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)
158 cycle_t ftrace_now(int cpu)
160 u64 ts = ring_buffer_time_stamp(cpu);
161 ring_buffer_normalize_time_stamp(cpu, &ts);
166 * The global_trace is the descriptor that holds the tracing
167 * buffers for the live tracing. For each CPU, it contains
168 * a link list of pages that will store trace entries. The
169 * page descriptor of the pages in the memory is used to hold
170 * the link list by linking the lru item in the page descriptor
171 * to each of the pages in the buffer per CPU.
173 * For each active CPU there is a data field that holds the
174 * pages for the buffer for that CPU. Each CPU has the same number
175 * of pages allocated for its buffer.
177 static struct trace_array global_trace;
179 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
182 * The max_tr is used to snapshot the global_trace when a maximum
183 * latency is reached. Some tracers will use this to store a maximum
184 * trace while it continues examining live traces.
186 * The buffers for the max_tr are set up the same as the global_trace.
187 * When a snapshot is taken, the link list of the max_tr is swapped
188 * with the link list of the global_trace and the buffers are reset for
189 * the global_trace so the tracing can continue.
191 static struct trace_array max_tr;
193 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
195 /* tracer_enabled is used to toggle activation of a tracer */
196 static int tracer_enabled = 1;
199 * tracing_is_enabled - return tracer_enabled status
201 * This function is used by other tracers to know the status
202 * of the tracer_enabled flag. Tracers may use this function
203 * to know if it should enable their features when starting
204 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
206 int tracing_is_enabled(void)
208 return tracer_enabled;
212 * trace_buf_size is the size in bytes that is allocated
213 * for a buffer. Note, the number of bytes is always rounded
216 * This number is purposely set to a low number of 16384.
217 * If the dump on oops happens, it will be much appreciated
218 * to not have to wait for all that output. Anyway this can be
219 * boot time and run time configurable.
221 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
223 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
225 /* trace_types holds a link list of available tracers. */
226 static struct tracer *trace_types __read_mostly;
228 /* current_trace points to the tracer that is currently active */
229 static struct tracer *current_trace __read_mostly;
232 * max_tracer_type_len is used to simplify the allocating of
233 * buffers to read userspace tracer names. We keep track of
234 * the longest tracer name registered.
236 static int max_tracer_type_len;
239 * trace_types_lock is used to protect the trace_types list.
240 * This lock is also used to keep user access serialized.
241 * Accesses from userspace will grab this lock while userspace
242 * activities happen inside the kernel.
244 static DEFINE_MUTEX(trace_types_lock);
246 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
247 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
249 /* trace_flags holds trace_options default values */
250 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
251 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO;
254 * trace_wake_up - wake up tasks waiting for trace input
256 * Simply wakes up any task that is blocked on the trace_wait
257 * queue. These is used with trace_poll for tasks polling the trace.
259 void trace_wake_up(void)
262 * The runqueue_is_locked() can fail, but this is the best we
265 if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
266 wake_up(&trace_wait);
269 static int __init set_buf_size(char *str)
271 unsigned long buf_size;
276 ret = strict_strtoul(str, 0, &buf_size);
277 /* nr_entries can not be zero */
278 if (ret < 0 || buf_size == 0)
280 trace_buf_size = buf_size;
283 __setup("trace_buf_size=", set_buf_size);
285 unsigned long nsecs_to_usecs(unsigned long nsecs)
290 /* These must match the bit postions in trace_iterator_flags */
291 static const char *trace_options[] = {
315 * ftrace_max_lock is used to protect the swapping of buffers
316 * when taking a max snapshot. The buffers themselves are
317 * protected by per_cpu spinlocks. But the action of the swap
318 * needs its own lock.
320 * This is defined as a raw_spinlock_t in order to help
321 * with performance when lockdep debugging is enabled.
323 static raw_spinlock_t ftrace_max_lock =
324 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
327 * Copy the new maximum trace into the separate maximum-trace
328 * structure. (this way the maximum trace is permanently saved,
329 * for later retrieval via /debugfs/tracing/latency_trace)
332 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
334 struct trace_array_cpu *data = tr->data[cpu];
337 max_tr.time_start = data->preempt_timestamp;
339 data = max_tr.data[cpu];
340 data->saved_latency = tracing_max_latency;
342 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
343 data->pid = tsk->pid;
344 data->uid = task_uid(tsk);
345 data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
346 data->policy = tsk->policy;
347 data->rt_priority = tsk->rt_priority;
349 /* record this tasks comm */
350 tracing_record_cmdline(tsk);
353 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
361 if (s->len <= s->readpos)
364 len = s->len - s->readpos;
367 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
377 ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
382 if (s->len <= s->readpos)
385 len = s->len - s->readpos;
388 ret = memcpy(buf, s->buffer + s->readpos, cnt);
397 trace_print_seq(struct seq_file *m, struct trace_seq *s)
399 int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
402 seq_puts(m, s->buffer);
408 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
410 * @tsk: the task with the latency
411 * @cpu: The cpu that initiated the trace.
413 * Flip the buffers between the @tr and the max_tr and record information
414 * about which task was the cause of this latency.
417 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
419 struct ring_buffer *buf = tr->buffer;
421 WARN_ON_ONCE(!irqs_disabled());
422 __raw_spin_lock(&ftrace_max_lock);
424 tr->buffer = max_tr.buffer;
427 ftrace_disable_cpu();
428 ring_buffer_reset(tr->buffer);
431 __update_max_tr(tr, tsk, cpu);
432 __raw_spin_unlock(&ftrace_max_lock);
436 * update_max_tr_single - only copy one trace over, and reset the rest
438 * @tsk - task with the latency
439 * @cpu - the cpu of the buffer to copy.
441 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
444 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
448 WARN_ON_ONCE(!irqs_disabled());
449 __raw_spin_lock(&ftrace_max_lock);
451 ftrace_disable_cpu();
453 ring_buffer_reset(max_tr.buffer);
454 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
458 WARN_ON_ONCE(ret && ret != -EAGAIN);
460 __update_max_tr(tr, tsk, cpu);
461 __raw_spin_unlock(&ftrace_max_lock);
465 * register_tracer - register a tracer with the ftrace system.
466 * @type - the plugin for the tracer
468 * Register a new plugin tracer.
470 int register_tracer(struct tracer *type)
471 __releases(kernel_lock)
472 __acquires(kernel_lock)
479 pr_info("Tracer must have a name\n");
484 * When this gets called we hold the BKL which means that
485 * preemption is disabled. Various trace selftests however
486 * need to disable and enable preemption for successful tests.
487 * So we drop the BKL here and grab it after the tests again.
490 mutex_lock(&trace_types_lock);
492 tracing_selftest_running = true;
494 for (t = trace_types; t; t = t->next) {
495 if (strcmp(type->name, t->name) == 0) {
497 pr_info("Trace %s already registered\n",
505 type->set_flag = &dummy_set_flag;
507 type->flags = &dummy_tracer_flags;
509 if (!type->flags->opts)
510 type->flags->opts = dummy_tracer_opt;
511 if (!type->wait_pipe)
512 type->wait_pipe = default_wait_pipe;
515 #ifdef CONFIG_FTRACE_STARTUP_TEST
516 if (type->selftest && !tracing_selftest_disabled) {
517 struct tracer *saved_tracer = current_trace;
518 struct trace_array *tr = &global_trace;
522 * Run a selftest on this tracer.
523 * Here we reset the trace buffer, and set the current
524 * tracer to be this tracer. The tracer can then run some
525 * internal tracing to verify that everything is in order.
526 * If we fail, we do not register this tracer.
528 for_each_tracing_cpu(i)
529 tracing_reset(tr, i);
531 current_trace = type;
532 /* the test is responsible for initializing and enabling */
533 pr_info("Testing tracer %s: ", type->name);
534 ret = type->selftest(type, tr);
535 /* the test is responsible for resetting too */
536 current_trace = saved_tracer;
538 printk(KERN_CONT "FAILED!\n");
541 /* Only reset on passing, to avoid touching corrupted buffers */
542 for_each_tracing_cpu(i)
543 tracing_reset(tr, i);
545 printk(KERN_CONT "PASSED\n");
549 type->next = trace_types;
551 len = strlen(type->name);
552 if (len > max_tracer_type_len)
553 max_tracer_type_len = len;
556 tracing_selftest_running = false;
557 mutex_unlock(&trace_types_lock);
559 if (ret || !default_bootup_tracer)
562 if (strncmp(default_bootup_tracer, type->name, BOOTUP_TRACER_SIZE))
565 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
566 /* Do we want this tracer to start on bootup? */
567 tracing_set_tracer(type->name);
568 default_bootup_tracer = NULL;
569 /* disable other selftests, since this will break it. */
570 tracing_selftest_disabled = 1;
571 #ifdef CONFIG_FTRACE_STARTUP_TEST
572 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
581 void unregister_tracer(struct tracer *type)
586 mutex_lock(&trace_types_lock);
587 for (t = &trace_types; *t; t = &(*t)->next) {
591 pr_info("Trace %s not registered\n", type->name);
597 if (type == current_trace && tracer_enabled) {
600 if (current_trace->stop)
601 current_trace->stop(&global_trace);
602 current_trace = &nop_trace;
605 if (strlen(type->name) != max_tracer_type_len)
608 max_tracer_type_len = 0;
609 for (t = &trace_types; *t; t = &(*t)->next) {
610 len = strlen((*t)->name);
611 if (len > max_tracer_type_len)
612 max_tracer_type_len = len;
615 mutex_unlock(&trace_types_lock);
618 void tracing_reset(struct trace_array *tr, int cpu)
620 ftrace_disable_cpu();
621 ring_buffer_reset_cpu(tr->buffer, cpu);
625 void tracing_reset_online_cpus(struct trace_array *tr)
629 tr->time_start = ftrace_now(tr->cpu);
631 for_each_online_cpu(cpu)
632 tracing_reset(tr, cpu);
635 #define SAVED_CMDLINES 128
636 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
637 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
638 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
639 static int cmdline_idx;
640 static raw_spinlock_t trace_cmdline_lock = __RAW_SPIN_LOCK_UNLOCKED;
642 /* temporary disable recording */
643 static atomic_t trace_record_cmdline_disabled __read_mostly;
645 static void trace_init_cmdlines(void)
647 memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
648 memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
652 static int trace_stop_count;
653 static DEFINE_SPINLOCK(tracing_start_lock);
656 * ftrace_off_permanent - disable all ftrace code permanently
658 * This should only be called when a serious anomally has
659 * been detected. This will turn off the function tracing,
660 * ring buffers, and other tracing utilites. It takes no
661 * locks and can be called from any context.
