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/ctype.h>
34 #include <linux/init.h>
35 #include <linux/poll.h>
36 #include <linux/gfp.h>
40 #include "trace_output.h"
42 #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
44 unsigned long __read_mostly tracing_max_latency;
45 unsigned long __read_mostly tracing_thresh;
48 * On boot up, the ring buffer is set to the minimum size, so that
49 * we do not waste memory on systems that are not using tracing.
51 static int ring_buffer_expanded;
54 * We need to change this state when a selftest is running.
55 * A selftest will lurk into the ring-buffer to count the
56 * entries inserted during the selftest although some concurrent
57 * insertions into the ring-buffer such as trace_printk could occurred
58 * at the same time, giving false positive or negative results.
60 static bool __read_mostly tracing_selftest_running;
63 * If a tracer is running, we do not want to run SELFTEST.
65 static bool __read_mostly tracing_selftest_disabled;
67 /* For tracers that don't implement custom flags */
68 static struct tracer_opt dummy_tracer_opt[] = {
72 static struct tracer_flags dummy_tracer_flags = {
74 .opts = dummy_tracer_opt
77 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
83 * Kill all tracing for good (never come back).
84 * It is initialized to 1 but will turn to zero if the initialization
85 * of the tracer is successful. But that is the only place that sets
88 static int tracing_disabled = 1;
90 static DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
92 static inline void ftrace_disable_cpu(void)
95 local_inc(&__get_cpu_var(ftrace_cpu_disabled));
98 static inline void ftrace_enable_cpu(void)
100 local_dec(&__get_cpu_var(ftrace_cpu_disabled));
104 static cpumask_var_t __read_mostly tracing_buffer_mask;
106 /* Define which cpu buffers are currently read in trace_pipe */
107 static cpumask_var_t tracing_reader_cpumask;
109 #define for_each_tracing_cpu(cpu) \
110 for_each_cpu(cpu, tracing_buffer_mask)
113 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
115 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
116 * is set, then ftrace_dump is called. This will output the contents
117 * of the ftrace buffers to the console. This is very useful for
118 * capturing traces that lead to crashes and outputing it to a
121 * It is default off, but you can enable it with either specifying
122 * "ftrace_dump_on_oops" in the kernel command line, or setting
123 * /proc/sys/kernel/ftrace_dump_on_oops to true.
125 int ftrace_dump_on_oops;
127 static int tracing_set_tracer(const char *buf);
129 #define BOOTUP_TRACER_SIZE 100
130 static char bootup_tracer_buf[BOOTUP_TRACER_SIZE] __initdata;
131 static char *default_bootup_tracer;
133 static int __init set_ftrace(char *str)
135 strncpy(bootup_tracer_buf, str, BOOTUP_TRACER_SIZE);
136 default_bootup_tracer = bootup_tracer_buf;
137 /* We are using ftrace early, expand it */
138 ring_buffer_expanded = 1;
141 __setup("ftrace=", set_ftrace);
143 static int __init set_ftrace_dump_on_oops(char *str)
145 ftrace_dump_on_oops = 1;
148 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
150 unsigned long long ns2usecs(cycle_t nsec)
158 * The global_trace is the descriptor that holds the tracing
159 * buffers for the live tracing. For each CPU, it contains
160 * a link list of pages that will store trace entries. The
161 * page descriptor of the pages in the memory is used to hold
162 * the link list by linking the lru item in the page descriptor
163 * to each of the pages in the buffer per CPU.
165 * For each active CPU there is a data field that holds the
166 * pages for the buffer for that CPU. Each CPU has the same number
167 * of pages allocated for its buffer.
169 static struct trace_array global_trace;
171 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
173 cycle_t ftrace_now(int cpu)
177 /* Early boot up does not have a buffer yet */
178 if (!global_trace.buffer)
179 return trace_clock_local();
181 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
182 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
188 * The max_tr is used to snapshot the global_trace when a maximum
189 * latency is reached. Some tracers will use this to store a maximum
190 * trace while it continues examining live traces.
192 * The buffers for the max_tr are set up the same as the global_trace.
193 * When a snapshot is taken, the link list of the max_tr is swapped
194 * with the link list of the global_trace and the buffers are reset for
195 * the global_trace so the tracing can continue.
197 static struct trace_array max_tr;
199 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
201 /* tracer_enabled is used to toggle activation of a tracer */
202 static int tracer_enabled = 1;
205 * tracing_is_enabled - return tracer_enabled status
207 * This function is used by other tracers to know the status
208 * of the tracer_enabled flag. Tracers may use this function
209 * to know if it should enable their features when starting
210 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
212 int tracing_is_enabled(void)
214 return tracer_enabled;
218 * trace_buf_size is the size in bytes that is allocated
219 * for a buffer. Note, the number of bytes is always rounded
222 * This number is purposely set to a low number of 16384.
223 * If the dump on oops happens, it will be much appreciated
224 * to not have to wait for all that output. Anyway this can be
225 * boot time and run time configurable.
227 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
229 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
231 /* trace_types holds a link list of available tracers. */
232 static struct tracer *trace_types __read_mostly;
234 /* current_trace points to the tracer that is currently active */
235 static struct tracer *current_trace __read_mostly;
238 * max_tracer_type_len is used to simplify the allocating of
239 * buffers to read userspace tracer names. We keep track of
240 * the longest tracer name registered.
242 static int max_tracer_type_len;
245 * trace_types_lock is used to protect the trace_types list.
246 * This lock is also used to keep user access serialized.
247 * Accesses from userspace will grab this lock while userspace
248 * activities happen inside the kernel.
250 static DEFINE_MUTEX(trace_types_lock);
252 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
253 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
255 /* trace_flags holds trace_options default values */
256 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
257 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME;
260 * trace_wake_up - wake up tasks waiting for trace input
262 * Simply wakes up any task that is blocked on the trace_wait
263 * queue. These is used with trace_poll for tasks polling the trace.
265 void trace_wake_up(void)
268 * The runqueue_is_locked() can fail, but this is the best we
271 if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
272 wake_up(&trace_wait);
275 static int __init set_buf_size(char *str)
277 unsigned long buf_size;
282 ret = strict_strtoul(str, 0, &buf_size);
283 /* nr_entries can not be zero */
284 if (ret < 0 || buf_size == 0)
286 trace_buf_size = buf_size;
289 __setup("trace_buf_size=", set_buf_size);
291 unsigned long nsecs_to_usecs(unsigned long nsecs)
296 /* These must match the bit postions in trace_iterator_flags */
297 static const char *trace_options[] = {
323 * ftrace_max_lock is used to protect the swapping of buffers
324 * when taking a max snapshot. The buffers themselves are
325 * protected by per_cpu spinlocks. But the action of the swap
326 * needs its own lock.
328 * This is defined as a raw_spinlock_t in order to help
329 * with performance when lockdep debugging is enabled.
331 static raw_spinlock_t ftrace_max_lock =
332 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
335 * Copy the new maximum trace into the separate maximum-trace
336 * structure. (this way the maximum trace is permanently saved,
337 * for later retrieval via /debugfs/tracing/latency_trace)
340 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
342 struct trace_array_cpu *data = tr->data[cpu];
345 max_tr.time_start = data->preempt_timestamp;
347 data = max_tr.data[cpu];
348 data->saved_latency = tracing_max_latency;
350 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
351 data->pid = tsk->pid;
352 data->uid = task_uid(tsk);
353 data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
354 data->policy = tsk->policy;
355 data->rt_priority = tsk->rt_priority;
357 /* record this tasks comm */
358 tracing_record_cmdline(tsk);
361 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
369 if (s->len <= s->readpos)
372 len = s->len - s->readpos;
375 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
385 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
390 if (s->len <= s->readpos)
393 len = s->len - s->readpos;
396 ret = memcpy(buf, s->buffer + s->readpos, cnt);
405 trace_print_seq(struct seq_file *m, struct trace_seq *s)
407 int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
410 seq_puts(m, s->buffer);
416 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
418 * @tsk: the task with the latency
419 * @cpu: The cpu that initiated the trace.
421 * Flip the buffers between the @tr and the max_tr and record information
422 * about which task was the cause of this latency.
425 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
427 struct ring_buffer *buf = tr->buffer;
429 WARN_ON_ONCE(!irqs_disabled());
430 __raw_spin_lock(&ftrace_max_lock);
432 tr->buffer = max_tr.buffer;
435 ftrace_disable_cpu();
436 ring_buffer_reset(tr->buffer);
439 __update_max_tr(tr, tsk, cpu);
440 __raw_spin_unlock(&ftrace_max_lock);
444 * update_max_tr_single - only copy one trace over, and reset the rest
446 * @tsk - task with the latency
447 * @cpu - the cpu of the buffer to copy.
449 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
452 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
456 WARN_ON_ONCE(!irqs_disabled());
457 __raw_spin_lock(&ftrace_max_lock);
459 ftrace_disable_cpu();
461 ring_buffer_reset(max_tr.buffer);
462 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
466 WARN_ON_ONCE(ret && ret != -EAGAIN);
468 __update_max_tr(tr, tsk, cpu);
469 __raw_spin_unlock(&ftrace_max_lock);
473 * register_tracer - register a tracer with the ftrace system.
474 * @type - the plugin for the tracer
476 * Register a new plugin tracer.
478 int register_tracer(struct tracer *type)
479 __releases(kernel_lock)
480 __acquires(kernel_lock)
487 pr_info("Tracer must have a name\n");
492 * When this gets called we hold the BKL which means that
493 * preemption is disabled. Various trace selftests however
494 * need to disable and enable preemption for successful tests.
495 * So we drop the BKL here and grab it after the tests again.
498 mutex_lock(&trace_types_lock);
500 tracing_selftest_running = true;
502 for (t = trace_types; t; t = t->next) {
503 if (strcmp(type->name, t->name) == 0) {
505 pr_info("Trace %s already registered\n",
513 type->set_flag = &dummy_set_flag;
515 type->flags = &dummy_tracer_flags;
517 if (!type->flags->opts)
518 type->flags->opts = dummy_tracer_opt;
519 if (!type->wait_pipe)
520 type->wait_pipe = default_wait_pipe;
523 #ifdef CONFIG_FTRACE_STARTUP_TEST
524 if (type->selftest && !tracing_selftest_disabled) {
525 struct tracer *saved_tracer = current_trace;
526 struct trace_array *tr = &global_trace;
530 * Run a selftest on this tracer.
531 * Here we reset the trace buffer, and set the current
532 * tracer to be this tracer. The tracer can then run some
533 * internal tracing to verify that everything is in order.
534 * If we fail, we do not register this tracer.
