2 * ring buffer based function tracer
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 * Based on code from the latency_tracer, that is:
11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 William Lee Irwin III
14 #include <linux/ring_buffer.h>
15 #include <generated/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/notifier.h>
21 #include <linux/irqflags.h>
22 #include <linux/debugfs.h>
23 #include <linux/pagemap.h>
24 #include <linux/hardirq.h>
25 #include <linux/linkage.h>
26 #include <linux/uaccess.h>
27 #include <linux/kprobes.h>
28 #include <linux/ftrace.h>
29 #include <linux/module.h>
30 #include <linux/percpu.h>
31 #include <linux/splice.h>
32 #include <linux/kdebug.h>
33 #include <linux/string.h>
34 #include <linux/rwsem.h>
35 #include <linux/slab.h>
36 #include <linux/ctype.h>
37 #include <linux/init.h>
38 #include <linux/poll.h>
39 #include <linux/nmi.h>
43 #include "trace_output.h"
46 * On boot up, the ring buffer is set to the minimum size, so that
47 * we do not waste memory on systems that are not using tracing.
49 int ring_buffer_expanded;
52 * We need to change this state when a selftest is running.
53 * A selftest will lurk into the ring-buffer to count the
54 * entries inserted during the selftest although some concurrent
55 * insertions into the ring-buffer such as trace_printk could occurred
56 * at the same time, giving false positive or negative results.
58 static bool __read_mostly tracing_selftest_running;
61 * If a tracer is running, we do not want to run SELFTEST.
63 bool __read_mostly tracing_selftest_disabled;
65 /* For tracers that don't implement custom flags */
66 static struct tracer_opt dummy_tracer_opt[] = {
70 static struct tracer_flags dummy_tracer_flags = {
72 .opts = dummy_tracer_opt
75 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
81 * Kill all tracing for good (never come back).
82 * It is initialized to 1 but will turn to zero if the initialization
83 * of the tracer is successful. But that is the only place that sets
86 static int tracing_disabled = 1;
88 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
90 static inline void ftrace_disable_cpu(void)
93 __this_cpu_inc(ftrace_cpu_disabled);
96 static inline void ftrace_enable_cpu(void)
98 __this_cpu_dec(ftrace_cpu_disabled);
102 cpumask_var_t __read_mostly tracing_buffer_mask;
105 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
107 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
108 * is set, then ftrace_dump is called. This will output the contents
109 * of the ftrace buffers to the console. This is very useful for
110 * capturing traces that lead to crashes and outputing it to a
113 * It is default off, but you can enable it with either specifying
114 * "ftrace_dump_on_oops" in the kernel command line, or setting
115 * /proc/sys/kernel/ftrace_dump_on_oops
116 * Set 1 if you want to dump buffers of all CPUs
117 * Set 2 if you want to dump the buffer of the CPU that triggered oops
120 enum ftrace_dump_mode ftrace_dump_on_oops;
122 static int tracing_set_tracer(const char *buf);
124 #define MAX_TRACER_SIZE 100
125 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
126 static char *default_bootup_tracer;
128 static int __init set_cmdline_ftrace(char *str)
130 strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
131 default_bootup_tracer = bootup_tracer_buf;
132 /* We are using ftrace early, expand it */
133 ring_buffer_expanded = 1;
136 __setup("ftrace=", set_cmdline_ftrace);
138 static int __init set_ftrace_dump_on_oops(char *str)
140 if (*str++ != '=' || !*str) {
141 ftrace_dump_on_oops = DUMP_ALL;
145 if (!strcmp("orig_cpu", str)) {
146 ftrace_dump_on_oops = DUMP_ORIG;
152 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
154 unsigned long long ns2usecs(cycle_t nsec)
162 * The global_trace is the descriptor that holds the tracing
163 * buffers for the live tracing. For each CPU, it contains
164 * a link list of pages that will store trace entries. The
165 * page descriptor of the pages in the memory is used to hold
166 * the link list by linking the lru item in the page descriptor
167 * to each of the pages in the buffer per CPU.
169 * For each active CPU there is a data field that holds the
170 * pages for the buffer for that CPU. Each CPU has the same number
171 * of pages allocated for its buffer.
173 static struct trace_array global_trace;
175 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
177 int filter_current_check_discard(struct ring_buffer *buffer,
178 struct ftrace_event_call *call, void *rec,
179 struct ring_buffer_event *event)
181 return filter_check_discard(call, rec, buffer, event);
183 EXPORT_SYMBOL_GPL(filter_current_check_discard);
185 cycle_t ftrace_now(int cpu)
189 /* Early boot up does not have a buffer yet */
190 if (!global_trace.buffer)
191 return trace_clock_local();
193 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
194 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
200 * The max_tr is used to snapshot the global_trace when a maximum
201 * latency is reached. Some tracers will use this to store a maximum
202 * trace while it continues examining live traces.
204 * The buffers for the max_tr are set up the same as the global_trace.
205 * When a snapshot is taken, the link list of the max_tr is swapped
206 * with the link list of the global_trace and the buffers are reset for
207 * the global_trace so the tracing can continue.
209 static struct trace_array max_tr;
211 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
213 /* tracer_enabled is used to toggle activation of a tracer */
214 static int tracer_enabled = 1;
217 * tracing_is_enabled - return tracer_enabled status
219 * This function is used by other tracers to know the status
220 * of the tracer_enabled flag. Tracers may use this function
221 * to know if it should enable their features when starting
222 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
224 int tracing_is_enabled(void)
226 return tracer_enabled;
230 * trace_buf_size is the size in bytes that is allocated
231 * for a buffer. Note, the number of bytes is always rounded
234 * This number is purposely set to a low number of 16384.
235 * If the dump on oops happens, it will be much appreciated
236 * to not have to wait for all that output. Anyway this can be
237 * boot time and run time configurable.
239 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
241 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
243 /* trace_types holds a link list of available tracers. */
244 static struct tracer *trace_types __read_mostly;
246 /* current_trace points to the tracer that is currently active */
247 static struct tracer *current_trace __read_mostly;
250 * trace_types_lock is used to protect the trace_types list.
252 static DEFINE_MUTEX(trace_types_lock);
255 * serialize the access of the ring buffer
257 * ring buffer serializes readers, but it is low level protection.
258 * The validity of the events (which returns by ring_buffer_peek() ..etc)
259 * are not protected by ring buffer.
261 * The content of events may become garbage if we allow other process consumes
262 * these events concurrently:
263 * A) the page of the consumed events may become a normal page
264 * (not reader page) in ring buffer, and this page will be rewrited
265 * by events producer.
266 * B) The page of the consumed events may become a page for splice_read,
267 * and this page will be returned to system.
269 * These primitives allow multi process access to different cpu ring buffer
272 * These primitives don't distinguish read-only and read-consume access.
273 * Multi read-only access are also serialized.
277 static DECLARE_RWSEM(all_cpu_access_lock);
278 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
280 static inline void trace_access_lock(int cpu)
282 if (cpu == TRACE_PIPE_ALL_CPU) {
283 /* gain it for accessing the whole ring buffer. */
284 down_write(&all_cpu_access_lock);
286 /* gain it for accessing a cpu ring buffer. */
288 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
289 down_read(&all_cpu_access_lock);
291 /* Secondly block other access to this @cpu ring buffer. */
292 mutex_lock(&per_cpu(cpu_access_lock, cpu));
296 static inline void trace_access_unlock(int cpu)
298 if (cpu == TRACE_PIPE_ALL_CPU) {
299 up_write(&all_cpu_access_lock);
301 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
302 up_read(&all_cpu_access_lock);
306 static inline void trace_access_lock_init(void)
310 for_each_possible_cpu(cpu)
311 mutex_init(&per_cpu(cpu_access_lock, cpu));
316 static DEFINE_MUTEX(access_lock);
318 static inline void trace_access_lock(int cpu)
321 mutex_lock(&access_lock);
324 static inline void trace_access_unlock(int cpu)
327 mutex_unlock(&access_lock);
330 static inline void trace_access_lock_init(void)
336 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
337 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
339 /* trace_flags holds trace_options default values */
340 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
341 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
342 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE |
345 static int trace_stop_count;
346 static DEFINE_RAW_SPINLOCK(tracing_start_lock);
348 static void wakeup_work_handler(struct work_struct *work)
350 wake_up(&trace_wait);
353 static DECLARE_DELAYED_WORK(wakeup_work, wakeup_work_handler);
356 * tracing_on - enable tracing buffers
358 * This function enables tracing buffers that may have been
359 * disabled with tracing_off.
361 void tracing_on(void)
363 if (global_trace.buffer)
364 ring_buffer_record_on(global_trace.buffer);
366 * This flag is only looked at when buffers haven't been
367 * allocated yet. We don't really care about the race
368 * between setting this flag and actually turning
371 global_trace.buffer_disabled = 0;
373 EXPORT_SYMBOL_GPL(tracing_on);
376 * tracing_off - turn off tracing buffers
378 * This function stops the tracing buffers from recording data.
379 * It does not disable any overhead the tracers themselves may
380 * be causing. This function simply causes all recording to
381 * the ring buffers to fail.
383 void tracing_off(void)
385 if (global_trace.buffer)
386 ring_buffer_record_off(global_trace.buffer);
388 * This flag is only looked at when buffers haven't been
389 * allocated yet. We don't really care about the race
390 * between setting this flag and actually turning
393 global_trace.buffer_disabled = 1;
395 EXPORT_SYMBOL_GPL(tracing_off);
398 * tracing_is_on - show state of ring buffers enabled
400 int tracing_is_on(void)
402 if (global_trace.buffer)
403 return ring_buffer_record_is_on(global_trace.buffer);
404 return !global_trace.buffer_disabled;
406 EXPORT_SYMBOL_GPL(tracing_is_on);
409 * trace_wake_up - wake up tasks waiting for trace input
411 * Schedules a delayed work to wake up any task that is blocked on the
412 * trace_wait queue. These is used with trace_poll for tasks polling the
415 void trace_wake_up(void)
417 const unsigned long delay = msecs_to_jiffies(2);
419 if (trace_flags & TRACE_ITER_BLOCK)
421 schedule_delayed_work(&wakeup_work, delay);
424 static int __init set_buf_size(char *str)
426 unsigned long buf_size;
430 buf_size = memparse(str, &str);
431 /* nr_entries can not be zero */
434 trace_buf_size = buf_size;
437 __setup("trace_buf_size=", set_buf_size);
439 static int __init set_tracing_thresh(char *str)
441 unsigned long threshhold;
446 ret = strict_strtoul(str, 0, &threshhold);
449 tracing_thresh = threshhold * 1000;
452 __setup("tracing_thresh=", set_tracing_thresh);
454 unsigned long nsecs_to_usecs(unsigned long nsecs)
459 /* These must match the bit postions in trace_iterator_flags */
460 static const char *trace_options[] = {
492 { trace_clock_local, "local" },
493 { trace_clock_global, "global" },
494 { trace_clock_counter, "counter" },
500 * trace_parser_get_init - gets the buffer for trace parser
502 int trace_parser_get_init(struct trace_parser *parser, int size)
504 memset(parser, 0, sizeof(*parser));
506 parser->buffer = kmalloc(size, GFP_KERNEL);
515 * trace_parser_put - frees the buffer for trace parser
517 void trace_parser_put(struct trace_parser *parser)
519 kfree(parser->buffer);
523 * trace_get_user - reads the user input string separated by space
524 * (matched by isspace(ch))
526 * For each string found the 'struct trace_parser' is updated,
527 * and the function returns.
529 * Returns number of bytes read.
531 * See kernel/trace/trace.h for 'struct trace_parser' details.
533 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
534 size_t cnt, loff_t *ppos)
541 trace_parser_clear(parser);
543 ret = get_user(ch, ubuf++);
551 * The parser is not finished with the last write,
552 * continue reading the user input without skipping spaces.
555 /* skip white space */
556 while (cnt && isspace(ch)) {
557 ret = get_user(ch, ubuf++);
564 /* only spaces were written */
574 /* read the non-space input */
575 while (cnt && !isspace(ch)) {
576 if (parser->idx < parser->size - 1)
577 parser->buffer[parser->idx++] = ch;
582 ret = get_user(ch, ubuf++);
589 /* We either got finished input or we have to wait for another call. */
591 parser->buffer[parser->idx] = 0;
592 parser->cont = false;
595 parser->buffer[parser->idx++] = ch;
605 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
613 if (s->len <= s->readpos)
616 len = s->len - s->readpos;
619 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
629 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
634 if (s->len <= s->readpos)
637 len = s->len - s->readpos;
640 ret = memcpy(buf, s->buffer + s->readpos, cnt);
649 * ftrace_max_lock is used to protect the swapping of buffers
650 * when taking a max snapshot. The buffers themselves are
651 * protected by per_cpu spinlocks. But the action of the swap
652 * needs its own lock.
654 * This is defined as a arch_spinlock_t in order to help
655 * with performance when lockdep debugging is enabled.
