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/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/ctype.h>
36 #include <linux/init.h>
37 #include <linux/poll.h>
38 #include <linux/gfp.h>
42 #include "trace_output.h"
44 #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
46 unsigned long __read_mostly tracing_max_latency;
47 unsigned long __read_mostly tracing_thresh;
50 * On boot up, the ring buffer is set to the minimum size, so that
51 * we do not waste memory on systems that are not using tracing.
53 int ring_buffer_expanded;
56 * We need to change this state when a selftest is running.
57 * A selftest will lurk into the ring-buffer to count the
58 * entries inserted during the selftest although some concurrent
59 * insertions into the ring-buffer such as trace_printk could occurred
60 * at the same time, giving false positive or negative results.
62 static bool __read_mostly tracing_selftest_running;
65 * If a tracer is running, we do not want to run SELFTEST.
67 bool __read_mostly tracing_selftest_disabled;
69 /* For tracers that don't implement custom flags */
70 static struct tracer_opt dummy_tracer_opt[] = {
74 static struct tracer_flags dummy_tracer_flags = {
76 .opts = dummy_tracer_opt
79 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
85 * Kill all tracing for good (never come back).
86 * It is initialized to 1 but will turn to zero if the initialization
87 * of the tracer is successful. But that is the only place that sets
90 static int tracing_disabled = 1;
92 DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
94 static inline void ftrace_disable_cpu(void)
97 local_inc(&__get_cpu_var(ftrace_cpu_disabled));
100 static inline void ftrace_enable_cpu(void)
102 local_dec(&__get_cpu_var(ftrace_cpu_disabled));
106 static cpumask_var_t __read_mostly tracing_buffer_mask;
108 /* Define which cpu buffers are currently read in trace_pipe */
109 static cpumask_var_t tracing_reader_cpumask;
111 #define for_each_tracing_cpu(cpu) \
112 for_each_cpu(cpu, tracing_buffer_mask)
115 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
117 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
118 * is set, then ftrace_dump is called. This will output the contents
119 * of the ftrace buffers to the console. This is very useful for
120 * capturing traces that lead to crashes and outputing it to a
123 * It is default off, but you can enable it with either specifying
124 * "ftrace_dump_on_oops" in the kernel command line, or setting
125 * /proc/sys/kernel/ftrace_dump_on_oops to true.
127 int ftrace_dump_on_oops;
129 static int tracing_set_tracer(const char *buf);
131 #define BOOTUP_TRACER_SIZE 100
132 static char bootup_tracer_buf[BOOTUP_TRACER_SIZE] __initdata;
133 static char *default_bootup_tracer;
135 static int __init set_ftrace(char *str)
137 strncpy(bootup_tracer_buf, str, BOOTUP_TRACER_SIZE);
138 default_bootup_tracer = bootup_tracer_buf;
139 /* We are using ftrace early, expand it */
140 ring_buffer_expanded = 1;
143 __setup("ftrace=", set_ftrace);
145 static int __init set_ftrace_dump_on_oops(char *str)
147 ftrace_dump_on_oops = 1;
150 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
152 unsigned long long ns2usecs(cycle_t nsec)
160 * The global_trace is the descriptor that holds the tracing
161 * buffers for the live tracing. For each CPU, it contains
162 * a link list of pages that will store trace entries. The
163 * page descriptor of the pages in the memory is used to hold
164 * the link list by linking the lru item in the page descriptor
165 * to each of the pages in the buffer per CPU.
167 * For each active CPU there is a data field that holds the
168 * pages for the buffer for that CPU. Each CPU has the same number
169 * of pages allocated for its buffer.
171 static struct trace_array global_trace;
173 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
175 int filter_current_check_discard(struct ftrace_event_call *call, void *rec,
176 struct ring_buffer_event *event)
178 return filter_check_discard(call, rec, global_trace.buffer, event);
180 EXPORT_SYMBOL_GPL(filter_current_check_discard);
182 cycle_t ftrace_now(int cpu)
186 /* Early boot up does not have a buffer yet */
187 if (!global_trace.buffer)
188 return trace_clock_local();
190 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
191 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
197 * The max_tr is used to snapshot the global_trace when a maximum
198 * latency is reached. Some tracers will use this to store a maximum
199 * trace while it continues examining live traces.
201 * The buffers for the max_tr are set up the same as the global_trace.
202 * When a snapshot is taken, the link list of the max_tr is swapped
203 * with the link list of the global_trace and the buffers are reset for
204 * the global_trace so the tracing can continue.
206 static struct trace_array max_tr;
208 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
210 /* tracer_enabled is used to toggle activation of a tracer */
211 static int tracer_enabled = 1;
214 * tracing_is_enabled - return tracer_enabled status
216 * This function is used by other tracers to know the status
217 * of the tracer_enabled flag. Tracers may use this function
218 * to know if it should enable their features when starting
219 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
221 int tracing_is_enabled(void)
223 return tracer_enabled;
227 * trace_buf_size is the size in bytes that is allocated
228 * for a buffer. Note, the number of bytes is always rounded
231 * This number is purposely set to a low number of 16384.
232 * If the dump on oops happens, it will be much appreciated
233 * to not have to wait for all that output. Anyway this can be
234 * boot time and run time configurable.
236 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
238 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
240 /* trace_types holds a link list of available tracers. */
241 static struct tracer *trace_types __read_mostly;
243 /* current_trace points to the tracer that is currently active */
244 static struct tracer *current_trace __read_mostly;
247 * max_tracer_type_len is used to simplify the allocating of
248 * buffers to read userspace tracer names. We keep track of
249 * the longest tracer name registered.
251 static int max_tracer_type_len;
254 * trace_types_lock is used to protect the trace_types list.
255 * This lock is also used to keep user access serialized.
256 * Accesses from userspace will grab this lock while userspace
257 * activities happen inside the kernel.
259 static DEFINE_MUTEX(trace_types_lock);
261 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
262 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
264 /* trace_flags holds trace_options default values */
265 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
266 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
267 TRACE_ITER_GRAPH_TIME;
270 * trace_wake_up - wake up tasks waiting for trace input
272 * Simply wakes up any task that is blocked on the trace_wait
273 * queue. These is used with trace_poll for tasks polling the trace.
275 void trace_wake_up(void)
278 * The runqueue_is_locked() can fail, but this is the best we
281 if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
282 wake_up(&trace_wait);
285 static int __init set_buf_size(char *str)
287 unsigned long buf_size;
291 buf_size = memparse(str, &str);
292 /* nr_entries can not be zero */
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 /sys/kernel/debug/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 void tracing_reset_current(int cpu)
644 tracing_reset(&global_trace, cpu);
647 void tracing_reset_current_online_cpus(void)
649 tracing_reset_online_cpus(&global_trace);
652 #define SAVED_CMDLINES 128
653 #define NO_CMDLINE_MAP UINT_MAX
654 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
655 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
656 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
657 static int cmdline_idx;
658 static raw_spinlock_t trace_cmdline_lock = __RAW_SPIN_LOCK_UNLOCKED;
660 /* temporary disable recording */
661 static atomic_t trace_record_cmdline_disabled __read_mostly;
663 static void trace_init_cmdlines(void)
665 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
666 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
670 static int trace_stop_count;
671 static DEFINE_SPINLOCK(tracing_start_lock);
674 * ftrace_off_permanent - disable all ftrace code permanently
676 * This should only be called when a serious anomally has
677 * been detected. This will turn off the function tracing,
678 * ring buffers, and other tracing utilites. It takes no
679 * locks and can be called from any context.
681 void ftrace_off_permanent(void)
683 tracing_disabled = 1;
685 tracing_off_permanent();
689 * tracing_start - quick start of the tracer
691 * If tracing is enabled but was stopped by tracing_stop,
692 * this will start the tracer back up.
694 void tracing_start(void)
696 struct ring_buffer *buffer;
699 if (tracing_disabled)
702 spin_lock_irqsave(&tracing_start_lock, flags);
703 if (--trace_stop_count) {
704 if (trace_stop_count < 0) {
705 /* Someone screwed up their debugging */
707 trace_stop_count = 0;
713 buffer = global_trace.buffer;
715 ring_buffer_record_enable(buffer);
717 buffer = max_tr.buffer;
719 ring_buffer_record_enable(buffer);
723 spin_unlock_irqrestore(&tracing_start_lock, flags);
727 * tracing_stop - quick stop of the tracer
729 * Light weight way to stop tracing. Use in conjunction with
732 void tracing_stop(void)
734 struct ring_buffer *buffer;
738 spin_lock_irqsave(&tracing_start_lock, flags);
739 if (trace_stop_count++)
742 buffer = global_trace.buffer;
744 ring_buffer_record_disable(buffer);
746 buffer = max_tr.buffer;
748 ring_buffer_record_disable(buffer);
751 spin_unlock_irqrestore(&tracing_start_lock, flags);
754 void trace_stop_cmdline_recording(void);
756 static void trace_save_cmdline(struct task_struct *tsk)
760 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
764 * It's not the end of the world if we don't get
765 * the lock, but we also don't want to spin
766 * nor do we want to disable interrupts,
767 * so if we miss here, then better luck next time.
769 if (!__raw_spin_trylock(&trace_cmdline_lock))
772 idx = map_pid_to_cmdline[tsk->pid];
773 if (idx == NO_CMDLINE_MAP) {
774 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
777 * Check whether the cmdline buffer at idx has a pid
778 * mapped. We are going to overwrite that entry so we
779 * need to clear the map_pid_to_cmdline. Otherwise we
780 * would read the new comm for the old pid.
