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/smp_lock.h>
21 #include <linux/notifier.h>
22 #include <linux/irqflags.h>
23 #include <linux/debugfs.h>
24 #include <linux/pagemap.h>
25 #include <linux/hardirq.h>
26 #include <linux/linkage.h>
27 #include <linux/uaccess.h>
28 #include <linux/kprobes.h>
29 #include <linux/ftrace.h>
30 #include <linux/module.h>
31 #include <linux/percpu.h>
32 #include <linux/splice.h>
33 #include <linux/kdebug.h>
34 #include <linux/string.h>
35 #include <linux/rwsem.h>
36 #include <linux/slab.h>
37 #include <linux/ctype.h>
38 #include <linux/init.h>
39 #include <linux/poll.h>
43 #include "trace_output.h"
45 #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
48 * On boot up, the ring buffer is set to the minimum size, so that
49 * we do not waste memory on systems that are not using tracing.
51 int ring_buffer_expanded;
54 * We need to change this state when a selftest is running.
55 * A selftest will lurk into the ring-buffer to count the
56 * entries inserted during the selftest although some concurrent
57 * insertions into the ring-buffer such as trace_printk could occurred
58 * at the same time, giving false positive or negative results.
60 static bool __read_mostly tracing_selftest_running;
63 * If a tracer is running, we do not want to run SELFTEST.
65 bool __read_mostly tracing_selftest_disabled;
67 /* For tracers that don't implement custom flags */
68 static struct tracer_opt dummy_tracer_opt[] = {
72 static struct tracer_flags dummy_tracer_flags = {
74 .opts = dummy_tracer_opt
77 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
83 * Kill all tracing for good (never come back).
84 * It is initialized to 1 but will turn to zero if the initialization
85 * of the tracer is successful. But that is the only place that sets
88 static int tracing_disabled = 1;
90 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
92 static inline void ftrace_disable_cpu(void)
95 __this_cpu_inc(ftrace_cpu_disabled);
98 static inline void ftrace_enable_cpu(void)
100 __this_cpu_dec(ftrace_cpu_disabled);
104 static cpumask_var_t __read_mostly tracing_buffer_mask;
106 #define for_each_tracing_cpu(cpu) \
107 for_each_cpu(cpu, tracing_buffer_mask)
110 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
112 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
113 * is set, then ftrace_dump is called. This will output the contents
114 * of the ftrace buffers to the console. This is very useful for
115 * capturing traces that lead to crashes and outputing it to a
118 * It is default off, but you can enable it with either specifying
119 * "ftrace_dump_on_oops" in the kernel command line, or setting
120 * /proc/sys/kernel/ftrace_dump_on_oops
121 * Set 1 if you want to dump buffers of all CPUs
122 * Set 2 if you want to dump the buffer of the CPU that triggered oops
125 enum ftrace_dump_mode ftrace_dump_on_oops;
127 static int tracing_set_tracer(const char *buf);
129 #define MAX_TRACER_SIZE 100
130 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
131 static char *default_bootup_tracer;
133 static int __init set_cmdline_ftrace(char *str)
135 strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
136 default_bootup_tracer = bootup_tracer_buf;
137 /* We are using ftrace early, expand it */
138 ring_buffer_expanded = 1;
141 __setup("ftrace=", set_cmdline_ftrace);
143 static int __init set_ftrace_dump_on_oops(char *str)
145 if (*str++ != '=' || !*str) {
146 ftrace_dump_on_oops = DUMP_ALL;
150 if (!strcmp("orig_cpu", str)) {
151 ftrace_dump_on_oops = DUMP_ORIG;
157 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
159 unsigned long long ns2usecs(cycle_t nsec)
167 * The global_trace is the descriptor that holds the tracing
168 * buffers for the live tracing. For each CPU, it contains
169 * a link list of pages that will store trace entries. The
170 * page descriptor of the pages in the memory is used to hold
171 * the link list by linking the lru item in the page descriptor
172 * to each of the pages in the buffer per CPU.
174 * For each active CPU there is a data field that holds the
175 * pages for the buffer for that CPU. Each CPU has the same number
176 * of pages allocated for its buffer.
178 static struct trace_array global_trace;
180 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
182 int filter_current_check_discard(struct ring_buffer *buffer,
183 struct ftrace_event_call *call, void *rec,
184 struct ring_buffer_event *event)
186 return filter_check_discard(call, rec, buffer, event);
188 EXPORT_SYMBOL_GPL(filter_current_check_discard);
190 cycle_t ftrace_now(int cpu)
194 /* Early boot up does not have a buffer yet */
195 if (!global_trace.buffer)
196 return trace_clock_local();
198 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
199 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
205 * The max_tr is used to snapshot the global_trace when a maximum
206 * latency is reached. Some tracers will use this to store a maximum
207 * trace while it continues examining live traces.
209 * The buffers for the max_tr are set up the same as the global_trace.
210 * When a snapshot is taken, the link list of the max_tr is swapped
211 * with the link list of the global_trace and the buffers are reset for
212 * the global_trace so the tracing can continue.
214 static struct trace_array max_tr;
216 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
218 /* tracer_enabled is used to toggle activation of a tracer */
219 static int tracer_enabled = 1;
222 * tracing_is_enabled - return tracer_enabled status
224 * This function is used by other tracers to know the status
225 * of the tracer_enabled flag. Tracers may use this function
226 * to know if it should enable their features when starting
227 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
229 int tracing_is_enabled(void)
231 return tracer_enabled;
235 * trace_buf_size is the size in bytes that is allocated
236 * for a buffer. Note, the number of bytes is always rounded
239 * This number is purposely set to a low number of 16384.
240 * If the dump on oops happens, it will be much appreciated
241 * to not have to wait for all that output. Anyway this can be
242 * boot time and run time configurable.
244 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
246 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
248 /* trace_types holds a link list of available tracers. */
249 static struct tracer *trace_types __read_mostly;
251 /* current_trace points to the tracer that is currently active */
252 static struct tracer *current_trace __read_mostly;
255 * trace_types_lock is used to protect the trace_types list.
257 static DEFINE_MUTEX(trace_types_lock);
260 * serialize the access of the ring buffer
262 * ring buffer serializes readers, but it is low level protection.
263 * The validity of the events (which returns by ring_buffer_peek() ..etc)
264 * are not protected by ring buffer.
266 * The content of events may become garbage if we allow other process consumes
267 * these events concurrently:
268 * A) the page of the consumed events may become a normal page
269 * (not reader page) in ring buffer, and this page will be rewrited
270 * by events producer.
271 * B) The page of the consumed events may become a page for splice_read,
272 * and this page will be returned to system.
274 * These primitives allow multi process access to different cpu ring buffer
277 * These primitives don't distinguish read-only and read-consume access.
278 * Multi read-only access are also serialized.
282 static DECLARE_RWSEM(all_cpu_access_lock);
283 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
285 static inline void trace_access_lock(int cpu)
287 if (cpu == TRACE_PIPE_ALL_CPU) {
288 /* gain it for accessing the whole ring buffer. */
289 down_write(&all_cpu_access_lock);
291 /* gain it for accessing a cpu ring buffer. */
293 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
294 down_read(&all_cpu_access_lock);
296 /* Secondly block other access to this @cpu ring buffer. */
297 mutex_lock(&per_cpu(cpu_access_lock, cpu));
301 static inline void trace_access_unlock(int cpu)
303 if (cpu == TRACE_PIPE_ALL_CPU) {
304 up_write(&all_cpu_access_lock);
306 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
307 up_read(&all_cpu_access_lock);
311 static inline void trace_access_lock_init(void)
315 for_each_possible_cpu(cpu)
316 mutex_init(&per_cpu(cpu_access_lock, cpu));
321 static DEFINE_MUTEX(access_lock);
323 static inline void trace_access_lock(int cpu)
326 mutex_lock(&access_lock);
329 static inline void trace_access_unlock(int cpu)
332 mutex_unlock(&access_lock);
335 static inline void trace_access_lock_init(void)
341 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
342 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
344 /* trace_flags holds trace_options default values */
345 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
346 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
347 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD;
349 static int trace_stop_count;
350 static DEFINE_SPINLOCK(tracing_start_lock);
353 * trace_wake_up - wake up tasks waiting for trace input
355 * Simply wakes up any task that is blocked on the trace_wait
356 * queue. These is used with trace_poll for tasks polling the trace.
358 void trace_wake_up(void)
362 if (trace_flags & TRACE_ITER_BLOCK)
365 * The runqueue_is_locked() can fail, but this is the best we
369 if (!runqueue_is_locked(cpu))
370 wake_up(&trace_wait);
374 static int __init set_buf_size(char *str)
376 unsigned long buf_size;
380 buf_size = memparse(str, &str);
381 /* nr_entries can not be zero */
384 trace_buf_size = buf_size;
387 __setup("trace_buf_size=", set_buf_size);
389 static int __init set_tracing_thresh(char *str)
391 unsigned long threshhold;
396 ret = strict_strtoul(str, 0, &threshhold);
399 tracing_thresh = threshhold * 1000;
402 __setup("tracing_thresh=", set_tracing_thresh);
404 unsigned long nsecs_to_usecs(unsigned long nsecs)
409 /* These must match the bit postions in trace_iterator_flags */
410 static const char *trace_options[] = {
439 { trace_clock_local, "local" },
440 { trace_clock_global, "global" },
446 * trace_parser_get_init - gets the buffer for trace parser
448 int trace_parser_get_init(struct trace_parser *parser, int size)
450 memset(parser, 0, sizeof(*parser));
452 parser->buffer = kmalloc(size, GFP_KERNEL);
461 * trace_parser_put - frees the buffer for trace parser
463 void trace_parser_put(struct trace_parser *parser)
465 kfree(parser->buffer);
469 * trace_get_user - reads the user input string separated by space
470 * (matched by isspace(ch))
472 * For each string found the 'struct trace_parser' is updated,
473 * and the function returns.
475 * Returns number of bytes read.
477 * See kernel/trace/trace.h for 'struct trace_parser' details.
479 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
480 size_t cnt, loff_t *ppos)
487 trace_parser_clear(parser);
489 ret = get_user(ch, ubuf++);
497 * The parser is not finished with the last write,
498 * continue reading the user input without skipping spaces.
501 /* skip white space */
502 while (cnt && isspace(ch)) {
503 ret = get_user(ch, ubuf++);
510 /* only spaces were written */
520 /* read the non-space input */
521 while (cnt && !isspace(ch)) {
522 if (parser->idx < parser->size - 1)
523 parser->buffer[parser->idx++] = ch;
528 ret = get_user(ch, ubuf++);
535 /* We either got finished input or we have to wait for another call. */
537 parser->buffer[parser->idx] = 0;
538 parser->cont = false;
541 parser->buffer[parser->idx++] = ch;
551 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
559 if (s->len <= s->readpos)
562 len = s->len - s->readpos;
565 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
575 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
580 if (s->len <= s->readpos)
583 len = s->len - s->readpos;
586 ret = memcpy(buf, s->buffer + s->readpos, cnt);
595 * ftrace_max_lock is used to protect the swapping of buffers
596 * when taking a max snapshot. The buffers themselves are
597 * protected by per_cpu spinlocks. But the action of the swap
598 * needs its own lock.
600 * This is defined as a arch_spinlock_t in order to help
601 * with performance when lockdep debugging is enabled.
603 * It is also used in other places outside the update_max_tr
604 * so it needs to be defined outside of the
605 * CONFIG_TRACER_MAX_TRACE.