657 * It is also used in other places outside the update_max_tr
658 * so it needs to be defined outside of the
659 * CONFIG_TRACER_MAX_TRACE.
661 static arch_spinlock_t ftrace_max_lock =
662 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
664 unsigned long __read_mostly tracing_thresh;
666 #ifdef CONFIG_TRACER_MAX_TRACE
667 unsigned long __read_mostly tracing_max_latency;
670 * Copy the new maximum trace into the separate maximum-trace
671 * structure. (this way the maximum trace is permanently saved,
672 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
675 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
677 struct trace_array_cpu *data = tr->data[cpu];
678 struct trace_array_cpu *max_data;
681 max_tr.time_start = data->preempt_timestamp;
683 max_data = max_tr.data[cpu];
684 max_data->saved_latency = tracing_max_latency;
685 max_data->critical_start = data->critical_start;
686 max_data->critical_end = data->critical_end;
688 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
689 max_data->pid = tsk->pid;
690 max_data->uid = task_uid(tsk);
691 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
692 max_data->policy = tsk->policy;
693 max_data->rt_priority = tsk->rt_priority;
695 /* record this tasks comm */
696 tracing_record_cmdline(tsk);
700 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
702 * @tsk: the task with the latency
703 * @cpu: The cpu that initiated the trace.
705 * Flip the buffers between the @tr and the max_tr and record information
706 * about which task was the cause of this latency.
709 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
711 struct ring_buffer *buf;
713 if (trace_stop_count)
716 WARN_ON_ONCE(!irqs_disabled());
717 if (!current_trace->use_max_tr) {
721 arch_spin_lock(&ftrace_max_lock);
724 tr->buffer = max_tr.buffer;
727 __update_max_tr(tr, tsk, cpu);
728 arch_spin_unlock(&ftrace_max_lock);
732 * update_max_tr_single - only copy one trace over, and reset the rest
734 * @tsk - task with the latency
735 * @cpu - the cpu of the buffer to copy.
737 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
740 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
744 if (trace_stop_count)
747 WARN_ON_ONCE(!irqs_disabled());
748 if (!current_trace->use_max_tr) {
753 arch_spin_lock(&ftrace_max_lock);
755 ftrace_disable_cpu();
757 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
761 * We failed to swap the buffer due to a commit taking
762 * place on this CPU. We fail to record, but we reset
763 * the max trace buffer (no one writes directly to it)
764 * and flag that it failed.
766 trace_array_printk(&max_tr, _THIS_IP_,
767 "Failed to swap buffers due to commit in progress\n");
772 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
774 __update_max_tr(tr, tsk, cpu);
775 arch_spin_unlock(&ftrace_max_lock);
777 #endif /* CONFIG_TRACER_MAX_TRACE */
780 * register_tracer - register a tracer with the ftrace system.
781 * @type - the plugin for the tracer
783 * Register a new plugin tracer.
785 int register_tracer(struct tracer *type)
786 __releases(kernel_lock)
787 __acquires(kernel_lock)
793 pr_info("Tracer must have a name\n");
797 if (strlen(type->name) >= MAX_TRACER_SIZE) {
798 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
802 mutex_lock(&trace_types_lock);
804 tracing_selftest_running = true;
806 for (t = trace_types; t; t = t->next) {
807 if (strcmp(type->name, t->name) == 0) {
809 pr_info("Tracer %s already registered\n",
817 type->set_flag = &dummy_set_flag;
819 type->flags = &dummy_tracer_flags;
821 if (!type->flags->opts)
822 type->flags->opts = dummy_tracer_opt;
823 if (!type->wait_pipe)
824 type->wait_pipe = default_wait_pipe;
827 #ifdef CONFIG_FTRACE_STARTUP_TEST
828 if (type->selftest && !tracing_selftest_disabled) {
829 struct tracer *saved_tracer = current_trace;
830 struct trace_array *tr = &global_trace;
833 * Run a selftest on this tracer.
834 * Here we reset the trace buffer, and set the current
835 * tracer to be this tracer. The tracer can then run some
836 * internal tracing to verify that everything is in order.
837 * If we fail, we do not register this tracer.
839 tracing_reset_online_cpus(tr);
841 current_trace = type;
843 /* If we expanded the buffers, make sure the max is expanded too */
844 if (ring_buffer_expanded && type->use_max_tr)
845 ring_buffer_resize(max_tr.buffer, trace_buf_size);
847 /* the test is responsible for initializing and enabling */
848 pr_info("Testing tracer %s: ", type->name);
849 ret = type->selftest(type, tr);
850 /* the test is responsible for resetting too */
851 current_trace = saved_tracer;
853 printk(KERN_CONT "FAILED!\n");
856 /* Only reset on passing, to avoid touching corrupted buffers */
857 tracing_reset_online_cpus(tr);
859 /* Shrink the max buffer again */
860 if (ring_buffer_expanded && type->use_max_tr)
861 ring_buffer_resize(max_tr.buffer, 1);
863 printk(KERN_CONT "PASSED\n");
867 type->next = trace_types;
871 tracing_selftest_running = false;
872 mutex_unlock(&trace_types_lock);
874 if (ret || !default_bootup_tracer)
877 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
880 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
881 /* Do we want this tracer to start on bootup? */
882 tracing_set_tracer(type->name);
883 default_bootup_tracer = NULL;
884 /* disable other selftests, since this will break it. */
885 tracing_selftest_disabled = 1;
886 #ifdef CONFIG_FTRACE_STARTUP_TEST
887 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
895 void unregister_tracer(struct tracer *type)
899 mutex_lock(&trace_types_lock);
900 for (t = &trace_types; *t; t = &(*t)->next) {
904 pr_info("Tracer %s not registered\n", type->name);
910 if (type == current_trace && tracer_enabled) {
913 if (current_trace->stop)
914 current_trace->stop(&global_trace);
915 current_trace = &nop_trace;
918 mutex_unlock(&trace_types_lock);
921 static void __tracing_reset(struct ring_buffer *buffer, int cpu)
923 ftrace_disable_cpu();
924 ring_buffer_reset_cpu(buffer, cpu);
928 void tracing_reset(struct trace_array *tr, int cpu)
930 struct ring_buffer *buffer = tr->buffer;
932 ring_buffer_record_disable(buffer);
934 /* Make sure all commits have finished */
936 __tracing_reset(buffer, cpu);
938 ring_buffer_record_enable(buffer);
941 void tracing_reset_online_cpus(struct trace_array *tr)
943 struct ring_buffer *buffer = tr->buffer;
946 ring_buffer_record_disable(buffer);
948 /* Make sure all commits have finished */
951 tr->time_start = ftrace_now(tr->cpu);
953 for_each_online_cpu(cpu)
954 __tracing_reset(buffer, cpu);
956 ring_buffer_record_enable(buffer);
959 void tracing_reset_current(int cpu)
961 tracing_reset(&global_trace, cpu);
964 void tracing_reset_current_online_cpus(void)
966 tracing_reset_online_cpus(&global_trace);
969 #define SAVED_CMDLINES 128
970 #define NO_CMDLINE_MAP UINT_MAX
971 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
972 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
973 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
974 static int cmdline_idx;
975 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
977 /* temporary disable recording */
978 static atomic_t trace_record_cmdline_disabled __read_mostly;
980 static void trace_init_cmdlines(void)
982 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
983 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
987 int is_tracing_stopped(void)
989 return trace_stop_count;
993 * ftrace_off_permanent - disable all ftrace code permanently
995 * This should only be called when a serious anomally has
996 * been detected. This will turn off the function tracing,
997 * ring buffers, and other tracing utilites. It takes no
998 * locks and can be called from any context.
1000 void ftrace_off_permanent(void)
1002 tracing_disabled = 1;
1004 tracing_off_permanent();
1008 * tracing_start - quick start of the tracer
1010 * If tracing is enabled but was stopped by tracing_stop,
1011 * this will start the tracer back up.
1013 void tracing_start(void)
1015 struct ring_buffer *buffer;
1016 unsigned long flags;
1018 if (tracing_disabled)
1021 raw_spin_lock_irqsave(&tracing_start_lock, flags);
1022 if (--trace_stop_count) {
1023 if (trace_stop_count < 0) {
1024 /* Someone screwed up their debugging */
1026 trace_stop_count = 0;
1031 /* Prevent the buffers from switching */
1032 arch_spin_lock(&ftrace_max_lock);
1034 buffer = global_trace.buffer;
1036 ring_buffer_record_enable(buffer);
1038 buffer = max_tr.buffer;
1040 ring_buffer_record_enable(buffer);
1042 arch_spin_unlock(&ftrace_max_lock);
1046 raw_spin_unlock_irqrestore(&tracing_start_lock, flags);
1050 * tracing_stop - quick stop of the tracer
1052 * Light weight way to stop tracing. Use in conjunction with
1055 void tracing_stop(void)
1057 struct ring_buffer *buffer;
1058 unsigned long flags;
1061 raw_spin_lock_irqsave(&tracing_start_lock, flags);
1062 if (trace_stop_count++)
1065 /* Prevent the buffers from switching */
1066 arch_spin_lock(&ftrace_max_lock);
1068 buffer = global_trace.buffer;
1070 ring_buffer_record_disable(buffer);
1072 buffer = max_tr.buffer;
1074 ring_buffer_record_disable(buffer);
1076 arch_spin_unlock(&ftrace_max_lock);
1079 raw_spin_unlock_irqrestore(&tracing_start_lock, flags);
1082 void trace_stop_cmdline_recording(void);
1084 static void trace_save_cmdline(struct task_struct *tsk)
1088 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1092 * It's not the end of the world if we don't get
1093 * the lock, but we also don't want to spin
1094 * nor do we want to disable interrupts,
1095 * so if we miss here, then better luck next time.
1097 if (!arch_spin_trylock(&trace_cmdline_lock))
1100 idx = map_pid_to_cmdline[tsk->pid];
1101 if (idx == NO_CMDLINE_MAP) {
1102 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1105 * Check whether the cmdline buffer at idx has a pid
1106 * mapped. We are going to overwrite that entry so we
1107 * need to clear the map_pid_to_cmdline. Otherwise we
1108 * would read the new comm for the old pid.
1110 pid = map_cmdline_to_pid[idx];
1111 if (pid != NO_CMDLINE_MAP)
1112 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1114 map_cmdline_to_pid[idx] = tsk->pid;
1115 map_pid_to_cmdline[tsk->pid] = idx;
1120 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1122 arch_spin_unlock(&trace_cmdline_lock);
1125 void trace_find_cmdline(int pid, char comm[])
1130 strcpy(comm, "<idle>");
1134 if (WARN_ON_ONCE(pid < 0)) {
1135 strcpy(comm, "<XXX>");
1139 if (pid > PID_MAX_DEFAULT) {
1140 strcpy(comm, "<...>");
1145 arch_spin_lock(&trace_cmdline_lock);
1146 map = map_pid_to_cmdline[pid];
1147 if (map != NO_CMDLINE_MAP)
1148 strcpy(comm, saved_cmdlines[map]);
1150 strcpy(comm, "<...>");
1152 arch_spin_unlock(&trace_cmdline_lock);
1156 void tracing_record_cmdline(struct task_struct *tsk)
1158 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1162 trace_save_cmdline(tsk);
1166 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1169 struct task_struct *tsk = current;
1171 entry->preempt_count = pc & 0xff;
1172 entry->pid = (tsk) ? tsk->pid : 0;
1175 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1176 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1178 TRACE_FLAG_IRQS_NOSUPPORT |
1180 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1181 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1182 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1184 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1186 struct ring_buffer_event *
1187 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1190 unsigned long flags, int pc)
1192 struct ring_buffer_event *event;
1194 event = ring_buffer_lock_reserve(buffer, len);
1195 if (event != NULL) {
1196 struct trace_entry *ent = ring_buffer_event_data(event);
1198 tracing_generic_entry_update(ent, flags, pc);
1206 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1207 struct ring_buffer_event *event,
1208 unsigned long flags, int pc,
1211 ring_buffer_unlock_commit(buffer, event);
1213 ftrace_trace_stack(buffer, flags, 6, pc);
1214 ftrace_trace_userstack(buffer, flags, pc);
1220 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1221 struct ring_buffer_event *event,
1222 unsigned long flags, int pc)
1224 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1227 struct ring_buffer_event *
1228 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1229 int type, unsigned long len,
1230 unsigned long flags, int pc)
1232 *current_rb = global_trace.buffer;
1233 return trace_buffer_lock_reserve(*current_rb,
1234 type, len, flags, pc);
1236 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1238 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1239 struct ring_buffer_event *event,
1240 unsigned long flags, int pc)
1242 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1244 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1246 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1247 struct ring_buffer_event *event,
1248 unsigned long flags, int pc)
1250 __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1252 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1254 void trace_nowake_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1255 struct ring_buffer_event *event,
1256 unsigned long flags, int pc,
1257 struct pt_regs *regs)
1259 ring_buffer_unlock_commit(buffer, event);
1261 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1262 ftrace_trace_userstack(buffer, flags, pc);
1264 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit_regs);
1266 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1267 struct ring_buffer_event *event)
1269 ring_buffer_discard_commit(buffer, event);
1271 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1274 trace_function(struct trace_array *tr,
1275 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1278 struct ftrace_event_call *call = &event_function;
1279 struct ring_buffer *buffer = tr->buffer;
1280 struct ring_buffer_event *event;
1281 struct ftrace_entry *entry;
1283 /* If we are reading the ring buffer, don't trace */
1284 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1287 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1291 entry = ring_buffer_event_data(event);
1293 entry->parent_ip = parent_ip;
1295 if (!filter_check_discard(call, entry, buffer, event))
1296 ring_buffer_unlock_commit(buffer, event);
1300 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1301 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1304 if (likely(!atomic_read(&data->disabled)))
1305 trace_function(tr, ip, parent_ip, flags, pc);
1308 #ifdef CONFIG_STACKTRACE
1310 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1311 struct ftrace_stack {
1312 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1315 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1316 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1318 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1319 unsigned long flags,
1320 int skip, int pc, struct pt_regs *regs)
1322 struct ftrace_event_call *call = &event_kernel_stack;
1323 struct ring_buffer_event *event;
1324 struct stack_entry *entry;
1325 struct stack_trace trace;
1327 int size = FTRACE_STACK_ENTRIES;
1329 trace.nr_entries = 0;
1333 * Since events can happen in NMIs there's no safe way to
1334 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1335 * or NMI comes in, it will just have to use the default
1336 * FTRACE_STACK_SIZE.