663 void ftrace_off_permanent(void)
665 tracing_disabled = 1;
667 tracing_off_permanent();
671 * tracing_start - quick start of the tracer
673 * If tracing is enabled but was stopped by tracing_stop,
674 * this will start the tracer back up.
676 void tracing_start(void)
678 struct ring_buffer *buffer;
681 if (tracing_disabled)
684 spin_lock_irqsave(&tracing_start_lock, flags);
685 if (--trace_stop_count) {
686 if (trace_stop_count < 0) {
687 /* Someone screwed up their debugging */
689 trace_stop_count = 0;
695 buffer = global_trace.buffer;
697 ring_buffer_record_enable(buffer);
699 buffer = max_tr.buffer;
701 ring_buffer_record_enable(buffer);
705 spin_unlock_irqrestore(&tracing_start_lock, flags);
709 * tracing_stop - quick stop of the tracer
711 * Light weight way to stop tracing. Use in conjunction with
714 void tracing_stop(void)
716 struct ring_buffer *buffer;
720 spin_lock_irqsave(&tracing_start_lock, flags);
721 if (trace_stop_count++)
724 buffer = global_trace.buffer;
726 ring_buffer_record_disable(buffer);
728 buffer = max_tr.buffer;
730 ring_buffer_record_disable(buffer);
733 spin_unlock_irqrestore(&tracing_start_lock, flags);
736 void trace_stop_cmdline_recording(void);
738 static void trace_save_cmdline(struct task_struct *tsk)
743 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
747 * It's not the end of the world if we don't get
748 * the lock, but we also don't want to spin
749 * nor do we want to disable interrupts,
750 * so if we miss here, then better luck next time.
752 if (!__raw_spin_trylock(&trace_cmdline_lock))
755 idx = map_pid_to_cmdline[tsk->pid];
756 if (idx >= SAVED_CMDLINES) {
757 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
759 map = map_cmdline_to_pid[idx];
760 if (map <= PID_MAX_DEFAULT)
761 map_pid_to_cmdline[map] = (unsigned)-1;
763 map_pid_to_cmdline[tsk->pid] = idx;
768 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
770 __raw_spin_unlock(&trace_cmdline_lock);
773 char *trace_find_cmdline(int pid)
775 char *cmdline = "<...>";
781 if (pid > PID_MAX_DEFAULT)
784 map = map_pid_to_cmdline[pid];
785 if (map >= SAVED_CMDLINES)
788 cmdline = saved_cmdlines[map];
794 void tracing_record_cmdline(struct task_struct *tsk)
796 if (atomic_read(&trace_record_cmdline_disabled))
799 trace_save_cmdline(tsk);
803 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
806 struct task_struct *tsk = current;
808 entry->preempt_count = pc & 0xff;
809 entry->pid = (tsk) ? tsk->pid : 0;
810 entry->tgid = (tsk) ? tsk->tgid : 0;
812 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
813 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
815 TRACE_FLAG_IRQS_NOSUPPORT |
817 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
818 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
819 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
822 struct ring_buffer_event *trace_buffer_lock_reserve(struct trace_array *tr,
825 unsigned long flags, int pc)
827 struct ring_buffer_event *event;
829 event = ring_buffer_lock_reserve(tr->buffer, len);
831 struct trace_entry *ent = ring_buffer_event_data(event);
833 tracing_generic_entry_update(ent, flags, pc);
839 static void ftrace_trace_stack(struct trace_array *tr,
840 unsigned long flags, int skip, int pc);
841 static void ftrace_trace_userstack(struct trace_array *tr,
842 unsigned long flags, int pc);
844 void trace_buffer_unlock_commit(struct trace_array *tr,
845 struct ring_buffer_event *event,
846 unsigned long flags, int pc)
848 ring_buffer_unlock_commit(tr->buffer, event);
850 ftrace_trace_stack(tr, flags, 6, pc);
851 ftrace_trace_userstack(tr, flags, pc);
855 struct ring_buffer_event *
856 trace_current_buffer_lock_reserve(unsigned char type, unsigned long len,
857 unsigned long flags, int pc)
859 return trace_buffer_lock_reserve(&global_trace,
860 type, len, flags, pc);
863 void trace_current_buffer_unlock_commit(struct ring_buffer_event *event,
864 unsigned long flags, int pc)
866 return trace_buffer_unlock_commit(&global_trace, event, flags, pc);
870 trace_function(struct trace_array *tr,
871 unsigned long ip, unsigned long parent_ip, unsigned long flags,
874 struct ring_buffer_event *event;
875 struct ftrace_entry *entry;
877 /* If we are reading the ring buffer, don't trace */
878 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
881 event = trace_buffer_lock_reserve(tr, TRACE_FN, sizeof(*entry),
885 entry = ring_buffer_event_data(event);
887 entry->parent_ip = parent_ip;
888 ring_buffer_unlock_commit(tr->buffer, event);
891 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
892 static void __trace_graph_entry(struct trace_array *tr,
893 struct ftrace_graph_ent *trace,
897 struct ring_buffer_event *event;
898 struct ftrace_graph_ent_entry *entry;
900 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
903 event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_ENT,
904 sizeof(*entry), flags, pc);
907 entry = ring_buffer_event_data(event);
908 entry->graph_ent = *trace;
909 ring_buffer_unlock_commit(global_trace.buffer, event);
912 static void __trace_graph_return(struct trace_array *tr,
913 struct ftrace_graph_ret *trace,
917 struct ring_buffer_event *event;
918 struct ftrace_graph_ret_entry *entry;
920 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
923 event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_RET,
924 sizeof(*entry), flags, pc);
927 entry = ring_buffer_event_data(event);
929 ring_buffer_unlock_commit(global_trace.buffer, event);
934 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
935 unsigned long ip, unsigned long parent_ip, unsigned long flags,
938 if (likely(!atomic_read(&data->disabled)))
939 trace_function(tr, ip, parent_ip, flags, pc);
942 static void __ftrace_trace_stack(struct trace_array *tr,
946 #ifdef CONFIG_STACKTRACE
947 struct ring_buffer_event *event;
948 struct stack_entry *entry;
949 struct stack_trace trace;
951 event = trace_buffer_lock_reserve(tr, TRACE_STACK,
952 sizeof(*entry), flags, pc);
955 entry = ring_buffer_event_data(event);
956 memset(&entry->caller, 0, sizeof(entry->caller));
958 trace.nr_entries = 0;
959 trace.max_entries = FTRACE_STACK_ENTRIES;
961 trace.entries = entry->caller;
963 save_stack_trace(&trace);
964 ring_buffer_unlock_commit(tr->buffer, event);
968 static void ftrace_trace_stack(struct trace_array *tr,
972 if (!(trace_flags & TRACE_ITER_STACKTRACE))
975 __ftrace_trace_stack(tr, flags, skip, pc);
978 void __trace_stack(struct trace_array *tr,
982 __ftrace_trace_stack(tr, flags, skip, pc);
985 static void ftrace_trace_userstack(struct trace_array *tr,
986 unsigned long flags, int pc)
988 #ifdef CONFIG_STACKTRACE
989 struct ring_buffer_event *event;
990 struct userstack_entry *entry;
991 struct stack_trace trace;
993 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
996 event = trace_buffer_lock_reserve(tr, TRACE_USER_STACK,
997 sizeof(*entry), flags, pc);
1000 entry = ring_buffer_event_data(event);
1002 memset(&entry->caller, 0, sizeof(entry->caller));
1004 trace.nr_entries = 0;
1005 trace.max_entries = FTRACE_STACK_ENTRIES;
1007 trace.entries = entry->caller;
1009 save_stack_trace_user(&trace);
1010 ring_buffer_unlock_commit(tr->buffer, event);
1015 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1017 ftrace_trace_userstack(tr, flags, preempt_count());
1022 ftrace_trace_special(void *__tr,
1023 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1026 struct ring_buffer_event *event;
1027 struct trace_array *tr = __tr;
1028 struct special_entry *entry;
1030 event = trace_buffer_lock_reserve(tr, TRACE_SPECIAL,
1031 sizeof(*entry), 0, pc);
1034 entry = ring_buffer_event_data(event);
1038 trace_buffer_unlock_commit(tr, event, 0, pc);
1042 __trace_special(void *__tr, void *__data,
