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);
151 ns2usecs(cycle_t nsec)
159 * The global_trace is the descriptor that holds the tracing
160 * buffers for the live tracing. For each CPU, it contains
161 * a link list of pages that will store trace entries. The
162 * page descriptor of the pages in the memory is used to hold
163 * the link list by linking the lru item in the page descriptor
164 * to each of the pages in the buffer per CPU.
166 * For each active CPU there is a data field that holds the
167 * pages for the buffer for that CPU. Each CPU has the same number
168 * of pages allocated for its buffer.
170 static struct trace_array global_trace;
172 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
174 cycle_t ftrace_now(int cpu)
178 /* Early boot up does not have a buffer yet */
179 if (!global_trace.buffer)
180 return trace_clock_local();
182 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
183 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
189 * The max_tr is used to snapshot the global_trace when a maximum
190 * latency is reached. Some tracers will use this to store a maximum
191 * trace while it continues examining live traces.
193 * The buffers for the max_tr are set up the same as the global_trace.
194 * When a snapshot is taken, the link list of the max_tr is swapped
195 * with the link list of the global_trace and the buffers are reset for
196 * the global_trace so the tracing can continue.
198 static struct trace_array max_tr;
200 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
202 /* tracer_enabled is used to toggle activation of a tracer */
203 static int tracer_enabled = 1;
206 * tracing_is_enabled - return tracer_enabled status
208 * This function is used by other tracers to know the status
209 * of the tracer_enabled flag. Tracers may use this function
210 * to know if it should enable their features when starting
211 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
213 int tracing_is_enabled(void)
215 return tracer_enabled;
219 * trace_buf_size is the size in bytes that is allocated
220 * for a buffer. Note, the number of bytes is always rounded
223 * This number is purposely set to a low number of 16384.
224 * If the dump on oops happens, it will be much appreciated
225 * to not have to wait for all that output. Anyway this can be
226 * boot time and run time configurable.
228 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
230 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
232 /* trace_types holds a link list of available tracers. */
233 static struct tracer *trace_types __read_mostly;
235 /* current_trace points to the tracer that is currently active */
236 static struct tracer *current_trace __read_mostly;
239 * max_tracer_type_len is used to simplify the allocating of
240 * buffers to read userspace tracer names. We keep track of
241 * the longest tracer name registered.
243 static int max_tracer_type_len;
246 * trace_types_lock is used to protect the trace_types list.
247 * This lock is also used to keep user access serialized.
248 * Accesses from userspace will grab this lock while userspace
249 * activities happen inside the kernel.
251 static DEFINE_MUTEX(trace_types_lock);
253 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
254 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
256 /* trace_flags holds trace_options default values */
257 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
258 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO;
261 * trace_wake_up - wake up tasks waiting for trace input
263 * Simply wakes up any task that is blocked on the trace_wait
264 * queue. These is used with trace_poll for tasks polling the trace.
266 void trace_wake_up(void)
269 * The runqueue_is_locked() can fail, but this is the best we
272 if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
273 wake_up(&trace_wait);
276 static int __init set_buf_size(char *str)
278 unsigned long buf_size;
283 ret = strict_strtoul(str, 0, &buf_size);
284 /* nr_entries can not be zero */
285 if (ret < 0 || buf_size == 0)
287 trace_buf_size = buf_size;
290 __setup("trace_buf_size=", set_buf_size);
292 unsigned long nsecs_to_usecs(unsigned long nsecs)
297 /* These must match the bit postions in trace_iterator_flags */
298 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 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 void __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);
947 static void __trace_graph_return(struct trace_array *tr,
948 struct ftrace_graph_ret *trace,
952 struct ring_buffer_event *event;
953 struct ftrace_graph_ret_entry *entry;
955 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
958 event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_RET,
959 sizeof(*entry), flags, pc);
962 entry = ring_buffer_event_data(event);
964 ring_buffer_unlock_commit(global_trace.buffer, event);
969 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
970 unsigned long ip, unsigned long parent_ip, unsigned long flags,
973 if (likely(!atomic_read(&data->disabled)))
974 trace_function(tr, ip, parent_ip, flags, pc);
977 static void __ftrace_trace_stack(struct trace_array *tr,
981 #ifdef CONFIG_STACKTRACE
982 struct ring_buffer_event *event;
983 struct stack_entry *entry;
984 struct stack_trace trace;
986 event = trace_buffer_lock_reserve(tr, TRACE_STACK,
987 sizeof(*entry), flags, pc);
990 entry = ring_buffer_event_data(event);
991 memset(&entry->caller, 0, sizeof(entry->caller));
993 trace.nr_entries = 0;
994 trace.max_entries = FTRACE_STACK_ENTRIES;
996 trace.entries = entry->caller;
998 save_stack_trace(&trace);
999 ring_buffer_unlock_commit(tr->buffer, event);
1003 static void ftrace_trace_stack(struct trace_array *tr,
1004 unsigned long flags,
1007 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1010 __ftrace_trace_stack(tr, flags, skip, pc);
1013 void __trace_stack(struct trace_array *tr,
1014 unsigned long flags,
1017 __ftrace_trace_stack(tr, flags, skip, pc);
1020 static void ftrace_trace_userstack(struct trace_array *tr,
1021 unsigned long flags, int pc)
1023 #ifdef CONFIG_STACKTRACE
1024 struct ring_buffer_event *event;
1025 struct userstack_entry *entry;
1026 struct stack_trace trace;
1028 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1031 event = trace_buffer_lock_reserve(tr, TRACE_USER_STACK,
1032 sizeof(*entry), flags, pc);
1035 entry = ring_buffer_event_data(event);
1037 memset(&entry->caller, 0, sizeof(entry->caller));
1039 trace.nr_entries = 0;
1040 trace.max_entries = FTRACE_STACK_ENTRIES;
1042 trace.entries = entry->caller;
1044 save_stack_trace_user(&trace);
1045 ring_buffer_unlock_commit(tr->buffer, event);
1050 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1052 ftrace_trace_userstack(tr, flags, preempt_count());
1057 ftrace_trace_special(void *__tr,
1058 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1061 struct ring_buffer_event *event;
1062 struct trace_array *tr = __tr;
1063 struct special_entry *entry;
1065 event = trace_buffer_lock_reserve(tr, TRACE_SPECIAL,
1066 sizeof(*entry), 0, pc);
1069 entry = ring_buffer_event_data(event);
1073 trace_buffer_unlock_commit(tr, event, 0, pc);
1077 __trace_special(void *__tr, void *__data,
1078 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1080 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1084 tracing_sched_switch_trace(struct trace_array *tr,
1085 struct task_struct *prev,
1086 struct task_struct *next,
1087 unsigned long flags, int pc)
1089 struct ring_buffer_event *event;
1090 struct ctx_switch_entry *entry;
1092 event = trace_buffer_lock_reserve(tr, TRACE_CTX,
1093 sizeof(*entry), flags, pc);
1096 entry = ring_buffer_event_data(event);
1097 entry->prev_pid = prev->pid;
1098 entry->prev_prio = prev->prio;
1099 entry->prev_state = prev->state;
1100 entry->next_pid = next->pid;
1101 entry->next_prio = next->prio;
1102 entry->next_state = next->state;
1103 entry->next_cpu = task_cpu(next);
1104 trace_buffer_unlock_commit(tr, event, flags, pc);
1108 tracing_sched_wakeup_trace(struct trace_array *tr,
1109 struct task_struct *wakee,
