2 * ring buffer based function tracer
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 * Based on code from the latency_tracer, that is:
11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 William Lee Irwin III
14 #include <linux/ring_buffer.h>
15 #include <generated/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/smp_lock.h>
21 #include <linux/notifier.h>
22 #include <linux/irqflags.h>
23 #include <linux/debugfs.h>
24 #include <linux/pagemap.h>
25 #include <linux/hardirq.h>
26 #include <linux/linkage.h>
27 #include <linux/uaccess.h>
28 #include <linux/kprobes.h>
29 #include <linux/ftrace.h>
30 #include <linux/module.h>
31 #include <linux/percpu.h>
32 #include <linux/splice.h>
33 #include <linux/kdebug.h>
34 #include <linux/string.h>
35 #include <linux/rwsem.h>
36 #include <linux/ctype.h>
37 #include <linux/init.h>
38 #include <linux/poll.h>
39 #include <linux/gfp.h>
43 #include "trace_output.h"
45 #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
48 * On boot up, the ring buffer is set to the minimum size, so that
49 * we do not waste memory on systems that are not using tracing.
51 int ring_buffer_expanded;
54 * We need to change this state when a selftest is running.
55 * A selftest will lurk into the ring-buffer to count the
56 * entries inserted during the selftest although some concurrent
57 * insertions into the ring-buffer such as trace_printk could occurred
58 * at the same time, giving false positive or negative results.
60 static bool __read_mostly tracing_selftest_running;
63 * If a tracer is running, we do not want to run SELFTEST.
65 bool __read_mostly tracing_selftest_disabled;
67 /* For tracers that don't implement custom flags */
68 static struct tracer_opt dummy_tracer_opt[] = {
72 static struct tracer_flags dummy_tracer_flags = {
74 .opts = dummy_tracer_opt
77 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
83 * Kill all tracing for good (never come back).
84 * It is initialized to 1 but will turn to zero if the initialization
85 * of the tracer is successful. But that is the only place that sets
88 static int tracing_disabled = 1;
90 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
92 static inline void ftrace_disable_cpu(void)
95 __this_cpu_inc(per_cpu_var(ftrace_cpu_disabled));
98 static inline void ftrace_enable_cpu(void)
100 __this_cpu_dec(per_cpu_var(ftrace_cpu_disabled));
104 static cpumask_var_t __read_mostly tracing_buffer_mask;
106 #define for_each_tracing_cpu(cpu) \
107 for_each_cpu(cpu, tracing_buffer_mask)
110 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
112 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
113 * is set, then ftrace_dump is called. This will output the contents
114 * of the ftrace buffers to the console. This is very useful for
115 * capturing traces that lead to crashes and outputing it to a
118 * It is default off, but you can enable it with either specifying
119 * "ftrace_dump_on_oops" in the kernel command line, or setting
120 * /proc/sys/kernel/ftrace_dump_on_oops to true.
122 int ftrace_dump_on_oops;
124 static int tracing_set_tracer(const char *buf);
126 #define MAX_TRACER_SIZE 100
127 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
128 static char *default_bootup_tracer;
130 static int __init set_cmdline_ftrace(char *str)
132 strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
133 default_bootup_tracer = bootup_tracer_buf;
134 /* We are using ftrace early, expand it */
135 ring_buffer_expanded = 1;
138 __setup("ftrace=", set_cmdline_ftrace);
140 static int __init set_ftrace_dump_on_oops(char *str)
142 ftrace_dump_on_oops = 1;
145 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
147 unsigned long long ns2usecs(cycle_t nsec)
155 * The global_trace is the descriptor that holds the tracing
156 * buffers for the live tracing. For each CPU, it contains
157 * a link list of pages that will store trace entries. The
158 * page descriptor of the pages in the memory is used to hold
159 * the link list by linking the lru item in the page descriptor
160 * to each of the pages in the buffer per CPU.
162 * For each active CPU there is a data field that holds the
163 * pages for the buffer for that CPU. Each CPU has the same number
164 * of pages allocated for its buffer.
166 static struct trace_array global_trace;
168 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
170 int filter_current_check_discard(struct ring_buffer *buffer,
171 struct ftrace_event_call *call, void *rec,
172 struct ring_buffer_event *event)
174 return filter_check_discard(call, rec, buffer, event);
176 EXPORT_SYMBOL_GPL(filter_current_check_discard);
178 cycle_t ftrace_now(int cpu)
182 /* Early boot up does not have a buffer yet */
183 if (!global_trace.buffer)
184 return trace_clock_local();
186 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
187 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
193 * The max_tr is used to snapshot the global_trace when a maximum
194 * latency is reached. Some tracers will use this to store a maximum
195 * trace while it continues examining live traces.
197 * The buffers for the max_tr are set up the same as the global_trace.
198 * When a snapshot is taken, the link list of the max_tr is swapped
199 * with the link list of the global_trace and the buffers are reset for
200 * the global_trace so the tracing can continue.
202 static struct trace_array max_tr;
204 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
206 /* tracer_enabled is used to toggle activation of a tracer */
207 static int tracer_enabled = 1;
210 * tracing_is_enabled - return tracer_enabled status
212 * This function is used by other tracers to know the status
213 * of the tracer_enabled flag. Tracers may use this function
214 * to know if it should enable their features when starting
215 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
217 int tracing_is_enabled(void)
219 return tracer_enabled;
223 * trace_buf_size is the size in bytes that is allocated
224 * for a buffer. Note, the number of bytes is always rounded
227 * This number is purposely set to a low number of 16384.
228 * If the dump on oops happens, it will be much appreciated
229 * to not have to wait for all that output. Anyway this can be
230 * boot time and run time configurable.
232 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
234 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
236 /* trace_types holds a link list of available tracers. */
237 static struct tracer *trace_types __read_mostly;
239 /* current_trace points to the tracer that is currently active */
240 static struct tracer *current_trace __read_mostly;
243 * trace_types_lock is used to protect the trace_types list.
245 static DEFINE_MUTEX(trace_types_lock);
248 * serialize the access of the ring buffer
250 * ring buffer serializes readers, but it is low level protection.
251 * The validity of the events (which returns by ring_buffer_peek() ..etc)
252 * are not protected by ring buffer.
254 * The content of events may become garbage if we allow other process consumes
255 * these events concurrently:
256 * A) the page of the consumed events may become a normal page
257 * (not reader page) in ring buffer, and this page will be rewrited
258 * by events producer.
259 * B) The page of the consumed events may become a page for splice_read,
260 * and this page will be returned to system.
262 * These primitives allow multi process access to different cpu ring buffer
265 * These primitives don't distinguish read-only and read-consume access.
266 * Multi read-only access are also serialized.
270 static DECLARE_RWSEM(all_cpu_access_lock);
271 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
273 static inline void trace_access_lock(int cpu)
275 if (cpu == TRACE_PIPE_ALL_CPU) {
276 /* gain it for accessing the whole ring buffer. */
277 down_write(&all_cpu_access_lock);
279 /* gain it for accessing a cpu ring buffer. */
281 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
282 down_read(&all_cpu_access_lock);
284 /* Secondly block other access to this @cpu ring buffer. */
285 mutex_lock(&per_cpu(cpu_access_lock, cpu));
289 static inline void trace_access_unlock(int cpu)
291 if (cpu == TRACE_PIPE_ALL_CPU) {
292 up_write(&all_cpu_access_lock);
294 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
295 up_read(&all_cpu_access_lock);
299 static inline void trace_access_lock_init(void)
303 for_each_possible_cpu(cpu)
304 mutex_init(&per_cpu(cpu_access_lock, cpu));
309 static DEFINE_MUTEX(access_lock);
311 static inline void trace_access_lock(int cpu)
314 mutex_lock(&access_lock);
317 static inline void trace_access_unlock(int cpu)
320 mutex_unlock(&access_lock);
323 static inline void trace_access_lock_init(void)
329 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
330 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
332 /* trace_flags holds trace_options default values */
333 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
334 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
335 TRACE_ITER_GRAPH_TIME;
337 static int trace_stop_count;
338 static DEFINE_SPINLOCK(tracing_start_lock);
341 * trace_wake_up - wake up tasks waiting for trace input
343 * Simply wakes up any task that is blocked on the trace_wait
344 * queue. These is used with trace_poll for tasks polling the trace.
346 void trace_wake_up(void)
350 if (trace_flags & TRACE_ITER_BLOCK)
353 * The runqueue_is_locked() can fail, but this is the best we
357 if (!runqueue_is_locked(cpu))
358 wake_up(&trace_wait);
362 static int __init set_buf_size(char *str)
364 unsigned long buf_size;
368 buf_size = memparse(str, &str);
369 /* nr_entries can not be zero */
372 trace_buf_size = buf_size;
375 __setup("trace_buf_size=", set_buf_size);
377 unsigned long nsecs_to_usecs(unsigned long nsecs)
382 /* These must match the bit postions in trace_iterator_flags */
383 static const char *trace_options[] = {
411 { trace_clock_local, "local" },
412 { trace_clock_global, "global" },
418 * trace_parser_get_init - gets the buffer for trace parser
420 int trace_parser_get_init(struct trace_parser *parser, int size)
422 memset(parser, 0, sizeof(*parser));
424 parser->buffer = kmalloc(size, GFP_KERNEL);
433 * trace_parser_put - frees the buffer for trace parser
435 void trace_parser_put(struct trace_parser *parser)
437 kfree(parser->buffer);
441 * trace_get_user - reads the user input string separated by space
442 * (matched by isspace(ch))
444 * For each string found the 'struct trace_parser' is updated,
445 * and the function returns.
447 * Returns number of bytes read.
449 * See kernel/trace/trace.h for 'struct trace_parser' details.
451 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
452 size_t cnt, loff_t *ppos)
459 trace_parser_clear(parser);
461 ret = get_user(ch, ubuf++);
469 * The parser is not finished with the last write,
470 * continue reading the user input without skipping spaces.
473 /* skip white space */
474 while (cnt && isspace(ch)) {
475 ret = get_user(ch, ubuf++);
482 /* only spaces were written */
492 /* read the non-space input */
493 while (cnt && !isspace(ch)) {
494 if (parser->idx < parser->size - 1)
495 parser->buffer[parser->idx++] = ch;
500 ret = get_user(ch, ubuf++);
507 /* We either got finished input or we have to wait for another call. */
509 parser->buffer[parser->idx] = 0;
510 parser->cont = false;
513 parser->buffer[parser->idx++] = ch;
523 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
531 if (s->len <= s->readpos)
534 len = s->len - s->readpos;
537 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
547 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
552 if (s->len <= s->readpos)
555 len = s->len - s->readpos;
558 ret = memcpy(buf, s->buffer + s->readpos, cnt);
567 * ftrace_max_lock is used to protect the swapping of buffers
568 * when taking a max snapshot. The buffers themselves are
569 * protected by per_cpu spinlocks. But the action of the swap
570 * needs its own lock.
572 * This is defined as a arch_spinlock_t in order to help
573 * with performance when lockdep debugging is enabled.
575 * It is also used in other places outside the update_max_tr
576 * so it needs to be defined outside of the
577 * CONFIG_TRACER_MAX_TRACE.