782 pid = map_cmdline_to_pid[idx];
783 if (pid != NO_CMDLINE_MAP)
784 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
786 map_cmdline_to_pid[idx] = tsk->pid;
787 map_pid_to_cmdline[tsk->pid] = idx;
792 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
794 __raw_spin_unlock(&trace_cmdline_lock);
797 void trace_find_cmdline(int pid, char comm[])
802 strcpy(comm, "<idle>");
806 if (pid > PID_MAX_DEFAULT) {
807 strcpy(comm, "<...>");
812 __raw_spin_lock(&trace_cmdline_lock);
813 map = map_pid_to_cmdline[pid];
814 if (map != NO_CMDLINE_MAP)
815 strcpy(comm, saved_cmdlines[map]);
817 strcpy(comm, "<...>");
819 __raw_spin_unlock(&trace_cmdline_lock);
823 void tracing_record_cmdline(struct task_struct *tsk)
825 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
829 trace_save_cmdline(tsk);
833 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
836 struct task_struct *tsk = current;
838 entry->preempt_count = pc & 0xff;
839 entry->pid = (tsk) ? tsk->pid : 0;
840 entry->tgid = (tsk) ? tsk->tgid : 0;
842 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
843 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
845 TRACE_FLAG_IRQS_NOSUPPORT |
847 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
848 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
849 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
852 struct ring_buffer_event *trace_buffer_lock_reserve(struct trace_array *tr,
855 unsigned long flags, int pc)
857 struct ring_buffer_event *event;
859 event = ring_buffer_lock_reserve(tr->buffer, len);
861 struct trace_entry *ent = ring_buffer_event_data(event);
863 tracing_generic_entry_update(ent, flags, pc);
870 static inline void __trace_buffer_unlock_commit(struct trace_array *tr,
871 struct ring_buffer_event *event,
872 unsigned long flags, int pc,
875 ring_buffer_unlock_commit(tr->buffer, event);
877 ftrace_trace_stack(tr, flags, 6, pc);
878 ftrace_trace_userstack(tr, flags, pc);
884 void trace_buffer_unlock_commit(struct trace_array *tr,
885 struct ring_buffer_event *event,
886 unsigned long flags, int pc)
888 __trace_buffer_unlock_commit(tr, event, flags, pc, 1);
891 struct ring_buffer_event *
892 trace_current_buffer_lock_reserve(int type, unsigned long len,
893 unsigned long flags, int pc)
895 return trace_buffer_lock_reserve(&global_trace,
896 type, len, flags, pc);
898 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
900 void trace_current_buffer_unlock_commit(struct ring_buffer_event *event,
901 unsigned long flags, int pc)
903 __trace_buffer_unlock_commit(&global_trace, event, flags, pc, 1);
905 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
907 void trace_nowake_buffer_unlock_commit(struct ring_buffer_event *event,
908 unsigned long flags, int pc)
910 __trace_buffer_unlock_commit(&global_trace, event, flags, pc, 0);
912 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
914 void trace_current_buffer_discard_commit(struct ring_buffer_event *event)
916 ring_buffer_discard_commit(global_trace.buffer, event);
918 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
921 trace_function(struct trace_array *tr,
922 unsigned long ip, unsigned long parent_ip, unsigned long flags,
925 struct ftrace_event_call *call = &event_function;
926 struct ring_buffer_event *event;
927 struct ftrace_entry *entry;
929 /* If we are reading the ring buffer, don't trace */
930 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
933 event = trace_buffer_lock_reserve(tr, TRACE_FN, sizeof(*entry),
937 entry = ring_buffer_event_data(event);
939 entry->parent_ip = parent_ip;
941 if (!filter_check_discard(call, entry, tr->buffer, event))
942 ring_buffer_unlock_commit(tr->buffer, event);
946 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
947 unsigned long ip, unsigned long parent_ip, unsigned long flags,
950 if (likely(!atomic_read(&data->disabled)))
951 trace_function(tr, ip, parent_ip, flags, pc);
954 #ifdef CONFIG_STACKTRACE
955 static void __ftrace_trace_stack(struct trace_array *tr,
959 struct ftrace_event_call *call = &event_kernel_stack;
960 struct ring_buffer_event *event;
961 struct stack_entry *entry;
962 struct stack_trace trace;
964 event = trace_buffer_lock_reserve(tr, TRACE_STACK,
965 sizeof(*entry), flags, pc);
968 entry = ring_buffer_event_data(event);
969 memset(&entry->caller, 0, sizeof(entry->caller));
971 trace.nr_entries = 0;
972 trace.max_entries = FTRACE_STACK_ENTRIES;
974 trace.entries = entry->caller;
976 save_stack_trace(&trace);
977 if (!filter_check_discard(call, entry, tr->buffer, event))
978 ring_buffer_unlock_commit(tr->buffer, event);
981 void ftrace_trace_stack(struct trace_array *tr, unsigned long flags, int skip,
984 if (!(trace_flags & TRACE_ITER_STACKTRACE))
987 __ftrace_trace_stack(tr, flags, skip, pc);
990 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
993 __ftrace_trace_stack(tr, flags, skip, pc);
996 void ftrace_trace_userstack(struct trace_array *tr, unsigned long flags, int pc)
998 struct ftrace_event_call *call = &event_user_stack;
999 struct ring_buffer_event *event;
1000 struct userstack_entry *entry;
1001 struct stack_trace trace;
1003 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1006 event = trace_buffer_lock_reserve(tr, TRACE_USER_STACK,
1007 sizeof(*entry), flags, pc);
1010 entry = ring_buffer_event_data(event);
1012 memset(&entry->caller, 0, sizeof(entry->caller));
1014 trace.nr_entries = 0;
1015 trace.max_entries = FTRACE_STACK_ENTRIES;
1017 trace.entries = entry->caller;
1019 save_stack_trace_user(&trace);
1020 if (!filter_check_discard(call, entry, tr->buffer, event))
1021 ring_buffer_unlock_commit(tr->buffer, event);
1025 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1027 ftrace_trace_userstack(tr, flags, preempt_count());
1031 #endif /* CONFIG_STACKTRACE */
1034 ftrace_trace_special(void *__tr,
1035 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1038 struct ring_buffer_event *event;
1039 struct trace_array *tr = __tr;
1040 struct special_entry *entry;
1042 event = trace_buffer_lock_reserve(tr, TRACE_SPECIAL,
1043 sizeof(*entry), 0, pc);
1046 entry = ring_buffer_event_data(event);
1050 trace_buffer_unlock_commit(tr, event, 0, pc);
1054 __trace_special(void *__tr, void *__data,
1055 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1057 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1061 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1063 struct trace_array *tr = &global_trace;
1064 struct trace_array_cpu *data;
1065 unsigned long flags;
1069 if (tracing_disabled)
1072 pc = preempt_count();
1073 local_irq_save(flags);
1074 cpu = raw_smp_processor_id();
1075 data = tr->data[cpu];
1077 if (likely(atomic_inc_return(&data->disabled) == 1))
1078 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1080 atomic_dec(&data->disabled);
1081 local_irq_restore(flags);
1085 * trace_vbprintk - write binary msg to tracing buffer
1088 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1090 static raw_spinlock_t trace_buf_lock =
1091 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
1092 static u32 trace_buf[TRACE_BUF_SIZE];
1094 struct ftrace_event_call *call = &event_bprint;
1095 struct ring_buffer_event *event;
1096 struct trace_array *tr = &global_trace;
1097 struct trace_array_cpu *data;
1098 struct bprint_entry *entry;
1099 unsigned long flags;
1102 int cpu, len = 0, size, pc;
1104 if (unlikely(tracing_selftest_running || tracing_disabled))
1107 /* Don't pollute graph traces with trace_vprintk internals */
1108 pause_graph_tracing();
1110 pc = preempt_count();
1111 resched = ftrace_preempt_disable();
1112 cpu = raw_smp_processor_id();
1113 data = tr->data[cpu];
1115 disable = atomic_inc_return(&data->disabled);
1116 if (unlikely(disable != 1))
1119 /* Lockdep uses trace_printk for lock tracing */
1120 local_irq_save(flags);
1121 __raw_spin_lock(&trace_buf_lock);
1122 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1124 if (len > TRACE_BUF_SIZE || len < 0)
1127 size = sizeof(*entry) + sizeof(u32) * len;
1128 event = trace_buffer_lock_reserve(tr, TRACE_BPRINT, size, flags, pc);
1131 entry = ring_buffer_event_data(event);
1135 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1136 if (!filter_check_discard(call, entry, tr->buffer, event))
1137 ring_buffer_unlock_commit(tr->buffer, event);
1140 __raw_spin_unlock(&trace_buf_lock);
1141 local_irq_restore(flags);
1144 atomic_dec_return(&data->disabled);
1145 ftrace_preempt_enable(resched);
1146 unpause_graph_tracing();
1150 EXPORT_SYMBOL_GPL(trace_vbprintk);
1152 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1154 static raw_spinlock_t trace_buf_lock = __RAW_SPIN_LOCK_UNLOCKED;
1155 static char trace_buf[TRACE_BUF_SIZE];
1157 struct ftrace_event_call *call = &event_print;
1158 struct ring_buffer_event *event;
1159 struct trace_array *tr = &global_trace;
1160 struct trace_array_cpu *data;
1161 int cpu, len = 0, size, pc;
1162 struct print_entry *entry;
1163 unsigned long irq_flags;
1166 if (tracing_disabled || tracing_selftest_running)
1169 pc = preempt_count();
1170 preempt_disable_notrace();
1171 cpu = raw_smp_processor_id();
1172 data = tr->data[cpu];
1174 disable = atomic_inc_return(&data->disabled);
1175 if (unlikely(disable != 1))
1178 pause_graph_tracing();
1179 raw_local_irq_save(irq_flags);
1180 __raw_spin_lock(&trace_buf_lock);
1181 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1183 len = min(len, TRACE_BUF_SIZE-1);
1186 size = sizeof(*entry) + len + 1;
1187 event = trace_buffer_lock_reserve(tr, TRACE_PRINT, size, irq_flags, pc);
1190 entry = ring_buffer_event_data(event);
1193 memcpy(&entry->buf, trace_buf, len);
1194 entry->buf[len] = 0;
1195 if (!filter_check_discard(call, entry, tr->buffer, event))
1196 ring_buffer_unlock_commit(tr->buffer, event);
1199 __raw_spin_unlock(&trace_buf_lock);
1200 raw_local_irq_restore(irq_flags);
1201 unpause_graph_tracing();
1203 atomic_dec_return(&data->disabled);
1204 preempt_enable_notrace();
1208 EXPORT_SYMBOL_GPL(trace_vprintk);
1210 enum trace_file_type {
1211 TRACE_FILE_LAT_FMT = 1,
1212 TRACE_FILE_ANNOTATE = 2,
1215 static void trace_iterator_increment(struct trace_iterator *iter)
1217 /* Don't allow ftrace to trace into the ring buffers */
1218 ftrace_disable_cpu();
1221 if (iter->buffer_iter[iter->cpu])
1222 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1224 ftrace_enable_cpu();
1227 static struct trace_entry *
1228 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1230 struct ring_buffer_event *event;
1231 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1233 /* Don't allow ftrace to trace into the ring buffers */
1234 ftrace_disable_cpu();
1237 event = ring_buffer_iter_peek(buf_iter, ts);
1239 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1241 ftrace_enable_cpu();
1243 return event ? ring_buffer_event_data(event) : NULL;
1246 static struct trace_entry *
1247 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1249 struct ring_buffer *buffer = iter->tr->buffer;
1250 struct trace_entry *ent, *next = NULL;
1251 int cpu_file = iter->cpu_file;
1252 u64 next_ts = 0, ts;
1257 * If we are in a per_cpu trace file, don't bother by iterating over
1258 * all cpu and peek directly.
