all around */
3327 mutex_lock(&trace_types_lock);
3328 if (unlikely(old_tracer != current_trace && current_trace)) {
3329 old_tracer = current_trace;
3330 *iter->trace = *current_trace;
3332 mutex_unlock(&trace_types_lock);
3334 mutex_lock(&iter->mutex);
3336 if (iter->trace->splice_read) {
3337 ret = iter->trace->splice_read(iter, filp,
3338 ppos, pipe, len, flags);
3343 ret = tracing_wait_pipe(filp);
3347 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
3352 trace_event_read_lock();
3353 trace_access_lock(iter->cpu_file);
3355 /* Fill as many pages as possible. */
3356 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3357 spd.pages[i] = alloc_page(GFP_KERNEL);
3361 rem = tracing_fill_pipe_page(rem, iter);
3363 /* Copy the data into the page, so we can start over. */
3364 ret = trace_seq_to_buffer(&iter->seq,
3365 page_address(spd.pages[i]),
3368 __free_page(spd.pages[i]);
3371 spd.partial[i].offset = 0;
3372 spd.partial[i].len = iter->seq.len;
3374 trace_seq_init(&iter->seq);
3377 trace_access_unlock(iter->cpu_file);
3378 trace_event_read_unlock();
3379 mutex_unlock(&iter->mutex);
3383 ret = splice_to_pipe(pipe, &spd);
3385 splice_shrink_spd(pipe, &spd);
3389 mutex_unlock(&iter->mutex);
3394 tracing_entries_read(struct file *filp, char __user *ubuf,
3395 size_t cnt, loff_t *ppos)
3397 struct trace_array *tr = filp->private_data;
3401 mutex_lock(&trace_types_lock);
3402 if (!ring_buffer_expanded)
3403 r = sprintf(buf, "%lu (expanded: %lu)\n",
3405 trace_buf_size >> 10);
3407 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3408 mutex_unlock(&trace_types_lock);
3410 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3414 tracing_entries_write(struct file *filp, const char __user *ubuf,
3415 size_t cnt, loff_t *ppos)
3421 if (cnt >= sizeof(buf))
3424 if (copy_from_user(&buf, ubuf, cnt))
3429 ret = strict_strtoul(buf, 10, &val);
3433 /* must have at least 1 entry */
3437 mutex_lock(&trace_types_lock);
3441 /* disable all cpu buffers */
3442 for_each_tracing_cpu(cpu) {
3443 if (global_trace.data[cpu])
3444 atomic_inc(&global_trace.data[cpu]->disabled);
3445 if (max_tr.data[cpu])
3446 atomic_inc(&max_tr.data[cpu]->disabled);
3449 /* value is in KB */
3452 if (val != global_trace.entries) {
3453 ret = tracing_resize_ring_buffer(val);
3462 /* If check pages failed, return ENOMEM */
3463 if (tracing_disabled)
3466 for_each_tracing_cpu(cpu) {
3467 if (global_trace.data[cpu])
3468 atomic_dec(&global_trace.data[cpu]->disabled);
3469 if (max_tr.data[cpu])
3470 atomic_dec(&max_tr.data[cpu]->disabled);
3474 mutex_unlock(&trace_types_lock);
3479 static int mark_printk(const char *fmt, ...)