579 static arch_spinlock_t ftrace_max_lock =
580 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
582 #ifdef CONFIG_TRACER_MAX_TRACE
583 unsigned long __read_mostly tracing_max_latency;
584 unsigned long __read_mostly tracing_thresh;
587 * Copy the new maximum trace into the separate maximum-trace
588 * structure. (this way the maximum trace is permanently saved,
589 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
592 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
594 struct trace_array_cpu *data = tr->data[cpu];
595 struct trace_array_cpu *max_data = tr->data[cpu];
598 max_tr.time_start = data->preempt_timestamp;
600 max_data = max_tr.data[cpu];
601 max_data->saved_latency = tracing_max_latency;
602 max_data->critical_start = data->critical_start;
603 max_data->critical_end = data->critical_end;
605 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
606 max_data->pid = tsk->pid;
607 max_data->uid = task_uid(tsk);
608 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
609 max_data->policy = tsk->policy;
610 max_data->rt_priority = tsk->rt_priority;
612 /* record this tasks comm */
613 tracing_record_cmdline(tsk);
617 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
619 * @tsk: the task with the latency
620 * @cpu: The cpu that initiated the trace.
622 * Flip the buffers between the @tr and the max_tr and record information
623 * about which task was the cause of this latency.
626 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
628 struct ring_buffer *buf = tr->buffer;
630 if (trace_stop_count)
633 WARN_ON_ONCE(!irqs_disabled());
634 arch_spin_lock(&ftrace_max_lock);
636 tr->buffer = max_tr.buffer;
639 __update_max_tr(tr, tsk, cpu);
640 arch_spin_unlock(&ftrace_max_lock);
644 * update_max_tr_single - only copy one trace over, and reset the rest
646 * @tsk - task with the latency
647 * @cpu - the cpu of the buffer to copy.
649 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
652 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
656 if (trace_stop_count)
659 WARN_ON_ONCE(!irqs_disabled());
660 arch_spin_lock(&ftrace_max_lock);
662 ftrace_disable_cpu();
664 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
668 * We failed to swap the buffer due to a commit taking
669 * place on this CPU. We fail to record, but we reset
670 * the max trace buffer (no one writes directly to it)
671 * and flag that it failed.
673 trace_array_printk(&max_tr, _THIS_IP_,
674 "Failed to swap buffers due to commit in progress\n");
679 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
681 __update_max_tr(tr, tsk, cpu);
682 arch_spin_unlock(&ftrace_max_lock);
684 #endif /* CONFIG_TRACER_MAX_TRACE */
687 * register_tracer - register a tracer with the ftrace system.
688 * @type - the plugin for the tracer
690 * Register a new plugin tracer.
692 int register_tracer(struct tracer *type)
693 __releases(kernel_lock)
694 __acquires(kernel_lock)
700 pr_info("Tracer must have a name\n");
704 if (strlen(type->name) > MAX_TRACER_SIZE) {
705 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
710 * When this gets called we hold the BKL which means that
711 * preemption is disabled. Various trace selftests however
712 * need to disable and enable preemption for successful tests.
713 * So we drop the BKL here and grab it after the tests again.
716 mutex_lock(&trace_types_lock);
718 tracing_selftest_running = true;
720 for (t = trace_types; t; t = t->next) {
721 if (strcmp(type->name, t->name) == 0) {
723 pr_info("Tracer %s already registered\n",
731 type->set_flag = &dummy_set_flag;
733 type->flags = &dummy_tracer_flags;
735 if (!type->flags->opts)
736 type->flags->opts = dummy_tracer_opt;
737 if (!type->wait_pipe)
738 type->wait_pipe = default_wait_pipe;
741 #ifdef CONFIG_FTRACE_STARTUP_TEST
742 if (type->selftest && !tracing_selftest_disabled) {
743 struct tracer *saved_tracer = current_trace;
744 struct trace_array *tr = &global_trace;
747 * Run a selftest on this tracer.
748 * Here we reset the trace buffer, and set the current
749 * tracer to be this tracer. The tracer can then run some
750 * internal tracing to verify that everything is in order.
751 * If we fail, we do not register this tracer.
753 tracing_reset_online_cpus(tr);
755 current_trace = type;
756 /* the test is responsible for initializing and enabling */
757 pr_info("Testing tracer %s: ", type->name);
758 ret = type->selftest(type, tr);
759 /* the test is responsible for resetting too */
760 current_trace = saved_tracer;
762 printk(KERN_CONT "FAILED!\n");
765 /* Only reset on passing, to avoid touching corrupted buffers */
766 tracing_reset_online_cpus(tr);
768 printk(KERN_CONT "PASSED\n");
772 type->next = trace_types;
776 tracing_selftest_running = false;
777 mutex_unlock(&trace_types_lock);
779 if (ret || !default_bootup_tracer)
782 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
785 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
786 /* Do we want this tracer to start on bootup? */
787 tracing_set_tracer(type->name);
788 default_bootup_tracer = NULL;
789 /* disable other selftests, since this will break it. */
790 tracing_selftest_disabled = 1;
791 #ifdef CONFIG_FTRACE_STARTUP_TEST
792 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
801 void unregister_tracer(struct tracer *type)
805 mutex_lock(&trace_types_lock);
806 for (t = &trace_types; *t; t = &(*t)->next) {
810 pr_info("Tracer %s not registered\n", type->name);
816 if (type == current_trace && tracer_enabled) {
819 if (current_trace->stop)
820 current_trace->stop(&global_trace);
821 current_trace = &nop_trace;
824 mutex_unlock(&trace_types_lock);
827 static void __tracing_reset(struct trace_array *tr, int cpu)
829 ftrace_disable_cpu();
830 ring_buffer_reset_cpu(tr->buffer, cpu);
834 void tracing_reset(struct trace_array *tr, int cpu)
836 struct ring_buffer *buffer = tr->buffer;
838 ring_buffer_record_disable(buffer);
840 /* Make sure all commits have finished */
842 __tracing_reset(tr, cpu);
844 ring_buffer_record_enable(buffer);
847 void tracing_reset_online_cpus(struct trace_array *tr)
849 struct ring_buffer *buffer = tr->buffer;
852 ring_buffer_record_disable(buffer);
854 /* Make sure all commits have finished */
857 tr->time_start = ftrace_now(tr->cpu);
859 for_each_online_cpu(cpu)
860 __tracing_reset(tr, cpu);
862 ring_buffer_record_enable(buffer);
865 void tracing_reset_current(int cpu)
867 tracing_reset(&global_trace, cpu);
870 void tracing_reset_current_online_cpus(void)
872 tracing_reset_online_cpus(&global_trace);
875 #define SAVED_CMDLINES 128
876 #define NO_CMDLINE_MAP UINT_MAX
877 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
878 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
879 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
880 static int cmdline_idx;
881 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
883 /* temporary disable recording */
884 static atomic_t trace_record_cmdline_disabled __read_mostly;
886 static void trace_init_cmdlines(void)
888 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
889 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
893 int is_tracing_stopped(void)
895 return trace_stop_count;
899 * ftrace_off_permanent - disable all ftrace code permanently
901 * This should only be called when a serious anomally has
902 * been detected. This will turn off the function tracing,
903 * ring buffers, and other tracing utilites. It takes no
904 * locks and can be called from any context.
906 void ftrace_off_permanent(void)
908 tracing_disabled = 1;
910 tracing_off_permanent();
914 * tracing_start - quick start of the tracer
916 * If tracing is enabled but was stopped by tracing_stop,
917 * this will start the tracer back up.
919 void tracing_start(void)
921 struct ring_buffer *buffer;
924 if (tracing_disabled)
927 spin_lock_irqsave(&tracing_start_lock, flags);
928 if (--trace_stop_count) {
929 if (trace_stop_count < 0) {
930 /* Someone screwed up their debugging */
932 trace_stop_count = 0;
938 buffer = global_trace.buffer;
940 ring_buffer_record_enable(buffer);
942 buffer = max_tr.buffer;
944 ring_buffer_record_enable(buffer);
948 spin_unlock_irqrestore(&tracing_start_lock, flags);
952 * tracing_stop - quick stop of the tracer
954 * Light weight way to stop tracing. Use in conjunction with
957 void tracing_stop(void)
959 struct ring_buffer *buffer;
963 spin_lock_irqsave(&tracing_start_lock, flags);
964 if (trace_stop_count++)
967 buffer = global_trace.buffer;
969 ring_buffer_record_disable(buffer);
971 buffer = max_tr.buffer;
973 ring_buffer_record_disable(buffer);
976 spin_unlock_irqrestore(&tracing_start_lock, flags);
979 void trace_stop_cmdline_recording(void);
981 static void trace_save_cmdline(struct task_struct *tsk)
985 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
989 * It's not the end of the world if we don't get
990 * the lock, but we also don't want to spin
991 * nor do we want to disable interrupts,
992 * so if we miss here, then better luck next time.
994 if (!arch_spin_trylock(&trace_cmdline_lock))
997 idx = map_pid_to_cmdline[tsk->pid];
998 if (idx == NO_CMDLINE_MAP) {
999 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1002 * Check whether the cmdline buffer at idx has a pid
1003 * mapped. We are going to overwrite that entry so we
1004 * need to clear the map_pid_to_cmdline. Otherwise we
1005 * would read the new comm for the old pid.
