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 <linux/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/notifier.h>
21 #include <linux/irqflags.h>
22 #include <linux/debugfs.h>
23 #include <linux/pagemap.h>
24 #include <linux/hardirq.h>
25 #include <linux/linkage.h>
26 #include <linux/uaccess.h>
27 #include <linux/kprobes.h>
28 #include <linux/ftrace.h>
29 #include <linux/module.h>
30 #include <linux/percpu.h>
31 #include <linux/splice.h>
32 #include <linux/kdebug.h>
33 #include <linux/string.h>
34 #include <linux/ctype.h>
35 #include <linux/init.h>
36 #include <linux/poll.h>
37 #include <linux/gfp.h>
41 #include "trace_output.h"
43 #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
45 unsigned long __read_mostly tracing_max_latency;
46 unsigned long __read_mostly tracing_thresh;
49 * On boot up, the ring buffer is set to the minimum size, so that
50 * we do not waste memory on systems that are not using tracing.
52 static int ring_buffer_expanded;
55 * We need to change this state when a selftest is running.
56 * A selftest will lurk into the ring-buffer to count the
57 * entries inserted during the selftest although some concurrent
58 * insertions into the ring-buffer such as trace_printk could occurred
59 * at the same time, giving false positive or negative results.
61 static bool __read_mostly tracing_selftest_running;
64 * If a tracer is running, we do not want to run SELFTEST.
66 static bool __read_mostly tracing_selftest_disabled;
68 /* For tracers that don't implement custom flags */
69 static struct tracer_opt dummy_tracer_opt[] = {
73 static struct tracer_flags dummy_tracer_flags = {
75 .opts = dummy_tracer_opt
78 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
84 * Kill all tracing for good (never come back).
85 * It is initialized to 1 but will turn to zero if the initialization
86 * of the tracer is successful. But that is the only place that sets
89 static int tracing_disabled = 1;
91 static DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
93 static inline void ftrace_disable_cpu(void)
96 local_inc(&__get_cpu_var(ftrace_cpu_disabled));
99 static inline void ftrace_enable_cpu(void)
101 local_dec(&__get_cpu_var(ftrace_cpu_disabled));
105 static cpumask_var_t __read_mostly tracing_buffer_mask;
107 /* Define which cpu buffers are currently read in trace_pipe */
108 static cpumask_var_t tracing_reader_cpumask;
110 #define for_each_tracing_cpu(cpu) \
111 for_each_cpu(cpu, tracing_buffer_mask)
114 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
116 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
117 * is set, then ftrace_dump is called. This will output the contents
118 * of the ftrace buffers to the console. This is very useful for
119 * capturing traces that lead to crashes and outputing it to a
122 * It is default off, but you can enable it with either specifying
123 * "ftrace_dump_on_oops" in the kernel command line, or setting
124 * /proc/sys/kernel/ftrace_dump_on_oops to true.
126 int ftrace_dump_on_oops;
128 static int tracing_set_tracer(const char *buf);
130 #define BOOTUP_TRACER_SIZE 100
131 static char bootup_tracer_buf[BOOTUP_TRACER_SIZE] __initdata;
132 static char *default_bootup_tracer;
134 static int __init set_ftrace(char *str)
136 strncpy(bootup_tracer_buf, str, BOOTUP_TRACER_SIZE);
137 default_bootup_tracer = bootup_tracer_buf;
138 /* We are using ftrace early, expand it */
139 ring_buffer_expanded = 1;
142 __setup("ftrace=", set_ftrace);
144 static int __init set_ftrace_dump_on_oops(char *str)
146 ftrace_dump_on_oops = 1;
149 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
151 unsigned long long ns2usecs(cycle_t nsec)
159 * The global_trace is the descriptor that holds the tracing
160 * buffers for the live tracing. For each CPU, it contains
161 * a link list of pages that will store trace entries. The
162 * page descriptor of the pages in the memory is used to hold
163 * the link list by linking the lru item in the page descriptor
164 * to each of the pages in the buffer per CPU.
166 * For each active CPU there is a data field that holds the
167 * pages for the buffer for that CPU. Each CPU has the same number
168 * of pages allocated for its buffer.
170 static struct trace_array global_trace;
172 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
174 int filter_current_check_discard(struct ftrace_event_call *call, void *rec,
175 struct ring_buffer_event *event)
177 return filter_check_discard(call, rec, global_trace.buffer, event);
179 EXPORT_SYMBOL_GPL(filter_current_check_discard);
181 cycle_t ftrace_now(int cpu)
185 /* Early boot up does not have a buffer yet */
186 if (!global_trace.buffer)
187 return trace_clock_local();
189 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
190 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
196 * The max_tr is used to snapshot the global_trace when a maximum
197 * latency is reached. Some tracers will use this to store a maximum
198 * trace while it continues examining live traces.
200 * The buffers for the max_tr are set up the same as the global_trace.
201 * When a snapshot is taken, the link list of the max_tr is swapped
202 * with the link list of the global_trace and the buffers are reset for
203 * the global_trace so the tracing can continue.
205 static struct trace_array max_tr;
207 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
209 /* tracer_enabled is used to toggle activation of a tracer */
210 static int tracer_enabled = 1;
213 * tracing_is_enabled - return tracer_enabled status
215 * This function is used by other tracers to know the status
216 * of the tracer_enabled flag. Tracers may use this function
217 * to know if it should enable their features when starting
218 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
220 int tracing_is_enabled(void)
222 return tracer_enabled;
226 * trace_buf_size is the size in bytes that is allocated
227 * for a buffer. Note, the number of bytes is always rounded
230 * This number is purposely set to a low number of 16384.
231 * If the dump on oops happens, it will be much appreciated
232 * to not have to wait for all that output. Anyway this can be
233 * boot time and run time configurable.
235 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
237 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
239 /* trace_types holds a link list of available tracers. */
240 static struct tracer *trace_types __read_mostly;
242 /* current_trace points to the tracer that is currently active */
243 static struct tracer *current_trace __read_mostly;
246 * max_tracer_type_len is used to simplify the allocating of
247 * buffers to read userspace tracer names. We keep track of
248 * the longest tracer name registered.
250 static int max_tracer_type_len;
253 * trace_types_lock is used to protect the trace_types list.
254 * This lock is also used to keep user access serialized.
255 * Accesses from userspace will grab this lock while userspace
256 * activities happen inside the kernel.
258 static DEFINE_MUTEX(trace_types_lock);
260 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
261 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
263 /* trace_flags holds trace_options default values */
264 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
265 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
266 TRACE_ITER_GRAPH_TIME;
269 * trace_wake_up - wake up tasks waiting for trace input
271 * Simply wakes up any task that is blocked on the trace_wait
272 * queue. These is used with trace_poll for tasks polling the trace.
274 void trace_wake_up(void)
277 * The runqueue_is_locked() can fail, but this is the best we
280 if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
281 wake_up(&trace_wait);
284 static int __init set_buf_size(char *str)
286 unsigned long buf_size;
291 ret = strict_strtoul(str, 0, &buf_size);
292 /* nr_entries can not be zero */
293 if (ret < 0 || buf_size == 0)
295 trace_buf_size = buf_size;
298 __setup("trace_buf_size=", set_buf_size);
300 unsigned long nsecs_to_usecs(unsigned long nsecs)
305 /* These must match the bit postions in trace_iterator_flags */
306 static const char *trace_options[] = {
333 * ftrace_max_lock is used to protect the swapping of buffers
334 * when taking a max snapshot. The buffers themselves are
335 * protected by per_cpu spinlocks. But the action of the swap
336 * needs its own lock.
338 * This is defined as a raw_spinlock_t in order to help
339 * with performance when lockdep debugging is enabled.
341 static raw_spinlock_t ftrace_max_lock =
342 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
345 * Copy the new maximum trace into the separate maximum-trace
346 * structure. (this way the maximum trace is permanently saved,
347 * for later retrieval via /debugfs/tracing/latency_trace)
350 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
352 struct trace_array_cpu *data = tr->data[cpu];
355 max_tr.time_start = data->preempt_timestamp;
357 data = max_tr.data[cpu];
358 data->saved_latency = tracing_max_latency;
360 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
361 data->pid = tsk->pid;
362 data->uid = task_uid(tsk);
363 data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
364 data->policy = tsk->policy;
365 data->rt_priority = tsk->rt_priority;
367 /* record this tasks comm */
368 tracing_record_cmdline(tsk);
371 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
379 if (s->len <= s->readpos)
382 len = s->len - s->readpos;
385 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
395 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
400 if (s->len <= s->readpos)
403 len = s->len - s->readpos;
406 ret = memcpy(buf, s->buffer + s->readpos, cnt);
415 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
417 * @tsk: the task with the latency
418 * @cpu: The cpu that initiated the trace.
420 * Flip the buffers between the @tr and the max_tr and record information
421 * about which task was the cause of this latency.
424 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
426 struct ring_buffer *buf = tr->buffer;
428 WARN_ON_ONCE(!irqs_disabled());
429 __raw_spin_lock(&ftrace_max_lock);
431 tr->buffer = max_tr.buffer;
434 ftrace_disable_cpu();
435 ring_buffer_reset(tr->buffer);
438 __update_max_tr(tr, tsk, cpu);
439 __raw_spin_unlock(&ftrace_max_lock);
443 * update_max_tr_single - only copy one trace over, and reset the rest
445 * @tsk - task with the latency
446 * @cpu - the cpu of the buffer to copy.
448 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
451 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
455 WARN_ON_ONCE(!irqs_disabled());
456 __raw_spin_lock(&ftrace_max_lock);
458 ftrace_disable_cpu();
460 ring_buffer_reset(max_tr.buffer);
461 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
465 WARN_ON_ONCE(ret && ret != -EAGAIN);
467 __update_max_tr(tr, tsk, cpu);
468 __raw_spin_unlock(&ftrace_max_lock);
472 * register_tracer - register a tracer with the ftrace system.
473 * @type - the plugin for the tracer
475 * Register a new plugin tracer.
477 int register_tracer(struct tracer *type)
478 __releases(kernel_lock)
479 __acquires(kernel_lock)
486 pr_info("Tracer must have a name\n");
491 * When this gets called we hold the BKL which means that
492 * preemption is disabled. Various trace selftests however
493 * need to disable and enable preemption for successful tests.
494 * So we drop the BKL here and grab it after the tests again.
497 mutex_lock(&trace_types_lock);
499 tracing_selftest_running = true;
501 for (t = trace_types; t; t = t->next) {
502 if (strcmp(type->name, t->name) == 0) {
504 pr_info("Trace %s already registered\n",
512 type->set_flag = &dummy_set_flag;
514 type->flags = &dummy_tracer_flags;
516 if (!type->flags->opts)
517 type->flags->opts = dummy_tracer_opt;
518 if (!type->wait_pipe)
519 type->wait_pipe = default_wait_pipe;
522 #ifdef CONFIG_FTRACE_STARTUP_TEST
523 if (type->selftest && !tracing_selftest_disabled) {
524 struct tracer *saved_tracer = current_trace;
525 struct trace_array *tr = &global_trace;
529 * Run a selftest on this tracer.
530 * Here we reset the trace buffer, and set the current
531 * tracer to be this tracer. The tracer can then run some
532 * internal tracing to verify that everything is in order.
533 * If we fail, we do not register this tracer.