1043 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1045 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1049 tracing_sched_switch_trace(struct trace_array *tr,
1050 struct task_struct *prev,
1051 struct task_struct *next,
1052 unsigned long flags, int pc)
1054 struct ring_buffer_event *event;
1055 struct ctx_switch_entry *entry;
1057 event = trace_buffer_lock_reserve(tr, TRACE_CTX,
1058 sizeof(*entry), flags, pc);
1061 entry = ring_buffer_event_data(event);
1062 entry->prev_pid = prev->pid;
1063 entry->prev_prio = prev->prio;
1064 entry->prev_state = prev->state;
1065 entry->next_pid = next->pid;
1066 entry->next_prio = next->prio;
1067 entry->next_state = next->state;
1068 entry->next_cpu = task_cpu(next);
1069 trace_buffer_unlock_commit(tr, event, flags, pc);
1073 tracing_sched_wakeup_trace(struct trace_array *tr,
1074 struct task_struct *wakee,
1075 struct task_struct *curr,
1076 unsigned long flags, int pc)
1078 struct ring_buffer_event *event;
1079 struct ctx_switch_entry *entry;
1081 event = trace_buffer_lock_reserve(tr, TRACE_WAKE,
1082 sizeof(*entry), flags, pc);
1085 entry = ring_buffer_event_data(event);
1086 entry->prev_pid = curr->pid;
1087 entry->prev_prio = curr->prio;
1088 entry->prev_state = curr->state;
1089 entry->next_pid = wakee->pid;
1090 entry->next_prio = wakee->prio;
1091 entry->next_state = wakee->state;
1092 entry->next_cpu = task_cpu(wakee);
1094 ring_buffer_unlock_commit(tr->buffer, event);
1095 ftrace_trace_stack(tr, flags, 6, pc);
1096 ftrace_trace_userstack(tr, flags, pc);
1100 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1102 struct trace_array *tr = &global_trace;
1103 struct trace_array_cpu *data;
1104 unsigned long flags;
1108 if (tracing_disabled)
1111 pc = preempt_count();
1112 local_irq_save(flags);
1113 cpu = raw_smp_processor_id();
1114 data = tr->data[cpu];
1116 if (likely(atomic_inc_return(&data->disabled) == 1))
1117 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1119 atomic_dec(&data->disabled);
1120 local_irq_restore(flags);
1123 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1124 int trace_graph_entry(struct ftrace_graph_ent *trace)
1126 struct trace_array *tr = &global_trace;
1127 struct trace_array_cpu *data;
1128 unsigned long flags;
1133 if (!ftrace_trace_task(current))
1136 if (!ftrace_graph_addr(trace->func))
1139 local_irq_save(flags);
1140 cpu = raw_smp_processor_id();
1141 data = tr->data[cpu];
1142 disabled = atomic_inc_return(&data->disabled);
1143 if (likely(disabled == 1)) {
1144 pc = preempt_count();
1145 __trace_graph_entry(tr, trace, flags, pc);
1147 /* Only do the atomic if it is not already set */
1148 if (!test_tsk_trace_graph(current))
1149 set_tsk_trace_graph(current);
1150 atomic_dec(&data->disabled);
1151 local_irq_restore(flags);
1156 void trace_graph_return(struct ftrace_graph_ret *trace)
1158 struct trace_array *tr = &global_trace;
1159 struct trace_array_cpu *data;
1160 unsigned long flags;
1165 local_irq_save(flags);
1166 cpu = raw_smp_processor_id();
1167 data = tr->data[cpu];
1168 disabled = atomic_inc_return(&data->disabled);
1169 if (likely(disabled == 1)) {
1170 pc = preempt_count();
1171 __trace_graph_return(tr, trace, flags, pc);
1174 clear_tsk_trace_graph(current);
1175 atomic_dec(&data->disabled);
1176 local_irq_restore(flags);
1178 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1182 * trace_vprintk - write binary msg to tracing buffer
1185 int trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args)
1187 static raw_spinlock_t trace_buf_lock =
1188 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
1189 static u32 trace_buf[TRACE_BUF_SIZE];
1191 struct ring_buffer_event *event;
1192 struct trace_array *tr = &global_trace;
1193 struct trace_array_cpu *data;
1194 struct print_entry *entry;
1195 unsigned long flags;
1197 int cpu, len = 0, size, pc;
1199 if (unlikely(tracing_selftest_running || tracing_disabled))
1202 /* Don't pollute graph traces with trace_vprintk internals */
1203 pause_graph_tracing();
1205 pc = preempt_count();
1206 resched = ftrace_preempt_disable();
1207 cpu = raw_smp_processor_id();
1208 data = tr->data[cpu];
1210 if (unlikely(atomic_read(&data->disabled)))
1213 /* Lockdep uses trace_printk for lock tracing */
1214 local_irq_save(flags);
1215 __raw_spin_lock(&trace_buf_lock);
1216 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1218 if (len > TRACE_BUF_SIZE || len < 0)
1221 size = sizeof(*entry) + sizeof(u32) * len;
1222 event = trace_buffer_lock_reserve(tr, TRACE_PRINT, size, flags, pc);
1225 entry = ring_buffer_event_data(event);
1227 entry->depth = depth;
1230 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1231 ring_buffer_unlock_commit(tr->buffer, event);
1234 __raw_spin_unlock(&trace_buf_lock);
1235 local_irq_restore(flags);
1238 ftrace_preempt_enable(resched);
1239 unpause_graph_tracing();
1243 EXPORT_SYMBOL_GPL(trace_vprintk);
1245 enum trace_file_type {
1246 TRACE_FILE_LAT_FMT = 1,
1247 TRACE_FILE_ANNOTATE = 2,
1250 static void trace_iterator_increment(struct trace_iterator *iter)
1252 /* Don't allow ftrace to trace into the ring buffers */
1253 ftrace_disable_cpu();
1256 if (iter->buffer_iter[iter->cpu])
1257 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1259 ftrace_enable_cpu();
1262 static struct trace_entry *
1263 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1265 struct ring_buffer_event *event;
1266 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1268 /* Don't allow ftrace to trace into the ring buffers */
1269 ftrace_disable_cpu();
1272 event = ring_buffer_iter_peek(buf_iter, ts);
1274 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1276 ftrace_enable_cpu();
1278 return event ? ring_buffer_event_data(event) : NULL;
1281 static struct trace_entry *
1282 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1284 struct ring_buffer *buffer = iter->tr->buffer;
1285 struct trace_entry *ent, *next = NULL;
1286 int cpu_file = iter->cpu_file;
1287 u64 next_ts = 0, ts;
1292 * If we are in a per_cpu trace file, don't bother by iterating over
1293 * all cpu and peek directly.
1295 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1296 if (ring_buffer_empty_cpu(buffer, cpu_file))
1298 ent = peek_next_entry(iter, cpu_file, ent_ts);
1300 *ent_cpu = cpu_file;
1305 for_each_tracing_cpu(cpu) {
1307 if (ring_buffer_empty_cpu(buffer, cpu))
1310 ent = peek_next_entry(iter, cpu, &ts);
1313 * Pick the entry with the smallest timestamp:
1315 if (ent && (!next || ts < next_ts)) {
1323 *ent_cpu = next_cpu;
1331 /* Find the next real entry, without updating the iterator itself */
1332 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1333 int *ent_cpu, u64 *ent_ts)
1335 return __find_next_entry(iter, ent_cpu, ent_ts);
1338 /* Find the next real entry, and increment the iterator to the next entry */
1339 static void *find_next_entry_inc(struct trace_iterator *iter)
1341 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1344 trace_iterator_increment(iter);
1346 return iter->ent ? iter : NULL;
1349 static void trace_consume(struct trace_iterator *iter)
1351 /* Don't allow ftrace to trace into the ring buffers */
1352 ftrace_disable_cpu();
1353 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1354 ftrace_enable_cpu();
1357 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1359 struct trace_iterator *iter = m->private;
1365 /* can't go backwards */
1370 ent = find_next_entry_inc(iter);
1374 while (ent && iter->idx < i)
1375 ent = find_next_entry_inc(iter);
1383 * No necessary locking here. The worst thing which can
1384 * happen is loosing events consumed at the same time
1385 * by a trace_pipe reader.
1386 * Other than that, we don't risk to crash the ring buffer
1387 * because it serializes the readers.