1007 pid = map_cmdline_to_pid[idx];
1008 if (pid != NO_CMDLINE_MAP)
1009 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1011 map_cmdline_to_pid[idx] = tsk->pid;
1012 map_pid_to_cmdline[tsk->pid] = idx;
1017 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1019 arch_spin_unlock(&trace_cmdline_lock);
1022 void trace_find_cmdline(int pid, char comm[])
1027 strcpy(comm, "<idle>");
1031 if (pid > PID_MAX_DEFAULT) {
1032 strcpy(comm, "<...>");
1037 arch_spin_lock(&trace_cmdline_lock);
1038 map = map_pid_to_cmdline[pid];
1039 if (map != NO_CMDLINE_MAP)
1040 strcpy(comm, saved_cmdlines[map]);
1042 strcpy(comm, "<...>");
1044 arch_spin_unlock(&trace_cmdline_lock);
1048 void tracing_record_cmdline(struct task_struct *tsk)
1050 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1054 trace_save_cmdline(tsk);
1058 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1061 struct task_struct *tsk = current;
1063 entry->preempt_count = pc & 0xff;
1064 entry->pid = (tsk) ? tsk->pid : 0;
1065 entry->lock_depth = (tsk) ? tsk->lock_depth : 0;
1067 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1068 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1070 TRACE_FLAG_IRQS_NOSUPPORT |
1072 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1073 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1074 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1076 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1078 struct ring_buffer_event *
1079 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1082 unsigned long flags, int pc)
1084 struct ring_buffer_event *event;
1086 event = ring_buffer_lock_reserve(buffer, len);
1087 if (event != NULL) {
1088 struct trace_entry *ent = ring_buffer_event_data(event);
1090 tracing_generic_entry_update(ent, flags, pc);
1098 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1099 struct ring_buffer_event *event,
1100 unsigned long flags, int pc,
1103 ring_buffer_unlock_commit(buffer, event);
1105 ftrace_trace_stack(buffer, flags, 6, pc);
1106 ftrace_trace_userstack(buffer, flags, pc);
1112 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1113 struct ring_buffer_event *event,
1114 unsigned long flags, int pc)
1116 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1119 struct ring_buffer_event *
1120 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1121 int type, unsigned long len,
1122 unsigned long flags, int pc)
1124 *current_rb = global_trace.buffer;
1125 return trace_buffer_lock_reserve(*current_rb,
1126 type, len, flags, pc);
1128 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1130 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1131 struct ring_buffer_event *event,
1132 unsigned long flags, int pc)
1134 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1136 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1138 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1139 struct ring_buffer_event *event,
1140 unsigned long flags, int pc)
1142 __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1144 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1146 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1147 struct ring_buffer_event *event)
1149 ring_buffer_discard_commit(buffer, event);
1151 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1154 trace_function(struct trace_array *tr,
1155 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1158 struct ftrace_event_call *call = &event_function;
1159 struct ring_buffer *buffer = tr->buffer;
1160 struct ring_buffer_event *event;
1161 struct ftrace_entry *entry;
1163 /* If we are reading the ring buffer, don't trace */
1164 if (unlikely(__this_cpu_read(per_cpu_var(ftrace_cpu_disabled))))
1167 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1171 entry = ring_buffer_event_data(event);
1173 entry->parent_ip = parent_ip;
1175 if (!filter_check_discard(call, entry, buffer, event))
1176 ring_buffer_unlock_commit(buffer, event);
1180 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1181 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1184 if (likely(!atomic_read(&data->disabled)))
1185 trace_function(tr, ip, parent_ip, flags, pc);
1188 #ifdef CONFIG_STACKTRACE
1189 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1190 unsigned long flags,
1193 struct ftrace_event_call *call = &event_kernel_stack;
1194 struct ring_buffer_event *event;
1195 struct stack_entry *entry;
1196 struct stack_trace trace;
1198 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1199 sizeof(*entry), flags, pc);
1202 entry = ring_buffer_event_data(event);
1203 memset(&entry->caller, 0, sizeof(entry->caller));
1205 trace.nr_entries = 0;
1206 trace.max_entries = FTRACE_STACK_ENTRIES;
1208 trace.entries = entry->caller;
1210 save_stack_trace(&trace);
1211 if (!filter_check_discard(call, entry, buffer, event))
1212 ring_buffer_unlock_commit(buffer, event);
1215 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1218 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1221 __ftrace_trace_stack(buffer, flags, skip, pc);
1224 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1227 __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1231 * trace_dump_stack - record a stack back trace in the trace buffer
1233 void trace_dump_stack(void)
1235 unsigned long flags;
1237 if (tracing_disabled || tracing_selftest_running)
1240 local_save_flags(flags);
1242 /* skipping 3 traces, seems to get us at the caller of this function */
1243 __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count());
1247 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1249 struct ftrace_event_call *call = &event_user_stack;
1250 struct ring_buffer_event *event;
1251 struct userstack_entry *entry;
1252 struct stack_trace trace;
1254 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1257 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1258 sizeof(*entry), flags, pc);
1261 entry = ring_buffer_event_data(event);
1263 entry->tgid = current->tgid;
1264 memset(&entry->caller, 0, sizeof(entry->caller));
1266 trace.nr_entries = 0;
1267 trace.max_entries = FTRACE_STACK_ENTRIES;
1269 trace.entries = entry->caller;
1271 save_stack_trace_user(&trace);
1272 if (!filter_check_discard(call, entry, buffer, event))
1273 ring_buffer_unlock_commit(buffer, event);
1277 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1279 ftrace_trace_userstack(tr, flags, preempt_count());
1283 #endif /* CONFIG_STACKTRACE */
1286 ftrace_trace_special(void *__tr,
1287 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1290 struct ftrace_event_call *call = &event_special;
1291 struct ring_buffer_event *event;
1292 struct trace_array *tr = __tr;
1293 struct ring_buffer *buffer = tr->buffer;
1294 struct special_entry *entry;
1296 event = trace_buffer_lock_reserve(buffer, TRACE_SPECIAL,
1297 sizeof(*entry), 0, pc);
1300 entry = ring_buffer_event_data(event);
1305 if (!filter_check_discard(call, entry, buffer, event))
1306 trace_buffer_unlock_commit(buffer, event, 0, pc);
1310 __trace_special(void *__tr, void *__data,
1311 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1313 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1317 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1319 struct trace_array *tr = &global_trace;
1320 struct trace_array_cpu *data;
1321 unsigned long flags;
1325 if (tracing_disabled)
1328 pc = preempt_count();
1329 local_irq_save(flags);
1330 cpu = raw_smp_processor_id();
1331 data = tr->data[cpu];
1333 if (likely(atomic_inc_return(&data->disabled) == 1))
1334 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1336 atomic_dec(&data->disabled);
1337 local_irq_restore(flags);
1341 * trace_vbprintk - write binary msg to tracing buffer
1344 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1346 static arch_spinlock_t trace_buf_lock =
1347 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1348 static u32 trace_buf[TRACE_BUF_SIZE];
1350 struct ftrace_event_call *call = &event_bprint;
1351 struct ring_buffer_event *event;
1352 struct ring_buffer *buffer;
1353 struct trace_array *tr = &global_trace;
1354 struct trace_array_cpu *data;
1355 struct bprint_entry *entry;
1356 unsigned long flags;
1359 int cpu, len = 0, size, pc;
1361 if (unlikely(tracing_selftest_running || tracing_disabled))
1364 /* Don't pollute graph traces with trace_vprintk internals */
1365 pause_graph_tracing();
1367 pc = preempt_count();
1368 resched = ftrace_preempt_disable();
1369 cpu = raw_smp_processor_id();
1370 data = tr->data[cpu];
1372 disable = atomic_inc_return(&data->disabled);
1373 if (unlikely(disable != 1))
1376 /* Lockdep uses trace_printk for lock tracing */
1377 local_irq_save(flags);
1378 arch_spin_lock(&trace_buf_lock);
1379 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1381 if (len > TRACE_BUF_SIZE || len < 0)
1384 size = sizeof(*entry) + sizeof(u32) * len;
1385 buffer = tr->buffer;
1386 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1390 entry = ring_buffer_event_data(event);
1394 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1395 if (!filter_check_discard(call, entry, buffer, event)) {
1396 ring_buffer_unlock_commit(buffer, event);
1397 ftrace_trace_stack(buffer, flags, 6, pc);
1401 arch_spin_unlock(&trace_buf_lock);
1402 local_irq_restore(flags);
1405 atomic_dec_return(&data->disabled);
1406 ftrace_preempt_enable(resched);
1407 unpause_graph_tracing();
1411 EXPORT_SYMBOL_GPL(trace_vbprintk);
1413 int trace_array_printk(struct trace_array *tr,
1414 unsigned long ip, const char *fmt, ...)
1419 if (!(trace_flags & TRACE_ITER_PRINTK))
1423 ret = trace_array_vprintk(tr, ip, fmt, ap);
1428 int trace_array_vprintk(struct trace_array *tr,
1429 unsigned long ip, const char *fmt, va_list args)
1431 static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1432 static char trace_buf[TRACE_BUF_SIZE];
1434 struct ftrace_event_call *call = &event_print;
1435 struct ring_buffer_event *event;
1436 struct ring_buffer *buffer;
1437 struct trace_array_cpu *data;
1438 int cpu, len = 0, size, pc;
1439 struct print_entry *entry;
1440 unsigned long irq_flags;
1443 if (tracing_disabled || tracing_selftest_running)
1446 pc = preempt_count();
1447 preempt_disable_notrace();
1448 cpu = raw_smp_processor_id();
1449 data = tr->data[cpu];
1451 disable = atomic_inc_return(&data->disabled);
1452 if (unlikely(disable != 1))
1455 pause_graph_tracing();
1456 raw_local_irq_save(irq_flags);
1457 arch_spin_lock(&trace_buf_lock);
1458 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1460 size = sizeof(*entry) + len + 1;
1461 buffer = tr->buffer;
1462 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1466 entry = ring_buffer_event_data(event);
1469 memcpy(&entry->buf, trace_buf, len);
1470 entry->buf[len] = '\0';
1471 if (!filter_check_discard(call, entry, buffer, event)) {
1472 ring_buffer_unlock_commit(buffer, event);
1473 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1477 arch_spin_unlock(&trace_buf_lock);
1478 raw_local_irq_restore(irq_flags);
1479 unpause_graph_tracing();
1481 atomic_dec_return(&data->disabled);
1482 preempt_enable_notrace();
1487 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1489 return trace_array_vprintk(&global_trace, ip, fmt, args);
1491 EXPORT_SYMBOL_GPL(trace_vprintk);
1493 enum trace_file_type {
1494 TRACE_FILE_LAT_FMT = 1,
1495 TRACE_FILE_ANNOTATE = 2,
1498 static void trace_iterator_increment(struct trace_iterator *iter)
1500 /* Don't allow ftrace to trace into the ring buffers */
1501 ftrace_disable_cpu();
1504 if (iter->buffer_iter[iter->cpu])
1505 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1507 ftrace_enable_cpu();
1510 static struct trace_entry *
1511 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1513 struct ring_buffer_event *event;
1514 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1516 /* Don't allow ftrace to trace into the ring buffers */
1517 ftrace_disable_cpu();
1520 event = ring_buffer_iter_peek(buf_iter, ts);
1522 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1524 ftrace_enable_cpu();
1526 return event ? ring_buffer_event_data(event) : NULL;
1529 static struct trace_entry *
1530 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1532 struct ring_buffer *buffer = iter->tr->buffer;
1533 struct trace_entry *ent, *next = NULL;
1534 int cpu_file = iter->cpu_file;
1535 u64 next_ts = 0, ts;
1540 * If we are in a per_cpu trace file, don't bother by iterating over
1541 * all cpu and peek directly.
1543 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1544 if (ring_buffer_empty_cpu(buffer, cpu_file))
1546 ent = peek_next_entry(iter, cpu_file, ent_ts);
1548 *ent_cpu = cpu_file;
1553 for_each_tracing_cpu(cpu) {
1555 if (ring_buffer_empty_cpu(buffer, cpu))
1558 ent = peek_next_entry(iter, cpu, &ts);
1561 * Pick the entry with the smallest timestamp:
1563 if (ent && (!next || ts < next_ts)) {
1571 *ent_cpu = next_cpu;
1579 /* Find the next real entry, without updating the iterator itself */
1580 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1581 int *ent_cpu, u64 *ent_ts)
1583 return __find_next_entry(iter, ent_cpu, ent_ts);
1586 /* Find the next real entry, and increment the iterator to the next entry */
1587 static void *find_next_entry_inc(struct trace_iterator *iter)
1589 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1592 trace_iterator_increment(iter);
1594 return iter->ent ? iter : NULL;
1597 static void trace_consume(struct trace_iterator *iter)
1599 /* Don't allow ftrace to trace into the ring buffers */
1600 ftrace_disable_cpu();
1601 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1602 ftrace_enable_cpu();
1605 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1607 struct trace_iterator *iter = m->private;
1611 WARN_ON_ONCE(iter->leftover);
1615 /* can't go backwards */
1620 ent = find_next_entry_inc(iter);
1624 while (ent && iter->idx < i)
1625 ent = find_next_entry_inc(iter);
1632 static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1634 struct trace_array *tr = iter->tr;
1635 struct ring_buffer_event *event;
1636 struct ring_buffer_iter *buf_iter;
1637 unsigned long entries = 0;
1640 tr->data[cpu]->skipped_entries = 0;
1642 if (!iter->buffer_iter[cpu])
1645 buf_iter = iter->buffer_iter[cpu];
1646 ring_buffer_iter_reset(buf_iter);
1649 * We could have the case with the max latency tracers
1650 * that a reset never took place on a cpu. This is evident
1651 * by the timestamp being before the start of the buffer.