1110 struct task_struct *curr,
1111 unsigned long flags, int pc)
1113 struct ring_buffer_event *event;
1114 struct ctx_switch_entry *entry;
1116 event = trace_buffer_lock_reserve(tr, TRACE_WAKE,
1117 sizeof(*entry), flags, pc);
1120 entry = ring_buffer_event_data(event);
1121 entry->prev_pid = curr->pid;
1122 entry->prev_prio = curr->prio;
1123 entry->prev_state = curr->state;
1124 entry->next_pid = wakee->pid;
1125 entry->next_prio = wakee->prio;
1126 entry->next_state = wakee->state;
1127 entry->next_cpu = task_cpu(wakee);
1129 ring_buffer_unlock_commit(tr->buffer, event);
1130 ftrace_trace_stack(tr, flags, 6, pc);
1131 ftrace_trace_userstack(tr, flags, pc);
1135 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1137 struct trace_array *tr = &global_trace;
1138 struct trace_array_cpu *data;
1139 unsigned long flags;
1143 if (tracing_disabled)
1146 pc = preempt_count();
1147 local_irq_save(flags);
1148 cpu = raw_smp_processor_id();
1149 data = tr->data[cpu];
1151 if (likely(atomic_inc_return(&data->disabled) == 1))
1152 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1154 atomic_dec(&data->disabled);
1155 local_irq_restore(flags);
1158 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1159 int trace_graph_entry(struct ftrace_graph_ent *trace)
1161 struct trace_array *tr = &global_trace;
1162 struct trace_array_cpu *data;
1163 unsigned long flags;
1168 if (!ftrace_trace_task(current))
1171 if (!ftrace_graph_addr(trace->func))
1174 local_irq_save(flags);
1175 cpu = raw_smp_processor_id();
1176 data = tr->data[cpu];
1177 disabled = atomic_inc_return(&data->disabled);
1178 if (likely(disabled == 1)) {
1179 pc = preempt_count();
1180 __trace_graph_entry(tr, trace, flags, pc);
1182 /* Only do the atomic if it is not already set */
1183 if (!test_tsk_trace_graph(current))
1184 set_tsk_trace_graph(current);
1185 atomic_dec(&data->disabled);
1186 local_irq_restore(flags);
1191 void trace_graph_return(struct ftrace_graph_ret *trace)
1193 struct trace_array *tr = &global_trace;
1194 struct trace_array_cpu *data;
1195 unsigned long flags;
1200 local_irq_save(flags);
1201 cpu = raw_smp_processor_id();
1202 data = tr->data[cpu];
1203 disabled = atomic_inc_return(&data->disabled);
1204 if (likely(disabled == 1)) {
1205 pc = preempt_count();
1206 __trace_graph_return(tr, trace, flags, pc);
1209 clear_tsk_trace_graph(current);
1210 atomic_dec(&data->disabled);
1211 local_irq_restore(flags);
1213 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1217 * trace_vbprintk - write binary msg to tracing buffer
1220 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1222 static raw_spinlock_t trace_buf_lock =
1223 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
1224 static u32 trace_buf[TRACE_BUF_SIZE];
1226 struct ring_buffer_event *event;
1227 struct trace_array *tr = &global_trace;
1228 struct trace_array_cpu *data;
1229 struct bprint_entry *entry;
1230 unsigned long flags;
1232 int cpu, len = 0, size, pc;
1234 if (unlikely(tracing_selftest_running || tracing_disabled))
1237 /* Don't pollute graph traces with trace_vprintk internals */
1238 pause_graph_tracing();
1240 pc = preempt_count();
1241 resched = ftrace_preempt_disable();
1242 cpu = raw_smp_processor_id();
1243 data = tr->data[cpu];
1245 if (unlikely(atomic_read(&data->disabled)))
1248 /* Lockdep uses trace_printk for lock tracing */
1249 local_irq_save(flags);
1250 __raw_spin_lock(&trace_buf_lock);
1251 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1253 if (len > TRACE_BUF_SIZE || len < 0)
1256 size = sizeof(*entry) + sizeof(u32) * len;
1257 event = trace_buffer_lock_reserve(tr, TRACE_BPRINT, size, flags, pc);
1260 entry = ring_buffer_event_data(event);
1264 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1265 ring_buffer_unlock_commit(tr->buffer, event);
1268 __raw_spin_unlock(&trace_buf_lock);
1269 local_irq_restore(flags);
1272 ftrace_preempt_enable(resched);
1273 unpause_graph_tracing();
1277 EXPORT_SYMBOL_GPL(trace_vbprintk);
1279 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1281 static raw_spinlock_t trace_buf_lock = __RAW_SPIN_LOCK_UNLOCKED;
1282 static char trace_buf[TRACE_BUF_SIZE];
1284 struct ring_buffer_event *event;
1285 struct trace_array *tr = &global_trace;
1286 struct trace_array_cpu *data;
1287 int cpu, len = 0, size, pc;
1288 struct print_entry *entry;
1289 unsigned long irq_flags;
1291 if (tracing_disabled || tracing_selftest_running)
1294 pc = preempt_count();
1295 preempt_disable_notrace();
1296 cpu = raw_smp_processor_id();
1297 data = tr->data[cpu];
1299 if (unlikely(atomic_read(&data->disabled)))
1302 pause_graph_tracing();
1303 raw_local_irq_save(irq_flags);
1304 __raw_spin_lock(&trace_buf_lock);
1305 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1307 len = min(len, TRACE_BUF_SIZE-1);
1310 size = sizeof(*entry) + len + 1;
1311 event = trace_buffer_lock_reserve(tr, TRACE_PRINT, size, irq_flags, pc);
1314 entry = ring_buffer_event_data(event);
1317 memcpy(&entry->buf, trace_buf, len);
1318 entry->buf[len] = 0;
1319 ring_buffer_unlock_commit(tr->buffer, event);
1322 __raw_spin_unlock(&trace_buf_lock);
1323 raw_local_irq_restore(irq_flags);
1324 unpause_graph_tracing();
1326 preempt_enable_notrace();
1330 EXPORT_SYMBOL_GPL(trace_vprintk);
1332 enum trace_file_type {
1333 TRACE_FILE_LAT_FMT = 1,
1334 TRACE_FILE_ANNOTATE = 2,
1337 static void trace_iterator_increment(struct trace_iterator *iter)
1339 /* Don't allow ftrace to trace into the ring buffers */
1340 ftrace_disable_cpu();
1343 if (iter->buffer_iter[iter->cpu])
1344 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1346 ftrace_enable_cpu();
1349 static struct trace_entry *
1350 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1352 struct ring_buffer_event *event;
1353 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1355 /* Don't allow ftrace to trace into the ring buffers */
1356 ftrace_disable_cpu();
1359 event = ring_buffer_iter_peek(buf_iter, ts);
1361 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1363 ftrace_enable_cpu();
1365 return event ? ring_buffer_event_data(event) : NULL;
1368 static struct trace_entry *
1369 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1371 struct ring_buffer *buffer = iter->tr->buffer;
1372 struct trace_entry *ent, *next = NULL;
1373 int cpu_file = iter->cpu_file;
1374 u64 next_ts = 0, ts;
1379 * If we are in a per_cpu trace file, don't bother by iterating over
1380 * all cpu and peek directly.
1382 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1383 if (ring_buffer_empty_cpu(buffer, cpu_file))
1385 ent = peek_next_entry(iter, cpu_file, ent_ts);
1387 *ent_cpu = cpu_file;
1392 for_each_tracing_cpu(cpu) {
1394 if (ring_buffer_empty_cpu(buffer, cpu))
1397 ent = peek_next_entry(iter, cpu, &ts);
1400 * Pick the entry with the smallest timestamp:
1402 if (ent && (!next || ts < next_ts)) {
1410 *ent_cpu = next_cpu;
1418 /* Find the next real entry, without updating the iterator itself */
1419 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1420 int *ent_cpu, u64 *ent_ts)
1422 return __find_next_entry(iter, ent_cpu, ent_ts);
1425 /* Find the next real entry, and increment the iterator to the next entry */
1426 static void *find_next_entry_inc(struct trace_iterator *iter)
1428 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1431 trace_iterator_increment(iter);
1433 return iter->ent ? iter : NULL;
1436 static void trace_consume(struct trace_iterator *iter)
1438 /* Don't allow ftrace to trace into the ring buffers */
1439 ftrace_disable_cpu();
1440 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1441 ftrace_enable_cpu();
1444 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1446 struct trace_iterator *iter = m->private;
1452 /* can't go backwards */
1457 ent = find_next_entry_inc(iter);
1461 while (ent && iter->idx < i)
1462 ent = find_next_entry_inc(iter);
1470 * No necessary locking here. The worst thing which can
1471 * happen is loosing events consumed at the same time
1472 * by a trace_pipe reader.
1473 * Other than that, we don't risk to crash the ring buffer
1474 * because it serializes the readers.