535 for_each_tracing_cpu(i)
536 tracing_reset(tr, i);
538 current_trace = type;
539 /* the test is responsible for initializing and enabling */
540 pr_info("Testing tracer %s: ", type->name);
541 ret = type->selftest(type, tr);
542 /* the test is responsible for resetting too */
543 current_trace = saved_tracer;
545 printk(KERN_CONT "FAILED!\n");
548 /* Only reset on passing, to avoid touching corrupted buffers */
549 for_each_tracing_cpu(i)
550 tracing_reset(tr, i);
552 printk(KERN_CONT "PASSED\n");
556 type->next = trace_types;
558 len = strlen(type->name);
559 if (len > max_tracer_type_len)
560 max_tracer_type_len = len;
563 tracing_selftest_running = false;
564 mutex_unlock(&trace_types_lock);
566 if (ret || !default_bootup_tracer)
569 if (strncmp(default_bootup_tracer, type->name, BOOTUP_TRACER_SIZE))
572 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
573 /* Do we want this tracer to start on bootup? */
574 tracing_set_tracer(type->name);
575 default_bootup_tracer = NULL;
576 /* disable other selftests, since this will break it. */
577 tracing_selftest_disabled = 1;
578 #ifdef CONFIG_FTRACE_STARTUP_TEST
579 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
588 void unregister_tracer(struct tracer *type)
593 mutex_lock(&trace_types_lock);
594 for (t = &trace_types; *t; t = &(*t)->next) {
598 pr_info("Trace %s not registered\n", type->name);
604 if (type == current_trace && tracer_enabled) {
607 if (current_trace->stop)
608 current_trace->stop(&global_trace);
609 current_trace = &nop_trace;
612 if (strlen(type->name) != max_tracer_type_len)
615 max_tracer_type_len = 0;
616 for (t = &trace_types; *t; t = &(*t)->next) {
617 len = strlen((*t)->name);
618 if (len > max_tracer_type_len)
619 max_tracer_type_len = len;
622 mutex_unlock(&trace_types_lock);
625 void tracing_reset(struct trace_array *tr, int cpu)
627 ftrace_disable_cpu();
628 ring_buffer_reset_cpu(tr->buffer, cpu);
632 void tracing_reset_online_cpus(struct trace_array *tr)
636 tr->time_start = ftrace_now(tr->cpu);
638 for_each_online_cpu(cpu)
639 tracing_reset(tr, cpu);
642 #define SAVED_CMDLINES 128
643 #define NO_CMDLINE_MAP UINT_MAX
644 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
645 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
646 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
647 static int cmdline_idx;
648 static raw_spinlock_t trace_cmdline_lock = __RAW_SPIN_LOCK_UNLOCKED;
650 /* temporary disable recording */
651 static atomic_t trace_record_cmdline_disabled __read_mostly;
653 static void trace_init_cmdlines(void)
655 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
656 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
660 static int trace_stop_count;
661 static DEFINE_SPINLOCK(tracing_start_lock);
664 * ftrace_off_permanent - disable all ftrace code permanently
666 * This should only be called when a serious anomally has
667 * been detected. This will turn off the function tracing,
668 * ring buffers, and other tracing utilites. It takes no
669 * locks and can be called from any context.
671 void ftrace_off_permanent(void)
673 tracing_disabled = 1;
675 tracing_off_permanent();
679 * tracing_start - quick start of the tracer
681 * If tracing is enabled but was stopped by tracing_stop,
682 * this will start the tracer back up.
684 void tracing_start(void)
686 struct ring_buffer *buffer;
689 if (tracing_disabled)
692 spin_lock_irqsave(&tracing_start_lock, flags);
693 if (--trace_stop_count) {
694 if (trace_stop_count < 0) {
695 /* Someone screwed up their debugging */
697 trace_stop_count = 0;
703 buffer = global_trace.buffer;
705 ring_buffer_record_enable(buffer);
707 buffer = max_tr.buffer;
709 ring_buffer_record_enable(buffer);
713 spin_unlock_irqrestore(&tracing_start_lock, flags);
717 * tracing_stop - quick stop of the tracer
719 * Light weight way to stop tracing. Use in conjunction with
722 void tracing_stop(void)
724 struct ring_buffer *buffer;
728 spin_lock_irqsave(&tracing_start_lock, flags);
729 if (trace_stop_count++)
732 buffer = global_trace.buffer;
734 ring_buffer_record_disable(buffer);
736 buffer = max_tr.buffer;
738 ring_buffer_record_disable(buffer);
741 spin_unlock_irqrestore(&tracing_start_lock, flags);
744 void trace_stop_cmdline_recording(void);
746 static void trace_save_cmdline(struct task_struct *tsk)
750 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
754 * It's not the end of the world if we don't get
755 * the lock, but we also don't want to spin
756 * nor do we want to disable interrupts,
757 * so if we miss here, then better luck next time.
759 if (!__raw_spin_trylock(&trace_cmdline_lock))
762 idx = map_pid_to_cmdline[tsk->pid];
763 if (idx == NO_CMDLINE_MAP) {
764 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
767 * Check whether the cmdline buffer at idx has a pid
768 * mapped. We are going to overwrite that entry so we
769 * need to clear the map_pid_to_cmdline. Otherwise we
770 * would read the new comm for the old pid.
772 pid = map_cmdline_to_pid[idx];
773 if (pid != NO_CMDLINE_MAP)
774 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
776 map_cmdline_to_pid[idx] = tsk->pid;
777 map_pid_to_cmdline[tsk->pid] = idx;
782 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
784 __raw_spin_unlock(&trace_cmdline_lock);
787 void trace_find_cmdline(int pid, char comm[])
792 strcpy(comm, "<idle>");
796 if (pid > PID_MAX_DEFAULT) {
797 strcpy(comm, "<...>");
801 __raw_spin_lock(&trace_cmdline_lock);
802 map = map_pid_to_cmdline[pid];
803 if (map != NO_CMDLINE_MAP)
804 strcpy(comm, saved_cmdlines[map]);
806 strcpy(comm, "<...>");
808 __raw_spin_unlock(&trace_cmdline_lock);
811 void tracing_record_cmdline(struct task_struct *tsk)
813 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
817 trace_save_cmdline(tsk);
821 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
824 struct task_struct *tsk = current;
826 entry->preempt_count = pc & 0xff;
827 entry->pid = (tsk) ? tsk->pid : 0;
828 entry->tgid = (tsk) ? tsk->tgid : 0;
830 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
831 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
833 TRACE_FLAG_IRQS_NOSUPPORT |
835 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
836 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
837 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
840 struct ring_buffer_event *trace_buffer_lock_reserve(struct trace_array *tr,
843 unsigned long flags, int pc)
845 struct ring_buffer_event *event;
847 event = ring_buffer_lock_reserve(tr->buffer, len);
849 struct trace_entry *ent = ring_buffer_event_data(event);
851 tracing_generic_entry_update(ent, flags, pc);
857 static void ftrace_trace_stack(struct trace_array *tr,
858 unsigned long flags, int skip, int pc);
859 static void ftrace_trace_userstack(struct trace_array *tr,
860 unsigned long flags, int pc);
862 static inline void __trace_buffer_unlock_commit(struct trace_array *tr,
863 struct ring_buffer_event *event,
864 unsigned long flags, int pc,
867 ring_buffer_unlock_commit(tr->buffer, event);
869 ftrace_trace_stack(tr, flags, 6, pc);
870 ftrace_trace_userstack(tr, flags, pc);
876 void trace_buffer_unlock_commit(struct trace_array *tr,
877 struct ring_buffer_event *event,
878 unsigned long flags, int pc)
880 __trace_buffer_unlock_commit(tr, event, flags, pc, 1);
883 struct ring_buffer_event *
884 trace_current_buffer_lock_reserve(int type, unsigned long len,
885 unsigned long flags, int pc)
887 return trace_buffer_lock_reserve(&global_trace,
888 type, len, flags, pc);
890 EXPORT_SYMBOL(trace_current_buffer_lock_reserve);
892 void trace_current_buffer_unlock_commit(struct ring_buffer_event *event,
893 unsigned long flags, int pc)
895 __trace_buffer_unlock_commit(&global_trace, event, flags, pc, 1);
897 EXPORT_SYMBOL(trace_current_buffer_unlock_commit);
899 void trace_nowake_buffer_unlock_commit(struct ring_buffer_event *event,
900 unsigned long flags, int pc)
902 __trace_buffer_unlock_commit(&global_trace, event, flags, pc, 0);
904 EXPORT_SYMBOL(trace_nowake_buffer_unlock_commit);
906 void trace_current_buffer_discard_commit(struct ring_buffer_event *event)
908 ring_buffer_discard_commit(global_trace.buffer, event);
910 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
913 trace_function(struct trace_array *tr,
914 unsigned long ip, unsigned long parent_ip, unsigned long flags,
917 struct ftrace_event_call *call = &event_function;
918 struct ring_buffer_event *event;
919 struct ftrace_entry *entry;
921 /* If we are reading the ring buffer, don't trace */
922 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
925 event = trace_buffer_lock_reserve(tr, TRACE_FN, sizeof(*entry),
929 entry = ring_buffer_event_data(event);
931 entry->parent_ip = parent_ip;
933 if (!filter_check_discard(call, entry, tr->buffer, event))
934 ring_buffer_unlock_commit(tr->buffer, event);
937 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
938 static int __trace_graph_entry(struct trace_array *tr,
939 struct ftrace_graph_ent *trace,
943 struct ftrace_event_call *call = &event_funcgraph_entry;
944 struct ring_buffer_event *event;
945 struct ftrace_graph_ent_entry *entry;
947 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
950 event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_ENT,
951 sizeof(*entry), flags, pc);
954 entry = ring_buffer_event_data(event);
955 entry->graph_ent = *trace;
956 if (!filter_current_check_discard(call, entry, event))
957 ring_buffer_unlock_commit(global_trace.buffer, event);
962 static void __trace_graph_return(struct trace_array *tr,
963 struct ftrace_graph_ret *trace,
967 struct ftrace_event_call *call = &event_funcgraph_exit;
968 struct ring_buffer_event *event;
969 struct ftrace_graph_ret_entry *entry;
971 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
974 event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_RET,
975 sizeof(*entry), flags, pc);
978 entry = ring_buffer_event_data(event);
980 if (!filter_current_check_discard(call, entry, event))
981 ring_buffer_unlock_commit(global_trace.buffer, event);
986 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
987 unsigned long ip, unsigned long parent_ip, unsigned long flags,
990 if (likely(!atomic_read(&data->disabled)))
991 trace_function(tr, ip, parent_ip, flags, pc);