1338 preempt_disable_notrace();
1340 use_stack = ++__get_cpu_var(ftrace_stack_reserve);
1342 * We don't need any atomic variables, just a barrier.
1343 * If an interrupt comes in, we don't care, because it would
1344 * have exited and put the counter back to what we want.
1345 * We just need a barrier to keep gcc from moving things
1349 if (use_stack == 1) {
1350 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1351 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1354 save_stack_trace_regs(regs, &trace);
1356 save_stack_trace(&trace);
1358 if (trace.nr_entries > size)
1359 size = trace.nr_entries;
1361 /* From now on, use_stack is a boolean */
1364 size *= sizeof(unsigned long);
1366 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1367 sizeof(*entry) + size, flags, pc);
1370 entry = ring_buffer_event_data(event);
1372 memset(&entry->caller, 0, size);
1375 memcpy(&entry->caller, trace.entries,
1376 trace.nr_entries * sizeof(unsigned long));
1378 trace.max_entries = FTRACE_STACK_ENTRIES;
1379 trace.entries = entry->caller;
1381 save_stack_trace_regs(regs, &trace);
1383 save_stack_trace(&trace);
1386 entry->size = trace.nr_entries;
1388 if (!filter_check_discard(call, entry, buffer, event))
1389 ring_buffer_unlock_commit(buffer, event);
1392 /* Again, don't let gcc optimize things here */
1394 __get_cpu_var(ftrace_stack_reserve)--;
1395 preempt_enable_notrace();
1399 void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1400 int skip, int pc, struct pt_regs *regs)
1402 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1405 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1408 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1411 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1414 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1417 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1420 __ftrace_trace_stack(tr->buffer, flags, skip, pc, NULL);
1424 * trace_dump_stack - record a stack back trace in the trace buffer
1426 void trace_dump_stack(void)
1428 unsigned long flags;
1430 if (tracing_disabled || tracing_selftest_running)
1433 local_save_flags(flags);
1435 /* skipping 3 traces, seems to get us at the caller of this function */
1436 __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count(), NULL);
1439 static DEFINE_PER_CPU(int, user_stack_count);
1442 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1444 struct ftrace_event_call *call = &event_user_stack;
1445 struct ring_buffer_event *event;
1446 struct userstack_entry *entry;
1447 struct stack_trace trace;
1449 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1453 * NMIs can not handle page faults, even with fix ups.
1454 * The save user stack can (and often does) fault.
1456 if (unlikely(in_nmi()))
1460 * prevent recursion, since the user stack tracing may
1461 * trigger other kernel events.
1464 if (__this_cpu_read(user_stack_count))
1467 __this_cpu_inc(user_stack_count);
1469 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1470 sizeof(*entry), flags, pc);
1472 goto out_drop_count;
1473 entry = ring_buffer_event_data(event);
1475 entry->tgid = current->tgid;
1476 memset(&entry->caller, 0, sizeof(entry->caller));
1478 trace.nr_entries = 0;
1479 trace.max_entries = FTRACE_STACK_ENTRIES;
1481 trace.entries = entry->caller;
1483 save_stack_trace_user(&trace);
1484 if (!filter_check_discard(call, entry, buffer, event))
1485 ring_buffer_unlock_commit(buffer, event);
1488 __this_cpu_dec(user_stack_count);
1494 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1496 ftrace_trace_userstack(tr, flags, preempt_count());
1500 #endif /* CONFIG_STACKTRACE */
1503 * trace_vbprintk - write binary msg to tracing buffer
1506 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1508 static arch_spinlock_t trace_buf_lock =
1509 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1510 static u32 trace_buf[TRACE_BUF_SIZE];
1512 struct ftrace_event_call *call = &event_bprint;
1513 struct ring_buffer_event *event;
1514 struct ring_buffer *buffer;
1515 struct trace_array *tr = &global_trace;
1516 struct trace_array_cpu *data;
1517 struct bprint_entry *entry;
1518 unsigned long flags;
1520 int cpu, len = 0, size, pc;
1522 if (unlikely(tracing_selftest_running || tracing_disabled))
1525 /* Don't pollute graph traces with trace_vprintk internals */
1526 pause_graph_tracing();
1528 pc = preempt_count();
1529 preempt_disable_notrace();
1530 cpu = raw_smp_processor_id();
1531 data = tr->data[cpu];
1533 disable = atomic_inc_return(&data->disabled);
1534 if (unlikely(disable != 1))
1537 /* Lockdep uses trace_printk for lock tracing */
1538 local_irq_save(flags);
1539 arch_spin_lock(&trace_buf_lock);
1540 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1542 if (len > TRACE_BUF_SIZE || len < 0)
1545 size = sizeof(*entry) + sizeof(u32) * len;
1546 buffer = tr->buffer;
1547 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1551 entry = ring_buffer_event_data(event);
1555 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1556 if (!filter_check_discard(call, entry, buffer, event)) {
1557 ring_buffer_unlock_commit(buffer, event);
1558 ftrace_trace_stack(buffer, flags, 6, pc);
1562 arch_spin_unlock(&trace_buf_lock);
1563 local_irq_restore(flags);
1566 atomic_dec_return(&data->disabled);
1567 preempt_enable_notrace();
1568 unpause_graph_tracing();
1572 EXPORT_SYMBOL_GPL(trace_vbprintk);
1574 int trace_array_printk(struct trace_array *tr,
1575 unsigned long ip, const char *fmt, ...)
1580 if (!(trace_flags & TRACE_ITER_PRINTK))
1584 ret = trace_array_vprintk(tr, ip, fmt, ap);
1589 int trace_array_vprintk(struct trace_array *tr,
1590 unsigned long ip, const char *fmt, va_list args)
1592 static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1593 static char trace_buf[TRACE_BUF_SIZE];
1595 struct ftrace_event_call *call = &event_print;
1596 struct ring_buffer_event *event;
1597 struct ring_buffer *buffer;
1598 struct trace_array_cpu *data;
1599 int cpu, len = 0, size, pc;
1600 struct print_entry *entry;
1601 unsigned long irq_flags;
1604 if (tracing_disabled || tracing_selftest_running)
1607 pc = preempt_count();
1608 preempt_disable_notrace();
1609 cpu = raw_smp_processor_id();
1610 data = tr->data[cpu];
1612 disable = atomic_inc_return(&data->disabled);
1613 if (unlikely(disable != 1))
1616 pause_graph_tracing();
1617 raw_local_irq_save(irq_flags);
1618 arch_spin_lock(&trace_buf_lock);
1619 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1621 size = sizeof(*entry) + len + 1;
1622 buffer = tr->buffer;
1623 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1627 entry = ring_buffer_event_data(event);
1630 memcpy(&entry->buf, trace_buf, len);
1631 entry->buf[len] = '\0';
1632 if (!filter_check_discard(call, entry, buffer, event)) {
1633 ring_buffer_unlock_commit(buffer, event);
1634 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1638 arch_spin_unlock(&trace_buf_lock);
1639 raw_local_irq_restore(irq_flags);
1640 unpause_graph_tracing();
1642 atomic_dec_return(&data->disabled);
1643 preempt_enable_notrace();
1648 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1650 return trace_array_vprintk(&global_trace, ip, fmt, args);
1652 EXPORT_SYMBOL_GPL(trace_vprintk);
1654 static void trace_iterator_increment(struct trace_iterator *iter)
1656 /* Don't allow ftrace to trace into the ring buffers */
1657 ftrace_disable_cpu();
1660 if (iter->buffer_iter[iter->cpu])
1661 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1663 ftrace_enable_cpu();
1666 static struct trace_entry *
1667 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1668 unsigned long *lost_events)
1670 struct ring_buffer_event *event;
1671 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1673 /* Don't allow ftrace to trace into the ring buffers */
1674 ftrace_disable_cpu();
1677 event = ring_buffer_iter_peek(buf_iter, ts);
1679 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1682 ftrace_enable_cpu();
1685 iter->ent_size = ring_buffer_event_length(event);
1686 return ring_buffer_event_data(event);
1692 static struct trace_entry *
1693 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1694 unsigned long *missing_events, u64 *ent_ts)
1696 struct ring_buffer *buffer = iter->tr->buffer;
1697 struct trace_entry *ent, *next = NULL;
1698 unsigned long lost_events = 0, next_lost = 0;
1699 int cpu_file = iter->cpu_file;
1700 u64 next_ts = 0, ts;
1706 * If we are in a per_cpu trace file, don't bother by iterating over
1707 * all cpu and peek directly.
1709 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1710 if (ring_buffer_empty_cpu(buffer, cpu_file))
1712 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1714 *ent_cpu = cpu_file;
1719 for_each_tracing_cpu(cpu) {
1721 if (ring_buffer_empty_cpu(buffer, cpu))
1724 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1727 * Pick the entry with the smallest timestamp:
1729 if (ent && (!next || ts < next_ts)) {
1733 next_lost = lost_events;
1734 next_size = iter->ent_size;
1738 iter->ent_size = next_size;
1741 *ent_cpu = next_cpu;
1747 *missing_events = next_lost;
1752 /* Find the next real entry, without updating the iterator itself */
1753 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1754 int *ent_cpu, u64 *ent_ts)
1756 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1759 /* Find the next real entry, and increment the iterator to the next entry */
1760 void *trace_find_next_entry_inc(struct trace_iterator *iter)
1762 iter->ent = __find_next_entry(iter, &iter->cpu,
1763 &iter->lost_events, &iter->ts);
1766 trace_iterator_increment(iter);
1768 return iter->ent ? iter : NULL;
1771 static void trace_consume(struct trace_iterator *iter)
1773 /* Don't allow ftrace to trace into the ring buffers */
1774 ftrace_disable_cpu();
1775 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1776 &iter->lost_events);
1777 ftrace_enable_cpu();
1780 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1782 struct trace_iterator *iter = m->private;
1786 WARN_ON_ONCE(iter->leftover);
1790 /* can't go backwards */
1795 ent = trace_find_next_entry_inc(iter);
1799 while (ent && iter->idx < i)
1800 ent = trace_find_next_entry_inc(iter);
1807 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1809 struct trace_array *tr = iter->tr;
1810 struct ring_buffer_event *event;
1811 struct ring_buffer_iter *buf_iter;
1812 unsigned long entries = 0;
1815 tr->data[cpu]->skipped_entries = 0;
1817 if (!iter->buffer_iter[cpu])
1820 buf_iter = iter->buffer_iter[cpu];
1821 ring_buffer_iter_reset(buf_iter);
1824 * We could have the case with the max latency tracers
1825 * that a reset never took place on a cpu. This is evident
1826 * by the timestamp being before the start of the buffer.
1828 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1829 if (ts >= iter->tr->time_start)
1832 ring_buffer_read(buf_iter, NULL);
1835 tr->data[cpu]->skipped_entries = entries;
1839 * The current tracer is copied to avoid a global locking
1842 static void *s_start(struct seq_file *m, loff_t *pos)
1844 struct trace_iterator *iter = m->private;
1845 static struct tracer *old_tracer;
1846 int cpu_file = iter->cpu_file;
1851 /* copy the tracer to avoid using a global lock all around */
1852 mutex_lock(&trace_types_lock);
1853 if (unlikely(old_tracer != current_trace && current_trace)) {
1854 old_tracer = current_trace;
1855 *iter->trace = *current_trace;
1857 mutex_unlock(&trace_types_lock);
1859 atomic_inc(&trace_record_cmdline_disabled);
1861 if (*pos != iter->pos) {
1866 ftrace_disable_cpu();
1868 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1869 for_each_tracing_cpu(cpu)
1870 tracing_iter_reset(iter, cpu);
1872 tracing_iter_reset(iter, cpu_file);
1874 ftrace_enable_cpu();
1877 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1882 * If we overflowed the seq_file before, then we want
1883 * to just reuse the trace_seq buffer again.