1476 * The current tracer is copied to avoid a global locking
1479 static void *s_start(struct seq_file *m, loff_t *pos)
1481 struct trace_iterator *iter = m->private;
1482 static struct tracer *old_tracer;
1483 int cpu_file = iter->cpu_file;
1488 /* copy the tracer to avoid using a global lock all around */
1489 mutex_lock(&trace_types_lock);
1490 if (unlikely(old_tracer != current_trace && current_trace)) {
1491 old_tracer = current_trace;
1492 *iter->trace = *current_trace;
1494 mutex_unlock(&trace_types_lock);
1496 atomic_inc(&trace_record_cmdline_disabled);
1498 if (*pos != iter->pos) {
1503 ftrace_disable_cpu();
1505 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1506 for_each_tracing_cpu(cpu)
1507 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1509 ring_buffer_iter_reset(iter->buffer_iter[cpu_file]);
1512 ftrace_enable_cpu();
1514 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1519 p = s_next(m, p, &l);
1525 static void s_stop(struct seq_file *m, void *p)
1527 atomic_dec(&trace_record_cmdline_disabled);
1530 static void print_lat_help_header(struct seq_file *m)
1532 seq_puts(m, "# _------=> CPU# \n");
1533 seq_puts(m, "# / _-----=> irqs-off \n");
1534 seq_puts(m, "# | / _----=> need-resched \n");
1535 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1536 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1537 seq_puts(m, "# |||| / \n");
1538 seq_puts(m, "# ||||| delay \n");
1539 seq_puts(m, "# cmd pid ||||| time | caller \n");
1540 seq_puts(m, "# \\ / ||||| \\ | / \n");
1543 static void print_func_help_header(struct seq_file *m)
1545 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1546 seq_puts(m, "# | | | | |\n");
1551 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1553 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1554 struct trace_array *tr = iter->tr;
1555 struct trace_array_cpu *data = tr->data[tr->cpu];
1556 struct tracer *type = current_trace;
1557 unsigned long total;
1558 unsigned long entries;
1559 const char *name = "preemption";
1564 entries = ring_buffer_entries(iter->tr->buffer);
1566 ring_buffer_overruns(iter->tr->buffer);
1568 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1570 seq_puts(m, "# -----------------------------------"
1571 "---------------------------------\n");
1572 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1573 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1574 nsecs_to_usecs(data->saved_latency),
1578 #if defined(CONFIG_PREEMPT_NONE)
1580 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1582 #elif defined(CONFIG_PREEMPT)
1587 /* These are reserved for later use */
1590 seq_printf(m, " #P:%d)\n", num_online_cpus());
1594 seq_puts(m, "# -----------------\n");
1595 seq_printf(m, "# | task: %.16s-%d "
1596 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1597 data->comm, data->pid, data->uid, data->nice,
1598 data->policy, data->rt_priority);
1599 seq_puts(m, "# -----------------\n");
1601 if (data->critical_start) {
1602 seq_puts(m, "# => started at: ");
1603 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1604 trace_print_seq(m, &iter->seq);
1605 seq_puts(m, "\n# => ended at: ");
1606 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1607 trace_print_seq(m, &iter->seq);
1614 static void test_cpu_buff_start(struct trace_iterator *iter)
1616 struct trace_seq *s = &iter->seq;
1618 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1621 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1624 if (cpumask_test_cpu(iter->cpu, iter->started))
1627 cpumask_set_cpu(iter->cpu, iter->started);
1628 trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1631 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1633 struct trace_seq *s = &iter->seq;
1634 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1635 struct trace_entry *entry;
1636 struct trace_event *event;
1640 test_cpu_buff_start(iter);
1642 event = ftrace_find_event(entry->type);
1644 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1645 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1646 if (!trace_print_lat_context(iter))
1649 if (!trace_print_context(iter))
1655 return event->trace(iter, sym_flags);
1657 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1660 return TRACE_TYPE_HANDLED;
1662 return TRACE_TYPE_PARTIAL_LINE;
1665 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1667 struct trace_seq *s = &iter->seq;
1668 struct trace_entry *entry;
1669 struct trace_event *event;
1673 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1674 if (!trace_seq_printf(s, "%d %d %llu ",
1675 entry->pid, iter->cpu, iter->ts))
1679 event = ftrace_find_event(entry->type);
1681 return event->raw(iter, 0);
1683 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1686 return TRACE_TYPE_HANDLED;
1688 return TRACE_TYPE_PARTIAL_LINE;
1691 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1693 struct trace_seq *s = &iter->seq;
1694 unsigned char newline = '\n';
1695 struct trace_entry *entry;
1696 struct trace_event *event;
1700 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1701 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1702 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1703 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1706 event = ftrace_find_event(entry->type);
1708 enum print_line_t ret = event->hex(iter, 0);
1709 if (ret != TRACE_TYPE_HANDLED)
1713 SEQ_PUT_FIELD_RET(s, newline);
1715 return TRACE_TYPE_HANDLED;
1718 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1720 struct trace_seq *s = &iter->seq;
1721 struct trace_entry *entry;
1722 struct trace_event *event;
1726 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1727 SEQ_PUT_FIELD_RET(s, entry->pid);
1728 SEQ_PUT_FIELD_RET(s, iter->cpu);
1729 SEQ_PUT_FIELD_RET(s, iter->ts);
1732 event = ftrace_find_event(entry->type);
1733 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1736 static int trace_empty(struct trace_iterator *iter)
1740 /* If we are looking at one CPU buffer, only check that one */
1741 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1742 cpu = iter->cpu_file;
1743 if (iter->buffer_iter[cpu]) {
1744 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1747 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1753 for_each_tracing_cpu(cpu) {
1754 if (iter->buffer_iter[cpu]) {
1755 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1758 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1766 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1768 enum print_line_t ret;
1770 if (iter->trace && iter->trace->print_line) {
1771 ret = iter->trace->print_line(iter);
1772 if (ret != TRACE_TYPE_UNHANDLED)
1776 if (iter->ent->type == TRACE_BPRINT &&
1777 trace_flags & TRACE_ITER_PRINTK &&
1778 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1779 return trace_print_bprintk_msg_only(iter);
1781 if (iter->ent->type == TRACE_PRINT &&
1782 trace_flags & TRACE_ITER_PRINTK &&
1783 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1784 return trace_print_printk_msg_only(iter);
1786 if (trace_flags & TRACE_ITER_BIN)
1787 return print_bin_fmt(iter);
1789 if (trace_flags & TRACE_ITER_HEX)
1790 return print_hex_fmt(iter);
1792 if (trace_flags & TRACE_ITER_RAW)
1793 return print_raw_fmt(iter);
1795 return print_trace_fmt(iter);
1798 static int s_show(struct seq_file *m, void *v)
1800 struct trace_iterator *iter = v;
1802 if (iter->ent == NULL) {
1804 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1807 if (iter->trace && iter->trace->print_header)
1808 iter->trace->print_header(m);
1809 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1810 /* print nothing if the buffers are empty */
1811 if (trace_empty(iter))
1813 print_trace_header(m, iter);
1814 if (!(trace_flags & TRACE_ITER_VERBOSE))
1815 print_lat_help_header(m);
1817 if (!(trace_flags & TRACE_ITER_VERBOSE))
1818 print_func_help_header(m);
1821 print_trace_line(iter);
1822 trace_print_seq(m, &iter->seq);
1828 static struct seq_operations tracer_seq_ops = {
1835 static struct trace_iterator *
1836 __tracing_open(struct inode *inode, struct file *file)
1838 long cpu_file = (long) inode->i_private;
1839 void *fail_ret = ERR_PTR(-ENOMEM);
1840 struct trace_iterator *iter;
1844 if (tracing_disabled)
1845 return ERR_PTR(-ENODEV);
1847 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1849 return ERR_PTR(-ENOMEM);
1852 * We make a copy of the current tracer to avoid concurrent
1853 * changes on it while we are reading.