994 static void __ftrace_trace_stack(struct trace_array *tr,
998 #ifdef CONFIG_STACKTRACE
999 struct ftrace_event_call *call = &event_kernel_stack;
1000 struct ring_buffer_event *event;
1001 struct stack_entry *entry;
1002 struct stack_trace trace;
1004 event = trace_buffer_lock_reserve(tr, TRACE_STACK,
1005 sizeof(*entry), flags, pc);
1008 entry = ring_buffer_event_data(event);
1009 memset(&entry->caller, 0, sizeof(entry->caller));
1011 trace.nr_entries = 0;
1012 trace.max_entries = FTRACE_STACK_ENTRIES;
1014 trace.entries = entry->caller;
1016 save_stack_trace(&trace);
1017 if (!filter_check_discard(call, entry, tr->buffer, event))
1018 ring_buffer_unlock_commit(tr->buffer, event);
1022 static void ftrace_trace_stack(struct trace_array *tr,
1023 unsigned long flags,
1026 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1029 __ftrace_trace_stack(tr, flags, skip, pc);
1032 void __trace_stack(struct trace_array *tr,
1033 unsigned long flags,
1036 __ftrace_trace_stack(tr, flags, skip, pc);
1039 static void ftrace_trace_userstack(struct trace_array *tr,
1040 unsigned long flags, int pc)
1042 #ifdef CONFIG_STACKTRACE
1043 struct ftrace_event_call *call = &event_user_stack;
1044 struct ring_buffer_event *event;
1045 struct userstack_entry *entry;
1046 struct stack_trace trace;
1048 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1051 event = trace_buffer_lock_reserve(tr, TRACE_USER_STACK,
1052 sizeof(*entry), flags, pc);
1055 entry = ring_buffer_event_data(event);
1057 memset(&entry->caller, 0, sizeof(entry->caller));
1059 trace.nr_entries = 0;
1060 trace.max_entries = FTRACE_STACK_ENTRIES;
1062 trace.entries = entry->caller;
1064 save_stack_trace_user(&trace);
1065 if (!filter_check_discard(call, entry, tr->buffer, event))
1066 ring_buffer_unlock_commit(tr->buffer, event);
1071 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1073 ftrace_trace_userstack(tr, flags, preempt_count());
1078 ftrace_trace_special(void *__tr,
1079 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1082 struct ring_buffer_event *event;
1083 struct trace_array *tr = __tr;
1084 struct special_entry *entry;
1086 event = trace_buffer_lock_reserve(tr, TRACE_SPECIAL,
1087 sizeof(*entry), 0, pc);
1090 entry = ring_buffer_event_data(event);
1094 trace_buffer_unlock_commit(tr, event, 0, pc);
1098 __trace_special(void *__tr, void *__data,
1099 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1101 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1105 tracing_sched_switch_trace(struct trace_array *tr,
1106 struct task_struct *prev,
1107 struct task_struct *next,
1108 unsigned long flags, int pc)
1110 struct ftrace_event_call *call = &event_context_switch;
1111 struct ring_buffer_event *event;
1112 struct ctx_switch_entry *entry;
1114 event = trace_buffer_lock_reserve(tr, TRACE_CTX,
1115 sizeof(*entry), flags, pc);
1118 entry = ring_buffer_event_data(event);
1119 entry->prev_pid = prev->pid;
1120 entry->prev_prio = prev->prio;
1121 entry->prev_state = prev->state;
1122 entry->next_pid = next->pid;
1123 entry->next_prio = next->prio;
1124 entry->next_state = next->state;
1125 entry->next_cpu = task_cpu(next);
1127 if (!filter_check_discard(call, entry, tr->buffer, event))
1128 trace_buffer_unlock_commit(tr, event, flags, pc);
1132 tracing_sched_wakeup_trace(struct trace_array *tr,
1133 struct task_struct *wakee,
1134 struct task_struct *curr,
1135 unsigned long flags, int pc)
1137 struct ftrace_event_call *call = &event_wakeup;
1138 struct ring_buffer_event *event;
1139 struct ctx_switch_entry *entry;
1141 event = trace_buffer_lock_reserve(tr, TRACE_WAKE,
1142 sizeof(*entry), flags, pc);
1145 entry = ring_buffer_event_data(event);
1146 entry->prev_pid = curr->pid;
1147 entry->prev_prio = curr->prio;
1148 entry->prev_state = curr->state;
1149 entry->next_pid = wakee->pid;
1150 entry->next_prio = wakee->prio;
1151 entry->next_state = wakee->state;
1152 entry->next_cpu = task_cpu(wakee);
1154 if (!filter_check_discard(call, entry, tr->buffer, event))
1155 ring_buffer_unlock_commit(tr->buffer, event);
1156 ftrace_trace_stack(tr, flags, 6, pc);
1157 ftrace_trace_userstack(tr, flags, pc);
1161 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1163 struct trace_array *tr = &global_trace;
1164 struct trace_array_cpu *data;
1165 unsigned long flags;
1169 if (tracing_disabled)
1172 pc = preempt_count();
1173 local_irq_save(flags);
1174 cpu = raw_smp_processor_id();
1175 data = tr->data[cpu];
1177 if (likely(atomic_inc_return(&data->disabled) == 1))
1178 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1180 atomic_dec(&data->disabled);
1181 local_irq_restore(flags);
1184 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1185 int trace_graph_entry(struct ftrace_graph_ent *trace)
1187 struct trace_array *tr = &global_trace;
1188 struct trace_array_cpu *data;
1189 unsigned long flags;
1195 if (!ftrace_trace_task(current))
1198 if (!ftrace_graph_addr(trace->func))
1201 local_irq_save(flags);
1202 cpu = raw_smp_processor_id();
1203 data = tr->data[cpu];
1204 disabled = atomic_inc_return(&data->disabled);
1205 if (likely(disabled == 1)) {
1206 pc = preempt_count();
1207 ret = __trace_graph_entry(tr, trace, flags, pc);
1211 /* Only do the atomic if it is not already set */
1212 if (!test_tsk_trace_graph(current))
1213 set_tsk_trace_graph(current);
1215 atomic_dec(&data->disabled);
1216 local_irq_restore(flags);
1221 void trace_graph_return(struct ftrace_graph_ret *trace)
1223 struct trace_array *tr = &global_trace;
1224 struct trace_array_cpu *data;
1225 unsigned long flags;
1230 local_irq_save(flags);
1231 cpu = raw_smp_processor_id();
1232 data = tr->data[cpu];
1233 disabled = atomic_inc_return(&data->disabled);
1234 if (likely(disabled == 1)) {
1235 pc = preempt_count();
1236 __trace_graph_return(tr, trace, flags, pc);
1239 clear_tsk_trace_graph(current);
1240 atomic_dec(&data->disabled);
1241 local_irq_restore(flags);
1243 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1247 * trace_vbprintk - write binary msg to tracing buffer
1250 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1252 static raw_spinlock_t trace_buf_lock =
1253 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
1254 static u32 trace_buf[TRACE_BUF_SIZE];
1256 struct ftrace_event_call *call = &event_bprint;
1257 struct ring_buffer_event *event;
1258 struct trace_array *tr = &global_trace;
1259 struct trace_array_cpu *data;
1260 struct bprint_entry *entry;
1261 unsigned long flags;
1264 int cpu, len = 0, size, pc;
1266 if (unlikely(tracing_selftest_running || tracing_disabled))
1269 /* Don't pollute graph traces with trace_vprintk internals */
1270 pause_graph_tracing();
1272 pc = preempt_count();
1273 resched = ftrace_preempt_disable();
1274 cpu = raw_smp_processor_id();
1275 data = tr->data[cpu];
1277 disable = atomic_inc_return(&data->disabled);
1278 if (unlikely(disable != 1))
1281 /* Lockdep uses trace_printk for lock tracing */
1282 local_irq_save(flags);
1283 __raw_spin_lock(&trace_buf_lock);
1284 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1286 if (len > TRACE_BUF_SIZE || len < 0)
1289 size = sizeof(*entry) + sizeof(u32) * len;
1290 event = trace_buffer_lock_reserve(tr, TRACE_BPRINT, size, flags, pc);
1293 entry = ring_buffer_event_data(event);
1297 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1298 if (!filter_check_discard(call, entry, tr->buffer, event))
1299 ring_buffer_unlock_commit(tr->buffer, event);
1302 __raw_spin_unlock(&trace_buf_lock);
1303 local_irq_restore(flags);
1306 atomic_dec_return(&data->disabled);
1307 ftrace_preempt_enable(resched);
1308 unpause_graph_tracing();
1312 EXPORT_SYMBOL_GPL(trace_vbprintk);
1314 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1316 static raw_spinlock_t trace_buf_lock = __RAW_SPIN_LOCK_UNLOCKED;
1317 static char trace_buf[TRACE_BUF_SIZE];
1319 struct ftrace_event_call *call = &event_print;
1320 struct ring_buffer_event *event;
1321 struct trace_array *tr = &global_trace;
1322 struct trace_array_cpu *data;
1323 int cpu, len = 0, size, pc;
1324 struct print_entry *entry;
1325 unsigned long irq_flags;
1328 if (tracing_disabled || tracing_selftest_running)
1331 pc = preempt_count();
1332 preempt_disable_notrace();
1333 cpu = raw_smp_processor_id();
1334 data = tr->data[cpu];
1336 disable = atomic_inc_return(&data->disabled);
1337 if (unlikely(disable != 1))
1340 pause_graph_tracing();
1341 raw_local_irq_save(irq_flags);
1342 __raw_spin_lock(&trace_buf_lock);
1343 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1345 len = min(len, TRACE_BUF_SIZE-1);
1348 size = sizeof(*entry) + len + 1;
1349 event = trace_buffer_lock_reserve(tr, TRACE_PRINT, size, irq_flags, pc);
1352 entry = ring_buffer_event_data(event);
1355 memcpy(&entry->buf, trace_buf, len);
1356 entry->buf[len] = 0;
1357 if (!filter_check_discard(call, entry, tr->buffer, event))
1358 ring_buffer_unlock_commit(tr->buffer, event);
1361 __raw_spin_unlock(&trace_buf_lock);
1362 raw_local_irq_restore(irq_flags);
1363 unpause_graph_tracing();
1365 atomic_dec_return(&data->disabled);
1366 preempt_enable_notrace();
1370 EXPORT_SYMBOL_GPL(trace_vprintk);
1372 enum trace_file_type {
1373 TRACE_FILE_LAT_FMT = 1,
1374 TRACE_FILE_ANNOTATE = 2,
1377 static void trace_iterator_increment(struct trace_iterator *iter)
1379 /* Don't allow ftrace to trace into the ring buffers */
1380 ftrace_disable_cpu();
1383 if (iter->buffer_iter[iter->cpu])
1384 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1386 ftrace_enable_cpu();
1389 static struct trace_entry *
1390 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1392 struct ring_buffer_event *event;
1393 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1395 /* Don't allow ftrace to trace into the ring buffers */
1396 ftrace_disable_cpu();
1399 event = ring_buffer_iter_peek(buf_iter, ts);
1401 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1403 ftrace_enable_cpu();
1405 return event ? ring_buffer_event_data(event) : NULL;
1408 static struct trace_entry *
1409 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1411 struct ring_buffer *buffer = iter->tr->buffer;
1412 struct trace_entry *ent, *next = NULL;
1413 int cpu_file = iter->cpu_file;
1414 u64 next_ts = 0, ts;
1419 * If we are in a per_cpu trace file, don't bother by iterating over
1420 * all cpu and peek directly.