607 static arch_spinlock_t ftrace_max_lock =
608 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
610 unsigned long __read_mostly tracing_thresh;
612 #ifdef CONFIG_TRACER_MAX_TRACE
613 unsigned long __read_mostly tracing_max_latency;
616 * Copy the new maximum trace into the separate maximum-trace
617 * structure. (this way the maximum trace is permanently saved,
618 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
621 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
623 struct trace_array_cpu *data = tr->data[cpu];
624 struct trace_array_cpu *max_data;
627 max_tr.time_start = data->preempt_timestamp;
629 max_data = max_tr.data[cpu];
630 max_data->saved_latency = tracing_max_latency;
631 max_data->critical_start = data->critical_start;
632 max_data->critical_end = data->critical_end;
634 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
635 max_data->pid = tsk->pid;
636 max_data->uid = task_uid(tsk);
637 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
638 max_data->policy = tsk->policy;
639 max_data->rt_priority = tsk->rt_priority;
641 /* record this tasks comm */
642 tracing_record_cmdline(tsk);
646 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
648 * @tsk: the task with the latency
649 * @cpu: The cpu that initiated the trace.
651 * Flip the buffers between the @tr and the max_tr and record information
652 * about which task was the cause of this latency.
655 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
657 struct ring_buffer *buf = tr->buffer;
659 if (trace_stop_count)
662 WARN_ON_ONCE(!irqs_disabled());
663 if (!current_trace->use_max_tr) {
667 arch_spin_lock(&ftrace_max_lock);
669 tr->buffer = max_tr.buffer;
672 __update_max_tr(tr, tsk, cpu);
673 arch_spin_unlock(&ftrace_max_lock);
677 * update_max_tr_single - only copy one trace over, and reset the rest
679 * @tsk - task with the latency
680 * @cpu - the cpu of the buffer to copy.
682 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
685 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
689 if (trace_stop_count)
692 WARN_ON_ONCE(!irqs_disabled());
693 if (!current_trace->use_max_tr) {
698 arch_spin_lock(&ftrace_max_lock);
700 ftrace_disable_cpu();
702 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
706 * We failed to swap the buffer due to a commit taking
707 * place on this CPU. We fail to record, but we reset
708 * the max trace buffer (no one writes directly to it)
709 * and flag that it failed.
711 trace_array_printk(&max_tr, _THIS_IP_,
712 "Failed to swap buffers due to commit in progress\n");
717 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
719 __update_max_tr(tr, tsk, cpu);
720 arch_spin_unlock(&ftrace_max_lock);
722 #endif /* CONFIG_TRACER_MAX_TRACE */
725 * register_tracer - register a tracer with the ftrace system.
726 * @type - the plugin for the tracer
728 * Register a new plugin tracer.
730 int register_tracer(struct tracer *type)
731 __releases(kernel_lock)
732 __acquires(kernel_lock)
738 pr_info("Tracer must have a name\n");
742 if (strlen(type->name) >= MAX_TRACER_SIZE) {
743 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
748 * When this gets called we hold the BKL which means that
749 * preemption is disabled. Various trace selftests however
750 * need to disable and enable preemption for successful tests.
751 * So we drop the BKL here and grab it after the tests again.
754 mutex_lock(&trace_types_lock);
756 tracing_selftest_running = true;
758 for (t = trace_types; t; t = t->next) {
759 if (strcmp(type->name, t->name) == 0) {
761 pr_info("Tracer %s already registered\n",
769 type->set_flag = &dummy_set_flag;
771 type->flags = &dummy_tracer_flags;
773 if (!type->flags->opts)
774 type->flags->opts = dummy_tracer_opt;
775 if (!type->wait_pipe)
776 type->wait_pipe = default_wait_pipe;
779 #ifdef CONFIG_FTRACE_STARTUP_TEST
780 if (type->selftest && !tracing_selftest_disabled) {
781 struct tracer *saved_tracer = current_trace;
782 struct trace_array *tr = &global_trace;
785 * Run a selftest on this tracer.
786 * Here we reset the trace buffer, and set the current
787 * tracer to be this tracer. The tracer can then run some
788 * internal tracing to verify that everything is in order.
789 * If we fail, we do not register this tracer.
791 tracing_reset_online_cpus(tr);
793 current_trace = type;
794 /* the test is responsible for initializing and enabling */
795 pr_info("Testing tracer %s: ", type->name);
796 ret = type->selftest(type, tr);
797 /* the test is responsible for resetting too */
798 current_trace = saved_tracer;
800 printk(KERN_CONT "FAILED!\n");
803 /* Only reset on passing, to avoid touching corrupted buffers */
804 tracing_reset_online_cpus(tr);
806 printk(KERN_CONT "PASSED\n");
810 type->next = trace_types;
814 tracing_selftest_running = false;
815 mutex_unlock(&trace_types_lock);
817 if (ret || !default_bootup_tracer)
820 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
823 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
824 /* Do we want this tracer to start on bootup? */
825 tracing_set_tracer(type->name);
826 default_bootup_tracer = NULL;
827 /* disable other selftests, since this will break it. */
828 tracing_selftest_disabled = 1;
829 #ifdef CONFIG_FTRACE_STARTUP_TEST
830 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
839 void unregister_tracer(struct tracer *type)
843 mutex_lock(&trace_types_lock);
844 for (t = &trace_types; *t; t = &(*t)->next) {
848 pr_info("Tracer %s not registered\n", type->name);
854 if (type == current_trace && tracer_enabled) {
857 if (current_trace->stop)
858 current_trace->stop(&global_trace);
859 current_trace = &nop_trace;
862 mutex_unlock(&trace_types_lock);
865 static void __tracing_reset(struct ring_buffer *buffer, int cpu)
867 ftrace_disable_cpu();
868 ring_buffer_reset_cpu(buffer, cpu);
872 void tracing_reset(struct trace_array *tr, int cpu)
874 struct ring_buffer *buffer = tr->buffer;
876 ring_buffer_record_disable(buffer);
878 /* Make sure all commits have finished */
880 __tracing_reset(buffer, cpu);
882 ring_buffer_record_enable(buffer);
885 void tracing_reset_online_cpus(struct trace_array *tr)
887 struct ring_buffer *buffer = tr->buffer;
890 ring_buffer_record_disable(buffer);
892 /* Make sure all commits have finished */
895 tr->time_start = ftrace_now(tr->cpu);
897 for_each_online_cpu(cpu)
898 __tracing_reset(buffer, cpu);
900 ring_buffer_record_enable(buffer);
903 void tracing_reset_current(int cpu)
905 tracing_reset(&global_trace, cpu);
908 void tracing_reset_current_online_cpus(void)
910 tracing_reset_online_cpus(&global_trace);
913 #define SAVED_CMDLINES 128
914 #define NO_CMDLINE_MAP UINT_MAX
915 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
916 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
917 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
918 static int cmdline_idx;
919 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
921 /* temporary disable recording */
922 static atomic_t trace_record_cmdline_disabled __read_mostly;
924 static void trace_init_cmdlines(void)
926 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
927 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
931 int is_tracing_stopped(void)
933 return trace_stop_count;
937 * ftrace_off_permanent - disable all ftrace code permanently
939 * This should only be called when a serious anomally has
940 * been detected. This will turn off the function tracing,
941 * ring buffers, and other tracing utilites. It takes no
942 * locks and can be called from any context.
944 void ftrace_off_permanent(void)
946 tracing_disabled = 1;
948 tracing_off_permanent();
952 * tracing_start - quick start of the tracer
954 * If tracing is enabled but was stopped by tracing_stop,
955 * this will start the tracer back up.
957 void tracing_start(void)
959 struct ring_buffer *buffer;
962 if (tracing_disabled)
965 spin_lock_irqsave(&tracing_start_lock, flags);
966 if (--trace_stop_count) {
967 if (trace_stop_count < 0) {
968 /* Someone screwed up their debugging */
970 trace_stop_count = 0;
975 /* Prevent the buffers from switching */
976 arch_spin_lock(&ftrace_max_lock);
978 buffer = global_trace.buffer;
980 ring_buffer_record_enable(buffer);
982 buffer = max_tr.buffer;
984 ring_buffer_record_enable(buffer);
986 arch_spin_unlock(&ftrace_max_lock);
990 spin_unlock_irqrestore(&tracing_start_lock, flags);
994 * tracing_stop - quick stop of the tracer
996 * Light weight way to stop tracing. Use in conjunction with
999 void tracing_stop(void)
1001 struct ring_buffer *buffer;
1002 unsigned long flags;
1005 spin_lock_irqsave(&tracing_start_lock, flags);
1006 if (trace_stop_count++)
1009 /* Prevent the buffers from switching */
1010 arch_spin_lock(&ftrace_max_lock);
1012 buffer = global_trace.buffer;
1014 ring_buffer_record_disable(buffer);
1016 buffer = max_tr.buffer;
1018 ring_buffer_record_disable(buffer);
1020 arch_spin_unlock(&ftrace_max_lock);
1023 spin_unlock_irqrestore(&tracing_start_lock, flags);
1026 void trace_stop_cmdline_recording(void);
1028 static void trace_save_cmdline(struct task_struct *tsk)
1032 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1036 * It's not the end of the world if we don't get
1037 * the lock, but we also don't want to spin
1038 * nor do we want to disable interrupts,
1039 * so if we miss here, then better luck next time.
1041 if (!arch_spin_trylock(&trace_cmdline_lock))
1044 idx = map_pid_to_cmdline[tsk->pid];
1045 if (idx == NO_CMDLINE_MAP) {
1046 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1049 * Check whether the cmdline buffer at idx has a pid
1050 * mapped. We are going to overwrite that entry so we
1051 * need to clear the map_pid_to_cmdline. Otherwise we
1052 * would read the new comm for the old pid.
