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[] = {
322 * ftrace_max_lock is used to protect the swapping of buffers
323 * when taking a max snapshot. The buffers themselves are
324 * protected by per_cpu spinlocks. But the action of the swap
325 * needs its own lock.
327 * This is defined as a raw_spinlock_t in order to help
328 * with performance when lockdep debugging is enabled.
330 static raw_spinlock_t ftrace_max_lock =
331 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
334 * Copy the new maximum trace into the separate maximum-trace
335 * structure. (this way the maximum trace is permanently saved,
336 * for later retrieval via /debugfs/tracing/latency_trace)
339 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
341 struct trace_array_cpu *data = tr->data[cpu];
344 max_tr.time_start = data->preempt_timestamp;
346 data = max_tr.data[cpu];
347 data->saved_latency = tracing_max_latency;
349 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
350 data->pid = tsk->pid;
351 data->uid = task_uid(tsk);
352 data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
353 data->policy = tsk->policy;
354 data->rt_priority = tsk->rt_priority;
356 /* record this tasks comm */
357 tracing_record_cmdline(tsk);
360 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
368 if (s->len <= s->readpos)
371 len = s->len - s->readpos;
374 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
384 ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
389 if (s->len <= s->readpos)
392 len = s->len - s->readpos;
395 ret = memcpy(buf, s->buffer + s->readpos, cnt);
404 trace_print_seq(struct seq_file *m, struct trace_seq *s)
406 int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
409 seq_puts(m, s->buffer);
415 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
417 * @tsk: the task with the latency
418 * @cpu: The cpu that initiated the trace.
420 * Flip the buffers between the @tr and the max_tr and record information
421 * about which task was the cause of this latency.
424 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
426 struct ring_buffer *buf = tr->buffer;
428 WARN_ON_ONCE(!irqs_disabled());
429 __raw_spin_lock(&ftrace_max_lock);
431 tr->buffer = max_tr.buffer;
434 ftrace_disable_cpu();
435 ring_buffer_reset(tr->buffer);
438 __update_max_tr(tr, tsk, cpu);
439 __raw_spin_unlock(&ftrace_max_lock);
443 * update_max_tr_single - only copy one trace over, and reset the rest
445 * @tsk - task with the latency
446 * @cpu - the cpu of the buffer to copy.
448 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
451 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
455 WARN_ON_ONCE(!irqs_disabled());
456 __raw_spin_lock(&ftrace_max_lock);
458 ftrace_disable_cpu();
460 ring_buffer_reset(max_tr.buffer);
461 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
465 WARN_ON_ONCE(ret && ret != -EAGAIN);
467 __update_max_tr(tr, tsk, cpu);
468 __raw_spin_unlock(&ftrace_max_lock);
472 * register_tracer - register a tracer with the ftrace system.
473 * @type - the plugin for the tracer
475 * Register a new plugin tracer.
477 int register_tracer(struct tracer *type)
478 __releases(kernel_lock)
479 __acquires(kernel_lock)
486 pr_info("Tracer must have a name\n");
491 * When this gets called we hold the BKL which means that
492 * preemption is disabled. Various trace selftests however
493 * need to disable and enable preemption for successful tests.
494 * So we drop the BKL here and grab it after the tests again.
497 mutex_lock(&trace_types_lock);
499 tracing_selftest_running = true;
501 for (t = trace_types; t; t = t->next) {
502 if (strcmp(type->name, t->name) == 0) {
504 pr_info("Trace %s already registered\n",
512 type->set_flag = &dummy_set_flag;
514 type->flags = &dummy_tracer_flags;
516 if (!type->flags->opts)
517 type->flags->opts = dummy_tracer_opt;
518 if (!type->wait_pipe)
519 type->wait_pipe = default_wait_pipe;
522 #ifdef CONFIG_FTRACE_STARTUP_TEST
523 if (type->selftest && !tracing_selftest_disabled) {
524 struct tracer *saved_tracer = current_trace;
525 struct trace_array *tr = &global_trace;
529 * Run a selftest on this tracer.
530 * Here we reset the trace buffer, and set the current
531 * tracer to be this tracer. The tracer can then run some
532 * internal tracing to verify that everything is in order.
533 * If we fail, we do not register this tracer.
535 for_each_tracing_cpu(i)
536 tracing_reset(tr, i);
538 current_trace = type;
539 /* the test is responsible for initializing and enabling */
540 pr_info("Testing tracer %s: ", type->name);
541 ret = type->selftest(type, tr);
542 /* the test is responsible for resetting too */
543 current_trace = saved_tracer;
545 printk(KERN_CONT "FAILED!\n");
548 /* Only reset on passing, to avoid touching corrupted buffers */
549 for_each_tracing_cpu(i)
550 tracing_reset(tr, i);
552 printk(KERN_CONT "PASSED\n");
556 type->next = trace_types;
558 len = strlen(type->name);
559 if (len > max_tracer_type_len)
560 max_tracer_type_len = len;
563 tracing_selftest_running = false;
564 mutex_unlock(&trace_types_lock);
566 if (ret || !default_bootup_tracer)
569 if (strncmp(default_bootup_tracer, type->name, BOOTUP_TRACER_SIZE))
572 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
573 /* Do we want this tracer to start on bootup? */
574 tracing_set_tracer(type->name);
575 default_bootup_tracer = NULL;
576 /* disable other selftests, since this will break it. */
577 tracing_selftest_disabled = 1;
578 #ifdef CONFIG_FTRACE_STARTUP_TEST
579 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
588 void unregister_tracer(struct tracer *type)
593 mutex_lock(&trace_types_lock);
594 for (t = &trace_types; *t; t = &(*t)->next) {
598 pr_info("Trace %s not registered\n", type->name);
604 if (type == current_trace && tracer_enabled) {
607 if (current_trace->stop)
608 current_trace->stop(&global_trace);
609 current_trace = &nop_trace;
612 if (strlen(type->name) != max_tracer_type_len)
615 max_tracer_type_len = 0;
616 for (t = &trace_types; *t; t = &(*t)->next) {
617 len = strlen((*t)->name);
618 if (len > max_tracer_type_len)
619 max_tracer_type_len = len;
622 mutex_unlock(&trace_types_lock);
625 void tracing_reset(struct trace_array *tr, int cpu)
627 ftrace_disable_cpu();
628 ring_buffer_reset_cpu(tr->buffer, cpu);
632 void tracing_reset_online_cpus(struct trace_array *tr)
636 tr->time_start = ftrace_now(tr->cpu);
638 for_each_online_cpu(cpu)
639 tracing_reset(tr, cpu);
642 #define SAVED_CMDLINES 128
643 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
644 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
645 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
646 static int cmdline_idx;
647 static raw_spinlock_t trace_cmdline_lock = __RAW_SPIN_LOCK_UNLOCKED;
649 /* temporary disable recording */
650 static atomic_t trace_record_cmdline_disabled __read_mostly;
652 static void trace_init_cmdlines(void)
654 memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
655 memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
659 static int trace_stop_count;
660 static DEFINE_SPINLOCK(tracing_start_lock);
663 * ftrace_off_permanent - disable all ftrace code permanently
665 * This should only be called when a serious anomally has
666 * been detected. This will turn off the function tracing,
667 * ring buffers, and other tracing utilites. It takes no
668 * locks and can be called from any context.
670 void ftrace_off_permanent(void)
672 tracing_disabled = 1;
674 tracing_off_permanent();
678 * tracing_start - quick start of the tracer
680 * If tracing is enabled but was stopped by tracing_stop,
681 * this will start the tracer back up.
683 void tracing_start(void)
685 struct ring_buffer *buffer;
688 if (tracing_disabled)
691 spin_lock_irqsave(&tracing_start_lock, flags);
692 if (--trace_stop_count) {
693 if (trace_stop_count < 0) {
694 /* Someone screwed up their debugging */
696 trace_stop_count = 0;
702 buffer = global_trace.buffer;
704 ring_buffer_record_enable(buffer);
706 buffer = max_tr.buffer;
708 ring_buffer_record_enable(buffer);
712 spin_unlock_irqrestore(&tracing_start_lock, flags);
716 * tracing_stop - quick stop of the tracer
718 * Light weight way to stop tracing. Use in conjunction with
721 void tracing_stop(void)
723 struct ring_buffer *buffer;
727 spin_lock_irqsave(&tracing_start_lock, flags);
728 if (trace_stop_count++)
731 buffer = global_trace.buffer;
733 ring_buffer_record_disable(buffer);
735 buffer = max_tr.buffer;
737 ring_buffer_record_disable(buffer);
740 spin_unlock_irqrestore(&tracing_start_lock, flags);
743 void trace_stop_cmdline_recording(void);
745 static void trace_save_cmdline(struct task_struct *tsk)
750 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
754 * It's not the end of the world if we don't get
755 * the lock, but we also don't want to spin
756 * nor do we want to disable interrupts,
757 * so if we miss here, then better luck next time.