1889 p = s_next(m, p, &l);
1893 trace_event_read_lock();
1894 trace_access_lock(cpu_file);
1898 static void s_stop(struct seq_file *m, void *p)
1900 struct trace_iterator *iter = m->private;
1902 atomic_dec(&trace_record_cmdline_disabled);
1903 trace_access_unlock(iter->cpu_file);
1904 trace_event_read_unlock();
1908 get_total_entries(struct trace_array *tr, unsigned long *total, unsigned long *entries)
1910 unsigned long count;
1916 for_each_tracing_cpu(cpu) {
1917 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1919 * If this buffer has skipped entries, then we hold all
1920 * entries for the trace and we need to ignore the
1921 * ones before the time stamp.
1923 if (tr->data[cpu]->skipped_entries) {
1924 count -= tr->data[cpu]->skipped_entries;
1925 /* total is the same as the entries */
1929 ring_buffer_overrun_cpu(tr->buffer, cpu);
1934 static void print_lat_help_header(struct seq_file *m)
1936 seq_puts(m, "# _------=> CPU# \n");
1937 seq_puts(m, "# / _-----=> irqs-off \n");
1938 seq_puts(m, "# | / _----=> need-resched \n");
1939 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1940 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1941 seq_puts(m, "# |||| / delay \n");
1942 seq_puts(m, "# cmd pid ||||| time | caller \n");
1943 seq_puts(m, "# \\ / ||||| \\ | / \n");
1946 static void print_event_info(struct trace_array *tr, struct seq_file *m)
1948 unsigned long total;
1949 unsigned long entries;
1951 get_total_entries(tr, &total, &entries);
1952 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
1953 entries, total, num_online_cpus());
1957 static void print_func_help_header(struct trace_array *tr, struct seq_file *m)
1959 print_event_info(tr, m);
1960 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1961 seq_puts(m, "# | | | | |\n");
1964 static void print_func_help_header_irq(struct trace_array *tr, struct seq_file *m)
1966 print_event_info(tr, m);
1967 seq_puts(m, "# _-----=> irqs-off\n");
1968 seq_puts(m, "# / _----=> need-resched\n");
1969 seq_puts(m, "# | / _---=> hardirq/softirq\n");
1970 seq_puts(m, "# || / _--=> preempt-depth\n");
1971 seq_puts(m, "# ||| / delay\n");
1972 seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
1973 seq_puts(m, "# | | | |||| | |\n");
1977 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1979 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1980 struct trace_array *tr = iter->tr;
1981 struct trace_array_cpu *data = tr->data[tr->cpu];
1982 struct tracer *type = current_trace;
1983 unsigned long entries;
1984 unsigned long total;
1985 const char *name = "preemption";
1990 get_total_entries(tr, &total, &entries);
1992 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1994 seq_puts(m, "# -----------------------------------"
1995 "---------------------------------\n");
1996 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1997 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1998 nsecs_to_usecs(data->saved_latency),
2002 #if defined(CONFIG_PREEMPT_NONE)
2004 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2006 #elif defined(CONFIG_PREEMPT)
2011 /* These are reserved for later use */
2014 seq_printf(m, " #P:%d)\n", num_online_cpus());
2018 seq_puts(m, "# -----------------\n");
2019 seq_printf(m, "# | task: %.16s-%d "
2020 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2021 data->comm, data->pid, data->uid, data->nice,
2022 data->policy, data->rt_priority);
2023 seq_puts(m, "# -----------------\n");
2025 if (data->critical_start) {
2026 seq_puts(m, "# => started at: ");
2027 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2028 trace_print_seq(m, &iter->seq);
2029 seq_puts(m, "\n# => ended at: ");
2030 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2031 trace_print_seq(m, &iter->seq);
2032 seq_puts(m, "\n#\n");
2038 static void test_cpu_buff_start(struct trace_iterator *iter)
2040 struct trace_seq *s = &iter->seq;
2042 if (!(trace_flags & TRACE_ITER_ANNOTATE))
2045 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2048 if (cpumask_test_cpu(iter->cpu, iter->started))
2051 if (iter->tr->data[iter->cpu]->skipped_entries)
2054 cpumask_set_cpu(iter->cpu, iter->started);
2056 /* Don't print started cpu buffer for the first entry of the trace */
2058 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2062 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2064 struct trace_seq *s = &iter->seq;
2065 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2066 struct trace_entry *entry;
2067 struct trace_event *event;
2071 test_cpu_buff_start(iter);
2073 event = ftrace_find_event(entry->type);
2075 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2076 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2077 if (!trace_print_lat_context(iter))
2080 if (!trace_print_context(iter))
2086 return event->funcs->trace(iter, sym_flags, event);
2088 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2091 return TRACE_TYPE_HANDLED;
2093 return TRACE_TYPE_PARTIAL_LINE;
2096 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2098 struct trace_seq *s = &iter->seq;
2099 struct trace_entry *entry;
2100 struct trace_event *event;
2104 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2105 if (!trace_seq_printf(s, "%d %d %llu ",
2106 entry->pid, iter->cpu, iter->ts))
2110 event = ftrace_find_event(entry->type);
2112 return event->funcs->raw(iter, 0, event);
2114 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2117 return TRACE_TYPE_HANDLED;
2119 return TRACE_TYPE_PARTIAL_LINE;
2122 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2124 struct trace_seq *s = &iter->seq;
2125 unsigned char newline = '\n';
2126 struct trace_entry *entry;
2127 struct trace_event *event;
2131 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2132 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2133 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2134 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2137 event = ftrace_find_event(entry->type);
2139 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2140 if (ret != TRACE_TYPE_HANDLED)
2144 SEQ_PUT_FIELD_RET(s, newline);
2146 return TRACE_TYPE_HANDLED;
2149 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2151 struct trace_seq *s = &iter->seq;
2152 struct trace_entry *entry;
2153 struct trace_event *event;
2157 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2158 SEQ_PUT_FIELD_RET(s, entry->pid);
2159 SEQ_PUT_FIELD_RET(s, iter->cpu);
2160 SEQ_PUT_FIELD_RET(s, iter->ts);
2163 event = ftrace_find_event(entry->type);
2164 return event ? event->funcs->binary(iter, 0, event) :
2168 int trace_empty(struct trace_iterator *iter)
2172 /* If we are looking at one CPU buffer, only check that one */
2173 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
2174 cpu = iter->cpu_file;
2175 if (iter->buffer_iter[cpu]) {
2176 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2179 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2185 for_each_tracing_cpu(cpu) {
2186 if (iter->buffer_iter[cpu]) {
2187 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2190 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2198 /* Called with trace_event_read_lock() held. */
2199 enum print_line_t print_trace_line(struct trace_iterator *iter)
2201 enum print_line_t ret;
2203 if (iter->lost_events &&
2204 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2205 iter->cpu, iter->lost_events))
2206 return TRACE_TYPE_PARTIAL_LINE;
2208 if (iter->trace && iter->trace->print_line) {
2209 ret = iter->trace->print_line(iter);
2210 if (ret != TRACE_TYPE_UNHANDLED)
2214 if (iter->ent->type == TRACE_BPRINT &&
2215 trace_flags & TRACE_ITER_PRINTK &&
2216 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2217 return trace_print_bprintk_msg_only(iter);
2219 if (iter->ent->type == TRACE_PRINT &&
2220 trace_flags & TRACE_ITER_PRINTK &&
2221 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2222 return trace_print_printk_msg_only(iter);
2224 if (trace_flags & TRACE_ITER_BIN)
2225 return print_bin_fmt(iter);
2227 if (trace_flags & TRACE_ITER_HEX)
2228 return print_hex_fmt(iter);
2230 if (trace_flags & TRACE_ITER_RAW)
2231 return print_raw_fmt(iter);
2233 return print_trace_fmt(iter);
2236 void trace_latency_header(struct seq_file *m)
2238 struct trace_iterator *iter = m->private;
2240 /* print nothing if the buffers are empty */
2241 if (trace_empty(iter))
2244 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2245 print_trace_header(m, iter);
2247 if (!(trace_flags & TRACE_ITER_VERBOSE))
2248 print_lat_help_header(m);
2251 void trace_default_header(struct seq_file *m)
2253 struct trace_iterator *iter = m->private;
2255 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2258 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2259 /* print nothing if the buffers are empty */
2260 if (trace_empty(iter))
2262 print_trace_header(m, iter);
2263 if (!(trace_flags & TRACE_ITER_VERBOSE))
2264 print_lat_help_header(m);
2266 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2267 if (trace_flags & TRACE_ITER_IRQ_INFO)
2268 print_func_help_header_irq(iter->tr, m);
2270 print_func_help_header(iter->tr, m);
2275 static void test_ftrace_alive(struct seq_file *m)
2277 if (!ftrace_is_dead())
2279 seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2280 seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2283 static int s_show(struct seq_file *m, void *v)
2285 struct trace_iterator *iter = v;
2288 if (iter->ent == NULL) {
2290 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2292 test_ftrace_alive(m);
2294 if (iter->trace && iter->trace->print_header)
2295 iter->trace->print_header(m);
2297 trace_default_header(m);
2299 } else if (iter->leftover) {
2301 * If we filled the seq_file buffer earlier, we
2302 * want to just show it now.
2304 ret = trace_print_seq(m, &iter->seq);
2306 /* ret should this time be zero, but you never know */
2307 iter->leftover = ret;
2310 print_trace_line(iter);
2311 ret = trace_print_seq(m, &iter->seq);
2313 * If we overflow the seq_file buffer, then it will
2314 * ask us for this data again at start up.
2316 * ret is 0 if seq_file write succeeded.
2319 iter->leftover = ret;
2325 static const struct seq_operations tracer_seq_ops = {
2332 static struct trace_iterator *
2333 __tracing_open(struct inode *inode, struct file *file)
2335 long cpu_file = (long) inode->i_private;
2336 void *fail_ret = ERR_PTR(-ENOMEM);
2337 struct trace_iterator *iter;
2341 if (tracing_disabled)
2342 return ERR_PTR(-ENODEV);
2344 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2346 return ERR_PTR(-ENOMEM);
2349 * We make a copy of the current tracer to avoid concurrent
2350 * changes on it while we are reading.