536 for_each_tracing_cpu(i)
537 tracing_reset(tr, i);
539 current_trace = type;
540 /* the test is responsible for initializing and enabling */
541 pr_info("Testing tracer %s: ", type->name);
542 ret = type->selftest(type, tr);
543 /* the test is responsible for resetting too */
544 current_trace = saved_tracer;
546 printk(KERN_CONT "FAILED!\n");
549 /* Only reset on passing, to avoid touching corrupted buffers */
550 for_each_tracing_cpu(i)
551 tracing_reset(tr, i);
553 printk(KERN_CONT "PASSED\n");
557 type->next = trace_types;
559 len = strlen(type->name);
560 if (len > max_tracer_type_len)
561 max_tracer_type_len = len;
564 tracing_selftest_running = false;
565 mutex_unlock(&trace_types_lock);
567 if (ret || !default_bootup_tracer)
570 if (strncmp(default_bootup_tracer, type->name, BOOTUP_TRACER_SIZE))
573 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
574 /* Do we want this tracer to start on bootup? */
575 tracing_set_tracer(type->name);
576 default_bootup_tracer = NULL;
577 /* disable other selftests, since this will break it. */
578 tracing_selftest_disabled = 1;
579 #ifdef CONFIG_FTRACE_STARTUP_TEST
580 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
589 void unregister_tracer(struct tracer *type)
594 mutex_lock(&trace_types_lock);
595 for (t = &trace_types; *t; t = &(*t)->next) {
599 pr_info("Trace %s not registered\n", type->name);
605 if (type == current_trace && tracer_enabled) {
608 if (current_trace->stop)
609 current_trace->stop(&global_trace);
610 current_trace = &nop_trace;
613 if (strlen(type->name) != max_tracer_type_len)
616 max_tracer_type_len = 0;
617 for (t = &trace_types; *t; t = &(*t)->next) {
618 len = strlen((*t)->name);
619 if (len > max_tracer_type_len)
620 max_tracer_type_len = len;
623 mutex_unlock(&trace_types_lock);
626 void tracing_reset(struct trace_array *tr, int cpu)
628 ftrace_disable_cpu();
629 ring_buffer_reset_cpu(tr->buffer, cpu);
633 void tracing_reset_online_cpus(struct trace_array *tr)
637 tr->time_start = ftrace_now(tr->cpu);
639 for_each_online_cpu(cpu)
640 tracing_reset(tr, cpu);
643 #define SAVED_CMDLINES 128
644 #define NO_CMDLINE_MAP UINT_MAX
645 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
646 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
647 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
648 static int cmdline_idx;
649 static raw_spinlock_t trace_cmdline_lock = __RAW_SPIN_LOCK_UNLOCKED;
651 /* temporary disable recording */
652 static atomic_t trace_record_cmdline_disabled __read_mostly;
654 static void trace_init_cmdlines(void)
656 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
657 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
661 static int trace_stop_count;
662 static DEFINE_SPINLOCK(tracing_start_lock);
665 * ftrace_off_permanent - disable all ftrace code permanently
667 * This should only be called when a serious anomally has
668 * been detected. This will turn off the function tracing,
669 * ring buffers, and other tracing utilites. It takes no
670 * locks and can be called from any context.
672 void ftrace_off_permanent(void)
674 tracing_disabled = 1;
676 tracing_off_permanent();
680 * tracing_start - quick start of the tracer
682 * If tracing is enabled but was stopped by tracing_stop,
683 * this will start the tracer back up.
685 void tracing_start(void)
687 struct ring_buffer *buffer;
690 if (tracing_disabled)
693 spin_lock_irqsave(&tracing_start_lock, flags);
694 if (--trace_stop_count) {
695 if (trace_stop_count < 0) {
696 /* Someone screwed up their debugging */
698 trace_stop_count = 0;
704 buffer = global_trace.buffer;
706 ring_buffer_record_enable(buffer);
708 buffer = max_tr.buffer;
710 ring_buffer_record_enable(buffer);
714 spin_unlock_irqrestore(&tracing_start_lock, flags);
718 * tracing_stop - quick stop of the tracer
720 * Light weight way to stop tracing. Use in conjunction with
723 void tracing_stop(void)
725 struct ring_buffer *buffer;
729 spin_lock_irqsave(&tracing_start_lock, flags);
730 if (trace_stop_count++)
733 buffer = global_trace.buffer;
735 ring_buffer_record_disable(buffer);
737 buffer = max_tr.buffer;
739 ring_buffer_record_disable(buffer);
742 spin_unlock_irqrestore(&tracing_start_lock, flags);
745 void trace_stop_cmdline_recording(void);
747 static void trace_save_cmdline(struct task_struct *tsk)
751 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
755 * It's not the end of the world if we don't get
756 * the lock, but we also don't want to spin
757 * nor do we want to disable interrupts,
758 * so if we miss here, then better luck next time.
760 if (!__raw_spin_trylock(&trace_cmdline_lock))
763 idx = map_pid_to_cmdline[tsk->pid];
764 if (idx == NO_CMDLINE_MAP) {
765 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
768 * Check whether the cmdline buffer at idx has a pid
769 * mapped. We are going to overwrite that entry so we
770 * need to clear the map_pid_to_cmdline. Otherwise we
771 * would read the new comm for the old pid.
773 pid = map_cmdline_to_pid[idx];
774 if (pid != NO_CMDLINE_MAP)
775 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
777 map_cmdline_to_pid[idx] = tsk->pid;
778 map_pid_to_cmdline[tsk->pid] = idx;
783 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
785 __raw_spin_unlock(&trace_cmdline_lock);
788 void trace_find_cmdline(int pid, char comm[])
793 strcpy(comm, "<idle>");
797 if (pid > PID_MAX_DEFAULT) {
798 strcpy(comm, "<...>");
802 __raw_spin_lock(&trace_cmdline_lock);
803 map = map_pid_to_cmdline[pid];
804 if (map != NO_CMDLINE_MAP)
805 strcpy(comm, saved_cmdlines[map]);
807 strcpy(comm, "<...>");
809 __raw_spin_unlock(&trace_cmdline_lock);
812 void tracing_record_cmdline(struct task_struct *tsk)
814 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
818 trace_save_cmdline(tsk);
822 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
825 struct task_struct *tsk = current;
827 entry->preempt_count = pc & 0xff;
828 entry->pid = (tsk) ? tsk->pid : 0;
829 entry->tgid = (tsk) ? tsk->tgid : 0;
831 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
832 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
834 TRACE_FLAG_IRQS_NOSUPPORT |
836 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
837 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
838 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
841 struct ring_buffer_event *trace_buffer_lock_reserve(struct trace_array *tr,
844 unsigned long flags, int pc)
846 struct ring_buffer_event *event;
848 event = ring_buffer_lock_reserve(tr->buffer, len);
850 struct trace_entry *ent = ring_buffer_event_data(event);
852 tracing_generic_entry_update(ent, flags, pc);
858 static void ftrace_trace_stack(struct trace_array *tr,
859 unsigned long flags, int skip, int pc);
860 static void ftrace_trace_userstack(struct trace_array *tr,
861 unsigned long flags, int pc);
863 static inline void __trace_buffer_unlock_commit(struct trace_array *tr,
864 struct ring_buffer_event *event,
865 unsigned long flags, int pc,
868 ring_buffer_unlock_commit(tr->buffer, event);
870 ftrace_trace_stack(tr, flags, 6, pc);
871 ftrace_trace_userstack(tr, flags, pc);
877 void trace_buffer_unlock_commit(struct trace_array *tr,
878 struct ring_buffer_event *event,
879 unsigned long flags, int pc)
881 __trace_buffer_unlock_commit(tr, event, flags, pc, 1);
884 struct ring_buffer_event *
885 trace_current_buffer_lock_reserve(unsigned char type, unsigned long len,
886 unsigned long flags, int pc)
888 return trace_buffer_lock_reserve(&global_trace,
889 type, len, flags, pc);
892 void trace_current_buffer_unlock_commit(struct ring_buffer_event *event,
893 unsigned long flags, int pc)
895 return __trace_buffer_unlock_commit(&global_trace, event, flags, pc, 1);
898 void trace_nowake_buffer_unlock_commit(struct ring_buffer_event *event,
899 unsigned long flags, int pc)
901 return __trace_buffer_unlock_commit(&global_trace, event, flags, pc, 0);
905 trace_function(struct trace_array *tr,
906 unsigned long ip, unsigned long parent_ip, unsigned long flags,
909 struct ring_buffer_event *event;
910 struct ftrace_entry *entry;
912 /* If we are reading the ring buffer, don't trace */
913 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
916 event = trace_buffer_lock_reserve(tr, TRACE_FN, sizeof(*entry),
920 entry = ring_buffer_event_data(event);
922 entry->parent_ip = parent_ip;
923 ring_buffer_unlock_commit(tr->buffer, event);
926 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
927 static int __trace_graph_entry(struct trace_array *tr,
928 struct ftrace_graph_ent *trace,
932 struct ring_buffer_event *event;
933 struct ftrace_graph_ent_entry *entry;
935 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
938 event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_ENT,
939 sizeof(*entry), flags, pc);
942 entry = ring_buffer_event_data(event);
943 entry->graph_ent = *trace;
944 ring_buffer_unlock_commit(global_trace.buffer, event);
949 static void __trace_graph_return(struct trace_array *tr,
950 struct ftrace_graph_ret *trace,
954 struct ring_buffer_event *event;
955 struct ftrace_graph_ret_entry *entry;
957 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
960 event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_RET,
961 sizeof(*entry), flags, pc);
964 entry = ring_buffer_event_data(event);
966 ring_buffer_unlock_commit(global_trace.buffer, event);
971 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
972 unsigned long ip, unsigned long parent_ip, unsigned long flags,
975 if (likely(!atomic_read(&data->disabled)))
976 trace_function(tr, ip, parent_ip, flags, pc);
979 static void __ftrace_trace_stack(struct trace_array *tr,
983 #ifdef CONFIG_STACKTRACE
984 struct ring_buffer_event *event;
985 struct stack_entry *entry;
986 struct stack_trace trace;
988 event = trace_buffer_lock_reserve(tr, TRACE_STACK,
989 sizeof(*entry), flags, pc);
992 entry = ring_buffer_event_data(event);
993 memset(&entry->caller, 0, sizeof(entry->caller));
995 trace.nr_entries = 0;
996 trace.max_entries = FTRACE_STACK_ENTRIES;
998 trace.entries = entry->caller;
1000 save_stack_trace(&trace);
1001 ring_buffer_unlock_commit(tr->buffer, event);
1005 static void ftrace_trace_stack(struct trace_array *tr,
1006 unsigned long flags,
1009 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1012 __ftrace_trace_stack(tr, flags, skip, pc);