759 if (!__raw_spin_trylock(&trace_cmdline_lock))
762 idx = map_pid_to_cmdline[tsk->pid];
763 if (idx >= SAVED_CMDLINES) {
764 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
766 map = map_cmdline_to_pid[idx];
767 if (map <= PID_MAX_DEFAULT)
768 map_pid_to_cmdline[map] = (unsigned)-1;
770 map_pid_to_cmdline[tsk->pid] = idx;
775 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
777 __raw_spin_unlock(&trace_cmdline_lock);
780 void trace_find_cmdline(int pid, char comm[])
785 strcpy(comm, "<idle>");
789 if (pid > PID_MAX_DEFAULT) {
790 strcpy(comm, "<...>");
794 __raw_spin_lock(&trace_cmdline_lock);
795 map = map_pid_to_cmdline[pid];
796 if (map >= SAVED_CMDLINES)
799 strcpy(comm, saved_cmdlines[map]);
802 __raw_spin_unlock(&trace_cmdline_lock);
805 void tracing_record_cmdline(struct task_struct *tsk)
807 if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on())
810 trace_save_cmdline(tsk);
814 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
817 struct task_struct *tsk = current;
819 entry->preempt_count = pc & 0xff;
820 entry->pid = (tsk) ? tsk->pid : 0;
821 entry->tgid = (tsk) ? tsk->tgid : 0;
823 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
824 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
826 TRACE_FLAG_IRQS_NOSUPPORT |
828 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
829 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
830 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
833 struct ring_buffer_event *trace_buffer_lock_reserve(struct trace_array *tr,
836 unsigned long flags, int pc)
838 struct ring_buffer_event *event;
840 event = ring_buffer_lock_reserve(tr->buffer, len);
842 struct trace_entry *ent = ring_buffer_event_data(event);
844 tracing_generic_entry_update(ent, flags, pc);
850 static void ftrace_trace_stack(struct trace_array *tr,
851 unsigned long flags, int skip, int pc);
852 static void ftrace_trace_userstack(struct trace_array *tr,
853 unsigned long flags, int pc);
855 void trace_buffer_unlock_commit(struct trace_array *tr,
856 struct ring_buffer_event *event,
857 unsigned long flags, int pc)
859 ring_buffer_unlock_commit(tr->buffer, event);
861 ftrace_trace_stack(tr, flags, 6, pc);
862 ftrace_trace_userstack(tr, flags, pc);
866 struct ring_buffer_event *
867 trace_current_buffer_lock_reserve(unsigned char type, unsigned long len,
868 unsigned long flags, int pc)
870 return trace_buffer_lock_reserve(&global_trace,
871 type, len, flags, pc);
874 void trace_current_buffer_unlock_commit(struct ring_buffer_event *event,
875 unsigned long flags, int pc)
877 return trace_buffer_unlock_commit(&global_trace, event, flags, pc);
881 trace_function(struct trace_array *tr,
882 unsigned long ip, unsigned long parent_ip, unsigned long flags,
885 struct ring_buffer_event *event;
886 struct ftrace_entry *entry;
888 /* If we are reading the ring buffer, don't trace */
889 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
892 event = trace_buffer_lock_reserve(tr, TRACE_FN, sizeof(*entry),
896 entry = ring_buffer_event_data(event);
898 entry->parent_ip = parent_ip;
899 ring_buffer_unlock_commit(tr->buffer, event);
902 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
903 static void __trace_graph_entry(struct trace_array *tr,
904 struct ftrace_graph_ent *trace,
908 struct ring_buffer_event *event;
909 struct ftrace_graph_ent_entry *entry;
911 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
914 event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_ENT,
915 sizeof(*entry), flags, pc);
918 entry = ring_buffer_event_data(event);
919 entry->graph_ent = *trace;
920 ring_buffer_unlock_commit(global_trace.buffer, event);
923 static void __trace_graph_return(struct trace_array *tr,
924 struct ftrace_graph_ret *trace,
928 struct ring_buffer_event *event;
929 struct ftrace_graph_ret_entry *entry;
931 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
934 event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_RET,
935 sizeof(*entry), flags, pc);
938 entry = ring_buffer_event_data(event);
940 ring_buffer_unlock_commit(global_trace.buffer, event);
945 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
946 unsigned long ip, unsigned long parent_ip, unsigned long flags,
949 if (likely(!atomic_read(&data->disabled)))
950 trace_function(tr, ip, parent_ip, flags, pc);
953 static void __ftrace_trace_stack(struct trace_array *tr,
957 #ifdef CONFIG_STACKTRACE
958 struct ring_buffer_event *event;
959 struct stack_entry *entry;
960 struct stack_trace trace;
962 event = trace_buffer_lock_reserve(tr, TRACE_STACK,
963 sizeof(*entry), flags, pc);
966 entry = ring_buffer_event_data(event);
967 memset(&entry->caller, 0, sizeof(entry->caller));
969 trace.nr_entries = 0;
970 trace.max_entries = FTRACE_STACK_ENTRIES;
972 trace.entries = entry->caller;
974 save_stack_trace(&trace);
975 ring_buffer_unlock_commit(tr->buffer, event);
979 static void ftrace_trace_stack(struct trace_array *tr,
983 if (!(trace_flags & TRACE_ITER_STACKTRACE))
986 __ftrace_trace_stack(tr, flags, skip, pc);
989 void __trace_stack(struct trace_array *tr,
993 __ftrace_trace_stack(tr, flags, skip, pc);
996 static void ftrace_trace_userstack(struct trace_array *tr,
997 unsigned long flags, int pc)
999 #ifdef CONFIG_STACKTRACE
1000 struct ring_buffer_event *event;
1001 struct userstack_entry *entry;
1002 struct stack_trace trace;
1004 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1007 event = trace_buffer_lock_reserve(tr, TRACE_USER_STACK,
1008 sizeof(*entry), flags, pc);
1011 entry = ring_buffer_event_data(event);
1013 memset(&entry->caller, 0, sizeof(entry->caller));
1015 trace.nr_entries = 0;
1016 trace.max_entries = FTRACE_STACK_ENTRIES;
1018 trace.entries = entry->caller;
1020 save_stack_trace_user(&trace);
1021 ring_buffer_unlock_commit(tr->buffer, event);
1026 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1028 ftrace_trace_userstack(tr, flags, preempt_count());
1033 ftrace_trace_special(void *__tr,
1034 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1037 struct ring_buffer_event *event;
1038 struct trace_array *tr = __tr;
1039 struct special_entry *entry;
1041 event = trace_buffer_lock_reserve(tr, TRACE_SPECIAL,
1042 sizeof(*entry), 0, pc);
1045 entry = ring_buffer_event_data(event);
1049 trace_buffer_unlock_commit(tr, event, 0, pc);
1053 __trace_special(void *__tr, void *__data,
1054 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1056 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1060 tracing_sched_switch_trace(struct trace_array *tr,
1061 struct task_struct *prev,
1062 struct task_struct *next,
1063 unsigned long flags, int pc)
1065 struct ring_buffer_event *event;
1066 struct ctx_switch_entry *entry;
1068 event = trace_buffer_lock_reserve(tr, TRACE_CTX,
1069 sizeof(*entry), flags, pc);
1072 entry = ring_buffer_event_data(event);
1073 entry->prev_pid = prev->pid;
1074 entry->prev_prio = prev->prio;
1075 entry->prev_state = prev->state;
1076 entry->next_pid = next->pid;
1077 entry->next_prio = next->prio;
1078 entry->next_state = next->state;
1079 entry->next_cpu = task_cpu(next);
1080 trace_buffer_unlock_commit(tr, event, flags, pc);
1084 tracing_sched_wakeup_trace(struct trace_array *tr,
1085 struct task_struct *wakee,
1086 struct task_struct *curr,
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_WAKE,
1093 sizeof(*entry), flags, pc);
1096 entry = ring_buffer_event_data(event);
1097 entry->prev_pid = curr->pid;
1098 entry->prev_prio = curr->prio;
1099 entry->prev_state = curr->state;
1100 entry->next_pid = wakee->pid;
1101 entry->next_prio = wakee->prio;
1102 entry->next_state = wakee->state;
1103 entry->next_cpu = task_cpu(wakee);
1105 ring_buffer_unlock_commit(tr->buffer, event);
1106 ftrace_trace_stack(tr, flags, 6, pc);
1107 ftrace_trace_userstack(tr, flags, pc);
1111 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1113 struct trace_array *tr = &global_trace;
1114 struct trace_array_cpu *data;
1115 unsigned long flags;
1119 if (tracing_disabled)
1122 pc = preempt_count();
1123 local_irq_save(flags);
1124 cpu = raw_smp_processor_id();
1125 data = tr->data[cpu];
1127 if (likely(atomic_inc_return(&data->disabled) == 1))
1128 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1130 atomic_dec(&data->disabled);
1131 local_irq_restore(flags);
1134 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1135 int trace_graph_entry(struct ftrace_graph_ent *trace)
1137 struct trace_array *tr = &global_trace;
1138 struct trace_array_cpu *data;
1139 unsigned long flags;
1144 if (!ftrace_trace_task(current))
1147 if (!ftrace_graph_addr(trace->func))
1150 local_irq_save(flags);
1151 cpu = raw_smp_processor_id();
1152 data = tr->data[cpu];
1153 disabled = atomic_inc_return(&data->disabled);
1154 if (likely(disabled == 1)) {
1155 pc = preempt_count();
1156 __trace_graph_entry(tr, trace, flags, pc);
1158 /* Only do the atomic if it is not already set */
1159 if (!test_tsk_trace_graph(current))
1160 set_tsk_trace_graph(current);
1161 atomic_dec(&data->disabled);
1162 local_irq_restore(flags);
1167 void trace_graph_return(struct ftrace_graph_ret *trace)
1169 struct trace_array *tr = &global_trace;
1170 struct trace_array_cpu *data;
1171 unsigned long flags;
1176 local_irq_save(flags);
1177 cpu = raw_smp_processor_id();
1178 data = tr->data[cpu];
1179 disabled = atomic_inc_return(&data->disabled);
1180 if (likely(disabled == 1)) {
1181 pc = preempt_count();
1182 __trace_graph_return(tr, trace, flags, pc);
1185 clear_tsk_trace_graph(current);
1186 atomic_dec(&data->disabled);
1187 local_irq_restore(flags);
1189 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1193 * trace_vbprintk - write binary msg to tracing buffer
1196 int trace_vbprintk(unsigned long ip, int depth, const char *fmt, va_list args)
1198 static raw_spinlock_t trace_buf_lock =
1199 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
1200 static u32 trace_buf[TRACE_BUF_SIZE];
1202 struct ring_buffer_event *event;
1203 struct trace_array *tr = &global_trace;
1204 struct trace_array_cpu *data;
1205 struct bprint_entry *entry;
1206 unsigned long flags;
1208 int cpu, len = 0, size, pc;