3483 va_start(args, fmt);
3484 ret = trace_vprintk(0, fmt, args);
3490 tracing_mark_write(struct file *filp, const char __user *ubuf,
3491 size_t cnt, loff_t *fpos)
3496 if (tracing_disabled)
3499 if (cnt > TRACE_BUF_SIZE)
3500 cnt = TRACE_BUF_SIZE;
3502 buf = kmalloc(cnt + 2, GFP_KERNEL);
3506 if (copy_from_user(buf, ubuf, cnt)) {
3510 if (buf[cnt-1] != '\n') {
3516 written = mark_printk("%s", buf);
3520 /* don't tell userspace we wrote more - it might confuse them */
3527 static int tracing_clock_show(struct seq_file *m, void *v)
3531 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3533 "%s%s%s%s", i ? " " : "",
3534 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3535 i == trace_clock_id ? "]" : "");
3541 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3542 size_t cnt, loff_t *fpos)
3545 const char *clockstr;
3548 if (cnt >= sizeof(buf))
3551 if (copy_from_user(&buf, ubuf, cnt))
3556 clockstr = strstrip(buf);
3558 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3559 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3562 if (i == ARRAY_SIZE(trace_clocks))
3567 mutex_lock(&trace_types_lock);
3569 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3571 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3573 mutex_unlock(&trace_types_lock);
3580 static int tracing_clock_open(struct inode *inode, struct file *file)
3582 if (tracing_disabled)
3584 return single_open(file, tracing_clock_show, NULL);
3587 static const struct file_operations tracing_max_lat_fops = {
3588 .open = tracing_open_generic,
3589 .read = tracing_max_lat_read,
3590 .write = tracing_max_lat_write,
3591 .llseek = generic_file_llseek,
3594 static const struct file_operations tracing_ctrl_fops = {
3595 .open = tracing_open_generic,
3596 .read = tracing_ctrl_read,
3597 .write = tracing_ctrl_write,
3598 .llseek = generic_file_llseek,
3601 static const struct file_operations set_tracer_fops = {
3602 .open = tracing_open_generic,
3603 .read = tracing_set_trace_read,
3604 .write = tracing_set_trace_write,
3605 .llseek = generic_file_llseek,
3608 static const struct file_operations tracing_pipe_fops = {
3609 .open = tracing_open_pipe,
3610 .poll = tracing_poll_pipe,
3611 .read = tracing_read_pipe,
3612 .splice_read = tracing_splice_read_pipe,
3613 .release = tracing_release_pipe,
3614 .llseek = no_llseek,
3617 static const struct file_operations tracing_entries_fops = {
3618 .open = tracing_open_generic,
3619 .read = tracing_entries_read,
3620 .write = tracing_entries_write,
3621 .llseek = generic_file_llseek,
3624 static const struct file_operations tracing_mark_fops = {
3625 .open = tracing_open_generic,
3626 .write = tracing_mark_write,
3627 .llseek = generic_file_llseek,
3630 static const struct file_operations trace_clock_fops = {
3631 .open = tracing_clock_open,
3633 .llseek = seq_lseek,
3634 .release = single_release,
3635 .write = tracing_clock_write,
3638 struct ftrace_buffer_info {
3639 struct trace_array *tr;
3645 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3647 int cpu = (int)(long)inode->i_private;
3648 struct ftrace_buffer_info *info;
3650 if (tracing_disabled)
3653 info = kzalloc(sizeof(*info), GFP_KERNEL);
3657 info->tr = &global_trace;
3660 /* Force reading ring buffer for first read */
3661 info->read = (unsigned int)-1;
3663 filp->private_data = info;
3665 return nonseekable_open(inode, filp);
3669 tracing_buffers_read(struct file *filp, char __user *ubuf,
3670 size_t count, loff_t *ppos)
3672 struct ftrace_buffer_info *info = filp->private_data;
3680 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3684 /* Do we have previous read data to read? */
3685 if (info->read < PAGE_SIZE)
3690 trace_access_lock(info->cpu);
3691 ret = ring_buffer_read_page(info->tr->buffer,
3695 trace_access_unlock(info->cpu);
3700 size = PAGE_SIZE - info->read;
3704 ret = copy_to_user(ubuf, info->spare + info->read, size);
3715 static int tracing_buffers_release(struct inode *inode, struct file *file)
3717 struct ftrace_buffer_info *info = file->private_data;
3720 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3727 struct ring_buffer *buffer;
3732 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3733 struct pipe_buffer *buf)
3735 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3740 ring_buffer_free_read_page(ref->buffer, ref->page);
3745 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3746 struct pipe_buffer *buf)
3751 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3752 struct pipe_buffer *buf)
3754 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3759 /* Pipe buffer operations for a buffer. */
3760 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
3762 .map = generic_pipe_buf_map,
3763 .unmap = generic_pipe_buf_unmap,
3764 .confirm = generic_pipe_buf_confirm,
3765 .release = buffer_pipe_buf_release,
3766 .steal = buffer_pipe_buf_steal,
3767 .get = buffer_pipe_buf_get,
3771 * Callback from splice_to_pipe(), if we need to release some pages
3772 * at the end of the spd in case we error'ed out in filling the pipe.