1015 void __trace_stack(struct trace_array *tr,
1016 unsigned long flags,
1019 __ftrace_trace_stack(tr, flags, skip, pc);
1022 static void ftrace_trace_userstack(struct trace_array *tr,
1023 unsigned long flags, int pc)
1025 #ifdef CONFIG_STACKTRACE
1026 struct ring_buffer_event *event;
1027 struct userstack_entry *entry;
1028 struct stack_trace trace;
1030 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1033 event = trace_buffer_lock_reserve(tr, TRACE_USER_STACK,
1034 sizeof(*entry), flags, pc);
1037 entry = ring_buffer_event_data(event);
1039 memset(&entry->caller, 0, sizeof(entry->caller));
1041 trace.nr_entries = 0;
1042 trace.max_entries = FTRACE_STACK_ENTRIES;
1044 trace.entries = entry->caller;
1046 save_stack_trace_user(&trace);
1047 ring_buffer_unlock_commit(tr->buffer, event);
1052 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1054 ftrace_trace_userstack(tr, flags, preempt_count());
1059 ftrace_trace_special(void *__tr,
1060 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1063 struct ring_buffer_event *event;
1064 struct trace_array *tr = __tr;
1065 struct special_entry *entry;
1067 event = trace_buffer_lock_reserve(tr, TRACE_SPECIAL,
1068 sizeof(*entry), 0, pc);
1071 entry = ring_buffer_event_data(event);
1075 trace_buffer_unlock_commit(tr, event, 0, pc);
1079 __trace_special(void *__tr, void *__data,
1080 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1082 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1086 tracing_sched_switch_trace(struct trace_array *tr,
1087 struct task_struct *prev,
1088 struct task_struct *next,
1089 unsigned long flags, int pc)
1091 struct ring_buffer_event *event;
1092 struct ctx_switch_entry *entry;
1094 event = trace_buffer_lock_reserve(tr, TRACE_CTX,
1095 sizeof(*entry), flags, pc);
1098 entry = ring_buffer_event_data(event);
1099 entry->prev_pid = prev->pid;
1100 entry->prev_prio = prev->prio;
1101 entry->prev_state = prev->state;
1102 entry->next_pid = next->pid;
1103 entry->next_prio = next->prio;
1104 entry->next_state = next->state;
1105 entry->next_cpu = task_cpu(next);
1106 trace_buffer_unlock_commit(tr, event, flags, pc);
1110 tracing_sched_wakeup_trace(struct trace_array *tr,
1111 struct task_struct *wakee,
1112 struct task_struct *curr,
1113 unsigned long flags, int pc)
1115 struct ring_buffer_event *event;
1116 struct ctx_switch_entry *entry;
1118 event = trace_buffer_lock_reserve(tr, TRACE_WAKE,
1119 sizeof(*entry), flags, pc);
1122 entry = ring_buffer_event_data(event);
1123 entry->prev_pid = curr->pid;
1124 entry->prev_prio = curr->prio;
1125 entry->prev_state = curr->state;
1126 entry->next_pid = wakee->pid;
1127 entry->next_prio = wakee->prio;
1128 entry->next_state = wakee->state;
1129 entry->next_cpu = task_cpu(wakee);
1131 ring_buffer_unlock_commit(tr->buffer, event);
1132 ftrace_trace_stack(tr, flags, 6, pc);
1133 ftrace_trace_userstack(tr, flags, pc);
1137 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1139 struct trace_array *tr = &global_trace;
1140 struct trace_array_cpu *data;
1141 unsigned long flags;
1145 if (tracing_disabled)
1148 pc = preempt_count();
1149 local_irq_save(flags);
1150 cpu = raw_smp_processor_id();
1151 data = tr->data[cpu];
1153 if (likely(atomic_inc_return(&data->disabled) == 1))
1154 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1156 atomic_dec(&data->disabled);
1157 local_irq_restore(flags);
1160 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1161 int trace_graph_entry(struct ftrace_graph_ent *trace)
1163 struct trace_array *tr = &global_trace;
1164 struct trace_array_cpu *data;
1165 unsigned long flags;
1171 if (!ftrace_trace_task(current))
1174 if (!ftrace_graph_addr(trace->func))
1177 local_irq_save(flags);
1178 cpu = raw_smp_processor_id();
1179 data = tr->data[cpu];
1180 disabled = atomic_inc_return(&data->disabled);
1181 if (likely(disabled == 1)) {
1182 pc = preempt_count();
1183 ret = __trace_graph_entry(tr, trace, flags, pc);
1187 /* Only do the atomic if it is not already set */
1188 if (!test_tsk_trace_graph(current))
1189 set_tsk_trace_graph(current);
1191 atomic_dec(&data->disabled);
1192 local_irq_restore(flags);
1197 void trace_graph_return(struct ftrace_graph_ret *trace)
1199 struct trace_array *tr = &global_trace;
1200 struct trace_array_cpu *data;
1201 unsigned long flags;
1206 local_irq_save(flags);
1207 cpu = raw_smp_processor_id();
1208 data = tr->data[cpu];
1209 disabled = atomic_inc_return(&data->disabled);
1210 if (likely(disabled == 1)) {
1211 pc = preempt_count();
1212 __trace_graph_return(tr, trace, flags, pc);
1215 clear_tsk_trace_graph(current);
1216 atomic_dec(&data->disabled);
1217 local_irq_restore(flags);
1219 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1223 * trace_vbprintk - write binary msg to tracing buffer
1226 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1228 static raw_spinlock_t trace_buf_lock =
1229 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
1230 static u32 trace_buf[TRACE_BUF_SIZE];
1232 struct ring_buffer_event *event;
1233 struct trace_array *tr = &global_trace;
1234 struct trace_array_cpu *data;
1235 struct bprint_entry *entry;
1236 unsigned long flags;
1238 int cpu, len = 0, size, pc;
1240 if (unlikely(tracing_selftest_running || tracing_disabled))
1243 /* Don't pollute graph traces with trace_vprintk internals */
1244 pause_graph_tracing();
1246 pc = preempt_count();
1247 resched = ftrace_preempt_disable();
1248 cpu = raw_smp_processor_id();
1249 data = tr->data[cpu];
1251 if (unlikely(atomic_read(&data->disabled)))
1254 /* Lockdep uses trace_printk for lock tracing */
1255 local_irq_save(flags);
1256 __raw_spin_lock(&trace_buf_lock);
1257 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1259 if (len > TRACE_BUF_SIZE || len < 0)
1262 size = sizeof(*entry) + sizeof(u32) * len;
1263 event = trace_buffer_lock_reserve(tr, TRACE_BPRINT, size, flags, pc);
1266 entry = ring_buffer_event_data(event);
1270 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1271 ring_buffer_unlock_commit(tr->buffer, event);
1274 __raw_spin_unlock(&trace_buf_lock);
1275 local_irq_restore(flags);
1278 ftrace_preempt_enable(resched);
1279 unpause_graph_tracing();
1283 EXPORT_SYMBOL_GPL(trace_vbprintk);
1285 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1287 static raw_spinlock_t trace_buf_lock = __RAW_SPIN_LOCK_UNLOCKED;
1288 static char trace_buf[TRACE_BUF_SIZE];
1290 struct ring_buffer_event *event;
1291 struct trace_array *tr = &global_trace;
1292 struct trace_array_cpu *data;
1293 int cpu, len = 0, size, pc;
1294 struct print_entry *entry;
1295 unsigned long irq_flags;
1297 if (tracing_disabled || tracing_selftest_running)
1300 pc = preempt_count();
1301 preempt_disable_notrace();
1302 cpu = raw_smp_processor_id();
1303 data = tr->data[cpu];
1305 if (unlikely(atomic_read(&data->disabled)))
1308 pause_graph_tracing();
1309 raw_local_irq_save(irq_flags);
1310 __raw_spin_lock(&trace_buf_lock);
1311 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1313 len = min(len, TRACE_BUF_SIZE-1);
1316 size = sizeof(*entry) + len + 1;
1317 event = trace_buffer_lock_reserve(tr, TRACE_PRINT, size, irq_flags, pc);
1320 entry = ring_buffer_event_data(event);
1323 memcpy(&entry->buf, trace_buf, len);
1324 entry->buf[len] = 0;
1325 ring_buffer_unlock_commit(tr->buffer, event);
1328 __raw_spin_unlock(&trace_buf_lock);
1329 raw_local_irq_restore(irq_flags);
1330 unpause_graph_tracing();
1332 preempt_enable_notrace();
1336 EXPORT_SYMBOL_GPL(trace_vprintk);
1338 enum trace_file_type {
1339 TRACE_FILE_LAT_FMT = 1,
1340 TRACE_FILE_ANNOTATE = 2,
1343 static void trace_iterator_increment(struct trace_iterator *iter)
1345 /* Don't allow ftrace to trace into the ring buffers */
1346 ftrace_disable_cpu();
1349 if (iter->buffer_iter[iter->cpu])
1350 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1352 ftrace_enable_cpu();
1355 static struct trace_entry *
1356 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1358 struct ring_buffer_event *event;
1359 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1361 /* Don't allow ftrace to trace into the ring buffers */
1362 ftrace_disable_cpu();
1365 event = ring_buffer_iter_peek(buf_iter, ts);
1367 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1369 ftrace_enable_cpu();
1371 return event ? ring_buffer_event_data(event) : NULL;
1374 static struct trace_entry *
1375 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1377 struct ring_buffer *buffer = iter->tr->buffer;
1378 struct trace_entry *ent, *next = NULL;
1379 int cpu_file = iter->cpu_file;
1380 u64 next_ts = 0, ts;
1385 * If we are in a per_cpu trace file, don't bother by iterating over
1386 * all cpu and peek directly.
1388 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1389 if (ring_buffer_empty_cpu(buffer, cpu_file))
1391 ent = peek_next_entry(iter, cpu_file, ent_ts);
1393 *ent_cpu = cpu_file;
1398 for_each_tracing_cpu(cpu) {
1400 if (ring_buffer_empty_cpu(buffer, cpu))
1403 ent = peek_next_entry(iter, cpu, &ts);
1406 * Pick the entry with the smallest timestamp:
1408 if (ent && (!next || ts < next_ts)) {
1416 *ent_cpu = next_cpu;
1424 /* Find the next real entry, without updating the iterator itself */
1425 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1426 int *ent_cpu, u64 *ent_ts)
1428 return __find_next_entry(iter, ent_cpu, ent_ts);
1431 /* Find the next real entry, and increment the iterator to the next entry */
1432 static void *find_next_entry_inc(struct trace_iterator *iter)
1434 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1437 trace_iterator_increment(iter);
1439 return iter->ent ? iter : NULL;
1442 static void trace_consume(struct trace_iterator *iter)
1444 /* Don't allow ftrace to trace into the ring buffers */
1445 ftrace_disable_cpu();
1446 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1447 ftrace_enable_cpu();
1450 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1452 struct trace_iterator *iter = m->private;
1458 /* can't go backwards */
1463 ent = find_next_entry_inc(iter);
1467 while (ent && iter->idx < i)
1468 ent = find_next_entry_inc(iter);
1476 * No necessary locking here. The worst thing which can
1477 * happen is loosing events consumed at the same time
1478 * by a trace_pipe reader.