2352 mutex_lock(&trace_types_lock);
2353 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2358 *iter->trace = *current_trace;
2360 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2363 if (current_trace && current_trace->print_max)
2366 iter->tr = &global_trace;
2368 mutex_init(&iter->mutex);
2369 iter->cpu_file = cpu_file;
2371 /* Notify the tracer early; before we stop tracing. */
2372 if (iter->trace && iter->trace->open)
2373 iter->trace->open(iter);
2375 /* Annotate start of buffers if we had overruns */
2376 if (ring_buffer_overruns(iter->tr->buffer))
2377 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2379 /* stop the trace while dumping */
2382 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2383 for_each_tracing_cpu(cpu) {
2384 iter->buffer_iter[cpu] =
2385 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2387 ring_buffer_read_prepare_sync();
2388 for_each_tracing_cpu(cpu) {
2389 ring_buffer_read_start(iter->buffer_iter[cpu]);
2390 tracing_iter_reset(iter, cpu);
2393 cpu = iter->cpu_file;
2394 iter->buffer_iter[cpu] =
2395 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2396 ring_buffer_read_prepare_sync();
2397 ring_buffer_read_start(iter->buffer_iter[cpu]);
2398 tracing_iter_reset(iter, cpu);
2401 ret = seq_open(file, &tracer_seq_ops);
2403 fail_ret = ERR_PTR(ret);
2407 m = file->private_data;
2410 mutex_unlock(&trace_types_lock);
2415 for_each_tracing_cpu(cpu) {
2416 if (iter->buffer_iter[cpu])
2417 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2419 free_cpumask_var(iter->started);
2422 mutex_unlock(&trace_types_lock);
2429 int tracing_open_generic(struct inode *inode, struct file *filp)
2431 if (tracing_disabled)
2434 filp->private_data = inode->i_private;
2438 static int tracing_release(struct inode *inode, struct file *file)
2440 struct seq_file *m = file->private_data;
2441 struct trace_iterator *iter;
2444 if (!(file->f_mode & FMODE_READ))
2449 mutex_lock(&trace_types_lock);
2450 for_each_tracing_cpu(cpu) {
2451 if (iter->buffer_iter[cpu])
2452 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2455 if (iter->trace && iter->trace->close)
2456 iter->trace->close(iter);
2458 /* reenable tracing if it was previously enabled */
2460 mutex_unlock(&trace_types_lock);
2462 seq_release(inode, file);
2463 mutex_destroy(&iter->mutex);
2464 free_cpumask_var(iter->started);
2470 static int tracing_open(struct inode *inode, struct file *file)
2472 struct trace_iterator *iter;
2475 /* If this file was open for write, then erase contents */
2476 if ((file->f_mode & FMODE_WRITE) &&
2477 (file->f_flags & O_TRUNC)) {
2478 long cpu = (long) inode->i_private;
2480 if (cpu == TRACE_PIPE_ALL_CPU)
2481 tracing_reset_online_cpus(&global_trace);
2483 tracing_reset(&global_trace, cpu);
2486 if (file->f_mode & FMODE_READ) {
2487 iter = __tracing_open(inode, file);
2489 ret = PTR_ERR(iter);
2490 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2491 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2497 t_next(struct seq_file *m, void *v, loff_t *pos)
2499 struct tracer *t = v;
2509 static void *t_start(struct seq_file *m, loff_t *pos)
2514 mutex_lock(&trace_types_lock);
2515 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2521 static void t_stop(struct seq_file *m, void *p)
2523 mutex_unlock(&trace_types_lock);
2526 static int t_show(struct seq_file *m, void *v)
2528 struct tracer *t = v;
2533 seq_printf(m, "%s", t->name);
2542 static const struct seq_operations show_traces_seq_ops = {
2549 static int show_traces_open(struct inode *inode, struct file *file)
2551 if (tracing_disabled)
2554 return seq_open(file, &show_traces_seq_ops);
2558 tracing_write_stub(struct file *filp, const char __user *ubuf,
2559 size_t count, loff_t *ppos)
2564 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
2566 if (file->f_mode & FMODE_READ)
2567 return seq_lseek(file, offset, origin);
2572 static const struct file_operations tracing_fops = {
2573 .open = tracing_open,
2575 .write = tracing_write_stub,
2576 .llseek = tracing_seek,
2577 .release = tracing_release,
2580 static const struct file_operations show_traces_fops = {
2581 .open = show_traces_open,
2583 .release = seq_release,
2584 .llseek = seq_lseek,
2588 * Only trace on a CPU if the bitmask is set:
2590 static cpumask_var_t tracing_cpumask;
2593 * The tracer itself will not take this lock, but still we want
2594 * to provide a consistent cpumask to user-space:
2596 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2599 * Temporary storage for the character representation of the
2600 * CPU bitmask (and one more byte for the newline):
2602 static char mask_str[NR_CPUS + 1];
2605 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2606 size_t count, loff_t *ppos)
2610 mutex_lock(&tracing_cpumask_update_lock);
2612 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2613 if (count - len < 2) {
2617 len += sprintf(mask_str + len, "\n");
2618 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2621 mutex_unlock(&tracing_cpumask_update_lock);
2627 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2628 size_t count, loff_t *ppos)
2631 cpumask_var_t tracing_cpumask_new;
2633 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2636 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2640 mutex_lock(&tracing_cpumask_update_lock);
2642 local_irq_disable();
2643 arch_spin_lock(&ftrace_max_lock);
2644 for_each_tracing_cpu(cpu) {
2646 * Increase/decrease the disabled counter if we are
2647 * about to flip a bit in the cpumask:
2649 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2650 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2651 atomic_inc(&global_trace.data[cpu]->disabled);
2652 ring_buffer_record_disable_cpu(global_trace.buffer, cpu);
2654 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2655 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2656 atomic_dec(&global_trace.data[cpu]->disabled);
2657 ring_buffer_record_enable_cpu(global_trace.buffer, cpu);
2660 arch_spin_unlock(&ftrace_max_lock);
2663 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2665 mutex_unlock(&tracing_cpumask_update_lock);
2666 free_cpumask_var(tracing_cpumask_new);
2671 free_cpumask_var(tracing_cpumask_new);
2676 static const struct file_operations tracing_cpumask_fops = {
2677 .open = tracing_open_generic,
2678 .read = tracing_cpumask_read,
2679 .write = tracing_cpumask_write,
2680 .llseek = generic_file_llseek,
2683 static int tracing_trace_options_show(struct seq_file *m, void *v)
2685 struct tracer_opt *trace_opts;
2689 mutex_lock(&trace_types_lock);
2690 tracer_flags = current_trace->flags->val;
2691 trace_opts = current_trace->flags->opts;
2693 for (i = 0; trace_options[i]; i++) {
2694 if (trace_flags & (1 << i))
2695 seq_printf(m, "%s\n", trace_options[i]);
2697 seq_printf(m, "no%s\n", trace_options[i]);
2700 for (i = 0; trace_opts[i].name; i++) {
2701 if (tracer_flags & trace_opts[i].bit)
2702 seq_printf(m, "%s\n", trace_opts[i].name);
2704 seq_printf(m, "no%s\n", trace_opts[i].name);
2706 mutex_unlock(&trace_types_lock);
2711 static int __set_tracer_option(struct tracer *trace,
2712 struct tracer_flags *tracer_flags,
2713 struct tracer_opt *opts, int neg)
2717 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2722 tracer_flags->val &= ~opts->bit;
2724 tracer_flags->val |= opts->bit;
2728 /* Try to assign a tracer specific option */
2729 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2731 struct tracer_flags *tracer_flags = trace->flags;
2732 struct tracer_opt *opts = NULL;
2735 for (i = 0; tracer_flags->opts[i].name; i++) {
2736 opts = &tracer_flags->opts[i];
2738 if (strcmp(cmp, opts->name) == 0)
2739 return __set_tracer_option(trace, trace->flags,
2746 static void set_tracer_flags(unsigned int mask, int enabled)
2748 /* do nothing if flag is already set */
2749 if (!!(trace_flags & mask) == !!enabled)
2753 trace_flags |= mask;
2755 trace_flags &= ~mask;
2757 if (mask == TRACE_ITER_RECORD_CMD)
2758 trace_event_enable_cmd_record(enabled);
2760 if (mask == TRACE_ITER_OVERWRITE)
2761 ring_buffer_change_overwrite(global_trace.buffer, enabled);
2765 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2766 size_t cnt, loff_t *ppos)
2774 if (cnt >= sizeof(buf))
2777 if (copy_from_user(&buf, ubuf, cnt))
2781 cmp = strstrip(buf);
2783 if (strncmp(cmp, "no", 2) == 0) {
2788 for (i = 0; trace_options[i]; i++) {
2789 if (strcmp(cmp, trace_options[i]) == 0) {
2790 set_tracer_flags(1 << i, !neg);
2795 /* If no option could be set, test the specific tracer options */
2796 if (!trace_options[i]) {
2797 mutex_lock(&trace_types_lock);
2798 ret = set_tracer_option(current_trace, cmp, neg);
2799 mutex_unlock(&trace_types_lock);
2809 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2811 if (tracing_disabled)
2813 return single_open(file, tracing_trace_options_show, NULL);
2816 static const struct file_operations tracing_iter_fops = {
2817 .open = tracing_trace_options_open,
2819 .llseek = seq_lseek,
2820 .release = single_release,
2821 .write = tracing_trace_options_write,
2824 static const char readme_msg[] =
2825 "tracing mini-HOWTO:\n\n"
2826 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2827 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2828 "wakeup wakeup_rt preemptirqsoff preemptoff irqsoff function nop\n\n"
2829 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2831 "# echo wakeup > /sys/kernel/debug/tracing/current_tracer\n"
2832 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2834 "# cat /sys/kernel/debug/tracing/trace_options\n"
2835 "noprint-parent nosym-offset nosym-addr noverbose\n"
2836 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2837 "# echo 1 > /sys/kernel/debug/tracing/tracing_on\n"
2838 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2839 "# echo 0 > /sys/kernel/debug/tracing/tracing_on\n"
2843 tracing_readme_read(struct file *filp, char __user *ubuf,
2844 size_t cnt, loff_t *ppos)
2846 return simple_read_from_buffer(ubuf, cnt, ppos,
2847 readme_msg, strlen(readme_msg));
2850 static const struct file_operations tracing_readme_fops = {
2851 .open = tracing_open_generic,
2852 .read = tracing_readme_read,
2853 .llseek = generic_file_llseek,
2857 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2858 size_t cnt, loff_t *ppos)
2867 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2871 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2879 for (i = 0; i < SAVED_CMDLINES; i++) {
2882 pid = map_cmdline_to_pid[i];
2883 if (pid == -1 || pid == NO_CMDLINE_MAP)
2886 trace_find_cmdline(pid, buf_comm);
2887 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2892 len = simple_read_from_buffer(ubuf, cnt, ppos,
2901 static const struct file_operations tracing_saved_cmdlines_fops = {
2902 .open = tracing_open_generic,
2903 .read = tracing_saved_cmdlines_read,
2904 .llseek = generic_file_llseek,
2908 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2909 size_t cnt, loff_t *ppos)
2914 r = sprintf(buf, "%u\n", tracer_enabled);
2915 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2919 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2920 size_t cnt, loff_t *ppos)
2922 struct trace_array *tr = filp->private_data;
2926 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
2932 mutex_lock(&trace_types_lock);
2933 if (tracer_enabled ^ val) {
2935 /* Only need to warn if this is used to change the state */
2936 WARN_ONCE(1, "tracing_enabled is deprecated. Use tracing_on");
2940 if (current_trace->start)
2941 current_trace->start(tr);
2946 if (current_trace->stop)
2947 current_trace->stop(tr);
2950 mutex_unlock(&trace_types_lock);
2958 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2959 size_t cnt, loff_t *ppos)
2961 char buf[MAX_TRACER_SIZE+2];
2964 mutex_lock(&trace_types_lock);
2966 r = sprintf(buf, "%s\n", current_trace->name);
2968 r = sprintf(buf, "\n");
2969 mutex_unlock(&trace_types_lock);
2971 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2974 int tracer_init(struct tracer *t, struct trace_array *tr)
2976 tracing_reset_online_cpus(tr);
2980 static int __tracing_resize_ring_buffer(unsigned long size)
2985 * If kernel or user changes the size of the ring buffer
2986 * we use the size that was given, and we can forget about
2987 * expanding it later.
2989 ring_buffer_expanded = 1;
2991 ret = ring_buffer_resize(global_trace.buffer, size);
2995 if (!current_trace->use_max_tr)
2998 ret = ring_buffer_resize(max_tr.buffer, size);
3002 r = ring_buffer_resize(global_trace.buffer,
3003 global_trace.entries);
3006 * AARGH! We are left with different
3007 * size max buffer!!!!
3008 * The max buffer is our "snapshot" buffer.
3009 * When a tracer needs a snapshot (one of the
3010 * latency tracers), it swaps the max buffer
3011 * with the saved snap shot. We succeeded to
3012 * update the size of the main buffer, but failed to
3013 * update the size of the max buffer. But when we tried
3014 * to reset the main buffer to the original size, we
3015 * failed there too. This is very unlikely to
3016 * happen, but if it does, warn and kill all
3020 tracing_disabled = 1;
3025 max_tr.entries = size;
3027 global_trace.entries = size;
3032 static ssize_t tracing_resize_ring_buffer(unsigned long size)
3034 int cpu, ret = size;
3036 mutex_lock(&trace_types_lock);
3040 /* disable all cpu buffers */
3041 for_each_tracing_cpu(cpu) {
3042 if (global_trace.data[cpu])
3043 atomic_inc(&global_trace.data[cpu]->disabled);
3044 if (max_tr.data[cpu])
3045 atomic_inc(&max_tr.data[cpu]->disabled);
3048 if (size != global_trace.entries)
3049 ret = __tracing_resize_ring_buffer(size);
3054 for_each_tracing_cpu(cpu) {
3055 if (global_trace.data[cpu])
3056 atomic_dec(&global_trace.data[cpu]->disabled);
3057 if (max_tr.data[cpu])
3058 atomic_dec(&max_tr.data[cpu]->disabled);
3062 mutex_unlock(&trace_types_lock);
3069 * tracing_update_buffers - used by tracing facility to expand ring buffers
3071 * To save on memory when the tracing is never used on a system with it
3072 * configured in. The ring buffers are set to a minimum size. But once
3073 * a user starts to use the tracing facility, then they need to grow
3074 * to their default size.
3076 * This function is to be called when a tracer is about to be used.
3078 int tracing_update_buffers(void)
3082 mutex_lock(&trace_types_lock);
3083 if (!ring_buffer_expanded)
3084 ret = __tracing_resize_ring_buffer(trace_buf_size);
3085 mutex_unlock(&trace_types_lock);
3090 struct trace_option_dentry;
3092 static struct trace_option_dentry *
3093 create_trace_option_files(struct tracer *tracer);
3096 destroy_trace_option_files(struct trace_option_dentry *topts);
3098 static int tracing_set_tracer(const char *buf)
3100 static struct trace_option_dentry *topts;
3101 struct trace_array *tr = &global_trace;
3105 mutex_lock(&trace_types_lock);
3107 if (!ring_buffer_expanded) {
3108 ret = __tracing_resize_ring_buffer(trace_buf_size);
3114 for (t = trace_types; t; t = t->next) {
3115 if (strcmp(t->name, buf) == 0)
3122 if (t == current_trace)
3125 trace_branch_disable();
3126 if (current_trace && current_trace->reset)
3127 current_trace->reset(tr);
3128 if (current_trace && current_trace->use_max_tr) {
3130 * We don't free the ring buffer. instead, resize it because
3131 * The max_tr ring buffer has some state (e.g. ring->clock) and
3132 * we want preserve it.