1653 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1654 if (ts >= iter->tr->time_start)
1657 ring_buffer_read(buf_iter, NULL);
1660 tr->data[cpu]->skipped_entries = entries;
1664 * The current tracer is copied to avoid a global locking
1667 static void *s_start(struct seq_file *m, loff_t *pos)
1669 struct trace_iterator *iter = m->private;
1670 static struct tracer *old_tracer;
1671 int cpu_file = iter->cpu_file;
1676 /* copy the tracer to avoid using a global lock all around */
1677 mutex_lock(&trace_types_lock);
1678 if (unlikely(old_tracer != current_trace && current_trace)) {
1679 old_tracer = current_trace;
1680 *iter->trace = *current_trace;
1682 mutex_unlock(&trace_types_lock);
1684 atomic_inc(&trace_record_cmdline_disabled);
1686 if (*pos != iter->pos) {
1691 ftrace_disable_cpu();
1693 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1694 for_each_tracing_cpu(cpu)
1695 tracing_iter_reset(iter, cpu);
1697 tracing_iter_reset(iter, cpu_file);
1699 ftrace_enable_cpu();
1701 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1706 * If we overflowed the seq_file before, then we want
1707 * to just reuse the trace_seq buffer again.
1713 p = s_next(m, p, &l);
1717 trace_event_read_lock();
1718 trace_access_lock(cpu_file);
1722 static void s_stop(struct seq_file *m, void *p)
1724 struct trace_iterator *iter = m->private;
1726 atomic_dec(&trace_record_cmdline_disabled);
1727 trace_access_unlock(iter->cpu_file);
1728 trace_event_read_unlock();
1731 static void print_lat_help_header(struct seq_file *m)
1733 seq_puts(m, "# _------=> CPU# \n");
1734 seq_puts(m, "# / _-----=> irqs-off \n");
1735 seq_puts(m, "# | / _----=> need-resched \n");
1736 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1737 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1738 seq_puts(m, "# |||| /_--=> lock-depth \n");
1739 seq_puts(m, "# |||||/ delay \n");
1740 seq_puts(m, "# cmd pid |||||| time | caller \n");
1741 seq_puts(m, "# \\ / |||||| \\ | / \n");
1744 static void print_func_help_header(struct seq_file *m)
1746 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1747 seq_puts(m, "# | | | | |\n");
1752 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1754 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1755 struct trace_array *tr = iter->tr;
1756 struct trace_array_cpu *data = tr->data[tr->cpu];
1757 struct tracer *type = current_trace;
1758 unsigned long entries = 0;
1759 unsigned long total = 0;
1760 unsigned long count;
1761 const char *name = "preemption";
1768 for_each_tracing_cpu(cpu) {
1769 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1771 * If this buffer has skipped entries, then we hold all
1772 * entries for the trace and we need to ignore the
1773 * ones before the time stamp.
1775 if (tr->data[cpu]->skipped_entries) {
1776 count -= tr->data[cpu]->skipped_entries;
1777 /* total is the same as the entries */
1781 ring_buffer_overrun_cpu(tr->buffer, cpu);
1785 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1787 seq_puts(m, "# -----------------------------------"
1788 "---------------------------------\n");
1789 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1790 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1791 nsecs_to_usecs(data->saved_latency),
1795 #if defined(CONFIG_PREEMPT_NONE)
1797 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1799 #elif defined(CONFIG_PREEMPT)
1804 /* These are reserved for later use */
1807 seq_printf(m, " #P:%d)\n", num_online_cpus());
1811 seq_puts(m, "# -----------------\n");
1812 seq_printf(m, "# | task: %.16s-%d "
1813 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1814 data->comm, data->pid, data->uid, data->nice,
1815 data->policy, data->rt_priority);
1816 seq_puts(m, "# -----------------\n");
1818 if (data->critical_start) {
1819 seq_puts(m, "# => started at: ");
1820 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1821 trace_print_seq(m, &iter->seq);
1822 seq_puts(m, "\n# => ended at: ");
1823 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1824 trace_print_seq(m, &iter->seq);
1825 seq_puts(m, "\n#\n");
1831 static void test_cpu_buff_start(struct trace_iterator *iter)
1833 struct trace_seq *s = &iter->seq;
1835 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1838 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1841 if (cpumask_test_cpu(iter->cpu, iter->started))
1844 if (iter->tr->data[iter->cpu]->skipped_entries)
1847 cpumask_set_cpu(iter->cpu, iter->started);
1849 /* Don't print started cpu buffer for the first entry of the trace */
1851 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1855 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1857 struct trace_seq *s = &iter->seq;
1858 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1859 struct trace_entry *entry;
1860 struct trace_event *event;
1864 test_cpu_buff_start(iter);
1866 event = ftrace_find_event(entry->type);
1868 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1869 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1870 if (!trace_print_lat_context(iter))
1873 if (!trace_print_context(iter))
1879 return event->trace(iter, sym_flags);
1881 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1884 return TRACE_TYPE_HANDLED;
1886 return TRACE_TYPE_PARTIAL_LINE;
1889 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1891 struct trace_seq *s = &iter->seq;
1892 struct trace_entry *entry;
1893 struct trace_event *event;
1897 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1898 if (!trace_seq_printf(s, "%d %d %llu ",
1899 entry->pid, iter->cpu, iter->ts))
1903 event = ftrace_find_event(entry->type);
1905 return event->raw(iter, 0);
1907 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1910 return TRACE_TYPE_HANDLED;
1912 return TRACE_TYPE_PARTIAL_LINE;
1915 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1917 struct trace_seq *s = &iter->seq;
1918 unsigned char newline = '\n';
1919 struct trace_entry *entry;
1920 struct trace_event *event;
1924 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1925 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1926 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1927 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1930 event = ftrace_find_event(entry->type);
1932 enum print_line_t ret = event->hex(iter, 0);
1933 if (ret != TRACE_TYPE_HANDLED)
1937 SEQ_PUT_FIELD_RET(s, newline);
1939 return TRACE_TYPE_HANDLED;
1942 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1944 struct trace_seq *s = &iter->seq;
1945 struct trace_entry *entry;
1946 struct trace_event *event;
1950 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1951 SEQ_PUT_FIELD_RET(s, entry->pid);
1952 SEQ_PUT_FIELD_RET(s, iter->cpu);
1953 SEQ_PUT_FIELD_RET(s, iter->ts);
1956 event = ftrace_find_event(entry->type);
1957 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1960 static int trace_empty(struct trace_iterator *iter)
1964 /* If we are looking at one CPU buffer, only check that one */
1965 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1966 cpu = iter->cpu_file;
1967 if (iter->buffer_iter[cpu]) {
1968 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1971 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1977 for_each_tracing_cpu(cpu) {
1978 if (iter->buffer_iter[cpu]) {
1979 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1982 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1990 /* Called with trace_event_read_lock() held. */
1991 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1993 enum print_line_t ret;
1995 if (iter->trace && iter->trace->print_line) {
1996 ret = iter->trace->print_line(iter);
1997 if (ret != TRACE_TYPE_UNHANDLED)
2001 if (iter->ent->type == TRACE_BPRINT &&
2002 trace_flags & TRACE_ITER_PRINTK &&
2003 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2004 return trace_print_bprintk_msg_only(iter);
2006 if (iter->ent->type == TRACE_PRINT &&
2007 trace_flags & TRACE_ITER_PRINTK &&
2008 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2009 return trace_print_printk_msg_only(iter);
2011 if (trace_flags & TRACE_ITER_BIN)
2012 return print_bin_fmt(iter);
2014 if (trace_flags & TRACE_ITER_HEX)
2015 return print_hex_fmt(iter);
2017 if (trace_flags & TRACE_ITER_RAW)
2018 return print_raw_fmt(iter);
2020 return print_trace_fmt(iter);
2023 static int s_show(struct seq_file *m, void *v)
2025 struct trace_iterator *iter = v;
2028 if (iter->ent == NULL) {
2030 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2033 if (iter->trace && iter->trace->print_header)
2034 iter->trace->print_header(m);
2035 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2036 /* print nothing if the buffers are empty */
2037 if (trace_empty(iter))
2039 print_trace_header(m, iter);
2040 if (!(trace_flags & TRACE_ITER_VERBOSE))
2041 print_lat_help_header(m);
2043 if (!(trace_flags & TRACE_ITER_VERBOSE))
2044 print_func_help_header(m);
2046 } else if (iter->leftover) {
2048 * If we filled the seq_file buffer earlier, we
2049 * want to just show it now.
2051 ret = trace_print_seq(m, &iter->seq);
2053 /* ret should this time be zero, but you never know */
2054 iter->leftover = ret;
2057 print_trace_line(iter);
2058 ret = trace_print_seq(m, &iter->seq);
2060 * If we overflow the seq_file buffer, then it will
2061 * ask us for this data again at start up.
2063 * ret is 0 if seq_file write succeeded.
2066 iter->leftover = ret;
2072 static const struct seq_operations tracer_seq_ops = {
2079 static struct trace_iterator *
2080 __tracing_open(struct inode *inode, struct file *file)
2082 long cpu_file = (long) inode->i_private;
2083 void *fail_ret = ERR_PTR(-ENOMEM);
2084 struct trace_iterator *iter;
2088 if (tracing_disabled)
2089 return ERR_PTR(-ENODEV);
2091 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2093 return ERR_PTR(-ENOMEM);
2096 * We make a copy of the current tracer to avoid concurrent
2097 * changes on it while we are reading.