1479 * Other than that, we don't risk to crash the ring buffer
1480 * because it serializes the readers.
1482 * The current tracer is copied to avoid a global locking
1485 static void *s_start(struct seq_file *m, loff_t *pos)
1487 struct trace_iterator *iter = m->private;
1488 static struct tracer *old_tracer;
1489 int cpu_file = iter->cpu_file;
1494 /* copy the tracer to avoid using a global lock all around */
1495 mutex_lock(&trace_types_lock);
1496 if (unlikely(old_tracer != current_trace && current_trace)) {
1497 old_tracer = current_trace;
1498 *iter->trace = *current_trace;
1500 mutex_unlock(&trace_types_lock);
1502 atomic_inc(&trace_record_cmdline_disabled);
1504 if (*pos != iter->pos) {
1509 ftrace_disable_cpu();
1511 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1512 for_each_tracing_cpu(cpu)
1513 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1515 ring_buffer_iter_reset(iter->buffer_iter[cpu_file]);
1518 ftrace_enable_cpu();
1520 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1525 p = s_next(m, p, &l);
1531 static void s_stop(struct seq_file *m, void *p)
1533 atomic_dec(&trace_record_cmdline_disabled);
1536 static void print_lat_help_header(struct seq_file *m)
1538 seq_puts(m, "# _------=> CPU# \n");
1539 seq_puts(m, "# / _-----=> irqs-off \n");
1540 seq_puts(m, "# | / _----=> need-resched \n");
1541 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1542 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1543 seq_puts(m, "# |||| / \n");
1544 seq_puts(m, "# ||||| delay \n");
1545 seq_puts(m, "# cmd pid ||||| time | caller \n");
1546 seq_puts(m, "# \\ / ||||| \\ | / \n");
1549 static void print_func_help_header(struct seq_file *m)
1551 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1552 seq_puts(m, "# | | | | |\n");
1557 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1559 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1560 struct trace_array *tr = iter->tr;
1561 struct trace_array_cpu *data = tr->data[tr->cpu];
1562 struct tracer *type = current_trace;
1563 unsigned long total;
1564 unsigned long entries;
1565 const char *name = "preemption";
1570 entries = ring_buffer_entries(iter->tr->buffer);
1572 ring_buffer_overruns(iter->tr->buffer);
1574 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1576 seq_puts(m, "# -----------------------------------"
1577 "---------------------------------\n");
1578 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1579 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1580 nsecs_to_usecs(data->saved_latency),
1584 #if defined(CONFIG_PREEMPT_NONE)
1586 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1588 #elif defined(CONFIG_PREEMPT)
1593 /* These are reserved for later use */
1596 seq_printf(m, " #P:%d)\n", num_online_cpus());
1600 seq_puts(m, "# -----------------\n");
1601 seq_printf(m, "# | task: %.16s-%d "
1602 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1603 data->comm, data->pid, data->uid, data->nice,
1604 data->policy, data->rt_priority);
1605 seq_puts(m, "# -----------------\n");
1607 if (data->critical_start) {
1608 seq_puts(m, "# => started at: ");
1609 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1610 trace_print_seq(m, &iter->seq);
1611 seq_puts(m, "\n# => ended at: ");
1612 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1613 trace_print_seq(m, &iter->seq);
1620 static void test_cpu_buff_start(struct trace_iterator *iter)
1622 struct trace_seq *s = &iter->seq;
1624 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1627 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1630 if (cpumask_test_cpu(iter->cpu, iter->started))
1633 cpumask_set_cpu(iter->cpu, iter->started);
1634 trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1637 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1639 struct trace_seq *s = &iter->seq;
1640 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1641 struct trace_entry *entry;
1642 struct trace_event *event;
1646 test_cpu_buff_start(iter);
1648 event = ftrace_find_event(entry->type);
1650 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1651 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1652 if (!trace_print_lat_context(iter))
1655 if (!trace_print_context(iter))
1661 return event->trace(iter, sym_flags);
1663 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1666 return TRACE_TYPE_HANDLED;
1668 return TRACE_TYPE_PARTIAL_LINE;
1671 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1673 struct trace_seq *s = &iter->seq;
1674 struct trace_entry *entry;
1675 struct trace_event *event;
1679 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1680 if (!trace_seq_printf(s, "%d %d %llu ",
1681 entry->pid, iter->cpu, iter->ts))
1685 event = ftrace_find_event(entry->type);
1687 return event->raw(iter, 0);
1689 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1692 return TRACE_TYPE_HANDLED;
1694 return TRACE_TYPE_PARTIAL_LINE;
1697 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1699 struct trace_seq *s = &iter->seq;
1700 unsigned char newline = '\n';
1701 struct trace_entry *entry;
1702 struct trace_event *event;
1706 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1707 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1708 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1709 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1712 event = ftrace_find_event(entry->type);
1714 enum print_line_t ret = event->hex(iter, 0);
1715 if (ret != TRACE_TYPE_HANDLED)
1719 SEQ_PUT_FIELD_RET(s, newline);
1721 return TRACE_TYPE_HANDLED;
1724 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1726 struct trace_seq *s = &iter->seq;
1727 struct trace_entry *entry;
1728 struct trace_event *event;
1732 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1733 SEQ_PUT_FIELD_RET(s, entry->pid);
1734 SEQ_PUT_FIELD_RET(s, iter->cpu);
1735 SEQ_PUT_FIELD_RET(s, iter->ts);
1738 event = ftrace_find_event(entry->type);
1739 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1742 static int trace_empty(struct trace_iterator *iter)
1746 /* If we are looking at one CPU buffer, only check that one */
1747 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1748 cpu = iter->cpu_file;
1749 if (iter->buffer_iter[cpu]) {
1750 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1753 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1759 for_each_tracing_cpu(cpu) {
1760 if (iter->buffer_iter[cpu]) {
1761 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1764 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1772 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1774 enum print_line_t ret;
1776 if (iter->trace && iter->trace->print_line) {
1777 ret = iter->trace->print_line(iter);
1778 if (ret != TRACE_TYPE_UNHANDLED)
1782 if (iter->ent->type == TRACE_BPRINT &&
1783 trace_flags & TRACE_ITER_PRINTK &&
1784 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1785 return trace_print_bprintk_msg_only(iter);
1787 if (iter->ent->type == TRACE_PRINT &&
1788 trace_flags & TRACE_ITER_PRINTK &&
1789 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1790 return trace_print_printk_msg_only(iter);
1792 if (trace_flags & TRACE_ITER_BIN)
1793 return print_bin_fmt(iter);
1795 if (trace_flags & TRACE_ITER_HEX)
1796 return print_hex_fmt(iter);
1798 if (trace_flags & TRACE_ITER_RAW)
1799 return print_raw_fmt(iter);
1801 return print_trace_fmt(iter);
1804 static int s_show(struct seq_file *m, void *v)
1806 struct trace_iterator *iter = v;
1808 if (iter->ent == NULL) {
1810 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1813 if (iter->trace && iter->trace->print_header)
1814 iter->trace->print_header(m);
1815 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1816 /* print nothing if the buffers are empty */
1817 if (trace_empty(iter))
1819 print_trace_header(m, iter);
1820 if (!(trace_flags & TRACE_ITER_VERBOSE))
1821 print_lat_help_header(m);
1823 if (!(trace_flags & TRACE_ITER_VERBOSE))
1824 print_func_help_header(m);
1827 print_trace_line(iter);
1828 trace_print_seq(m, &iter->seq);
1834 static struct seq_operations tracer_seq_ops = {
1841 static struct trace_iterator *
1842 __tracing_open(struct inode *inode, struct file *file)
1844 long cpu_file = (long) inode->i_private;
1845 void *fail_ret = ERR_PTR(-ENOMEM);
1846 struct trace_iterator *iter;
1850 if (tracing_disabled)
1851 return ERR_PTR(-ENODEV);
1853 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1855 return ERR_PTR(-ENOMEM);
1858 * We make a copy of the current tracer to avoid concurrent
1859 * changes on it while we are reading.