1389 * The current tracer is copied to avoid a global locking
1392 static void *s_start(struct seq_file *m, loff_t *pos)
1394 struct trace_iterator *iter = m->private;
1395 static struct tracer *old_tracer;
1396 int cpu_file = iter->cpu_file;
1401 /* copy the tracer to avoid using a global lock all around */
1402 mutex_lock(&trace_types_lock);
1403 if (unlikely(old_tracer != current_trace && current_trace)) {
1404 old_tracer = current_trace;
1405 *iter->trace = *current_trace;
1407 mutex_unlock(&trace_types_lock);
1409 atomic_inc(&trace_record_cmdline_disabled);
1411 if (*pos != iter->pos) {
1416 ftrace_disable_cpu();
1418 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1419 for_each_tracing_cpu(cpu)
1420 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1422 ring_buffer_iter_reset(iter->buffer_iter[cpu_file]);
1425 ftrace_enable_cpu();
1427 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1432 p = s_next(m, p, &l);
1438 static void s_stop(struct seq_file *m, void *p)
1440 atomic_dec(&trace_record_cmdline_disabled);
1443 static void print_lat_help_header(struct seq_file *m)
1445 seq_puts(m, "# _------=> CPU# \n");
1446 seq_puts(m, "# / _-----=> irqs-off \n");
1447 seq_puts(m, "# | / _----=> need-resched \n");
1448 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1449 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1450 seq_puts(m, "# |||| / \n");
1451 seq_puts(m, "# ||||| delay \n");
1452 seq_puts(m, "# cmd pid ||||| time | caller \n");
1453 seq_puts(m, "# \\ / ||||| \\ | / \n");
1456 static void print_func_help_header(struct seq_file *m)
1458 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1459 seq_puts(m, "# | | | | |\n");
1464 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1466 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1467 struct trace_array *tr = iter->tr;
1468 struct trace_array_cpu *data = tr->data[tr->cpu];
1469 struct tracer *type = current_trace;
1470 unsigned long total;
1471 unsigned long entries;
1472 const char *name = "preemption";
1477 entries = ring_buffer_entries(iter->tr->buffer);
1479 ring_buffer_overruns(iter->tr->buffer);
1481 seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1483 seq_puts(m, "-----------------------------------"
1484 "---------------------------------\n");
1485 seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1486 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1487 nsecs_to_usecs(data->saved_latency),
1491 #if defined(CONFIG_PREEMPT_NONE)
1493 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1495 #elif defined(CONFIG_PREEMPT)
1500 /* These are reserved for later use */
1503 seq_printf(m, " #P:%d)\n", num_online_cpus());
1507 seq_puts(m, " -----------------\n");
1508 seq_printf(m, " | task: %.16s-%d "
1509 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1510 data->comm, data->pid, data->uid, data->nice,
1511 data->policy, data->rt_priority);
1512 seq_puts(m, " -----------------\n");
1514 if (data->critical_start) {
1515 seq_puts(m, " => started at: ");
1516 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1517 trace_print_seq(m, &iter->seq);
1518 seq_puts(m, "\n => ended at: ");
1519 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1520 trace_print_seq(m, &iter->seq);
1527 static void test_cpu_buff_start(struct trace_iterator *iter)
1529 struct trace_seq *s = &iter->seq;
1531 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1534 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1537 if (cpumask_test_cpu(iter->cpu, iter->started))
1540 cpumask_set_cpu(iter->cpu, iter->started);
1541 trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1544 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1546 struct trace_seq *s = &iter->seq;
1547 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1548 struct trace_entry *entry;
1549 struct trace_event *event;
1553 test_cpu_buff_start(iter);
1555 event = ftrace_find_event(entry->type);
1557 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1558 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1559 if (!trace_print_lat_context(iter))
1562 if (!trace_print_context(iter))
1568 return event->trace(iter, sym_flags);
1570 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1573 return TRACE_TYPE_HANDLED;
1575 return TRACE_TYPE_PARTIAL_LINE;
1578 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1580 struct trace_seq *s = &iter->seq;
1581 struct trace_entry *entry;
1582 struct trace_event *event;
1586 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1587 if (!trace_seq_printf(s, "%d %d %llu ",
1588 entry->pid, iter->cpu, iter->ts))
1592 event = ftrace_find_event(entry->type);
1594 return event->raw(iter, 0);
1596 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1599 return TRACE_TYPE_HANDLED;
1601 return TRACE_TYPE_PARTIAL_LINE;
1604 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1606 struct trace_seq *s = &iter->seq;
1607 unsigned char newline = '\n';
1608 struct trace_entry *entry;
1609 struct trace_event *event;
1613 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1614 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1615 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1616 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1619 event = ftrace_find_event(entry->type);
1621 enum print_line_t ret = event->hex(iter, 0);
1622 if (ret != TRACE_TYPE_HANDLED)
1626 SEQ_PUT_FIELD_RET(s, newline);
1628 return TRACE_TYPE_HANDLED;
1631 static enum print_line_t print_printk_msg_only(struct trace_iterator *iter)
1633 struct trace_seq *s = &iter->seq;
1634 struct trace_entry *entry = iter->ent;
1635 struct print_entry *field;
1638 trace_assign_type(field, entry);
1640 ret = trace_seq_bprintf(s, field->fmt, field->buf);
1642 return TRACE_TYPE_PARTIAL_LINE;
1644 return TRACE_TYPE_HANDLED;
1647 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1649 struct trace_seq *s = &iter->seq;
1650 struct trace_entry *entry;
1651 struct trace_event *event;
1655 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1656 SEQ_PUT_FIELD_RET(s, entry->pid);
1657 SEQ_PUT_FIELD_RET(s, iter->cpu);
1658 SEQ_PUT_FIELD_RET(s, iter->ts);
1661 event = ftrace_find_event(entry->type);
1662 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1665 static int trace_empty(struct trace_iterator *iter)
1669 /* If we are looking at one CPU buffer, only check that one */
1670 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1671 cpu = iter->cpu_file;
1672 if (iter->buffer_iter[cpu]) {
1673 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1676 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1682 for_each_tracing_cpu(cpu) {
1683 if (iter->buffer_iter[cpu]) {
1684 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1687 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1695 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1697 enum print_line_t ret;
1699 if (iter->trace && iter->trace->print_line) {
1700 ret = iter->trace->print_line(iter);
1701 if (ret != TRACE_TYPE_UNHANDLED)
1705 if (iter->ent->type == TRACE_PRINT &&
1706 trace_flags & TRACE_ITER_PRINTK &&
1707 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1708 return print_printk_msg_only(iter);
1710 if (trace_flags & TRACE_ITER_BIN)
1711 return print_bin_fmt(iter);
1713 if (trace_flags & TRACE_ITER_HEX)
1714 return print_hex_fmt(iter);
1716 if (trace_flags & TRACE_ITER_RAW)
1717 return print_raw_fmt(iter);
1719 return print_trace_fmt(iter);
1722 static int s_show(struct seq_file *m, void *v)
1724 struct trace_iterator *iter = v;
1726 if (iter->ent == NULL) {
1728 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1731 if (iter->trace && iter->trace->print_header)
1732 iter->trace->print_header(m);
1733 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1734 /* print nothing if the buffers are empty */
1735 if (trace_empty(iter))
1737 print_trace_header(m, iter);
1738 if (!(trace_flags & TRACE_ITER_VERBOSE))
1739 print_lat_help_header(m);
1741 if (!(trace_flags & TRACE_ITER_VERBOSE))
1742 print_func_help_header(m);
1745 print_trace_line(iter);
1746 trace_print_seq(m, &iter->seq);
1752 static struct seq_operations tracer_seq_ops = {
1759 static struct trace_iterator *
1760 __tracing_open(struct inode *inode, struct file *file)
1762 long cpu_file = (long) inode->i_private;
1763 void *fail_ret = ERR_PTR(-ENOMEM);
1764 struct trace_iterator *iter;
1768 if (tracing_disabled)
1769 return ERR_PTR(-ENODEV);
1771 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1773 return ERR_PTR(-ENOMEM);
1776 * We make a copy of the current tracer to avoid concurrent
1777 * changes on it while we are reading.
1779 mutex_lock(&trace_types_lock);
1780 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
1785 *iter->trace = *current_trace;
1787 if (current_trace && current_trace->print_max)
1790 iter->tr = &global_trace;
1792 mutex_init(&iter->mutex);
1793 iter->cpu_file = cpu_file;
1795 /* Notify the tracer early; before we stop tracing. */
1796 if (iter->trace && iter->trace->open)
1797 iter->trace->open(iter);
1799 /* Annotate start of buffers if we had overruns */
1800 if (ring_buffer_overruns(iter->tr->buffer))
1801 iter->iter_flags |= TRACE_FILE_ANNOTATE;
1803 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
1804 for_each_tracing_cpu(cpu) {
1806 iter->buffer_iter[cpu] =
1807 ring_buffer_read_start(iter->tr->buffer, cpu);
1810 cpu = iter->cpu_file;
1811 iter->buffer_iter[cpu] =
1812 ring_buffer_read_start(iter->tr->buffer, cpu);
1815 /* TODO stop tracer */
1816 ret = seq_open(file, &tracer_seq_ops);
1818 fail_ret = ERR_PTR(ret);
1822 m = file->private_data;
1825 /* stop the trace while dumping */
1828 mutex_unlock(&trace_types_lock);
1833 for_each_tracing_cpu(cpu) {
1834 if (iter->buffer_iter[cpu])
1835 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1838 mutex_unlock(&trace_types_lock);
1845 int tracing_open_generic(struct inode *inode, struct file *filp)
1847 if (tracing_disabled)
1850 filp->private_data = inode->i_private;
1854 static int tracing_release(struct inode *inode, struct file *file)
1856 struct seq_file *m = (struct seq_file *)file->private_data;
1857 struct trace_iterator *iter = m->private;
1860 mutex_lock(&trace_types_lock);
1861 for_each_tracing_cpu(cpu) {
1862 if (iter->buffer_iter[cpu])
1863 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1866 if (iter->trace && iter->trace->close)
1867 iter->trace->close(iter);
1869 /* reenable tracing if it was previously enabled */
1871 mutex_unlock(&trace_types_lock);
1873 seq_release(inode, file);
1874 mutex_destroy(&iter->mutex);
1880 static int tracing_open(struct inode *inode, struct file *file)
1882 struct trace_iterator *iter;
1885 iter = __tracing_open(inode, file);
1887 ret = PTR_ERR(iter);
1888 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
1889 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1895 t_next(struct seq_file *m, void *v, loff_t *pos)
1897 struct tracer *t = m->private;
1909 static void *t_start(struct seq_file *m, loff_t *pos)
1911 struct tracer *t = m->private;
1914 mutex_lock(&trace_types_lock);
1915 for (; t && l < *pos; t = t_next(m, t, &l))
1921 static void t_stop(struct seq_file *m, void *p)
1923 mutex_unlock(&trace_types_lock);
1926 static int t_show(struct seq_file *m, void *v)
1928 struct tracer *t = v;
1933 seq_printf(m, "%s", t->name);
1942 static struct seq_operations show_traces_seq_ops = {
1949 static int show_traces_open(struct inode *inode, struct file *file)
1953 if (tracing_disabled)
1956 ret = seq_open(file, &show_traces_seq_ops);
1958 struct seq_file *m = file->private_data;
1959 m->private = trace_types;
1965 static const struct file_operations tracing_fops = {
1966 .open = tracing_open,
1968 .llseek = seq_lseek,
1969 .release = tracing_release,
1972 static const struct file_operations show_traces_fops = {
1973 .open = show_traces_open,
1975 .release = seq_release,
1979 * Only trace on a CPU if the bitmask is set:
1981 static cpumask_var_t tracing_cpumask;
1984 * The tracer itself will not take this lock, but still we want
1985 * to provide a consistent cpumask to user-space:
1987 static DEFINE_MUTEX(tracing_cpumask_update_lock);
1990 * Temporary storage for the character representation of the
1991 * CPU bitmask (and one more byte for the newline):
1993 static char mask_str[NR_CPUS + 1];
1996 tracing_cpumask_read(struct file *filp, char __user *ubuf,
1997 size_t count, loff_t *ppos)
2001 mutex_lock(&tracing_cpumask_update_lock);
2003 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2004 if (count - len < 2) {
2008 len += sprintf(mask_str + len, "\n");
2009 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2012 mutex_unlock(&tracing_cpumask_update_lock);
2018 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2019 size_t count, loff_t *ppos)
2022 cpumask_var_t tracing_cpumask_new;
2024 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2027 mutex_lock(&tracing_cpumask_update_lock);
2028 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2032 local_irq_disable();
2033 __raw_spin_lock(&ftrace_max_lock);
2034 for_each_tracing_cpu(cpu) {
2036 * Increase/decrease the disabled counter if we are
2037 * about to flip a bit in the cpumask:
2039 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2040 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2041 atomic_inc(&global_trace.data[cpu]->disabled);
2043 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2044 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2045 atomic_dec(&global_trace.data[cpu]->disabled);
2048 __raw_spin_unlock(&ftrace_max_lock);
2051 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2053 mutex_unlock(&tracing_cpumask_update_lock);
2054 free_cpumask_var(tracing_cpumask_new);
2059 mutex_unlock(&tracing_cpumask_update_lock);
2060 free_cpumask_var(tracing_cpumask);
2065 static const struct file_operations tracing_cpumask_fops = {
2066 .open = tracing_open_generic,
2067 .read = tracing_cpumask_read,
2068 .write = tracing_cpumask_write,
2072 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2073 size_t cnt, loff_t *ppos)
2075 struct tracer_opt *trace_opts;
2083 /* calculate max size */
2084 for (i = 0; trace_options[i]; i++) {
2085 len += strlen(trace_options[i]);
2086 len += 3; /* "no" and newline */
2089 mutex_lock(&trace_types_lock);
2090 tracer_flags = current_trace->flags->val;
2091 trace_opts = current_trace->flags->opts;
2094 * Increase the size with names of options specific
2095 * of the current tracer.