1855 mutex_lock(&trace_types_lock);
1856 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
1861 *iter->trace = *current_trace;
1863 if (current_trace && current_trace->print_max)
1866 iter->tr = &global_trace;
1868 mutex_init(&iter->mutex);
1869 iter->cpu_file = cpu_file;
1871 /* Notify the tracer early; before we stop tracing. */
1872 if (iter->trace && iter->trace->open)
1873 iter->trace->open(iter);
1875 /* Annotate start of buffers if we had overruns */
1876 if (ring_buffer_overruns(iter->tr->buffer))
1877 iter->iter_flags |= TRACE_FILE_ANNOTATE;
1879 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
1880 for_each_tracing_cpu(cpu) {
1882 iter->buffer_iter[cpu] =
1883 ring_buffer_read_start(iter->tr->buffer, cpu);
1886 cpu = iter->cpu_file;
1887 iter->buffer_iter[cpu] =
1888 ring_buffer_read_start(iter->tr->buffer, cpu);
1891 /* TODO stop tracer */
1892 ret = seq_open(file, &tracer_seq_ops);
1894 fail_ret = ERR_PTR(ret);
1898 m = file->private_data;
1901 /* stop the trace while dumping */
1904 mutex_unlock(&trace_types_lock);
1909 for_each_tracing_cpu(cpu) {
1910 if (iter->buffer_iter[cpu])
1911 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1914 mutex_unlock(&trace_types_lock);
1921 int tracing_open_generic(struct inode *inode, struct file *filp)
1923 if (tracing_disabled)
1926 filp->private_data = inode->i_private;
1930 static int tracing_release(struct inode *inode, struct file *file)
1932 struct seq_file *m = (struct seq_file *)file->private_data;
1933 struct trace_iterator *iter;
1936 if (!(file->f_mode & FMODE_READ))
1941 mutex_lock(&trace_types_lock);
1942 for_each_tracing_cpu(cpu) {
1943 if (iter->buffer_iter[cpu])
1944 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1947 if (iter->trace && iter->trace->close)
1948 iter->trace->close(iter);
1950 /* reenable tracing if it was previously enabled */
1952 mutex_unlock(&trace_types_lock);
1954 seq_release(inode, file);
1955 mutex_destroy(&iter->mutex);
1961 static int tracing_open(struct inode *inode, struct file *file)
1963 struct trace_iterator *iter;
1966 /* If this file was open for write, then erase contents */
1967 if ((file->f_mode & FMODE_WRITE) &&
1968 !(file->f_flags & O_APPEND)) {
1969 long cpu = (long) inode->i_private;
1971 if (cpu == TRACE_PIPE_ALL_CPU)
1972 tracing_reset_online_cpus(&global_trace);
1974 tracing_reset(&global_trace, cpu);
1977 if (file->f_mode & FMODE_READ) {
1978 iter = __tracing_open(inode, file);
1980 ret = PTR_ERR(iter);
1981 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
1982 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1988 t_next(struct seq_file *m, void *v, loff_t *pos)
1990 struct tracer *t = m->private;
2002 static void *t_start(struct seq_file *m, loff_t *pos)
2004 struct tracer *t = m->private;
2007 mutex_lock(&trace_types_lock);
2008 for (; t && l < *pos; t = t_next(m, t, &l))
2014 static void t_stop(struct seq_file *m, void *p)
2016 mutex_unlock(&trace_types_lock);
2019 static int t_show(struct seq_file *m, void *v)
2021 struct tracer *t = v;
2026 seq_printf(m, "%s", t->name);
2035 static struct seq_operations show_traces_seq_ops = {
2042 static int show_traces_open(struct inode *inode, struct file *file)
2046 if (tracing_disabled)
2049 ret = seq_open(file, &show_traces_seq_ops);
2051 struct seq_file *m = file->private_data;
2052 m->private = trace_types;
2059 tracing_write_stub(struct file *filp, const char __user *ubuf,
2060 size_t count, loff_t *ppos)
2065 static const struct file_operations tracing_fops = {
2066 .open = tracing_open,
2068 .write = tracing_write_stub,
2069 .llseek = seq_lseek,
2070 .release = tracing_release,
2073 static const struct file_operations show_traces_fops = {
2074 .open = show_traces_open,
2076 .release = seq_release,
2080 * Only trace on a CPU if the bitmask is set:
2082 static cpumask_var_t tracing_cpumask;
2085 * The tracer itself will not take this lock, but still we want
2086 * to provide a consistent cpumask to user-space:
2088 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2091 * Temporary storage for the character representation of the
2092 * CPU bitmask (and one more byte for the newline):
2094 static char mask_str[NR_CPUS + 1];
2097 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2098 size_t count, loff_t *ppos)
2102 mutex_lock(&tracing_cpumask_update_lock);
2104 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2105 if (count - len < 2) {
2109 len += sprintf(mask_str + len, "\n");
2110 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2113 mutex_unlock(&tracing_cpumask_update_lock);
2119 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2120 size_t count, loff_t *ppos)
2123 cpumask_var_t tracing_cpumask_new;
2125 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2128 mutex_lock(&tracing_cpumask_update_lock);
2129 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2133 local_irq_disable();
2134 __raw_spin_lock(&ftrace_max_lock);
2135 for_each_tracing_cpu(cpu) {
2137 * Increase/decrease the disabled counter if we are
2138 * about to flip a bit in the cpumask:
2140 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2141 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2142 atomic_inc(&global_trace.data[cpu]->disabled);
2144 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2145 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2146 atomic_dec(&global_trace.data[cpu]->disabled);
2149 __raw_spin_unlock(&ftrace_max_lock);
2152 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2154 mutex_unlock(&tracing_cpumask_update_lock);
2155 free_cpumask_var(tracing_cpumask_new);
2160 mutex_unlock(&tracing_cpumask_update_lock);
2161 free_cpumask_var(tracing_cpumask);
2166 static const struct file_operations tracing_cpumask_fops = {
2167 .open = tracing_open_generic,
2168 .read = tracing_cpumask_read,
2169 .write = tracing_cpumask_write,
2173 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2174 size_t cnt, loff_t *ppos)
2176 struct tracer_opt *trace_opts;
2184 /* calculate max size */
2185 for (i = 0; trace_options[i]; i++) {
2186 len += strlen(trace_options[i]);
2187 len += 3; /* "no" and newline */
2190 mutex_lock(&trace_types_lock);
2191 tracer_flags = current_trace->flags->val;
2192 trace_opts = current_trace->flags->opts;
2195 * Increase the size with names of options specific
2196 * of the current tracer.
2198 for (i = 0; trace_opts[i].name; i++) {
2199 len += strlen(trace_opts[i].name);
2200 len += 3; /* "no" and newline */
2203 /* +2 for \n and \0 */
2204 buf = kmalloc(len + 2, GFP_KERNEL);
2206 mutex_unlock(&trace_types_lock);
2210 for (i = 0; trace_options[i]; i++) {
2211 if (trace_flags & (1 << i))
2212 r += sprintf(buf + r, "%s\n", trace_options[i]);
2214 r += sprintf(buf + r, "no%s\n", trace_options[i]);
2217 for (i = 0; trace_opts[i].name; i++) {
2218 if (tracer_flags & trace_opts[i].bit)
2219 r += sprintf(buf + r, "%s\n",
2220 trace_opts[i].name);
2222 r += sprintf(buf + r, "no%s\n",
2223 trace_opts[i].name);
2225 mutex_unlock(&trace_types_lock);
2227 WARN_ON(r >= len + 2);
2229 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2235 /* Try to assign a tracer specific option */
2236 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2238 struct tracer_flags *trace_flags = trace->flags;
2239 struct tracer_opt *opts = NULL;
2243 for (i = 0; trace_flags->opts[i].name; i++) {
2244 opts = &trace_flags->opts[i];
2245 len = strlen(opts->name);
2247 if (strncmp(cmp, opts->name, len) == 0) {
2248 ret = trace->set_flag(trace_flags->val,
2254 if (!trace_flags->opts[i].name)
2257 /* Refused to handle */
2262 trace_flags->val &= ~opts->bit;
2264 trace_flags->val |= opts->bit;
2269 static void set_tracer_flags(unsigned int mask, int enabled)
2271 /* do nothing if flag is already set */
2272 if (!!(trace_flags & mask) == !!enabled)
2276 trace_flags |= mask;
2278 trace_flags &= ~mask;
2280 if (mask == TRACE_ITER_GLOBAL_CLK) {
2284 func = trace_clock_global;
2286 func = trace_clock_local;
2288 mutex_lock(&trace_types_lock);
2289 ring_buffer_set_clock(global_trace.buffer, func);
2292 ring_buffer_set_clock(max_tr.buffer, func);
2293 mutex_unlock(&trace_types_lock);
2298 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2299 size_t cnt, loff_t *ppos)
2307 if (cnt >= sizeof(buf))
2310 if (copy_from_user(&buf, ubuf, cnt))
2315 if (strncmp(buf, "no", 2) == 0) {
2320 for (i = 0; trace_options[i]; i++) {
2321 int len = strlen(trace_options[i]);
2323 if (strncmp(cmp, trace_options[i], len) == 0) {
2324 set_tracer_flags(1 << i, !neg);
2329 /* If no option could be set, test the specific tracer options */
2330 if (!trace_options[i]) {
2331 mutex_lock(&trace_types_lock);
2332 ret = set_tracer_option(current_trace, cmp, neg);
2333 mutex_unlock(&trace_types_lock);
2343 static const struct file_operations tracing_iter_fops = {
2344 .open = tracing_open_generic,
2345 .read = tracing_trace_options_read,
2346 .write = tracing_trace_options_write,
2349 static const char readme_msg[] =
2350 "tracing mini-HOWTO:\n\n"
2352 "# mount -t debugfs nodev /debug\n\n"
2353 "# cat /debug/tracing/available_tracers\n"
2354 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2355 "# cat /debug/tracing/current_tracer\n"
2357 "# echo sched_switch > /debug/tracing/current_tracer\n"
2358 "# cat /debug/tracing/current_tracer\n"
2360 "# cat /debug/tracing/trace_options\n"
2361 "noprint-parent nosym-offset nosym-addr noverbose\n"
2362 "# echo print-parent > /debug/tracing/trace_options\n"
2363 "# echo 1 > /debug/tracing/tracing_enabled\n"
2364 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2365 "echo 0 > /debug/tracing/tracing_enabled\n"
2369 tracing_readme_read(struct file *filp, char __user *ubuf,
2370 size_t cnt, loff_t *ppos)
2372 return simple_read_from_buffer(ubuf, cnt, ppos,
2373 readme_msg, strlen(readme_msg));
2376 static const struct file_operations tracing_readme_fops = {
2377 .open = tracing_open_generic,
2378 .read = tracing_readme_read,
2382 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2383 size_t cnt, loff_t *ppos)
2388 r = sprintf(buf, "%u\n", tracer_enabled);
2389 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2393 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2394 size_t cnt, loff_t *ppos)
2396 struct trace_array *tr = filp->private_data;
2401 if (cnt >= sizeof(buf))
2404 if (copy_from_user(&buf, ubuf, cnt))
2409 ret = strict_strtoul(buf, 10, &val);
2415 mutex_lock(&trace_types_lock);
2416 if (tracer_enabled ^ val) {
2419 if (current_trace->start)
2420 current_trace->start(tr);
2425 if (current_trace->stop)
2426 current_trace->stop(tr);
2429 mutex_unlock(&trace_types_lock);
2437 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2438 size_t cnt, loff_t *ppos)
2440 char buf[max_tracer_type_len+2];
2443 mutex_lock(&trace_types_lock);
2445 r = sprintf(buf, "%s\n", current_trace->name);
2447 r = sprintf(buf, "\n");
2448 mutex_unlock(&trace_types_lock);
2450 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2453 int tracer_init(struct tracer *t, struct trace_array *tr)
2455 tracing_reset_online_cpus(tr);
2459 static int tracing_resize_ring_buffer(unsigned long size)
2464 * If kernel or user changes the size of the ring buffer
2465 * we use the size that was given, and we can forget about
2466 * expanding it later.