1210 if (unlikely(tracing_selftest_running || tracing_disabled))
1213 /* Don't pollute graph traces with trace_vprintk internals */
1214 pause_graph_tracing();
1216 pc = preempt_count();
1217 resched = ftrace_preempt_disable();
1218 cpu = raw_smp_processor_id();
1219 data = tr->data[cpu];
1221 if (unlikely(atomic_read(&data->disabled)))
1224 /* Lockdep uses trace_printk for lock tracing */
1225 local_irq_save(flags);
1226 __raw_spin_lock(&trace_buf_lock);
1227 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1229 if (len > TRACE_BUF_SIZE || len < 0)
1232 size = sizeof(*entry) + sizeof(u32) * len;
1233 event = trace_buffer_lock_reserve(tr, TRACE_BPRINT, size, flags, pc);
1236 entry = ring_buffer_event_data(event);
1238 entry->depth = depth;
1241 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1242 ring_buffer_unlock_commit(tr->buffer, event);
1245 __raw_spin_unlock(&trace_buf_lock);
1246 local_irq_restore(flags);
1249 ftrace_preempt_enable(resched);
1250 unpause_graph_tracing();
1254 EXPORT_SYMBOL_GPL(trace_vbprintk);
1256 int trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args)
1258 static raw_spinlock_t trace_buf_lock = __RAW_SPIN_LOCK_UNLOCKED;
1259 static char trace_buf[TRACE_BUF_SIZE];
1261 struct ring_buffer_event *event;
1262 struct trace_array *tr = &global_trace;
1263 struct trace_array_cpu *data;
1264 int cpu, len = 0, size, pc;
1265 struct print_entry *entry;
1266 unsigned long irq_flags;
1268 if (tracing_disabled || tracing_selftest_running)
1271 pc = preempt_count();
1272 preempt_disable_notrace();
1273 cpu = raw_smp_processor_id();
1274 data = tr->data[cpu];
1276 if (unlikely(atomic_read(&data->disabled)))
1279 pause_graph_tracing();
1280 raw_local_irq_save(irq_flags);
1281 __raw_spin_lock(&trace_buf_lock);
1282 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1284 len = min(len, TRACE_BUF_SIZE-1);
1287 size = sizeof(*entry) + len + 1;
1288 event = trace_buffer_lock_reserve(tr, TRACE_PRINT, size, irq_flags, pc);
1291 entry = ring_buffer_event_data(event);
1293 entry->depth = depth;
1295 memcpy(&entry->buf, trace_buf, len);
1296 entry->buf[len] = 0;
1297 ring_buffer_unlock_commit(tr->buffer, event);
1300 __raw_spin_unlock(&trace_buf_lock);
1301 raw_local_irq_restore(irq_flags);
1302 unpause_graph_tracing();
1304 preempt_enable_notrace();
1308 EXPORT_SYMBOL_GPL(trace_vprintk);
1310 enum trace_file_type {
1311 TRACE_FILE_LAT_FMT = 1,
1312 TRACE_FILE_ANNOTATE = 2,
1315 static void trace_iterator_increment(struct trace_iterator *iter)
1317 /* Don't allow ftrace to trace into the ring buffers */
1318 ftrace_disable_cpu();
1321 if (iter->buffer_iter[iter->cpu])
1322 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1324 ftrace_enable_cpu();
1327 static struct trace_entry *
1328 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1330 struct ring_buffer_event *event;
1331 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1333 /* Don't allow ftrace to trace into the ring buffers */
1334 ftrace_disable_cpu();
1337 event = ring_buffer_iter_peek(buf_iter, ts);
1339 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1341 ftrace_enable_cpu();
1343 return event ? ring_buffer_event_data(event) : NULL;
1346 static struct trace_entry *
1347 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1349 struct ring_buffer *buffer = iter->tr->buffer;
1350 struct trace_entry *ent, *next = NULL;
1351 int cpu_file = iter->cpu_file;
1352 u64 next_ts = 0, ts;
1357 * If we are in a per_cpu trace file, don't bother by iterating over
1358 * all cpu and peek directly.
1360 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1361 if (ring_buffer_empty_cpu(buffer, cpu_file))
1363 ent = peek_next_entry(iter, cpu_file, ent_ts);
1365 *ent_cpu = cpu_file;
1370 for_each_tracing_cpu(cpu) {
1372 if (ring_buffer_empty_cpu(buffer, cpu))
1375 ent = peek_next_entry(iter, cpu, &ts);
1378 * Pick the entry with the smallest timestamp:
1380 if (ent && (!next || ts < next_ts)) {
1388 *ent_cpu = next_cpu;
1396 /* Find the next real entry, without updating the iterator itself */
1397 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1398 int *ent_cpu, u64 *ent_ts)
1400 return __find_next_entry(iter, ent_cpu, ent_ts);
1403 /* Find the next real entry, and increment the iterator to the next entry */
1404 static void *find_next_entry_inc(struct trace_iterator *iter)
1406 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1409 trace_iterator_increment(iter);
1411 return iter->ent ? iter : NULL;
1414 static void trace_consume(struct trace_iterator *iter)
1416 /* Don't allow ftrace to trace into the ring buffers */
1417 ftrace_disable_cpu();
1418 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1419 ftrace_enable_cpu();
1422 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1424 struct trace_iterator *iter = m->private;
1430 /* can't go backwards */
1435 ent = find_next_entry_inc(iter);
1439 while (ent && iter->idx < i)
1440 ent = find_next_entry_inc(iter);
1448 * No necessary locking here. The worst thing which can
1449 * happen is loosing events consumed at the same time
1450 * by a trace_pipe reader.
1451 * Other than that, we don't risk to crash the ring buffer
1452 * because it serializes the readers.
1454 * The current tracer is copied to avoid a global locking
1457 static void *s_start(struct seq_file *m, loff_t *pos)
1459 struct trace_iterator *iter = m->private;
1460 static struct tracer *old_tracer;
1461 int cpu_file = iter->cpu_file;
1466 /* copy the tracer to avoid using a global lock all around */
1467 mutex_lock(&trace_types_lock);
1468 if (unlikely(old_tracer != current_trace && current_trace)) {
1469 old_tracer = current_trace;
1470 *iter->trace = *current_trace;
1472 mutex_unlock(&trace_types_lock);
1474 atomic_inc(&trace_record_cmdline_disabled);
1476 if (*pos != iter->pos) {
1481 ftrace_disable_cpu();
1483 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1484 for_each_tracing_cpu(cpu)
1485 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1487 ring_buffer_iter_reset(iter->buffer_iter[cpu_file]);
1490 ftrace_enable_cpu();
1492 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1497 p = s_next(m, p, &l);
1503 static void s_stop(struct seq_file *m, void *p)
1505 atomic_dec(&trace_record_cmdline_disabled);
1508 static void print_lat_help_header(struct seq_file *m)
1510 seq_puts(m, "# _------=> CPU# \n");
1511 seq_puts(m, "# / _-----=> irqs-off \n");
1512 seq_puts(m, "# | / _----=> need-resched \n");
1513 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1514 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1515 seq_puts(m, "# |||| / \n");
1516 seq_puts(m, "# ||||| delay \n");
1517 seq_puts(m, "# cmd pid ||||| time | caller \n");
1518 seq_puts(m, "# \\ / ||||| \\ | / \n");
1521 static void print_func_help_header(struct seq_file *m)
1523 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1524 seq_puts(m, "# | | | | |\n");
1529 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1531 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1532 struct trace_array *tr = iter->tr;
1533 struct trace_array_cpu *data = tr->data[tr->cpu];
1534 struct tracer *type = current_trace;
1535 unsigned long total;
1536 unsigned long entries;
1537 const char *name = "preemption";
1542 entries = ring_buffer_entries(iter->tr->buffer);
1544 ring_buffer_overruns(iter->tr->buffer);
1546 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1548 seq_puts(m, "# -----------------------------------"
1549 "---------------------------------\n");
1550 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1551 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1552 nsecs_to_usecs(data->saved_latency),
1556 #if defined(CONFIG_PREEMPT_NONE)
1558 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1560 #elif defined(CONFIG_PREEMPT)
1565 /* These are reserved for later use */
1568 seq_printf(m, " #P:%d)\n", num_online_cpus());
1572 seq_puts(m, "# -----------------\n");
1573 seq_printf(m, "# | task: %.16s-%d "
1574 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1575 data->comm, data->pid, data->uid, data->nice,
1576 data->policy, data->rt_priority);
1577 seq_puts(m, "# -----------------\n");
1579 if (data->critical_start) {
1580 seq_puts(m, "# => started at: ");
1581 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1582 trace_print_seq(m, &iter->seq);
1583 seq_puts(m, "\n# => ended at: ");
1584 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1585 trace_print_seq(m, &iter->seq);
1592 static void test_cpu_buff_start(struct trace_iterator *iter)
1594 struct trace_seq *s = &iter->seq;
1596 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1599 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1602 if (cpumask_test_cpu(iter->cpu, iter->started))
1605 cpumask_set_cpu(iter->cpu, iter->started);
1606 trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1609 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1611 struct trace_seq *s = &iter->seq;
1612 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1613 struct trace_entry *entry;
1614 struct trace_event *event;
1618 test_cpu_buff_start(iter);
1620 event = ftrace_find_event(entry->type);
1622 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1623 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1624 if (!trace_print_lat_context(iter))
1627 if (!trace_print_context(iter))
1633 return event->trace(iter, sym_flags);
1635 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1638 return TRACE_TYPE_HANDLED;
1640 return TRACE_TYPE_PARTIAL_LINE;
1643 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1645 struct trace_seq *s = &iter->seq;
1646 struct trace_entry *entry;
1647 struct trace_event *event;
1651 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1652 if (!trace_seq_printf(s, "%d %d %llu ",
1653 entry->pid, iter->cpu, iter->ts))
1657 event = ftrace_find_event(entry->type);
1659 return event->raw(iter, 0);
1661 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1664 return TRACE_TYPE_HANDLED;
1666 return TRACE_TYPE_PARTIAL_LINE;
1669 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1671 struct trace_seq *s = &iter->seq;