1260 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1261 if (ring_buffer_empty_cpu(buffer, cpu_file))
1263 ent = peek_next_entry(iter, cpu_file, ent_ts);
1265 *ent_cpu = cpu_file;
1270 for_each_tracing_cpu(cpu) {
1272 if (ring_buffer_empty_cpu(buffer, cpu))
1275 ent = peek_next_entry(iter, cpu, &ts);
1278 * Pick the entry with the smallest timestamp:
1280 if (ent && (!next || ts < next_ts)) {
1288 *ent_cpu = next_cpu;
1296 /* Find the next real entry, without updating the iterator itself */
1297 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1298 int *ent_cpu, u64 *ent_ts)
1300 return __find_next_entry(iter, ent_cpu, ent_ts);
1303 /* Find the next real entry, and increment the iterator to the next entry */
1304 static void *find_next_entry_inc(struct trace_iterator *iter)
1306 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1309 trace_iterator_increment(iter);
1311 return iter->ent ? iter : NULL;
1314 static void trace_consume(struct trace_iterator *iter)
1316 /* Don't allow ftrace to trace into the ring buffers */
1317 ftrace_disable_cpu();
1318 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1319 ftrace_enable_cpu();
1322 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1324 struct trace_iterator *iter = m->private;
1330 /* can't go backwards */
1335 ent = find_next_entry_inc(iter);
1339 while (ent && iter->idx < i)
1340 ent = find_next_entry_inc(iter);
1348 * No necessary locking here. The worst thing which can
1349 * happen is loosing events consumed at the same time
1350 * by a trace_pipe reader.
1351 * Other than that, we don't risk to crash the ring buffer
1352 * because it serializes the readers.
1354 * The current tracer is copied to avoid a global locking
1357 static void *s_start(struct seq_file *m, loff_t *pos)
1359 struct trace_iterator *iter = m->private;
1360 static struct tracer *old_tracer;
1361 int cpu_file = iter->cpu_file;
1366 /* copy the tracer to avoid using a global lock all around */
1367 mutex_lock(&trace_types_lock);
1368 if (unlikely(old_tracer != current_trace && current_trace)) {
1369 old_tracer = current_trace;
1370 *iter->trace = *current_trace;
1372 mutex_unlock(&trace_types_lock);
1374 atomic_inc(&trace_record_cmdline_disabled);
1376 if (*pos != iter->pos) {
1381 ftrace_disable_cpu();
1383 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1384 for_each_tracing_cpu(cpu)
1385 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1387 ring_buffer_iter_reset(iter->buffer_iter[cpu_file]);
1390 ftrace_enable_cpu();
1392 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1397 p = s_next(m, p, &l);
1400 trace_event_read_lock();
1404 static void s_stop(struct seq_file *m, void *p)
1406 atomic_dec(&trace_record_cmdline_disabled);
1407 trace_event_read_unlock();
1410 static void print_lat_help_header(struct seq_file *m)
1412 seq_puts(m, "# _------=> CPU# \n");
1413 seq_puts(m, "# / _-----=> irqs-off \n");
1414 seq_puts(m, "# | / _----=> need-resched \n");
1415 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1416 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1417 seq_puts(m, "# |||| / \n");
1418 seq_puts(m, "# ||||| delay \n");
1419 seq_puts(m, "# cmd pid ||||| time | caller \n");
1420 seq_puts(m, "# \\ / ||||| \\ | / \n");
1423 static void print_func_help_header(struct seq_file *m)
1425 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1426 seq_puts(m, "# | | | | |\n");
1431 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1433 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1434 struct trace_array *tr = iter->tr;
1435 struct trace_array_cpu *data = tr->data[tr->cpu];
1436 struct tracer *type = current_trace;
1437 unsigned long total;
1438 unsigned long entries;
1439 const char *name = "preemption";
1444 entries = ring_buffer_entries(iter->tr->buffer);
1446 ring_buffer_overruns(iter->tr->buffer);
1448 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1450 seq_puts(m, "# -----------------------------------"
1451 "---------------------------------\n");
1452 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1453 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1454 nsecs_to_usecs(data->saved_latency),
1458 #if defined(CONFIG_PREEMPT_NONE)
1460 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1462 #elif defined(CONFIG_PREEMPT)
1467 /* These are reserved for later use */
1470 seq_printf(m, " #P:%d)\n", num_online_cpus());
1474 seq_puts(m, "# -----------------\n");
1475 seq_printf(m, "# | task: %.16s-%d "
1476 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1477 data->comm, data->pid, data->uid, data->nice,
1478 data->policy, data->rt_priority);
1479 seq_puts(m, "# -----------------\n");
1481 if (data->critical_start) {
1482 seq_puts(m, "# => started at: ");
1483 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1484 trace_print_seq(m, &iter->seq);
1485 seq_puts(m, "\n# => ended at: ");
1486 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1487 trace_print_seq(m, &iter->seq);
1494 static void test_cpu_buff_start(struct trace_iterator *iter)
1496 struct trace_seq *s = &iter->seq;
1498 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1501 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1504 if (cpumask_test_cpu(iter->cpu, iter->started))
1507 cpumask_set_cpu(iter->cpu, iter->started);
1509 /* Don't print started cpu buffer for the first entry of the trace */
1511 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1515 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1517 struct trace_seq *s = &iter->seq;
1518 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1519 struct trace_entry *entry;
1520 struct trace_event *event;
1524 test_cpu_buff_start(iter);
1526 event = ftrace_find_event(entry->type);
1528 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1529 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1530 if (!trace_print_lat_context(iter))
1533 if (!trace_print_context(iter))
1539 return event->trace(iter, sym_flags);
1541 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1544 return TRACE_TYPE_HANDLED;
1546 return TRACE_TYPE_PARTIAL_LINE;
1549 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1551 struct trace_seq *s = &iter->seq;
1552 struct trace_entry *entry;
1553 struct trace_event *event;
1557 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1558 if (!trace_seq_printf(s, "%d %d %llu ",
1559 entry->pid, iter->cpu, iter->ts))
1563 event = ftrace_find_event(entry->type);
1565 return event->raw(iter, 0);
1567 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1570 return TRACE_TYPE_HANDLED;
1572 return TRACE_TYPE_PARTIAL_LINE;
1575 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1577 struct trace_seq *s = &iter->seq;
1578 unsigned char newline = '\n';
1579 struct trace_entry *entry;
1580 struct trace_event *event;
1584 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1585 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1586 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1587 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1590 event = ftrace_find_event(entry->type);
1592 enum print_line_t ret = event->hex(iter, 0);
1593 if (ret != TRACE_TYPE_HANDLED)
1597 SEQ_PUT_FIELD_RET(s, newline);
1599 return TRACE_TYPE_HANDLED;
1602 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1604 struct trace_seq *s = &iter->seq;
1605 struct trace_entry *entry;
1606 struct trace_event *event;
1610 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1611 SEQ_PUT_FIELD_RET(s, entry->pid);
1612 SEQ_PUT_FIELD_RET(s, iter->cpu);
1613 SEQ_PUT_FIELD_RET(s, iter->ts);
1616 event = ftrace_find_event(entry->type);
1617 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1620 static int trace_empty(struct trace_iterator *iter)
1624 /* If we are looking at one CPU buffer, only check that one */
1625 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1626 cpu = iter->cpu_file;
1627 if (iter->buffer_iter[cpu]) {
1628 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1631 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1637 for_each_tracing_cpu(cpu) {
1638 if (iter->buffer_iter[cpu]) {
1639 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1642 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1650 /* Called with trace_event_read_lock() held. */
1651 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1653 enum print_line_t ret;
1655 if (iter->trace && iter->trace->print_line) {
1656 ret = iter->trace->print_line(iter);
1657 if (ret != TRACE_TYPE_UNHANDLED)
1661 if (iter->ent->type == TRACE_BPRINT &&
1662 trace_flags & TRACE_ITER_PRINTK &&
1663 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1664 return trace_print_bprintk_msg_only(iter);
1666 if (iter->ent->type == TRACE_PRINT &&
1667 trace_flags & TRACE_ITER_PRINTK &&
1668 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1669 return trace_print_printk_msg_only(iter);
1671 if (trace_flags & TRACE_ITER_BIN)
1672 return print_bin_fmt(iter);
1674 if (trace_flags & TRACE_ITER_HEX)
1675 return print_hex_fmt(iter);
1677 if (trace_flags & TRACE_ITER_RAW)
1678 return print_raw_fmt(iter);
1680 return print_trace_fmt(iter);
1683 static int s_show(struct seq_file *m, void *v)
1685 struct trace_iterator *iter = v;
1687 if (iter->ent == NULL) {
1689 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1692 if (iter->trace && iter->trace->print_header)
1693 iter->trace->print_header(m);
1694 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1695 /* print nothing if the buffers are empty */
1696 if (trace_empty(iter))
1698 print_trace_header(m, iter);
1699 if (!(trace_flags & TRACE_ITER_VERBOSE))
1700 print_lat_help_header(m);
1702 if (!(trace_flags & TRACE_ITER_VERBOSE))
1703 print_func_help_header(m);
1706 print_trace_line(iter);
1707 trace_print_seq(m, &iter->seq);
1713 static struct seq_operations tracer_seq_ops = {
1720 static struct trace_iterator *
1721 __tracing_open(struct inode *inode, struct file *file)
1723 long cpu_file = (long) inode->i_private;
1724 void *fail_ret = ERR_PTR(-ENOMEM);
1725 struct trace_iterator *iter;
1729 if (tracing_disabled)
1730 return ERR_PTR(-ENODEV);
1732 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1734 return ERR_PTR(-ENOMEM);
1737 * We make a copy of the current tracer to avoid concurrent
1738 * changes on it while we are reading.