1861 mutex_lock(&trace_types_lock);
1862 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
1867 *iter->trace = *current_trace;
1869 if (current_trace && current_trace->print_max)
1872 iter->tr = &global_trace;
1874 mutex_init(&iter->mutex);
1875 iter->cpu_file = cpu_file;
1877 /* Notify the tracer early; before we stop tracing. */
1878 if (iter->trace && iter->trace->open)
1879 iter->trace->open(iter);
1881 /* Annotate start of buffers if we had overruns */
1882 if (ring_buffer_overruns(iter->tr->buffer))
1883 iter->iter_flags |= TRACE_FILE_ANNOTATE;
1885 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
1886 for_each_tracing_cpu(cpu) {
1888 iter->buffer_iter[cpu] =
1889 ring_buffer_read_start(iter->tr->buffer, cpu);
1892 cpu = iter->cpu_file;
1893 iter->buffer_iter[cpu] =
1894 ring_buffer_read_start(iter->tr->buffer, cpu);
1897 /* TODO stop tracer */
1898 ret = seq_open(file, &tracer_seq_ops);
1900 fail_ret = ERR_PTR(ret);
1904 m = file->private_data;
1907 /* stop the trace while dumping */
1910 mutex_unlock(&trace_types_lock);
1915 for_each_tracing_cpu(cpu) {
1916 if (iter->buffer_iter[cpu])
1917 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1920 mutex_unlock(&trace_types_lock);
1927 int tracing_open_generic(struct inode *inode, struct file *filp)
1929 if (tracing_disabled)
1932 filp->private_data = inode->i_private;
1936 static int tracing_release(struct inode *inode, struct file *file)
1938 struct seq_file *m = (struct seq_file *)file->private_data;
1939 struct trace_iterator *iter;
1942 if (!(file->f_mode & FMODE_READ))
1947 mutex_lock(&trace_types_lock);
1948 for_each_tracing_cpu(cpu) {
1949 if (iter->buffer_iter[cpu])
1950 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1953 if (iter->trace && iter->trace->close)
1954 iter->trace->close(iter);
1956 /* reenable tracing if it was previously enabled */
1958 mutex_unlock(&trace_types_lock);
1960 seq_release(inode, file);
1961 mutex_destroy(&iter->mutex);
1967 static int tracing_open(struct inode *inode, struct file *file)
1969 struct trace_iterator *iter;
1972 /* If this file was open for write, then erase contents */
1973 if ((file->f_mode & FMODE_WRITE) &&
1974 !(file->f_flags & O_APPEND)) {
1975 long cpu = (long) inode->i_private;
1977 if (cpu == TRACE_PIPE_ALL_CPU)
1978 tracing_reset_online_cpus(&global_trace);
1980 tracing_reset(&global_trace, cpu);
1983 if (file->f_mode & FMODE_READ) {
1984 iter = __tracing_open(inode, file);
1986 ret = PTR_ERR(iter);
1987 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
1988 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1994 t_next(struct seq_file *m, void *v, loff_t *pos)
1996 struct tracer *t = m->private;
2008 static void *t_start(struct seq_file *m, loff_t *pos)
2010 struct tracer *t = m->private;
2013 mutex_lock(&trace_types_lock);
2014 for (; t && l < *pos; t = t_next(m, t, &l))
2020 static void t_stop(struct seq_file *m, void *p)
2022 mutex_unlock(&trace_types_lock);
2025 static int t_show(struct seq_file *m, void *v)
2027 struct tracer *t = v;
2032 seq_printf(m, "%s", t->name);
2041 static struct seq_operations show_traces_seq_ops = {
2048 static int show_traces_open(struct inode *inode, struct file *file)
2052 if (tracing_disabled)
2055 ret = seq_open(file, &show_traces_seq_ops);
2057 struct seq_file *m = file->private_data;
2058 m->private = trace_types;
2065 tracing_write_stub(struct file *filp, const char __user *ubuf,
2066 size_t count, loff_t *ppos)
2071 static const struct file_operations tracing_fops = {
2072 .open = tracing_open,
2074 .write = tracing_write_stub,
2075 .llseek = seq_lseek,
2076 .release = tracing_release,
2079 static const struct file_operations show_traces_fops = {
2080 .open = show_traces_open,
2082 .release = seq_release,
2086 * Only trace on a CPU if the bitmask is set:
2088 static cpumask_var_t tracing_cpumask;
2091 * The tracer itself will not take this lock, but still we want
2092 * to provide a consistent cpumask to user-space:
2094 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2097 * Temporary storage for the character representation of the
2098 * CPU bitmask (and one more byte for the newline):
2100 static char mask_str[NR_CPUS + 1];
2103 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2104 size_t count, loff_t *ppos)
2108 mutex_lock(&tracing_cpumask_update_lock);
2110 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2111 if (count - len < 2) {
2115 len += sprintf(mask_str + len, "\n");
2116 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2119 mutex_unlock(&tracing_cpumask_update_lock);
2125 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2126 size_t count, loff_t *ppos)
2129 cpumask_var_t tracing_cpumask_new;
2131 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2134 mutex_lock(&tracing_cpumask_update_lock);
2135 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2139 local_irq_disable();
2140 __raw_spin_lock(&ftrace_max_lock);
2141 for_each_tracing_cpu(cpu) {
2143 * Increase/decrease the disabled counter if we are
2144 * about to flip a bit in the cpumask:
2146 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2147 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2148 atomic_inc(&global_trace.data[cpu]->disabled);
2150 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2151 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2152 atomic_dec(&global_trace.data[cpu]->disabled);
2155 __raw_spin_unlock(&ftrace_max_lock);
2158 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2160 mutex_unlock(&tracing_cpumask_update_lock);
2161 free_cpumask_var(tracing_cpumask_new);
2166 mutex_unlock(&tracing_cpumask_update_lock);
2167 free_cpumask_var(tracing_cpumask);
2172 static const struct file_operations tracing_cpumask_fops = {
2173 .open = tracing_open_generic,
2174 .read = tracing_cpumask_read,
2175 .write = tracing_cpumask_write,
2179 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2180 size_t cnt, loff_t *ppos)
2182 struct tracer_opt *trace_opts;
2190 /* calculate max size */
2191 for (i = 0; trace_options[i]; i++) {
2192 len += strlen(trace_options[i]);
2193 len += 3; /* "no" and newline */
2196 mutex_lock(&trace_types_lock);
2197 tracer_flags = current_trace->flags->val;
2198 trace_opts = current_trace->flags->opts;
2201 * Increase the size with names of options specific
2202 * of the current tracer.
2204 for (i = 0; trace_opts[i].name; i++) {
2205 len += strlen(trace_opts[i].name);
2206 len += 3; /* "no" and newline */
2209 /* +2 for \n and \0 */
2210 buf = kmalloc(len + 2, GFP_KERNEL);
2212 mutex_unlock(&trace_types_lock);
2216 for (i = 0; trace_options[i]; i++) {
2217 if (trace_flags & (1 << i))
2218 r += sprintf(buf + r, "%s\n", trace_options[i]);
2220 r += sprintf(buf + r, "no%s\n", trace_options[i]);
2223 for (i = 0; trace_opts[i].name; i++) {
2224 if (tracer_flags & trace_opts[i].bit)
2225 r += sprintf(buf + r, "%s\n",
2226 trace_opts[i].name);
2228 r += sprintf(buf + r, "no%s\n",
2229 trace_opts[i].name);
2231 mutex_unlock(&trace_types_lock);
2233 WARN_ON(r >= len + 2);
2235 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2241 /* Try to assign a tracer specific option */
2242 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2244 struct tracer_flags *trace_flags = trace->flags;
2245 struct tracer_opt *opts = NULL;
2249 for (i = 0; trace_flags->opts[i].name; i++) {
2250 opts = &trace_flags->opts[i];
2251 len = strlen(opts->name);
2253 if (strncmp(cmp, opts->name, len) == 0) {
2254 ret = trace->set_flag(trace_flags->val,
2260 if (!trace_flags->opts[i].name)
2263 /* Refused to handle */
2268 trace_flags->val &= ~opts->bit;
2270 trace_flags->val |= opts->bit;
2275 static void set_tracer_flags(unsigned int mask, int enabled)
2277 /* do nothing if flag is already set */
2278 if (!!(trace_flags & mask) == !!enabled)
2282 trace_flags |= mask;
2284 trace_flags &= ~mask;
2286 if (mask == TRACE_ITER_GLOBAL_CLK) {
2290 func = trace_clock_global;
2292 func = trace_clock_local;
2294 mutex_lock(&trace_types_lock);
2295 ring_buffer_set_clock(global_trace.buffer, func);
2298 ring_buffer_set_clock(max_tr.buffer, func);
2299 mutex_unlock(&trace_types_lock);
2304 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2305 size_t cnt, loff_t *ppos)
2313 if (cnt >= sizeof(buf))
2316 if (copy_from_user(&buf, ubuf, cnt))
2321 if (strncmp(buf, "no", 2) == 0) {
2326 for (i = 0; trace_options[i]; i++) {
2327 int len = strlen(trace_options[i]);
2329 if (strncmp(cmp, trace_options[i], len) == 0) {
2330 set_tracer_flags(1 << i, !neg);
2335 /* If no option could be set, test the specific tracer options */
2336 if (!trace_options[i]) {
2337 mutex_lock(&trace_types_lock);
2338 ret = set_tracer_option(current_trace, cmp, neg);
2339 mutex_unlock(&trace_types_lock);
2349 static const struct file_operations tracing_iter_fops = {
2350 .open = tracing_open_generic,
2351 .read = tracing_trace_options_read,
2352 .write = tracing_trace_options_write,
2355 static const char readme_msg[] =
2356 "tracing mini-HOWTO:\n\n"
2358 "# mount -t debugfs nodev /debug\n\n"
2359 "# cat /debug/tracing/available_tracers\n"
2360 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2361 "# cat /debug/tracing/current_tracer\n"
2363 "# echo sched_switch > /debug/tracing/current_tracer\n"
2364 "# cat /debug/tracing/current_tracer\n"
2366 "# cat /debug/tracing/trace_options\n"
2367 "noprint-parent nosym-offset nosym-addr noverbose\n"
2368 "# echo print-parent > /debug/tracing/trace_options\n"
2369 "# echo 1 > /debug/tracing/tracing_enabled\n"
2370 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2371 "echo 0 > /debug/tracing/tracing_enabled\n"
2375 tracing_readme_read(struct file *filp, char __user *ubuf,
2376 size_t cnt, loff_t *ppos)
2378 return simple_read_from_buffer(ubuf, cnt, ppos,
2379 readme_msg, strlen(readme_msg));
2382 static const struct file_operations tracing_readme_fops = {
2383 .open = tracing_open_generic,
2384 .read = tracing_readme_read,
2388 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2389 size_t cnt, loff_t *ppos)
2394 r = sprintf(buf, "%u\n", tracer_enabled);
2395 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2399 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2400 size_t cnt, loff_t *ppos)
2402 struct trace_array *tr = filp->private_data;
2407 if (cnt >= sizeof(buf))
2410 if (copy_from_user(&buf, ubuf, cnt))
2415 ret = strict_strtoul(buf, 10, &val);
2421 mutex_lock(&trace_types_lock);
2422 if (tracer_enabled ^ val) {
2425 if (current_trace->start)
2426 current_trace->start(tr);
2431 if (current_trace->stop)
2432 current_trace->stop(tr);
2435 mutex_unlock(&trace_types_lock);
2443 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2444 size_t cnt, loff_t *ppos)
2446 char buf[max_tracer_type_len+2];
2449 mutex_lock(&trace_types_lock);
2451 r = sprintf(buf, "%s\n", current_trace->name);
2453 r = sprintf(buf, "\n");
2454 mutex_unlock(&trace_types_lock);
2456 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2459 int tracer_init(struct tracer *t, struct trace_array *tr)
2461 tracing_reset_online_cpus(tr);
2465 static int tracing_resize_ring_buffer(unsigned long size)
2470 * If kernel or user changes the size of the ring buffer
2471 * we use the size that was given, and we can forget about
2472 * expanding it later.
2474 ring_buffer_expanded = 1;
2476 ret = ring_buffer_resize(global_trace.buffer, size);
2480 ret = ring_buffer_resize(max_tr.buffer, size);
2484 r = ring_buffer_resize(global_trace.buffer,
2485 global_trace.entries);
2488 * AARGH! We are left with different
2489 * size max buffer!!!!
2490 * The max buffer is our "snapshot" buffer.
2491 * When a tracer needs a snapshot (one of the
2492 * latency tracers), it swaps the max buffer
2493 * with the saved snap shot. We succeeded to
2494 * update the size of the main buffer, but failed to
2495 * update the size of the max buffer. But when we tried
2496 * to reset the main buffer to the original size, we
2497 * failed there too. This is very unlikely to
2498 * happen, but if it does, warn and kill all
2502 tracing_disabled = 1;
2507 global_trace.entries = size;
2513 * tracing_update_buffers - used by tracing facility to expand ring buffers
2515 * To save on memory when the tracing is never used on a system with it
2516 * configured in. The ring buffers are set to a minimum size. But once
2517 * a user starts to use the tracing facility, then they need to grow
2518 * to their default size.
2520 * This function is to be called when a tracer is about to be used.