1422 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1423 if (ring_buffer_empty_cpu(buffer, cpu_file))
1425 ent = peek_next_entry(iter, cpu_file, ent_ts);
1427 *ent_cpu = cpu_file;
1432 for_each_tracing_cpu(cpu) {
1434 if (ring_buffer_empty_cpu(buffer, cpu))
1437 ent = peek_next_entry(iter, cpu, &ts);
1440 * Pick the entry with the smallest timestamp:
1442 if (ent && (!next || ts < next_ts)) {
1450 *ent_cpu = next_cpu;
1458 /* Find the next real entry, without updating the iterator itself */
1459 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1460 int *ent_cpu, u64 *ent_ts)
1462 return __find_next_entry(iter, ent_cpu, ent_ts);
1465 /* Find the next real entry, and increment the iterator to the next entry */
1466 static void *find_next_entry_inc(struct trace_iterator *iter)
1468 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1471 trace_iterator_increment(iter);
1473 return iter->ent ? iter : NULL;
1476 static void trace_consume(struct trace_iterator *iter)
1478 /* Don't allow ftrace to trace into the ring buffers */
1479 ftrace_disable_cpu();
1480 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1481 ftrace_enable_cpu();
1484 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1486 struct trace_iterator *iter = m->private;
1492 /* can't go backwards */
1497 ent = find_next_entry_inc(iter);
1501 while (ent && iter->idx < i)
1502 ent = find_next_entry_inc(iter);
1510 * No necessary locking here. The worst thing which can
1511 * happen is loosing events consumed at the same time
1512 * by a trace_pipe reader.
1513 * Other than that, we don't risk to crash the ring buffer
1514 * because it serializes the readers.
1516 * The current tracer is copied to avoid a global locking
1519 static void *s_start(struct seq_file *m, loff_t *pos)
1521 struct trace_iterator *iter = m->private;
1522 static struct tracer *old_tracer;
1523 int cpu_file = iter->cpu_file;
1528 /* copy the tracer to avoid using a global lock all around */
1529 mutex_lock(&trace_types_lock);
1530 if (unlikely(old_tracer != current_trace && current_trace)) {
1531 old_tracer = current_trace;
1532 *iter->trace = *current_trace;
1534 mutex_unlock(&trace_types_lock);
1536 atomic_inc(&trace_record_cmdline_disabled);
1538 if (*pos != iter->pos) {
1543 ftrace_disable_cpu();
1545 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1546 for_each_tracing_cpu(cpu)
1547 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1549 ring_buffer_iter_reset(iter->buffer_iter[cpu_file]);
1552 ftrace_enable_cpu();
1554 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1559 p = s_next(m, p, &l);
1565 static void s_stop(struct seq_file *m, void *p)
1567 atomic_dec(&trace_record_cmdline_disabled);
1570 static void print_lat_help_header(struct seq_file *m)
1572 seq_puts(m, "# _------=> CPU# \n");
1573 seq_puts(m, "# / _-----=> irqs-off \n");
1574 seq_puts(m, "# | / _----=> need-resched \n");
1575 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1576 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1577 seq_puts(m, "# |||| / \n");
1578 seq_puts(m, "# ||||| delay \n");
1579 seq_puts(m, "# cmd pid ||||| time | caller \n");
1580 seq_puts(m, "# \\ / ||||| \\ | / \n");
1583 static void print_func_help_header(struct seq_file *m)
1585 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1586 seq_puts(m, "# | | | | |\n");
1591 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1593 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1594 struct trace_array *tr = iter->tr;
1595 struct trace_array_cpu *data = tr->data[tr->cpu];
1596 struct tracer *type = current_trace;
1597 unsigned long total;
1598 unsigned long entries;
1599 const char *name = "preemption";
1604 entries = ring_buffer_entries(iter->tr->buffer);
1606 ring_buffer_overruns(iter->tr->buffer);
1608 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1610 seq_puts(m, "# -----------------------------------"
1611 "---------------------------------\n");
1612 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1613 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1614 nsecs_to_usecs(data->saved_latency),
1618 #if defined(CONFIG_PREEMPT_NONE)
1620 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1622 #elif defined(CONFIG_PREEMPT)
1627 /* These are reserved for later use */
1630 seq_printf(m, " #P:%d)\n", num_online_cpus());
1634 seq_puts(m, "# -----------------\n");
1635 seq_printf(m, "# | task: %.16s-%d "
1636 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1637 data->comm, data->pid, data->uid, data->nice,
1638 data->policy, data->rt_priority);
1639 seq_puts(m, "# -----------------\n");
1641 if (data->critical_start) {
1642 seq_puts(m, "# => started at: ");
1643 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1644 trace_print_seq(m, &iter->seq);
1645 seq_puts(m, "\n# => ended at: ");
1646 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1647 trace_print_seq(m, &iter->seq);
1654 static void test_cpu_buff_start(struct trace_iterator *iter)
1656 struct trace_seq *s = &iter->seq;
1658 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1661 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1664 if (cpumask_test_cpu(iter->cpu, iter->started))
1667 cpumask_set_cpu(iter->cpu, iter->started);
1669 /* Don't print started cpu buffer for the first entry of the trace */
1671 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1675 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1677 struct trace_seq *s = &iter->seq;
1678 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1679 struct trace_entry *entry;
1680 struct trace_event *event;
1684 test_cpu_buff_start(iter);
1686 event = ftrace_find_event(entry->type);
1688 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1689 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1690 if (!trace_print_lat_context(iter))
1693 if (!trace_print_context(iter))
1699 return event->trace(iter, sym_flags);
1701 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1704 return TRACE_TYPE_HANDLED;
1706 return TRACE_TYPE_PARTIAL_LINE;
1709 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1711 struct trace_seq *s = &iter->seq;
1712 struct trace_entry *entry;
1713 struct trace_event *event;
1717 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1718 if (!trace_seq_printf(s, "%d %d %llu ",
1719 entry->pid, iter->cpu, iter->ts))
1723 event = ftrace_find_event(entry->type);
1725 return event->raw(iter, 0);
1727 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1730 return TRACE_TYPE_HANDLED;
1732 return TRACE_TYPE_PARTIAL_LINE;
1735 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1737 struct trace_seq *s = &iter->seq;
1738 unsigned char newline = '\n';
1739 struct trace_entry *entry;
1740 struct trace_event *event;
1744 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1745 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1746 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1747 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1750 event = ftrace_find_event(entry->type);
1752 enum print_line_t ret = event->hex(iter, 0);
1753 if (ret != TRACE_TYPE_HANDLED)
1757 SEQ_PUT_FIELD_RET(s, newline);
1759 return TRACE_TYPE_HANDLED;
1762 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1764 struct trace_seq *s = &iter->seq;
1765 struct trace_entry *entry;
1766 struct trace_event *event;
1770 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1771 SEQ_PUT_FIELD_RET(s, entry->pid);
1772 SEQ_PUT_FIELD_RET(s, iter->cpu);
1773 SEQ_PUT_FIELD_RET(s, iter->ts);
1776 event = ftrace_find_event(entry->type);
1777 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1780 static int trace_empty(struct trace_iterator *iter)
1784 /* If we are looking at one CPU buffer, only check that one */
1785 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1786 cpu = iter->cpu_file;
1787 if (iter->buffer_iter[cpu]) {
1788 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1791 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1797 for_each_tracing_cpu(cpu) {
1798 if (iter->buffer_iter[cpu]) {
1799 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1802 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1810 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1812 enum print_line_t ret;
1814 if (iter->trace && iter->trace->print_line) {
1815 ret = iter->trace->print_line(iter);
1816 if (ret != TRACE_TYPE_UNHANDLED)
1820 if (iter->ent->type == TRACE_BPRINT &&
1821 trace_flags & TRACE_ITER_PRINTK &&
1822 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1823 return trace_print_bprintk_msg_only(iter);
1825 if (iter->ent->type == TRACE_PRINT &&
1826 trace_flags & TRACE_ITER_PRINTK &&
1827 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1828 return trace_print_printk_msg_only(iter);
1830 if (trace_flags & TRACE_ITER_BIN)
1831 return print_bin_fmt(iter);
1833 if (trace_flags & TRACE_ITER_HEX)
1834 return print_hex_fmt(iter);
1836 if (trace_flags & TRACE_ITER_RAW)
1837 return print_raw_fmt(iter);
1839 return print_trace_fmt(iter);
1842 static int s_show(struct seq_file *m, void *v)
1844 struct trace_iterator *iter = v;
1846 if (iter->ent == NULL) {
1848 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1851 if (iter->trace && iter->trace->print_header)
1852 iter->trace->print_header(m);
1853 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1854 /* print nothing if the buffers are empty */
1855 if (trace_empty(iter))
1857 print_trace_header(m, iter);
1858 if (!(trace_flags & TRACE_ITER_VERBOSE))
1859 print_lat_help_header(m);
1861 if (!(trace_flags & TRACE_ITER_VERBOSE))
1862 print_func_help_header(m);
1865 print_trace_line(iter);
1866 trace_print_seq(m, &iter->seq);
1872 static struct seq_operations tracer_seq_ops = {
1879 static struct trace_iterator *
1880 __tracing_open(struct inode *inode, struct file *file)
1882 long cpu_file = (long) inode->i_private;
1883 void *fail_ret = ERR_PTR(-ENOMEM);
1884 struct trace_iterator *iter;
1888 if (tracing_disabled)
1889 return ERR_PTR(-ENODEV);
1891 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1893 return ERR_PTR(-ENOMEM);
1896 * We make a copy of the current tracer to avoid concurrent
1897 * changes on it while we are reading.