2468 ring_buffer_expanded = 1;
2470 ret = ring_buffer_resize(global_trace.buffer, size);
2474 ret = ring_buffer_resize(max_tr.buffer, size);
2478 r = ring_buffer_resize(global_trace.buffer,
2479 global_trace.entries);
2482 * AARGH! We are left with different
2483 * size max buffer!!!!
2484 * The max buffer is our "snapshot" buffer.
2485 * When a tracer needs a snapshot (one of the
2486 * latency tracers), it swaps the max buffer
2487 * with the saved snap shot. We succeeded to
2488 * update the size of the main buffer, but failed to
2489 * update the size of the max buffer. But when we tried
2490 * to reset the main buffer to the original size, we
2491 * failed there too. This is very unlikely to
2492 * happen, but if it does, warn and kill all
2496 tracing_disabled = 1;
2501 global_trace.entries = size;
2507 * tracing_update_buffers - used by tracing facility to expand ring buffers
2509 * To save on memory when the tracing is never used on a system with it
2510 * configured in. The ring buffers are set to a minimum size. But once
2511 * a user starts to use the tracing facility, then they need to grow
2512 * to their default size.
2514 * This function is to be called when a tracer is about to be used.
2516 int tracing_update_buffers(void)
2520 mutex_lock(&trace_types_lock);
2521 if (!ring_buffer_expanded)
2522 ret = tracing_resize_ring_buffer(trace_buf_size);
2523 mutex_unlock(&trace_types_lock);
2528 struct trace_option_dentry;
2530 static struct trace_option_dentry *
2531 create_trace_option_files(struct tracer *tracer);
2534 destroy_trace_option_files(struct trace_option_dentry *topts);
2536 static int tracing_set_tracer(const char *buf)
2538 static struct trace_option_dentry *topts;
2539 struct trace_array *tr = &global_trace;
2543 mutex_lock(&trace_types_lock);
2545 if (!ring_buffer_expanded) {
2546 ret = tracing_resize_ring_buffer(trace_buf_size);
2552 for (t = trace_types; t; t = t->next) {
2553 if (strcmp(t->name, buf) == 0)
2560 if (t == current_trace)
2563 trace_branch_disable();
2564 if (current_trace && current_trace->reset)
2565 current_trace->reset(tr);
2567 destroy_trace_option_files(topts);
2571 topts = create_trace_option_files(current_trace);
2574 ret = tracer_init(t, tr);
2579 trace_branch_enable(tr);
2581 mutex_unlock(&trace_types_lock);
2587 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2588 size_t cnt, loff_t *ppos)
2590 char buf[max_tracer_type_len+1];
2597 if (cnt > max_tracer_type_len)
2598 cnt = max_tracer_type_len;
2600 if (copy_from_user(&buf, ubuf, cnt))
2605 /* strip ending whitespace. */
2606 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2609 err = tracing_set_tracer(buf);
2619 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2620 size_t cnt, loff_t *ppos)
2622 unsigned long *ptr = filp->private_data;
2626 r = snprintf(buf, sizeof(buf), "%ld\n",
2627 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2628 if (r > sizeof(buf))
2630 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2634 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2635 size_t cnt, loff_t *ppos)
2637 unsigned long *ptr = filp->private_data;
2642 if (cnt >= sizeof(buf))
2645 if (copy_from_user(&buf, ubuf, cnt))
2650 ret = strict_strtoul(buf, 10, &val);
2659 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2661 long cpu_file = (long) inode->i_private;
2662 struct trace_iterator *iter;
2665 if (tracing_disabled)
2668 mutex_lock(&trace_types_lock);
2670 /* We only allow one reader per cpu */
2671 if (cpu_file == TRACE_PIPE_ALL_CPU) {
2672 if (!cpumask_empty(tracing_reader_cpumask)) {
2676 cpumask_setall(tracing_reader_cpumask);
2678 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2679 cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2686 /* create a buffer to store the information to pass to userspace */
2687 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2694 * We make a copy of the current tracer to avoid concurrent
2695 * changes on it while we are reading.
2697 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2703 *iter->trace = *current_trace;
2705 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2710 /* trace pipe does not show start of buffer */
2711 cpumask_setall(iter->started);
2713 iter->cpu_file = cpu_file;
2714 iter->tr = &global_trace;
2715 mutex_init(&iter->mutex);
2716 filp->private_data = iter;
2718 if (iter->trace->pipe_open)
2719 iter->trace->pipe_open(iter);
2722 mutex_unlock(&trace_types_lock);
2728 mutex_unlock(&trace_types_lock);
2732 static int tracing_release_pipe(struct inode *inode, struct file *file)
2734 struct trace_iterator *iter = file->private_data;
2736 mutex_lock(&trace_types_lock);
2738 if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2739 cpumask_clear(tracing_reader_cpumask);
2741 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2743 mutex_unlock(&trace_types_lock);
2745 free_cpumask_var(iter->started);
2746 mutex_destroy(&iter->mutex);
2754 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2756 struct trace_iterator *iter = filp->private_data;
2758 if (trace_flags & TRACE_ITER_BLOCK) {
2760 * Always select as readable when in blocking mode
2762 return POLLIN | POLLRDNORM;
2764 if (!trace_empty(iter))
2765 return POLLIN | POLLRDNORM;
2766 poll_wait(filp, &trace_wait, poll_table);
2767 if (!trace_empty(iter))
2768 return POLLIN | POLLRDNORM;
2775 void default_wait_pipe(struct trace_iterator *iter)
2779 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2781 if (trace_empty(iter))
2784 finish_wait(&trace_wait, &wait);
2788 * This is a make-shift waitqueue.
2789 * A tracer might use this callback on some rare cases:
2791 * 1) the current tracer might hold the runqueue lock when it wakes up
2792 * a reader, hence a deadlock (sched, function, and function graph tracers)
2793 * 2) the function tracers, trace all functions, we don't want
2794 * the overhead of calling wake_up and friends
2795 * (and tracing them too)
2797 * Anyway, this is really very primitive wakeup.
2799 void poll_wait_pipe(struct trace_iterator *iter)
2801 set_current_state(TASK_INTERRUPTIBLE);
2802 /* sleep for 100 msecs, and try again. */
2803 schedule_timeout(HZ / 10);
2806 /* Must be called with trace_types_lock mutex held. */
2807 static int tracing_wait_pipe(struct file *filp)
2809 struct trace_iterator *iter = filp->private_data;
2811 while (trace_empty(iter)) {
2813 if ((filp->f_flags & O_NONBLOCK)) {
2817 mutex_unlock(&iter->mutex);
2819 iter->trace->wait_pipe(iter);
2821 mutex_lock(&iter->mutex);
2823 if (signal_pending(current))
2827 * We block until we read something and tracing is disabled.
2828 * We still block if tracing is disabled, but we have never
2829 * read anything. This allows a user to cat this file, and
2830 * then enable tracing. But after we have read something,
2831 * we give an EOF when tracing is again disabled.
2833 * iter->pos will be 0 if we haven't read anything.