3134 ring_buffer_resize(max_tr.buffer, 1);
3137 destroy_trace_option_files(topts);
3141 topts = create_trace_option_files(current_trace);
3142 if (current_trace->use_max_tr) {
3143 ret = ring_buffer_resize(max_tr.buffer, global_trace.entries);
3146 max_tr.entries = global_trace.entries;
3150 ret = tracer_init(t, tr);
3155 trace_branch_enable(tr);
3157 mutex_unlock(&trace_types_lock);
3163 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3164 size_t cnt, loff_t *ppos)
3166 char buf[MAX_TRACER_SIZE+1];
3173 if (cnt > MAX_TRACER_SIZE)
3174 cnt = MAX_TRACER_SIZE;
3176 if (copy_from_user(&buf, ubuf, cnt))
3181 /* strip ending whitespace. */
3182 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3185 err = tracing_set_tracer(buf);
3195 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3196 size_t cnt, loff_t *ppos)
3198 unsigned long *ptr = filp->private_data;
3202 r = snprintf(buf, sizeof(buf), "%ld\n",
3203 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3204 if (r > sizeof(buf))
3206 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3210 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3211 size_t cnt, loff_t *ppos)
3213 unsigned long *ptr = filp->private_data;
3217 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3226 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3228 long cpu_file = (long) inode->i_private;
3229 struct trace_iterator *iter;
3232 if (tracing_disabled)
3235 mutex_lock(&trace_types_lock);
3237 /* create a buffer to store the information to pass to userspace */
3238 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3245 * We make a copy of the current tracer to avoid concurrent
3246 * changes on it while we are reading.
3248 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3254 *iter->trace = *current_trace;
3256 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3261 /* trace pipe does not show start of buffer */
3262 cpumask_setall(iter->started);
3264 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3265 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3267 iter->cpu_file = cpu_file;
3268 iter->tr = &global_trace;
3269 mutex_init(&iter->mutex);
3270 filp->private_data = iter;
3272 if (iter->trace->pipe_open)
3273 iter->trace->pipe_open(iter);
3275 nonseekable_open(inode, filp);
3277 mutex_unlock(&trace_types_lock);
3283 mutex_unlock(&trace_types_lock);
3287 static int tracing_release_pipe(struct inode *inode, struct file *file)
3289 struct trace_iterator *iter = file->private_data;
3291 mutex_lock(&trace_types_lock);
3293 if (iter->trace->pipe_close)
3294 iter->trace->pipe_close(iter);
3296 mutex_unlock(&trace_types_lock);
3298 free_cpumask_var(iter->started);
3299 mutex_destroy(&iter->mutex);
3307 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3309 struct trace_iterator *iter = filp->private_data;
3311 if (trace_flags & TRACE_ITER_BLOCK) {
3313 * Always select as readable when in blocking mode
3315 return POLLIN | POLLRDNORM;
3317 if (!trace_empty(iter))
3318 return POLLIN | POLLRDNORM;
3319 poll_wait(filp, &trace_wait, poll_table);
3320 if (!trace_empty(iter))
3321 return POLLIN | POLLRDNORM;
3328 void default_wait_pipe(struct trace_iterator *iter)
3332 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3334 if (trace_empty(iter))
3337 finish_wait(&trace_wait, &wait);
3341 * This is a make-shift waitqueue.
3342 * A tracer might use this callback on some rare cases:
3344 * 1) the current tracer might hold the runqueue lock when it wakes up
3345 * a reader, hence a deadlock (sched, function, and function graph tracers)
3346 * 2) the function tracers, trace all functions, we don't want
3347 * the overhead of calling wake_up and friends
3348 * (and tracing them too)
3350 * Anyway, this is really very primitive wakeup.
3352 void poll_wait_pipe(struct trace_iterator *iter)
3354 set_current_state(TASK_INTERRUPTIBLE);
3355 /* sleep for 100 msecs, and try again. */
3356 schedule_timeout(HZ / 10);
3359 /* Must be called with trace_types_lock mutex held. */
3360 static int tracing_wait_pipe(struct file *filp)
3362 struct trace_iterator *iter = filp->private_data;
3364 while (trace_empty(iter)) {
3366 if ((filp->f_flags & O_NONBLOCK)) {
3370 mutex_unlock(&iter->mutex);
3372 iter->trace->wait_pipe(iter);
3374 mutex_lock(&iter->mutex);
3376 if (signal_pending(current))
3380 * We block until we read something and tracing is disabled.
3381 * We still block if tracing is disabled, but we have never
3382 * read anything. This allows a user to cat this file, and
3383 * then enable tracing. But after we have read something,
3384 * we give an EOF when tracing is again disabled.
3386 * iter->pos will be 0 if we haven't read anything.
3388 if (!tracer_enabled && iter->pos)
3399 tracing_read_pipe(struct file *filp, char __user *ubuf,
3400 size_t cnt, loff_t *ppos)
3402 struct trace_iterator *iter = filp->private_data;
3403 static struct tracer *old_tracer;
3406 /* return any leftover data */
3407 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3411 trace_seq_init(&iter->seq);
3413 /* copy the tracer to avoid using a global lock all around */
3414 mutex_lock(&trace_types_lock);
3415 if (unlikely(old_tracer != current_trace && current_trace)) {
3416 old_tracer = current_trace;
3417 *iter->trace = *current_trace;
3419 mutex_unlock(&trace_types_lock);
3422 * Avoid more than one consumer on a single file descriptor
3423 * This is just a matter of traces coherency, the ring buffer itself
3426 mutex_lock(&iter->mutex);
3427 if (iter->trace->read) {
3428 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3434 sret = tracing_wait_pipe(filp);
3438 /* stop when tracing is finished */
3439 if (trace_empty(iter)) {
3444 if (cnt >= PAGE_SIZE)
3445 cnt = PAGE_SIZE - 1;
3447 /* reset all but tr, trace, and overruns */
3448 memset(&iter->seq, 0,
3449 sizeof(struct trace_iterator) -
3450 offsetof(struct trace_iterator, seq));
3453 trace_event_read_lock();
3454 trace_access_lock(iter->cpu_file);
3455 while (trace_find_next_entry_inc(iter) != NULL) {
3456 enum print_line_t ret;
3457 int len = iter->seq.len;
3459 ret = print_trace_line(iter);
3460 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3461 /* don't print partial lines */
3462 iter->seq.len = len;
3465 if (ret != TRACE_TYPE_NO_CONSUME)
3466 trace_consume(iter);
3468 if (iter->seq.len >= cnt)
3472 * Setting the full flag means we reached the trace_seq buffer
3473 * size and we should leave by partial output condition above.
3474 * One of the trace_seq_* functions is not used properly.
3476 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
3479 trace_access_unlock(iter->cpu_file);
3480 trace_event_read_unlock();
3482 /* Now copy what we have to the user */
3483 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3484 if (iter->seq.readpos >= iter->seq.len)
3485 trace_seq_init(&iter->seq);
3488 * If there was nothing to send to user, in spite of consuming trace
3489 * entries, go back to wait for more entries.
3495 mutex_unlock(&iter->mutex);
3500 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3501 struct pipe_buffer *buf)
3503 __free_page(buf->page);
3506 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3509 __free_page(spd->pages[idx]);
3512 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3514 .map = generic_pipe_buf_map,
3515 .unmap = generic_pipe_buf_unmap,
3516 .confirm = generic_pipe_buf_confirm,
3517 .release = tracing_pipe_buf_release,
3518 .steal = generic_pipe_buf_steal,
3519 .get = generic_pipe_buf_get,
3523 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3528 /* Seq buffer is page-sized, exactly what we need. */
3530 count = iter->seq.len;
3531 ret = print_trace_line(iter);
3532 count = iter->seq.len - count;
3535 iter->seq.len -= count;
3538 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3539 iter->seq.len -= count;
3543 if (ret != TRACE_TYPE_NO_CONSUME)
3544 trace_consume(iter);
3546 if (!trace_find_next_entry_inc(iter)) {
3556 static ssize_t tracing_splice_read_pipe(struct file *filp,
3558 struct pipe_inode_info *pipe,
3562 struct page *pages_def[PIPE_DEF_BUFFERS];
3563 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3564 struct trace_iterator *iter = filp->private_data;
3565 struct splice_pipe_desc spd = {
3567 .partial = partial_def,
3568 .nr_pages = 0, /* This gets updated below. */
3569 .nr_pages_max = PIPE_DEF_BUFFERS,
3571 .ops = &tracing_pipe_buf_ops,
3572 .spd_release = tracing_spd_release_pipe,
3574 static struct tracer *old_tracer;
3579 if (splice_grow_spd(pipe, &spd))
3582 /* copy the tracer to avoid using a global lock all around */
3583 mutex_lock(&trace_types_lock);
3584 if (unlikely(old_tracer != current_trace && current_trace)) {
3585 old_tracer = current_trace;
3586 *iter->trace = *current_trace;
3588 mutex_unlock(&trace_types_lock);
3590 mutex_lock(&iter->mutex);
3592 if (iter->trace->splice_read) {
3593 ret = iter->trace->splice_read(iter, filp,
3594 ppos, pipe, len, flags);
3599 ret = tracing_wait_pipe(filp);
3603 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
3608 trace_event_read_lock();
3609 trace_access_lock(iter->cpu_file);
3611 /* Fill as many pages as possible. */
3612 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3613 spd.pages[i] = alloc_page(GFP_KERNEL);
3617 rem = tracing_fill_pipe_page(rem, iter);
3619 /* Copy the data into the page, so we can start over. */
3620 ret = trace_seq_to_buffer(&iter->seq,
3621 page_address(spd.pages[i]),
3624 __free_page(spd.pages[i]);
3627 spd.partial[i].offset = 0;
3628 spd.partial[i].len = iter->seq.len;
3630 trace_seq_init(&iter->seq);
3633 trace_access_unlock(iter->cpu_file);
3634 trace_event_read_unlock();
3635 mutex_unlock(&iter->mutex);
3639 ret = splice_to_pipe(pipe, &spd);
3641 splice_shrink_spd(&spd);
3645 mutex_unlock(&iter->mutex);
3650 tracing_entries_read(struct file *filp, char __user *ubuf,
3651 size_t cnt, loff_t *ppos)
3653 struct trace_array *tr = filp->private_data;
3657 mutex_lock(&trace_types_lock);
3658 if (!ring_buffer_expanded)
3659 r = sprintf(buf, "%lu (expanded: %lu)\n",
3661 trace_buf_size >> 10);
3663 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3664 mutex_unlock(&trace_types_lock);
3666 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3670 tracing_entries_write(struct file *filp, const char __user *ubuf,
3671 size_t cnt, loff_t *ppos)
3676 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3680 /* must have at least 1 entry */
3684 /* value is in KB */
3687 ret = tracing_resize_ring_buffer(val);
3697 tracing_total_entries_read(struct file *filp, char __user *ubuf,
3698 size_t cnt, loff_t *ppos)
3700 struct trace_array *tr = filp->private_data;
3703 unsigned long size = 0, expanded_size = 0;
3705 mutex_lock(&trace_types_lock);
3706 for_each_tracing_cpu(cpu) {
3707 size += tr->entries >> 10;
3708 if (!ring_buffer_expanded)
3709 expanded_size += trace_buf_size >> 10;
3711 if (ring_buffer_expanded)
3712 r = sprintf(buf, "%lu\n", size);
3714 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
3715 mutex_unlock(&trace_types_lock);
3717 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3721 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
3722 size_t cnt, loff_t *ppos)
3725 * There is no need to read what the user has written, this function
3726 * is just to make sure that there is no error when "echo" is used
3735 tracing_free_buffer_release(struct inode *inode, struct file *filp)
3737 /* disable tracing ? */
3738 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
3740 /* resize the ring buffer to 0 */
3741 tracing_resize_ring_buffer(0);
3747 tracing_mark_write(struct file *filp, const char __user *ubuf,
3748 size_t cnt, loff_t *fpos)
3750 unsigned long addr = (unsigned long)ubuf;
3751 struct ring_buffer_event *event;
3752 struct ring_buffer *buffer;
3753 struct print_entry *entry;
3754 unsigned long irq_flags;
3755 struct page *pages[2];
3765 if (tracing_disabled)
3768 if (cnt > TRACE_BUF_SIZE)
3769 cnt = TRACE_BUF_SIZE;
3772 * Userspace is injecting traces into the kernel trace buffer.