1740 mutex_lock(&trace_types_lock);
1741 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
1746 *iter->trace = *current_trace;
1748 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL))
1751 cpumask_clear(iter->started);
1753 if (current_trace && current_trace->print_max)
1756 iter->tr = &global_trace;
1758 mutex_init(&iter->mutex);
1759 iter->cpu_file = cpu_file;
1761 /* Notify the tracer early; before we stop tracing. */
1762 if (iter->trace && iter->trace->open)
1763 iter->trace->open(iter);
1765 /* Annotate start of buffers if we had overruns */
1766 if (ring_buffer_overruns(iter->tr->buffer))
1767 iter->iter_flags |= TRACE_FILE_ANNOTATE;
1769 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
1770 for_each_tracing_cpu(cpu) {
1772 iter->buffer_iter[cpu] =
1773 ring_buffer_read_start(iter->tr->buffer, cpu);
1776 cpu = iter->cpu_file;
1777 iter->buffer_iter[cpu] =
1778 ring_buffer_read_start(iter->tr->buffer, cpu);
1781 /* TODO stop tracer */
1782 ret = seq_open(file, &tracer_seq_ops);
1784 fail_ret = ERR_PTR(ret);
1788 m = file->private_data;
1791 /* stop the trace while dumping */
1794 mutex_unlock(&trace_types_lock);
1799 for_each_tracing_cpu(cpu) {
1800 if (iter->buffer_iter[cpu])
1801 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1803 free_cpumask_var(iter->started);
1805 mutex_unlock(&trace_types_lock);
1812 int tracing_open_generic(struct inode *inode, struct file *filp)
1814 if (tracing_disabled)
1817 filp->private_data = inode->i_private;
1821 static int tracing_release(struct inode *inode, struct file *file)
1823 struct seq_file *m = (struct seq_file *)file->private_data;
1824 struct trace_iterator *iter;
1827 if (!(file->f_mode & FMODE_READ))
1832 mutex_lock(&trace_types_lock);
1833 for_each_tracing_cpu(cpu) {
1834 if (iter->buffer_iter[cpu])
1835 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1838 if (iter->trace && iter->trace->close)
1839 iter->trace->close(iter);
1841 /* reenable tracing if it was previously enabled */
1843 mutex_unlock(&trace_types_lock);
1845 seq_release(inode, file);
1846 mutex_destroy(&iter->mutex);
1847 free_cpumask_var(iter->started);
1853 static int tracing_open(struct inode *inode, struct file *file)
1855 struct trace_iterator *iter;
1858 /* If this file was open for write, then erase contents */
1859 if ((file->f_mode & FMODE_WRITE) &&
1860 !(file->f_flags & O_APPEND)) {
1861 long cpu = (long) inode->i_private;
1863 if (cpu == TRACE_PIPE_ALL_CPU)
1864 tracing_reset_online_cpus(&global_trace);
1866 tracing_reset(&global_trace, cpu);
1869 if (file->f_mode & FMODE_READ) {
1870 iter = __tracing_open(inode, file);
1872 ret = PTR_ERR(iter);
1873 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
1874 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1880 t_next(struct seq_file *m, void *v, loff_t *pos)
1882 struct tracer *t = v;
1892 static void *t_start(struct seq_file *m, loff_t *pos)
1897 mutex_lock(&trace_types_lock);
1898 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
1904 static void t_stop(struct seq_file *m, void *p)
1906 mutex_unlock(&trace_types_lock);
1909 static int t_show(struct seq_file *m, void *v)
1911 struct tracer *t = v;
1916 seq_printf(m, "%s", t->name);
1925 static struct seq_operations show_traces_seq_ops = {
1932 static int show_traces_open(struct inode *inode, struct file *file)
1934 if (tracing_disabled)
1937 return seq_open(file, &show_traces_seq_ops);
1941 tracing_write_stub(struct file *filp, const char __user *ubuf,
1942 size_t count, loff_t *ppos)
1947 static const struct file_operations tracing_fops = {
1948 .open = tracing_open,
1950 .write = tracing_write_stub,
1951 .llseek = seq_lseek,
1952 .release = tracing_release,
1955 static const struct file_operations show_traces_fops = {
1956 .open = show_traces_open,
1958 .release = seq_release,
1962 * Only trace on a CPU if the bitmask is set:
1964 static cpumask_var_t tracing_cpumask;
1967 * The tracer itself will not take this lock, but still we want
1968 * to provide a consistent cpumask to user-space:
1970 static DEFINE_MUTEX(tracing_cpumask_update_lock);
1973 * Temporary storage for the character representation of the
1974 * CPU bitmask (and one more byte for the newline):
1976 static char mask_str[NR_CPUS + 1];
1979 tracing_cpumask_read(struct file *filp, char __user *ubuf,
1980 size_t count, loff_t *ppos)
1984 mutex_lock(&tracing_cpumask_update_lock);
1986 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
1987 if (count - len < 2) {
1991 len += sprintf(mask_str + len, "\n");
1992 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
1995 mutex_unlock(&tracing_cpumask_update_lock);
2001 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2002 size_t count, loff_t *ppos)
2005 cpumask_var_t tracing_cpumask_new;
2007 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2010 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2014 mutex_lock(&tracing_cpumask_update_lock);
2016 local_irq_disable();
2017 __raw_spin_lock(&ftrace_max_lock);
2018 for_each_tracing_cpu(cpu) {
2020 * Increase/decrease the disabled counter if we are
2021 * about to flip a bit in the cpumask:
2023 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2024 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2025 atomic_inc(&global_trace.data[cpu]->disabled);
2027 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2028 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2029 atomic_dec(&global_trace.data[cpu]->disabled);
2032 __raw_spin_unlock(&ftrace_max_lock);
2035 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2037 mutex_unlock(&tracing_cpumask_update_lock);
2038 free_cpumask_var(tracing_cpumask_new);
2043 free_cpumask_var(tracing_cpumask_new);
2048 static const struct file_operations tracing_cpumask_fops = {
2049 .open = tracing_open_generic,
2050 .read = tracing_cpumask_read,
2051 .write = tracing_cpumask_write,
2055 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2056 size_t cnt, loff_t *ppos)
2058 struct tracer_opt *trace_opts;
2066 /* calculate max size */
2067 for (i = 0; trace_options[i]; i++) {
2068 len += strlen(trace_options[i]);
2069 len += 3; /* "no" and newline */
2072 mutex_lock(&trace_types_lock);
2073 tracer_flags = current_trace->flags->val;
2074 trace_opts = current_trace->flags->opts;
2077 * Increase the size with names of options specific
2078 * of the current tracer.