2835 if (!tracer_enabled && iter->pos)
2846 tracing_read_pipe(struct file *filp, char __user *ubuf,
2847 size_t cnt, loff_t *ppos)
2849 struct trace_iterator *iter = filp->private_data;
2850 static struct tracer *old_tracer;
2853 /* return any leftover data */
2854 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2858 trace_seq_init(&iter->seq);
2860 /* copy the tracer to avoid using a global lock all around */
2861 mutex_lock(&trace_types_lock);
2862 if (unlikely(old_tracer != current_trace && current_trace)) {
2863 old_tracer = current_trace;
2864 *iter->trace = *current_trace;
2866 mutex_unlock(&trace_types_lock);
2869 * Avoid more than one consumer on a single file descriptor
2870 * This is just a matter of traces coherency, the ring buffer itself
2873 mutex_lock(&iter->mutex);
2874 if (iter->trace->read) {
2875 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2881 sret = tracing_wait_pipe(filp);
2885 /* stop when tracing is finished */
2886 if (trace_empty(iter)) {
2891 if (cnt >= PAGE_SIZE)
2892 cnt = PAGE_SIZE - 1;
2894 /* reset all but tr, trace, and overruns */
2895 memset(&iter->seq, 0,
2896 sizeof(struct trace_iterator) -
2897 offsetof(struct trace_iterator, seq));
2900 while (find_next_entry_inc(iter) != NULL) {
2901 enum print_line_t ret;
2902 int len = iter->seq.len;
2904 ret = print_trace_line(iter);
2905 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2906 /* don't print partial lines */
2907 iter->seq.len = len;
2910 if (ret != TRACE_TYPE_NO_CONSUME)
2911 trace_consume(iter);
2913 if (iter->seq.len >= cnt)
2917 /* Now copy what we have to the user */
2918 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2919 if (iter->seq.readpos >= iter->seq.len)
2920 trace_seq_init(&iter->seq);
2923 * If there was nothing to send to user, inspite of consuming trace
2924 * entries, go back to wait for more entries.
2930 mutex_unlock(&iter->mutex);
2935 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
2936 struct pipe_buffer *buf)
2938 __free_page(buf->page);
2941 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
2944 __free_page(spd->pages[idx]);
2947 static struct pipe_buf_operations tracing_pipe_buf_ops = {
2949 .map = generic_pipe_buf_map,
2950 .unmap = generic_pipe_buf_unmap,
2951 .confirm = generic_pipe_buf_confirm,
2952 .release = tracing_pipe_buf_release,
2953 .steal = generic_pipe_buf_steal,
2954 .get = generic_pipe_buf_get,
2958 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
2963 /* Seq buffer is page-sized, exactly what we need. */
2965 count = iter->seq.len;
2966 ret = print_trace_line(iter);
2967 count = iter->seq.len - count;
2970 iter->seq.len -= count;
2973 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2974 iter->seq.len -= count;
2978 trace_consume(iter);
2980 if (!find_next_entry_inc(iter)) {
2990 static ssize_t tracing_splice_read_pipe(struct file *filp,
2992 struct pipe_inode_info *pipe,
2996 struct page *pages[PIPE_BUFFERS];
2997 struct partial_page partial[PIPE_BUFFERS];
2998 struct trace_iterator *iter = filp->private_data;
2999 struct splice_pipe_desc spd = {
3002 .nr_pages = 0, /* This gets updated below. */
3004 .ops = &tracing_pipe_buf_ops,
3005 .spd_release = tracing_spd_release_pipe,
3007 static struct tracer *old_tracer;
3012 /* copy the tracer to avoid using a global lock all around */
3013 mutex_lock(&trace_types_lock);
3014 if (unlikely(old_tracer != current_trace && current_trace)) {
3015 old_tracer = current_trace;
3016 *iter->trace = *current_trace;
3018 mutex_unlock(&trace_types_lock);
3020 mutex_lock(&iter->mutex);
3022 if (iter->trace->splice_read) {
3023 ret = iter->trace->splice_read(iter, filp,
3024 ppos, pipe, len, flags);
3029 ret = tracing_wait_pipe(filp);
3033 if (!iter->ent && !find_next_entry_inc(iter)) {
3038 /* Fill as many pages as possible. */
3039 for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3040 pages[i] = alloc_page(GFP_KERNEL);
3044 rem = tracing_fill_pipe_page(rem, iter);
3046 /* Copy the data into the page, so we can start over. */
3047 ret = trace_seq_to_buffer(&iter->seq,
3048 page_address(pages[i]),
3051 __free_page(pages[i]);
3054 partial[i].offset = 0;
3055 partial[i].len = iter->seq.len;
3057 trace_seq_init(&iter->seq);
3060 mutex_unlock(&iter->mutex);
3064 return splice_to_pipe(pipe, &spd);
3067 mutex_unlock(&iter->mutex);
3073 tracing_entries_read(struct file *filp, char __user *ubuf,
3074 size_t cnt, loff_t *ppos)
3076 struct trace_array *tr = filp->private_data;
3080 mutex_lock(&trace_types_lock);
3081 if (!ring_buffer_expanded)
3082 r = sprintf(buf, "%lu (expanded: %lu)\n",
3084 trace_buf_size >> 10);
3086 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3087 mutex_unlock(&trace_types_lock);
3089 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3093 tracing_entries_write(struct file *filp, const char __user *ubuf,
3094 size_t cnt, loff_t *ppos)
3100 if (cnt >= sizeof(buf))
3103 if (copy_from_user(&buf, ubuf, cnt))
3108 ret = strict_strtoul(buf, 10, &val);
3112 /* must have at least 1 entry */
3116 mutex_lock(&trace_types_lock);
3120 /* disable all cpu buffers */
3121 for_each_tracing_cpu(cpu) {
3122 if (global_trace.data[cpu])
3123 atomic_inc(&global_trace.data[cpu]->disabled);
3124 if (max_tr.data[cpu])
3125 atomic_inc(&max_tr.data[cpu]->disabled);
3128 /* value is in KB */
3131 if (val != global_trace.entries) {
3132 ret = tracing_resize_ring_buffer(val);
3141 /* If check pages failed, return ENOMEM */
3142 if (tracing_disabled)
3145 for_each_tracing_cpu(cpu) {
3146 if (global_trace.data[cpu])
3147 atomic_dec(&global_trace.data[cpu]->disabled);
3148 if (max_tr.data[cpu])
3149 atomic_dec(&max_tr.data[cpu]->disabled);
3153 max_tr.entries = global_trace.entries;
3154 mutex_unlock(&trace_types_lock);
3159 static int mark_printk(const char *fmt, ...)
3163 va_start(args, fmt);
3164 ret = trace_vprintk(0, fmt, args);
3170 tracing_mark_write(struct file *filp, const char __user *ubuf,
3171 size_t cnt, loff_t *fpos)
3176 if (tracing_disabled)
3179 if (cnt > TRACE_BUF_SIZE)
3180 cnt = TRACE_BUF_SIZE;
3182 buf = kmalloc(cnt + 1, GFP_KERNEL);
3186 if (copy_from_user(buf, ubuf, cnt)) {
3191 /* Cut from the first nil or newline. */
3193 end = strchr(buf, '\n');
3197 cnt = mark_printk("%s\n", buf);
3204 static const struct file_operations tracing_max_lat_fops = {
3205 .open = tracing_open_generic,
3206 .read = tracing_max_lat_read,
3207 .write = tracing_max_lat_write,
3210 static const struct file_operations tracing_ctrl_fops = {
3211 .open = tracing_open_generic,
3212 .read = tracing_ctrl_read,
3213 .write = tracing_ctrl_write,
3216 static const struct file_operations set_tracer_fops = {
3217 .open = tracing_open_generic,
3218 .read = tracing_set_trace_read,
3219 .write = tracing_set_trace_write,
3222 static const struct file_operations tracing_pipe_fops = {
3223 .open = tracing_open_pipe,
3224 .poll = tracing_poll_pipe,
3225 .read = tracing_read_pipe,
3226 .splice_read = tracing_splice_read_pipe,
3227 .release = tracing_release_pipe,
3230 static const struct file_operations tracing_entries_fops = {
3231 .open = tracing_open_generic,
3232 .read = tracing_entries_read,
3233 .write = tracing_entries_write,
3236 static const struct file_operations tracing_mark_fops = {
3237 .open = tracing_open_generic,
3238 .write = tracing_mark_write,
3241 struct ftrace_buffer_info {
3242 struct trace_array *tr;
3248 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3250 int cpu = (int)(long)inode->i_private;
3251 struct ftrace_buffer_info *info;
3253 if (tracing_disabled)
3256 info = kzalloc(sizeof(*info), GFP_KERNEL);
3260 info->tr = &global_trace;
3262 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3263 /* Force reading ring buffer for first read */
3264 info->read = (unsigned int)-1;
3268 filp->private_data = info;
3278 tracing_buffers_read(struct file *filp, char __user *ubuf,
3279 size_t count, loff_t *ppos)
3281 struct ftrace_buffer_info *info = filp->private_data;
3289 /* Do we have previous read data to read? */
3290 if (info->read < PAGE_SIZE)
3295 ret = ring_buffer_read_page(info->tr->buffer,
3302 pos = ring_buffer_page_len(info->spare);
3304 if (pos < PAGE_SIZE)
3305 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3308 size = PAGE_SIZE - info->read;
3312 ret = copy_to_user(ubuf, info->spare + info->read, size);
3323 static int tracing_buffers_release(struct inode *inode, struct file *file)
3325 struct ftrace_buffer_info *info = file->private_data;
3327 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3334 struct ring_buffer *buffer;
3339 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3340 struct pipe_buffer *buf)
3342 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3347 ring_buffer_free_read_page(ref->buffer, ref->page);
3352 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3353 struct pipe_buffer *buf)
3358 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3359 struct pipe_buffer *buf)
3361 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3366 /* Pipe buffer operations for a buffer. */
3367 static struct pipe_buf_operations buffer_pipe_buf_ops = {
3369 .map = generic_pipe_buf_map,
3370 .unmap = generic_pipe_buf_unmap,
3371 .confirm = generic_pipe_buf_confirm,
3372 .release = buffer_pipe_buf_release,
3373 .steal = buffer_pipe_buf_steal,
3374 .get = buffer_pipe_buf_get,
3378 * Callback from splice_to_pipe(), if we need to release some pages
3379 * at the end of the spd in case we error'ed out in filling the pipe.