2099 mutex_lock(&trace_types_lock);
2100 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2105 *iter->trace = *current_trace;
2107 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2110 if (current_trace && current_trace->print_max)
2113 iter->tr = &global_trace;
2115 mutex_init(&iter->mutex);
2116 iter->cpu_file = cpu_file;
2118 /* Notify the tracer early; before we stop tracing. */
2119 if (iter->trace && iter->trace->open)
2120 iter->trace->open(iter);
2122 /* Annotate start of buffers if we had overruns */
2123 if (ring_buffer_overruns(iter->tr->buffer))
2124 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2126 /* stop the trace while dumping */
2129 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2130 for_each_tracing_cpu(cpu) {
2132 iter->buffer_iter[cpu] =
2133 ring_buffer_read_start(iter->tr->buffer, cpu);
2134 tracing_iter_reset(iter, cpu);
2137 cpu = iter->cpu_file;
2138 iter->buffer_iter[cpu] =
2139 ring_buffer_read_start(iter->tr->buffer, cpu);
2140 tracing_iter_reset(iter, cpu);
2143 ret = seq_open(file, &tracer_seq_ops);
2145 fail_ret = ERR_PTR(ret);
2149 m = file->private_data;
2152 mutex_unlock(&trace_types_lock);
2157 for_each_tracing_cpu(cpu) {
2158 if (iter->buffer_iter[cpu])
2159 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2161 free_cpumask_var(iter->started);
2164 mutex_unlock(&trace_types_lock);
2171 int tracing_open_generic(struct inode *inode, struct file *filp)
2173 if (tracing_disabled)
2176 filp->private_data = inode->i_private;
2180 static int tracing_release(struct inode *inode, struct file *file)
2182 struct seq_file *m = (struct seq_file *)file->private_data;
2183 struct trace_iterator *iter;
2186 if (!(file->f_mode & FMODE_READ))
2191 mutex_lock(&trace_types_lock);
2192 for_each_tracing_cpu(cpu) {
2193 if (iter->buffer_iter[cpu])
2194 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2197 if (iter->trace && iter->trace->close)
2198 iter->trace->close(iter);
2200 /* reenable tracing if it was previously enabled */
2202 mutex_unlock(&trace_types_lock);
2204 seq_release(inode, file);
2205 mutex_destroy(&iter->mutex);
2206 free_cpumask_var(iter->started);
2212 static int tracing_open(struct inode *inode, struct file *file)
2214 struct trace_iterator *iter;
2217 /* If this file was open for write, then erase contents */
2218 if ((file->f_mode & FMODE_WRITE) &&
2219 (file->f_flags & O_TRUNC)) {
2220 long cpu = (long) inode->i_private;
2222 if (cpu == TRACE_PIPE_ALL_CPU)
2223 tracing_reset_online_cpus(&global_trace);
2225 tracing_reset(&global_trace, cpu);
2228 if (file->f_mode & FMODE_READ) {
2229 iter = __tracing_open(inode, file);
2231 ret = PTR_ERR(iter);
2232 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2233 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2239 t_next(struct seq_file *m, void *v, loff_t *pos)
2241 struct tracer *t = v;
2251 static void *t_start(struct seq_file *m, loff_t *pos)
2256 mutex_lock(&trace_types_lock);
2257 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2263 static void t_stop(struct seq_file *m, void *p)
2265 mutex_unlock(&trace_types_lock);
2268 static int t_show(struct seq_file *m, void *v)
2270 struct tracer *t = v;
2275 seq_printf(m, "%s", t->name);
2284 static const struct seq_operations show_traces_seq_ops = {
2291 static int show_traces_open(struct inode *inode, struct file *file)
2293 if (tracing_disabled)
2296 return seq_open(file, &show_traces_seq_ops);
2300 tracing_write_stub(struct file *filp, const char __user *ubuf,
2301 size_t count, loff_t *ppos)
2306 static const struct file_operations tracing_fops = {
2307 .open = tracing_open,
2309 .write = tracing_write_stub,
2310 .llseek = seq_lseek,
2311 .release = tracing_release,
2314 static const struct file_operations show_traces_fops = {
2315 .open = show_traces_open,
2317 .release = seq_release,
2321 * Only trace on a CPU if the bitmask is set:
2323 static cpumask_var_t tracing_cpumask;
2326 * The tracer itself will not take this lock, but still we want
2327 * to provide a consistent cpumask to user-space:
2329 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2332 * Temporary storage for the character representation of the
2333 * CPU bitmask (and one more byte for the newline):
2335 static char mask_str[NR_CPUS + 1];
2338 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2339 size_t count, loff_t *ppos)
2343 mutex_lock(&tracing_cpumask_update_lock);
2345 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2346 if (count - len < 2) {
2350 len += sprintf(mask_str + len, "\n");
2351 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2354 mutex_unlock(&tracing_cpumask_update_lock);
2360 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2361 size_t count, loff_t *ppos)
2364 cpumask_var_t tracing_cpumask_new;
2366 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2369 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2373 mutex_lock(&tracing_cpumask_update_lock);
2375 local_irq_disable();
2376 arch_spin_lock(&ftrace_max_lock);
2377 for_each_tracing_cpu(cpu) {
2379 * Increase/decrease the disabled counter if we are
2380 * about to flip a bit in the cpumask:
2382 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2383 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2384 atomic_inc(&global_trace.data[cpu]->disabled);
2386 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2387 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2388 atomic_dec(&global_trace.data[cpu]->disabled);
2391 arch_spin_unlock(&ftrace_max_lock);
2394 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2396 mutex_unlock(&tracing_cpumask_update_lock);
2397 free_cpumask_var(tracing_cpumask_new);
2402 free_cpumask_var(tracing_cpumask_new);
2407 static const struct file_operations tracing_cpumask_fops = {
2408 .open = tracing_open_generic,
2409 .read = tracing_cpumask_read,
2410 .write = tracing_cpumask_write,
2413 static int tracing_trace_options_show(struct seq_file *m, void *v)
2415 struct tracer_opt *trace_opts;
2419 mutex_lock(&trace_types_lock);
2420 tracer_flags = current_trace->flags->val;
2421 trace_opts = current_trace->flags->opts;
2423 for (i = 0; trace_options[i]; i++) {
2424 if (trace_flags & (1 << i))
2425 seq_printf(m, "%s\n", trace_options[i]);
2427 seq_printf(m, "no%s\n", trace_options[i]);
2430 for (i = 0; trace_opts[i].name; i++) {
2431 if (tracer_flags & trace_opts[i].bit)
2432 seq_printf(m, "%s\n", trace_opts[i].name);
2434 seq_printf(m, "no%s\n", trace_opts[i].name);
2436 mutex_unlock(&trace_types_lock);
2441 static int __set_tracer_option(struct tracer *trace,
2442 struct tracer_flags *tracer_flags,
2443 struct tracer_opt *opts, int neg)
2447 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2452 tracer_flags->val &= ~opts->bit;
2454 tracer_flags->val |= opts->bit;
2458 /* Try to assign a tracer specific option */
2459 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2461 struct tracer_flags *tracer_flags = trace->flags;
2462 struct tracer_opt *opts = NULL;
2465 for (i = 0; tracer_flags->opts[i].name; i++) {
2466 opts = &tracer_flags->opts[i];
2468 if (strcmp(cmp, opts->name) == 0)
2469 return __set_tracer_option(trace, trace->flags,
2476 static void set_tracer_flags(unsigned int mask, int enabled)
2478 /* do nothing if flag is already set */
2479 if (!!(trace_flags & mask) == !!enabled)
2483 trace_flags |= mask;
2485 trace_flags &= ~mask;
2489 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2490 size_t cnt, loff_t *ppos)
2498 if (cnt >= sizeof(buf))
2501 if (copy_from_user(&buf, ubuf, cnt))
2505 cmp = strstrip(buf);
2507 if (strncmp(cmp, "no", 2) == 0) {
2512 for (i = 0; trace_options[i]; i++) {
2513 if (strcmp(cmp, trace_options[i]) == 0) {
2514 set_tracer_flags(1 << i, !neg);
2519 /* If no option could be set, test the specific tracer options */
2520 if (!trace_options[i]) {
2521 mutex_lock(&trace_types_lock);
2522 ret = set_tracer_option(current_trace, cmp, neg);
2523 mutex_unlock(&trace_types_lock);
2533 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2535 if (tracing_disabled)
2537 return single_open(file, tracing_trace_options_show, NULL);
2540 static const struct file_operations tracing_iter_fops = {
2541 .open = tracing_trace_options_open,
2543 .llseek = seq_lseek,
2544 .release = single_release,
2545 .write = tracing_trace_options_write,
2548 static const char readme_msg[] =
2549 "tracing mini-HOWTO:\n\n"
2550 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2551 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2552 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2553 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2555 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2556 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2558 "# cat /sys/kernel/debug/tracing/trace_options\n"
2559 "noprint-parent nosym-offset nosym-addr noverbose\n"
2560 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2561 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2562 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2563 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2567 tracing_readme_read(struct file *filp, char __user *ubuf,
2568 size_t cnt, loff_t *ppos)
2570 return simple_read_from_buffer(ubuf, cnt, ppos,
2571 readme_msg, strlen(readme_msg));
2574 static const struct file_operations tracing_readme_fops = {
2575 .open = tracing_open_generic,
2576 .read = tracing_readme_read,
2580 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2581 size_t cnt, loff_t *ppos)
2590 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2594 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2602 for (i = 0; i < SAVED_CMDLINES; i++) {
2605 pid = map_cmdline_to_pid[i];
2606 if (pid == -1 || pid == NO_CMDLINE_MAP)
2609 trace_find_cmdline(pid, buf_comm);
2610 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2615 len = simple_read_from_buffer(ubuf, cnt, ppos,
2624 static const struct file_operations tracing_saved_cmdlines_fops = {
2625 .open = tracing_open_generic,
2626 .read = tracing_saved_cmdlines_read,
2630 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2631 size_t cnt, loff_t *ppos)
2636 r = sprintf(buf, "%u\n", tracer_enabled);
2637 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2641 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2642 size_t cnt, loff_t *ppos)
2644 struct trace_array *tr = filp->private_data;
2649 if (cnt >= sizeof(buf))
2652 if (copy_from_user(&buf, ubuf, cnt))
2657 ret = strict_strtoul(buf, 10, &val);
2663 mutex_lock(&trace_types_lock);
2664 if (tracer_enabled ^ val) {
2667 if (current_trace->start)
2668 current_trace->start(tr);
2673 if (current_trace->stop)
2674 current_trace->stop(tr);
2677 mutex_unlock(&trace_types_lock);
2685 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2686 size_t cnt, loff_t *ppos)
2688 char buf[MAX_TRACER_SIZE+2];
2691 mutex_lock(&trace_types_lock);
2693 r = sprintf(buf, "%s\n", current_trace->name);
2695 r = sprintf(buf, "\n");
2696 mutex_unlock(&trace_types_lock);
2698 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2701 int tracer_init(struct tracer *t, struct trace_array *tr)
2703 tracing_reset_online_cpus(tr);
2707 static int tracing_resize_ring_buffer(unsigned long size)
2712 * If kernel or user changes the size of the ring buffer
2713 * we use the size that was given, and we can forget about
2714 * expanding it later.
2716 ring_buffer_expanded = 1;
2718 ret = ring_buffer_resize(global_trace.buffer, size);
2722 ret = ring_buffer_resize(max_tr.buffer, size);
2726 r = ring_buffer_resize(global_trace.buffer,
2727 global_trace.entries);
2730 * AARGH! We are left with different
2731 * size max buffer!!!!
2732 * The max buffer is our "snapshot" buffer.
2733 * When a tracer needs a snapshot (one of the
2734 * latency tracers), it swaps the max buffer
2735 * with the saved snap shot. We succeeded to
2736 * update the size of the main buffer, but failed to
2737 * update the size of the max buffer. But when we tried
2738 * to reset the main buffer to the original size, we
2739 * failed there too. This is very unlikely to
2740 * happen, but if it does, warn and kill all
2744 tracing_disabled = 1;
2749 global_trace.entries = size;
2755 * tracing_update_buffers - used by tracing facility to expand ring buffers
2757 * To save on memory when the tracing is never used on a system with it
2758 * configured in. The ring buffers are set to a minimum size. But once
2759 * a user starts to use the tracing facility, then they need to grow
2760 * to their default size.
2762 * This function is to be called when a tracer is about to be used.