1672 unsigned char newline = '\n';
1673 struct trace_entry *entry;
1674 struct trace_event *event;
1678 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1679 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1680 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1681 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1684 event = ftrace_find_event(entry->type);
1686 enum print_line_t ret = event->hex(iter, 0);
1687 if (ret != TRACE_TYPE_HANDLED)
1691 SEQ_PUT_FIELD_RET(s, newline);
1693 return TRACE_TYPE_HANDLED;
1696 static enum print_line_t print_bprintk_msg_only(struct trace_iterator *iter)
1698 struct trace_seq *s = &iter->seq;
1699 struct trace_entry *entry = iter->ent;
1700 struct bprint_entry *field;
1703 trace_assign_type(field, entry);
1705 ret = trace_seq_bprintf(s, field->fmt, field->buf);
1707 return TRACE_TYPE_PARTIAL_LINE;
1709 return TRACE_TYPE_HANDLED;
1712 static enum print_line_t print_printk_msg_only(struct trace_iterator *iter)
1714 struct trace_seq *s = &iter->seq;
1715 struct trace_entry *entry = iter->ent;
1716 struct print_entry *field;
1719 trace_assign_type(field, entry);
1721 ret = trace_seq_printf(s, "%s", field->buf);
1723 return TRACE_TYPE_PARTIAL_LINE;
1725 return TRACE_TYPE_HANDLED;
1728 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1730 struct trace_seq *s = &iter->seq;
1731 struct trace_entry *entry;
1732 struct trace_event *event;
1736 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1737 SEQ_PUT_FIELD_RET(s, entry->pid);
1738 SEQ_PUT_FIELD_RET(s, iter->cpu);
1739 SEQ_PUT_FIELD_RET(s, iter->ts);
1742 event = ftrace_find_event(entry->type);
1743 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1746 static int trace_empty(struct trace_iterator *iter)
1750 /* If we are looking at one CPU buffer, only check that one */
1751 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1752 cpu = iter->cpu_file;
1753 if (iter->buffer_iter[cpu]) {
1754 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1757 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1763 for_each_tracing_cpu(cpu) {
1764 if (iter->buffer_iter[cpu]) {
1765 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1768 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1776 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1778 enum print_line_t ret;
1780 if (iter->trace && iter->trace->print_line) {
1781 ret = iter->trace->print_line(iter);
1782 if (ret != TRACE_TYPE_UNHANDLED)
1786 if (iter->ent->type == TRACE_BPRINT &&
1787 trace_flags & TRACE_ITER_PRINTK &&
1788 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1789 return print_bprintk_msg_only(iter);
1791 if (iter->ent->type == TRACE_PRINT &&
1792 trace_flags & TRACE_ITER_PRINTK &&
1793 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1794 return print_printk_msg_only(iter);
1796 if (trace_flags & TRACE_ITER_BIN)
1797 return print_bin_fmt(iter);
1799 if (trace_flags & TRACE_ITER_HEX)
1800 return print_hex_fmt(iter);
1802 if (trace_flags & TRACE_ITER_RAW)
1803 return print_raw_fmt(iter);
1805 return print_trace_fmt(iter);
1808 static int s_show(struct seq_file *m, void *v)
1810 struct trace_iterator *iter = v;
1812 if (iter->ent == NULL) {
1814 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1817 if (iter->trace && iter->trace->print_header)
1818 iter->trace->print_header(m);
1819 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1820 /* print nothing if the buffers are empty */
1821 if (trace_empty(iter))
1823 print_trace_header(m, iter);
1824 if (!(trace_flags & TRACE_ITER_VERBOSE))
1825 print_lat_help_header(m);
1827 if (!(trace_flags & TRACE_ITER_VERBOSE))
1828 print_func_help_header(m);
1831 print_trace_line(iter);
1832 trace_print_seq(m, &iter->seq);
1838 static struct seq_operations tracer_seq_ops = {
1845 static struct trace_iterator *
1846 __tracing_open(struct inode *inode, struct file *file)
1848 long cpu_file = (long) inode->i_private;
1849 void *fail_ret = ERR_PTR(-ENOMEM);
1850 struct trace_iterator *iter;
1854 if (tracing_disabled)
1855 return ERR_PTR(-ENODEV);
1857 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1859 return ERR_PTR(-ENOMEM);
1862 * We make a copy of the current tracer to avoid concurrent
1863 * changes on it while we are reading.
1865 mutex_lock(&trace_types_lock);
1866 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
1871 *iter->trace = *current_trace;
1873 if (current_trace && current_trace->print_max)
1876 iter->tr = &global_trace;
1878 mutex_init(&iter->mutex);
1879 iter->cpu_file = cpu_file;
1881 /* Notify the tracer early; before we stop tracing. */
1882 if (iter->trace && iter->trace->open)
1883 iter->trace->open(iter);
1885 /* Annotate start of buffers if we had overruns */
1886 if (ring_buffer_overruns(iter->tr->buffer))
1887 iter->iter_flags |= TRACE_FILE_ANNOTATE;
1889 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
1890 for_each_tracing_cpu(cpu) {
1892 iter->buffer_iter[cpu] =
1893 ring_buffer_read_start(iter->tr->buffer, cpu);
1896 cpu = iter->cpu_file;
1897 iter->buffer_iter[cpu] =
1898 ring_buffer_read_start(iter->tr->buffer, cpu);
1901 /* TODO stop tracer */
1902 ret = seq_open(file, &tracer_seq_ops);
1904 fail_ret = ERR_PTR(ret);
1908 m = file->private_data;
1911 /* stop the trace while dumping */
1914 mutex_unlock(&trace_types_lock);
1919 for_each_tracing_cpu(cpu) {
1920 if (iter->buffer_iter[cpu])
1921 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1924 mutex_unlock(&trace_types_lock);
1931 int tracing_open_generic(struct inode *inode, struct file *filp)
1933 if (tracing_disabled)
1936 filp->private_data = inode->i_private;
1940 static int tracing_release(struct inode *inode, struct file *file)
1942 struct seq_file *m = (struct seq_file *)file->private_data;
1943 struct trace_iterator *iter = m->private;
1946 mutex_lock(&trace_types_lock);
1947 for_each_tracing_cpu(cpu) {
1948 if (iter->buffer_iter[cpu])
1949 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1952 if (iter->trace && iter->trace->close)
1953 iter->trace->close(iter);
1955 /* reenable tracing if it was previously enabled */
1957 mutex_unlock(&trace_types_lock);
1959 seq_release(inode, file);
1960 mutex_destroy(&iter->mutex);
1966 static int tracing_open(struct inode *inode, struct file *file)
1968 struct trace_iterator *iter;
1971 iter = __tracing_open(inode, file);
1973 ret = PTR_ERR(iter);
1974 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
1975 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1981 t_next(struct seq_file *m, void *v, loff_t *pos)
1983 struct tracer *t = m->private;
1995 static void *t_start(struct seq_file *m, loff_t *pos)
1997 struct tracer *t = m->private;
2000 mutex_lock(&trace_types_lock);
2001 for (; t && l < *pos; t = t_next(m, t, &l))
2007 static void t_stop(struct seq_file *m, void *p)
2009 mutex_unlock(&trace_types_lock);
2012 static int t_show(struct seq_file *m, void *v)
2014 struct tracer *t = v;
2019 seq_printf(m, "%s", t->name);
2028 static struct seq_operations show_traces_seq_ops = {
2035 static int show_traces_open(struct inode *inode, struct file *file)
2039 if (tracing_disabled)
2042 ret = seq_open(file, &show_traces_seq_ops);
2044 struct seq_file *m = file->private_data;
2045 m->private = trace_types;
2051 static const struct file_operations tracing_fops = {
2052 .open = tracing_open,
2054 .llseek = seq_lseek,
2055 .release = tracing_release,
2058 static const struct file_operations show_traces_fops = {
2059 .open = show_traces_open,
2061 .release = seq_release,
2065 * Only trace on a CPU if the bitmask is set:
2067 static cpumask_var_t tracing_cpumask;
2070 * The tracer itself will not take this lock, but still we want
2071 * to provide a consistent cpumask to user-space:
2073 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2076 * Temporary storage for the character representation of the
2077 * CPU bitmask (and one more byte for the newline):
2079 static char mask_str[NR_CPUS + 1];
2082 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2083 size_t count, loff_t *ppos)
2087 mutex_lock(&tracing_cpumask_update_lock);
2089 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2090 if (count - len < 2) {
2094 len += sprintf(mask_str + len, "\n");
2095 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2098 mutex_unlock(&tracing_cpumask_update_lock);
2104 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2105 size_t count, loff_t *ppos)
2108 cpumask_var_t tracing_cpumask_new;
2110 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2113 mutex_lock(&tracing_cpumask_update_lock);
2114 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2118 local_irq_disable();
2119 __raw_spin_lock(&ftrace_max_lock);
2120 for_each_tracing_cpu(cpu) {
2122 * Increase/decrease the disabled counter if we are
2123 * about to flip a bit in the cpumask:
2125 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2126 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2127 atomic_inc(&global_trace.data[cpu]->disabled);
2129 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2130 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2131 atomic_dec(&global_trace.data[cpu]->disabled);
2134 __raw_spin_unlock(&ftrace_max_lock);
2137 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2139 mutex_unlock(&tracing_cpumask_update_lock);
2140 free_cpumask_var(tracing_cpumask_new);
2145 mutex_unlock(&tracing_cpumask_update_lock);
2146 free_cpumask_var(tracing_cpumask);
2151 static const struct file_operations tracing_cpumask_fops = {
2152 .open = tracing_open_generic,
2153 .read = tracing_cpumask_read,
2154 .write = tracing_cpumask_write,
2158 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2159 size_t cnt, loff_t *ppos)
2161 struct tracer_opt *trace_opts;
2169 /* calculate max size */
2170 for (i = 0; trace_options[i]; i++) {
2171 len += strlen(trace_options[i]);
2172 len += 3; /* "no" and newline */
2175 mutex_lock(&trace_types_lock);
2176 tracer_flags = current_trace->flags->val;
2177 trace_opts = current_trace->flags->opts;
2180 * Increase the size with names of options specific
2181 * of the current tracer.