3381 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3383 struct buffer_ref *ref =
3384 (struct buffer_ref *)spd->partial[i].private;
3389 ring_buffer_free_read_page(ref->buffer, ref->page);
3391 spd->partial[i].private = 0;
3395 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3396 struct pipe_inode_info *pipe, size_t len,
3399 struct ftrace_buffer_info *info = file->private_data;
3400 struct partial_page partial[PIPE_BUFFERS];
3401 struct page *pages[PIPE_BUFFERS];
3402 struct splice_pipe_desc spd = {
3406 .ops = &buffer_pipe_buf_ops,
3407 .spd_release = buffer_spd_release,
3409 struct buffer_ref *ref;
3414 * We can't seek on a buffer input
3416 if (unlikely(*ppos))
3420 for (i = 0; i < PIPE_BUFFERS && len; i++, len -= size) {
3424 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3428 ref->buffer = info->tr->buffer;
3429 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3435 r = ring_buffer_read_page(ref->buffer, &ref->page,
3438 ring_buffer_free_read_page(ref->buffer,
3445 * zero out any left over data, this is going to
3448 size = ring_buffer_page_len(ref->page);
3449 if (size < PAGE_SIZE)
3450 memset(ref->page + size, 0, PAGE_SIZE - size);
3452 page = virt_to_page(ref->page);
3454 spd.pages[i] = page;
3455 spd.partial[i].len = PAGE_SIZE;
3456 spd.partial[i].offset = 0;
3457 spd.partial[i].private = (unsigned long)ref;
3463 /* did we read anything? */
3464 if (!spd.nr_pages) {
3465 if (flags & SPLICE_F_NONBLOCK)
3473 ret = splice_to_pipe(pipe, &spd);
3478 static const struct file_operations tracing_buffers_fops = {
3479 .open = tracing_buffers_open,
3480 .read = tracing_buffers_read,
3481 .release = tracing_buffers_release,
3482 .splice_read = tracing_buffers_splice_read,
3483 .llseek = no_llseek,
3486 #ifdef CONFIG_DYNAMIC_FTRACE
3488 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3494 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3495 size_t cnt, loff_t *ppos)
3497 static char ftrace_dyn_info_buffer[1024];
3498 static DEFINE_MUTEX(dyn_info_mutex);
3499 unsigned long *p = filp->private_data;
3500 char *buf = ftrace_dyn_info_buffer;
3501 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3504 mutex_lock(&dyn_info_mutex);
3505 r = sprintf(buf, "%ld ", *p);
3507 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3510 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3512 mutex_unlock(&dyn_info_mutex);
3517 static const struct file_operations tracing_dyn_info_fops = {
3518 .open = tracing_open_generic,
3519 .read = tracing_read_dyn_info,
3523 static struct dentry *d_tracer;
3525 struct dentry *tracing_init_dentry(void)
3532 d_tracer = debugfs_create_dir("tracing", NULL);
3534 if (!d_tracer && !once) {
3536 pr_warning("Could not create debugfs directory 'tracing'\n");
3543 static struct dentry *d_percpu;
3545 struct dentry *tracing_dentry_percpu(void)
3548 struct dentry *d_tracer;
3553 d_tracer = tracing_init_dentry();
3558 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3560 if (!d_percpu && !once) {
3562 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3569 static void tracing_init_debugfs_percpu(long cpu)
3571 struct dentry *d_percpu = tracing_dentry_percpu();
3572 struct dentry *entry, *d_cpu;
3573 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3576 if (cpu > 999 || cpu < 0)
3579 sprintf(cpu_dir, "cpu%ld", cpu);
3580 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3582 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3586 /* per cpu trace_pipe */
3587 entry = debugfs_create_file("trace_pipe", 0444, d_cpu,
3588 (void *) cpu, &tracing_pipe_fops);
3590 pr_warning("Could not create debugfs 'trace_pipe' entry\n");
3593 entry = debugfs_create_file("trace", 0644, d_cpu,
3594 (void *) cpu, &tracing_fops);
3596 pr_warning("Could not create debugfs 'trace' entry\n");
3598 entry = debugfs_create_file("trace_pipe_raw", 0444, d_cpu,
3599 (void *) cpu, &tracing_buffers_fops);
3601 pr_warning("Could not create debugfs 'trace_pipe_raw' entry\n");
3604 #ifdef CONFIG_FTRACE_SELFTEST
3605 /* Let selftest have access to static functions in this file */
3606 #include "trace_selftest.c"
3609 struct trace_option_dentry {
3610 struct tracer_opt *opt;
3611 struct tracer_flags *flags;
3612 struct dentry *entry;
3616 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3619 struct trace_option_dentry *topt = filp->private_data;
3622 if (topt->flags->val & topt->opt->bit)
3627 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3631 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3634 struct trace_option_dentry *topt = filp->private_data;
3639 if (cnt >= sizeof(buf))
3642 if (copy_from_user(&buf, ubuf, cnt))
3647 ret = strict_strtoul(buf, 10, &val);
3654 /* do nothing if already cleared */
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;
3668 /* do nothing if already set */
3669 if (topt->flags->val & topt->opt->bit)
3672 mutex_lock(&trace_types_lock);
3673 if (current_trace->set_flag)
3674 ret = current_trace->set_flag(topt->flags->val,
3676 mutex_unlock(&trace_types_lock);
3679 topt->flags->val |= topt->opt->bit;
3692 static const struct file_operations trace_options_fops = {
3693 .open = tracing_open_generic,
3694 .read = trace_options_read,
3695 .write = trace_options_write,
3699 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
3702 long index = (long)filp->private_data;
3705 if (trace_flags & (1 << index))
3710 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3714 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
3717 long index = (long)filp->private_data;
3722 if (cnt >= sizeof(buf))
3725 if (copy_from_user(&buf, ubuf, cnt))
3730 ret = strict_strtoul(buf, 10, &val);
3736 trace_flags &= ~(1 << index);
3739 trace_flags |= 1 << index;
3751 static const struct file_operations trace_options_core_fops = {
3752 .open = tracing_open_generic,
3753 .read = trace_options_core_read,
3754 .write = trace_options_core_write,
3757 static struct dentry *trace_options_init_dentry(void)
3759 struct dentry *d_tracer;
3760 static struct dentry *t_options;
3765 d_tracer = tracing_init_dentry();
3769 t_options = debugfs_create_dir("options", d_tracer);
3771 pr_warning("Could not create debugfs directory 'options'\n");
3779 create_trace_option_file(struct trace_option_dentry *topt,
3780 struct tracer_flags *flags,
3781 struct tracer_opt *opt)
3783 struct dentry *t_options;
3784 struct dentry *entry;
3786 t_options = trace_options_init_dentry();
3790 topt->flags = flags;
3793 entry = debugfs_create_file(opt->name, 0644, t_options, topt,
3794 &trace_options_fops);
3796 topt->entry = entry;
3800 static struct trace_option_dentry *
3801 create_trace_option_files(struct tracer *tracer)
3803 struct trace_option_dentry *topts;
3804 struct tracer_flags *flags;
3805 struct tracer_opt *opts;
3811 flags = tracer->flags;
3813 if (!flags || !flags->opts)
3818 for (cnt = 0; opts[cnt].name; cnt++)
3821 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
3825 for (cnt = 0; opts[cnt].name; cnt++)
3826 create_trace_option_file(&topts[cnt], flags,
3833 destroy_trace_option_files(struct trace_option_dentry *topts)
3840 for (cnt = 0; topts[cnt].opt; cnt++) {
3841 if (topts[cnt].entry)
3842 debugfs_remove(topts[cnt].entry);
3848 static struct dentry *
3849 create_trace_option_core_file(const char *option, long index)
3851 struct dentry *t_options;
3852 struct dentry *entry;
3854 t_options = trace_options_init_dentry();
3858 entry = debugfs_create_file(option, 0644, t_options, (void *)index,
3859 &trace_options_core_fops);
3864 static __init void create_trace_options_dir(void)
3866 struct dentry *t_options;
3867 struct dentry *entry;
3870 t_options = trace_options_init_dentry();
3874 for (i = 0; trace_options[i]; i++) {
3875 entry = create_trace_option_core_file(trace_options[i], i);
3877 pr_warning("Could not create debugfs %s entry\n",
3882 static __init int tracer_init_debugfs(void)
3884 struct dentry *d_tracer;
3885 struct dentry *entry;
3888 d_tracer = tracing_init_dentry();
3890 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3891 &global_trace, &tracing_ctrl_fops);