2097 for (i = 0; trace_opts[i].name; i++) {
2098 len += strlen(trace_opts[i].name);
2099 len += 3; /* "no" and newline */
2102 /* +2 for \n and \0 */
2103 buf = kmalloc(len + 2, GFP_KERNEL);
2105 mutex_unlock(&trace_types_lock);
2109 for (i = 0; trace_options[i]; i++) {
2110 if (trace_flags & (1 << i))
2111 r += sprintf(buf + r, "%s\n", trace_options[i]);
2113 r += sprintf(buf + r, "no%s\n", trace_options[i]);
2116 for (i = 0; trace_opts[i].name; i++) {
2117 if (tracer_flags & trace_opts[i].bit)
2118 r += sprintf(buf + r, "%s\n",
2119 trace_opts[i].name);
2121 r += sprintf(buf + r, "no%s\n",
2122 trace_opts[i].name);
2124 mutex_unlock(&trace_types_lock);
2126 WARN_ON(r >= len + 2);
2128 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2134 /* Try to assign a tracer specific option */
2135 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2137 struct tracer_flags *trace_flags = trace->flags;
2138 struct tracer_opt *opts = NULL;
2142 for (i = 0; trace_flags->opts[i].name; i++) {
2143 opts = &trace_flags->opts[i];
2144 len = strlen(opts->name);
2146 if (strncmp(cmp, opts->name, len) == 0) {
2147 ret = trace->set_flag(trace_flags->val,
2153 if (!trace_flags->opts[i].name)
2156 /* Refused to handle */
2161 trace_flags->val &= ~opts->bit;
2163 trace_flags->val |= opts->bit;
2169 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2170 size_t cnt, loff_t *ppos)
2178 if (cnt >= sizeof(buf))
2181 if (copy_from_user(&buf, ubuf, cnt))
2186 if (strncmp(buf, "no", 2) == 0) {
2191 for (i = 0; trace_options[i]; i++) {
2192 int len = strlen(trace_options[i]);
2194 if (strncmp(cmp, trace_options[i], len) == 0) {
2196 trace_flags &= ~(1 << i);
2198 trace_flags |= (1 << i);
2203 /* If no option could be set, test the specific tracer options */
2204 if (!trace_options[i]) {
2205 mutex_lock(&trace_types_lock);
2206 ret = set_tracer_option(current_trace, cmp, neg);
2207 mutex_unlock(&trace_types_lock);
2217 static const struct file_operations tracing_iter_fops = {
2218 .open = tracing_open_generic,
2219 .read = tracing_trace_options_read,
2220 .write = tracing_trace_options_write,
2223 static const char readme_msg[] =
2224 "tracing mini-HOWTO:\n\n"
2226 "# mount -t debugfs nodev /debug\n\n"
2227 "# cat /debug/tracing/available_tracers\n"
2228 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2229 "# cat /debug/tracing/current_tracer\n"
2231 "# echo sched_switch > /debug/tracing/current_tracer\n"
2232 "# cat /debug/tracing/current_tracer\n"
2234 "# cat /debug/tracing/trace_options\n"
2235 "noprint-parent nosym-offset nosym-addr noverbose\n"
2236 "# echo print-parent > /debug/tracing/trace_options\n"
2237 "# echo 1 > /debug/tracing/tracing_enabled\n"
2238 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2239 "echo 0 > /debug/tracing/tracing_enabled\n"
2243 tracing_readme_read(struct file *filp, char __user *ubuf,
2244 size_t cnt, loff_t *ppos)
2246 return simple_read_from_buffer(ubuf, cnt, ppos,
2247 readme_msg, strlen(readme_msg));
2250 static const struct file_operations tracing_readme_fops = {
2251 .open = tracing_open_generic,
2252 .read = tracing_readme_read,
2256 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2257 size_t cnt, loff_t *ppos)
2262 r = sprintf(buf, "%u\n", tracer_enabled);
2263 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2267 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2268 size_t cnt, loff_t *ppos)
2270 struct trace_array *tr = filp->private_data;
2275 if (cnt >= sizeof(buf))
2278 if (copy_from_user(&buf, ubuf, cnt))
2283 ret = strict_strtoul(buf, 10, &val);
2289 mutex_lock(&trace_types_lock);
2290 if (tracer_enabled ^ val) {
2293 if (current_trace->start)
2294 current_trace->start(tr);
2299 if (current_trace->stop)
2300 current_trace->stop(tr);
2303 mutex_unlock(&trace_types_lock);
2311 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2312 size_t cnt, loff_t *ppos)
2314 char buf[max_tracer_type_len+2];
2317 mutex_lock(&trace_types_lock);
2319 r = sprintf(buf, "%s\n", current_trace->name);
2321 r = sprintf(buf, "\n");
2322 mutex_unlock(&trace_types_lock);
2324 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2327 int tracer_init(struct tracer *t, struct trace_array *tr)
2329 tracing_reset_online_cpus(tr);
2333 static int tracing_resize_ring_buffer(unsigned long size)
2338 * If kernel or user changes the size of the ring buffer
2341 ring_buffer_expanded = 1;
2343 ret = ring_buffer_resize(global_trace.buffer, size);
2347 ret = ring_buffer_resize(max_tr.buffer, size);
2351 r = ring_buffer_resize(global_trace.buffer,
2352 global_trace.entries);
2354 /* AARGH! We are left with different
2355 * size max buffer!!!! */
2357 tracing_disabled = 1;
2362 global_trace.entries = size;
2368 * tracing_update_buffers - used by tracing facility to expand ring buffers
2370 * To save on memory when the tracing is never used on a system with it
2371 * configured in. The ring buffers are set to a minimum size. But once
2372 * a user starts to use the tracing facility, then they need to grow
2373 * to their default size.
2375 * This function is to be called when a tracer is about to be used.
2377 int tracing_update_buffers(void)
2381 if (!ring_buffer_expanded)
2382 ret = tracing_resize_ring_buffer(trace_buf_size);
2387 struct trace_option_dentry;
2389 static struct trace_option_dentry *
2390 create_trace_option_files(struct tracer *tracer);
2393 destroy_trace_option_files(struct trace_option_dentry *topts);
2395 static int tracing_set_tracer(const char *buf)
2397 static struct trace_option_dentry *topts;
2398 struct trace_array *tr = &global_trace;
2402 if (!ring_buffer_expanded) {
2403 ret = tracing_resize_ring_buffer(trace_buf_size);
2409 mutex_lock(&trace_types_lock);
2410 for (t = trace_types; t; t = t->next) {
2411 if (strcmp(t->name, buf) == 0)
2418 if (t == current_trace)
2421 trace_branch_disable();
2422 if (current_trace && current_trace->reset)
2423 current_trace->reset(tr);
2425 destroy_trace_option_files(topts);
2429 topts = create_trace_option_files(current_trace);
2432 ret = tracer_init(t, tr);
2437 trace_branch_enable(tr);
2439 mutex_unlock(&trace_types_lock);
2445 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2446 size_t cnt, loff_t *ppos)
2448 char buf[max_tracer_type_len+1];
2455 if (cnt > max_tracer_type_len)
2456 cnt = max_tracer_type_len;
2458 if (copy_from_user(&buf, ubuf, cnt))
2463 /* strip ending whitespace. */
2464 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2467 err = tracing_set_tracer(buf);
2477 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2478 size_t cnt, loff_t *ppos)
2480 unsigned long *ptr = filp->private_data;
2484 r = snprintf(buf, sizeof(buf), "%ld\n",
2485 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2486 if (r > sizeof(buf))
2488 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2492 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2493 size_t cnt, loff_t *ppos)
2495 unsigned long *ptr = filp->private_data;
2500 if (cnt >= sizeof(buf))
2503 if (copy_from_user(&buf, ubuf, cnt))
2508 ret = strict_strtoul(buf, 10, &val);
2517 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2519 long cpu_file = (long) inode->i_private;
2520 struct trace_iterator *iter;
2523 if (tracing_disabled)
2526 mutex_lock(&trace_types_lock);
2528 /* We only allow one reader per cpu */
2529 if (cpu_file == TRACE_PIPE_ALL_CPU) {
2530 if (!cpumask_empty(tracing_reader_cpumask)) {
2534 cpumask_setall(tracing_reader_cpumask);
2536 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2537 cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2544 /* create a buffer to store the information to pass to userspace */
2545 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2552 * We make a copy of the current tracer to avoid concurrent
2553 * changes on it while we are reading.