1899 mutex_lock(&trace_types_lock);
1900 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
1905 *iter->trace = *current_trace;
1907 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL))
1910 cpumask_clear(iter->started);
1912 if (current_trace && current_trace->print_max)
1915 iter->tr = &global_trace;
1917 mutex_init(&iter->mutex);
1918 iter->cpu_file = cpu_file;
1920 /* Notify the tracer early; before we stop tracing. */
1921 if (iter->trace && iter->trace->open)
1922 iter->trace->open(iter);
1924 /* Annotate start of buffers if we had overruns */
1925 if (ring_buffer_overruns(iter->tr->buffer))
1926 iter->iter_flags |= TRACE_FILE_ANNOTATE;
1928 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
1929 for_each_tracing_cpu(cpu) {
1931 iter->buffer_iter[cpu] =
1932 ring_buffer_read_start(iter->tr->buffer, cpu);
1935 cpu = iter->cpu_file;
1936 iter->buffer_iter[cpu] =
1937 ring_buffer_read_start(iter->tr->buffer, cpu);
1940 /* TODO stop tracer */
1941 ret = seq_open(file, &tracer_seq_ops);
1943 fail_ret = ERR_PTR(ret);
1947 m = file->private_data;
1950 /* stop the trace while dumping */
1953 mutex_unlock(&trace_types_lock);
1958 for_each_tracing_cpu(cpu) {
1959 if (iter->buffer_iter[cpu])
1960 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1962 free_cpumask_var(iter->started);
1964 mutex_unlock(&trace_types_lock);
1971 int tracing_open_generic(struct inode *inode, struct file *filp)
1973 if (tracing_disabled)
1976 filp->private_data = inode->i_private;
1980 static int tracing_release(struct inode *inode, struct file *file)
1982 struct seq_file *m = (struct seq_file *)file->private_data;
1983 struct trace_iterator *iter;
1986 if (!(file->f_mode & FMODE_READ))
1991 mutex_lock(&trace_types_lock);
1992 for_each_tracing_cpu(cpu) {
1993 if (iter->buffer_iter[cpu])
1994 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1997 if (iter->trace && iter->trace->close)
1998 iter->trace->close(iter);
2000 /* reenable tracing if it was previously enabled */
2002 mutex_unlock(&trace_types_lock);
2004 seq_release(inode, file);
2005 mutex_destroy(&iter->mutex);
2006 free_cpumask_var(iter->started);
2012 static int tracing_open(struct inode *inode, struct file *file)
2014 struct trace_iterator *iter;
2017 /* If this file was open for write, then erase contents */
2018 if ((file->f_mode & FMODE_WRITE) &&
2019 !(file->f_flags & O_APPEND)) {
2020 long cpu = (long) inode->i_private;
2022 if (cpu == TRACE_PIPE_ALL_CPU)
2023 tracing_reset_online_cpus(&global_trace);
2025 tracing_reset(&global_trace, cpu);
2028 if (file->f_mode & FMODE_READ) {
2029 iter = __tracing_open(inode, file);
2031 ret = PTR_ERR(iter);
2032 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2033 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2039 t_next(struct seq_file *m, void *v, loff_t *pos)
2041 struct tracer *t = m->private;
2053 static void *t_start(struct seq_file *m, loff_t *pos)
2055 struct tracer *t = m->private;
2058 mutex_lock(&trace_types_lock);
2059 for (; t && l < *pos; t = t_next(m, t, &l))
2065 static void t_stop(struct seq_file *m, void *p)
2067 mutex_unlock(&trace_types_lock);
2070 static int t_show(struct seq_file *m, void *v)
2072 struct tracer *t = v;
2077 seq_printf(m, "%s", t->name);
2086 static struct seq_operations show_traces_seq_ops = {
2093 static int show_traces_open(struct inode *inode, struct file *file)
2097 if (tracing_disabled)
2100 ret = seq_open(file, &show_traces_seq_ops);
2102 struct seq_file *m = file->private_data;
2103 m->private = trace_types;
2110 tracing_write_stub(struct file *filp, const char __user *ubuf,
2111 size_t count, loff_t *ppos)
2116 static const struct file_operations tracing_fops = {
2117 .open = tracing_open,
2119 .write = tracing_write_stub,
2120 .llseek = seq_lseek,
2121 .release = tracing_release,
2124 static const struct file_operations show_traces_fops = {
2125 .open = show_traces_open,
2127 .release = seq_release,
2131 * Only trace on a CPU if the bitmask is set:
2133 static cpumask_var_t tracing_cpumask;
2136 * The tracer itself will not take this lock, but still we want
2137 * to provide a consistent cpumask to user-space:
2139 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2142 * Temporary storage for the character representation of the
2143 * CPU bitmask (and one more byte for the newline):
2145 static char mask_str[NR_CPUS + 1];
2148 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2149 size_t count, loff_t *ppos)
2153 mutex_lock(&tracing_cpumask_update_lock);
2155 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2156 if (count - len < 2) {
2160 len += sprintf(mask_str + len, "\n");
2161 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2164 mutex_unlock(&tracing_cpumask_update_lock);
2170 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2171 size_t count, loff_t *ppos)
2174 cpumask_var_t tracing_cpumask_new;
2176 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2179 mutex_lock(&tracing_cpumask_update_lock);
2180 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2184 local_irq_disable();
2185 __raw_spin_lock(&ftrace_max_lock);
2186 for_each_tracing_cpu(cpu) {
2188 * Increase/decrease the disabled counter if we are
2189 * about to flip a bit in the cpumask:
2191 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2192 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2193 atomic_inc(&global_trace.data[cpu]->disabled);
2195 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2196 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2197 atomic_dec(&global_trace.data[cpu]->disabled);
2200 __raw_spin_unlock(&ftrace_max_lock);
2203 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2205 mutex_unlock(&tracing_cpumask_update_lock);
2206 free_cpumask_var(tracing_cpumask_new);
2211 mutex_unlock(&tracing_cpumask_update_lock);
2212 free_cpumask_var(tracing_cpumask);
2217 static const struct file_operations tracing_cpumask_fops = {
2218 .open = tracing_open_generic,
2219 .read = tracing_cpumask_read,
2220 .write = tracing_cpumask_write,
2224 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2225 size_t cnt, loff_t *ppos)
2227 struct tracer_opt *trace_opts;
2235 /* calculate max size */
2236 for (i = 0; trace_options[i]; i++) {
2237 len += strlen(trace_options[i]);
2238 len += 3; /* "no" and newline */
2241 mutex_lock(&trace_types_lock);
2242 tracer_flags = current_trace->flags->val;
2243 trace_opts = current_trace->flags->opts;
2246 * Increase the size with names of options specific
2247 * of the current tracer.
2249 for (i = 0; trace_opts[i].name; i++) {
2250 len += strlen(trace_opts[i].name);
2251 len += 3; /* "no" and newline */
2254 /* +2 for \n and \0 */
2255 buf = kmalloc(len + 2, GFP_KERNEL);
2257 mutex_unlock(&trace_types_lock);
2261 for (i = 0; trace_options[i]; i++) {
2262 if (trace_flags & (1 << i))
2263 r += sprintf(buf + r, "%s\n", trace_options[i]);
2265 r += sprintf(buf + r, "no%s\n", trace_options[i]);
2268 for (i = 0; trace_opts[i].name; i++) {
2269 if (tracer_flags & trace_opts[i].bit)
2270 r += sprintf(buf + r, "%s\n",
2271 trace_opts[i].name);
2273 r += sprintf(buf + r, "no%s\n",
2274 trace_opts[i].name);
2276 mutex_unlock(&trace_types_lock);
2278 WARN_ON(r >= len + 2);
2280 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2286 /* Try to assign a tracer specific option */
2287 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2289 struct tracer_flags *trace_flags = trace->flags;
2290 struct tracer_opt *opts = NULL;
2294 for (i = 0; trace_flags->opts[i].name; i++) {
2295 opts = &trace_flags->opts[i];
2296 len = strlen(opts->name);
2298 if (strncmp(cmp, opts->name, len) == 0) {
2299 ret = trace->set_flag(trace_flags->val,
2305 if (!trace_flags->opts[i].name)
2308 /* Refused to handle */
2313 trace_flags->val &= ~opts->bit;
2315 trace_flags->val |= opts->bit;
2320 static void set_tracer_flags(unsigned int mask, int enabled)
2322 /* do nothing if flag is already set */
2323 if (!!(trace_flags & mask) == !!enabled)
2327 trace_flags |= mask;
2329 trace_flags &= ~mask;
2331 if (mask == TRACE_ITER_GLOBAL_CLK) {
2335 func = trace_clock_global;
2337 func = trace_clock_local;
2339 mutex_lock(&trace_types_lock);
2340 ring_buffer_set_clock(global_trace.buffer, func);
2343 ring_buffer_set_clock(max_tr.buffer, func);
2344 mutex_unlock(&trace_types_lock);
2349 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2350 size_t cnt, loff_t *ppos)
2358 if (cnt >= sizeof(buf))
2361 if (copy_from_user(&buf, ubuf, cnt))
2366 if (strncmp(buf, "no", 2) == 0) {
2371 for (i = 0; trace_options[i]; i++) {
2372 int len = strlen(trace_options[i]);
2374 if (strncmp(cmp, trace_options[i], len) == 0) {
2375 set_tracer_flags(1 << i, !neg);
2380 /* If no option could be set, test the specific tracer options */
2381 if (!trace_options[i]) {
2382 mutex_lock(&trace_types_lock);
2383 ret = set_tracer_option(current_trace, cmp, neg);
2384 mutex_unlock(&trace_types_lock);
2394 static const struct file_operations tracing_iter_fops = {
2395 .open = tracing_open_generic,
2396 .read = tracing_trace_options_read,
2397 .write = tracing_trace_options_write,
2400 static const char readme_msg[] =
2401 "tracing mini-HOWTO:\n\n"
2403 "# mount -t debugfs nodev /debug\n\n"
2404 "# cat /debug/tracing/available_tracers\n"
2405 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2406 "# cat /debug/tracing/current_tracer\n"
2408 "# echo sched_switch > /debug/tracing/current_tracer\n"
2409 "# cat /debug/tracing/current_tracer\n"
2411 "# cat /debug/tracing/trace_options\n"
2412 "noprint-parent nosym-offset nosym-addr noverbose\n"
2413 "# echo print-parent > /debug/tracing/trace_options\n"
2414 "# echo 1 > /debug/tracing/tracing_enabled\n"
2415 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2416 "echo 0 > /debug/tracing/tracing_enabled\n"
2420 tracing_readme_read(struct file *filp, char __user *ubuf,
2421 size_t cnt, loff_t *ppos)
2423 return simple_read_from_buffer(ubuf, cnt, ppos,
2424 readme_msg, strlen(readme_msg));
2427 static const struct file_operations tracing_readme_fops = {
2428 .open = tracing_open_generic,
2429 .read = tracing_readme_read,
2433 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2434 size_t cnt, loff_t *ppos)
2443 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2447 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2455 for (i = 0; i < SAVED_CMDLINES; i++) {
2458 pid = map_cmdline_to_pid[i];
2459 if (pid == -1 || pid == NO_CMDLINE_MAP)
2462 trace_find_cmdline(pid, buf_comm);
2463 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2468 len = simple_read_from_buffer(ubuf, cnt, ppos,
2477 static const struct file_operations tracing_saved_cmdlines_fops = {
2478 .open = tracing_open_generic,
2479 .read = tracing_saved_cmdlines_read,
2483 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2484 size_t cnt, loff_t *ppos)
2489 r = sprintf(buf, "%u\n", tracer_enabled);
2490 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2494 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2495 size_t cnt, loff_t *ppos)
2497 struct trace_array *tr = filp->private_data;
2502 if (cnt >= sizeof(buf))
2505 if (copy_from_user(&buf, ubuf, cnt))
2510 ret = strict_strtoul(buf, 10, &val);
2516 mutex_lock(&trace_types_lock);
2517 if (tracer_enabled ^ val) {
2520 if (current_trace->start)
2521 current_trace->start(tr);
2526 if (current_trace->stop)
2527 current_trace->stop(tr);
2530 mutex_unlock(&trace_types_lock);
2538 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2539 size_t cnt, loff_t *ppos)
2541 char buf[max_tracer_type_len+2];
2544 mutex_lock(&trace_types_lock);
2546 r = sprintf(buf, "%s\n", current_trace->name);
2548 r = sprintf(buf, "\n");
2549 mutex_unlock(&trace_types_lock);
2551 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2554 int tracer_init(struct tracer *t, struct trace_array *tr)
2556 tracing_reset_online_cpus(tr);
2560 static int tracing_resize_ring_buffer(unsigned long size)
2565 * If kernel or user changes the size of the ring buffer
2566 * we use the size that was given, and we can forget about
2567 * expanding it later.
2569 ring_buffer_expanded = 1;
2571 ret = ring_buffer_resize(global_trace.buffer, size);
2575 ret = ring_buffer_resize(max_tr.buffer, size);
2579 r = ring_buffer_resize(global_trace.buffer,
2580 global_trace.entries);
2583 * AARGH! We are left with different
2584 * size max buffer!!!!
2585 * The max buffer is our "snapshot" buffer.