3774 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3776 struct buffer_ref *ref =
3777 (struct buffer_ref *)spd->partial[i].private;
3782 ring_buffer_free_read_page(ref->buffer, ref->page);
3784 spd->partial[i].private = 0;
3788 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3789 struct pipe_inode_info *pipe, size_t len,
3792 struct ftrace_buffer_info *info = file->private_data;
3793 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3794 struct page *pages_def[PIPE_DEF_BUFFERS];
3795 struct splice_pipe_desc spd = {
3797 .partial = partial_def,
3799 .ops = &buffer_pipe_buf_ops,
3800 .spd_release = buffer_spd_release,
3802 struct buffer_ref *ref;
3803 int entries, size, i;
3806 if (splice_grow_spd(pipe, &spd))
3809 if (*ppos & (PAGE_SIZE - 1)) {
3810 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3815 if (len & (PAGE_SIZE - 1)) {
3816 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3817 if (len < PAGE_SIZE) {
3824 trace_access_lock(info->cpu);
3825 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3827 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
3831 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3836 ref->buffer = info->tr->buffer;
3837 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3843 r = ring_buffer_read_page(ref->buffer, &ref->page,
3846 ring_buffer_free_read_page(ref->buffer,
3853 * zero out any left over data, this is going to
3856 size = ring_buffer_page_len(ref->page);
3857 if (size < PAGE_SIZE)
3858 memset(ref->page + size, 0, PAGE_SIZE - size);
3860 page = virt_to_page(ref->page);
3862 spd.pages[i] = page;
3863 spd.partial[i].len = PAGE_SIZE;
3864 spd.partial[i].offset = 0;
3865 spd.partial[i].private = (unsigned long)ref;
3869 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3872 trace_access_unlock(info->cpu);
3875 /* did we read anything? */
3876 if (!spd.nr_pages) {
3877 if (flags & SPLICE_F_NONBLOCK)
3885 ret = splice_to_pipe(pipe, &spd);
3886 splice_shrink_spd(pipe, &spd);
3891 static const struct file_operations tracing_buffers_fops = {
3892 .open = tracing_buffers_open,
3893 .read = tracing_buffers_read,
3894 .release = tracing_buffers_release,
3895 .splice_read = tracing_buffers_splice_read,
3896 .llseek = no_llseek,
3900 tracing_stats_read(struct file *filp, char __user *ubuf,
3901 size_t count, loff_t *ppos)
3903 unsigned long cpu = (unsigned long)filp->private_data;
3904 struct trace_array *tr = &global_trace;
3905 struct trace_seq *s;
3908 s = kmalloc(sizeof(*s), GFP_KERNEL);
3914 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3915 trace_seq_printf(s, "entries: %ld\n", cnt);
3917 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3918 trace_seq_printf(s, "overrun: %ld\n", cnt);
3920 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3921 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3923 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3930 static const struct file_operations tracing_stats_fops = {
3931 .open = tracing_open_generic,
3932 .read = tracing_stats_read,
3933 .llseek = generic_file_llseek,
3936 #ifdef CONFIG_DYNAMIC_FTRACE
3938 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3944 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3945 size_t cnt, loff_t *ppos)
3947 static char ftrace_dyn_info_buffer[1024];
3948 static DEFINE_MUTEX(dyn_info_mutex);
3949 unsigned long *p = filp->private_data;
3950 char *buf = ftrace_dyn_info_buffer;
3951 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3954 mutex_lock(&dyn_info_mutex);
3955 r = sprintf(buf, "%ld ", *p);
3957 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3960 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3962 mutex_unlock(&dyn_info_mutex);
3967 static const struct file_operations tracing_dyn_info_fops = {
3968 .open = tracing_open_generic,
3969 .read = tracing_read_dyn_info,
3970 .llseek = generic_file_llseek,
3974 static struct dentry *d_tracer;
3976 struct dentry *tracing_init_dentry(void)
3983 if (!debugfs_initialized())
3986 d_tracer = debugfs_create_dir("tracing", NULL);
3988 if (!d_tracer && !once) {
3990 pr_warning("Could not create debugfs directory 'tracing'\n");
3997 static struct dentry *d_percpu;
3999 struct dentry *tracing_dentry_percpu(void)
4002 struct dentry *d_tracer;
4007 d_tracer = tracing_init_dentry();
4012 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
4014 if (!d_percpu && !once) {
4016 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4023 static void tracing_init_debugfs_percpu(long cpu)