2080 for (i = 0; trace_opts[i].name; i++) {
2081 len += strlen(trace_opts[i].name);
2082 len += 3; /* "no" and newline */
2086 buf = kmalloc(len + 1, GFP_KERNEL);
2088 mutex_unlock(&trace_types_lock);
2092 for (i = 0; trace_options[i]; i++) {
2093 if (trace_flags & (1 << i))
2094 r += sprintf(buf + r, "%s\n", trace_options[i]);
2096 r += sprintf(buf + r, "no%s\n", trace_options[i]);
2099 for (i = 0; trace_opts[i].name; i++) {
2100 if (tracer_flags & trace_opts[i].bit)
2101 r += sprintf(buf + r, "%s\n",
2102 trace_opts[i].name);
2104 r += sprintf(buf + r, "no%s\n",
2105 trace_opts[i].name);
2107 mutex_unlock(&trace_types_lock);
2109 WARN_ON(r >= len + 1);
2111 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2117 /* Try to assign a tracer specific option */
2118 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2120 struct tracer_flags *trace_flags = trace->flags;
2121 struct tracer_opt *opts = NULL;
2125 for (i = 0; trace_flags->opts[i].name; i++) {
2126 opts = &trace_flags->opts[i];
2127 len = strlen(opts->name);
2129 if (strncmp(cmp, opts->name, len) == 0) {
2130 ret = trace->set_flag(trace_flags->val,
2136 if (!trace_flags->opts[i].name)
2139 /* Refused to handle */
2144 trace_flags->val &= ~opts->bit;
2146 trace_flags->val |= opts->bit;
2151 static void set_tracer_flags(unsigned int mask, int enabled)
2153 /* do nothing if flag is already set */
2154 if (!!(trace_flags & mask) == !!enabled)
2158 trace_flags |= mask;
2160 trace_flags &= ~mask;
2162 if (mask == TRACE_ITER_GLOBAL_CLK) {
2166 func = trace_clock_global;
2168 func = trace_clock_local;
2170 mutex_lock(&trace_types_lock);
2171 ring_buffer_set_clock(global_trace.buffer, func);
2174 ring_buffer_set_clock(max_tr.buffer, func);
2175 mutex_unlock(&trace_types_lock);
2180 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2181 size_t cnt, loff_t *ppos)
2189 if (cnt >= sizeof(buf))
2192 if (copy_from_user(&buf, ubuf, cnt))
2197 if (strncmp(buf, "no", 2) == 0) {
2202 for (i = 0; trace_options[i]; i++) {
2203 int len = strlen(trace_options[i]);
2205 if (strncmp(cmp, trace_options[i], len) == 0) {
2206 set_tracer_flags(1 << i, !neg);
2211 /* If no option could be set, test the specific tracer options */
2212 if (!trace_options[i]) {
2213 mutex_lock(&trace_types_lock);
2214 ret = set_tracer_option(current_trace, cmp, neg);
2215 mutex_unlock(&trace_types_lock);
2225 static const struct file_operations tracing_iter_fops = {
2226 .open = tracing_open_generic,
2227 .read = tracing_trace_options_read,
2228 .write = tracing_trace_options_write,
2231 static const char readme_msg[] =
2232 "tracing mini-HOWTO:\n\n"
2233 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2234 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2235 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2236 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2238 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2239 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2241 "# cat /sys/kernel/debug/tracing/trace_options\n"
2242 "noprint-parent nosym-offset nosym-addr noverbose\n"
2243 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2244 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2245 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2246 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2250 tracing_readme_read(struct file *filp, char __user *ubuf,
2251 size_t cnt, loff_t *ppos)
2253 return simple_read_from_buffer(ubuf, cnt, ppos,
2254 readme_msg, strlen(readme_msg));
2257 static const struct file_operations tracing_readme_fops = {
2258 .open = tracing_open_generic,
2259 .read = tracing_readme_read,
2263 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2264 size_t cnt, loff_t *ppos)
2273 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2277 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2285 for (i = 0; i < SAVED_CMDLINES; i++) {
2288 pid = map_cmdline_to_pid[i];
2289 if (pid == -1 || pid == NO_CMDLINE_MAP)
2292 trace_find_cmdline(pid, buf_comm);
2293 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2298 len = simple_read_from_buffer(ubuf, cnt, ppos,
2307 static const struct file_operations tracing_saved_cmdlines_fops = {
2308 .open = tracing_open_generic,
2309 .read = tracing_saved_cmdlines_read,
2313 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2314 size_t cnt, loff_t *ppos)
2319 r = sprintf(buf, "%u\n", tracer_enabled);
2320 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2324 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2325 size_t cnt, loff_t *ppos)
2327 struct trace_array *tr = filp->private_data;
2332 if (cnt >= sizeof(buf))
2335 if (copy_from_user(&buf, ubuf, cnt))
2340 ret = strict_strtoul(buf, 10, &val);
2346 mutex_lock(&trace_types_lock);
2347 if (tracer_enabled ^ val) {
2350 if (current_trace->start)
2351 current_trace->start(tr);
2356 if (current_trace->stop)
2357 current_trace->stop(tr);
2360 mutex_unlock(&trace_types_lock);
2368 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2369 size_t cnt, loff_t *ppos)
2371 char buf[max_tracer_type_len+2];
2374 mutex_lock(&trace_types_lock);
2376 r = sprintf(buf, "%s\n", current_trace->name);
2378 r = sprintf(buf, "\n");
2379 mutex_unlock(&trace_types_lock);
2381 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2384 int tracer_init(struct tracer *t, struct trace_array *tr)
2386 tracing_reset_online_cpus(tr);
2390 static int tracing_resize_ring_buffer(unsigned long size)
2395 * If kernel or user changes the size of the ring buffer
2396 * we use the size that was given, and we can forget about
2397 * expanding it later.
2399 ring_buffer_expanded = 1;
2401 ret = ring_buffer_resize(global_trace.buffer, size);
2405 ret = ring_buffer_resize(max_tr.buffer, size);
2409 r = ring_buffer_resize(global_trace.buffer,
2410 global_trace.entries);
2413 * AARGH! We are left with different
2414 * size max buffer!!!!
2415 * The max buffer is our "snapshot" buffer.
2416 * When a tracer needs a snapshot (one of the
2417 * latency tracers), it swaps the max buffer
2418 * with the saved snap shot. We succeeded to
2419 * update the size of the main buffer, but failed to
2420 * update the size of the max buffer. But when we tried
2421 * to reset the main buffer to the original size, we
2422 * failed there too. This is very unlikely to
2423 * happen, but if it does, warn and kill all
2427 tracing_disabled = 1;
2432 global_trace.entries = size;
2438 * tracing_update_buffers - used by tracing facility to expand ring buffers
2440 * To save on memory when the tracing is never used on a system with it
2441 * configured in. The ring buffers are set to a minimum size. But once
2442 * a user starts to use the tracing facility, then they need to grow
2443 * to their default size.
2445 * This function is to be called when a tracer is about to be used.
2447 int tracing_update_buffers(void)
2451 mutex_lock(&trace_types_lock);
2452 if (!ring_buffer_expanded)
2453 ret = tracing_resize_ring_buffer(trace_buf_size);
2454 mutex_unlock(&trace_types_lock);
2459 struct trace_option_dentry;
2461 static struct trace_option_dentry *
2462 create_trace_option_files(struct tracer *tracer);
2465 destroy_trace_option_files(struct trace_option_dentry *topts);
2467 static int tracing_set_tracer(const char *buf)
2469 static struct trace_option_dentry *topts;
2470 struct trace_array *tr = &global_trace;
2474 mutex_lock(&trace_types_lock);
2476 if (!ring_buffer_expanded) {
2477 ret = tracing_resize_ring_buffer(trace_buf_size);
2483 for (t = trace_types; t; t = t->next) {
2484 if (strcmp(t->name, buf) == 0)
2491 if (t == current_trace)
2494 trace_branch_disable();
2495 if (current_trace && current_trace->reset)
2496 current_trace->reset(tr);
2498 destroy_trace_option_files(topts);
2502 topts = create_trace_option_files(current_trace);
2505 ret = tracer_init(t, tr);
2510 trace_branch_enable(tr);
2512 mutex_unlock(&trace_types_lock);
2518 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2519 size_t cnt, loff_t *ppos)
2521 char buf[max_tracer_type_len+1];
2528 if (cnt > max_tracer_type_len)
2529 cnt = max_tracer_type_len;
2531 if (copy_from_user(&buf, ubuf, cnt))
2536 /* strip ending whitespace. */
2537 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2540 err = tracing_set_tracer(buf);
2550 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2551 size_t cnt, loff_t *ppos)
2553 unsigned long *ptr = filp->private_data;
2557 r = snprintf(buf, sizeof(buf), "%ld\n",
2558 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2559 if (r > sizeof(buf))
2561 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2565 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2566 size_t cnt, loff_t *ppos)
2568 unsigned long *ptr = filp->private_data;
2573 if (cnt >= sizeof(buf))
2576 if (copy_from_user(&buf, ubuf, cnt))
2581 ret = strict_strtoul(buf, 10, &val);
2590 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2592 long cpu_file = (long) inode->i_private;
2593 struct trace_iterator *iter;
2596 if (tracing_disabled)
2599 mutex_lock(&trace_types_lock);
2601 /* We only allow one reader per cpu */
2602 if (cpu_file == TRACE_PIPE_ALL_CPU) {
2603 if (!cpumask_empty(tracing_reader_cpumask)) {
2607 cpumask_setall(tracing_reader_cpumask);
2609 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2610 cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2617 /* create a buffer to store the information to pass to userspace */
2618 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2625 * We make a copy of the current tracer to avoid concurrent
2626 * changes on it while we are reading.
2628 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2634 *iter->trace = *current_trace;
2636 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2641 /* trace pipe does not show start of buffer */
2642 cpumask_setall(iter->started);
2644 if (trace_flags & TRACE_ITER_LATENCY_FMT)
2645 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2647 iter->cpu_file = cpu_file;
2648 iter->tr = &global_trace;
2649 mutex_init(&iter->mutex);
2650 filp->private_data = iter;
2652 if (iter->trace->pipe_open)
2653 iter->trace->pipe_open(iter);
2656 mutex_unlock(&trace_types_lock);
2662 mutex_unlock(&trace_types_lock);
2666 static int tracing_release_pipe(struct inode *inode, struct file *file)
2668 struct trace_iterator *iter = file->private_data;
2670 mutex_lock(&trace_types_lock);
2672 if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2673 cpumask_clear(tracing_reader_cpumask);
2675 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2677 mutex_unlock(&trace_types_lock);
2679 free_cpumask_var(iter->started);
2680 mutex_destroy(&iter->mutex);
2688 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2690 struct trace_iterator *iter = filp->private_data;
2692 if (trace_flags & TRACE_ITER_BLOCK) {
2694 * Always select as readable when in blocking mode
2696 return POLLIN | POLLRDNORM;
2698 if (!trace_empty(iter))
2699 return POLLIN | POLLRDNORM;
2700 poll_wait(filp, &trace_wait, poll_table);
2701 if (!trace_empty(iter))
2702 return POLLIN | POLLRDNORM;
2709 void default_wait_pipe(struct trace_iterator *iter)
2713 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2715 if (trace_empty(iter))
2718 finish_wait(&trace_wait, &wait);
2722 * This is a make-shift waitqueue.