2555 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2561 *iter->trace = *current_trace;
2563 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2568 /* trace pipe does not show start of buffer */
2569 cpumask_setall(iter->started);
2571 iter->cpu_file = cpu_file;
2572 iter->tr = &global_trace;
2573 mutex_init(&iter->mutex);
2574 filp->private_data = iter;
2576 if (iter->trace->pipe_open)
2577 iter->trace->pipe_open(iter);
2580 mutex_unlock(&trace_types_lock);
2586 mutex_unlock(&trace_types_lock);
2590 static int tracing_release_pipe(struct inode *inode, struct file *file)
2592 struct trace_iterator *iter = file->private_data;
2594 mutex_lock(&trace_types_lock);
2596 if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2597 cpumask_clear(tracing_reader_cpumask);
2599 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2601 mutex_unlock(&trace_types_lock);
2603 free_cpumask_var(iter->started);
2604 mutex_destroy(&iter->mutex);
2612 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2614 struct trace_iterator *iter = filp->private_data;
2616 if (trace_flags & TRACE_ITER_BLOCK) {
2618 * Always select as readable when in blocking mode
2620 return POLLIN | POLLRDNORM;
2622 if (!trace_empty(iter))
2623 return POLLIN | POLLRDNORM;
2624 poll_wait(filp, &trace_wait, poll_table);
2625 if (!trace_empty(iter))
2626 return POLLIN | POLLRDNORM;
2633 void default_wait_pipe(struct trace_iterator *iter)
2637 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2639 if (trace_empty(iter))
2642 finish_wait(&trace_wait, &wait);
2646 * This is a make-shift waitqueue.
2647 * A tracer might use this callback on some rare cases:
2649 * 1) the current tracer might hold the runqueue lock when it wakes up
2650 * a reader, hence a deadlock (sched, function, and function graph tracers)
2651 * 2) the function tracers, trace all functions, we don't want
2652 * the overhead of calling wake_up and friends
2653 * (and tracing them too)
2655 * Anyway, this is really very primitive wakeup.
2657 void poll_wait_pipe(struct trace_iterator *iter)
2659 set_current_state(TASK_INTERRUPTIBLE);
2660 /* sleep for 100 msecs, and try again. */
2661 schedule_timeout(HZ / 10);
2664 /* Must be called with trace_types_lock mutex held. */
2665 static int tracing_wait_pipe(struct file *filp)
2667 struct trace_iterator *iter = filp->private_data;
2669 while (trace_empty(iter)) {
2671 if ((filp->f_flags & O_NONBLOCK)) {
2675 mutex_unlock(&iter->mutex);
2677 iter->trace->wait_pipe(iter);
2679 mutex_lock(&iter->mutex);
2681 if (signal_pending(current))
2685 * We block until we read something and tracing is disabled.
2686 * We still block if tracing is disabled, but we have never
2687 * read anything. This allows a user to cat this file, and
2688 * then enable tracing. But after we have read something,
2689 * we give an EOF when tracing is again disabled.
2691 * iter->pos will be 0 if we haven't read anything.
2693 if (!tracer_enabled && iter->pos)
2704 tracing_read_pipe(struct file *filp, char __user *ubuf,
2705 size_t cnt, loff_t *ppos)
2707 struct trace_iterator *iter = filp->private_data;
2708 static struct tracer *old_tracer;
2711 /* return any leftover data */
2712 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2716 trace_seq_init(&iter->seq);
2718 /* copy the tracer to avoid using a global lock all around */
2719 mutex_lock(&trace_types_lock);
2720 if (unlikely(old_tracer != current_trace && current_trace)) {
2721 old_tracer = current_trace;
2722 *iter->trace = *current_trace;
2724 mutex_unlock(&trace_types_lock);
2727 * Avoid more than one consumer on a single file descriptor
2728 * This is just a matter of traces coherency, the ring buffer itself
2731 mutex_lock(&iter->mutex);
2732 if (iter->trace->read) {
2733 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2739 sret = tracing_wait_pipe(filp);
2743 /* stop when tracing is finished */
2744 if (trace_empty(iter)) {
2749 if (cnt >= PAGE_SIZE)
2750 cnt = PAGE_SIZE - 1;
2752 /* reset all but tr, trace, and overruns */
2753 memset(&iter->seq, 0,
2754 sizeof(struct trace_iterator) -
2755 offsetof(struct trace_iterator, seq));
2758 while (find_next_entry_inc(iter) != NULL) {
2759 enum print_line_t ret;
2760 int len = iter->seq.len;
2762 ret = print_trace_line(iter);
2763 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2764 /* don't print partial lines */
2765 iter->seq.len = len;
2768 if (ret != TRACE_TYPE_NO_CONSUME)
2769 trace_consume(iter);
2771 if (iter->seq.len >= cnt)
2775 /* Now copy what we have to the user */
2776 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2777 if (iter->seq.readpos >= iter->seq.len)
2778 trace_seq_init(&iter->seq);
2781 * If there was nothing to send to user, inspite of consuming trace
2782 * entries, go back to wait for more entries.
2788 mutex_unlock(&iter->mutex);
2793 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
2794 struct pipe_buffer *buf)
2796 __free_page(buf->page);
2799 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
2802 __free_page(spd->pages[idx]);
2805 static struct pipe_buf_operations tracing_pipe_buf_ops = {
2807 .map = generic_pipe_buf_map,
2808 .unmap = generic_pipe_buf_unmap,
2809 .confirm = generic_pipe_buf_confirm,
2810 .release = tracing_pipe_buf_release,
2811 .steal = generic_pipe_buf_steal,
2812 .get = generic_pipe_buf_get,
2816 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
2821 /* Seq buffer is page-sized, exactly what we need. */
2823 count = iter->seq.len;
2824 ret = print_trace_line(iter);
2825 count = iter->seq.len - count;
2828 iter->seq.len -= count;
2831 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2832 iter->seq.len -= count;
2836 trace_consume(iter);
2838 if (!find_next_entry_inc(iter)) {
2848 static ssize_t tracing_splice_read_pipe(struct file *filp,
2850 struct pipe_inode_info *pipe,
2854 struct page *pages[PIPE_BUFFERS];
2855 struct partial_page partial[PIPE_BUFFERS];
2856 struct trace_iterator *iter = filp->private_data;
2857 struct splice_pipe_desc spd = {
2860 .nr_pages = 0, /* This gets updated below. */
2862 .ops = &tracing_pipe_buf_ops,
2863 .spd_release = tracing_spd_release_pipe,
2865 static struct tracer *old_tracer;
2870 /* copy the tracer to avoid using a global lock all around */
2871 mutex_lock(&trace_types_lock);
2872 if (unlikely(old_tracer != current_trace && current_trace)) {
2873 old_tracer = current_trace;
2874 *iter->trace = *current_trace;
2876 mutex_unlock(&trace_types_lock);
2878 mutex_lock(&iter->mutex);
2880 if (iter->trace->splice_read) {
2881 ret = iter->trace->splice_read(iter, filp,
2882 ppos, pipe, len, flags);
2887 ret = tracing_wait_pipe(filp);
2891 if (!iter->ent && !find_next_entry_inc(iter)) {
2896 /* Fill as many pages as possible. */
2897 for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
2898 pages[i] = alloc_page(GFP_KERNEL);
2902 rem = tracing_fill_pipe_page(rem, iter);
2904 /* Copy the data into the page, so we can start over. */
2905 ret = trace_seq_to_buffer(&iter->seq,
2906 page_address(pages[i]),
2909 __free_page(pages[i]);
2912 partial[i].offset = 0;
2913 partial[i].len = iter->seq.len;
2915 trace_seq_init(&iter->seq);
2918 mutex_unlock(&iter->mutex);
2922 return splice_to_pipe(pipe, &spd);
2925 mutex_unlock(&iter->mutex);
2931 tracing_entries_read(struct file *filp, char __user *ubuf,
2932 size_t cnt, loff_t *ppos)
2934 struct trace_array *tr = filp->private_data;
2938 r = sprintf(buf, "%lu\n", tr->entries >> 10);
2939 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2943 tracing_entries_write(struct file *filp, const char __user *ubuf,
2944 size_t cnt, loff_t *ppos)
2950 if (cnt >= sizeof(buf))
2953 if (copy_from_user(&buf, ubuf, cnt))
2958 ret = strict_strtoul(buf, 10, &val);
2962 /* must have at least 1 entry */
2966 mutex_lock(&trace_types_lock);
2970 /* disable all cpu buffers */
2971 for_each_tracing_cpu(cpu) {
2972 if (global_trace.data[cpu])
2973 atomic_inc(&global_trace.data[cpu]->disabled);
2974 if (max_tr.data[cpu])
2975 atomic_inc(&max_tr.data[cpu]->disabled);
2978 /* value is in KB */
2981 if (val != global_trace.entries) {
2982 ret = tracing_resize_ring_buffer(val);
2991 /* If check pages failed, return ENOMEM */
2992 if (tracing_disabled)
2995 for_each_tracing_cpu(cpu) {
2996 if (global_trace.data[cpu])
2997 atomic_dec(&global_trace.data[cpu]->disabled);
2998 if (max_tr.data[cpu])
2999 atomic_dec(&max_tr.data[cpu]->disabled);
3003 max_tr.entries = global_trace.entries;
3004 mutex_unlock(&trace_types_lock);
3009 static int mark_printk(const char *fmt, ...)