1054 pid = map_cmdline_to_pid[idx];
1055 if (pid != NO_CMDLINE_MAP)
1056 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1058 map_cmdline_to_pid[idx] = tsk->pid;
1059 map_pid_to_cmdline[tsk->pid] = idx;
1064 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1066 arch_spin_unlock(&trace_cmdline_lock);
1069 void trace_find_cmdline(int pid, char comm[])
1074 strcpy(comm, "<idle>");
1078 if (WARN_ON_ONCE(pid < 0)) {
1079 strcpy(comm, "<XXX>");
1083 if (pid > PID_MAX_DEFAULT) {
1084 strcpy(comm, "<...>");
1089 arch_spin_lock(&trace_cmdline_lock);
1090 map = map_pid_to_cmdline[pid];
1091 if (map != NO_CMDLINE_MAP)
1092 strcpy(comm, saved_cmdlines[map]);
1094 strcpy(comm, "<...>");
1096 arch_spin_unlock(&trace_cmdline_lock);
1100 void tracing_record_cmdline(struct task_struct *tsk)
1102 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1106 trace_save_cmdline(tsk);
1110 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1113 struct task_struct *tsk = current;
1115 entry->preempt_count = pc & 0xff;
1116 entry->pid = (tsk) ? tsk->pid : 0;
1117 entry->lock_depth = (tsk) ? tsk->lock_depth : 0;
1119 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1120 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1122 TRACE_FLAG_IRQS_NOSUPPORT |
1124 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1125 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1126 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1128 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1130 struct ring_buffer_event *
1131 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1134 unsigned long flags, int pc)
1136 struct ring_buffer_event *event;
1138 event = ring_buffer_lock_reserve(buffer, len);
1139 if (event != NULL) {
1140 struct trace_entry *ent = ring_buffer_event_data(event);
1142 tracing_generic_entry_update(ent, flags, pc);
1150 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1151 struct ring_buffer_event *event,
1152 unsigned long flags, int pc,
1155 ring_buffer_unlock_commit(buffer, event);
1157 ftrace_trace_stack(buffer, flags, 6, pc);
1158 ftrace_trace_userstack(buffer, flags, pc);
1164 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1165 struct ring_buffer_event *event,
1166 unsigned long flags, int pc)
1168 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1171 struct ring_buffer_event *
1172 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1173 int type, unsigned long len,
1174 unsigned long flags, int pc)
1176 *current_rb = global_trace.buffer;
1177 return trace_buffer_lock_reserve(*current_rb,
1178 type, len, flags, pc);
1180 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1182 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1183 struct ring_buffer_event *event,
1184 unsigned long flags, int pc)
1186 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1188 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1190 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1191 struct ring_buffer_event *event,
1192 unsigned long flags, int pc)
1194 __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1196 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1198 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1199 struct ring_buffer_event *event)
1201 ring_buffer_discard_commit(buffer, event);
1203 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1206 trace_function(struct trace_array *tr,
1207 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1210 struct ftrace_event_call *call = &event_function;
1211 struct ring_buffer *buffer = tr->buffer;
1212 struct ring_buffer_event *event;
1213 struct ftrace_entry *entry;
1215 /* If we are reading the ring buffer, don't trace */
1216 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1219 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1223 entry = ring_buffer_event_data(event);
1225 entry->parent_ip = parent_ip;
1227 if (!filter_check_discard(call, entry, buffer, event))
1228 ring_buffer_unlock_commit(buffer, event);
1232 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1233 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1236 if (likely(!atomic_read(&data->disabled)))
1237 trace_function(tr, ip, parent_ip, flags, pc);
1240 #ifdef CONFIG_STACKTRACE
1241 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1242 unsigned long flags,
1245 struct ftrace_event_call *call = &event_kernel_stack;
1246 struct ring_buffer_event *event;
1247 struct stack_entry *entry;
1248 struct stack_trace trace;
1250 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1251 sizeof(*entry), flags, pc);
1254 entry = ring_buffer_event_data(event);
1255 memset(&entry->caller, 0, sizeof(entry->caller));
1257 trace.nr_entries = 0;
1258 trace.max_entries = FTRACE_STACK_ENTRIES;
1260 trace.entries = entry->caller;
1262 save_stack_trace(&trace);
1263 if (!filter_check_discard(call, entry, buffer, event))
1264 ring_buffer_unlock_commit(buffer, event);
1267 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1270 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1273 __ftrace_trace_stack(buffer, flags, skip, pc);
1276 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1279 __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1283 * trace_dump_stack - record a stack back trace in the trace buffer
1285 void trace_dump_stack(void)
1287 unsigned long flags;
1289 if (tracing_disabled || tracing_selftest_running)
1292 local_save_flags(flags);
1294 /* skipping 3 traces, seems to get us at the caller of this function */
1295 __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count());
1299 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1301 struct ftrace_event_call *call = &event_user_stack;
1302 struct ring_buffer_event *event;
1303 struct userstack_entry *entry;
1304 struct stack_trace trace;
1306 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1310 * NMIs can not handle page faults, even with fix ups.
1311 * The save user stack can (and often does) fault.
1313 if (unlikely(in_nmi()))
1316 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1317 sizeof(*entry), flags, pc);
1320 entry = ring_buffer_event_data(event);
1322 entry->tgid = current->tgid;
1323 memset(&entry->caller, 0, sizeof(entry->caller));
1325 trace.nr_entries = 0;
1326 trace.max_entries = FTRACE_STACK_ENTRIES;
1328 trace.entries = entry->caller;
1330 save_stack_trace_user(&trace);
1331 if (!filter_check_discard(call, entry, buffer, event))
1332 ring_buffer_unlock_commit(buffer, event);
1336 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1338 ftrace_trace_userstack(tr, flags, preempt_count());
1342 #endif /* CONFIG_STACKTRACE */
1345 ftrace_trace_special(void *__tr,
1346 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1349 struct ftrace_event_call *call = &event_special;
1350 struct ring_buffer_event *event;
1351 struct trace_array *tr = __tr;
1352 struct ring_buffer *buffer = tr->buffer;
1353 struct special_entry *entry;
1355 event = trace_buffer_lock_reserve(buffer, TRACE_SPECIAL,
1356 sizeof(*entry), 0, pc);
1359 entry = ring_buffer_event_data(event);
1364 if (!filter_check_discard(call, entry, buffer, event))
1365 trace_buffer_unlock_commit(buffer, event, 0, pc);
1369 __trace_special(void *__tr, void *__data,
1370 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1372 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1376 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1378 struct trace_array *tr = &global_trace;
1379 struct trace_array_cpu *data;
1380 unsigned long flags;
1384 if (tracing_disabled)
1387 pc = preempt_count();
1388 local_irq_save(flags);
1389 cpu = raw_smp_processor_id();
1390 data = tr->data[cpu];
1392 if (likely(atomic_inc_return(&data->disabled) == 1))
1393 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1395 atomic_dec(&data->disabled);
1396 local_irq_restore(flags);
1400 * trace_vbprintk - write binary msg to tracing buffer
1403 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1405 static arch_spinlock_t trace_buf_lock =
1406 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1407 static u32 trace_buf[TRACE_BUF_SIZE];
1409 struct ftrace_event_call *call = &event_bprint;
1410 struct ring_buffer_event *event;
1411 struct ring_buffer *buffer;
1412 struct trace_array *tr = &global_trace;
1413 struct trace_array_cpu *data;
1414 struct bprint_entry *entry;
1415 unsigned long flags;
1417 int cpu, len = 0, size, pc;
1419 if (unlikely(tracing_selftest_running || tracing_disabled))
1422 /* Don't pollute graph traces with trace_vprintk internals */
1423 pause_graph_tracing();
1425 pc = preempt_count();
1426 preempt_disable_notrace();
1427 cpu = raw_smp_processor_id();
1428 data = tr->data[cpu];
1430 disable = atomic_inc_return(&data->disabled);
1431 if (unlikely(disable != 1))
1434 /* Lockdep uses trace_printk for lock tracing */
1435 local_irq_save(flags);
1436 arch_spin_lock(&trace_buf_lock);
1437 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1439 if (len > TRACE_BUF_SIZE || len < 0)
1442 size = sizeof(*entry) + sizeof(u32) * len;
1443 buffer = tr->buffer;
1444 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1448 entry = ring_buffer_event_data(event);
1452 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1453 if (!filter_check_discard(call, entry, buffer, event)) {
1454 ring_buffer_unlock_commit(buffer, event);
1455 ftrace_trace_stack(buffer, flags, 6, pc);
1459 arch_spin_unlock(&trace_buf_lock);
1460 local_irq_restore(flags);
1463 atomic_dec_return(&data->disabled);
1464 preempt_enable_notrace();
1465 unpause_graph_tracing();
1469 EXPORT_SYMBOL_GPL(trace_vbprintk);
1471 int trace_array_printk(struct trace_array *tr,
1472 unsigned long ip, const char *fmt, ...)
1477 if (!(trace_flags & TRACE_ITER_PRINTK))
1481 ret = trace_array_vprintk(tr, ip, fmt, ap);
1486 int trace_array_vprintk(struct trace_array *tr,
1487 unsigned long ip, const char *fmt, va_list args)
1489 static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1490 static char trace_buf[TRACE_BUF_SIZE];
1492 struct ftrace_event_call *call = &event_print;
1493 struct ring_buffer_event *event;
1494 struct ring_buffer *buffer;
1495 struct trace_array_cpu *data;
1496 int cpu, len = 0, size, pc;
1497 struct print_entry *entry;
1498 unsigned long irq_flags;
1501 if (tracing_disabled || tracing_selftest_running)
1504 pc = preempt_count();
1505 preempt_disable_notrace();
1506 cpu = raw_smp_processor_id();
1507 data = tr->data[cpu];
1509 disable = atomic_inc_return(&data->disabled);
1510 if (unlikely(disable != 1))
1513 pause_graph_tracing();
1514 raw_local_irq_save(irq_flags);
1515 arch_spin_lock(&trace_buf_lock);
1516 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1518 size = sizeof(*entry) + len + 1;
1519 buffer = tr->buffer;
1520 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1524 entry = ring_buffer_event_data(event);
1527 memcpy(&entry->buf, trace_buf, len);
1528 entry->buf[len] = '\0';
1529 if (!filter_check_discard(call, entry, buffer, event)) {
1530 ring_buffer_unlock_commit(buffer, event);
1531 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1535 arch_spin_unlock(&trace_buf_lock);
1536 raw_local_irq_restore(irq_flags);
1537 unpause_graph_tracing();
1539 atomic_dec_return(&data->disabled);
1540 preempt_enable_notrace();
1545 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1547 return trace_array_vprintk(&global_trace, ip, fmt, args);
1549 EXPORT_SYMBOL_GPL(trace_vprintk);
1551 enum trace_file_type {
1552 TRACE_FILE_LAT_FMT = 1,
1553 TRACE_FILE_ANNOTATE = 2,
1556 static void trace_iterator_increment(struct trace_iterator *iter)
1558 /* Don't allow ftrace to trace into the ring buffers */
1559 ftrace_disable_cpu();
1562 if (iter->buffer_iter[iter->cpu])
1563 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1565 ftrace_enable_cpu();
1568 static struct trace_entry *
1569 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1570 unsigned long *lost_events)
1572 struct ring_buffer_event *event;
1573 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1575 /* Don't allow ftrace to trace into the ring buffers */
1576 ftrace_disable_cpu();
1579 event = ring_buffer_iter_peek(buf_iter, ts);
1581 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1584 ftrace_enable_cpu();
1586 return event ? ring_buffer_event_data(event) : NULL;
1589 static struct trace_entry *
1590 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1591 unsigned long *missing_events, u64 *ent_ts)
1593 struct ring_buffer *buffer = iter->tr->buffer;
1594 struct trace_entry *ent, *next = NULL;
1595 unsigned long lost_events = 0, next_lost = 0;
1596 int cpu_file = iter->cpu_file;
1597 u64 next_ts = 0, ts;
1602 * If we are in a per_cpu trace file, don't bother by iterating over
1603 * all cpu and peek directly.
1605 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1606 if (ring_buffer_empty_cpu(buffer, cpu_file))
1608 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1610 *ent_cpu = cpu_file;
1615 for_each_tracing_cpu(cpu) {
1617 if (ring_buffer_empty_cpu(buffer, cpu))
1620 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1623 * Pick the entry with the smallest timestamp:
1625 if (ent && (!next || ts < next_ts)) {
1629 next_lost = lost_events;
1634 *ent_cpu = next_cpu;
1640 *missing_events = next_lost;
1645 /* Find the next real entry, without updating the iterator itself */
1646 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1647 int *ent_cpu, u64 *ent_ts)
1649 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1652 /* Find the next real entry, and increment the iterator to the next entry */
1653 static void *find_next_entry_inc(struct trace_iterator *iter)
1655 iter->ent = __find_next_entry(iter, &iter->cpu,
1656 &iter->lost_events, &iter->ts);
1659 trace_iterator_increment(iter);
1661 return iter->ent ? iter : NULL;
1664 static void trace_consume(struct trace_iterator *iter)
1666 /* Don't allow ftrace to trace into the ring buffers */
1667 ftrace_disable_cpu();
1668 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1669 &iter->lost_events);
1670 ftrace_enable_cpu();
1673 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1675 struct trace_iterator *iter = m->private;
1679 WARN_ON_ONCE(iter->leftover);
1683 /* can't go backwards */
1688 ent = find_next_entry_inc(iter);
1692 while (ent && iter->idx < i)
1693 ent = find_next_entry_inc(iter);
1700 static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1702 struct trace_array *tr = iter->tr;
1703 struct ring_buffer_event *event;
1704 struct ring_buffer_iter *buf_iter;
1705 unsigned long entries = 0;
1708 tr->data[cpu]->skipped_entries = 0;
1710 if (!iter->buffer_iter[cpu])
1713 buf_iter = iter->buffer_iter[cpu];
1714 ring_buffer_iter_reset(buf_iter);
1717 * We could have the case with the max latency tracers
1718 * that a reset never took place on a cpu. This is evident
1719 * by the timestamp being before the start of the buffer.