2586 * When a tracer needs a snapshot (one of the
2587 * latency tracers), it swaps the max buffer
2588 * with the saved snap shot. We succeeded to
2589 * update the size of the main buffer, but failed to
2590 * update the size of the max buffer. But when we tried
2591 * to reset the main buffer to the original size, we
2592 * failed there too. This is very unlikely to
2593 * happen, but if it does, warn and kill all
2597 tracing_disabled = 1;
2602 global_trace.entries = size;
2608 * tracing_update_buffers - used by tracing facility to expand ring buffers
2610 * To save on memory when the tracing is never used on a system with it
2611 * configured in. The ring buffers are set to a minimum size. But once
2612 * a user starts to use the tracing facility, then they need to grow
2613 * to their default size.
2615 * This function is to be called when a tracer is about to be used.
2617 int tracing_update_buffers(void)
2621 mutex_lock(&trace_types_lock);
2622 if (!ring_buffer_expanded)
2623 ret = tracing_resize_ring_buffer(trace_buf_size);
2624 mutex_unlock(&trace_types_lock);
2629 struct trace_option_dentry;
2631 static struct trace_option_dentry *
2632 create_trace_option_files(struct tracer *tracer);
2635 destroy_trace_option_files(struct trace_option_dentry *topts);
2637 static int tracing_set_tracer(const char *buf)
2639 static struct trace_option_dentry *topts;
2640 struct trace_array *tr = &global_trace;
2644 mutex_lock(&trace_types_lock);
2646 if (!ring_buffer_expanded) {
2647 ret = tracing_resize_ring_buffer(trace_buf_size);
2653 for (t = trace_types; t; t = t->next) {
2654 if (strcmp(t->name, buf) == 0)
2661 if (t == current_trace)
2664 trace_branch_disable();
2665 if (current_trace && current_trace->reset)
2666 current_trace->reset(tr);
2668 destroy_trace_option_files(topts);
2672 topts = create_trace_option_files(current_trace);
2675 ret = tracer_init(t, tr);
2680 trace_branch_enable(tr);
2682 mutex_unlock(&trace_types_lock);
2688 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2689 size_t cnt, loff_t *ppos)
2691 char buf[max_tracer_type_len+1];
2698 if (cnt > max_tracer_type_len)
2699 cnt = max_tracer_type_len;
2701 if (copy_from_user(&buf, ubuf, cnt))
2706 /* strip ending whitespace. */
2707 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2710 err = tracing_set_tracer(buf);
2720 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2721 size_t cnt, loff_t *ppos)
2723 unsigned long *ptr = filp->private_data;
2727 r = snprintf(buf, sizeof(buf), "%ld\n",
2728 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2729 if (r > sizeof(buf))
2731 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2735 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2736 size_t cnt, loff_t *ppos)
2738 unsigned long *ptr = filp->private_data;
2743 if (cnt >= sizeof(buf))
2746 if (copy_from_user(&buf, ubuf, cnt))
2751 ret = strict_strtoul(buf, 10, &val);
2760 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2762 long cpu_file = (long) inode->i_private;
2763 struct trace_iterator *iter;
2766 if (tracing_disabled)
2769 mutex_lock(&trace_types_lock);
2771 /* We only allow one reader per cpu */
2772 if (cpu_file == TRACE_PIPE_ALL_CPU) {
2773 if (!cpumask_empty(tracing_reader_cpumask)) {
2777 cpumask_setall(tracing_reader_cpumask);
2779 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2780 cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2787 /* create a buffer to store the information to pass to userspace */
2788 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2795 * We make a copy of the current tracer to avoid concurrent
2796 * changes on it while we are reading.
2798 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2804 *iter->trace = *current_trace;
2806 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2811 /* trace pipe does not show start of buffer */
2812 cpumask_setall(iter->started);
2814 iter->cpu_file = cpu_file;
2815 iter->tr = &global_trace;
2816 mutex_init(&iter->mutex);
2817 filp->private_data = iter;
2819 if (iter->trace->pipe_open)
2820 iter->trace->pipe_open(iter);
2823 mutex_unlock(&trace_types_lock);
2829 mutex_unlock(&trace_types_lock);
2833 static int tracing_release_pipe(struct inode *inode, struct file *file)
2835 struct trace_iterator *iter = file->private_data;
2837 mutex_lock(&trace_types_lock);
2839 if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2840 cpumask_clear(tracing_reader_cpumask);
2842 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2844 mutex_unlock(&trace_types_lock);
2846 free_cpumask_var(iter->started);
2847 mutex_destroy(&iter->mutex);
2855 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2857 struct trace_iterator *iter = filp->private_data;
2859 if (trace_flags & TRACE_ITER_BLOCK) {
2861 * Always select as readable when in blocking mode
2863 return POLLIN | POLLRDNORM;
2865 if (!trace_empty(iter))
2866 return POLLIN | POLLRDNORM;
2867 poll_wait(filp, &trace_wait, poll_table);
2868 if (!trace_empty(iter))
2869 return POLLIN | POLLRDNORM;
2876 void default_wait_pipe(struct trace_iterator *iter)
2880 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2882 if (trace_empty(iter))
2885 finish_wait(&trace_wait, &wait);
2889 * This is a make-shift waitqueue.
2890 * A tracer might use this callback on some rare cases:
2892 * 1) the current tracer might hold the runqueue lock when it wakes up
2893 * a reader, hence a deadlock (sched, function, and function graph tracers)
2894 * 2) the function tracers, trace all functions, we don't want
2895 * the overhead of calling wake_up and friends
2896 * (and tracing them too)
2898 * Anyway, this is really very primitive wakeup.
2900 void poll_wait_pipe(struct trace_iterator *iter)
2902 set_current_state(TASK_INTERRUPTIBLE);
2903 /* sleep for 100 msecs, and try again. */
2904 schedule_timeout(HZ / 10);
2907 /* Must be called with trace_types_lock mutex held. */
2908 static int tracing_wait_pipe(struct file *filp)
2910 struct trace_iterator *iter = filp->private_data;
2912 while (trace_empty(iter)) {
2914 if ((filp->f_flags & O_NONBLOCK)) {
2918 mutex_unlock(&iter->mutex);
2920 iter->trace->wait_pipe(iter);
2922 mutex_lock(&iter->mutex);
2924 if (signal_pending(current))
2928 * We block until we read something and tracing is disabled.
2929 * We still block if tracing is disabled, but we have never
2930 * read anything. This allows a user to cat this file, and
2931 * then enable tracing. But after we have read something,
2932 * we give an EOF when tracing is again disabled.
2934 * iter->pos will be 0 if we haven't read anything.
2936 if (!tracer_enabled && iter->pos)
2947 tracing_read_pipe(struct file *filp, char __user *ubuf,
2948 size_t cnt, loff_t *ppos)
2950 struct trace_iterator *iter = filp->private_data;
2951 static struct tracer *old_tracer;
2954 /* return any leftover data */
2955 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2959 trace_seq_init(&iter->seq);
2961 /* copy the tracer to avoid using a global lock all around */
2962 mutex_lock(&trace_types_lock);
2963 if (unlikely(old_tracer != current_trace && current_trace)) {
2964 old_tracer = current_trace;
2965 *iter->trace = *current_trace;
2967 mutex_unlock(&trace_types_lock);
2970 * Avoid more than one consumer on a single file descriptor
2971 * This is just a matter of traces coherency, the ring buffer itself
2974 mutex_lock(&iter->mutex);
2975 if (iter->trace->read) {
2976 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2982 sret = tracing_wait_pipe(filp);
2986 /* stop when tracing is finished */
2987 if (trace_empty(iter)) {
2992 if (cnt >= PAGE_SIZE)
2993 cnt = PAGE_SIZE - 1;
2995 /* reset all but tr, trace, and overruns */
2996 memset(&iter->seq, 0,
2997 sizeof(struct trace_iterator) -
2998 offsetof(struct trace_iterator, seq));
3001 while (find_next_entry_inc(iter) != NULL) {
3002 enum print_line_t ret;
3003 int len = iter->seq.len;
3005 ret = print_trace_line(iter);
3006 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3007 /* don't print partial lines */
3008 iter->seq.len = len;
3011 if (ret != TRACE_TYPE_NO_CONSUME)
3012 trace_consume(iter);
3014 if (iter->seq.len >= cnt)
3018 /* Now copy what we have to the user */
3019 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3020 if (iter->seq.readpos >= iter->seq.len)
3021 trace_seq_init(&iter->seq);
3024 * If there was nothing to send to user, inspite of consuming trace
3025 * entries, go back to wait for more entries.
3031 mutex_unlock(&iter->mutex);
3036 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3037 struct pipe_buffer *buf)
3039 __free_page(buf->page);
3042 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3045 __free_page(spd->pages[idx]);
3048 static struct pipe_buf_operations tracing_pipe_buf_ops = {
3050 .map = generic_pipe_buf_map,
3051 .unmap = generic_pipe_buf_unmap,
3052 .confirm = generic_pipe_buf_confirm,
3053 .release = tracing_pipe_buf_release,
3054 .steal = generic_pipe_buf_steal,
3055 .get = generic_pipe_buf_get,
3059 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3064 /* Seq buffer is page-sized, exactly what we need. */
3066 count = iter->seq.len;
3067 ret = print_trace_line(iter);
3068 count = iter->seq.len - count;
3071 iter->seq.len -= count;
3074 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3075 iter->seq.len -= count;
3079 trace_consume(iter);
3081 if (!find_next_entry_inc(iter)) {
3091 static ssize_t tracing_splice_read_pipe(struct file *filp,
3093 struct pipe_inode_info *pipe,
3097 struct page *pages[PIPE_BUFFERS];
3098 struct partial_page partial[PIPE_BUFFERS];
3099 struct trace_iterator *iter = filp->private_data;
3100 struct splice_pipe_desc spd = {
3103 .nr_pages = 0, /* This gets updated below. */
3105 .ops = &tracing_pipe_buf_ops,
3106 .spd_release = tracing_spd_release_pipe,
3108 static struct tracer *old_tracer;
3113 /* copy the tracer to avoid using a global lock all around */
3114 mutex_lock(&trace_types_lock);
3115 if (unlikely(old_tracer != current_trace && current_trace)) {
3116 old_tracer = current_trace;
3117 *iter->trace = *current_trace;
3119 mutex_unlock(&trace_types_lock);
3121 mutex_lock(&iter->mutex);
3123 if (iter->trace->splice_read) {
3124 ret = iter->trace->splice_read(iter, filp,
3125 ppos, pipe, len, flags);
3130 ret = tracing_wait_pipe(filp);
3134 if (!iter->ent && !find_next_entry_inc(iter)) {
3139 /* Fill as many pages as possible. */
3140 for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3141 pages[i] = alloc_page(GFP_KERNEL);
3145 rem = tracing_fill_pipe_page(rem, iter);
3147 /* Copy the data into the page, so we can start over. */
3148 ret = trace_seq_to_buffer(&iter->seq,
3149 page_address(pages[i]),
3152 __free_page(pages[i]);
3155 partial[i].offset = 0;
3156 partial[i].len = iter->seq.len;
3158 trace_seq_init(&iter->seq);
3161 mutex_unlock(&iter->mutex);
3165 return splice_to_pipe(pipe, &spd);
3168 mutex_unlock(&iter->mutex);
3174 tracing_entries_read(struct file *filp, char __user *ubuf,
3175 size_t cnt, loff_t *ppos)
3177 struct trace_array *tr = filp->private_data;
3181 mutex_lock(&trace_types_lock);
3182 if (!ring_buffer_expanded)
3183 r = sprintf(buf, "%lu (expanded: %lu)\n",
3185 trace_buf_size >> 10);
3187 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3188 mutex_unlock(&trace_types_lock);
3190 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3194 tracing_entries_write(struct file *filp, const char __user *ubuf,
3195 size_t cnt, loff_t *ppos)
3201 if (cnt >= sizeof(buf))
3204 if (copy_from_user(&buf, ubuf, cnt))
3209 ret = strict_strtoul(buf, 10, &val);
3213 /* must have at least 1 entry */
3217 mutex_lock(&trace_types_lock);
3221 /* disable all cpu buffers */
3222 for_each_tracing_cpu(cpu) {
3223 if (global_trace.data[cpu])
3224 atomic_inc(&global_trace.data[cpu]->disabled);
3225 if (max_tr.data[cpu])
3226 atomic_inc(&max_tr.data[cpu]->disabled);
3229 /* value is in KB */
3232 if (val != global_trace.entries) {
3233 ret = tracing_resize_ring_buffer(val);
3242 /* If check pages failed, return ENOMEM */
3243 if (tracing_disabled)
3246 for_each_tracing_cpu(cpu) {
3247 if (global_trace.data[cpu])
3248 atomic_dec(&global_trace.data[cpu]->disabled);
3249 if (max_tr.data[cpu])
3250 atomic_dec(&max_tr.data[cpu]->disabled);
3254 max_tr.entries = global_trace.entries;
3255 mutex_unlock(&trace_types_lock);
3260 static int mark_printk(const char *fmt, ...)