2764 int tracing_update_buffers(void)
2768 mutex_lock(&trace_types_lock);
2769 if (!ring_buffer_expanded)
2770 ret = tracing_resize_ring_buffer(trace_buf_size);
2771 mutex_unlock(&trace_types_lock);
2776 struct trace_option_dentry;
2778 static struct trace_option_dentry *
2779 create_trace_option_files(struct tracer *tracer);
2782 destroy_trace_option_files(struct trace_option_dentry *topts);
2784 static int tracing_set_tracer(const char *buf)
2786 static struct trace_option_dentry *topts;
2787 struct trace_array *tr = &global_trace;
2791 mutex_lock(&trace_types_lock);
2793 if (!ring_buffer_expanded) {
2794 ret = tracing_resize_ring_buffer(trace_buf_size);
2800 for (t = trace_types; t; t = t->next) {
2801 if (strcmp(t->name, buf) == 0)
2808 if (t == current_trace)
2811 trace_branch_disable();
2812 if (current_trace && current_trace->reset)
2813 current_trace->reset(tr);
2815 destroy_trace_option_files(topts);
2819 topts = create_trace_option_files(current_trace);
2822 ret = tracer_init(t, tr);
2827 trace_branch_enable(tr);
2829 mutex_unlock(&trace_types_lock);
2835 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2836 size_t cnt, loff_t *ppos)
2838 char buf[MAX_TRACER_SIZE+1];
2845 if (cnt > MAX_TRACER_SIZE)
2846 cnt = MAX_TRACER_SIZE;
2848 if (copy_from_user(&buf, ubuf, cnt))
2853 /* strip ending whitespace. */
2854 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2857 err = tracing_set_tracer(buf);
2867 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2868 size_t cnt, loff_t *ppos)
2870 unsigned long *ptr = filp->private_data;
2874 r = snprintf(buf, sizeof(buf), "%ld\n",
2875 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2876 if (r > sizeof(buf))
2878 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2882 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2883 size_t cnt, loff_t *ppos)
2885 unsigned long *ptr = filp->private_data;
2890 if (cnt >= sizeof(buf))
2893 if (copy_from_user(&buf, ubuf, cnt))
2898 ret = strict_strtoul(buf, 10, &val);
2907 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2909 long cpu_file = (long) inode->i_private;
2910 struct trace_iterator *iter;
2913 if (tracing_disabled)
2916 mutex_lock(&trace_types_lock);
2918 /* create a buffer to store the information to pass to userspace */
2919 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2926 * We make a copy of the current tracer to avoid concurrent
2927 * changes on it while we are reading.
2929 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2935 *iter->trace = *current_trace;
2937 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2942 /* trace pipe does not show start of buffer */
2943 cpumask_setall(iter->started);
2945 if (trace_flags & TRACE_ITER_LATENCY_FMT)
2946 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2948 iter->cpu_file = cpu_file;
2949 iter->tr = &global_trace;
2950 mutex_init(&iter->mutex);
2951 filp->private_data = iter;
2953 if (iter->trace->pipe_open)
2954 iter->trace->pipe_open(iter);
2957 mutex_unlock(&trace_types_lock);
2963 mutex_unlock(&trace_types_lock);
2967 static int tracing_release_pipe(struct inode *inode, struct file *file)
2969 struct trace_iterator *iter = file->private_data;
2971 mutex_lock(&trace_types_lock);
2973 if (iter->trace->pipe_close)
2974 iter->trace->pipe_close(iter);
2976 mutex_unlock(&trace_types_lock);
2978 free_cpumask_var(iter->started);
2979 mutex_destroy(&iter->mutex);
2987 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2989 struct trace_iterator *iter = filp->private_data;
2991 if (trace_flags & TRACE_ITER_BLOCK) {
2993 * Always select as readable when in blocking mode
2995 return POLLIN | POLLRDNORM;
2997 if (!trace_empty(iter))
2998 return POLLIN | POLLRDNORM;
2999 poll_wait(filp, &trace_wait, poll_table);
3000 if (!trace_empty(iter))
3001 return POLLIN | POLLRDNORM;
3008 void default_wait_pipe(struct trace_iterator *iter)
3012 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3014 if (trace_empty(iter))
3017 finish_wait(&trace_wait, &wait);
3021 * This is a make-shift waitqueue.
3022 * A tracer might use this callback on some rare cases:
3024 * 1) the current tracer might hold the runqueue lock when it wakes up
3025 * a reader, hence a deadlock (sched, function, and function graph tracers)
3026 * 2) the function tracers, trace all functions, we don't want
3027 * the overhead of calling wake_up and friends
3028 * (and tracing them too)
3030 * Anyway, this is really very primitive wakeup.
3032 void poll_wait_pipe(struct trace_iterator *iter)
3034 set_current_state(TASK_INTERRUPTIBLE);
3035 /* sleep for 100 msecs, and try again. */
3036 schedule_timeout(HZ / 10);
3039 /* Must be called with trace_types_lock mutex held. */
3040 static int tracing_wait_pipe(struct file *filp)
3042 struct trace_iterator *iter = filp->private_data;
3044 while (trace_empty(iter)) {
3046 if ((filp->f_flags & O_NONBLOCK)) {
3050 mutex_unlock(&iter->mutex);
3052 iter->trace->wait_pipe(iter);
3054 mutex_lock(&iter->mutex);
3056 if (signal_pending(current))
3060 * We block until we read something and tracing is disabled.
3061 * We still block if tracing is disabled, but we have never
3062 * read anything. This allows a user to cat this file, and
3063 * then enable tracing. But after we have read something,
3064 * we give an EOF when tracing is again disabled.
3066 * iter->pos will be 0 if we haven't read anything.
3068 if (!tracer_enabled && iter->pos)
3079 tracing_read_pipe(struct file *filp, char __user *ubuf,
3080 size_t cnt, loff_t *ppos)
3082 struct trace_iterator *iter = filp->private_data;
3083 static struct tracer *old_tracer;
3086 /* return any leftover data */
3087 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3091 trace_seq_init(&iter->seq);
3093 /* copy the tracer to avoid using a global lock all around */
3094 mutex_lock(&trace_types_lock);
3095 if (unlikely(old_tracer != current_trace && current_trace)) {
3096 old_tracer = current_trace;
3097 *iter->trace = *current_trace;
3099 mutex_unlock(&trace_types_lock);
3102 * Avoid more than one consumer on a single file descriptor
3103 * This is just a matter of traces coherency, the ring buffer itself
3106 mutex_lock(&iter->mutex);
3107 if (iter->trace->read) {
3108 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3114 sret = tracing_wait_pipe(filp);
3118 /* stop when tracing is finished */
3119 if (trace_empty(iter)) {
3124 if (cnt >= PAGE_SIZE)
3125 cnt = PAGE_SIZE - 1;
3127 /* reset all but tr, trace, and overruns */
3128 memset(&iter->seq, 0,
3129 sizeof(struct trace_iterator) -
3130 offsetof(struct trace_iterator, seq));
3133 trace_event_read_lock();
3134 trace_access_lock(iter->cpu_file);
3135 while (find_next_entry_inc(iter) != NULL) {
3136 enum print_line_t ret;
3137 int len = iter->seq.len;
3139 ret = print_trace_line(iter);
3140 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3141 /* don't print partial lines */
3142 iter->seq.len = len;
3145 if (ret != TRACE_TYPE_NO_CONSUME)
3146 trace_consume(iter);
3148 if (iter->seq.len >= cnt)
3151 trace_access_unlock(iter->cpu_file);
3152 trace_event_read_unlock();
3154 /* Now copy what we have to the user */
3155 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3156 if (iter->seq.readpos >= iter->seq.len)
3157 trace_seq_init(&iter->seq);
3160 * If there was nothing to send to user, inspite of consuming trace
3161 * entries, go back to wait for more entries.
3167 mutex_unlock(&iter->mutex);
3172 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3173 struct pipe_buffer *buf)
3175 __free_page(buf->page);
3178 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3181 __free_page(spd->pages[idx]);
3184 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3186 .map = generic_pipe_buf_map,
3187 .unmap = generic_pipe_buf_unmap,
3188 .confirm = generic_pipe_buf_confirm,
3189 .release = tracing_pipe_buf_release,
3190 .steal = generic_pipe_buf_steal,
3191 .get = generic_pipe_buf_get,
3195 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3200 /* Seq buffer is page-sized, exactly what we need. */
3202 count = iter->seq.len;
3203 ret = print_trace_line(iter);
3204 count = iter->seq.len - count;
3207 iter->seq.len -= count;
3210 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3211 iter->seq.len -= count;
3215 if (ret != TRACE_TYPE_NO_CONSUME)
3216 trace_consume(iter);
3218 if (!find_next_entry_inc(iter)) {
3228 static ssize_t tracing_splice_read_pipe(struct file *filp,
3230 struct pipe_inode_info *pipe,
3234 struct page *pages[PIPE_BUFFERS];
3235 struct partial_page partial[PIPE_BUFFERS];
3236 struct trace_iterator *iter = filp->private_data;
3237 struct splice_pipe_desc spd = {
3240 .nr_pages = 0, /* This gets updated below. */
3242 .ops = &tracing_pipe_buf_ops,
3243 .spd_release = tracing_spd_release_pipe,
3245 static struct tracer *old_tracer;
3250 /* copy the tracer to avoid using a global lock all around */
3251 mutex_lock(&trace_types_lock);
3252 if (unlikely(old_tracer != current_trace && current_trace)) {
3253 old_tracer = current_trace;
3254 *iter->trace = *current_trace;
3256 mutex_unlock(&trace_types_lock);
3258 mutex_lock(&iter->mutex);
3260 if (iter->trace->splice_read) {
3261 ret = iter->trace->splice_read(iter, filp,
3262 ppos, pipe, len, flags);
3267 ret = tracing_wait_pipe(filp);
3271 if (!iter->ent && !find_next_entry_inc(iter)) {
3276 trace_event_read_lock();
3277 trace_access_lock(iter->cpu_file);
3279 /* Fill as many pages as possible. */
3280 for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3281 pages[i] = alloc_page(GFP_KERNEL);
3285 rem = tracing_fill_pipe_page(rem, iter);
3287 /* Copy the data into the page, so we can start over. */
3288 ret = trace_seq_to_buffer(&iter->seq,
3289 page_address(pages[i]),
3292 __free_page(pages[i]);
3295 partial[i].offset = 0;
3296 partial[i].len = iter->seq.len;
3298 trace_seq_init(&iter->seq);
3301 trace_access_unlock(iter->cpu_file);
3302 trace_event_read_unlock();
3303 mutex_unlock(&iter->mutex);
3307 return splice_to_pipe(pipe, &spd);
3310 mutex_unlock(&iter->mutex);
3316 tracing_entries_read(struct file *filp, char __user *ubuf,
3317 size_t cnt, loff_t *ppos)
3319 struct trace_array *tr = filp->private_data;
3323 mutex_lock(&trace_types_lock);
3324 if (!ring_buffer_expanded)
3325 r = sprintf(buf, "%lu (expanded: %lu)\n",
3327 trace_buf_size >> 10);
3329 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3330 mutex_unlock(&trace_types_lock);
3332 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3336 tracing_entries_write(struct file *filp, const char __user *ubuf,
3337 size_t cnt, loff_t *ppos)
3343 if (cnt >= sizeof(buf))
3346 if (copy_from_user(&buf, ubuf, cnt))
3351 ret = strict_strtoul(buf, 10, &val);
3355 /* must have at least 1 entry */
3359 mutex_lock(&trace_types_lock);
3363 /* disable all cpu buffers */
3364 for_each_tracing_cpu(cpu) {
3365 if (global_trace.data[cpu])
3366 atomic_inc(&global_trace.data[cpu]->disabled);
3367 if (max_tr.data[cpu])
3368 atomic_inc(&max_tr.data[cpu]->disabled);
3371 /* value is in KB */
3374 if (val != global_trace.entries) {
3375 ret = tracing_resize_ring_buffer(val);
3384 /* If check pages failed, return ENOMEM */
3385 if (tracing_disabled)
3388 for_each_tracing_cpu(cpu) {
3389 if (global_trace.data[cpu])
3390 atomic_dec(&global_trace.data[cpu]->disabled);
3391 if (max_tr.data[cpu])
3392 atomic_dec(&max_tr.data[cpu]->disabled);
3396 max_tr.entries = global_trace.entries;
3397 mutex_unlock(&trace_types_lock);
3402 static int mark_printk(const char *fmt, ...)