4025 struct dentry *d_percpu = tracing_dentry_percpu();
4026 struct dentry *d_cpu;
4027 char cpu_dir[30]; /* 30 characters should be more than enough */
4029 snprintf(cpu_dir, 30, "cpu%ld", cpu);
4030 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4032 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4036 /* per cpu trace_pipe */
4037 trace_create_file("trace_pipe", 0444, d_cpu,
4038 (void *) cpu, &tracing_pipe_fops);
4041 trace_create_file("trace", 0644, d_cpu,
4042 (void *) cpu, &tracing_fops);
4044 trace_create_file("trace_pipe_raw", 0444, d_cpu,
4045 (void *) cpu, &tracing_buffers_fops);
4047 trace_create_file("stats", 0444, d_cpu,
4048 (void *) cpu, &tracing_stats_fops);
4051 #ifdef CONFIG_FTRACE_SELFTEST
4052 /* Let selftest have access to static functions in this file */
4053 #include "trace_selftest.c"
4056 struct trace_option_dentry {
4057 struct tracer_opt *opt;
4058 struct tracer_flags *flags;
4059 struct dentry *entry;
4063 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4066 struct trace_option_dentry *topt = filp->private_data;
4069 if (topt->flags->val & topt->opt->bit)
4074 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4078 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4081 struct trace_option_dentry *topt = filp->private_data;
4086 if (cnt >= sizeof(buf))
4089 if (copy_from_user(&buf, ubuf, cnt))
4094 ret = strict_strtoul(buf, 10, &val);
4098 if (val != 0 && val != 1)
4101 if (!!(topt->flags->val & topt->opt->bit) != val) {
4102 mutex_lock(&trace_types_lock);
4103 ret = __set_tracer_option(current_trace, topt->flags,
4105 mutex_unlock(&trace_types_lock);
4116 static const struct file_operations trace_options_fops = {
4117 .open = tracing_open_generic,
4118 .read = trace_options_read,
4119 .write = trace_options_write,
4120 .llseek = generic_file_llseek,
4124 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4127 long index = (long)filp->private_data;
4130 if (trace_flags & (1 << index))
4135 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4139 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4142 long index = (long)filp->private_data;
4147 if (cnt >= sizeof(buf))
4150 if (copy_from_user(&buf, ubuf, cnt))
4155 ret = strict_strtoul(buf, 10, &val);
4159 if (val != 0 && val != 1)
4161 set_tracer_flags(1 << index, val);
4168 static const struct file_operations trace_options_core_fops = {
4169 .open = tracing_open_generic,
4170 .read = trace_options_core_read,
4171 .write = trace_options_core_write,
4172 .llseek = generic_file_llseek,
4175 struct dentry *trace_create_file(const char *name,
4177 struct dentry *parent,
4179 const struct file_operations *fops)
4183 ret = debugfs_create_file(name, mode, parent, data, fops);
4185 pr_warning("Could not create debugfs '%s' entry\n", name);
4191 static struct dentry *trace_options_init_dentry(void)
4193 struct dentry *d_tracer;
4194 static struct dentry *t_options;
4199 d_tracer = tracing_init_dentry();
4203 t_options = debugfs_create_dir("options", d_tracer);
4205 pr_warning("Could not create debugfs directory 'options'\n");
4213 create_trace_option_file(struct trace_option_dentry *topt,
4214 struct tracer_flags *flags,
4215 struct tracer_opt *opt)
4217 struct dentry *t_options;
4219 t_options = trace_options_init_dentry();
4223 topt->flags = flags;
4226 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4227 &trace_options_fops);
4231 static struct trace_option_dentry *
4232 create_trace_option_files(struct tracer *tracer)
4234 struct trace_option_dentry *topts;
4235 struct tracer_flags *flags;
4236 struct tracer_opt *opts;
4242 flags = tracer->flags;
4244 if (!flags || !flags->opts)
4249 for (cnt = 0; opts[cnt].name; cnt++)
4252 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4256 for (cnt = 0; opts[cnt].name; cnt++)
4257 create_trace_option_file(&topts[cnt], flags,
4264 destroy_trace_option_files(struct trace_option_dentry *topts)
4271 for (cnt = 0; topts[cnt].opt; cnt++) {
4272 if (topts[cnt].entry)
4273 debugfs_remove(topts[cnt].entry);
4279 static struct dentry *
4280 create_trace_option_core_file(const char *option, long index)
4282 struct dentry *t_options;
4284 t_options = trace_options_init_dentry();
4288 return trace_create_file(option, 0644, t_options, (void *)index,
4289 &trace_options_core_fops);
4292 static __init void create_trace_options_dir(void)
4294 struct dentry *t_options;
4297 t_options = trace_options_init_dentry();
4301 for (i = 0; trace_options[i]; i++)