3893 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3895 entry = debugfs_create_file("trace_options", 0644, d_tracer,
3896 NULL, &tracing_iter_fops);
3898 pr_warning("Could not create debugfs 'trace_options' entry\n");
3900 create_trace_options_dir();
3902 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3903 NULL, &tracing_cpumask_fops);
3905 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3907 entry = debugfs_create_file("trace", 0644, d_tracer,
3908 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
3910 pr_warning("Could not create debugfs 'trace' entry\n");
3912 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3913 &global_trace, &show_traces_fops);
3915 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3917 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3918 &global_trace, &set_tracer_fops);
3920 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3922 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3923 &tracing_max_latency,
3924 &tracing_max_lat_fops);
3926 pr_warning("Could not create debugfs "
3927 "'tracing_max_latency' entry\n");
3929 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3930 &tracing_thresh, &tracing_max_lat_fops);
3932 pr_warning("Could not create debugfs "
3933 "'tracing_thresh' entry\n");
3934 entry = debugfs_create_file("README", 0644, d_tracer,
3935 NULL, &tracing_readme_fops);
3937 pr_warning("Could not create debugfs 'README' entry\n");
3939 entry = debugfs_create_file("trace_pipe", 0444, d_tracer,
3940 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
3942 pr_warning("Could not create debugfs "
3943 "'trace_pipe' entry\n");
3945 entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
3946 &global_trace, &tracing_entries_fops);
3948 pr_warning("Could not create debugfs "
3949 "'buffer_size_kb' entry\n");
3951 entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3952 NULL, &tracing_mark_fops);
3954 pr_warning("Could not create debugfs "
3955 "'trace_marker' entry\n");
3957 #ifdef CONFIG_DYNAMIC_FTRACE
3958 entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3959 &ftrace_update_tot_cnt,
3960 &tracing_dyn_info_fops);
3962 pr_warning("Could not create debugfs "
3963 "'dyn_ftrace_total_info' entry\n");
3965 #ifdef CONFIG_SYSPROF_TRACER
3966 init_tracer_sysprof_debugfs(d_tracer);
3969 for_each_tracing_cpu(cpu)
3970 tracing_init_debugfs_percpu(cpu);
3975 static int trace_panic_handler(struct notifier_block *this,
3976 unsigned long event, void *unused)
3978 if (ftrace_dump_on_oops)
3983 static struct notifier_block trace_panic_notifier = {
3984 .notifier_call = trace_panic_handler,
3986 .priority = 150 /* priority: INT_MAX >= x >= 0 */
3989 static int trace_die_handler(struct notifier_block *self,
3995 if (ftrace_dump_on_oops)
4004 static struct notifier_block trace_die_notifier = {
4005 .notifier_call = trace_die_handler,
4010 * printk is set to max of 1024, we really don't need it that big.
4011 * Nothing should be printing 1000 characters anyway.
4013 #define TRACE_MAX_PRINT 1000
4016 * Define here KERN_TRACE so that we have one place to modify
4017 * it if we decide to change what log level the ftrace dump
4020 #define KERN_TRACE KERN_EMERG
4023 trace_printk_seq(struct trace_seq *s)
4025 /* Probably should print a warning here. */
4029 /* should be zero ended, but we are paranoid. */
4030 s->buffer[s->len] = 0;
4032 printk(KERN_TRACE "%s", s->buffer);
4037 static void __ftrace_dump(bool disable_tracing)
4039 static DEFINE_SPINLOCK(ftrace_dump_lock);
4040 /* use static because iter can be a bit big for the stack */
4041 static struct trace_iterator iter;
4042 unsigned int old_userobj;
4043 static int dump_ran;
4044 unsigned long flags;
4048 spin_lock_irqsave(&ftrace_dump_lock, flags);
4056 if (disable_tracing)
4059 for_each_tracing_cpu(cpu) {
4060 atomic_inc(&global_trace.data[cpu]->disabled);
4063 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4065 /* don't look at user memory in panic mode */
4066 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4068 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4070 /* Simulate the iterator */
4071 iter.tr = &global_trace;
4072 iter.trace = current_trace;
4073 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4076 * We need to stop all tracing on all CPUS to read the
4077 * the next buffer. This is a bit expensive, but is
4078 * not done often. We fill all what we can read,
4079 * and then release the locks again.
4082 while (!trace_empty(&iter)) {
4085 printk(KERN_TRACE "---------------------------------\n");
4089 /* reset all but tr, trace, and overruns */
4090 memset(&iter.seq, 0,
4091 sizeof(struct trace_iterator) -
4092 offsetof(struct trace_iterator, seq));
4093 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4096 if (find_next_entry_inc(&iter) != NULL) {
4097 print_trace_line(&iter);
4098 trace_consume(&iter);
4101 trace_printk_seq(&iter.seq);
4105 printk(KERN_TRACE " (ftrace buffer empty)\n");
4107 printk(KERN_TRACE "---------------------------------\n");
4109 /* Re-enable tracing if requested */
4110 if (!disable_tracing) {
4111 trace_flags |= old_userobj;
4113 for_each_tracing_cpu(cpu) {
4114 atomic_dec(&global_trace.data[cpu]->disabled);
4120 spin_unlock_irqrestore(&ftrace_dump_lock, flags);
4123 /* By default: disable tracing after the dump */
4124 void ftrace_dump(void)
4126 __ftrace_dump(true);
4129 __init static int tracer_alloc_buffers(void)
4131 struct trace_array_cpu *data;
4136 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4139 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4140 goto out_free_buffer_mask;
4142 if (!alloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
4143 goto out_free_tracing_cpumask;
4145 /* To save memory, keep the ring buffer size to its minimum */
4146 if (ring_buffer_expanded)
4147 ring_buf_size = trace_buf_size;
4151 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4152 cpumask_copy(tracing_cpumask, cpu_all_mask);
4153 cpumask_clear(tracing_reader_cpumask);
4155 /* TODO: make the number of buffers hot pluggable with CPUS */
4156 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4157 TRACE_BUFFER_FLAGS);
4158 if (!global_trace.buffer) {
4159 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4161 goto out_free_cpumask;
4163 global_trace.entries = ring_buffer_size(global_trace.buffer);
4166 #ifdef CONFIG_TRACER_MAX_TRACE
4167 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4168 TRACE_BUFFER_FLAGS);
4169 if (!max_tr.buffer) {
4170 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4172 ring_buffer_free(global_trace.buffer);
4173 goto out_free_cpumask;
4175 max_tr.entries = ring_buffer_size(max_tr.buffer);
4176 WARN_ON(max_tr.entries != global_trace.entries);
4179 /* Allocate the first page for all buffers */
4180 for_each_tracing_cpu(i) {
4181 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4182 max_tr.data[i] = &per_cpu(max_data, i);
4185 trace_init_cmdlines();
4187 register_tracer(&nop_trace);
4188 current_trace = &nop_trace;
4189 #ifdef CONFIG_BOOT_TRACER
4190 register_tracer(&boot_tracer);
4192 /* All seems OK, enable tracing */
4193 tracing_disabled = 0;
4195 atomic_notifier_chain_register(&panic_notifier_list,
4196 &trace_panic_notifier);
4198 register_die_notifier(&trace_die_notifier);
4203 free_cpumask_var(tracing_reader_cpumask);
4204 out_free_tracing_cpumask:
4205 free_cpumask_var(tracing_cpumask);
4206 out_free_buffer_mask:
4207 free_cpumask_var(tracing_buffer_mask);
4212 __init static int clear_boot_tracer(void)
4215 * The default tracer at boot buffer is an init section.
4216 * This function is called in lateinit. If we did not
4217 * find the boot tracer, then clear it out, to prevent
4218 * later registration from accessing the buffer that is
4219 * about to be freed.
4221 if (!default_bootup_tracer)
4224 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4225 default_bootup_tracer);
4226 default_bootup_tracer = NULL;
4231 early_initcall(tracer_alloc_buffers);
4232 fs_initcall(tracer_init_debugfs);
4233 late_initcall(clear_boot_tracer);