2723 * A tracer might use this callback on some rare cases:
2725 * 1) the current tracer might hold the runqueue lock when it wakes up
2726 * a reader, hence a deadlock (sched, function, and function graph tracers)
2727 * 2) the function tracers, trace all functions, we don't want
2728 * the overhead of calling wake_up and friends
2729 * (and tracing them too)
2731 * Anyway, this is really very primitive wakeup.
2733 void poll_wait_pipe(struct trace_iterator *iter)
2735 set_current_state(TASK_INTERRUPTIBLE);
2736 /* sleep for 100 msecs, and try again. */
2737 schedule_timeout(HZ / 10);
2740 /* Must be called with trace_types_lock mutex held. */
2741 static int tracing_wait_pipe(struct file *filp)
2743 struct trace_iterator *iter = filp->private_data;
2745 while (trace_empty(iter)) {
2747 if ((filp->f_flags & O_NONBLOCK)) {
2751 mutex_unlock(&iter->mutex);
2753 iter->trace->wait_pipe(iter);
2755 mutex_lock(&iter->mutex);
2757 if (signal_pending(current))
2761 * We block until we read something and tracing is disabled.
2762 * We still block if tracing is disabled, but we have never
2763 * read anything. This allows a user to cat this file, and
2764 * then enable tracing. But after we have read something,
2765 * we give an EOF when tracing is again disabled.
2767 * iter->pos will be 0 if we haven't read anything.
2769 if (!tracer_enabled && iter->pos)
2780 tracing_read_pipe(struct file *filp, char __user *ubuf,
2781 size_t cnt, loff_t *ppos)
2783 struct trace_iterator *iter = filp->private_data;
2784 static struct tracer *old_tracer;
2787 /* return any leftover data */
2788 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2792 trace_seq_init(&iter->seq);
2794 /* copy the tracer to avoid using a global lock all around */
2795 mutex_lock(&trace_types_lock);
2796 if (unlikely(old_tracer != current_trace && current_trace)) {
2797 old_tracer = current_trace;
2798 *iter->trace = *current_trace;
2800 mutex_unlock(&trace_types_lock);
2803 * Avoid more than one consumer on a single file descriptor
2804 * This is just a matter of traces coherency, the ring buffer itself
2807 mutex_lock(&iter->mutex);
2808 if (iter->trace->read) {
2809 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2815 sret = tracing_wait_pipe(filp);
2819 /* stop when tracing is finished */
2820 if (trace_empty(iter)) {
2825 if (cnt >= PAGE_SIZE)
2826 cnt = PAGE_SIZE - 1;
2828 /* reset all but tr, trace, and overruns */
2829 memset(&iter->seq, 0,
2830 sizeof(struct trace_iterator) -
2831 offsetof(struct trace_iterator, seq));
2834 trace_event_read_lock();
2835 while (find_next_entry_inc(iter) != NULL) {
2836 enum print_line_t ret;
2837 int len = iter->seq.len;
2839 ret = print_trace_line(iter);
2840 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2841 /* don't print partial lines */
2842 iter->seq.len = len;
2845 if (ret != TRACE_TYPE_NO_CONSUME)
2846 trace_consume(iter);
2848 if (iter->seq.len >= cnt)
2851 trace_event_read_unlock();
2853 /* Now copy what we have to the user */
2854 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2855 if (iter->seq.readpos >= iter->seq.len)
2856 trace_seq_init(&iter->seq);
2859 * If there was nothing to send to user, inspite of consuming trace
2860 * entries, go back to wait for more entries.
2866 mutex_unlock(&iter->mutex);
2871 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
2872 struct pipe_buffer *buf)
2874 __free_page(buf->page);
2877 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
2880 __free_page(spd->pages[idx]);
2883 static struct pipe_buf_operations tracing_pipe_buf_ops = {
2885 .map = generic_pipe_buf_map,
2886 .unmap = generic_pipe_buf_unmap,
2887 .confirm = generic_pipe_buf_confirm,
2888 .release = tracing_pipe_buf_release,
2889 .steal = generic_pipe_buf_steal,
2890 .get = generic_pipe_buf_get,
2894 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
2899 /* Seq buffer is page-sized, exactly what we need. */
2901 count = iter->seq.len;
2902 ret = print_trace_line(iter);
2903 count = iter->seq.len - count;
2906 iter->seq.len -= count;
2909 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2910 iter->seq.len -= count;
2914 trace_consume(iter);
2916 if (!find_next_entry_inc(iter)) {
2926 static ssize_t tracing_splice_read_pipe(struct file *filp,
2928 struct pipe_inode_info *pipe,
2932 struct page *pages[PIPE_BUFFERS];
2933 struct partial_page partial[PIPE_BUFFERS];
2934 struct trace_iterator *iter = filp->private_data;
2935 struct splice_pipe_desc spd = {
2938 .nr_pages = 0, /* This gets updated below. */
2940 .ops = &tracing_pipe_buf_ops,
2941 .spd_release = tracing_spd_release_pipe,
2943 static struct tracer *old_tracer;
2948 /* copy the tracer to avoid using a global lock all around */
2949 mutex_lock(&trace_types_lock);
2950 if (unlikely(old_tracer != current_trace && current_trace)) {
2951 old_tracer = current_trace;
2952 *iter->trace = *current_trace;
2954 mutex_unlock(&trace_types_lock);
2956 mutex_lock(&iter->mutex);
2958 if (iter->trace->splice_read) {
2959 ret = iter->trace->splice_read(iter, filp,
2960 ppos, pipe, len, flags);
2965 ret = tracing_wait_pipe(filp);
2969 if (!iter->ent && !find_next_entry_inc(iter)) {
2974 trace_event_read_lock();
2976 /* Fill as many pages as possible. */
2977 for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
2978 pages[i] = alloc_page(GFP_KERNEL);
2982 rem = tracing_fill_pipe_page(rem, iter);
2984 /* Copy the data into the page, so we can start over. */
2985 ret = trace_seq_to_buffer(&iter->seq,
2986 page_address(pages[i]),
2989 __free_page(pages[i]);
2992 partial[i].offset = 0;
2993 partial[i].len = iter->seq.len;
2995 trace_seq_init(&iter->seq);
2998 trace_event_read_unlock();
2999 mutex_unlock(&iter->mutex);
3003 return splice_to_pipe(pipe, &spd);
3006 mutex_unlock(&iter->mutex);
3012 tracing_entries_read(struct file *filp, char __user *ubuf,
3013 size_t cnt, loff_t *ppos)
3015 struct trace_array *tr = filp->private_data;
3019 mutex_lock(&trace_types_lock);
3020 if (!ring_buffer_expanded)
3021 r = sprintf(buf, "%lu (expanded: %lu)\n",
3023 trace_buf_size >> 10);
3025 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3026 mutex_unlock(&trace_types_lock);
3028 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3032 tracing_entries_write(struct file *filp, const char __user *ubuf,
3033 size_t cnt, loff_t *ppos)
3039 if (cnt >= sizeof(buf))
3042 if (copy_from_user(&buf, ubuf, cnt))
3047 ret = strict_strtoul(buf, 10, &val);
3051 /* must have at least 1 entry */
3055 mutex_lock(&trace_types_lock);
3059 /* disable all cpu buffers */
3060 for_each_tracing_cpu(cpu) {
3061 if (global_trace.data[cpu])
3062 atomic_inc(&global_trace.data[cpu]->disabled);
3063 if (max_tr.data[cpu])
3064 atomic_inc(&max_tr.data[cpu]->disabled);
3067 /* value is in KB */
3070 if (val != global_trace.entries) {
3071 ret = tracing_resize_ring_buffer(val);
3080 /* If check pages failed, return ENOMEM */
3081 if (tracing_disabled)
3084 for_each_tracing_cpu(cpu) {
3085 if (global_trace.data[cpu])
3086 atomic_dec(&global_trace.data[cpu]->disabled);
3087 if (max_tr.data[cpu])
3088 atomic_dec(&max_tr.data[cpu]->disabled);
3092 max_tr.entries = global_trace.entries;
3093 mutex_unlock(&trace_types_lock);
3098 static int mark_printk(const char *fmt, ...)