3773 * We want to be as non intrusive as possible.
3774 * To do so, we do not want to allocate any special buffers
3775 * or take any locks, but instead write the userspace data
3776 * straight into the ring buffer.
3778 * First we need to pin the userspace buffer into memory,
3779 * which, most likely it is, because it just referenced it.
3780 * But there's no guarantee that it is. By using get_user_pages_fast()
3781 * and kmap_atomic/kunmap_atomic() we can get access to the
3782 * pages directly. We then write the data directly into the
3785 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
3787 /* check if we cross pages */
3788 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
3791 offset = addr & (PAGE_SIZE - 1);
3794 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
3795 if (ret < nr_pages) {
3797 put_page(pages[ret]);
3802 page1 = kmap_atomic(pages[0]);
3804 page2 = kmap_atomic(pages[1]);
3806 local_save_flags(irq_flags);
3807 size = sizeof(*entry) + cnt + 2; /* possible \n added */
3808 buffer = global_trace.buffer;
3809 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3810 irq_flags, preempt_count());
3812 /* Ring buffer disabled, return as if not open for write */
3817 entry = ring_buffer_event_data(event);
3818 entry->ip = _THIS_IP_;
3820 if (nr_pages == 2) {
3821 len = PAGE_SIZE - offset;
3822 memcpy(&entry->buf, page1 + offset, len);
3823 memcpy(&entry->buf[len], page2, cnt - len);
3825 memcpy(&entry->buf, page1 + offset, cnt);
3827 if (entry->buf[cnt - 1] != '\n') {
3828 entry->buf[cnt] = '\n';
3829 entry->buf[cnt + 1] = '\0';
3831 entry->buf[cnt] = '\0';
3833 ring_buffer_unlock_commit(buffer, event);
3841 kunmap_atomic(page2);
3842 kunmap_atomic(page1);
3843 while (nr_pages > 0)
3844 put_page(pages[--nr_pages]);
3849 static int tracing_clock_show(struct seq_file *m, void *v)
3853 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3855 "%s%s%s%s", i ? " " : "",
3856 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3857 i == trace_clock_id ? "]" : "");
3863 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3864 size_t cnt, loff_t *fpos)
3867 const char *clockstr;
3870 if (cnt >= sizeof(buf))
3873 if (copy_from_user(&buf, ubuf, cnt))
3878 clockstr = strstrip(buf);
3880 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3881 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3884 if (i == ARRAY_SIZE(trace_clocks))
3889 mutex_lock(&trace_types_lock);
3891 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3893 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3895 mutex_unlock(&trace_types_lock);
3902 static int tracing_clock_open(struct inode *inode, struct file *file)
3904 if (tracing_disabled)
3906 return single_open(file, tracing_clock_show, NULL);
3909 static const struct file_operations tracing_max_lat_fops = {
3910 .open = tracing_open_generic,
3911 .read = tracing_max_lat_read,
3912 .write = tracing_max_lat_write,
3913 .llseek = generic_file_llseek,
3916 static const struct file_operations tracing_ctrl_fops = {
3917 .open = tracing_open_generic,
3918 .read = tracing_ctrl_read,
3919 .write = tracing_ctrl_write,
3920 .llseek = generic_file_llseek,
3923 static const struct file_operations set_tracer_fops = {
3924 .open = tracing_open_generic,
3925 .read = tracing_set_trace_read,
3926 .write = tracing_set_trace_write,
3927 .llseek = generic_file_llseek,
3930 static const struct file_operations tracing_pipe_fops = {
3931 .open = tracing_open_pipe,
3932 .poll = tracing_poll_pipe,
3933 .read = tracing_read_pipe,
3934 .splice_read = tracing_splice_read_pipe,
3935 .release = tracing_release_pipe,
3936 .llseek = no_llseek,
3939 static const struct file_operations tracing_entries_fops = {
3940 .open = tracing_open_generic,
3941 .read = tracing_entries_read,
3942 .write = tracing_entries_write,
3943 .llseek = generic_file_llseek,
3946 static const struct file_operations tracing_total_entries_fops = {
3947 .open = tracing_open_generic,
3948 .read = tracing_total_entries_read,
3949 .llseek = generic_file_llseek,
3952 static const struct file_operations tracing_free_buffer_fops = {
3953 .write = tracing_free_buffer_write,
3954 .release = tracing_free_buffer_release,
3957 static const struct file_operations tracing_mark_fops = {
3958 .open = tracing_open_generic,
3959 .write = tracing_mark_write,
3960 .llseek = generic_file_llseek,
3963 static const struct file_operations trace_clock_fops = {
3964 .open = tracing_clock_open,
3966 .llseek = seq_lseek,
3967 .release = single_release,
3968 .write = tracing_clock_write,
3971 struct ftrace_buffer_info {
3972 struct trace_array *tr;
3978 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3980 int cpu = (int)(long)inode->i_private;
3981 struct ftrace_buffer_info *info;
3983 if (tracing_disabled)
3986 info = kzalloc(sizeof(*info), GFP_KERNEL);
3990 info->tr = &global_trace;
3993 /* Force reading ring buffer for first read */
3994 info->read = (unsigned int)-1;
3996 filp->private_data = info;
3998 return nonseekable_open(inode, filp);
4002 tracing_buffers_read(struct file *filp, char __user *ubuf,
4003 size_t count, loff_t *ppos)
4005 struct ftrace_buffer_info *info = filp->private_data;
4013 info->spare = ring_buffer_alloc_read_page(info->tr->buffer, info->cpu);
4017 /* Do we have previous read data to read? */
4018 if (info->read < PAGE_SIZE)
4021 trace_access_lock(info->cpu);
4022 ret = ring_buffer_read_page(info->tr->buffer,
4026 trace_access_unlock(info->cpu);
4033 size = PAGE_SIZE - info->read;
4037 ret = copy_to_user(ubuf, info->spare + info->read, size);
4048 static int tracing_buffers_release(struct inode *inode, struct file *file)
4050 struct ftrace_buffer_info *info = file->private_data;
4053 ring_buffer_free_read_page(info->tr->buffer, info->spare);
4060 struct ring_buffer *buffer;
4065 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
4066 struct pipe_buffer *buf)
4068 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4073 ring_buffer_free_read_page(ref->buffer, ref->page);
4078 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
4079 struct pipe_buffer *buf)
4084 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
4085 struct pipe_buffer *buf)
4087 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4092 /* Pipe buffer operations for a buffer. */
4093 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
4095 .map = generic_pipe_buf_map,
4096 .unmap = generic_pipe_buf_unmap,
4097 .confirm = generic_pipe_buf_confirm,
4098 .release = buffer_pipe_buf_release,
4099 .steal = buffer_pipe_buf_steal,
4100 .get = buffer_pipe_buf_get,
4104 * Callback from splice_to_pipe(), if we need to release some pages
4105 * at the end of the spd in case we error'ed out in filling the pipe.
4107 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
4109 struct buffer_ref *ref =
4110 (struct buffer_ref *)spd->partial[i].private;
4115 ring_buffer_free_read_page(ref->buffer, ref->page);
4117 spd->partial[i].private = 0;
4121 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4122 struct pipe_inode_info *pipe, size_t len,
4125 struct ftrace_buffer_info *info = file->private_data;
4126 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4127 struct page *pages_def[PIPE_DEF_BUFFERS];
4128 struct splice_pipe_desc spd = {
4130 .partial = partial_def,
4131 .nr_pages_max = PIPE_DEF_BUFFERS,
4133 .ops = &buffer_pipe_buf_ops,
4134 .spd_release = buffer_spd_release,
4136 struct buffer_ref *ref;
4137 int entries, size, i;
4140 if (splice_grow_spd(pipe, &spd))
4143 if (*ppos & (PAGE_SIZE - 1)) {
4144 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
4149 if (len & (PAGE_SIZE - 1)) {
4150 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
4151 if (len < PAGE_SIZE) {
4158 trace_access_lock(info->cpu);
4159 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
4161 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
4165 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
4170 ref->buffer = info->tr->buffer;
4171 ref->page = ring_buffer_alloc_read_page(ref->buffer, info->cpu);
4177 r = ring_buffer_read_page(ref->buffer, &ref->page,
4180 ring_buffer_free_read_page(ref->buffer, ref->page);
4186 * zero out any left over data, this is going to
4189 size = ring_buffer_page_len(ref->page);
4190 if (size < PAGE_SIZE)
4191 memset(ref->page + size, 0, PAGE_SIZE - size);
4193 page = virt_to_page(ref->page);
4195 spd.pages[i] = page;
4196 spd.partial[i].len = PAGE_SIZE;
4197 spd.partial[i].offset = 0;
4198 spd.partial[i].private = (unsigned long)ref;
4202 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
4205 trace_access_unlock(info->cpu);
4208 /* did we read anything? */
4209 if (!spd.nr_pages) {
4210 if (flags & SPLICE_F_NONBLOCK)
4218 ret = splice_to_pipe(pipe, &spd);
4219 splice_shrink_spd(&spd);
4224 static const struct file_operations tracing_buffers_fops = {
4225 .open = tracing_buffers_open,
4226 .read = tracing_buffers_read,
4227 .release = tracing_buffers_release,
4228 .splice_read = tracing_buffers_splice_read,
4229 .llseek = no_llseek,
4233 tracing_stats_read(struct file *filp, char __user *ubuf,
4234 size_t count, loff_t *ppos)
4236 unsigned long cpu = (unsigned long)filp->private_data;
4237 struct trace_array *tr = &global_trace;
4238 struct trace_seq *s;
4240 unsigned long long t;
4241 unsigned long usec_rem;
4243 s = kmalloc(sizeof(*s), GFP_KERNEL);
4249 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
4250 trace_seq_printf(s, "entries: %ld\n", cnt);
4252 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
4253 trace_seq_printf(s, "overrun: %ld\n", cnt);
4255 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
4256 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
4258 cnt = ring_buffer_bytes_cpu(tr->buffer, cpu);
4259 trace_seq_printf(s, "bytes: %ld\n", cnt);
4261 t = ns2usecs(ring_buffer_oldest_event_ts(tr->buffer, cpu));
4262 usec_rem = do_div(t, USEC_PER_SEC);
4263 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n", t, usec_rem);
4265 t = ns2usecs(ring_buffer_time_stamp(tr->buffer, cpu));
4266 usec_rem = do_div(t, USEC_PER_SEC);
4267 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
4269 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
4276 static const struct file_operations tracing_stats_fops = {
4277 .open = tracing_open_generic,
4278 .read = tracing_stats_read,
4279 .llseek = generic_file_llseek,
4282 #ifdef CONFIG_DYNAMIC_FTRACE
4284 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
4290 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
4291 size_t cnt, loff_t *ppos)
4293 static char ftrace_dyn_info_buffer[1024];
4294 static DEFINE_MUTEX(dyn_info_mutex);
4295 unsigned long *p = filp->private_data;
4296 char *buf = ftrace_dyn_info_buffer;
4297 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
4300 mutex_lock(&dyn_info_mutex);
4301 r = sprintf(buf, "%ld ", *p);
4303 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
4306 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4308 mutex_unlock(&dyn_info_mutex);
4313 static const struct file_operations tracing_dyn_info_fops = {
4314 .open = tracing_open_generic,
4315 .read = tracing_read_dyn_info,
4316 .llseek = generic_file_llseek,
4320 static struct dentry *d_tracer;
4322 struct dentry *tracing_init_dentry(void)
4329 if (!debugfs_initialized())
4332 d_tracer = debugfs_create_dir("tracing", NULL);
4334 if (!d_tracer && !once) {
4336 pr_warning("Could not create debugfs directory 'tracing'\n");
4343 static struct dentry *d_percpu;
4345 struct dentry *tracing_dentry_percpu(void)
4348 struct dentry *d_tracer;
4353 d_tracer = tracing_init_dentry();
4358 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
4360 if (!d_percpu && !once) {
4362 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4369 static void tracing_init_debugfs_percpu(long cpu)
4371 struct dentry *d_percpu = tracing_dentry_percpu();
4372 struct dentry *d_cpu;
4373 char cpu_dir[30]; /* 30 characters should be more than enough */
4375 snprintf(cpu_dir, 30, "cpu%ld", cpu);
4376 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4378 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4382 /* per cpu trace_pipe */
4383 trace_create_file("trace_pipe", 0444, d_cpu,
4384 (void *) cpu, &tracing_pipe_fops);
4387 trace_create_file("trace", 0644, d_cpu,
4388 (void *) cpu, &tracing_fops);
4390 trace_create_file("trace_pipe_raw", 0444, d_cpu,
4391 (void *) cpu, &tracing_buffers_fops);
4393 trace_create_file("stats", 0444, d_cpu,
4394 (void *) cpu, &tracing_stats_fops);
4397 #ifdef CONFIG_FTRACE_SELFTEST
4398 /* Let selftest have access to static functions in this file */
4399 #include "trace_selftest.c"
4402 struct trace_option_dentry {
4403 struct tracer_opt *opt;