1721 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1722 if (ts >= iter->tr->time_start)
1725 ring_buffer_read(buf_iter, NULL);
1728 tr->data[cpu]->skipped_entries = entries;
1732 * The current tracer is copied to avoid a global locking
1735 static void *s_start(struct seq_file *m, loff_t *pos)
1737 struct trace_iterator *iter = m->private;
1738 static struct tracer *old_tracer;
1739 int cpu_file = iter->cpu_file;
1744 /* copy the tracer to avoid using a global lock all around */
1745 mutex_lock(&trace_types_lock);
1746 if (unlikely(old_tracer != current_trace && current_trace)) {
1747 old_tracer = current_trace;
1748 *iter->trace = *current_trace;
1750 mutex_unlock(&trace_types_lock);
1752 atomic_inc(&trace_record_cmdline_disabled);
1754 if (*pos != iter->pos) {
1759 ftrace_disable_cpu();
1761 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1762 for_each_tracing_cpu(cpu)
1763 tracing_iter_reset(iter, cpu);
1765 tracing_iter_reset(iter, cpu_file);
1767 ftrace_enable_cpu();
1770 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1775 * If we overflowed the seq_file before, then we want
1776 * to just reuse the trace_seq buffer again.
1782 p = s_next(m, p, &l);
1786 trace_event_read_lock();
1787 trace_access_lock(cpu_file);
1791 static void s_stop(struct seq_file *m, void *p)
1793 struct trace_iterator *iter = m->private;
1795 atomic_dec(&trace_record_cmdline_disabled);
1796 trace_access_unlock(iter->cpu_file);
1797 trace_event_read_unlock();
1800 static void print_lat_help_header(struct seq_file *m)
1802 seq_puts(m, "# _------=> CPU# \n");
1803 seq_puts(m, "# / _-----=> irqs-off \n");
1804 seq_puts(m, "# | / _----=> need-resched \n");
1805 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1806 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1807 seq_puts(m, "# |||| /_--=> lock-depth \n");
1808 seq_puts(m, "# |||||/ delay \n");
1809 seq_puts(m, "# cmd pid |||||| time | caller \n");
1810 seq_puts(m, "# \\ / |||||| \\ | / \n");
1813 static void print_func_help_header(struct seq_file *m)
1815 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1816 seq_puts(m, "# | | | | |\n");
1821 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1823 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1824 struct trace_array *tr = iter->tr;
1825 struct trace_array_cpu *data = tr->data[tr->cpu];
1826 struct tracer *type = current_trace;
1827 unsigned long entries = 0;
1828 unsigned long total = 0;
1829 unsigned long count;
1830 const char *name = "preemption";
1837 for_each_tracing_cpu(cpu) {
1838 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1840 * If this buffer has skipped entries, then we hold all
1841 * entries for the trace and we need to ignore the
1842 * ones before the time stamp.
1844 if (tr->data[cpu]->skipped_entries) {
1845 count -= tr->data[cpu]->skipped_entries;
1846 /* total is the same as the entries */
1850 ring_buffer_overrun_cpu(tr->buffer, cpu);
1854 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1856 seq_puts(m, "# -----------------------------------"
1857 "---------------------------------\n");
1858 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1859 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1860 nsecs_to_usecs(data->saved_latency),
1864 #if defined(CONFIG_PREEMPT_NONE)
1866 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1868 #elif defined(CONFIG_PREEMPT)
1873 /* These are reserved for later use */
1876 seq_printf(m, " #P:%d)\n", num_online_cpus());
1880 seq_puts(m, "# -----------------\n");
1881 seq_printf(m, "# | task: %.16s-%d "
1882 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1883 data->comm, data->pid, data->uid, data->nice,
1884 data->policy, data->rt_priority);
1885 seq_puts(m, "# -----------------\n");
1887 if (data->critical_start) {
1888 seq_puts(m, "# => started at: ");
1889 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1890 trace_print_seq(m, &iter->seq);
1891 seq_puts(m, "\n# => ended at: ");
1892 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1893 trace_print_seq(m, &iter->seq);
1894 seq_puts(m, "\n#\n");
1900 static void test_cpu_buff_start(struct trace_iterator *iter)
1902 struct trace_seq *s = &iter->seq;
1904 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1907 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1910 if (cpumask_test_cpu(iter->cpu, iter->started))
1913 if (iter->tr->data[iter->cpu]->skipped_entries)
1916 cpumask_set_cpu(iter->cpu, iter->started);
1918 /* Don't print started cpu buffer for the first entry of the trace */
1920 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1924 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1926 struct trace_seq *s = &iter->seq;
1927 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1928 struct trace_entry *entry;
1929 struct trace_event *event;
1933 test_cpu_buff_start(iter);
1935 event = ftrace_find_event(entry->type);
1937 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1938 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1939 if (!trace_print_lat_context(iter))
1942 if (!trace_print_context(iter))
1948 return event->funcs->trace(iter, sym_flags, event);
1950 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1953 return TRACE_TYPE_HANDLED;
1955 return TRACE_TYPE_PARTIAL_LINE;
1958 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1960 struct trace_seq *s = &iter->seq;
1961 struct trace_entry *entry;
1962 struct trace_event *event;
1966 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1967 if (!trace_seq_printf(s, "%d %d %llu ",
1968 entry->pid, iter->cpu, iter->ts))
1972 event = ftrace_find_event(entry->type);
1974 return event->funcs->raw(iter, 0, event);
1976 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1979 return TRACE_TYPE_HANDLED;
1981 return TRACE_TYPE_PARTIAL_LINE;
1984 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1986 struct trace_seq *s = &iter->seq;
1987 unsigned char newline = '\n';
1988 struct trace_entry *entry;
1989 struct trace_event *event;
1993 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1994 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1995 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1996 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1999 event = ftrace_find_event(entry->type);
2001 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2002 if (ret != TRACE_TYPE_HANDLED)
2006 SEQ_PUT_FIELD_RET(s, newline);
2008 return TRACE_TYPE_HANDLED;
2011 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2013 struct trace_seq *s = &iter->seq;
2014 struct trace_entry *entry;
2015 struct trace_event *event;
2019 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2020 SEQ_PUT_FIELD_RET(s, entry->pid);
2021 SEQ_PUT_FIELD_RET(s, iter->cpu);
2022 SEQ_PUT_FIELD_RET(s, iter->ts);
2025 event = ftrace_find_event(entry->type);
2026 return event ? event->funcs->binary(iter, 0, event) :
2030 int trace_empty(struct trace_iterator *iter)
2034 /* If we are looking at one CPU buffer, only check that one */
2035 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
2036 cpu = iter->cpu_file;
2037 if (iter->buffer_iter[cpu]) {
2038 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2041 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2047 for_each_tracing_cpu(cpu) {
2048 if (iter->buffer_iter[cpu]) {
2049 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2052 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2060 /* Called with trace_event_read_lock() held. */
2061 static enum print_line_t print_trace_line(struct trace_iterator *iter)
2063 enum print_line_t ret;
2065 if (iter->lost_events)
2066 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2067 iter->cpu, iter->lost_events);
2069 if (iter->trace && iter->trace->print_line) {
2070 ret = iter->trace->print_line(iter);
2071 if (ret != TRACE_TYPE_UNHANDLED)
2075 if (iter->ent->type == TRACE_BPRINT &&
2076 trace_flags & TRACE_ITER_PRINTK &&
2077 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2078 return trace_print_bprintk_msg_only(iter);
2080 if (iter->ent->type == TRACE_PRINT &&
2081 trace_flags & TRACE_ITER_PRINTK &&
2082 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2083 return trace_print_printk_msg_only(iter);
2085 if (trace_flags & TRACE_ITER_BIN)
2086 return print_bin_fmt(iter);
2088 if (trace_flags & TRACE_ITER_HEX)
2089 return print_hex_fmt(iter);
2091 if (trace_flags & TRACE_ITER_RAW)
2092 return print_raw_fmt(iter);
2094 return print_trace_fmt(iter);
2097 void trace_default_header(struct seq_file *m)
2099 struct trace_iterator *iter = m->private;
2101 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2102 /* print nothing if the buffers are empty */
2103 if (trace_empty(iter))
2105 print_trace_header(m, iter);
2106 if (!(trace_flags & TRACE_ITER_VERBOSE))
2107 print_lat_help_header(m);
2109 if (!(trace_flags & TRACE_ITER_VERBOSE))
2110 print_func_help_header(m);
2114 static int s_show(struct seq_file *m, void *v)
2116 struct trace_iterator *iter = v;
2119 if (iter->ent == NULL) {
2121 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2124 if (iter->trace && iter->trace->print_header)
2125 iter->trace->print_header(m);
2127 trace_default_header(m);
2129 } else if (iter->leftover) {
2131 * If we filled the seq_file buffer earlier, we
2132 * want to just show it now.
2134 ret = trace_print_seq(m, &iter->seq);
2136 /* ret should this time be zero, but you never know */
2137 iter->leftover = ret;
2140 print_trace_line(iter);
2141 ret = trace_print_seq(m, &iter->seq);
2143 * If we overflow the seq_file buffer, then it will
2144 * ask us for this data again at start up.
2146 * ret is 0 if seq_file write succeeded.
2149 iter->leftover = ret;
2155 static const struct seq_operations tracer_seq_ops = {
2162 static struct trace_iterator *
2163 __tracing_open(struct inode *inode, struct file *file)
2165 long cpu_file = (long) inode->i_private;
2166 void *fail_ret = ERR_PTR(-ENOMEM);
2167 struct trace_iterator *iter;
2171 if (tracing_disabled)
2172 return ERR_PTR(-ENODEV);
2174 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2176 return ERR_PTR(-ENOMEM);
2179 * We make a copy of the current tracer to avoid concurrent
2180 * changes on it while we are reading.