2183 for (i = 0; trace_opts[i].name; i++) {
2184 len += strlen(trace_opts[i].name);
2185 len += 3; /* "no" and newline */
2188 /* +2 for \n and \0 */
2189 buf = kmalloc(len + 2, GFP_KERNEL);
2191 mutex_unlock(&trace_types_lock);
2195 for (i = 0; trace_options[i]; i++) {
2196 if (trace_flags & (1 << i))
2197 r += sprintf(buf + r, "%s\n", trace_options[i]);
2199 r += sprintf(buf + r, "no%s\n", trace_options[i]);
2202 for (i = 0; trace_opts[i].name; i++) {
2203 if (tracer_flags & trace_opts[i].bit)
2204 r += sprintf(buf + r, "%s\n",
2205 trace_opts[i].name);
2207 r += sprintf(buf + r, "no%s\n",
2208 trace_opts[i].name);
2210 mutex_unlock(&trace_types_lock);
2212 WARN_ON(r >= len + 2);
2214 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2220 /* Try to assign a tracer specific option */
2221 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2223 struct tracer_flags *trace_flags = trace->flags;
2224 struct tracer_opt *opts = NULL;
2228 for (i = 0; trace_flags->opts[i].name; i++) {
2229 opts = &trace_flags->opts[i];
2230 len = strlen(opts->name);
2232 if (strncmp(cmp, opts->name, len) == 0) {
2233 ret = trace->set_flag(trace_flags->val,
2239 if (!trace_flags->opts[i].name)
2242 /* Refused to handle */
2247 trace_flags->val &= ~opts->bit;
2249 trace_flags->val |= opts->bit;
2255 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2256 size_t cnt, loff_t *ppos)
2264 if (cnt >= sizeof(buf))
2267 if (copy_from_user(&buf, ubuf, cnt))
2272 if (strncmp(buf, "no", 2) == 0) {
2277 for (i = 0; trace_options[i]; i++) {
2278 int len = strlen(trace_options[i]);
2280 if (strncmp(cmp, trace_options[i], len) == 0) {
2282 trace_flags &= ~(1 << i);
2284 trace_flags |= (1 << i);
2289 /* If no option could be set, test the specific tracer options */
2290 if (!trace_options[i]) {
2291 mutex_lock(&trace_types_lock);
2292 ret = set_tracer_option(current_trace, cmp, neg);
2293 mutex_unlock(&trace_types_lock);
2303 static const struct file_operations tracing_iter_fops = {
2304 .open = tracing_open_generic,
2305 .read = tracing_trace_options_read,
2306 .write = tracing_trace_options_write,
2309 static const char readme_msg[] =
2310 "tracing mini-HOWTO:\n\n"
2312 "# mount -t debugfs nodev /debug\n\n"
2313 "# cat /debug/tracing/available_tracers\n"
2314 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2315 "# cat /debug/tracing/current_tracer\n"
2317 "# echo sched_switch > /debug/tracing/current_tracer\n"
2318 "# cat /debug/tracing/current_tracer\n"
2320 "# cat /debug/tracing/trace_options\n"
2321 "noprint-parent nosym-offset nosym-addr noverbose\n"
2322 "# echo print-parent > /debug/tracing/trace_options\n"
2323 "# echo 1 > /debug/tracing/tracing_enabled\n"
2324 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2325 "echo 0 > /debug/tracing/tracing_enabled\n"
2329 tracing_readme_read(struct file *filp, char __user *ubuf,
2330 size_t cnt, loff_t *ppos)
2332 return simple_read_from_buffer(ubuf, cnt, ppos,
2333 readme_msg, strlen(readme_msg));
2336 static const struct file_operations tracing_readme_fops = {
2337 .open = tracing_open_generic,
2338 .read = tracing_readme_read,
2342 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2343 size_t cnt, loff_t *ppos)
2348 r = sprintf(buf, "%u\n", tracer_enabled);
2349 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2353 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2354 size_t cnt, loff_t *ppos)
2356 struct trace_array *tr = filp->private_data;
2361 if (cnt >= sizeof(buf))
2364 if (copy_from_user(&buf, ubuf, cnt))
2369 ret = strict_strtoul(buf, 10, &val);
2375 mutex_lock(&trace_types_lock);
2376 if (tracer_enabled ^ val) {
2379 if (current_trace->start)
2380 current_trace->start(tr);
2385 if (current_trace->stop)
2386 current_trace->stop(tr);
2389 mutex_unlock(&trace_types_lock);
2397 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2398 size_t cnt, loff_t *ppos)
2400 char buf[max_tracer_type_len+2];
2403 mutex_lock(&trace_types_lock);
2405 r = sprintf(buf, "%s\n", current_trace->name);
2407 r = sprintf(buf, "\n");
2408 mutex_unlock(&trace_types_lock);
2410 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2413 int tracer_init(struct tracer *t, struct trace_array *tr)
2415 tracing_reset_online_cpus(tr);
2419 static int tracing_resize_ring_buffer(unsigned long size)
2424 * If kernel or user changes the size of the ring buffer
2425 * we use the size that was given, and we can forget about
2426 * expanding it later.
2428 ring_buffer_expanded = 1;
2430 ret = ring_buffer_resize(global_trace.buffer, size);
2434 ret = ring_buffer_resize(max_tr.buffer, size);
2438 r = ring_buffer_resize(global_trace.buffer,
2439 global_trace.entries);
2442 * AARGH! We are left with different
2443 * size max buffer!!!!
2444 * The max buffer is our "snapshot" buffer.
2445 * When a tracer needs a snapshot (one of the
2446 * latency tracers), it swaps the max buffer
2447 * with the saved snap shot. We succeeded to
2448 * update the size of the main buffer, but failed to
2449 * update the size of the max buffer. But when we tried
2450 * to reset the main buffer to the original size, we
2451 * failed there too. This is very unlikely to
2452 * happen, but if it does, warn and kill all
2456 tracing_disabled = 1;
2461 global_trace.entries = size;
2467 * tracing_update_buffers - used by tracing facility to expand ring buffers
2469 * To save on memory when the tracing is never used on a system with it
2470 * configured in. The ring buffers are set to a minimum size. But once
2471 * a user starts to use the tracing facility, then they need to grow
2472 * to their default size.
2474 * This function is to be called when a tracer is about to be used.
2476 int tracing_update_buffers(void)
2480 mutex_lock(&trace_types_lock);
2481 if (!ring_buffer_expanded)
2482 ret = tracing_resize_ring_buffer(trace_buf_size);
2483 mutex_unlock(&trace_types_lock);
2488 struct trace_option_dentry;
2490 static struct trace_option_dentry *
2491 create_trace_option_files(struct tracer *tracer);
2494 destroy_trace_option_files(struct trace_option_dentry *topts);
2496 static int tracing_set_tracer(const char *buf)
2498 static struct trace_option_dentry *topts;
2499 struct trace_array *tr = &global_trace;
2503 mutex_lock(&trace_types_lock);
2505 if (!ring_buffer_expanded) {
2506 ret = tracing_resize_ring_buffer(trace_buf_size);
2512 for (t = trace_types; t; t = t->next) {
2513 if (strcmp(t->name, buf) == 0)
2520 if (t == current_trace)
2523 trace_branch_disable();
2524 if (current_trace && current_trace->reset)
2525 current_trace->reset(tr);
2527 destroy_trace_option_files(topts);
2531 topts = create_trace_option_files(current_trace);
2534 ret = tracer_init(t, tr);
2539 trace_branch_enable(tr);
2541 mutex_unlock(&trace_types_lock);
2547 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2548 size_t cnt, loff_t *ppos)
2550 char buf[max_tracer_type_len+1];
2557 if (cnt > max_tracer_type_len)
2558 cnt = max_tracer_type_len;
2560 if (copy_from_user(&buf, ubuf, cnt))
2565 /* strip ending whitespace. */
2566 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2569 err = tracing_set_tracer(buf);
2579 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2580 size_t cnt, loff_t *ppos)
2582 unsigned long *ptr = filp->private_data;
2586 r = snprintf(buf, sizeof(buf), "%ld\n",
2587 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2588 if (r > sizeof(buf))
2590 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2594 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2595 size_t cnt, loff_t *ppos)
2597 unsigned long *ptr = filp->private_data;
2602 if (cnt >= sizeof(buf))
2605 if (copy_from_user(&buf, ubuf, cnt))
2610 ret = strict_strtoul(buf, 10, &val);
2619 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2621 long cpu_file = (long) inode->i_private;
2622 struct trace_iterator *iter;
2625 if (tracing_disabled)
2628 mutex_lock(&trace_types_lock);
2630 /* We only allow one reader per cpu */
2631 if (cpu_file == TRACE_PIPE_ALL_CPU) {
2632 if (!cpumask_empty(tracing_reader_cpumask)) {
2636 cpumask_setall(tracing_reader_cpumask);
2638 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2639 cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2646 /* create a buffer to store the information to pass to userspace */
2647 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2654 * We make a copy of the current tracer to avoid concurrent
2655 * changes on it while we are reading.