4302 create_trace_option_core_file(trace_options[i], i);
4305 static __init int tracer_init_debugfs(void)
4307 struct dentry *d_tracer;
4310 trace_access_lock_init();
4312 d_tracer = tracing_init_dentry();
4314 trace_create_file("tracing_enabled", 0644, d_tracer,
4315 &global_trace, &tracing_ctrl_fops);
4317 trace_create_file("trace_options", 0644, d_tracer,
4318 NULL, &tracing_iter_fops);
4320 trace_create_file("tracing_cpumask", 0644, d_tracer,
4321 NULL, &tracing_cpumask_fops);
4323 trace_create_file("trace", 0644, d_tracer,
4324 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4326 trace_create_file("available_tracers", 0444, d_tracer,
4327 &global_trace, &show_traces_fops);
4329 trace_create_file("current_tracer", 0644, d_tracer,
4330 &global_trace, &set_tracer_fops);
4332 #ifdef CONFIG_TRACER_MAX_TRACE
4333 trace_create_file("tracing_max_latency", 0644, d_tracer,
4334 &tracing_max_latency, &tracing_max_lat_fops);
4337 trace_create_file("tracing_thresh", 0644, d_tracer,
4338 &tracing_thresh, &tracing_max_lat_fops);
4340 trace_create_file("README", 0444, d_tracer,
4341 NULL, &tracing_readme_fops);
4343 trace_create_file("trace_pipe", 0444, d_tracer,
4344 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4346 trace_create_file("buffer_size_kb", 0644, d_tracer,
4347 &global_trace, &tracing_entries_fops);
4349 trace_create_file("trace_marker", 0220, d_tracer,
4350 NULL, &tracing_mark_fops);
4352 trace_create_file("saved_cmdlines", 0444, d_tracer,
4353 NULL, &tracing_saved_cmdlines_fops);
4355 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4358 #ifdef CONFIG_DYNAMIC_FTRACE
4359 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4360 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4363 create_trace_options_dir();
4365 for_each_tracing_cpu(cpu)
4366 tracing_init_debugfs_percpu(cpu);
4371 static int trace_panic_handler(struct notifier_block *this,
4372 unsigned long event, void *unused)
4374 if (ftrace_dump_on_oops)
4375 ftrace_dump(ftrace_dump_on_oops);
4379 static struct notifier_block trace_panic_notifier = {
4380 .notifier_call = trace_panic_handler,
4382 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4385 static int trace_die_handler(struct notifier_block *self,
4391 if (ftrace_dump_on_oops)
4392 ftrace_dump(ftrace_dump_on_oops);
4400 static struct notifier_block trace_die_notifier = {
4401 .notifier_call = trace_die_handler,
4406 * printk is set to max of 1024, we really don't need it that big.
4407 * Nothing should be printing 1000 characters anyway.
4409 #define TRACE_MAX_PRINT 1000
4412 * Define here KERN_TRACE so that we have one place to modify
4413 * it if we decide to change what log level the ftrace dump
4416 #define KERN_TRACE KERN_EMERG
4419 trace_printk_seq(struct trace_seq *s)
4421 /* Probably should print a warning here. */
4425 /* should be zero ended, but we are paranoid. */
4426 s->buffer[s->len] = 0;
4428 printk(KERN_TRACE "%s", s->buffer);
4433 void trace_init_global_iter(struct trace_iterator *iter)
4435 iter->tr = &global_trace;
4436 iter->trace = current_trace;
4437 iter->cpu_file = TRACE_PIPE_ALL_CPU;
4441 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4443 static arch_spinlock_t ftrace_dump_lock =
4444 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4445 /* use static because iter can be a bit big for the stack */
4446 static struct trace_iterator iter;
4447 unsigned int old_userobj;
4448 static int dump_ran;
4449 unsigned long flags;
4453 local_irq_save(flags);
4454 arch_spin_lock(&ftrace_dump_lock);
4462 if (disable_tracing)
4465 trace_init_global_iter(&iter);
4467 for_each_tracing_cpu(cpu) {
4468 atomic_inc(&iter.tr->data[cpu]->disabled);
4471 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4473 /* don't look at user memory in panic mode */
4474 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4476 /* Simulate the iterator */
4477 iter.tr = &global_trace;
4478 iter.trace = current_trace;
4480 switch (oops_dump_mode) {
4482 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4485 iter.cpu_file = raw_smp_processor_id();
4490 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
4491 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4494 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4497 * We need to stop all tracing on all CPUS to read the
4498 * the next buffer. This is a bit expensive, but is
4499 * not done often. We fill all what we can read,
4500 * and then release the locks again.