2522 int tracing_update_buffers(void)
2526 mutex_lock(&trace_types_lock);
2527 if (!ring_buffer_expanded)
2528 ret = tracing_resize_ring_buffer(trace_buf_size);
2529 mutex_unlock(&trace_types_lock);
2534 struct trace_option_dentry;
2536 static struct trace_option_dentry *
2537 create_trace_option_files(struct tracer *tracer);
2540 destroy_trace_option_files(struct trace_option_dentry *topts);
2542 static int tracing_set_tracer(const char *buf)
2544 static struct trace_option_dentry *topts;
2545 struct trace_array *tr = &global_trace;
2549 mutex_lock(&trace_types_lock);
2551 if (!ring_buffer_expanded) {
2552 ret = tracing_resize_ring_buffer(trace_buf_size);
2558 for (t = trace_types; t; t = t->next) {
2559 if (strcmp(t->name, buf) == 0)
2566 if (t == current_trace)
2569 trace_branch_disable();
2570 if (current_trace && current_trace->reset)
2571 current_trace->reset(tr);
2573 destroy_trace_option_files(topts);
2577 topts = create_trace_option_files(current_trace);
2580 ret = tracer_init(t, tr);
2585 trace_branch_enable(tr);
2587 mutex_unlock(&trace_types_lock);
2593 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2594 size_t cnt, loff_t *ppos)
2596 char buf[max_tracer_type_len+1];
2603 if (cnt > max_tracer_type_len)
2604 cnt = max_tracer_type_len;
2606 if (copy_from_user(&buf, ubuf, cnt))
2611 /* strip ending whitespace. */
2612 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2615 err = tracing_set_tracer(buf);
2625 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2626 size_t cnt, loff_t *ppos)
2628 unsigned long *ptr = filp->private_data;
2632 r = snprintf(buf, sizeof(buf), "%ld\n",
2633 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2634 if (r > sizeof(buf))
2636 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2640 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2641 size_t cnt, loff_t *ppos)
2643 unsigned long *ptr = filp->private_data;
2648 if (cnt >= sizeof(buf))
2651 if (copy_from_user(&buf, ubuf, cnt))
2656 ret = strict_strtoul(buf, 10, &val);
2665 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2667 long cpu_file = (long) inode->i_private;
2668 struct trace_iterator *iter;
2671 if (tracing_disabled)
2674 mutex_lock(&trace_types_lock);
2676 /* We only allow one reader per cpu */
2677 if (cpu_file == TRACE_PIPE_ALL_CPU) {
2678 if (!cpumask_empty(tracing_reader_cpumask)) {
2682 cpumask_setall(tracing_reader_cpumask);
2684 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2685 cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2692 /* create a buffer to store the information to pass to userspace */
2693 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2700 * We make a copy of the current tracer to avoid concurrent
2701 * changes on it while we are reading.
2703 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2709 *iter->trace = *current_trace;
2711 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2716 /* trace pipe does not show start of buffer */
2717 cpumask_setall(iter->started);
2719 iter->cpu_file = cpu_file;
2720 iter->tr = &global_trace;
2721 mutex_init(&iter->mutex);
2722 filp->private_data = iter;
2724 if (iter->trace->pipe_open)
2725 iter->trace->pipe_open(iter);
2728 mutex_unlock(&trace_types_lock);
2734 mutex_unlock(&trace_types_lock);
2738 static int tracing_release_pipe(struct inode *inode, struct file *file)
2740 struct trace_iterator *iter = file->private_data;
2742 mutex_lock(&trace_types_lock);
2744 if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2745 cpumask_clear(tracing_reader_cpumask);
2747 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2749 mutex_unlock(&trace_types_lock);
2751 free_cpumask_var(iter->started);
2752 mutex_destroy(&iter->mutex);
2760 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2762 struct trace_iterator *iter = filp->private_data;
2764 if (trace_flags & TRACE_ITER_BLOCK) {
2766 * Always select as readable when in blocking mode
2768 return POLLIN | POLLRDNORM;
2770 if (!trace_empty(iter))
2771 return POLLIN | POLLRDNORM;
2772 poll_wait(filp, &trace_wait, poll_table);
2773 if (!trace_empty(iter))
2774 return POLLIN | POLLRDNORM;
2781 void default_wait_pipe(struct trace_iterator *iter)
2785 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2787 if (trace_empty(iter))
2790 finish_wait(&trace_wait, &wait);
2794 * This is a make-shift waitqueue.
2795 * A tracer might use this callback on some rare cases:
2797 * 1) the current tracer might hold the runqueue lock when it wakes up
2798 * a reader, hence a deadlock (sched, function, and function graph tracers)
2799 * 2) the function tracers, trace all functions, we don't want
2800 * the overhead of calling wake_up and friends
2801 * (and tracing them too)
2803 * Anyway, this is really very primitive wakeup.
2805 void poll_wait_pipe(struct trace_iterator *iter)
2807 set_current_state(TASK_INTERRUPTIBLE);
2808 /* sleep for 100 msecs, and try again. */
2809 schedule_timeout(HZ / 10);
2812 /* Must be called with trace_types_lock mutex held. */
2813 static int tracing_wait_pipe(struct file *filp)
2815 struct trace_iterator *iter = filp->private_data;
2817 while (trace_empty(iter)) {
2819 if ((filp->f_flags & O_NONBLOCK)) {
2823 mutex_unlock(&iter->mutex);
2825 iter->trace->wait_pipe(iter);
2827 mutex_lock(&iter->mutex);
2829 if (signal_pending(current))
2833 * We block until we read something and tracing is disabled.
2834 * We still block if tracing is disabled, but we have never
2835 * read anything. This allows a user to cat this file, and
2836 * then enable tracing. But after we have read something,
2837 * we give an EOF when tracing is again disabled.
2839 * iter->pos will be 0 if we haven't read anything.
2841 if (!tracer_enabled && iter->pos)
2852 tracing_read_pipe(struct file *filp, char __user *ubuf,
2853 size_t cnt, loff_t *ppos)
2855 struct trace_iterator *iter = filp->private_data;
2856 static struct tracer *old_tracer;
2859 /* return any leftover data */
2860 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2864 trace_seq_init(&iter->seq);
2866 /* copy the tracer to avoid using a global lock all around */
2867 mutex_lock(&trace_types_lock);
2868 if (unlikely(old_tracer != current_trace && current_trace)) {
2869 old_tracer = current_trace;
2870 *iter->trace = *current_trace;
2872 mutex_unlock(&trace_types_lock);
2875 * Avoid more than one consumer on a single file descriptor
2876 * This is just a matter of traces coherency, the ring buffer itself
2879 mutex_lock(&iter->mutex);
2880 if (iter->trace->read) {
2881 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2887 sret = tracing_wait_pipe(filp);
2891 /* stop when tracing is finished */
2892 if (trace_empty(iter)) {
2897 if (cnt >= PAGE_SIZE)
2898 cnt = PAGE_SIZE - 1;
2900 /* reset all but tr, trace, and overruns */
2901 memset(&iter->seq, 0,
2902 sizeof(struct trace_iterator) -
2903 offsetof(struct trace_iterator, seq));
2906 while (find_next_entry_inc(iter) != NULL) {
2907 enum print_line_t ret;
2908 int len = iter->seq.len;
2910 ret = print_trace_line(iter);
2911 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2912 /* don't print partial lines */
2913 iter->seq.len = len;
2916 if (ret != TRACE_TYPE_NO_CONSUME)
2917 trace_consume(iter);
2919 if (iter->seq.len >= cnt)
2923 /* Now copy what we have to the user */
2924 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2925 if (iter->seq.readpos >= iter->seq.len)
2926 trace_seq_init(&iter->seq);
2929 * If there was nothing to send to user, inspite of consuming trace
2930 * entries, go back to wait for more entries.
2936 mutex_unlock(&iter->mutex);
2941 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
2942 struct pipe_buffer *buf)
2944 __free_page(buf->page);
2947 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
2950 __free_page(spd->pages[idx]);
2953 static struct pipe_buf_operations tracing_pipe_buf_ops = {
2955 .map = generic_pipe_buf_map,
2956 .unmap = generic_pipe_buf_unmap,
2957 .confirm = generic_pipe_buf_confirm,
2958 .release = tracing_pipe_buf_release,
2959 .steal = generic_pipe_buf_steal,
2960 .get = generic_pipe_buf_get,
2964 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
2969 /* Seq buffer is page-sized, exactly what we need. */
2971 count = iter->seq.len;
2972 ret = print_trace_line(iter);
2973 count = iter->seq.len - count;
2976 iter->seq.len -= count;
2979 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2980 iter->seq.len -= count;
2984 trace_consume(iter);
2986 if (!find_next_entry_inc(iter)) {
2996 static ssize_t tracing_splice_read_pipe(struct file *filp,
2998 struct pipe_inode_info *pipe,
3002 struct page *pages[PIPE_BUFFERS];
3003 struct partial_page partial[PIPE_BUFFERS];
3004 struct trace_iterator *iter = filp->private_data;
3005 struct splice_pipe_desc spd = {
3008 .nr_pages = 0, /* This gets updated below. */
3010 .ops = &tracing_pipe_buf_ops,
3011 .spd_release = tracing_spd_release_pipe,
3013 static struct tracer *old_tracer;
3018 /* copy the tracer to avoid using a global lock all around */
3019 mutex_lock(&trace_types_lock);
3020 if (unlikely(old_tracer != current_trace && current_trace)) {
3021 old_tracer = current_trace;
3022 *iter->trace = *current_trace;
3024 mutex_unlock(&trace_types_lock);
3026 mutex_lock(&iter->mutex);
3028 if (iter->trace->splice_read) {
3029 ret = iter->trace->splice_read(iter, filp,
3030 ppos, pipe, len, flags);
3035 ret = tracing_wait_pipe(filp);
3039 if (!iter->ent && !find_next_entry_inc(iter)) {
3044 /* Fill as many pages as possible. */
3045 for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3046 pages[i] = alloc_page(GFP_KERNEL);
3050 rem = tracing_fill_pipe_page(rem, iter);
3052 /* Copy the data into the page, so we can start over. */
3053 ret = trace_seq_to_buffer(&iter->seq,
3054 page_address(pages[i]),
3057 __free_page(pages[i]);
3060 partial[i].offset = 0;
3061 partial[i].len = iter->seq.len;
3063 trace_seq_init(&iter->seq);
3066 mutex_unlock(&iter->mutex);
3070 return splice_to_pipe(pipe, &spd);
3073 mutex_unlock(&iter->mutex);
3079 tracing_entries_read(struct file *filp, char __user *ubuf,
3080 size_t cnt, loff_t *ppos)
3082 struct trace_array *tr = filp->private_data;
3086 mutex_lock(&trace_types_lock);
3087 if (!ring_buffer_expanded)
3088 r = sprintf(buf, "%lu (expanded: %lu)\n",
3090 trace_buf_size >> 10);
3092 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3093 mutex_unlock(&trace_types_lock);
3095 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3099 tracing_entries_write(struct file *filp, const char __user *ubuf,
3100 size_t cnt, loff_t *ppos)
3106 if (cnt >= sizeof(buf))
3109 if (copy_from_user(&buf, ubuf, cnt))
3114 ret = strict_strtoul(buf, 10, &val);
3118 /* must have at least 1 entry */
3122 mutex_lock(&trace_types_lock);
3126 /* disable all cpu buffers */
3127 for_each_tracing_cpu(cpu) {
3128 if (global_trace.data[cpu])
3129 atomic_inc(&global_trace.data[cpu]->disabled);
3130 if (max_tr.data[cpu])
3131 atomic_inc(&max_tr.data[cpu]->disabled);
3134 /* value is in KB */
3137 if (val != global_trace.entries) {
3138 ret = tracing_resize_ring_buffer(val);
3147 /* If check pages failed, return ENOMEM */
3148 if (tracing_disabled)
3151 for_each_tracing_cpu(cpu) {
3152 if (global_trace.data[cpu])
3153 atomic_dec(&global_trace.data[cpu]->disabled);
3154 if (max_tr.data[cpu])
3155 atomic_dec(&max_tr.data[cpu]->disabled);
3159 max_tr.entries = global_trace.entries;
3160 mutex_unlock(&trace_types_lock);
3165 static int mark_printk(const char *fmt, ...)