4404 struct tracer_flags *flags;
4405 struct dentry *entry;
4409 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4412 struct trace_option_dentry *topt = filp->private_data;
4415 if (topt->flags->val & topt->opt->bit)
4420 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4424 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4427 struct trace_option_dentry *topt = filp->private_data;
4431 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4435 if (val != 0 && val != 1)
4438 if (!!(topt->flags->val & topt->opt->bit) != val) {
4439 mutex_lock(&trace_types_lock);
4440 ret = __set_tracer_option(current_trace, topt->flags,
4442 mutex_unlock(&trace_types_lock);
4453 static const struct file_operations trace_options_fops = {
4454 .open = tracing_open_generic,
4455 .read = trace_options_read,
4456 .write = trace_options_write,
4457 .llseek = generic_file_llseek,
4461 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4464 long index = (long)filp->private_data;
4467 if (trace_flags & (1 << index))
4472 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4476 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4479 long index = (long)filp->private_data;
4483 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4487 if (val != 0 && val != 1)
4489 set_tracer_flags(1 << index, val);
4496 static const struct file_operations trace_options_core_fops = {
4497 .open = tracing_open_generic,
4498 .read = trace_options_core_read,
4499 .write = trace_options_core_write,
4500 .llseek = generic_file_llseek,
4503 struct dentry *trace_create_file(const char *name,
4505 struct dentry *parent,
4507 const struct file_operations *fops)
4511 ret = debugfs_create_file(name, mode, parent, data, fops);
4513 pr_warning("Could not create debugfs '%s' entry\n", name);
4519 static struct dentry *trace_options_init_dentry(void)
4521 struct dentry *d_tracer;
4522 static struct dentry *t_options;
4527 d_tracer = tracing_init_dentry();
4531 t_options = debugfs_create_dir("options", d_tracer);
4533 pr_warning("Could not create debugfs directory 'options'\n");
4541 create_trace_option_file(struct trace_option_dentry *topt,
4542 struct tracer_flags *flags,
4543 struct tracer_opt *opt)
4545 struct dentry *t_options;
4547 t_options = trace_options_init_dentry();
4551 topt->flags = flags;
4554 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4555 &trace_options_fops);
4559 static struct trace_option_dentry *
4560 create_trace_option_files(struct tracer *tracer)
4562 struct trace_option_dentry *topts;
4563 struct tracer_flags *flags;
4564 struct tracer_opt *opts;
4570 flags = tracer->flags;
4572 if (!flags || !flags->opts)
4577 for (cnt = 0; opts[cnt].name; cnt++)
4580 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4584 for (cnt = 0; opts[cnt].name; cnt++)
4585 create_trace_option_file(&topts[cnt], flags,
4592 destroy_trace_option_files(struct trace_option_dentry *topts)
4599 for (cnt = 0; topts[cnt].opt; cnt++) {
4600 if (topts[cnt].entry)
4601 debugfs_remove(topts[cnt].entry);
4607 static struct dentry *
4608 create_trace_option_core_file(const char *option, long index)
4610 struct dentry *t_options;
4612 t_options = trace_options_init_dentry();
4616 return trace_create_file(option, 0644, t_options, (void *)index,
4617 &trace_options_core_fops);
4620 static __init void create_trace_options_dir(void)
4622 struct dentry *t_options;
4625 t_options = trace_options_init_dentry();
4629 for (i = 0; trace_options[i]; i++)
4630 create_trace_option_core_file(trace_options[i], i);
4634 rb_simple_read(struct file *filp, char __user *ubuf,
4635 size_t cnt, loff_t *ppos)
4637 struct trace_array *tr = filp->private_data;
4638 struct ring_buffer *buffer = tr->buffer;
4643 r = ring_buffer_record_is_on(buffer);
4647 r = sprintf(buf, "%d\n", r);
4649 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4653 rb_simple_write(struct file *filp, const char __user *ubuf,
4654 size_t cnt, loff_t *ppos)
4656 struct trace_array *tr = filp->private_data;
4657 struct ring_buffer *buffer = tr->buffer;
4661 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4667 ring_buffer_record_on(buffer);
4669 ring_buffer_record_off(buffer);
4677 static const struct file_operations rb_simple_fops = {
4678 .open = tracing_open_generic,
4679 .read = rb_simple_read,
4680 .write = rb_simple_write,
4681 .llseek = default_llseek,
4684 static __init int tracer_init_debugfs(void)
4686 struct dentry *d_tracer;
4689 trace_access_lock_init();
4691 d_tracer = tracing_init_dentry();
4693 trace_create_file("tracing_enabled", 0644, d_tracer,
4694 &global_trace, &tracing_ctrl_fops);
4696 trace_create_file("trace_options", 0644, d_tracer,
4697 NULL, &tracing_iter_fops);
4699 trace_create_file("tracing_cpumask", 0644, d_tracer,
4700 NULL, &tracing_cpumask_fops);
4702 trace_create_file("trace", 0644, d_tracer,
4703 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4705 trace_create_file("available_tracers", 0444, d_tracer,
4706 &global_trace, &show_traces_fops);
4708 trace_create_file("current_tracer", 0644, d_tracer,
4709 &global_trace, &set_tracer_fops);
4711 #ifdef CONFIG_TRACER_MAX_TRACE
4712 trace_create_file("tracing_max_latency", 0644, d_tracer,
4713 &tracing_max_latency, &tracing_max_lat_fops);
4716 trace_create_file("tracing_thresh", 0644, d_tracer,
4717 &tracing_thresh, &tracing_max_lat_fops);
4719 trace_create_file("README", 0444, d_tracer,
4720 NULL, &tracing_readme_fops);
4722 trace_create_file("trace_pipe", 0444, d_tracer,
4723 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4725 trace_create_file("buffer_size_kb", 0644, d_tracer,
4726 &global_trace, &tracing_entries_fops);
4728 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
4729 &global_trace, &tracing_total_entries_fops);
4731 trace_create_file("free_buffer", 0644, d_tracer,
4732 &global_trace, &tracing_free_buffer_fops);
4734 trace_create_file("trace_marker", 0220, d_tracer,
4735 NULL, &tracing_mark_fops);
4737 trace_create_file("saved_cmdlines", 0444, d_tracer,
4738 NULL, &tracing_saved_cmdlines_fops);
4740 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4743 trace_create_file("tracing_on", 0644, d_tracer,
4744 &global_trace, &rb_simple_fops);
4746 #ifdef CONFIG_DYNAMIC_FTRACE
4747 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4748 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4751 create_trace_options_dir();
4753 for_each_tracing_cpu(cpu)
4754 tracing_init_debugfs_percpu(cpu);
4759 static int trace_panic_handler(struct notifier_block *this,
4760 unsigned long event, void *unused)
4762 if (ftrace_dump_on_oops)
4763 ftrace_dump(ftrace_dump_on_oops);
4767 static struct notifier_block trace_panic_notifier = {
4768 .notifier_call = trace_panic_handler,
4770 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4773 static int trace_die_handler(struct notifier_block *self,
4779 if (ftrace_dump_on_oops)
4780 ftrace_dump(ftrace_dump_on_oops);
4788 static struct notifier_block trace_die_notifier = {
4789 .notifier_call = trace_die_handler,
4794 * printk is set to max of 1024, we really don't need it that big.
4795 * Nothing should be printing 1000 characters anyway.
4797 #define TRACE_MAX_PRINT 1000
4800 * Define here KERN_TRACE so that we have one place to modify
4801 * it if we decide to change what log level the ftrace dump
4804 #define KERN_TRACE KERN_EMERG
4807 trace_printk_seq(struct trace_seq *s)
4809 /* Probably should print a warning here. */
4813 /* should be zero ended, but we are paranoid. */
4814 s->buffer[s->len] = 0;
4816 printk(KERN_TRACE "%s", s->buffer);
4821 void trace_init_global_iter(struct trace_iterator *iter)
4823 iter->tr = &global_trace;
4824 iter->trace = current_trace;
4825 iter->cpu_file = TRACE_PIPE_ALL_CPU;
4829 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4831 static arch_spinlock_t ftrace_dump_lock =
4832 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4833 /* use static because iter can be a bit big for the stack */
4834 static struct trace_iterator iter;
4835 unsigned int old_userobj;
4836 static int dump_ran;
4837 unsigned long flags;
4841 local_irq_save(flags);
4842 arch_spin_lock(&ftrace_dump_lock);
4850 /* Did function tracer already get disabled? */
4851 if (ftrace_is_dead()) {
4852 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
4853 printk("# MAY BE MISSING FUNCTION EVENTS\n");
4856 if (disable_tracing)
4859 trace_init_global_iter(&iter);
4861 for_each_tracing_cpu(cpu) {
4862 atomic_inc(&iter.tr->data[cpu]->disabled);
4865 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4867 /* don't look at user memory in panic mode */
4868 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4870 /* Simulate the iterator */
4871 iter.tr = &global_trace;
4872 iter.trace = current_trace;
4874 switch (oops_dump_mode) {
4876 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4879 iter.cpu_file = raw_smp_processor_id();
4884 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
4885 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4888 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4891 * We need to stop all tracing on all CPUS to read the
4892 * the next buffer. This is a bit expensive, but is
4893 * not done often. We fill all what we can read,
4894 * and then release the locks again.
4897 while (!trace_empty(&iter)) {
4900 printk(KERN_TRACE "---------------------------------\n");
4904 /* reset all but tr, trace, and overruns */
4905 memset(&iter.seq, 0,
4906 sizeof(struct trace_iterator) -
4907 offsetof(struct trace_iterator, seq));
4908 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4911 if (trace_find_next_entry_inc(&iter) != NULL) {
4914 ret = print_trace_line(&iter);
4915 if (ret != TRACE_TYPE_NO_CONSUME)
4916 trace_consume(&iter);
4918 touch_nmi_watchdog();
4920 trace_printk_seq(&iter.seq);
4924 printk(KERN_TRACE " (ftrace buffer empty)\n");
4926 printk(KERN_TRACE "---------------------------------\n");
4929 /* Re-enable tracing if requested */
4930 if (!disable_tracing) {
4931 trace_flags |= old_userobj;
4933 for_each_tracing_cpu(cpu) {
4934 atomic_dec(&iter.tr->data[cpu]->disabled);
4940 arch_spin_unlock(&ftrace_dump_lock);
4941 local_irq_restore(flags);
4944 /* By default: disable tracing after the dump */
4945 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
4947 __ftrace_dump(true, oops_dump_mode);
4949 EXPORT_SYMBOL_GPL(ftrace_dump);
4951 __init static int tracer_alloc_buffers(void)
4954 enum ring_buffer_flags rb_flags;
4959 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4962 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4963 goto out_free_buffer_mask;
4965 /* To save memory, keep the ring buffer size to its minimum */
4966 if (ring_buffer_expanded)
4967 ring_buf_size = trace_buf_size;
4971 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
4973 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4974 cpumask_copy(tracing_cpumask, cpu_all_mask);
4976 /* TODO: make the number of buffers hot pluggable with CPUS */
4977 global_trace.buffer = ring_buffer_alloc(ring_buf_size, rb_flags);
4978 if (!global_trace.buffer) {
4979 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4981 goto out_free_cpumask;
4983 global_trace.entries = ring_buffer_size(global_trace.buffer);
4984 if (global_trace.buffer_disabled)
4988 #ifdef CONFIG_TRACER_MAX_TRACE
4989 max_tr.buffer = ring_buffer_alloc(1, rb_flags);
4990 if (!max_tr.buffer) {
4991 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4993 ring_buffer_free(global_trace.buffer);
4994 goto out_free_cpumask;
4999 /* Allocate the first page for all buffers */
5000 for_each_tracing_cpu(i) {
5001 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
5002 max_tr.data[i] = &per_cpu(max_tr_data, i);
5005 trace_init_cmdlines();
5007 register_tracer(&nop_trace);
5008 current_trace = &nop_trace;
5009 /* All seems OK, enable tracing */
5010 tracing_disabled = 0;
5012 atomic_notifier_chain_register(&panic_notifier_list,
5013 &trace_panic_notifier);
5015 register_die_notifier(&trace_die_notifier);
5020 free_cpumask_var(tracing_cpumask);
5021 out_free_buffer_mask:
5022 free_cpumask_var(tracing_buffer_mask);
5027 __init static int clear_boot_tracer(void)
5030 * The default tracer at boot buffer is an init section.
5031 * This function is called in lateinit. If we did not
5032 * find the boot tracer, then clear it out, to prevent
5033 * later registration from accessing the buffer that is
5034 * about to be freed.
5036 if (!default_bootup_tracer)
5039 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
5040 default_bootup_tracer);
5041 default_bootup_tracer = NULL;
5046 early_initcall(tracer_alloc_buffers);
5047 fs_initcall(tracer_init_debugfs);
5048 late_initcall(clear_boot_tracer);