3013 va_start(args, fmt);
3014 ret = trace_vprintk(0, -1, fmt, args);
3020 tracing_mark_write(struct file *filp, const char __user *ubuf,
3021 size_t cnt, loff_t *fpos)
3026 if (tracing_disabled)
3029 if (cnt > TRACE_BUF_SIZE)
3030 cnt = TRACE_BUF_SIZE;
3032 buf = kmalloc(cnt + 1, GFP_KERNEL);
3036 if (copy_from_user(buf, ubuf, cnt)) {
3041 /* Cut from the first nil or newline. */
3043 end = strchr(buf, '\n');
3047 cnt = mark_printk("%s\n", buf);
3054 static const struct file_operations tracing_max_lat_fops = {
3055 .open = tracing_open_generic,
3056 .read = tracing_max_lat_read,
3057 .write = tracing_max_lat_write,
3060 static const struct file_operations tracing_ctrl_fops = {
3061 .open = tracing_open_generic,
3062 .read = tracing_ctrl_read,
3063 .write = tracing_ctrl_write,
3066 static const struct file_operations set_tracer_fops = {
3067 .open = tracing_open_generic,
3068 .read = tracing_set_trace_read,
3069 .write = tracing_set_trace_write,
3072 static const struct file_operations tracing_pipe_fops = {
3073 .open = tracing_open_pipe,
3074 .poll = tracing_poll_pipe,
3075 .read = tracing_read_pipe,
3076 .splice_read = tracing_splice_read_pipe,
3077 .release = tracing_release_pipe,
3080 static const struct file_operations tracing_entries_fops = {
3081 .open = tracing_open_generic,
3082 .read = tracing_entries_read,
3083 .write = tracing_entries_write,
3086 static const struct file_operations tracing_mark_fops = {
3087 .open = tracing_open_generic,
3088 .write = tracing_mark_write,
3091 struct ftrace_buffer_info {
3092 struct trace_array *tr;
3098 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3100 int cpu = (int)(long)inode->i_private;
3101 struct ftrace_buffer_info *info;
3103 if (tracing_disabled)
3106 info = kzalloc(sizeof(*info), GFP_KERNEL);
3110 info->tr = &global_trace;
3112 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3113 /* Force reading ring buffer for first read */
3114 info->read = (unsigned int)-1;
3118 filp->private_data = info;
3128 tracing_buffers_read(struct file *filp, char __user *ubuf,
3129 size_t count, loff_t *ppos)
3131 struct ftrace_buffer_info *info = filp->private_data;
3139 /* Do we have previous read data to read? */
3140 if (info->read < PAGE_SIZE)
3145 ret = ring_buffer_read_page(info->tr->buffer,
3152 pos = ring_buffer_page_len(info->spare);
3154 if (pos < PAGE_SIZE)
3155 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3158 size = PAGE_SIZE - info->read;
3162 ret = copy_to_user(ubuf, info->spare + info->read, size);
3173 static int tracing_buffers_release(struct inode *inode, struct file *file)
3175 struct ftrace_buffer_info *info = file->private_data;
3177 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3184 struct ring_buffer *buffer;
3189 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3190 struct pipe_buffer *buf)
3192 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3197 ring_buffer_free_read_page(ref->buffer, ref->page);
3202 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3203 struct pipe_buffer *buf)
3208 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3209 struct pipe_buffer *buf)
3211 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3216 /* Pipe buffer operations for a buffer. */
3217 static struct pipe_buf_operations buffer_pipe_buf_ops = {
3219 .map = generic_pipe_buf_map,
3220 .unmap = generic_pipe_buf_unmap,
3221 .confirm = generic_pipe_buf_confirm,
3222 .release = buffer_pipe_buf_release,
3223 .steal = buffer_pipe_buf_steal,
3224 .get = buffer_pipe_buf_get,
3228 * Callback from splice_to_pipe(), if we need to release some pages
3229 * at the end of the spd in case we error'ed out in filling the pipe.
3231 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3233 struct buffer_ref *ref =
3234 (struct buffer_ref *)spd->partial[i].private;
3239 ring_buffer_free_read_page(ref->buffer, ref->page);
3241 spd->partial[i].private = 0;
3245 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3246 struct pipe_inode_info *pipe, size_t len,
3249 struct ftrace_buffer_info *info = file->private_data;
3250 struct partial_page partial[PIPE_BUFFERS];
3251 struct page *pages[PIPE_BUFFERS];
3252 struct splice_pipe_desc spd = {
3256 .ops = &buffer_pipe_buf_ops,
3257 .spd_release = buffer_spd_release,
3259 struct buffer_ref *ref;
3264 * We can't seek on a buffer input
3266 if (unlikely(*ppos))
3270 for (i = 0; i < PIPE_BUFFERS && len; i++, len -= size) {
3274 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3278 ref->buffer = info->tr->buffer;
3279 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3285 r = ring_buffer_read_page(ref->buffer, &ref->page,
3288 ring_buffer_free_read_page(ref->buffer,
3295 * zero out any left over data, this is going to
3298 size = ring_buffer_page_len(ref->page);
3299 if (size < PAGE_SIZE)
3300 memset(ref->page + size, 0, PAGE_SIZE - size);
3302 page = virt_to_page(ref->page);
3304 spd.pages[i] = page;
3305 spd.partial[i].len = PAGE_SIZE;
3306 spd.partial[i].offset = 0;
3307 spd.partial[i].private = (unsigned long)ref;
3313 /* did we read anything? */
3314 if (!spd.nr_pages) {
3315 if (flags & SPLICE_F_NONBLOCK)
3323 ret = splice_to_pipe(pipe, &spd);
3328 static const struct file_operations tracing_buffers_fops = {
3329 .open = tracing_buffers_open,
3330 .read = tracing_buffers_read,
3331 .release = tracing_buffers_release,
3332 .splice_read = tracing_buffers_splice_read,
3333 .llseek = no_llseek,
3336 #ifdef CONFIG_DYNAMIC_FTRACE
3338 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3344 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3345 size_t cnt, loff_t *ppos)
3347 static char ftrace_dyn_info_buffer[1024];
3348 static DEFINE_MUTEX(dyn_info_mutex);
3349 unsigned long *p = filp->private_data;
3350 char *buf = ftrace_dyn_info_buffer;
3351 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3354 mutex_lock(&dyn_info_mutex);
3355 r = sprintf(buf, "%ld ", *p);
3357 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3360 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3362 mutex_unlock(&dyn_info_mutex);
3367 static const struct file_operations tracing_dyn_info_fops = {
3368 .open = tracing_open_generic,
3369 .read = tracing_read_dyn_info,
3373 static struct dentry *d_tracer;
3375 struct dentry *tracing_init_dentry(void)
3382 d_tracer = debugfs_create_dir("tracing", NULL);
3384 if (!d_tracer && !once) {
3386 pr_warning("Could not create debugfs directory 'tracing'\n");
3393 static struct dentry *d_percpu;
3395 struct dentry *tracing_dentry_percpu(void)
3398 struct dentry *d_tracer;
3403 d_tracer = tracing_init_dentry();
3408 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3410 if (!d_percpu && !once) {
3412 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3419 static void tracing_init_debugfs_percpu(long cpu)
3421 struct dentry *d_percpu = tracing_dentry_percpu();
3422 struct dentry *entry, *d_cpu;
3423 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3426 if (cpu > 999 || cpu < 0)
3429 sprintf(cpu_dir, "cpu%ld", cpu);
3430 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3432 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3436 /* per cpu trace_pipe */
3437 entry = debugfs_create_file("trace_pipe", 0444, d_cpu,
3438 (void *) cpu, &tracing_pipe_fops);
3440 pr_warning("Could not create debugfs 'trace_pipe' entry\n");
3443 entry = debugfs_create_file("trace", 0444, d_cpu,
3444 (void *) cpu, &tracing_fops);
3446 pr_warning("Could not create debugfs 'trace' entry\n");
3449 #ifdef CONFIG_FTRACE_SELFTEST
3450 /* Let selftest have access to static functions in this file */
3451 #include "trace_selftest.c"
3454 struct trace_option_dentry {
3455 struct tracer_opt *opt;
3456 struct tracer_flags *flags;
3457 struct dentry *entry;
3461 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3464 struct trace_option_dentry *topt = filp->private_data;
3467 if (topt->flags->val & topt->opt->bit)
3472 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3476 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3479 struct trace_option_dentry *topt = filp->private_data;
3484 if (cnt >= sizeof(buf))
3487 if (copy_from_user(&buf, ubuf, cnt))
3492 ret = strict_strtoul(buf, 10, &val);
3499 /* do nothing if already cleared */
3500 if (!(topt->flags->val & topt->opt->bit))
3503 mutex_lock(&trace_types_lock);
3504 if (current_trace->set_flag)
3505 ret = current_trace->set_flag(topt->flags->val,
3507 mutex_unlock(&trace_types_lock);
3510 topt->flags->val &= ~topt->opt->bit;
3513 /* do nothing if already set */
3514 if (topt->flags->val & topt->opt->bit)
3517 mutex_lock(&trace_types_lock);
3518 if (current_trace->set_flag)
3519 ret = current_trace->set_flag(topt->flags->val,
3521 mutex_unlock(&trace_types_lock);
3524 topt->flags->val |= topt->opt->bit;
3537 static const struct file_operations trace_options_fops = {
3538 .open = tracing_open_generic,
3539 .read = trace_options_read,
3540 .write = trace_options_write,
3544 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
3547 long index = (long)filp->private_data;
3550 if (trace_flags & (1 << index))
3555 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3559 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
3562 long index = (long)filp->private_data;
3567 if (cnt >= sizeof(buf))
3570 if (copy_from_user(&buf, ubuf, cnt))
3575 ret = strict_strtoul(buf, 10, &val);
3581 trace_flags &= ~(1 << index);
3584 trace_flags |= 1 << index;
3596 static const struct file_operations trace_options_core_fops = {
3597 .open = tracing_open_generic,
3598 .read = trace_options_core_read,
3599 .write = trace_options_core_write,
3602 static struct dentry *trace_options_init_dentry(void)
3604 struct dentry *d_tracer;
3605 static struct dentry *t_options;
3610 d_tracer = tracing_init_dentry();
3614 t_options = debugfs_create_dir("options", d_tracer);
3616 pr_warning("Could not create debugfs directory 'options'\n");