3169 va_start(args, fmt);
3170 ret = trace_vprintk(0, fmt, args);
3176 tracing_mark_write(struct file *filp, const char __user *ubuf,
3177 size_t cnt, loff_t *fpos)
3182 if (tracing_disabled)
3185 if (cnt > TRACE_BUF_SIZE)
3186 cnt = TRACE_BUF_SIZE;
3188 buf = kmalloc(cnt + 1, GFP_KERNEL);
3192 if (copy_from_user(buf, ubuf, cnt)) {
3197 /* Cut from the first nil or newline. */
3199 end = strchr(buf, '\n');
3203 cnt = mark_printk("%s\n", buf);
3210 static const struct file_operations tracing_max_lat_fops = {
3211 .open = tracing_open_generic,
3212 .read = tracing_max_lat_read,
3213 .write = tracing_max_lat_write,
3216 static const struct file_operations tracing_ctrl_fops = {
3217 .open = tracing_open_generic,
3218 .read = tracing_ctrl_read,
3219 .write = tracing_ctrl_write,
3222 static const struct file_operations set_tracer_fops = {
3223 .open = tracing_open_generic,
3224 .read = tracing_set_trace_read,
3225 .write = tracing_set_trace_write,
3228 static const struct file_operations tracing_pipe_fops = {
3229 .open = tracing_open_pipe,
3230 .poll = tracing_poll_pipe,
3231 .read = tracing_read_pipe,
3232 .splice_read = tracing_splice_read_pipe,
3233 .release = tracing_release_pipe,
3236 static const struct file_operations tracing_entries_fops = {
3237 .open = tracing_open_generic,
3238 .read = tracing_entries_read,
3239 .write = tracing_entries_write,
3242 static const struct file_operations tracing_mark_fops = {
3243 .open = tracing_open_generic,
3244 .write = tracing_mark_write,
3247 struct ftrace_buffer_info {
3248 struct trace_array *tr;
3254 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3256 int cpu = (int)(long)inode->i_private;
3257 struct ftrace_buffer_info *info;
3259 if (tracing_disabled)
3262 info = kzalloc(sizeof(*info), GFP_KERNEL);
3266 info->tr = &global_trace;
3268 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3269 /* Force reading ring buffer for first read */
3270 info->read = (unsigned int)-1;
3274 filp->private_data = info;
3284 tracing_buffers_read(struct file *filp, char __user *ubuf,
3285 size_t count, loff_t *ppos)
3287 struct ftrace_buffer_info *info = filp->private_data;
3295 /* Do we have previous read data to read? */
3296 if (info->read < PAGE_SIZE)
3301 ret = ring_buffer_read_page(info->tr->buffer,
3308 pos = ring_buffer_page_len(info->spare);
3310 if (pos < PAGE_SIZE)
3311 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3314 size = PAGE_SIZE - info->read;
3318 ret = copy_to_user(ubuf, info->spare + info->read, size);
3329 static int tracing_buffers_release(struct inode *inode, struct file *file)
3331 struct ftrace_buffer_info *info = file->private_data;
3333 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3340 struct ring_buffer *buffer;
3345 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3346 struct pipe_buffer *buf)
3348 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3353 ring_buffer_free_read_page(ref->buffer, ref->page);
3358 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3359 struct pipe_buffer *buf)
3364 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3365 struct pipe_buffer *buf)
3367 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3372 /* Pipe buffer operations for a buffer. */
3373 static struct pipe_buf_operations buffer_pipe_buf_ops = {
3375 .map = generic_pipe_buf_map,
3376 .unmap = generic_pipe_buf_unmap,
3377 .confirm = generic_pipe_buf_confirm,
3378 .release = buffer_pipe_buf_release,
3379 .steal = buffer_pipe_buf_steal,
3380 .get = buffer_pipe_buf_get,
3384 * Callback from splice_to_pipe(), if we need to release some pages
3385 * at the end of the spd in case we error'ed out in filling the pipe.
3387 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3389 struct buffer_ref *ref =
3390 (struct buffer_ref *)spd->partial[i].private;
3395 ring_buffer_free_read_page(ref->buffer, ref->page);
3397 spd->partial[i].private = 0;
3401 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3402 struct pipe_inode_info *pipe, size_t len,
3405 struct ftrace_buffer_info *info = file->private_data;
3406 struct partial_page partial[PIPE_BUFFERS];
3407 struct page *pages[PIPE_BUFFERS];
3408 struct splice_pipe_desc spd = {
3412 .ops = &buffer_pipe_buf_ops,
3413 .spd_release = buffer_spd_release,
3415 struct buffer_ref *ref;
3420 * We can't seek on a buffer input
3422 if (unlikely(*ppos))
3426 for (i = 0; i < PIPE_BUFFERS && len; i++, len -= size) {
3430 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3434 ref->buffer = info->tr->buffer;
3435 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3441 r = ring_buffer_read_page(ref->buffer, &ref->page,
3444 ring_buffer_free_read_page(ref->buffer,
3451 * zero out any left over data, this is going to
3454 size = ring_buffer_page_len(ref->page);
3455 if (size < PAGE_SIZE)
3456 memset(ref->page + size, 0, PAGE_SIZE - size);
3458 page = virt_to_page(ref->page);
3460 spd.pages[i] = page;
3461 spd.partial[i].len = PAGE_SIZE;
3462 spd.partial[i].offset = 0;
3463 spd.partial[i].private = (unsigned long)ref;
3469 /* did we read anything? */
3470 if (!spd.nr_pages) {
3471 if (flags & SPLICE_F_NONBLOCK)
3479 ret = splice_to_pipe(pipe, &spd);
3484 static const struct file_operations tracing_buffers_fops = {
3485 .open = tracing_buffers_open,
3486 .read = tracing_buffers_read,
3487 .release = tracing_buffers_release,
3488 .splice_read = tracing_buffers_splice_read,
3489 .llseek = no_llseek,
3492 #ifdef CONFIG_DYNAMIC_FTRACE
3494 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3500 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3501 size_t cnt, loff_t *ppos)
3503 static char ftrace_dyn_info_buffer[1024];
3504 static DEFINE_MUTEX(dyn_info_mutex);
3505 unsigned long *p = filp->private_data;
3506 char *buf = ftrace_dyn_info_buffer;
3507 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3510 mutex_lock(&dyn_info_mutex);
3511 r = sprintf(buf, "%ld ", *p);
3513 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3516 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3518 mutex_unlock(&dyn_info_mutex);
3523 static const struct file_operations tracing_dyn_info_fops = {
3524 .open = tracing_open_generic,
3525 .read = tracing_read_dyn_info,
3529 static struct dentry *d_tracer;
3531 struct dentry *tracing_init_dentry(void)
3538 if (!debugfs_initialized())
3541 d_tracer = debugfs_create_dir("tracing", NULL);
3543 if (!d_tracer && !once) {
3545 pr_warning("Could not create debugfs directory 'tracing'\n");
3552 static struct dentry *d_percpu;
3554 struct dentry *tracing_dentry_percpu(void)
3557 struct dentry *d_tracer;
3562 d_tracer = tracing_init_dentry();
3567 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3569 if (!d_percpu && !once) {
3571 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3578 static void tracing_init_debugfs_percpu(long cpu)
3580 struct dentry *d_percpu = tracing_dentry_percpu();
3581 struct dentry *entry, *d_cpu;
3582 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3585 if (cpu > 999 || cpu < 0)
3588 sprintf(cpu_dir, "cpu%ld", cpu);
3589 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3591 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3595 /* per cpu trace_pipe */
3596 entry = debugfs_create_file("trace_pipe", 0444, d_cpu,
3597 (void *) cpu, &tracing_pipe_fops);
3599 pr_warning("Could not create debugfs 'trace_pipe' entry\n");
3602 entry = debugfs_create_file("trace", 0644, d_cpu,
3603 (void *) cpu, &tracing_fops);
3605 pr_warning("Could not create debugfs 'trace' entry\n");
3607 entry = debugfs_create_file("trace_pipe_raw", 0444, d_cpu,
3608 (void *) cpu, &tracing_buffers_fops);
3610 pr_warning("Could not create debugfs 'trace_pipe_raw' entry\n");
3613 #ifdef CONFIG_FTRACE_SELFTEST
3614 /* Let selftest have access to static functions in this file */
3615 #include "trace_selftest.c"
3618 struct trace_option_dentry {
3619 struct tracer_opt *opt;
3620 struct tracer_flags *flags;
3621 struct dentry *entry;
3625 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3628 struct trace_option_dentry *topt = filp->private_data;
3631 if (topt->flags->val & topt->opt->bit)
3636 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3640 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3643 struct trace_option_dentry *topt = filp->private_data;
3648 if (cnt >= sizeof(buf))
3651 if (copy_from_user(&buf, ubuf, cnt))
3656 ret = strict_strtoul(buf, 10, &val);
3663 /* do nothing if already cleared */
3664 if (!(topt->flags->val & topt->opt->bit))
3667 mutex_lock(&trace_types_lock);
3668 if (current_trace->set_flag)
3669 ret = current_trace->set_flag(topt->flags->val,
3671 mutex_unlock(&trace_types_lock);
3674 topt->flags->val &= ~topt->opt->bit;
3677 /* do nothing if already set */
3678 if (topt->flags->val & topt->opt->bit)
3681 mutex_lock(&trace_types_lock);
3682 if (current_trace->set_flag)
3683 ret = current_trace->set_flag(topt->flags->val,
3685 mutex_unlock(&trace_types_lock);
3688 topt->flags->val |= topt->opt->bit;
3701 static const struct file_operations trace_options_fops = {
3702 .open = tracing_open_generic,
3703 .read = trace_options_read,
3704 .write = trace_options_write,
3708 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
3711 long index = (long)filp->private_data;
3714 if (trace_flags & (1 << index))
3719 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3723 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
3726 long index = (long)filp->private_data;
3731 if (cnt >= sizeof(buf))
3734 if (copy_from_user(&buf, ubuf, cnt))
3739 ret = strict_strtoul(buf, 10, &val);
3745 trace_flags &= ~(1 << index);
3748 trace_flags |= 1 << index;
3760 static const struct file_operations trace_options_core_fops = {
3761 .open = tracing_open_generic,
3762 .read = trace_options_core_read,
3763 .write = trace_options_core_write,
3766 static struct dentry *trace_options_init_dentry(void)
3768 struct dentry *d_tracer;
3769 static struct dentry *t_options;
3774 d_tracer = tracing_init_dentry();
3778 t_options = debugfs_create_dir("options", d_tracer);