3406 va_start(args, fmt);
3407 ret = trace_vprintk(0, fmt, args);
3413 tracing_mark_write(struct file *filp, const char __user *ubuf,
3414 size_t cnt, loff_t *fpos)
3418 if (tracing_disabled)
3421 if (cnt > TRACE_BUF_SIZE)
3422 cnt = TRACE_BUF_SIZE;
3424 buf = kmalloc(cnt + 2, GFP_KERNEL);
3428 if (copy_from_user(buf, ubuf, cnt)) {
3432 if (buf[cnt-1] != '\n') {
3438 cnt = mark_printk("%s", buf);
3445 static int tracing_clock_show(struct seq_file *m, void *v)
3449 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3451 "%s%s%s%s", i ? " " : "",
3452 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3453 i == trace_clock_id ? "]" : "");
3459 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3460 size_t cnt, loff_t *fpos)
3463 const char *clockstr;
3466 if (cnt >= sizeof(buf))
3469 if (copy_from_user(&buf, ubuf, cnt))
3474 clockstr = strstrip(buf);
3476 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3477 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3480 if (i == ARRAY_SIZE(trace_clocks))
3485 mutex_lock(&trace_types_lock);
3487 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3489 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3491 mutex_unlock(&trace_types_lock);
3498 static int tracing_clock_open(struct inode *inode, struct file *file)
3500 if (tracing_disabled)
3502 return single_open(file, tracing_clock_show, NULL);
3505 static const struct file_operations tracing_max_lat_fops = {
3506 .open = tracing_open_generic,
3507 .read = tracing_max_lat_read,
3508 .write = tracing_max_lat_write,
3511 static const struct file_operations tracing_ctrl_fops = {
3512 .open = tracing_open_generic,
3513 .read = tracing_ctrl_read,
3514 .write = tracing_ctrl_write,
3517 static const struct file_operations set_tracer_fops = {
3518 .open = tracing_open_generic,
3519 .read = tracing_set_trace_read,
3520 .write = tracing_set_trace_write,
3523 static const struct file_operations tracing_pipe_fops = {
3524 .open = tracing_open_pipe,
3525 .poll = tracing_poll_pipe,
3526 .read = tracing_read_pipe,
3527 .splice_read = tracing_splice_read_pipe,
3528 .release = tracing_release_pipe,
3531 static const struct file_operations tracing_entries_fops = {
3532 .open = tracing_open_generic,
3533 .read = tracing_entries_read,
3534 .write = tracing_entries_write,
3537 static const struct file_operations tracing_mark_fops = {
3538 .open = tracing_open_generic,
3539 .write = tracing_mark_write,
3542 static const struct file_operations trace_clock_fops = {
3543 .open = tracing_clock_open,
3545 .llseek = seq_lseek,
3546 .release = single_release,
3547 .write = tracing_clock_write,
3550 struct ftrace_buffer_info {
3551 struct trace_array *tr;
3557 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3559 int cpu = (int)(long)inode->i_private;
3560 struct ftrace_buffer_info *info;
3562 if (tracing_disabled)
3565 info = kzalloc(sizeof(*info), GFP_KERNEL);
3569 info->tr = &global_trace;
3572 /* Force reading ring buffer for first read */
3573 info->read = (unsigned int)-1;
3575 filp->private_data = info;
3577 return nonseekable_open(inode, filp);
3581 tracing_buffers_read(struct file *filp, char __user *ubuf,
3582 size_t count, loff_t *ppos)
3584 struct ftrace_buffer_info *info = filp->private_data;
3593 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3597 /* Do we have previous read data to read? */
3598 if (info->read < PAGE_SIZE)
3603 trace_access_lock(info->cpu);
3604 ret = ring_buffer_read_page(info->tr->buffer,
3608 trace_access_unlock(info->cpu);
3612 pos = ring_buffer_page_len(info->spare);
3614 if (pos < PAGE_SIZE)
3615 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3618 size = PAGE_SIZE - info->read;
3622 ret = copy_to_user(ubuf, info->spare + info->read, size);
3633 static int tracing_buffers_release(struct inode *inode, struct file *file)
3635 struct ftrace_buffer_info *info = file->private_data;
3638 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3645 struct ring_buffer *buffer;
3650 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3651 struct pipe_buffer *buf)
3653 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3658 ring_buffer_free_read_page(ref->buffer, ref->page);
3663 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3664 struct pipe_buffer *buf)
3669 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3670 struct pipe_buffer *buf)
3672 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3677 /* Pipe buffer operations for a buffer. */
3678 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
3680 .map = generic_pipe_buf_map,
3681 .unmap = generic_pipe_buf_unmap,
3682 .confirm = generic_pipe_buf_confirm,
3683 .release = buffer_pipe_buf_release,
3684 .steal = buffer_pipe_buf_steal,
3685 .get = buffer_pipe_buf_get,
3689 * Callback from splice_to_pipe(), if we need to release some pages
3690 * at the end of the spd in case we error'ed out in filling the pipe.
3692 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3694 struct buffer_ref *ref =
3695 (struct buffer_ref *)spd->partial[i].private;
3700 ring_buffer_free_read_page(ref->buffer, ref->page);
3702 spd->partial[i].private = 0;
3706 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3707 struct pipe_inode_info *pipe, size_t len,
3710 struct ftrace_buffer_info *info = file->private_data;
3711 struct partial_page partial[PIPE_BUFFERS];
3712 struct page *pages[PIPE_BUFFERS];
3713 struct splice_pipe_desc spd = {
3717 .ops = &buffer_pipe_buf_ops,
3718 .spd_release = buffer_spd_release,
3720 struct buffer_ref *ref;
3721 int entries, size, i;
3724 if (*ppos & (PAGE_SIZE - 1)) {
3725 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3729 if (len & (PAGE_SIZE - 1)) {
3730 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3731 if (len < PAGE_SIZE)
3736 trace_access_lock(info->cpu);
3737 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3739 for (i = 0; i < PIPE_BUFFERS && len && entries; i++, len -= PAGE_SIZE) {
3743 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3748 ref->buffer = info->tr->buffer;
3749 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3755 r = ring_buffer_read_page(ref->buffer, &ref->page,
3758 ring_buffer_free_read_page(ref->buffer,
3765 * zero out any left over data, this is going to
3768 size = ring_buffer_page_len(ref->page);
3769 if (size < PAGE_SIZE)
3770 memset(ref->page + size, 0, PAGE_SIZE - size);
3772 page = virt_to_page(ref->page);
3774 spd.pages[i] = page;
3775 spd.partial[i].len = PAGE_SIZE;
3776 spd.partial[i].offset = 0;
3777 spd.partial[i].private = (unsigned long)ref;
3781 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3784 trace_access_unlock(info->cpu);
3787 /* did we read anything? */
3788 if (!spd.nr_pages) {
3789 if (flags & SPLICE_F_NONBLOCK)
3797 ret = splice_to_pipe(pipe, &spd);
3802 static const struct file_operations tracing_buffers_fops = {
3803 .open = tracing_buffers_open,
3804 .read = tracing_buffers_read,
3805 .release = tracing_buffers_release,
3806 .splice_read = tracing_buffers_splice_read,
3807 .llseek = no_llseek,
3811 tracing_stats_read(struct file *filp, char __user *ubuf,
3812 size_t count, loff_t *ppos)
3814 unsigned long cpu = (unsigned long)filp->private_data;
3815 struct trace_array *tr = &global_trace;
3816 struct trace_seq *s;
3819 s = kmalloc(sizeof(*s), GFP_KERNEL);
3825 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3826 trace_seq_printf(s, "entries: %ld\n", cnt);
3828 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3829 trace_seq_printf(s, "overrun: %ld\n", cnt);
3831 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3832 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3834 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3841 static const struct file_operations tracing_stats_fops = {
3842 .open = tracing_open_generic,
3843 .read = tracing_stats_read,
3846 #ifdef CONFIG_DYNAMIC_FTRACE
3848 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3854 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3855 size_t cnt, loff_t *ppos)
3857 static char ftrace_dyn_info_buffer[1024];
3858 static DEFINE_MUTEX(dyn_info_mutex);
3859 unsigned long *p = filp->private_data;
3860 char *buf = ftrace_dyn_info_buffer;
3861 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3864 mutex_lock(&dyn_info_mutex);
3865 r = sprintf(buf, "%ld ", *p);
3867 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3870 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3872 mutex_unlock(&dyn_info_mutex);
3877 static const struct file_operations tracing_dyn_info_fops = {
3878 .open = tracing_open_generic,
3879 .read = tracing_read_dyn_info,
3883 static struct dentry *d_tracer;
3885 struct dentry *tracing_init_dentry(void)
3892 if (!debugfs_initialized())
3895 d_tracer = debugfs_create_dir("tracing", NULL);
3897 if (!d_tracer && !once) {
3899 pr_warning("Could not create debugfs directory 'tracing'\n");
3906 static struct dentry *d_percpu;
3908 struct dentry *tracing_dentry_percpu(void)
3911 struct dentry *d_tracer;
3916 d_tracer = tracing_init_dentry();
3921 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3923 if (!d_percpu && !once) {
3925 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3932 static void tracing_init_debugfs_percpu(long cpu)
3934 struct dentry *d_percpu = tracing_dentry_percpu();
3935 struct dentry *d_cpu;
3936 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3939 if (cpu > 999 || cpu < 0)
3942 sprintf(cpu_dir, "cpu%ld", cpu);
3943 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3945 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3949 /* per cpu trace_pipe */
3950 trace_create_file("trace_pipe", 0444, d_cpu,
3951 (void *) cpu, &tracing_pipe_fops);
3954 trace_create_file("trace", 0644, d_cpu,
3955 (void *) cpu, &tracing_fops);
3957 trace_create_file("trace_pipe_raw", 0444, d_cpu,
3958 (void *) cpu, &tracing_buffers_fops);
3960 trace_create_file("stats", 0444, d_cpu,
3961 (void *) cpu, &tracing_stats_fops);
3964 #ifdef CONFIG_FTRACE_SELFTEST
3965 /* Let selftest have access to static functions in this file */
3966 #include "trace_selftest.c"
3969 struct trace_option_dentry {
3970 struct tracer_opt *opt;
3971 struct tracer_flags *flags;
3972 struct dentry *entry;
3976 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3979 struct trace_option_dentry *topt = filp->private_data;
3982 if (topt->flags->val & topt->opt->bit)
3987 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3991 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3994 struct trace_option_dentry *topt = filp->private_data;
3999 if (cnt >= sizeof(buf))