2657 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2663 *iter->trace = *current_trace;
2665 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2670 /* trace pipe does not show start of buffer */
2671 cpumask_setall(iter->started);
2673 iter->cpu_file = cpu_file;
2674 iter->tr = &global_trace;
2675 mutex_init(&iter->mutex);
2676 filp->private_data = iter;
2678 if (iter->trace->pipe_open)
2679 iter->trace->pipe_open(iter);
2682 mutex_unlock(&trace_types_lock);
2688 mutex_unlock(&trace_types_lock);
2692 static int tracing_release_pipe(struct inode *inode, struct file *file)
2694 struct trace_iterator *iter = file->private_data;
2696 mutex_lock(&trace_types_lock);
2698 if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2699 cpumask_clear(tracing_reader_cpumask);
2701 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2703 mutex_unlock(&trace_types_lock);
2705 free_cpumask_var(iter->started);
2706 mutex_destroy(&iter->mutex);
2714 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2716 struct trace_iterator *iter = filp->private_data;
2718 if (trace_flags & TRACE_ITER_BLOCK) {
2720 * Always select as readable when in blocking mode
2722 return POLLIN | POLLRDNORM;
2724 if (!trace_empty(iter))
2725 return POLLIN | POLLRDNORM;
2726 poll_wait(filp, &trace_wait, poll_table);
2727 if (!trace_empty(iter))
2728 return POLLIN | POLLRDNORM;
2735 void default_wait_pipe(struct trace_iterator *iter)
2739 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2741 if (trace_empty(iter))
2744 finish_wait(&trace_wait, &wait);
2748 * This is a make-shift waitqueue.
2749 * A tracer might use this callback on some rare cases:
2751 * 1) the current tracer might hold the runqueue lock when it wakes up
2752 * a reader, hence a deadlock (sched, function, and function graph tracers)
2753 * 2) the function tracers, trace all functions, we don't want
2754 * the overhead of calling wake_up and friends
2755 * (and tracing them too)
2757 * Anyway, this is really very primitive wakeup.
2759 void poll_wait_pipe(struct trace_iterator *iter)
2761 set_current_state(TASK_INTERRUPTIBLE);
2762 /* sleep for 100 msecs, and try again. */
2763 schedule_timeout(HZ / 10);
2766 /* Must be called with trace_types_lock mutex held. */
2767 static int tracing_wait_pipe(struct file *filp)
2769 struct trace_iterator *iter = filp->private_data;
2771 while (trace_empty(iter)) {
2773 if ((filp->f_flags & O_NONBLOCK)) {
2777 mutex_unlock(&iter->mutex);
2779 iter->trace->wait_pipe(iter);
2781 mutex_lock(&iter->mutex);
2783 if (signal_pending(current))
2787 * We block until we read something and tracing is disabled.
2788 * We still block if tracing is disabled, but we have never
2789 * read anything. This allows a user to cat this file, and
2790 * then enable tracing. But after we have read something,
2791 * we give an EOF when tracing is again disabled.
2793 * iter->pos will be 0 if we haven't read anything.
2795 if (!tracer_enabled && iter->pos)
2806 tracing_read_pipe(struct file *filp, char __user *ubuf,
2807 size_t cnt, loff_t *ppos)
2809 struct trace_iterator *iter = filp->private_data;
2810 static struct tracer *old_tracer;
2813 /* return any leftover data */
2814 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2818 trace_seq_init(&iter->seq);
2820 /* copy the tracer to avoid using a global lock all around */
2821 mutex_lock(&trace_types_lock);
2822 if (unlikely(old_tracer != current_trace && current_trace)) {
2823 old_tracer = current_trace;
2824 *iter->trace = *current_trace;
2826 mutex_unlock(&trace_types_lock);
2829 * Avoid more than one consumer on a single file descriptor
2830 * This is just a matter of traces coherency, the ring buffer itself
2833 mutex_lock(&iter->mutex);
2834 if (iter->trace->read) {
2835 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2841 sret = tracing_wait_pipe(filp);
2845 /* stop when tracing is finished */
2846 if (trace_empty(iter)) {
2851 if (cnt >= PAGE_SIZE)
2852 cnt = PAGE_SIZE - 1;
2854 /* reset all but tr, trace, and overruns */
2855 memset(&iter->seq, 0,
2856 sizeof(struct trace_iterator) -
2857 offsetof(struct trace_iterator, seq));
2860 while (find_next_entry_inc(iter) != NULL) {
2861 enum print_line_t ret;
2862 int len = iter->seq.len;
2864 ret = print_trace_line(iter);
2865 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2866 /* don't print partial lines */
2867 iter->seq.len = len;
2870 if (ret != TRACE_TYPE_NO_CONSUME)
2871 trace_consume(iter);
2873 if (iter->seq.len >= cnt)
2877 /* Now copy what we have to the user */
2878 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2879 if (iter->seq.readpos >= iter->seq.len)
2880 trace_seq_init(&iter->seq);
2883 * If there was nothing to send to user, inspite of consuming trace
2884 * entries, go back to wait for more entries.
2890 mutex_unlock(&iter->mutex);
2895 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
2896 struct pipe_buffer *buf)
2898 __free_page(buf->page);
2901 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
2904 __free_page(spd->pages[idx]);
2907 static struct pipe_buf_operations tracing_pipe_buf_ops = {
2909 .map = generic_pipe_buf_map,
2910 .unmap = generic_pipe_buf_unmap,
2911 .confirm = generic_pipe_buf_confirm,
2912 .release = tracing_pipe_buf_release,
2913 .steal = generic_pipe_buf_steal,
2914 .get = generic_pipe_buf_get,
2918 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
2923 /* Seq buffer is page-sized, exactly what we need. */
2925 count = iter->seq.len;
2926 ret = print_trace_line(iter);
2927 count = iter->seq.len - count;
2930 iter->seq.len -= count;
2933 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2934 iter->seq.len -= count;
2938 trace_consume(iter);
2940 if (!find_next_entry_inc(iter)) {
2950 static ssize_t tracing_splice_read_pipe(struct file *filp,
2952 struct pipe_inode_info *pipe,
2956 struct page *pages[PIPE_BUFFERS];
2957 struct partial_page partial[PIPE_BUFFERS];
2958 struct trace_iterator *iter = filp->private_data;
2959 struct splice_pipe_desc spd = {
2962 .nr_pages = 0, /* This gets updated below. */
2964 .ops = &tracing_pipe_buf_ops,
2965 .spd_release = tracing_spd_release_pipe,
2967 static struct tracer *old_tracer;
2972 /* copy the tracer to avoid using a global lock all around */
2973 mutex_lock(&trace_types_lock);
2974 if (unlikely(old_tracer != current_trace && current_trace)) {
2975 old_tracer = current_trace;
2976 *iter->trace = *current_trace;
2978 mutex_unlock(&trace_types_lock);
2980 mutex_lock(&iter->mutex);
2982 if (iter->trace->splice_read) {
2983 ret = iter->trace->splice_read(iter, filp,
2984 ppos, pipe, len, flags);
2989 ret = tracing_wait_pipe(filp);
2993 if (!iter->ent && !find_next_entry_inc(iter)) {
2998 /* Fill as many pages as possible. */
2999 for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3000 pages[i] = alloc_page(GFP_KERNEL);
3004 rem = tracing_fill_pipe_page(rem, iter);
3006 /* Copy the data into the page, so we can start over. */
3007 ret = trace_seq_to_buffer(&iter->seq,
3008 page_address(pages[i]),
3011 __free_page(pages[i]);
3014 partial[i].offset = 0;
3015 partial[i].len = iter->seq.len;
3017 trace_seq_init(&iter->seq);
3020 mutex_unlock(&iter->mutex);
3024 return splice_to_pipe(pipe, &spd);
3027 mutex_unlock(&iter->mutex);
3033 tracing_entries_read(struct file *filp, char __user *ubuf,
3034 size_t cnt, loff_t *ppos)
3036 struct trace_array *tr = filp->private_data;
3040 mutex_lock(&trace_types_lock);
3041 if (!ring_buffer_expanded)
3042 r = sprintf(buf, "%lu (expanded: %lu)\n",
3044 trace_buf_size >> 10);
3046 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3047 mutex_unlock(&trace_types_lock);
3049 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3053 tracing_entries_write(struct file *filp, const char __user *ubuf,
3054 size_t cnt, loff_t *ppos)
3060 if (cnt >= sizeof(buf))
3063 if (copy_from_user(&buf, ubuf, cnt))
3068 ret = strict_strtoul(buf, 10, &val);
3072 /* must have at least 1 entry */
3076 mutex_lock(&trace_types_lock);
3080 /* disable all cpu buffers */
3081 for_each_tracing_cpu(cpu) {
3082 if (global_trace.data[cpu])
3083 atomic_inc(&global_trace.data[cpu]->disabled);
3084 if (max_tr.data[cpu])
3085 atomic_inc(&max_tr.data[cpu]->disabled);
3088 /* value is in KB */
3091 if (val != global_trace.entries) {
3092 ret = tracing_resize_ring_buffer(val);
3101 /* If check pages failed, return ENOMEM */
3102 if (tracing_disabled)
3105 for_each_tracing_cpu(cpu) {
3106 if (global_trace.data[cpu])
3107 atomic_dec(&global_trace.data[cpu]->disabled);
3108 if (max_tr.data[cpu])
3109 atomic_dec(&max_tr.data[cpu]->disabled);
3113 max_tr.entries = global_trace.entries;
3114 mutex_unlock(&trace_types_lock);
3119 static int mark_printk(const char *fmt, ...)