3264 va_start(args, fmt);
3265 ret = trace_vprintk(0, fmt, args);
3271 tracing_mark_write(struct file *filp, const char __user *ubuf,
3272 size_t cnt, loff_t *fpos)
3277 if (tracing_disabled)
3280 if (cnt > TRACE_BUF_SIZE)
3281 cnt = TRACE_BUF_SIZE;
3283 buf = kmalloc(cnt + 1, GFP_KERNEL);
3287 if (copy_from_user(buf, ubuf, cnt)) {
3292 /* Cut from the first nil or newline. */
3294 end = strchr(buf, '\n');
3298 cnt = mark_printk("%s\n", buf);
3305 static const struct file_operations tracing_max_lat_fops = {
3306 .open = tracing_open_generic,
3307 .read = tracing_max_lat_read,
3308 .write = tracing_max_lat_write,
3311 static const struct file_operations tracing_ctrl_fops = {
3312 .open = tracing_open_generic,
3313 .read = tracing_ctrl_read,
3314 .write = tracing_ctrl_write,
3317 static const struct file_operations set_tracer_fops = {
3318 .open = tracing_open_generic,
3319 .read = tracing_set_trace_read,
3320 .write = tracing_set_trace_write,
3323 static const struct file_operations tracing_pipe_fops = {
3324 .open = tracing_open_pipe,
3325 .poll = tracing_poll_pipe,
3326 .read = tracing_read_pipe,
3327 .splice_read = tracing_splice_read_pipe,
3328 .release = tracing_release_pipe,
3331 static const struct file_operations tracing_entries_fops = {
3332 .open = tracing_open_generic,
3333 .read = tracing_entries_read,
3334 .write = tracing_entries_write,
3337 static const struct file_operations tracing_mark_fops = {
3338 .open = tracing_open_generic,
3339 .write = tracing_mark_write,
3342 struct ftrace_buffer_info {
3343 struct trace_array *tr;
3349 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3351 int cpu = (int)(long)inode->i_private;
3352 struct ftrace_buffer_info *info;
3354 if (tracing_disabled)
3357 info = kzalloc(sizeof(*info), GFP_KERNEL);
3361 info->tr = &global_trace;
3364 /* Force reading ring buffer for first read */
3365 info->read = (unsigned int)-1;
3367 filp->private_data = info;
3369 return nonseekable_open(inode, filp);
3373 tracing_buffers_read(struct file *filp, char __user *ubuf,
3374 size_t count, loff_t *ppos)
3376 struct ftrace_buffer_info *info = filp->private_data;
3385 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3389 /* Do we have previous read data to read? */
3390 if (info->read < PAGE_SIZE)
3395 ret = ring_buffer_read_page(info->tr->buffer,
3402 pos = ring_buffer_page_len(info->spare);
3404 if (pos < PAGE_SIZE)
3405 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3408 size = PAGE_SIZE - info->read;
3412 ret = copy_to_user(ubuf, info->spare + info->read, size);
3423 static int tracing_buffers_release(struct inode *inode, struct file *file)
3425 struct ftrace_buffer_info *info = file->private_data;
3428 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3435 struct ring_buffer *buffer;
3440 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3441 struct pipe_buffer *buf)
3443 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3448 ring_buffer_free_read_page(ref->buffer, ref->page);
3453 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3454 struct pipe_buffer *buf)
3459 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3460 struct pipe_buffer *buf)
3462 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3467 /* Pipe buffer operations for a buffer. */
3468 static struct pipe_buf_operations buffer_pipe_buf_ops = {
3470 .map = generic_pipe_buf_map,
3471 .unmap = generic_pipe_buf_unmap,
3472 .confirm = generic_pipe_buf_confirm,
3473 .release = buffer_pipe_buf_release,
3474 .steal = buffer_pipe_buf_steal,
3475 .get = buffer_pipe_buf_get,
3479 * Callback from splice_to_pipe(), if we need to release some pages
3480 * at the end of the spd in case we error'ed out in filling the pipe.
3482 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3484 struct buffer_ref *ref =
3485 (struct buffer_ref *)spd->partial[i].private;
3490 ring_buffer_free_read_page(ref->buffer, ref->page);
3492 spd->partial[i].private = 0;
3496 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3497 struct pipe_inode_info *pipe, size_t len,
3500 struct ftrace_buffer_info *info = file->private_data;
3501 struct partial_page partial[PIPE_BUFFERS];
3502 struct page *pages[PIPE_BUFFERS];
3503 struct splice_pipe_desc spd = {
3507 .ops = &buffer_pipe_buf_ops,
3508 .spd_release = buffer_spd_release,
3510 struct buffer_ref *ref;
3511 int entries, size, i;
3514 if (*ppos & (PAGE_SIZE - 1)) {
3515 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3519 if (len & (PAGE_SIZE - 1)) {
3520 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3521 if (len < PAGE_SIZE)
3526 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3528 for (i = 0; i < PIPE_BUFFERS && len && entries; i++, len -= PAGE_SIZE) {
3532 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3537 ref->buffer = info->tr->buffer;
3538 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3544 r = ring_buffer_read_page(ref->buffer, &ref->page,
3547 ring_buffer_free_read_page(ref->buffer,
3554 * zero out any left over data, this is going to
3557 size = ring_buffer_page_len(ref->page);
3558 if (size < PAGE_SIZE)
3559 memset(ref->page + size, 0, PAGE_SIZE - size);
3561 page = virt_to_page(ref->page);
3563 spd.pages[i] = page;
3564 spd.partial[i].len = PAGE_SIZE;
3565 spd.partial[i].offset = 0;
3566 spd.partial[i].private = (unsigned long)ref;
3570 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3575 /* did we read anything? */
3576 if (!spd.nr_pages) {
3577 if (flags & SPLICE_F_NONBLOCK)
3585 ret = splice_to_pipe(pipe, &spd);
3590 static const struct file_operations tracing_buffers_fops = {
3591 .open = tracing_buffers_open,
3592 .read = tracing_buffers_read,
3593 .release = tracing_buffers_release,
3594 .splice_read = tracing_buffers_splice_read,
3595 .llseek = no_llseek,
3599 tracing_stats_read(struct file *filp, char __user *ubuf,
3600 size_t count, loff_t *ppos)
3602 unsigned long cpu = (unsigned long)filp->private_data;
3603 struct trace_array *tr = &global_trace;
3604 struct trace_seq *s;
3607 s = kmalloc(sizeof(*s), GFP_ATOMIC);
3613 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3614 trace_seq_printf(s, "entries: %ld\n", cnt);
3616 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3617 trace_seq_printf(s, "overrun: %ld\n", cnt);
3619 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3620 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3622 cnt = ring_buffer_nmi_dropped_cpu(tr->buffer, cpu);
3623 trace_seq_printf(s, "nmi dropped: %ld\n", cnt);
3625 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3632 static const struct file_operations tracing_stats_fops = {
3633 .open = tracing_open_generic,
3634 .read = tracing_stats_read,
3637 #ifdef CONFIG_DYNAMIC_FTRACE
3639 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3645 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3646 size_t cnt, loff_t *ppos)
3648 static char ftrace_dyn_info_buffer[1024];
3649 static DEFINE_MUTEX(dyn_info_mutex);
3650 unsigned long *p = filp->private_data;
3651 char *buf = ftrace_dyn_info_buffer;
3652 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3655 mutex_lock(&dyn_info_mutex);
3656 r = sprintf(buf, "%ld ", *p);
3658 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3661 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3663 mutex_unlock(&dyn_info_mutex);
3668 static const struct file_operations tracing_dyn_info_fops = {
3669 .open = tracing_open_generic,
3670 .read = tracing_read_dyn_info,
3674 static struct dentry *d_tracer;
3676 struct dentry *tracing_init_dentry(void)
3683 if (!debugfs_initialized())
3686 d_tracer = debugfs_create_dir("tracing", NULL);
3688 if (!d_tracer && !once) {
3690 pr_warning("Could not create debugfs directory 'tracing'\n");
3697 static struct dentry *d_percpu;
3699 struct dentry *tracing_dentry_percpu(void)
3702 struct dentry *d_tracer;
3707 d_tracer = tracing_init_dentry();
3712 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3714 if (!d_percpu && !once) {
3716 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3723 static void tracing_init_debugfs_percpu(long cpu)
3725 struct dentry *d_percpu = tracing_dentry_percpu();
3726 struct dentry *d_cpu;
3727 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3730 if (cpu > 999 || cpu < 0)
3733 sprintf(cpu_dir, "cpu%ld", cpu);
3734 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3736 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3740 /* per cpu trace_pipe */
3741 trace_create_file("trace_pipe", 0444, d_cpu,
3742 (void *) cpu, &tracing_pipe_fops);
3745 trace_create_file("trace", 0644, d_cpu,
3746 (void *) cpu, &tracing_fops);
3748 trace_create_file("trace_pipe_raw", 0444, d_cpu,
3749 (void *) cpu, &tracing_buffers_fops);
3751 trace_create_file("stats", 0444, d_cpu,
3752 (void *) cpu, &tracing_stats_fops);
3755 #ifdef CONFIG_FTRACE_SELFTEST
3756 /* Let selftest have access to static functions in this file */
3757 #include "trace_selftest.c"
3760 struct trace_option_dentry {
3761 struct tracer_opt *opt;
3762 struct tracer_flags *flags;
3763 struct dentry *entry;
3767 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3770 struct trace_option_dentry *topt = filp->private_data;
3773 if (topt->flags->val & topt->opt->bit)
3778 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3782 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3785 struct trace_option_dentry *topt = filp->private_data;
3790 if (cnt >= sizeof(buf))
3793 if (copy_from_user(&buf, ubuf, cnt))
3798 ret = strict_strtoul(buf, 10, &val);
3805 /* do nothing if already cleared */
3806 if (!(topt->flags->val & topt->opt->bit))
3809 mutex_lock(&trace_types_lock);
3810 if (current_trace->set_flag)
3811 ret = current_trace->set_flag(topt->flags->val,
3813 mutex_unlock(&trace_types_lock);
3816 topt->flags->val &= ~topt->opt->bit;
3819 /* do nothing if already set */
3820 if (topt->flags->val & topt->opt->bit)
3823 mutex_lock(&trace_types_lock);
3824 if (current_trace->set_flag)
3825 ret = current_trace->set_flag(topt->flags->val,
3827 mutex_unlock(&trace_types_lock);
3830 topt->flags->val |= topt->opt->bit;
3843 static const struct file_operations trace_options_fops = {
3844 .open = tracing_open_generic,
3845 .read = trace_options_read,