4002 if (copy_from_user(&buf, ubuf, cnt))
4007 ret = strict_strtoul(buf, 10, &val);
4011 if (val != 0 && val != 1)
4014 if (!!(topt->flags->val & topt->opt->bit) != val) {
4015 mutex_lock(&trace_types_lock);
4016 ret = __set_tracer_option(current_trace, topt->flags,
4018 mutex_unlock(&trace_types_lock);
4029 static const struct file_operations trace_options_fops = {
4030 .open = tracing_open_generic,
4031 .read = trace_options_read,
4032 .write = trace_options_write,
4036 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4039 long index = (long)filp->private_data;
4042 if (trace_flags & (1 << index))
4047 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4051 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4054 long index = (long)filp->private_data;
4059 if (cnt >= sizeof(buf))
4062 if (copy_from_user(&buf, ubuf, cnt))
4067 ret = strict_strtoul(buf, 10, &val);
4071 if (val != 0 && val != 1)
4073 set_tracer_flags(1 << index, val);
4080 static const struct file_operations trace_options_core_fops = {
4081 .open = tracing_open_generic,
4082 .read = trace_options_core_read,
4083 .write = trace_options_core_write,
4086 struct dentry *trace_create_file(const char *name,
4088 struct dentry *parent,
4090 const struct file_operations *fops)
4094 ret = debugfs_create_file(name, mode, parent, data, fops);
4096 pr_warning("Could not create debugfs '%s' entry\n", name);
4102 static struct dentry *trace_options_init_dentry(void)
4104 struct dentry *d_tracer;
4105 static struct dentry *t_options;
4110 d_tracer = tracing_init_dentry();
4114 t_options = debugfs_create_dir("options", d_tracer);
4116 pr_warning("Could not create debugfs directory 'options'\n");
4124 create_trace_option_file(struct trace_option_dentry *topt,
4125 struct tracer_flags *flags,
4126 struct tracer_opt *opt)
4128 struct dentry *t_options;
4130 t_options = trace_options_init_dentry();
4134 topt->flags = flags;
4137 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4138 &trace_options_fops);
4142 static struct trace_option_dentry *
4143 create_trace_option_files(struct tracer *tracer)
4145 struct trace_option_dentry *topts;
4146 struct tracer_flags *flags;
4147 struct tracer_opt *opts;
4153 flags = tracer->flags;
4155 if (!flags || !flags->opts)
4160 for (cnt = 0; opts[cnt].name; cnt++)
4163 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4167 for (cnt = 0; opts[cnt].name; cnt++)
4168 create_trace_option_file(&topts[cnt], flags,
4175 destroy_trace_option_files(struct trace_option_dentry *topts)
4182 for (cnt = 0; topts[cnt].opt; cnt++) {
4183 if (topts[cnt].entry)
4184 debugfs_remove(topts[cnt].entry);
4190 static struct dentry *
4191 create_trace_option_core_file(const char *option, long index)
4193 struct dentry *t_options;
4195 t_options = trace_options_init_dentry();
4199 return trace_create_file(option, 0644, t_options, (void *)index,
4200 &trace_options_core_fops);
4203 static __init void create_trace_options_dir(void)
4205 struct dentry *t_options;
4208 t_options = trace_options_init_dentry();
4212 for (i = 0; trace_options[i]; i++)
4213 create_trace_option_core_file(trace_options[i], i);
4216 static __init int tracer_init_debugfs(void)
4218 struct dentry *d_tracer;
4221 trace_access_lock_init();
4223 d_tracer = tracing_init_dentry();
4225 trace_create_file("tracing_enabled", 0644, d_tracer,
4226 &global_trace, &tracing_ctrl_fops);
4228 trace_create_file("trace_options", 0644, d_tracer,
4229 NULL, &tracing_iter_fops);
4231 trace_create_file("tracing_cpumask", 0644, d_tracer,
4232 NULL, &tracing_cpumask_fops);
4234 trace_create_file("trace", 0644, d_tracer,
4235 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4237 trace_create_file("available_tracers", 0444, d_tracer,
4238 &global_trace, &show_traces_fops);
4240 trace_create_file("current_tracer", 0644, d_tracer,
4241 &global_trace, &set_tracer_fops);
4243 #ifdef CONFIG_TRACER_MAX_TRACE
4244 trace_create_file("tracing_max_latency", 0644, d_tracer,
4245 &tracing_max_latency, &tracing_max_lat_fops);
4247 trace_create_file("tracing_thresh", 0644, d_tracer,
4248 &tracing_thresh, &tracing_max_lat_fops);
4251 trace_create_file("README", 0444, d_tracer,
4252 NULL, &tracing_readme_fops);
4254 trace_create_file("trace_pipe", 0444, d_tracer,
4255 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4257 trace_create_file("buffer_size_kb", 0644, d_tracer,
4258 &global_trace, &tracing_entries_fops);
4260 trace_create_file("trace_marker", 0220, d_tracer,
4261 NULL, &tracing_mark_fops);
4263 trace_create_file("saved_cmdlines", 0444, d_tracer,
4264 NULL, &tracing_saved_cmdlines_fops);
4266 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4269 #ifdef CONFIG_DYNAMIC_FTRACE
4270 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4271 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4273 #ifdef CONFIG_SYSPROF_TRACER
4274 init_tracer_sysprof_debugfs(d_tracer);
4277 create_trace_options_dir();
4279 for_each_tracing_cpu(cpu)
4280 tracing_init_debugfs_percpu(cpu);
4285 static int trace_panic_handler(struct notifier_block *this,
4286 unsigned long event, void *unused)
4288 if (ftrace_dump_on_oops)
4293 static struct notifier_block trace_panic_notifier = {
4294 .notifier_call = trace_panic_handler,
4296 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4299 static int trace_die_handler(struct notifier_block *self,
4305 if (ftrace_dump_on_oops)
4314 static struct notifier_block trace_die_notifier = {
4315 .notifier_call = trace_die_handler,
4320 * printk is set to max of 1024, we really don't need it that big.
4321 * Nothing should be printing 1000 characters anyway.
4323 #define TRACE_MAX_PRINT 1000
4326 * Define here KERN_TRACE so that we have one place to modify
4327 * it if we decide to change what log level the ftrace dump
4330 #define KERN_TRACE KERN_EMERG
4333 trace_printk_seq(struct trace_seq *s)
4335 /* Probably should print a warning here. */
4339 /* should be zero ended, but we are paranoid. */
4340 s->buffer[s->len] = 0;
4342 printk(KERN_TRACE "%s", s->buffer);
4347 static void __ftrace_dump(bool disable_tracing)
4349 static arch_spinlock_t ftrace_dump_lock =
4350 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4351 /* use static because iter can be a bit big for the stack */
4352 static struct trace_iterator iter;
4353 unsigned int old_userobj;
4354 static int dump_ran;
4355 unsigned long flags;
4359 local_irq_save(flags);
4360 arch_spin_lock(&ftrace_dump_lock);
4368 if (disable_tracing)
4371 for_each_tracing_cpu(cpu) {
4372 atomic_inc(&global_trace.data[cpu]->disabled);
4375 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4377 /* don't look at user memory in panic mode */
4378 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4380 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4382 /* Simulate the iterator */
4383 iter.tr = &global_trace;
4384 iter.trace = current_trace;
4385 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4388 * We need to stop all tracing on all CPUS to read the
4389 * the next buffer. This is a bit expensive, but is
4390 * not done often. We fill all what we can read,
4391 * and then release the locks again.
4394 while (!trace_empty(&iter)) {
4397 printk(KERN_TRACE "---------------------------------\n");
4401 /* reset all but tr, trace, and overruns */
4402 memset(&iter.seq, 0,
4403 sizeof(struct trace_iterator) -
4404 offsetof(struct trace_iterator, seq));
4405 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4408 if (find_next_entry_inc(&iter) != NULL) {
4411 ret = print_trace_line(&iter);
4412 if (ret != TRACE_TYPE_NO_CONSUME)
4413 trace_consume(&iter);
4416 trace_printk_seq(&iter.seq);
4420 printk(KERN_TRACE " (ftrace buffer empty)\n");
4422 printk(KERN_TRACE "---------------------------------\n");
4424 /* Re-enable tracing if requested */
4425 if (!disable_tracing) {
4426 trace_flags |= old_userobj;
4428 for_each_tracing_cpu(cpu) {
4429 atomic_dec(&global_trace.data[cpu]->disabled);
4435 arch_spin_unlock(&ftrace_dump_lock);
4436 local_irq_restore(flags);
4439 /* By default: disable tracing after the dump */
4440 void ftrace_dump(void)
4442 __ftrace_dump(true);
4445 __init static int tracer_alloc_buffers(void)
4451 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4454 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4455 goto out_free_buffer_mask;
4457 /* To save memory, keep the ring buffer size to its minimum */
4458 if (ring_buffer_expanded)
4459 ring_buf_size = trace_buf_size;
4463 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4464 cpumask_copy(tracing_cpumask, cpu_all_mask);
4466 /* TODO: make the number of buffers hot pluggable with CPUS */
4467 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4468 TRACE_BUFFER_FLAGS);
4469 if (!global_trace.buffer) {
4470 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4472 goto out_free_cpumask;
4474 global_trace.entries = ring_buffer_size(global_trace.buffer);
4477 #ifdef CONFIG_TRACER_MAX_TRACE
4478 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4479 TRACE_BUFFER_FLAGS);
4480 if (!max_tr.buffer) {
4481 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4483 ring_buffer_free(global_trace.buffer);
4484 goto out_free_cpumask;
4486 max_tr.entries = ring_buffer_size(max_tr.buffer);
4487 WARN_ON(max_tr.entries != global_trace.entries);
4490 /* Allocate the first page for all buffers */
4491 for_each_tracing_cpu(i) {
4492 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4493 max_tr.data[i] = &per_cpu(max_tr_data, i);
4496 trace_init_cmdlines();
4498 register_tracer(&nop_trace);
4499 current_trace = &nop_trace;
4500 #ifdef CONFIG_BOOT_TRACER
4501 register_tracer(&boot_tracer);
4503 /* All seems OK, enable tracing */
4504 tracing_disabled = 0;
4506 atomic_notifier_chain_register(&panic_notifier_list,
4507 &trace_panic_notifier);
4509 register_die_notifier(&trace_die_notifier);
4514 free_cpumask_var(tracing_cpumask);
4515 out_free_buffer_mask:
4516 free_cpumask_var(tracing_buffer_mask);
4521 __init static int clear_boot_tracer(void)
4524 * The default tracer at boot buffer is an init section.
4525 * This function is called in lateinit. If we did not
4526 * find the boot tracer, then clear it out, to prevent
4527 * later registration from accessing the buffer that is
4528 * about to be freed.
4530 if (!default_bootup_tracer)
4533 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4534 default_bootup_tracer);
4535 default_bootup_tracer = NULL;
4540 early_initcall(tracer_alloc_buffers);
4541 fs_initcall(tracer_init_debugfs);
4542 late_initcall(clear_boot_tracer);