3123 va_start(args, fmt);
3124 ret = trace_vprintk(0, -1, fmt, args);
3130 tracing_mark_write(struct file *filp, const char __user *ubuf,
3131 size_t cnt, loff_t *fpos)
3136 if (tracing_disabled)
3139 if (cnt > TRACE_BUF_SIZE)
3140 cnt = TRACE_BUF_SIZE;
3142 buf = kmalloc(cnt + 1, GFP_KERNEL);
3146 if (copy_from_user(buf, ubuf, cnt)) {
3151 /* Cut from the first nil or newline. */
3153 end = strchr(buf, '\n');
3157 cnt = mark_printk("%s\n", buf);
3164 static const struct file_operations tracing_max_lat_fops = {
3165 .open = tracing_open_generic,
3166 .read = tracing_max_lat_read,
3167 .write = tracing_max_lat_write,
3170 static const struct file_operations tracing_ctrl_fops = {
3171 .open = tracing_open_generic,
3172 .read = tracing_ctrl_read,
3173 .write = tracing_ctrl_write,
3176 static const struct file_operations set_tracer_fops = {
3177 .open = tracing_open_generic,
3178 .read = tracing_set_trace_read,
3179 .write = tracing_set_trace_write,
3182 static const struct file_operations tracing_pipe_fops = {
3183 .open = tracing_open_pipe,
3184 .poll = tracing_poll_pipe,
3185 .read = tracing_read_pipe,
3186 .splice_read = tracing_splice_read_pipe,
3187 .release = tracing_release_pipe,
3190 static const struct file_operations tracing_entries_fops = {
3191 .open = tracing_open_generic,
3192 .read = tracing_entries_read,
3193 .write = tracing_entries_write,
3196 static const struct file_operations tracing_mark_fops = {
3197 .open = tracing_open_generic,
3198 .write = tracing_mark_write,
3201 struct ftrace_buffer_info {
3202 struct trace_array *tr;
3208 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3210 int cpu = (int)(long)inode->i_private;
3211 struct ftrace_buffer_info *info;
3213 if (tracing_disabled)
3216 info = kzalloc(sizeof(*info), GFP_KERNEL);
3220 info->tr = &global_trace;
3222 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3223 /* Force reading ring buffer for first read */
3224 info->read = (unsigned int)-1;
3228 filp->private_data = info;
3238 tracing_buffers_read(struct file *filp, char __user *ubuf,
3239 size_t count, loff_t *ppos)
3241 struct ftrace_buffer_info *info = filp->private_data;
3249 /* Do we have previous read data to read? */
3250 if (info->read < PAGE_SIZE)
3255 ret = ring_buffer_read_page(info->tr->buffer,
3262 pos = ring_buffer_page_len(info->spare);
3264 if (pos < PAGE_SIZE)
3265 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3268 size = PAGE_SIZE - info->read;
3272 ret = copy_to_user(ubuf, info->spare + info->read, size);
3283 static int tracing_buffers_release(struct inode *inode, struct file *file)
3285 struct ftrace_buffer_info *info = file->private_data;
3287 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3294 struct ring_buffer *buffer;
3299 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3300 struct pipe_buffer *buf)
3302 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3307 ring_buffer_free_read_page(ref->buffer, ref->page);
3312 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3313 struct pipe_buffer *buf)
3318 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3319 struct pipe_buffer *buf)
3321 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3326 /* Pipe buffer operations for a buffer. */
3327 static struct pipe_buf_operations buffer_pipe_buf_ops = {
3329 .map = generic_pipe_buf_map,
3330 .unmap = generic_pipe_buf_unmap,
3331 .confirm = generic_pipe_buf_confirm,
3332 .release = buffer_pipe_buf_release,
3333 .steal = buffer_pipe_buf_steal,
3334 .get = buffer_pipe_buf_get,
3338 * Callback from splice_to_pipe(), if we need to release some pages
3339 * at the end of the spd in case we error'ed out in filling the pipe.
3341 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3343 struct buffer_ref *ref =
3344 (struct buffer_ref *)spd->partial[i].private;
3349 ring_buffer_free_read_page(ref->buffer, ref->page);
3351 spd->partial[i].private = 0;
3355 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3356 struct pipe_inode_info *pipe, size_t len,
3359 struct ftrace_buffer_info *info = file->private_data;
3360 struct partial_page partial[PIPE_BUFFERS];
3361 struct page *pages[PIPE_BUFFERS];
3362 struct splice_pipe_desc spd = {
3366 .ops = &buffer_pipe_buf_ops,
3367 .spd_release = buffer_spd_release,
3369 struct buffer_ref *ref;
3374 * We can't seek on a buffer input
3376 if (unlikely(*ppos))
3380 for (i = 0; i < PIPE_BUFFERS && len; i++, len -= size) {
3384 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3388 ref->buffer = info->tr->buffer;
3389 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3395 r = ring_buffer_read_page(ref->buffer, &ref->page,
3398 ring_buffer_free_read_page(ref->buffer,
3405 * zero out any left over data, this is going to
3408 size = ring_buffer_page_len(ref->page);
3409 if (size < PAGE_SIZE)
3410 memset(ref->page + size, 0, PAGE_SIZE - size);
3412 page = virt_to_page(ref->page);
3414 spd.pages[i] = page;
3415 spd.partial[i].len = PAGE_SIZE;
3416 spd.partial[i].offset = 0;
3417 spd.partial[i].private = (unsigned long)ref;
3423 /* did we read anything? */
3424 if (!spd.nr_pages) {
3425 if (flags & SPLICE_F_NONBLOCK)
3433 ret = splice_to_pipe(pipe, &spd);
3438 static const struct file_operations tracing_buffers_fops = {
3439 .open = tracing_buffers_open,
3440 .read = tracing_buffers_read,
3441 .release = tracing_buffers_release,
3442 .splice_read = tracing_buffers_splice_read,
3443 .llseek = no_llseek,
3446 #ifdef CONFIG_DYNAMIC_FTRACE
3448 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3454 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3455 size_t cnt, loff_t *ppos)
3457 static char ftrace_dyn_info_buffer[1024];
3458 static DEFINE_MUTEX(dyn_info_mutex);
3459 unsigned long *p = filp->private_data;
3460 char *buf = ftrace_dyn_info_buffer;
3461 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3464 mutex_lock(&dyn_info_mutex);
3465 r = sprintf(buf, "%ld ", *p);
3467 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3470 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3472 mutex_unlock(&dyn_info_mutex);
3477 static const struct file_operations tracing_dyn_info_fops = {
3478 .open = tracing_open_generic,
3479 .read = tracing_read_dyn_info,
3483 static struct dentry *d_tracer;
3485 struct dentry *tracing_init_dentry(void)
3492 d_tracer = debugfs_create_dir("tracing", NULL);
3494 if (!d_tracer && !once) {
3496 pr_warning("Could not create debugfs directory 'tracing'\n");
3503 static struct dentry *d_percpu;
3505 struct dentry *tracing_dentry_percpu(void)
3508 struct dentry *d_tracer;
3513 d_tracer = tracing_init_dentry();
3518 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3520 if (!d_percpu && !once) {
3522 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3529 static void tracing_init_debugfs_percpu(long cpu)
3531 struct dentry *d_percpu = tracing_dentry_percpu();
3532 struct dentry *entry, *d_cpu;
3533 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3536 if (cpu > 999 || cpu < 0)
3539 sprintf(cpu_dir, "cpu%ld", cpu);
3540 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3542 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3546 /* per cpu trace_pipe */
3547 entry = debugfs_create_file("trace_pipe", 0444, d_cpu,
3548 (void *) cpu, &tracing_pipe_fops);
3550 pr_warning("Could not create debugfs 'trace_pipe' entry\n");
3553 entry = debugfs_create_file("trace", 0444, d_cpu,
3554 (void *) cpu, &tracing_fops);
3556 pr_warning("Could not create debugfs 'trace' entry\n");
3558 entry = debugfs_create_file("trace_pipe_raw", 0444, d_cpu,
3559 (void *) cpu, &tracing_buffers_fops);
3561 pr_warning("Could not create debugfs 'trace_pipe_raw' entry\n");
3564 #ifdef CONFIG_FTRACE_SELFTEST
3565 /* Let selftest have access to static functions in this file */
3566 #include "trace_selftest.c"
3569 struct trace_option_dentry {
3570 struct tracer_opt *opt;
3571 struct tracer_flags *flags;
3572 struct dentry *entry;
3576 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3579 struct trace_option_dentry *topt = filp->private_data;
3582 if (topt->flags->val & topt->opt->bit)
3587 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3591 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3594 struct trace_option_dentry *topt = filp->private_data;
3599 if (cnt >= sizeof(buf))
3602 if (copy_from_user(&buf, ubuf, cnt))
3607 ret = strict_strtoul(buf, 10, &val);
3614 /* do nothing if already cleared */
3615 if (!(topt->flags->val & topt->opt->bit))
3618 mutex_lock(&trace_types_lock);
3619 if (current_trace->set_flag)
3620 ret = current_trace->set_flag(topt->flags->val,
3622 mutex_unlock(&trace_types_lock);
3625 topt->flags->val &= ~topt->opt->bit;
3628 /* do nothing if already set */
3629 if (topt->flags->val & topt->opt->bit)
3632 mutex_lock(&trace_types_lock);
3633 if (current_trace->set_flag)
3634 ret = current_trace->set_flag(topt->flags->val,
3636 mutex_unlock(&trace_types_lock);
3639 topt->flags->val |= topt->opt->bit;
3652 static const struct file_operations trace_options_fops = {
3653 .open = tracing_open_generic,
3654 .read = trace_options_read,
3655 .write = trace_options_write,
3659 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
3662 long index = (long)filp->private_data;
3665 if (trace_flags & (1 << index))
3670 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3674 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
3677 long index = (long)filp->private_data;
3682 if (cnt >= sizeof(buf))
3685 if (copy_from_user(&buf, ubuf, cnt))
3690 ret = strict_strtoul(buf, 10, &val);
3696 trace_flags &= ~(1 << index);
3699 trace_flags |= 1 << index;
3711 static const struct file_operations trace_options_core_fops = {
3712 .open = tracing_open_generic,
3713 .read = trace_options_core_read,
3714 .write = trace_options_core_write,