2182 mutex_lock(&trace_types_lock);
2183 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2188 *iter->trace = *current_trace;
2190 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2193 if (current_trace && current_trace->print_max)
2196 iter->tr = &global_trace;
2198 mutex_init(&iter->mutex);
2199 iter->cpu_file = cpu_file;
2201 /* Notify the tracer early; before we stop tracing. */
2202 if (iter->trace && iter->trace->open)
2203 iter->trace->open(iter);
2205 /* Annotate start of buffers if we had overruns */
2206 if (ring_buffer_overruns(iter->tr->buffer))
2207 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2209 /* stop the trace while dumping */
2212 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2213 for_each_tracing_cpu(cpu) {
2214 iter->buffer_iter[cpu] =
2215 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2217 ring_buffer_read_prepare_sync();
2218 for_each_tracing_cpu(cpu) {
2219 ring_buffer_read_start(iter->buffer_iter[cpu]);
2220 tracing_iter_reset(iter, cpu);
2223 cpu = iter->cpu_file;
2224 iter->buffer_iter[cpu] =
2225 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2226 ring_buffer_read_prepare_sync();
2227 ring_buffer_read_start(iter->buffer_iter[cpu]);
2228 tracing_iter_reset(iter, cpu);
2231 ret = seq_open(file, &tracer_seq_ops);
2233 fail_ret = ERR_PTR(ret);
2237 m = file->private_data;
2240 mutex_unlock(&trace_types_lock);
2245 for_each_tracing_cpu(cpu) {
2246 if (iter->buffer_iter[cpu])
2247 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2249 free_cpumask_var(iter->started);
2252 mutex_unlock(&trace_types_lock);
2259 int tracing_open_generic(struct inode *inode, struct file *filp)
2261 if (tracing_disabled)
2264 filp->private_data = inode->i_private;
2268 static int tracing_release(struct inode *inode, struct file *file)
2270 struct seq_file *m = (struct seq_file *)file->private_data;
2271 struct trace_iterator *iter;
2274 if (!(file->f_mode & FMODE_READ))
2279 mutex_lock(&trace_types_lock);
2280 for_each_tracing_cpu(cpu) {
2281 if (iter->buffer_iter[cpu])
2282 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2285 if (iter->trace && iter->trace->close)
2286 iter->trace->close(iter);
2288 /* reenable tracing if it was previously enabled */
2290 mutex_unlock(&trace_types_lock);
2292 seq_release(inode, file);
2293 mutex_destroy(&iter->mutex);
2294 free_cpumask_var(iter->started);
2300 static int tracing_open(struct inode *inode, struct file *file)
2302 struct trace_iterator *iter;
2305 /* If this file was open for write, then erase contents */
2306 if ((file->f_mode & FMODE_WRITE) &&
2307 (file->f_flags & O_TRUNC)) {
2308 long cpu = (long) inode->i_private;
2310 if (cpu == TRACE_PIPE_ALL_CPU)
2311 tracing_reset_online_cpus(&global_trace);
2313 tracing_reset(&global_trace, cpu);
2316 if (file->f_mode & FMODE_READ) {
2317 iter = __tracing_open(inode, file);
2319 ret = PTR_ERR(iter);
2320 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2321 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2327 t_next(struct seq_file *m, void *v, loff_t *pos)
2329 struct tracer *t = v;
2339 static void *t_start(struct seq_file *m, loff_t *pos)
2344 mutex_lock(&trace_types_lock);
2345 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2351 static void t_stop(struct seq_file *m, void *p)
2353 mutex_unlock(&trace_types_lock);
2356 static int t_show(struct seq_file *m, void *v)
2358 struct tracer *t = v;
2363 seq_printf(m, "%s", t->name);
2372 static const struct seq_operations show_traces_seq_ops = {
2379 static int show_traces_open(struct inode *inode, struct file *file)
2381 if (tracing_disabled)
2384 return seq_open(file, &show_traces_seq_ops);
2388 tracing_write_stub(struct file *filp, const char __user *ubuf,
2389 size_t count, loff_t *ppos)
2394 static const struct file_operations tracing_fops = {
2395 .open = tracing_open,
2397 .write = tracing_write_stub,
2398 .llseek = seq_lseek,
2399 .release = tracing_release,
2402 static const struct file_operations show_traces_fops = {
2403 .open = show_traces_open,
2405 .release = seq_release,
2409 * Only trace on a CPU if the bitmask is set:
2411 static cpumask_var_t tracing_cpumask;
2414 * The tracer itself will not take this lock, but still we want
2415 * to provide a consistent cpumask to user-space:
2417 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2420 * Temporary storage for the character representation of the
2421 * CPU bitmask (and one more byte for the newline):
2423 static char mask_str[NR_CPUS + 1];
2426 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2427 size_t count, loff_t *ppos)
2431 mutex_lock(&tracing_cpumask_update_lock);
2433 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2434 if (count - len < 2) {
2438 len += sprintf(mask_str + len, "\n");
2439 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2442 mutex_unlock(&tracing_cpumask_update_lock);
2448 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2449 size_t count, loff_t *ppos)
2452 cpumask_var_t tracing_cpumask_new;
2454 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2457 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2461 mutex_lock(&tracing_cpumask_update_lock);
2463 local_irq_disable();
2464 arch_spin_lock(&ftrace_max_lock);
2465 for_each_tracing_cpu(cpu) {
2467 * Increase/decrease the disabled counter if we are
2468 * about to flip a bit in the cpumask:
2470 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2471 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2472 atomic_inc(&global_trace.data[cpu]->disabled);
2474 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2475 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2476 atomic_dec(&global_trace.data[cpu]->disabled);
2479 arch_spin_unlock(&ftrace_max_lock);
2482 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2484 mutex_unlock(&tracing_cpumask_update_lock);
2485 free_cpumask_var(tracing_cpumask_new);
2490 free_cpumask_var(tracing_cpumask_new);
2495 static const struct file_operations tracing_cpumask_fops = {
2496 .open = tracing_open_generic,
2497 .read = tracing_cpumask_read,
2498 .write = tracing_cpumask_write,
2501 static int tracing_trace_options_show(struct seq_file *m, void *v)
2503 struct tracer_opt *trace_opts;
2507 mutex_lock(&trace_types_lock);
2508 tracer_flags = current_trace->flags->val;
2509 trace_opts = current_trace->flags->opts;
2511 for (i = 0; trace_options[i]; i++) {
2512 if (trace_flags & (1 << i))
2513 seq_printf(m, "%s\n", trace_options[i]);
2515 seq_printf(m, "no%s\n", trace_options[i]);
2518 for (i = 0; trace_opts[i].name; i++) {
2519 if (tracer_flags & trace_opts[i].bit)
2520 seq_printf(m, "%s\n", trace_opts[i].name);
2522 seq_printf(m, "no%s\n", trace_opts[i].name);
2524 mutex_unlock(&trace_types_lock);
2529 static int __set_tracer_option(struct tracer *trace,
2530 struct tracer_flags *tracer_flags,
2531 struct tracer_opt *opts, int neg)
2535 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2540 tracer_flags->val &= ~opts->bit;
2542 tracer_flags->val |= opts->bit;
2546 /* Try to assign a tracer specific option */
2547 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2549 struct tracer_flags *tracer_flags = trace->flags;
2550 struct tracer_opt *opts = NULL;
2553 for (i = 0; tracer_flags->opts[i].name; i++) {
2554 opts = &tracer_flags->opts[i];
2556 if (strcmp(cmp, opts->name) == 0)
2557 return __set_tracer_option(trace, trace->flags,
2564 static void set_tracer_flags(unsigned int mask, int enabled)
2566 /* do nothing if flag is already set */
2567 if (!!(trace_flags & mask) == !!enabled)
2571 trace_flags |= mask;
2573 trace_flags &= ~mask;
2575 if (mask == TRACE_ITER_RECORD_CMD)
2576 trace_event_enable_cmd_record(enabled);
2580 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2581 size_t cnt, loff_t *ppos)
2589 if (cnt >= sizeof(buf))
2592 if (copy_from_user(&buf, ubuf, cnt))
2596 cmp = strstrip(buf);
2598 if (strncmp(cmp, "no", 2) == 0) {
2603 for (i = 0; trace_options[i]; i++) {
2604 if (strcmp(cmp, trace_options[i]) == 0) {
2605 set_tracer_flags(1 << i, !neg);
2610 /* If no option could be set, test the specific tracer options */
2611 if (!trace_options[i]) {
2612 mutex_lock(&trace_types_lock);
2613 ret = set_tracer_option(current_trace, cmp, neg);
2614 mutex_unlock(&trace_types_lock);
2624 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2626 if (tracing_disabled)
2628 return single_open(file, tracing_trace_options_show, NULL);
2631 static const struct file_operations tracing_iter_fops = {
2632 .open = tracing_trace_options_open,
2634 .llseek = seq_lseek,
2635 .release = single_release,
2636 .write = tracing_trace_options_write,
2639 static const char readme_msg[] =
2640 "tracing mini-HOWTO:\n\n"
2641 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2642 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2643 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2644 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2646 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2647 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2649 "# cat /sys/kernel/debug/tracing/trace_options\n"
2650 "noprint-parent nosym-offset nosym-addr noverbose\n"
2651 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2652 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2653 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2654 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2658 tracing_readme_read(struct file *filp, char __user *ubuf,
2659 size_t cnt, loff_t *ppos)
2661 return simple_read_from_buffer(ubuf, cnt, ppos,
2662 readme_msg, strlen(readme_msg));
2665 static const struct file_operations tracing_readme_fops = {
2666 .open = tracing_open_generic,
2667 .read = tracing_readme_read,
2671 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2672 size_t cnt, loff_t *ppos)
2681 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2685 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2693 for (i = 0; i < SAVED_CMDLINES; i++) {
2696 pid = map_cmdline_to_pid[i];
2697 if (pid == -1 || pid == NO_CMDLINE_MAP)
2700 trace_find_cmdline(pid, buf_comm);
2701 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2706 len = simple_read_from_buffer(ubuf, cnt, ppos,
2715 static const struct file_operations tracing_saved_cmdlines_fops = {
2716 .open = tracing_open_generic,
2717 .read = tracing_saved_cmdlines_read,
2721 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2722 size_t cnt, loff_t *ppos)
2727 r = sprintf(buf, "%u\n", tracer_enabled);
2728 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2732 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2733 size_t cnt, loff_t *ppos)
2735 struct trace_array *tr = filp->private_data;
2740 if (cnt >= sizeof(buf))
2743 if (copy_from_user(&buf, ubuf, cnt))
2748 ret = strict_strtoul(buf, 10, &val);
2754 mutex_lock(&trace_types_lock);
2755 if (tracer_enabled ^ val) {
2758 if (current_trace->start)
2759 current_trace->start(tr);
2764 if (current_trace->stop)
2765 current_trace->stop(tr);
2768 mutex_unlock(&trace_types_lock);
2776 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2777 size_t cnt, loff_t *ppos)
2779 char buf[MAX_TRACER_SIZE+2];
2782 mutex_lock(&trace_types_lock);
2784 r = sprintf(buf, "%s\n", current_trace->name);
2786 r = sprintf(buf, "\n");
2787 mutex_unlock(&trace_types_lock);
2789 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2792 int tracer_init(struct tracer *t, struct trace_array *tr)
2794 tracing_reset_online_cpus(tr);
2798 static int tracing_resize_ring_buffer(unsigned long size)
2803 * If kernel or user changes the size of the ring buffer
2804 * we use the size that was given, and we can forget about
2805 * expanding it later.
2807 ring_buffer_expanded = 1;
2809 ret = ring_buffer_resize(global_trace.buffer, size);
2813 if (!current_trace->use_max_tr)
2816 ret = ring_buffer_resize(max_tr.buffer, size);
2820 r = ring_buffer_resize(global_trace.buffer,
2821 global_trace.entries);
2824 * AARGH! We are left with different
2825 * size max buffer!!!!
2826 * The max buffer is our "snapshot" buffer.
2827 * When a tracer needs a snapshot (one of the
2828 * latency tracers), it swaps the max buffer
2829 * with the saved snap shot. We succeeded to
2830 * update the size of the main buffer, but failed to
2831 * update the size of the max buffer. But when we tried
2832 * to reset the main buffer to the original size, we
2833 * failed there too. This is very unlikely to
2834 * happen, but if it does, warn and kill all
2838 tracing_disabled = 1;
2843 max_tr.entries = size;
2845 global_trace.entries = size;
2852 * tracing_update_buffers - used by tracing facility to expand ring buffers
2854 * To save on memory when the tracing is never used on a system with it
2855 * configured in. The ring buffers are set to a minimum size. But once
2856 * a user starts to use the tracing facility, then they need to grow
2857 * to their default size.