3102 va_start(args, fmt);
3103 ret = trace_vprintk(0, fmt, args);
3109 tracing_mark_write(struct file *filp, const char __user *ubuf,
3110 size_t cnt, loff_t *fpos)
3115 if (tracing_disabled)
3118 if (cnt > TRACE_BUF_SIZE)
3119 cnt = TRACE_BUF_SIZE;
3121 buf = kmalloc(cnt + 1, GFP_KERNEL);
3125 if (copy_from_user(buf, ubuf, cnt)) {
3130 /* Cut from the first nil or newline. */
3132 end = strchr(buf, '\n');
3136 cnt = mark_printk("%s\n", buf);
3143 static const struct file_operations tracing_max_lat_fops = {
3144 .open = tracing_open_generic,
3145 .read = tracing_max_lat_read,
3146 .write = tracing_max_lat_write,
3149 static const struct file_operations tracing_ctrl_fops = {
3150 .open = tracing_open_generic,
3151 .read = tracing_ctrl_read,
3152 .write = tracing_ctrl_write,
3155 static const struct file_operations set_tracer_fops = {
3156 .open = tracing_open_generic,
3157 .read = tracing_set_trace_read,
3158 .write = tracing_set_trace_write,
3161 static const struct file_operations tracing_pipe_fops = {
3162 .open = tracing_open_pipe,
3163 .poll = tracing_poll_pipe,
3164 .read = tracing_read_pipe,
3165 .splice_read = tracing_splice_read_pipe,
3166 .release = tracing_release_pipe,
3169 static const struct file_operations tracing_entries_fops = {
3170 .open = tracing_open_generic,
3171 .read = tracing_entries_read,
3172 .write = tracing_entries_write,
3175 static const struct file_operations tracing_mark_fops = {
3176 .open = tracing_open_generic,
3177 .write = tracing_mark_write,
3180 struct ftrace_buffer_info {
3181 struct trace_array *tr;
3187 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3189 int cpu = (int)(long)inode->i_private;
3190 struct ftrace_buffer_info *info;
3192 if (tracing_disabled)
3195 info = kzalloc(sizeof(*info), GFP_KERNEL);
3199 info->tr = &global_trace;
3202 /* Force reading ring buffer for first read */
3203 info->read = (unsigned int)-1;
3205 filp->private_data = info;
3207 return nonseekable_open(inode, filp);
3211 tracing_buffers_read(struct file *filp, char __user *ubuf,
3212 size_t count, loff_t *ppos)
3214 struct ftrace_buffer_info *info = filp->private_data;
3223 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3227 /* Do we have previous read data to read? */
3228 if (info->read < PAGE_SIZE)
3233 ret = ring_buffer_read_page(info->tr->buffer,
3240 pos = ring_buffer_page_len(info->spare);
3242 if (pos < PAGE_SIZE)
3243 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3246 size = PAGE_SIZE - info->read;
3250 ret = copy_to_user(ubuf, info->spare + info->read, size);
3261 static int tracing_buffers_release(struct inode *inode, struct file *file)
3263 struct ftrace_buffer_info *info = file->private_data;
3266 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3273 struct ring_buffer *buffer;
3278 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3279 struct pipe_buffer *buf)
3281 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3286 ring_buffer_free_read_page(ref->buffer, ref->page);
3291 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3292 struct pipe_buffer *buf)
3297 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3298 struct pipe_buffer *buf)
3300 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3305 /* Pipe buffer operations for a buffer. */
3306 static struct pipe_buf_operations buffer_pipe_buf_ops = {
3308 .map = generic_pipe_buf_map,
3309 .unmap = generic_pipe_buf_unmap,
3310 .confirm = generic_pipe_buf_confirm,
3311 .release = buffer_pipe_buf_release,
3312 .steal = buffer_pipe_buf_steal,
3313 .get = buffer_pipe_buf_get,
3317 * Callback from splice_to_pipe(), if we need to release some pages
3318 * at the end of the spd in case we error'ed out in filling the pipe.
3320 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3322 struct buffer_ref *ref =
3323 (struct buffer_ref *)spd->partial[i].private;
3328 ring_buffer_free_read_page(ref->buffer, ref->page);
3330 spd->partial[i].private = 0;
3334 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3335 struct pipe_inode_info *pipe, size_t len,
3338 struct ftrace_buffer_info *info = file->private_data;
3339 struct partial_page partial[PIPE_BUFFERS];
3340 struct page *pages[PIPE_BUFFERS];
3341 struct splice_pipe_desc spd = {
3345 .ops = &buffer_pipe_buf_ops,
3346 .spd_release = buffer_spd_release,
3348 struct buffer_ref *ref;
3349 int entries, size, i;
3352 if (*ppos & (PAGE_SIZE - 1)) {
3353 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3357 if (len & (PAGE_SIZE - 1)) {
3358 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3359 if (len < PAGE_SIZE)
3364 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3366 for (i = 0; i < PIPE_BUFFERS && len && entries; i++, len -= PAGE_SIZE) {
3370 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3375 ref->buffer = info->tr->buffer;
3376 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3382 r = ring_buffer_read_page(ref->buffer, &ref->page,
3385 ring_buffer_free_read_page(ref->buffer,
3392 * zero out any left over data, this is going to
3395 size = ring_buffer_page_len(ref->page);
3396 if (size < PAGE_SIZE)
3397 memset(ref->page + size, 0, PAGE_SIZE - size);
3399 page = virt_to_page(ref->page);
3401 spd.pages[i] = page;
3402 spd.partial[i].len = PAGE_SIZE;
3403 spd.partial[i].offset = 0;
3404 spd.partial[i].private = (unsigned long)ref;
3408 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3413 /* did we read anything? */
3414 if (!spd.nr_pages) {
3415 if (flags & SPLICE_F_NONBLOCK)
3423 ret = splice_to_pipe(pipe, &spd);
3428 static const struct file_operations tracing_buffers_fops = {
3429 .open = tracing_buffers_open,
3430 .read = tracing_buffers_read,
3431 .release = tracing_buffers_release,
3432 .splice_read = tracing_buffers_splice_read,
3433 .llseek = no_llseek,
3437 tracing_stats_read(struct file *filp, char __user *ubuf,
3438 size_t count, loff_t *ppos)
3440 unsigned long cpu = (unsigned long)filp->private_data;
3441 struct trace_array *tr = &global_trace;
3442 struct trace_seq *s;
3445 s = kmalloc(sizeof(*s), GFP_KERNEL);
3451 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3452 trace_seq_printf(s, "entries: %ld\n", cnt);
3454 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3455 trace_seq_printf(s, "overrun: %ld\n", cnt);
3457 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3458 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3460 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3467 static const struct file_operations tracing_stats_fops = {
3468 .open = tracing_open_generic,
3469 .read = tracing_stats_read,
3472 #ifdef CONFIG_DYNAMIC_FTRACE
3474 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3480 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3481 size_t cnt, loff_t *ppos)
3483 static char ftrace_dyn_info_buffer[1024];
3484 static DEFINE_MUTEX(dyn_info_mutex);
3485 unsigned long *p = filp->private_data;
3486 char *buf = ftrace_dyn_info_buffer;
3487 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3490 mutex_lock(&dyn_info_mutex);
3491 r = sprintf(buf, "%ld ", *p);
3493 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3496 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3498 mutex_unlock(&dyn_info_mutex);
3503 static const struct file_operations tracing_dyn_info_fops = {
3504 .open = tracing_open_generic,
3505 .read = tracing_read_dyn_info,
3509 static struct dentry *d_tracer;
3511 struct dentry *tracing_init_dentry(void)
3518 if (!debugfs_initialized())
3521 d_tracer = debugfs_create_dir("tracing", NULL);
3523 if (!d_tracer && !once) {
3525 pr_warning("Could not create debugfs directory 'tracing'\n");
3532 static struct dentry *d_percpu;
3534 struct dentry *tracing_dentry_percpu(void)
3537 struct dentry *d_tracer;
3542 d_tracer = tracing_init_dentry();
3547 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3549 if (!d_percpu && !once) {
3551 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3558 static void tracing_init_debugfs_percpu(long cpu)
3560 struct dentry *d_percpu = tracing_dentry_percpu();
3561 struct dentry *d_cpu;
3562 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3565 if (cpu > 999 || cpu < 0)
3568 sprintf(cpu_dir, "cpu%ld", cpu);
3569 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3571 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3575 /* per cpu trace_pipe */
3576 trace_create_file("trace_pipe", 0444, d_cpu,
3577 (void *) cpu, &tracing_pipe_fops);
3580 trace_create_file("trace", 0644, d_cpu,
3581 (void *) cpu, &tracing_fops);
3583 trace_create_file("trace_pipe_raw", 0444, d_cpu,
3584 (void *) cpu, &tracing_buffers_fops);
3586 trace_create_file("stats", 0444, d_cpu,
3587 (void *) cpu, &tracing_stats_fops);
3590 #ifdef CONFIG_FTRACE_SELFTEST
3591 /* Let selftest have access to static functions in this file */
3592 #include "trace_selftest.c"
3595 struct trace_option_dentry {
3596 struct tracer_opt *opt;
3597 struct tracer_flags *flags;
3598 struct dentry *entry;
3602 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3605 struct trace_option_dentry *topt = filp->private_data;
3608 if (topt->flags->val & topt->opt->bit)
3613 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3617 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3620 struct trace_option_dentry *topt = filp->private_data;
3625 if (cnt >= sizeof(buf))
3628 if (copy_from_user(&buf, ubuf, cnt))
3633 ret = strict_strtoul(buf, 10, &val);
3640 /* do nothing if already cleared */
3641 if (!(topt->flags->val & topt->opt->bit))
3644 mutex_lock(&trace_types_lock);
3645 if (current_trace->set_flag)
3646 ret = current_trace->set_flag(topt->flags->val,
3648 mutex_unlock(&trace_types_lock);
3651 topt->flags->val &= ~topt->opt->bit;
3654 /* do nothing if already set */
3655 if (topt->flags->val & topt->opt->bit)
3658 mutex_lock(&trace_types_lock);
3659 if (current_trace->set_flag)
3660 ret = current_trace->set_flag(topt->flags->val,
3662 mutex_unlock(&trace_types_lock);
3665 topt->flags->val |= topt->opt->bit;
3678 static const struct file_operations trace_options_fops = {
3679 .open = tracing_open_generic,
3680 .read = trace_options_read,
3681 .write = trace_options_write,