3624 create_trace_option_file(struct trace_option_dentry *topt,
3625 struct tracer_flags *flags,
3626 struct tracer_opt *opt)
3628 struct dentry *t_options;
3629 struct dentry *entry;
3631 t_options = trace_options_init_dentry();
3635 topt->flags = flags;
3638 entry = debugfs_create_file(opt->name, 0644, t_options, topt,
3639 &trace_options_fops);
3641 topt->entry = entry;
3645 static struct trace_option_dentry *
3646 create_trace_option_files(struct tracer *tracer)
3648 struct trace_option_dentry *topts;
3649 struct tracer_flags *flags;
3650 struct tracer_opt *opts;
3656 flags = tracer->flags;
3658 if (!flags || !flags->opts)
3663 for (cnt = 0; opts[cnt].name; cnt++)
3666 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
3670 for (cnt = 0; opts[cnt].name; cnt++)
3671 create_trace_option_file(&topts[cnt], flags,
3678 destroy_trace_option_files(struct trace_option_dentry *topts)
3685 for (cnt = 0; topts[cnt].opt; cnt++) {
3686 if (topts[cnt].entry)
3687 debugfs_remove(topts[cnt].entry);
3693 static struct dentry *
3694 create_trace_option_core_file(const char *option, long index)
3696 struct dentry *t_options;
3697 struct dentry *entry;
3699 t_options = trace_options_init_dentry();
3703 entry = debugfs_create_file(option, 0644, t_options, (void *)index,
3704 &trace_options_core_fops);
3709 static __init void create_trace_options_dir(void)
3711 struct dentry *t_options;
3712 struct dentry *entry;
3715 t_options = trace_options_init_dentry();
3719 for (i = 0; trace_options[i]; i++) {
3720 entry = create_trace_option_core_file(trace_options[i], i);
3722 pr_warning("Could not create debugfs %s entry\n",
3727 static __init int tracer_init_debugfs(void)
3729 struct dentry *d_tracer;
3730 struct dentry *buffers;
3731 struct dentry *entry;
3734 d_tracer = tracing_init_dentry();
3736 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3737 &global_trace, &tracing_ctrl_fops);
3739 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3741 entry = debugfs_create_file("trace_options", 0644, d_tracer,
3742 NULL, &tracing_iter_fops);
3744 pr_warning("Could not create debugfs 'trace_options' entry\n");
3746 create_trace_options_dir();
3748 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3749 NULL, &tracing_cpumask_fops);
3751 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3753 entry = debugfs_create_file("trace", 0444, d_tracer,
3754 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
3756 pr_warning("Could not create debugfs 'trace' entry\n");
3758 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3759 &global_trace, &show_traces_fops);
3761 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3763 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3764 &global_trace, &set_tracer_fops);
3766 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3768 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3769 &tracing_max_latency,
3770 &tracing_max_lat_fops);
3772 pr_warning("Could not create debugfs "
3773 "'tracing_max_latency' entry\n");
3775 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3776 &tracing_thresh, &tracing_max_lat_fops);
3778 pr_warning("Could not create debugfs "
3779 "'tracing_thresh' entry\n");
3780 entry = debugfs_create_file("README", 0644, d_tracer,
3781 NULL, &tracing_readme_fops);
3783 pr_warning("Could not create debugfs 'README' entry\n");
3785 entry = debugfs_create_file("trace_pipe", 0444, d_tracer,
3786 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
3788 pr_warning("Could not create debugfs "
3789 "'trace_pipe' entry\n");
3791 entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
3792 &global_trace, &tracing_entries_fops);
3794 pr_warning("Could not create debugfs "
3795 "'buffer_size_kb' entry\n");
3797 entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3798 NULL, &tracing_mark_fops);
3800 pr_warning("Could not create debugfs "
3801 "'trace_marker' entry\n");
3803 buffers = debugfs_create_dir("binary_buffers", d_tracer);
3806 pr_warning("Could not create buffers directory\n");
3811 for_each_tracing_cpu(cpu) {
3812 sprintf(buf, "%d", cpu);
3814 entry = debugfs_create_file(buf, 0444, buffers,
3816 &tracing_buffers_fops);
3818 pr_warning("Could not create debugfs buffers "
3819 "'%s' entry\n", buf);
3823 #ifdef CONFIG_DYNAMIC_FTRACE
3824 entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3825 &ftrace_update_tot_cnt,
3826 &tracing_dyn_info_fops);
3828 pr_warning("Could not create debugfs "
3829 "'dyn_ftrace_total_info' entry\n");
3831 #ifdef CONFIG_SYSPROF_TRACER
3832 init_tracer_sysprof_debugfs(d_tracer);
3835 for_each_tracing_cpu(cpu)
3836 tracing_init_debugfs_percpu(cpu);
3841 static int trace_panic_handler(struct notifier_block *this,
3842 unsigned long event, void *unused)
3844 if (ftrace_dump_on_oops)
3849 static struct notifier_block trace_panic_notifier = {
3850 .notifier_call = trace_panic_handler,
3852 .priority = 150 /* priority: INT_MAX >= x >= 0 */
3855 static int trace_die_handler(struct notifier_block *self,
3861 if (ftrace_dump_on_oops)
3870 static struct notifier_block trace_die_notifier = {
3871 .notifier_call = trace_die_handler,
3876 * printk is set to max of 1024, we really don't need it that big.
3877 * Nothing should be printing 1000 characters anyway.
3879 #define TRACE_MAX_PRINT 1000
3882 * Define here KERN_TRACE so that we have one place to modify
3883 * it if we decide to change what log level the ftrace dump
3886 #define KERN_TRACE KERN_EMERG
3889 trace_printk_seq(struct trace_seq *s)
3891 /* Probably should print a warning here. */
3895 /* should be zero ended, but we are paranoid. */
3896 s->buffer[s->len] = 0;
3898 printk(KERN_TRACE "%s", s->buffer);
3903 void ftrace_dump(void)
3905 static DEFINE_SPINLOCK(ftrace_dump_lock);
3906 /* use static because iter can be a bit big for the stack */
3907 static struct trace_iterator iter;
3908 static int dump_ran;
3909 unsigned long flags;
3913 spin_lock_irqsave(&ftrace_dump_lock, flags);
3919 /* No turning back! */
3923 for_each_tracing_cpu(cpu) {
3924 atomic_inc(&global_trace.data[cpu]->disabled);
3927 /* don't look at user memory in panic mode */
3928 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
3930 printk(KERN_TRACE "Dumping ftrace buffer:\n");
3932 /* Simulate the iterator */
3933 iter.tr = &global_trace;
3934 iter.trace = current_trace;
3935 iter.cpu_file = TRACE_PIPE_ALL_CPU;
3938 * We need to stop all tracing on all CPUS to read the
3939 * the next buffer. This is a bit expensive, but is
3940 * not done often. We fill all what we can read,
3941 * and then release the locks again.
3944 while (!trace_empty(&iter)) {
3947 printk(KERN_TRACE "---------------------------------\n");
3951 /* reset all but tr, trace, and overruns */
3952 memset(&iter.seq, 0,
3953 sizeof(struct trace_iterator) -
3954 offsetof(struct trace_iterator, seq));
3955 iter.iter_flags |= TRACE_FILE_LAT_FMT;
3958 if (find_next_entry_inc(&iter) != NULL) {
3959 print_trace_line(&iter);
3960 trace_consume(&iter);
3963 trace_printk_seq(&iter.seq);
3967 printk(KERN_TRACE " (ftrace buffer empty)\n");
3969 printk(KERN_TRACE "---------------------------------\n");
3972 spin_unlock_irqrestore(&ftrace_dump_lock, flags);
3975 __init static int tracer_alloc_buffers(void)
3977 struct trace_array_cpu *data;
3982 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
3985 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
3986 goto out_free_buffer_mask;
3988 if (!alloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
3989 goto out_free_tracing_cpumask;
3991 /* To save memory, keep the ring buffer size to its minimum */
3992 if (ring_buffer_expanded)
3993 ring_buf_size = trace_buf_size;
3997 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
3998 cpumask_copy(tracing_cpumask, cpu_all_mask);
3999 cpumask_clear(tracing_reader_cpumask);
4001 /* TODO: make the number of buffers hot pluggable with CPUS */
4002 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4003 TRACE_BUFFER_FLAGS);
4004 if (!global_trace.buffer) {
4005 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4007 goto out_free_cpumask;
4009 global_trace.entries = ring_buffer_size(global_trace.buffer);
4012 #ifdef CONFIG_TRACER_MAX_TRACE
4013 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4014 TRACE_BUFFER_FLAGS);
4015 if (!max_tr.buffer) {
4016 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4018 ring_buffer_free(global_trace.buffer);
4019 goto out_free_cpumask;
4021 max_tr.entries = ring_buffer_size(max_tr.buffer);
4022 WARN_ON(max_tr.entries != global_trace.entries);
4025 /* Allocate the first page for all buffers */
4026 for_each_tracing_cpu(i) {
4027 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4028 max_tr.data[i] = &per_cpu(max_data, i);
4031 trace_init_cmdlines();
4033 register_tracer(&nop_trace);
4034 current_trace = &nop_trace;
4035 #ifdef CONFIG_BOOT_TRACER
4036 register_tracer(&boot_tracer);
4038 /* All seems OK, enable tracing */
4039 tracing_disabled = 0;
4041 atomic_notifier_chain_register(&panic_notifier_list,
4042 &trace_panic_notifier);
4044 register_die_notifier(&trace_die_notifier);
4048 free_cpumask_var(tracing_reader_cpumask);
4049 out_free_tracing_cpumask:
4050 free_cpumask_var(tracing_cpumask);
4051 out_free_buffer_mask:
4052 free_cpumask_var(tracing_buffer_mask);
4057 __init static int clear_boot_tracer(void)
4060 * The default tracer at boot buffer is an init section.
4061 * This function is called in lateinit. If we did not
4062 * find the boot tracer, then clear it out, to prevent
4063 * later registration from accessing the buffer that is
4064 * about to be freed.
4066 if (!default_bootup_tracer)
4069 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4070 default_bootup_tracer);
4071 default_bootup_tracer = NULL;
4076 early_initcall(tracer_alloc_buffers);
4077 fs_initcall(tracer_init_debugfs);
4078 late_initcall(clear_boot_tracer);