2859 * This function is to be called when a tracer is about to be used.
2861 int tracing_update_buffers(void)
2865 mutex_lock(&trace_types_lock);
2866 if (!ring_buffer_expanded)
2867 ret = tracing_resize_ring_buffer(trace_buf_size);
2868 mutex_unlock(&trace_types_lock);
2873 struct trace_option_dentry;
2875 static struct trace_option_dentry *
2876 create_trace_option_files(struct tracer *tracer);
2879 destroy_trace_option_files(struct trace_option_dentry *topts);
2881 static int tracing_set_tracer(const char *buf)
2883 static struct trace_option_dentry *topts;
2884 struct trace_array *tr = &global_trace;
2888 mutex_lock(&trace_types_lock);
2890 if (!ring_buffer_expanded) {
2891 ret = tracing_resize_ring_buffer(trace_buf_size);
2897 for (t = trace_types; t; t = t->next) {
2898 if (strcmp(t->name, buf) == 0)
2905 if (t == current_trace)
2908 trace_branch_disable();
2909 if (current_trace && current_trace->reset)
2910 current_trace->reset(tr);
2911 if (current_trace && current_trace->use_max_tr) {
2913 * We don't free the ring buffer. instead, resize it because
2914 * The max_tr ring buffer has some state (e.g. ring->clock) and
2915 * we want preserve it.
2917 ring_buffer_resize(max_tr.buffer, 1);
2920 destroy_trace_option_files(topts);
2924 topts = create_trace_option_files(current_trace);
2925 if (current_trace->use_max_tr) {
2926 ret = ring_buffer_resize(max_tr.buffer, global_trace.entries);
2929 max_tr.entries = global_trace.entries;
2933 ret = tracer_init(t, tr);
2938 trace_branch_enable(tr);
2940 mutex_unlock(&trace_types_lock);
2946 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2947 size_t cnt, loff_t *ppos)
2949 char buf[MAX_TRACER_SIZE+1];
2956 if (cnt > MAX_TRACER_SIZE)
2957 cnt = MAX_TRACER_SIZE;
2959 if (copy_from_user(&buf, ubuf, cnt))
2964 /* strip ending whitespace. */
2965 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2968 err = tracing_set_tracer(buf);
2978 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2979 size_t cnt, loff_t *ppos)
2981 unsigned long *ptr = filp->private_data;
2985 r = snprintf(buf, sizeof(buf), "%ld\n",
2986 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2987 if (r > sizeof(buf))
2989 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2993 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2994 size_t cnt, loff_t *ppos)
2996 unsigned long *ptr = filp->private_data;
3001 if (cnt >= sizeof(buf))
3004 if (copy_from_user(&buf, ubuf, cnt))
3009 ret = strict_strtoul(buf, 10, &val);
3018 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3020 long cpu_file = (long) inode->i_private;
3021 struct trace_iterator *iter;
3024 if (tracing_disabled)
3027 mutex_lock(&trace_types_lock);
3029 /* create a buffer to store the information to pass to userspace */
3030 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3037 * We make a copy of the current tracer to avoid concurrent
3038 * changes on it while we are reading.
3040 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3046 *iter->trace = *current_trace;
3048 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3053 /* trace pipe does not show start of buffer */
3054 cpumask_setall(iter->started);
3056 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3057 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3059 iter->cpu_file = cpu_file;
3060 iter->tr = &global_trace;
3061 mutex_init(&iter->mutex);
3062 filp->private_data = iter;
3064 if (iter->trace->pipe_open)
3065 iter->trace->pipe_open(iter);
3068 mutex_unlock(&trace_types_lock);
3074 mutex_unlock(&trace_types_lock);
3078 static int tracing_release_pipe(struct inode *inode, struct file *file)
3080 struct trace_iterator *iter = file->private_data;
3082 mutex_lock(&trace_types_lock);
3084 if (iter->trace->pipe_close)
3085 iter->trace->pipe_close(iter);
3087 mutex_unlock(&trace_types_lock);
3089 free_cpumask_var(iter->started);
3090 mutex_destroy(&iter->mutex);
3098 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3100 struct trace_iterator *iter = filp->private_data;
3102 if (trace_flags & TRACE_ITER_BLOCK) {
3104 * Always select as readable when in blocking mode
3106 return POLLIN | POLLRDNORM;
3108 if (!trace_empty(iter))
3109 return POLLIN | POLLRDNORM;
3110 poll_wait(filp, &trace_wait, poll_table);
3111 if (!trace_empty(iter))
3112 return POLLIN | POLLRDNORM;
3119 void default_wait_pipe(struct trace_iterator *iter)
3123 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3125 if (trace_empty(iter))
3128 finish_wait(&trace_wait, &wait);
3132 * This is a make-shift waitqueue.
3133 * A tracer might use this callback on some rare cases:
3135 * 1) the current tracer might hold the runqueue lock when it wakes up
3136 * a reader, hence a deadlock (sched, function, and function graph tracers)
3137 * 2) the function tracers, trace all functions, we don't want
3138 * the overhead of calling wake_up and friends
3139 * (and tracing them too)
3141 * Anyway, this is really very primitive wakeup.
3143 void poll_wait_pipe(struct trace_iterator *iter)
3145 set_current_state(TASK_INTERRUPTIBLE);
3146 /* sleep for 100 msecs, and try again. */
3147 schedule_timeout(HZ / 10);
3150 /* Must be called with trace_types_lock mutex held. */
3151 static int tracing_wait_pipe(struct file *filp)
3153 struct trace_iterator *iter = filp->private_data;
3155 while (trace_empty(iter)) {
3157 if ((filp->f_flags & O_NONBLOCK)) {
3161 mutex_unlock(&iter->mutex);
3163 iter->trace->wait_pipe(iter);
3165 mutex_lock(&iter->mutex);
3167 if (signal_pending(current))
3171 * We block until we read something and tracing is disabled.
3172 * We still block if tracing is disabled, but we have never
3173 * read anything. This allows a user to cat this file, and
3174 * then enable tracing. But after we have read something,
3175 * we give an EOF when tracing is again disabled.
3177 * iter->pos will be 0 if we haven't read anything.
3179 if (!tracer_enabled && iter->pos)
3190 tracing_read_pipe(struct file *filp, char __user *ubuf,
3191 size_t cnt, loff_t *ppos)
3193 struct trace_iterator *iter = filp->private_data;
3194 static struct tracer *old_tracer;
3197 /* return any leftover data */
3198 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3202 trace_seq_init(&iter->seq);
3204 /* copy the tracer to avoid using a global lock all around */
3205 mutex_lock(&trace_types_lock);
3206 if (unlikely(old_tracer != current_trace && current_trace)) {
3207 old_tracer = current_trace;
3208 *iter->trace = *current_trace;
3210 mutex_unlock(&trace_types_lock);
3213 * Avoid more than one consumer on a single file descriptor
3214 * This is just a matter of traces coherency, the ring buffer itself
3217 mutex_lock(&iter->mutex);
3218 if (iter->trace->read) {
3219 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3225 sret = tracing_wait_pipe(filp);
3229 /* stop when tracing is finished */
3230 if (trace_empty(iter)) {
3235 if (cnt >= PAGE_SIZE)
3236 cnt = PAGE_SIZE - 1;
3238 /* reset all but tr, trace, and overruns */
3239 memset(&iter->seq, 0,
3240 sizeof(struct trace_iterator) -
3241 offsetof(struct trace_iterator, seq));
3244 trace_event_read_lock();
3245 trace_access_lock(iter->cpu_file);
3246 while (find_next_entry_inc(iter) != NULL) {
3247 enum print_line_t ret;
3248 int len = iter->seq.len;
3250 ret = print_trace_line(iter);
3251 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3252 /* don't print partial lines */
3253 iter->seq.len = len;
3256 if (ret != TRACE_TYPE_NO_CONSUME)
3257 trace_consume(iter);
3259 if (iter->seq.len >= cnt)
3262 trace_access_unlock(iter->cpu_file);
3263 trace_event_read_unlock();
3265 /* Now copy what we have to the user */
3266 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3267 if (iter->seq.readpos >= iter->seq.len)
3268 trace_seq_init(&iter->seq);
3271 * If there was nothing to send to user, inspite of consuming trace
3272 * entries, go back to wait for more entries.
3278 mutex_unlock(&iter->mutex);
3283 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3284 struct pipe_buffer *buf)
3286 __free_page(buf->page);
3289 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3292 __free_page(spd->pages[idx]);
3295 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3297 .map = generic_pipe_buf_map,
3298 .unmap = generic_pipe_buf_unmap,
3299 .confirm = generic_pipe_buf_confirm,
3300 .release = tracing_pipe_buf_release,
3301 .steal = generic_pipe_buf_steal,
3302 .get = generic_pipe_buf_get,
3306 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3311 /* Seq buffer is page-sized, exactly what we need. */
3313 count = iter->seq.len;
3314 ret = print_trace_line(iter);
3315 count = iter->seq.len - count;
3318 iter->seq.len -= count;
3321 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3322 iter->seq.len -= count;
3326 if (ret != TRACE_TYPE_NO_CONSUME)
3327 trace_consume(iter);
3329 if (!find_next_entry_inc(iter)) {
3339 static ssize_t tracing_splice_read_pipe(struct file *filp,
3341 struct pipe_inode_info *pipe,
3345 struct page *pages_def[PIPE_DEF_BUFFERS];
3346 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3347 struct trace_iterator *iter = filp->private_data;
3348 struct splice_pipe_desc spd = {
3350 .partial = partial_def,
3351 .nr_pages = 0, /* This gets updated below. */
3353 .ops = &tracing_pipe_buf_ops,
3354 .spd_release = tracing_spd_release_pipe,
3356 static struct tracer *old_tracer;
3361 if (splice_grow_spd(pipe, &spd))
3364 /* copy the tracer to avoid using a global lock all around */
3365 mutex_lock(&trace_types_lock);
3366 if (unlikely(old_tracer != current_trace && current_trace)) {
3367 old_tracer = current_trace;
3368 *iter->trace = *current_trace;
3370 mutex_unlock(&trace_types_lock);
3372 mutex_lock(&iter->mutex);
3374 if (iter->trace->splice_read) {
3375 ret = iter->trace->splice_read(iter, filp,
3376 ppos, pipe, len, flags);
3381 ret = tracing_wait_pipe(filp);
3385 if (!iter->ent && !find_next_entry_inc(iter)) {
3390 trace_event_read_lock();
3391 trace_access_lock(iter->cpu_file);
3393 /* Fill as many pages as possible. */
3394 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3395 spd.pages[i] = alloc_page(GFP_KERNEL);
3399 rem = tracing_fill_pipe_page(rem, iter);
3401 /* Copy the data into the page, so we can start over. */
3402 ret = trace_seq_to_buffer(&iter->seq,
3403 page_address(spd.pages[i]),
3406 __free_page(spd.pages[i]);
3409 spd.partial[i].offset = 0;
3410 spd.partial[i].len = iter->seq.len;
3412 trace_seq_init(&iter->seq);
3415 trace_access_unlock(iter->cpu_file);
3416 trace_event_read_unlock();
3417 mutex_unlock(&iter->mutex);
3421 ret = splice_to_pipe(pipe, &spd);
3423 splice_shrink_spd(pipe, &spd);
3427 mutex_unlock(&iter->mutex);
3432 tracing_entries_read(struct file *filp, char __user *ubuf,
3433 size_t cnt, loff_t *ppos)
3435 struct trace_array *tr = filp->private_data;
3439 mutex_lock(&trace_types_lock);
3440 if (!ring_buffer_expanded)
3441 r = sprintf(buf, "%lu (expanded: %lu)\n",
3443 trace_buf_size >> 10);
3445 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3446 mutex_unlock(&trace_types_lock);
3448 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3452 tracing_entries_write(struct file *filp, const char __user *ubuf,
3453 size_t cnt, loff_t *ppos)
3459 if (cnt >= sizeof(buf))
3462 if (copy_from_user(&buf, ubuf, cnt))
3467 ret = strict_strtoul(buf, 10, &val);
3471 /* must have at least 1 entry */
3475 mutex_lock(&trace_types_lock);
3479 /* disable all cpu buffers */
3480 for_each_tracing_cpu(cpu) {
3481 if (global_trace.data[cpu])
3482 atomic_inc(&global_trace.data[cpu]->disabled);
3483 if (max_tr.data[cpu])
3484 atomic_inc(&max_tr.data[cpu]->disabled);
3487 /* value is in KB */
3490 if (val != global_trace.entries) {
3491 ret = tracing_resize_ring_buffer(val);
3500 /* If check pages failed, return ENOMEM */
3501 if (tracing_disabled)
3504 for_each_tracing_cpu(cpu) {
3505 if (global_trace.data[cpu])
3506 atomic_dec(&global_trace.data[cpu]->disabled);
3507 if (max_tr.data[cpu])
3508 atomic_dec(&max_tr.data[cpu]->disabled);
3512 mutex_unlock(&trace_types_lock);
3517 static int mark_printk(const char *fmt, ...)