4503 while (!trace_empty(&iter)) {
4506 printk(KERN_TRACE "---------------------------------\n");
4510 /* reset all but tr, trace, and overruns */
4511 memset(&iter.seq, 0,
4512 sizeof(struct trace_iterator) -
4513 offsetof(struct trace_iterator, seq));
4514 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4517 if (trace_find_next_entry_inc(&iter) != NULL) {
4520 ret = print_trace_line(&iter);
4521 if (ret != TRACE_TYPE_NO_CONSUME)
4522 trace_consume(&iter);
4525 trace_printk_seq(&iter.seq);
4529 printk(KERN_TRACE " (ftrace buffer empty)\n");
4531 printk(KERN_TRACE "---------------------------------\n");
4534 /* Re-enable tracing if requested */
4535 if (!disable_tracing) {
4536 trace_flags |= old_userobj;
4538 for_each_tracing_cpu(cpu) {
4539 atomic_dec(&iter.tr->data[cpu]->disabled);
4545 arch_spin_unlock(&ftrace_dump_lock);
4546 local_irq_restore(flags);
4549 /* By default: disable tracing after the dump */
4550 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
4552 __ftrace_dump(true, oops_dump_mode);
4555 __init static int tracer_alloc_buffers(void)
4558 enum ring_buffer_flags rb_flags;
4563 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4566 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4567 goto out_free_buffer_mask;
4569 /* To save memory, keep the ring buffer size to its minimum */
4570 if (ring_buffer_expanded)
4571 ring_buf_size = trace_buf_size;
4575 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
4577 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4578 cpumask_copy(tracing_cpumask, cpu_all_mask);
4580 /* TODO: make the number of buffers hot pluggable with CPUS */
4581 global_trace.buffer = ring_buffer_alloc(ring_buf_size, rb_flags);
4582 if (!global_trace.buffer) {
4583 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4585 goto out_free_cpumask;
4587 global_trace.entries = ring_buffer_size(global_trace.buffer);
4590 #ifdef CONFIG_TRACER_MAX_TRACE
4591 max_tr.buffer = ring_buffer_alloc(1, rb_flags);
4592 if (!max_tr.buffer) {
4593 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4595 ring_buffer_free(global_trace.buffer);
4596 goto out_free_cpumask;
4601 /* Allocate the first page for all buffers */
4602 for_each_tracing_cpu(i) {
4603 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4604 max_tr.data[i] = &per_cpu(max_tr_data, i);
4607 trace_init_cmdlines();
4609 register_tracer(&nop_trace);
4610 current_trace = &nop_trace;
4611 /* All seems OK, enable tracing */
4612 tracing_disabled = 0;
4614 atomic_notifier_chain_register(&panic_notifier_list,
4615 &trace_panic_notifier);
4617 register_die_notifier(&trace_die_notifier);
4622 free_cpumask_var(tracing_cpumask);
4623 out_free_buffer_mask:
4624 free_cpumask_var(tracing_buffer_mask);
4629 __init static int clear_boot_tracer(void)
4632 * The default tracer at boot buffer is an init section.
4633 * This function is called in lateinit. If we did not
4634 * find the boot tracer, then clear it out, to prevent
4635 * later registration from accessing the buffer that is
4636 * about to be freed.
4638 if (!default_bootup_tracer)
4641 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4642 default_bootup_tracer);
4643 default_bootup_tracer = NULL;
4648 early_initcall(tracer_alloc_buffers);
4649 fs_initcall(tracer_init_debugfs);
4650 late_initcall(clear_boot_tracer);