3717 static struct dentry *trace_options_init_dentry(void)
3719 struct dentry *d_tracer;
3720 static struct dentry *t_options;
3725 d_tracer = tracing_init_dentry();
3729 t_options = debugfs_create_dir("options", d_tracer);
3731 pr_warning("Could not create debugfs directory 'options'\n");
3739 create_trace_option_file(struct trace_option_dentry *topt,
3740 struct tracer_flags *flags,
3741 struct tracer_opt *opt)
3743 struct dentry *t_options;
3744 struct dentry *entry;
3746 t_options = trace_options_init_dentry();
3750 topt->flags = flags;
3753 entry = debugfs_create_file(opt->name, 0644, t_options, topt,
3754 &trace_options_fops);
3756 topt->entry = entry;
3760 static struct trace_option_dentry *
3761 create_trace_option_files(struct tracer *tracer)
3763 struct trace_option_dentry *topts;
3764 struct tracer_flags *flags;
3765 struct tracer_opt *opts;
3771 flags = tracer->flags;
3773 if (!flags || !flags->opts)
3778 for (cnt = 0; opts[cnt].name; cnt++)
3781 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
3785 for (cnt = 0; opts[cnt].name; cnt++)
3786 create_trace_option_file(&topts[cnt], flags,
3793 destroy_trace_option_files(struct trace_option_dentry *topts)
3800 for (cnt = 0; topts[cnt].opt; cnt++) {
3801 if (topts[cnt].entry)
3802 debugfs_remove(topts[cnt].entry);
3808 static struct dentry *
3809 create_trace_option_core_file(const char *option, long index)
3811 struct dentry *t_options;
3812 struct dentry *entry;
3814 t_options = trace_options_init_dentry();
3818 entry = debugfs_create_file(option, 0644, t_options, (void *)index,
3819 &trace_options_core_fops);
3824 static __init void create_trace_options_dir(void)
3826 struct dentry *t_options;
3827 struct dentry *entry;
3830 t_options = trace_options_init_dentry();
3834 for (i = 0; trace_options[i]; i++) {
3835 entry = create_trace_option_core_file(trace_options[i], i);
3837 pr_warning("Could not create debugfs %s entry\n",
3842 static __init int tracer_init_debugfs(void)
3844 struct dentry *d_tracer;
3845 struct dentry *entry;
3848 d_tracer = tracing_init_dentry();
3850 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3851 &global_trace, &tracing_ctrl_fops);
3853 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3855 entry = debugfs_create_file("trace_options", 0644, d_tracer,
3856 NULL, &tracing_iter_fops);
3858 pr_warning("Could not create debugfs 'trace_options' entry\n");
3860 create_trace_options_dir();
3862 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3863 NULL, &tracing_cpumask_fops);
3865 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3867 entry = debugfs_create_file("trace", 0444, d_tracer,
3868 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
3870 pr_warning("Could not create debugfs 'trace' entry\n");
3872 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3873 &global_trace, &show_traces_fops);
3875 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3877 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3878 &global_trace, &set_tracer_fops);
3880 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3882 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3883 &tracing_max_latency,
3884 &tracing_max_lat_fops);
3886 pr_warning("Could not create debugfs "
3887 "'tracing_max_latency' entry\n");
3889 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3890 &tracing_thresh, &tracing_max_lat_fops);
3892 pr_warning("Could not create debugfs "
3893 "'tracing_thresh' entry\n");
3894 entry = debugfs_create_file("README", 0644, d_tracer,
3895 NULL, &tracing_readme_fops);
3897 pr_warning("Could not create debugfs 'README' entry\n");
3899 entry = debugfs_create_file("trace_pipe", 0444, d_tracer,
3900 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
3902 pr_warning("Could not create debugfs "
3903 "'trace_pipe' entry\n");
3905 entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
3906 &global_trace, &tracing_entries_fops);
3908 pr_warning("Could not create debugfs "
3909 "'buffer_size_kb' entry\n");
3911 entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3912 NULL, &tracing_mark_fops);
3914 pr_warning("Could not create debugfs "
3915 "'trace_marker' entry\n");
3917 #ifdef CONFIG_DYNAMIC_FTRACE
3918 entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3919 &ftrace_update_tot_cnt,
3920 &tracing_dyn_info_fops);
3922 pr_warning("Could not create debugfs "
3923 "'dyn_ftrace_total_info' entry\n");
3925 #ifdef CONFIG_SYSPROF_TRACER
3926 init_tracer_sysprof_debugfs(d_tracer);
3929 for_each_tracing_cpu(cpu)
3930 tracing_init_debugfs_percpu(cpu);
3935 static int trace_panic_handler(struct notifier_block *this,
3936 unsigned long event, void *unused)
3938 if (ftrace_dump_on_oops)
3943 static struct notifier_block trace_panic_notifier = {
3944 .notifier_call = trace_panic_handler,
3946 .priority = 150 /* priority: INT_MAX >= x >= 0 */
3949 static int trace_die_handler(struct notifier_block *self,
3955 if (ftrace_dump_on_oops)
3964 static struct notifier_block trace_die_notifier = {
3965 .notifier_call = trace_die_handler,
3970 * printk is set to max of 1024, we really don't need it that big.
3971 * Nothing should be printing 1000 characters anyway.
3973 #define TRACE_MAX_PRINT 1000
3976 * Define here KERN_TRACE so that we have one place to modify
3977 * it if we decide to change what log level the ftrace dump
3980 #define KERN_TRACE KERN_EMERG
3983 trace_printk_seq(struct trace_seq *s)
3985 /* Probably should print a warning here. */
3989 /* should be zero ended, but we are paranoid. */
3990 s->buffer[s->len] = 0;
3992 printk(KERN_TRACE "%s", s->buffer);
3997 void ftrace_dump(void)
3999 static DEFINE_SPINLOCK(ftrace_dump_lock);
4000 /* use static because iter can be a bit big for the stack */
4001 static struct trace_iterator iter;
4002 static int dump_ran;
4003 unsigned long flags;
4007 spin_lock_irqsave(&ftrace_dump_lock, flags);
4013 /* No turning back! */
4017 for_each_tracing_cpu(cpu) {
4018 atomic_inc(&global_trace.data[cpu]->disabled);
4021 /* don't look at user memory in panic mode */
4022 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4024 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4026 /* Simulate the iterator */
4027 iter.tr = &global_trace;
4028 iter.trace = current_trace;
4029 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4032 * We need to stop all tracing on all CPUS to read the
4033 * the next buffer. This is a bit expensive, but is
4034 * not done often. We fill all what we can read,
4035 * and then release the locks again.
4038 while (!trace_empty(&iter)) {
4041 printk(KERN_TRACE "---------------------------------\n");
4045 /* reset all but tr, trace, and overruns */
4046 memset(&iter.seq, 0,
4047 sizeof(struct trace_iterator) -
4048 offsetof(struct trace_iterator, seq));
4049 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4052 if (find_next_entry_inc(&iter) != NULL) {
4053 print_trace_line(&iter);
4054 trace_consume(&iter);
4057 trace_printk_seq(&iter.seq);
4061 printk(KERN_TRACE " (ftrace buffer empty)\n");
4063 printk(KERN_TRACE "---------------------------------\n");
4066 spin_unlock_irqrestore(&ftrace_dump_lock, flags);
4069 __init static int tracer_alloc_buffers(void)
4071 struct trace_array_cpu *data;
4076 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4079 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4080 goto out_free_buffer_mask;
4082 if (!alloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
4083 goto out_free_tracing_cpumask;
4085 /* To save memory, keep the ring buffer size to its minimum */
4086 if (ring_buffer_expanded)
4087 ring_buf_size = trace_buf_size;
4091 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4092 cpumask_copy(tracing_cpumask, cpu_all_mask);
4093 cpumask_clear(tracing_reader_cpumask);
4095 /* TODO: make the number of buffers hot pluggable with CPUS */
4096 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4097 TRACE_BUFFER_FLAGS);
4098 if (!global_trace.buffer) {
4099 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4101 goto out_free_cpumask;
4103 global_trace.entries = ring_buffer_size(global_trace.buffer);
4106 #ifdef CONFIG_TRACER_MAX_TRACE
4107 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4108 TRACE_BUFFER_FLAGS);
4109 if (!max_tr.buffer) {
4110 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4112 ring_buffer_free(global_trace.buffer);
4113 goto out_free_cpumask;
4115 max_tr.entries = ring_buffer_size(max_tr.buffer);
4116 WARN_ON(max_tr.entries != global_trace.entries);
4119 /* Allocate the first page for all buffers */
4120 for_each_tracing_cpu(i) {
4121 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4122 max_tr.data[i] = &per_cpu(max_data, i);
4125 trace_init_cmdlines();
4127 register_tracer(&nop_trace);
4128 current_trace = &nop_trace;
4129 #ifdef CONFIG_BOOT_TRACER
4130 register_tracer(&boot_tracer);
4132 /* All seems OK, enable tracing */
4133 tracing_disabled = 0;
4135 atomic_notifier_chain_register(&panic_notifier_list,
4136 &trace_panic_notifier);
4138 register_die_notifier(&trace_die_notifier);
4142 free_cpumask_var(tracing_reader_cpumask);
4143 out_free_tracing_cpumask:
4144 free_cpumask_var(tracing_cpumask);
4145 out_free_buffer_mask:
4146 free_cpumask_var(tracing_buffer_mask);
4151 __init static int clear_boot_tracer(void)
4154 * The default tracer at boot buffer is an init section.
4155 * This function is called in lateinit. If we did not
4156 * find the boot tracer, then clear it out, to prevent
4157 * later registration from accessing the buffer that is
4158 * about to be freed.
4160 if (!default_bootup_tracer)
4163 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4164 default_bootup_tracer);
4165 default_bootup_tracer = NULL;
4170 early_initcall(tracer_alloc_buffers);
4171 fs_initcall(tracer_init_debugfs);
4172 late_initcall(clear_boot_tracer);