3846 .write = trace_options_write,
3850 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
3853 long index = (long)filp->private_data;
3856 if (trace_flags & (1 << index))
3861 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3865 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
3868 long index = (long)filp->private_data;
3873 if (cnt >= sizeof(buf))
3876 if (copy_from_user(&buf, ubuf, cnt))
3881 ret = strict_strtoul(buf, 10, &val);
3887 trace_flags &= ~(1 << index);
3890 trace_flags |= 1 << index;
3902 static const struct file_operations trace_options_core_fops = {
3903 .open = tracing_open_generic,
3904 .read = trace_options_core_read,
3905 .write = trace_options_core_write,
3908 struct dentry *trace_create_file(const char *name,
3910 struct dentry *parent,
3912 const struct file_operations *fops)
3916 ret = debugfs_create_file(name, mode, parent, data, fops);
3918 pr_warning("Could not create debugfs '%s' entry\n", name);
3924 static struct dentry *trace_options_init_dentry(void)
3926 struct dentry *d_tracer;
3927 static struct dentry *t_options;
3932 d_tracer = tracing_init_dentry();
3936 t_options = debugfs_create_dir("options", d_tracer);
3938 pr_warning("Could not create debugfs directory 'options'\n");
3946 create_trace_option_file(struct trace_option_dentry *topt,
3947 struct tracer_flags *flags,
3948 struct tracer_opt *opt)
3950 struct dentry *t_options;
3952 t_options = trace_options_init_dentry();
3956 topt->flags = flags;
3959 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
3960 &trace_options_fops);
3964 static struct trace_option_dentry *
3965 create_trace_option_files(struct tracer *tracer)
3967 struct trace_option_dentry *topts;
3968 struct tracer_flags *flags;
3969 struct tracer_opt *opts;
3975 flags = tracer->flags;
3977 if (!flags || !flags->opts)
3982 for (cnt = 0; opts[cnt].name; cnt++)
3985 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
3989 for (cnt = 0; opts[cnt].name; cnt++)
3990 create_trace_option_file(&topts[cnt], flags,
3997 destroy_trace_option_files(struct trace_option_dentry *topts)
4004 for (cnt = 0; topts[cnt].opt; cnt++) {
4005 if (topts[cnt].entry)
4006 debugfs_remove(topts[cnt].entry);
4012 static struct dentry *
4013 create_trace_option_core_file(const char *option, long index)
4015 struct dentry *t_options;
4017 t_options = trace_options_init_dentry();
4021 return trace_create_file(option, 0644, t_options, (void *)index,
4022 &trace_options_core_fops);
4025 static __init void create_trace_options_dir(void)
4027 struct dentry *t_options;
4030 t_options = trace_options_init_dentry();
4034 for (i = 0; trace_options[i]; i++)
4035 create_trace_option_core_file(trace_options[i], i);
4038 static __init int tracer_init_debugfs(void)
4040 struct dentry *d_tracer;
4043 d_tracer = tracing_init_dentry();
4045 trace_create_file("tracing_enabled", 0644, d_tracer,
4046 &global_trace, &tracing_ctrl_fops);
4048 trace_create_file("trace_options", 0644, d_tracer,
4049 NULL, &tracing_iter_fops);
4051 trace_create_file("tracing_cpumask", 0644, d_tracer,
4052 NULL, &tracing_cpumask_fops);
4054 trace_create_file("trace", 0644, d_tracer,
4055 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4057 trace_create_file("available_tracers", 0444, d_tracer,
4058 &global_trace, &show_traces_fops);
4060 trace_create_file("current_tracer", 0644, d_tracer,
4061 &global_trace, &set_tracer_fops);
4063 trace_create_file("tracing_max_latency", 0644, d_tracer,
4064 &tracing_max_latency, &tracing_max_lat_fops);
4066 trace_create_file("tracing_thresh", 0644, d_tracer,
4067 &tracing_thresh, &tracing_max_lat_fops);
4069 trace_create_file("README", 0444, d_tracer,
4070 NULL, &tracing_readme_fops);
4072 trace_create_file("trace_pipe", 0444, d_tracer,
4073 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4075 trace_create_file("buffer_size_kb", 0644, d_tracer,
4076 &global_trace, &tracing_entries_fops);
4078 trace_create_file("trace_marker", 0220, d_tracer,
4079 NULL, &tracing_mark_fops);
4081 trace_create_file("saved_cmdlines", 0444, d_tracer,
4082 NULL, &tracing_saved_cmdlines_fops);
4084 #ifdef CONFIG_DYNAMIC_FTRACE
4085 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4086 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4088 #ifdef CONFIG_SYSPROF_TRACER
4089 init_tracer_sysprof_debugfs(d_tracer);
4092 create_trace_options_dir();
4094 for_each_tracing_cpu(cpu)
4095 tracing_init_debugfs_percpu(cpu);
4100 static int trace_panic_handler(struct notifier_block *this,
4101 unsigned long event, void *unused)
4103 if (ftrace_dump_on_oops)
4108 static struct notifier_block trace_panic_notifier = {
4109 .notifier_call = trace_panic_handler,
4111 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4114 static int trace_die_handler(struct notifier_block *self,
4120 if (ftrace_dump_on_oops)
4129 static struct notifier_block trace_die_notifier = {
4130 .notifier_call = trace_die_handler,
4135 * printk is set to max of 1024, we really don't need it that big.
4136 * Nothing should be printing 1000 characters anyway.
4138 #define TRACE_MAX_PRINT 1000
4141 * Define here KERN_TRACE so that we have one place to modify
4142 * it if we decide to change what log level the ftrace dump
4145 #define KERN_TRACE KERN_EMERG
4148 trace_printk_seq(struct trace_seq *s)
4150 /* Probably should print a warning here. */
4154 /* should be zero ended, but we are paranoid. */
4155 s->buffer[s->len] = 0;
4157 printk(KERN_TRACE "%s", s->buffer);
4162 static void __ftrace_dump(bool disable_tracing)
4164 static raw_spinlock_t ftrace_dump_lock =
4165 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
4166 /* use static because iter can be a bit big for the stack */
4167 static struct trace_iterator iter;
4168 unsigned int old_userobj;
4169 static int dump_ran;
4170 unsigned long flags;
4174 local_irq_save(flags);
4175 __raw_spin_lock(&ftrace_dump_lock);
4183 if (disable_tracing)
4186 for_each_tracing_cpu(cpu) {
4187 atomic_inc(&global_trace.data[cpu]->disabled);
4190 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4192 /* don't look at user memory in panic mode */
4193 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4195 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4197 /* Simulate the iterator */
4198 iter.tr = &global_trace;
4199 iter.trace = current_trace;
4200 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4203 * We need to stop all tracing on all CPUS to read the
4204 * the next buffer. This is a bit expensive, but is
4205 * not done often. We fill all what we can read,
4206 * and then release the locks again.
4209 while (!trace_empty(&iter)) {
4212 printk(KERN_TRACE "---------------------------------\n");
4216 /* reset all but tr, trace, and overruns */
4217 memset(&iter.seq, 0,
4218 sizeof(struct trace_iterator) -
4219 offsetof(struct trace_iterator, seq));
4220 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4223 if (find_next_entry_inc(&iter) != NULL) {
4224 print_trace_line(&iter);
4225 trace_consume(&iter);
4228 trace_printk_seq(&iter.seq);
4232 printk(KERN_TRACE " (ftrace buffer empty)\n");
4234 printk(KERN_TRACE "---------------------------------\n");
4236 /* Re-enable tracing if requested */
4237 if (!disable_tracing) {
4238 trace_flags |= old_userobj;
4240 for_each_tracing_cpu(cpu) {
4241 atomic_dec(&global_trace.data[cpu]->disabled);
4247 __raw_spin_unlock(&ftrace_dump_lock);
4248 local_irq_restore(flags);
4251 /* By default: disable tracing after the dump */
4252 void ftrace_dump(void)
4254 __ftrace_dump(true);
4257 __init static int tracer_alloc_buffers(void)
4259 struct trace_array_cpu *data;
4264 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4267 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4268 goto out_free_buffer_mask;
4270 if (!alloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
4271 goto out_free_tracing_cpumask;
4273 /* To save memory, keep the ring buffer size to its minimum */
4274 if (ring_buffer_expanded)
4275 ring_buf_size = trace_buf_size;
4279 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4280 cpumask_copy(tracing_cpumask, cpu_all_mask);
4281 cpumask_clear(tracing_reader_cpumask);
4283 /* TODO: make the number of buffers hot pluggable with CPUS */
4284 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4285 TRACE_BUFFER_FLAGS);
4286 if (!global_trace.buffer) {
4287 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4289 goto out_free_cpumask;
4291 global_trace.entries = ring_buffer_size(global_trace.buffer);
4294 #ifdef CONFIG_TRACER_MAX_TRACE
4295 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4296 TRACE_BUFFER_FLAGS);
4297 if (!max_tr.buffer) {
4298 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4300 ring_buffer_free(global_trace.buffer);
4301 goto out_free_cpumask;
4303 max_tr.entries = ring_buffer_size(max_tr.buffer);
4304 WARN_ON(max_tr.entries != global_trace.entries);
4307 /* Allocate the first page for all buffers */
4308 for_each_tracing_cpu(i) {
4309 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4310 max_tr.data[i] = &per_cpu(max_data, i);
4313 trace_init_cmdlines();
4315 register_tracer(&nop_trace);
4316 current_trace = &nop_trace;
4317 #ifdef CONFIG_BOOT_TRACER
4318 register_tracer(&boot_tracer);
4320 /* All seems OK, enable tracing */
4321 tracing_disabled = 0;
4323 atomic_notifier_chain_register(&panic_notifier_list,
4324 &trace_panic_notifier);
4326 register_die_notifier(&trace_die_notifier);
4331 free_cpumask_var(tracing_reader_cpumask);
4332 out_free_tracing_cpumask:
4333 free_cpumask_var(tracing_cpumask);
4334 out_free_buffer_mask:
4335 free_cpumask_var(tracing_buffer_mask);
4340 __init static int clear_boot_tracer(void)
4343 * The default tracer at boot buffer is an init section.
4344 * This function is called in lateinit. If we did not
4345 * find the boot tracer, then clear it out, to prevent
4346 * later registration from accessing the buffer that is
4347 * about to be freed.
4349 if (!default_bootup_tracer)
4352 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4353 default_bootup_tracer);
4354 default_bootup_tracer = NULL;
4359 early_initcall(tracer_alloc_buffers);
4360 fs_initcall(tracer_init_debugfs);
4361 late_initcall(clear_boot_tracer);