3685 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
3688 long index = (long)filp->private_data;
3691 if (trace_flags & (1 << index))
3696 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3700 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
3703 long index = (long)filp->private_data;
3708 if (cnt >= sizeof(buf))
3711 if (copy_from_user(&buf, ubuf, cnt))
3716 ret = strict_strtoul(buf, 10, &val);
3722 trace_flags &= ~(1 << index);
3725 trace_flags |= 1 << index;
3737 static const struct file_operations trace_options_core_fops = {
3738 .open = tracing_open_generic,
3739 .read = trace_options_core_read,
3740 .write = trace_options_core_write,
3743 struct dentry *trace_create_file(const char *name,
3745 struct dentry *parent,
3747 const struct file_operations *fops)
3751 ret = debugfs_create_file(name, mode, parent, data, fops);
3753 pr_warning("Could not create debugfs '%s' entry\n", name);
3759 static struct dentry *trace_options_init_dentry(void)
3761 struct dentry *d_tracer;
3762 static struct dentry *t_options;
3767 d_tracer = tracing_init_dentry();
3771 t_options = debugfs_create_dir("options", d_tracer);
3773 pr_warning("Could not create debugfs directory 'options'\n");
3781 create_trace_option_file(struct trace_option_dentry *topt,
3782 struct tracer_flags *flags,
3783 struct tracer_opt *opt)
3785 struct dentry *t_options;
3787 t_options = trace_options_init_dentry();
3791 topt->flags = flags;
3794 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
3795 &trace_options_fops);
3799 static struct trace_option_dentry *
3800 create_trace_option_files(struct tracer *tracer)
3802 struct trace_option_dentry *topts;
3803 struct tracer_flags *flags;
3804 struct tracer_opt *opts;
3810 flags = tracer->flags;
3812 if (!flags || !flags->opts)
3817 for (cnt = 0; opts[cnt].name; cnt++)
3820 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
3824 for (cnt = 0; opts[cnt].name; cnt++)
3825 create_trace_option_file(&topts[cnt], flags,
3832 destroy_trace_option_files(struct trace_option_dentry *topts)
3839 for (cnt = 0; topts[cnt].opt; cnt++) {
3840 if (topts[cnt].entry)
3841 debugfs_remove(topts[cnt].entry);
3847 static struct dentry *
3848 create_trace_option_core_file(const char *option, long index)
3850 struct dentry *t_options;
3852 t_options = trace_options_init_dentry();
3856 return trace_create_file(option, 0644, t_options, (void *)index,
3857 &trace_options_core_fops);
3860 static __init void create_trace_options_dir(void)
3862 struct dentry *t_options;
3865 t_options = trace_options_init_dentry();
3869 for (i = 0; trace_options[i]; i++)
3870 create_trace_option_core_file(trace_options[i], i);
3873 static __init int tracer_init_debugfs(void)
3875 struct dentry *d_tracer;
3878 d_tracer = tracing_init_dentry();
3880 trace_create_file("tracing_enabled", 0644, d_tracer,
3881 &global_trace, &tracing_ctrl_fops);
3883 trace_create_file("trace_options", 0644, d_tracer,
3884 NULL, &tracing_iter_fops);
3886 trace_create_file("tracing_cpumask", 0644, d_tracer,
3887 NULL, &tracing_cpumask_fops);
3889 trace_create_file("trace", 0644, d_tracer,
3890 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
3892 trace_create_file("available_tracers", 0444, d_tracer,
3893 &global_trace, &show_traces_fops);
3895 trace_create_file("current_tracer", 0644, d_tracer,
3896 &global_trace, &set_tracer_fops);
3898 trace_create_file("tracing_max_latency", 0644, d_tracer,
3899 &tracing_max_latency, &tracing_max_lat_fops);
3901 trace_create_file("tracing_thresh", 0644, d_tracer,
3902 &tracing_thresh, &tracing_max_lat_fops);
3904 trace_create_file("README", 0444, d_tracer,
3905 NULL, &tracing_readme_fops);
3907 trace_create_file("trace_pipe", 0444, d_tracer,
3908 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
3910 trace_create_file("buffer_size_kb", 0644, d_tracer,
3911 &global_trace, &tracing_entries_fops);
3913 trace_create_file("trace_marker", 0220, d_tracer,
3914 NULL, &tracing_mark_fops);
3916 trace_create_file("saved_cmdlines", 0444, d_tracer,
3917 NULL, &tracing_saved_cmdlines_fops);
3919 #ifdef CONFIG_DYNAMIC_FTRACE
3920 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3921 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
3923 #ifdef CONFIG_SYSPROF_TRACER
3924 init_tracer_sysprof_debugfs(d_tracer);
3927 create_trace_options_dir();
3929 for_each_tracing_cpu(cpu)
3930 tracing_init_debugfs_percpu(cpu);
3935 static int trace_panic_handler(struct notifier_block *this,
3936 unsigned long event, void *unused)
3938 if (ftrace_dump_on_oops)
3943 static struct notifier_block trace_panic_notifier = {
3944 .notifier_call = trace_panic_handler,
3946 .priority = 150 /* priority: INT_MAX >= x >= 0 */
3949 static int trace_die_handler(struct notifier_block *self,
3955 if (ftrace_dump_on_oops)
3964 static struct notifier_block trace_die_notifier = {
3965 .notifier_call = trace_die_handler,
3970 * printk is set to max of 1024, we really don't need it that big.
3971 * Nothing should be printing 1000 characters anyway.
3973 #define TRACE_MAX_PRINT 1000
3976 * Define here KERN_TRACE so that we have one place to modify
3977 * it if we decide to change what log level the ftrace dump
3980 #define KERN_TRACE KERN_EMERG
3983 trace_printk_seq(struct trace_seq *s)
3985 /* Probably should print a warning here. */
3989 /* should be zero ended, but we are paranoid. */
3990 s->buffer[s->len] = 0;
3992 printk(KERN_TRACE "%s", s->buffer);
3997 static void __ftrace_dump(bool disable_tracing)
3999 static raw_spinlock_t ftrace_dump_lock =
4000 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
4001 /* use static because iter can be a bit big for the stack */
4002 static struct trace_iterator iter;
4003 unsigned int old_userobj;
4004 static int dump_ran;
4005 unsigned long flags;
4009 local_irq_save(flags);
4010 __raw_spin_lock(&ftrace_dump_lock);
4018 if (disable_tracing)
4021 for_each_tracing_cpu(cpu) {
4022 atomic_inc(&global_trace.data[cpu]->disabled);
4025 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4027 /* don't look at user memory in panic mode */
4028 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4030 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4032 /* Simulate the iterator */
4033 iter.tr = &global_trace;
4034 iter.trace = current_trace;
4035 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4038 * We need to stop all tracing on all CPUS to read the
4039 * the next buffer. This is a bit expensive, but is
4040 * not done often. We fill all what we can read,
4041 * and then release the locks again.
4044 while (!trace_empty(&iter)) {
4047 printk(KERN_TRACE "---------------------------------\n");
4051 /* reset all but tr, trace, and overruns */
4052 memset(&iter.seq, 0,
4053 sizeof(struct trace_iterator) -
4054 offsetof(struct trace_iterator, seq));
4055 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4058 if (find_next_entry_inc(&iter) != NULL) {
4059 print_trace_line(&iter);
4060 trace_consume(&iter);
4063 trace_printk_seq(&iter.seq);
4067 printk(KERN_TRACE " (ftrace buffer empty)\n");
4069 printk(KERN_TRACE "---------------------------------\n");
4071 /* Re-enable tracing if requested */
4072 if (!disable_tracing) {
4073 trace_flags |= old_userobj;
4075 for_each_tracing_cpu(cpu) {
4076 atomic_dec(&global_trace.data[cpu]->disabled);
4082 __raw_spin_unlock(&ftrace_dump_lock);
4083 local_irq_restore(flags);
4086 /* By default: disable tracing after the dump */
4087 void ftrace_dump(void)
4089 __ftrace_dump(true);
4092 __init static int tracer_alloc_buffers(void)
4098 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4101 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4102 goto out_free_buffer_mask;
4104 if (!alloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
4105 goto out_free_tracing_cpumask;
4107 /* To save memory, keep the ring buffer size to its minimum */
4108 if (ring_buffer_expanded)
4109 ring_buf_size = trace_buf_size;
4113 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4114 cpumask_copy(tracing_cpumask, cpu_all_mask);
4115 cpumask_clear(tracing_reader_cpumask);
4117 /* TODO: make the number of buffers hot pluggable with CPUS */
4118 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4119 TRACE_BUFFER_FLAGS);
4120 if (!global_trace.buffer) {
4121 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4123 goto out_free_cpumask;
4125 global_trace.entries = ring_buffer_size(global_trace.buffer);
4128 #ifdef CONFIG_TRACER_MAX_TRACE
4129 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4130 TRACE_BUFFER_FLAGS);
4131 if (!max_tr.buffer) {
4132 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4134 ring_buffer_free(global_trace.buffer);
4135 goto out_free_cpumask;
4137 max_tr.entries = ring_buffer_size(max_tr.buffer);
4138 WARN_ON(max_tr.entries != global_trace.entries);
4141 /* Allocate the first page for all buffers */
4142 for_each_tracing_cpu(i) {
4143 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4144 max_tr.data[i] = &per_cpu(max_data, i);
4147 trace_init_cmdlines();
4149 register_tracer(&nop_trace);
4150 current_trace = &nop_trace;
4151 #ifdef CONFIG_BOOT_TRACER
4152 register_tracer(&boot_tracer);
4154 /* All seems OK, enable tracing */
4155 tracing_disabled = 0;
4157 atomic_notifier_chain_register(&panic_notifier_list,
4158 &trace_panic_notifier);
4160 register_die_notifier(&trace_die_notifier);
4165 free_cpumask_var(tracing_reader_cpumask);
4166 out_free_tracing_cpumask:
4167 free_cpumask_var(tracing_cpumask);
4168 out_free_buffer_mask:
4169 free_cpumask_var(tracing_buffer_mask);
4174 __init static int clear_boot_tracer(void)
4177 * The default tracer at boot buffer is an init section.
4178 * This function is called in lateinit. If we did not
4179 * find the boot tracer, then clear it out, to prevent
4180 * later registration from accessing the buffer that is
4181 * about to be freed.
4183 if (!default_bootup_tracer)
4186 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4187 default_bootup_tracer);
4188 default_bootup_tracer = NULL;
4193 early_initcall(tracer_alloc_buffers);
4194 fs_initcall(tracer_init_debugfs);
4195 late_initcall(clear_boot_tracer);