3780 pr_warning("Could not create debugfs directory 'options'\n");
3788 create_trace_option_file(struct trace_option_dentry *topt,
3789 struct tracer_flags *flags,
3790 struct tracer_opt *opt)
3792 struct dentry *t_options;
3793 struct dentry *entry;
3795 t_options = trace_options_init_dentry();
3799 topt->flags = flags;
3802 entry = debugfs_create_file(opt->name, 0644, t_options, topt,
3803 &trace_options_fops);
3805 topt->entry = entry;
3809 static struct trace_option_dentry *
3810 create_trace_option_files(struct tracer *tracer)
3812 struct trace_option_dentry *topts;
3813 struct tracer_flags *flags;
3814 struct tracer_opt *opts;
3820 flags = tracer->flags;
3822 if (!flags || !flags->opts)
3827 for (cnt = 0; opts[cnt].name; cnt++)
3830 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
3834 for (cnt = 0; opts[cnt].name; cnt++)
3835 create_trace_option_file(&topts[cnt], flags,
3842 destroy_trace_option_files(struct trace_option_dentry *topts)
3849 for (cnt = 0; topts[cnt].opt; cnt++) {
3850 if (topts[cnt].entry)
3851 debugfs_remove(topts[cnt].entry);
3857 static struct dentry *
3858 create_trace_option_core_file(const char *option, long index)
3860 struct dentry *t_options;
3861 struct dentry *entry;
3863 t_options = trace_options_init_dentry();
3867 entry = debugfs_create_file(option, 0644, t_options, (void *)index,
3868 &trace_options_core_fops);
3873 static __init void create_trace_options_dir(void)
3875 struct dentry *t_options;
3876 struct dentry *entry;
3879 t_options = trace_options_init_dentry();
3883 for (i = 0; trace_options[i]; i++) {
3884 entry = create_trace_option_core_file(trace_options[i], i);
3886 pr_warning("Could not create debugfs %s entry\n",
3891 static __init int tracer_init_debugfs(void)
3893 struct dentry *d_tracer;
3894 struct dentry *entry;
3897 d_tracer = tracing_init_dentry();
3899 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3900 &global_trace, &tracing_ctrl_fops);
3902 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3904 entry = debugfs_create_file("trace_options", 0644, d_tracer,
3905 NULL, &tracing_iter_fops);
3907 pr_warning("Could not create debugfs 'trace_options' entry\n");
3909 create_trace_options_dir();
3911 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3912 NULL, &tracing_cpumask_fops);
3914 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3916 entry = debugfs_create_file("trace", 0644, d_tracer,
3917 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
3919 pr_warning("Could not create debugfs 'trace' entry\n");
3921 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3922 &global_trace, &show_traces_fops);
3924 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3926 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3927 &global_trace, &set_tracer_fops);
3929 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3931 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3932 &tracing_max_latency,
3933 &tracing_max_lat_fops);
3935 pr_warning("Could not create debugfs "
3936 "'tracing_max_latency' entry\n");
3938 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3939 &tracing_thresh, &tracing_max_lat_fops);
3941 pr_warning("Could not create debugfs "
3942 "'tracing_thresh' entry\n");
3943 entry = debugfs_create_file("README", 0644, d_tracer,
3944 NULL, &tracing_readme_fops);
3946 pr_warning("Could not create debugfs 'README' entry\n");
3948 entry = debugfs_create_file("trace_pipe", 0444, d_tracer,
3949 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
3951 pr_warning("Could not create debugfs "
3952 "'trace_pipe' entry\n");
3954 entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
3955 &global_trace, &tracing_entries_fops);
3957 pr_warning("Could not create debugfs "
3958 "'buffer_size_kb' entry\n");
3960 entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3961 NULL, &tracing_mark_fops);
3963 pr_warning("Could not create debugfs "
3964 "'trace_marker' entry\n");
3966 #ifdef CONFIG_DYNAMIC_FTRACE
3967 entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3968 &ftrace_update_tot_cnt,
3969 &tracing_dyn_info_fops);
3971 pr_warning("Could not create debugfs "
3972 "'dyn_ftrace_total_info' entry\n");
3974 #ifdef CONFIG_SYSPROF_TRACER
3975 init_tracer_sysprof_debugfs(d_tracer);
3978 for_each_tracing_cpu(cpu)
3979 tracing_init_debugfs_percpu(cpu);
3984 static int trace_panic_handler(struct notifier_block *this,
3985 unsigned long event, void *unused)
3987 if (ftrace_dump_on_oops)
3992 static struct notifier_block trace_panic_notifier = {
3993 .notifier_call = trace_panic_handler,
3995 .priority = 150 /* priority: INT_MAX >= x >= 0 */
3998 static int trace_die_handler(struct notifier_block *self,
4004 if (ftrace_dump_on_oops)
4013 static struct notifier_block trace_die_notifier = {
4014 .notifier_call = trace_die_handler,
4019 * printk is set to max of 1024, we really don't need it that big.
4020 * Nothing should be printing 1000 characters anyway.
4022 #define TRACE_MAX_PRINT 1000
4025 * Define here KERN_TRACE so that we have one place to modify
4026 * it if we decide to change what log level the ftrace dump
4029 #define KERN_TRACE KERN_EMERG
4032 trace_printk_seq(struct trace_seq *s)
4034 /* Probably should print a warning here. */
4038 /* should be zero ended, but we are paranoid. */
4039 s->buffer[s->len] = 0;
4041 printk(KERN_TRACE "%s", s->buffer);
4046 static void __ftrace_dump(bool disable_tracing)
4048 static DEFINE_SPINLOCK(ftrace_dump_lock);
4049 /* use static because iter can be a bit big for the stack */
4050 static struct trace_iterator iter;
4051 unsigned int old_userobj;
4052 static int dump_ran;
4053 unsigned long flags;
4057 spin_lock_irqsave(&ftrace_dump_lock, flags);
4065 if (disable_tracing)
4068 for_each_tracing_cpu(cpu) {
4069 atomic_inc(&global_trace.data[cpu]->disabled);
4072 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4074 /* don't look at user memory in panic mode */
4075 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4077 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4079 /* Simulate the iterator */
4080 iter.tr = &global_trace;
4081 iter.trace = current_trace;
4082 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4085 * We need to stop all tracing on all CPUS to read the
4086 * the next buffer. This is a bit expensive, but is
4087 * not done often. We fill all what we can read,
4088 * and then release the locks again.
4091 while (!trace_empty(&iter)) {
4094 printk(KERN_TRACE "---------------------------------\n");
4098 /* reset all but tr, trace, and overruns */
4099 memset(&iter.seq, 0,
4100 sizeof(struct trace_iterator) -
4101 offsetof(struct trace_iterator, seq));
4102 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4105 if (find_next_entry_inc(&iter) != NULL) {
4106 print_trace_line(&iter);
4107 trace_consume(&iter);
4110 trace_printk_seq(&iter.seq);
4114 printk(KERN_TRACE " (ftrace buffer empty)\n");
4116 printk(KERN_TRACE "---------------------------------\n");
4118 /* Re-enable tracing if requested */
4119 if (!disable_tracing) {
4120 trace_flags |= old_userobj;
4122 for_each_tracing_cpu(cpu) {
4123 atomic_dec(&global_trace.data[cpu]->disabled);
4129 spin_unlock_irqrestore(&ftrace_dump_lock, flags);
4132 /* By default: disable tracing after the dump */
4133 void ftrace_dump(void)
4135 __ftrace_dump(true);
4138 __init static int tracer_alloc_buffers(void)
4140 struct trace_array_cpu *data;
4145 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4148 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4149 goto out_free_buffer_mask;
4151 if (!alloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
4152 goto out_free_tracing_cpumask;
4154 /* To save memory, keep the ring buffer size to its minimum */
4155 if (ring_buffer_expanded)
4156 ring_buf_size = trace_buf_size;
4160 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4161 cpumask_copy(tracing_cpumask, cpu_all_mask);
4162 cpumask_clear(tracing_reader_cpumask);
4164 /* TODO: make the number of buffers hot pluggable with CPUS */
4165 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4166 TRACE_BUFFER_FLAGS);
4167 if (!global_trace.buffer) {
4168 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4170 goto out_free_cpumask;
4172 global_trace.entries = ring_buffer_size(global_trace.buffer);
4175 #ifdef CONFIG_TRACER_MAX_TRACE
4176 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4177 TRACE_BUFFER_FLAGS);
4178 if (!max_tr.buffer) {
4179 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4181 ring_buffer_free(global_trace.buffer);
4182 goto out_free_cpumask;
4184 max_tr.entries = ring_buffer_size(max_tr.buffer);
4185 WARN_ON(max_tr.entries != global_trace.entries);
4188 /* Allocate the first page for all buffers */
4189 for_each_tracing_cpu(i) {
4190 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4191 max_tr.data[i] = &per_cpu(max_data, i);
4194 trace_init_cmdlines();
4196 register_tracer(&nop_trace);
4197 current_trace = &nop_trace;
4198 #ifdef CONFIG_BOOT_TRACER
4199 register_tracer(&boot_tracer);
4201 /* All seems OK, enable tracing */
4202 tracing_disabled = 0;
4204 atomic_notifier_chain_register(&panic_notifier_list,
4205 &trace_panic_notifier);
4207 register_die_notifier(&trace_die_notifier);
4212 free_cpumask_var(tracing_reader_cpumask);
4213 out_free_tracing_cpumask:
4214 free_cpumask_var(tracing_cpumask);
4215 out_free_buffer_mask:
4216 free_cpumask_var(tracing_buffer_mask);
4221 __init static int clear_boot_tracer(void)
4224 * The default tracer at boot buffer is an init section.
4225 * This function is called in lateinit. If we did not
4226 * find the boot tracer, then clear it out, to prevent
4227 * later registration from accessing the buffer that is
4228 * about to be freed.
4230 if (!default_bootup_tracer)
4233 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4234 default_bootup_tracer);
4235 default_bootup_tracer = NULL;
4240 early_initcall(tracer_alloc_buffers);
4241 fs_initcall(tracer_init_debugfs);
4242 late_initcall(clear_boot_tracer);