3521 va_start(args, fmt);
3522 ret = trace_vprintk(0, fmt, args);
3528 tracing_mark_write(struct file *filp, const char __user *ubuf,
3529 size_t cnt, loff_t *fpos)
3533 if (tracing_disabled)
3536 if (cnt > TRACE_BUF_SIZE)
3537 cnt = TRACE_BUF_SIZE;
3539 buf = kmalloc(cnt + 2, GFP_KERNEL);
3543 if (copy_from_user(buf, ubuf, cnt)) {
3547 if (buf[cnt-1] != '\n') {
3553 cnt = mark_printk("%s", buf);
3560 static int tracing_clock_show(struct seq_file *m, void *v)
3564 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3566 "%s%s%s%s", i ? " " : "",
3567 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3568 i == trace_clock_id ? "]" : "");
3574 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3575 size_t cnt, loff_t *fpos)
3578 const char *clockstr;
3581 if (cnt >= sizeof(buf))
3584 if (copy_from_user(&buf, ubuf, cnt))
3589 clockstr = strstrip(buf);
3591 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3592 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3595 if (i == ARRAY_SIZE(trace_clocks))
3600 mutex_lock(&trace_types_lock);
3602 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3604 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3606 mutex_unlock(&trace_types_lock);
3613 static int tracing_clock_open(struct inode *inode, struct file *file)
3615 if (tracing_disabled)
3617 return single_open(file, tracing_clock_show, NULL);
3620 static const struct file_operations tracing_max_lat_fops = {
3621 .open = tracing_open_generic,
3622 .read = tracing_max_lat_read,
3623 .write = tracing_max_lat_write,
3626 static const struct file_operations tracing_ctrl_fops = {
3627 .open = tracing_open_generic,
3628 .read = tracing_ctrl_read,
3629 .write = tracing_ctrl_write,
3632 static const struct file_operations set_tracer_fops = {
3633 .open = tracing_open_generic,
3634 .read = tracing_set_trace_read,
3635 .write = tracing_set_trace_write,
3638 static const struct file_operations tracing_pipe_fops = {
3639 .open = tracing_open_pipe,
3640 .poll = tracing_poll_pipe,
3641 .read = tracing_read_pipe,
3642 .splice_read = tracing_splice_read_pipe,
3643 .release = tracing_release_pipe,
3646 static const struct file_operations tracing_entries_fops = {
3647 .open = tracing_open_generic,
3648 .read = tracing_entries_read,
3649 .write = tracing_entries_write,
3652 static const struct file_operations tracing_mark_fops = {
3653 .open = tracing_open_generic,
3654 .write = tracing_mark_write,
3657 static const struct file_operations trace_clock_fops = {
3658 .open = tracing_clock_open,
3660 .llseek = seq_lseek,
3661 .release = single_release,
3662 .write = tracing_clock_write,
3665 struct ftrace_buffer_info {
3666 struct trace_array *tr;
3672 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3674 int cpu = (int)(long)inode->i_private;
3675 struct ftrace_buffer_info *info;
3677 if (tracing_disabled)
3680 info = kzalloc(sizeof(*info), GFP_KERNEL);
3684 info->tr = &global_trace;
3687 /* Force reading ring buffer for first read */
3688 info->read = (unsigned int)-1;
3690 filp->private_data = info;
3692 return nonseekable_open(inode, filp);
3696 tracing_buffers_read(struct file *filp, char __user *ubuf,
3697 size_t count, loff_t *ppos)
3699 struct ftrace_buffer_info *info = filp->private_data;
3707 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3711 /* Do we have previous read data to read? */
3712 if (info->read < PAGE_SIZE)
3717 trace_access_lock(info->cpu);
3718 ret = ring_buffer_read_page(info->tr->buffer,
3722 trace_access_unlock(info->cpu);
3727 size = PAGE_SIZE - info->read;
3731 ret = copy_to_user(ubuf, info->spare + info->read, size);
3742 static int tracing_buffers_release(struct inode *inode, struct file *file)
3744 struct ftrace_buffer_info *info = file->private_data;
3747 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3754 struct ring_buffer *buffer;
3759 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3760 struct pipe_buffer *buf)
3762 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3767 ring_buffer_free_read_page(ref->buffer, ref->page);
3772 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3773 struct pipe_buffer *buf)
3778 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3779 struct pipe_buffer *buf)
3781 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3786 /* Pipe buffer operations for a buffer. */
3787 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
3789 .map = generic_pipe_buf_map,
3790 .unmap = generic_pipe_buf_unmap,
3791 .confirm = generic_pipe_buf_confirm,
3792 .release = buffer_pipe_buf_release,
3793 .steal = buffer_pipe_buf_steal,
3794 .get = buffer_pipe_buf_get,
3798 * Callback from splice_to_pipe(), if we need to release some pages
3799 * at the end of the spd in case we error'ed out in filling the pipe.
3801 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3803 struct buffer_ref *ref =
3804 (struct buffer_ref *)spd->partial[i].private;
3809 ring_buffer_free_read_page(ref->buffer, ref->page);
3811 spd->partial[i].private = 0;
3815 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3816 struct pipe_inode_info *pipe, size_t len,
3819 struct ftrace_buffer_info *info = file->private_data;
3820 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3821 struct page *pages_def[PIPE_DEF_BUFFERS];
3822 struct splice_pipe_desc spd = {
3824 .partial = partial_def,
3826 .ops = &buffer_pipe_buf_ops,
3827 .spd_release = buffer_spd_release,
3829 struct buffer_ref *ref;
3830 int entries, size, i;
3833 if (splice_grow_spd(pipe, &spd))
3836 if (*ppos & (PAGE_SIZE - 1)) {
3837 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3842 if (len & (PAGE_SIZE - 1)) {
3843 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3844 if (len < PAGE_SIZE) {
3851 trace_access_lock(info->cpu);
3852 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3854 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
3858 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3863 ref->buffer = info->tr->buffer;
3864 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3870 r = ring_buffer_read_page(ref->buffer, &ref->page,
3873 ring_buffer_free_read_page(ref->buffer,
3880 * zero out any left over data, this is going to
3883 size = ring_buffer_page_len(ref->page);
3884 if (size < PAGE_SIZE)
3885 memset(ref->page + size, 0, PAGE_SIZE - size);
3887 page = virt_to_page(ref->page);
3889 spd.pages[i] = page;
3890 spd.partial[i].len = PAGE_SIZE;
3891 spd.partial[i].offset = 0;
3892 spd.partial[i].private = (unsigned long)ref;
3896 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3899 trace_access_unlock(info->cpu);
3902 /* did we read anything? */
3903 if (!spd.nr_pages) {
3904 if (flags & SPLICE_F_NONBLOCK)
3912 ret = splice_to_pipe(pipe, &spd);
3913 splice_shrink_spd(pipe, &spd);
3918 static const struct file_operations tracing_buffers_fops = {
3919 .open = tracing_buffers_open,
3920 .read = tracing_buffers_read,
3921 .release = tracing_buffers_release,
3922 .splice_read = tracing_buffers_splice_read,
3923 .llseek = no_llseek,
3927 tracing_stats_read(struct file *filp, char __user *ubuf,
3928 size_t count, loff_t *ppos)
3930 unsigned long cpu = (unsigned long)filp->private_data;
3931 struct trace_array *tr = &global_trace;
3932 struct trace_seq *s;
3935 s = kmalloc(sizeof(*s), GFP_KERNEL);
3941 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3942 trace_seq_printf(s, "entries: %ld\n", cnt);
3944 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3945 trace_seq_printf(s, "overrun: %ld\n", cnt);
3947 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3948 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3950 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3957 static const struct file_operations tracing_stats_fops = {
3958 .open = tracing_open_generic,
3959 .read = tracing_stats_read,
3962 #ifdef CONFIG_DYNAMIC_FTRACE
3964 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3970 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3971 size_t cnt, loff_t *ppos)
3973 static char ftrace_dyn_info_buffer[1024];
3974 static DEFINE_MUTEX(dyn_info_mutex);
3975 unsigned long *p = filp->private_data;
3976 char *buf = ftrace_dyn_info_buffer;
3977 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3980 mutex_lock(&dyn_info_mutex);
3981 r = sprintf(buf, "%ld ", *p);
3983 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3986 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3988 mutex_unlock(&dyn_info_mutex);
3993 static const struct file_operations tracing_dyn_info_fops = {
3994 .open = tracing_open_generic,
3995 .read = tracing_read_dyn_info,
3999 static struct dentry *d_tracer;
4001 struct dentry *tracing_init_dentry(void)
4008 if (!debugfs_initialized())
4011 d_tracer = debugfs_create_dir("tracing", NULL);
4013 if (!d_tracer && !once) {
4015 pr_warning("Could not create debugfs directory 'tracing'\n");
4022 static struct dentry *d_percpu;
4024 struct dentry *tracing_dentry_percpu(void)
4027 struct dentry *d_tracer;
4032 d_tracer = tracing_init_dentry();
4037 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
4039 if (!d_percpu && !once) {
4041 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4048 static void tracing_init_debugfs_percpu(long cpu)
4050 struct dentry *d_percpu = tracing_dentry_percpu();
4051 struct dentry *d_cpu;
4052 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
4055 if (cpu > 999 || cpu < 0)
4058 sprintf(cpu_dir, "cpu%ld", cpu);
4059 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4061 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4065 /* per cpu trace_pipe */
4066 trace_create_file("trace_pipe", 0444, d_cpu,
4067 (void *) cpu, &tracing_pipe_fops);
4070 trace_create_file("trace", 0644, d_cpu,
4071 (void *) cpu, &tracing_fops);
4073 trace_create_file("trace_pipe_raw", 0444, d_cpu,
4074 (void *) cpu, &tracing_buffers_fops);
4076 trace_create_file("stats", 0444, d_cpu,
4077 (void *) cpu, &tracing_stats_fops);
4080 #ifdef CONFIG_FTRACE_SELFTEST
4081 /* Let selftest have access to static functions in this file */
4082 #include "trace_selftest.c"
4085 struct trace_option_dentry {
4086 struct tracer_opt *opt;
4087 struct tracer_flags *flags;
4088 struct dentry *entry;
4092 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4095 struct trace_option_dentry *topt = filp->private_data;
4098 if (topt->flags->val & topt->opt->bit)
4103 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4107 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4110 struct trace_option_dentry *topt = filp->private_data;
4115 if (cnt >= sizeof(buf))
4118 if (copy_from_user(&buf, ubuf, cnt))
4123 ret = strict_strtoul(buf, 10, &val);
4127 if (val != 0 && val != 1)
4130 if (!!(topt->flags->val & topt->opt->bit) != val) {
4131 mutex_lock(&trace_types_lock);
4132 ret = __set_tracer_option(current_trace, topt->flags,
4134 mutex_unlock(&trace_types_lock);
4145 static const struct file_operations trace_options_fops = {
4146 .open = tracing_open_generic,
4147 .read = trace_options_read,
4148 .write = trace_options_write,
4152 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4155 long index = (long)filp->private_data;
4158 if (trace_flags & (1 << index))
4163 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4167 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4170 long index = (long)filp->private_data;
4175 if (cnt >= sizeof(buf))
4178 if (copy_from_user(&buf, ubuf, cnt))
4183 ret = strict_strtoul(buf, 10, &val);
4187 if (val != 0 && val != 1)
4189 set_tracer_flags(1 << index, val);
4196 static const struct file_operations trace_options_core_fops = {
4197 .open = tracing_open_generic,
4198 .read = trace_options_core_read,
4199 .write = trace_options_core_write,
4202 struct dentry *trace_create_file(const char *name,
4204 struct dentry *parent,
4206 const struct file_operations *fops)
4210 ret = debugfs_create_file(name, mode, parent, data, fops);
4212 pr_warning("Could not create debugfs '%s' entry\n", name);
4218 static struct dentry *trace_options_init_dentry(void)
4220 struct dentry *d_tracer;
4221 static struct dentry *t_options;
4226 d_tracer = tracing_init_dentry();
4230 t_options = debugfs_create_dir("options", d_tracer);
4232 pr_warning("Could not create debugfs directory 'options'\n");
4240 create_trace_option_file(struct trace_option_dentry *topt,
4241 struct tracer_flags *flags,
4242 struct tracer_opt *opt)
4244 struct dentry *t_options;
4246 t_options = trace_options_init_dentry();
4250 topt->flags = flags;
4253 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4254 &trace_options_fops);
4258 static struct trace_option_dentry *
4259 create_trace_option_files(struct tracer *tracer)
4261 struct trace_option_dentry *topts;
4262 struct tracer_flags *flags;
4263 struct tracer_opt *opts;
4269 flags = tracer->flags;
4271 if (!flags || !flags->opts)
4276 for (cnt = 0; opts[cnt].name; cnt++)
4279 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4283 for (cnt = 0; opts[cnt].name; cnt++)
4284 create_trace_option_file(&topts[cnt], flags,
4291 destroy_trace_option_files(struct trace_option_dentry *topts)
4298 for (cnt = 0; topts[cnt].opt; cnt++) {
4299 if (topts[cnt].entry)
4300 debugfs_remove(topts[cnt].entry);
4306 static struct dentry *
4307 create_trace_option_core_file(const char *option, long index)
4309 struct dentry *t_options;
4311 t_options = trace_options_init_dentry();
4315 return trace_create_file(option, 0644, t_options, (void *)index,
4316 &trace_options_core_fops);
4319 static __init void create_trace_options_dir(void)
4321 struct dentry *t_options;
4324 t_options = trace_options_init_dentry();
4328 for (i = 0; trace_options[i]; i++)
4329 create_trace_option_core_file(trace_options[i], i);
4332 static __init int tracer_init_debugfs(void)
4334 struct dentry *d_tracer;
4337 trace_access_lock_init();
4339 d_tracer = tracing_init_dentry();
4341 trace_create_file("tracing_enabled", 0644, d_tracer,
4342 &global_trace, &tracing_ctrl_fops);
4344 trace_create_file("trace_options", 0644, d_tracer,
4345 NULL, &tracing_iter_fops);
4347 trace_create_file("tracing_cpumask", 0644, d_tracer,
4348 NULL, &tracing_cpumask_fops);
4350 trace_create_file("trace", 0644, d_tracer,
4351 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4353 trace_create_file("available_tracers", 0444, d_tracer,
4354 &global_trace, &show_traces_fops);
4356 trace_create_file("current_tracer", 0644, d_tracer,
4357 &global_trace, &set_tracer_fops);
4359 #ifdef CONFIG_TRACER_MAX_TRACE
4360 trace_create_file("tracing_max_latency", 0644, d_tracer,
4361 &tracing_max_latency, &tracing_max_lat_fops);
4364 trace_create_file("tracing_thresh", 0644, d_tracer,
4365 &tracing_thresh, &tracing_max_lat_fops);
4367 trace_create_file("README", 0444, d_tracer,
4368 NULL, &tracing_readme_fops);
4370 trace_create_file("trace_pipe", 0444, d_tracer,
4371 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4373 trace_create_file("buffer_size_kb", 0644, d_tracer,
4374 &global_trace, &tracing_entries_fops);
4376 trace_create_file("trace_marker", 0220, d_tracer,
4377 NULL, &tracing_mark_fops);
4379 trace_create_file("saved_cmdlines", 0444, d_tracer,
4380 NULL, &tracing_saved_cmdlines_fops);
4382 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4385 #ifdef CONFIG_DYNAMIC_FTRACE
4386 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4387 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4389 #ifdef CONFIG_SYSPROF_TRACER
4390 init_tracer_sysprof_debugfs(d_tracer);
4393 create_trace_options_dir();
4395 for_each_tracing_cpu(cpu)
4396 tracing_init_debugfs_percpu(cpu);
4401 static int trace_panic_handler(struct notifier_block *this,
4402 unsigned long event, void *unused)
4404 if (ftrace_dump_on_oops)
4405 ftrace_dump(ftrace_dump_on_oops);
4409 static struct notifier_block trace_panic_notifier = {
4410 .notifier_call = trace_panic_handler,
4412 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4415 static int trace_die_handler(struct notifier_block *self,
4421 if (ftrace_dump_on_oops)
4422 ftrace_dump(ftrace_dump_on_oops);
4430 static struct notifier_block trace_die_notifier = {
4431 .notifier_call = trace_die_handler,
4436 * printk is set to max of 1024, we really don't need it that big.
4437 * Nothing should be printing 1000 characters anyway.
4439 #define TRACE_MAX_PRINT 1000
4442 * Define here KERN_TRACE so that we have one place to modify
4443 * it if we decide to change what log level the ftrace dump
4446 #define KERN_TRACE KERN_EMERG
4449 trace_printk_seq(struct trace_seq *s)
4451 /* Probably should print a warning here. */
4455 /* should be zero ended, but we are paranoid. */
4456 s->buffer[s->len] = 0;
4458 printk(KERN_TRACE "%s", s->buffer);
4464 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4466 static arch_spinlock_t ftrace_dump_lock =
4467 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4468 /* use static because iter can be a bit big for the stack */
4469 static struct trace_iterator iter;
4470 unsigned int old_userobj;
4471 static int dump_ran;
4472 unsigned long flags;
4476 local_irq_save(flags);
4477 arch_spin_lock(&ftrace_dump_lock);
4485 if (disable_tracing)
4488 for_each_tracing_cpu(cpu) {
4489 atomic_inc(&global_trace.data[cpu]->disabled);
4492 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4494 /* don't look at user memory in panic mode */
4495 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4497 /* Simulate the iterator */
4498 iter.tr = &global_trace;
4499 iter.trace = current_trace;
4501 switch (oops_dump_mode) {
4503 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4506 iter.cpu_file = raw_smp_processor_id();
4511 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
4512 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4515 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4518 * We need to stop all tracing on all CPUS to read the
4519 * the next buffer. This is a bit expensive, but is
4520 * not done often. We fill all what we can read,
4521 * and then release the locks again.
4524 while (!trace_empty(&iter)) {
4527 printk(KERN_TRACE "---------------------------------\n");
4531 /* reset all but tr, trace, and overruns */
4532 memset(&iter.seq, 0,
4533 sizeof(struct trace_iterator) -
4534 offsetof(struct trace_iterator, seq));
4535 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4538 if (find_next_entry_inc(&iter) != NULL) {
4541 ret = print_trace_line(&iter);
4542 if (ret != TRACE_TYPE_NO_CONSUME)
4543 trace_consume(&iter);
4546 trace_printk_seq(&iter.seq);
4550 printk(KERN_TRACE " (ftrace buffer empty)\n");
4552 printk(KERN_TRACE "---------------------------------\n");
4555 /* Re-enable tracing if requested */
4556 if (!disable_tracing) {
4557 trace_flags |= old_userobj;
4559 for_each_tracing_cpu(cpu) {
4560 atomic_dec(&global_trace.data[cpu]->disabled);
4566 arch_spin_unlock(&ftrace_dump_lock);
4567 local_irq_restore(flags);
4570 /* By default: disable tracing after the dump */
4571 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
4573 __ftrace_dump(true, oops_dump_mode);
4576 __init static int tracer_alloc_buffers(void)
4582 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4585 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4586 goto out_free_buffer_mask;
4588 /* To save memory, keep the ring buffer size to its minimum */
4589 if (ring_buffer_expanded)
4590 ring_buf_size = trace_buf_size;
4594 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4595 cpumask_copy(tracing_cpumask, cpu_all_mask);
4597 /* TODO: make the number of buffers hot pluggable with CPUS */
4598 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4599 TRACE_BUFFER_FLAGS);
4600 if (!global_trace.buffer) {
4601 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4603 goto out_free_cpumask;
4605 global_trace.entries = ring_buffer_size(global_trace.buffer);
4608 #ifdef CONFIG_TRACER_MAX_TRACE
4609 max_tr.buffer = ring_buffer_alloc(1, TRACE_BUFFER_FLAGS);
4610 if (!max_tr.buffer) {
4611 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4613 ring_buffer_free(global_trace.buffer);
4614 goto out_free_cpumask;
4619 /* Allocate the first page for all buffers */
4620 for_each_tracing_cpu(i) {
4621 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4622 max_tr.data[i] = &per_cpu(max_tr_data, i);
4625 trace_init_cmdlines();
4627 register_tracer(&nop_trace);
4628 current_trace = &nop_trace;
4629 /* All seems OK, enable tracing */
4630 tracing_disabled = 0;
4632 atomic_notifier_chain_register(&panic_notifier_list,
4633 &trace_panic_notifier);
4635 register_die_notifier(&trace_die_notifier);
4640 free_cpumask_var(tracing_cpumask);
4641 out_free_buffer_mask:
4642 free_cpumask_var(tracing_buffer_mask);
4647 __init static int clear_boot_tracer(void)
4650 * The default tracer at boot buffer is an init section.
4651 * This function is called in lateinit. If we did not
4652 * find the boot tracer, then clear it out, to prevent
4653 * later registration from accessing the buffer that is
4654 * about to be freed.
4656 if (!default_bootup_tracer)
4659 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4660 default_bootup_tracer);
4661 default_bootup_tracer = NULL;
4666 early_initcall(tracer_alloc_buffers);
4667 fs_initcall(tracer_init_debugfs);
4668 late_initcall(clear_boot_tracer);