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 | TRACE_ITER_SLEEP_TIME |
259 TRACE_ITER_GRAPH_TIME;
262 * trace_wake_up - wake up tasks waiting for trace input
264 * Simply wakes up any task that is blocked on the trace_wait
265 * queue. These is used with trace_poll for tasks polling the trace.
267 void trace_wake_up(void)
270 * The runqueue_is_locked() can fail, but this is the best we
273 if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
274 wake_up(&trace_wait);
277 static int __init set_buf_size(char *str)
279 unsigned long buf_size;
284 ret = strict_strtoul(str, 0, &buf_size);
285 /* nr_entries can not be zero */
286 if (ret < 0 || buf_size == 0)
288 trace_buf_size = buf_size;
291 __setup("trace_buf_size=", set_buf_size);
293 unsigned long nsecs_to_usecs(unsigned long nsecs)
298 /* These must match the bit postions in trace_iterator_flags */
299 static const char *trace_options[] = {
326 * ftrace_max_lock is used to protect the swapping of buffers
327 * when taking a max snapshot. The buffers themselves are
328 * protected by per_cpu spinlocks. But the action of the swap
329 * needs its own lock.
331 * This is defined as a raw_spinlock_t in order to help
332 * with performance when lockdep debugging is enabled.
334 static raw_spinlock_t ftrace_max_lock =
335 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
338 * Copy the new maximum trace into the separate maximum-trace
339 * structure. (this way the maximum trace is permanently saved,
340 * for later retrieval via /debugfs/tracing/latency_trace)
343 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
345 struct trace_array_cpu *data = tr->data[cpu];
348 max_tr.time_start = data->preempt_timestamp;
350 data = max_tr.data[cpu];
351 data->saved_latency = tracing_max_latency;
353 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
354 data->pid = tsk->pid;
355 data->uid = task_uid(tsk);
356 data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
357 data->policy = tsk->policy;
358 data->rt_priority = tsk->rt_priority;
360 /* record this tasks comm */
361 tracing_record_cmdline(tsk);
364 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
372 if (s->len <= s->readpos)
375 len = s->len - s->readpos;
378 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
388 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
393 if (s->len <= s->readpos)
396 len = s->len - s->readpos;
399 ret = memcpy(buf, s->buffer + s->readpos, cnt);
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 #define NO_CMDLINE_MAP UINT_MAX
637 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
638 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
639 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
640 static int cmdline_idx;
641 static raw_spinlock_t trace_cmdline_lock = __RAW_SPIN_LOCK_UNLOCKED;
643 /* temporary disable recording */
644 static atomic_t trace_record_cmdline_disabled __read_mostly;
646 static void trace_init_cmdlines(void)
648 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
649 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
653 static int trace_stop_count;
654 static DEFINE_SPINLOCK(tracing_start_lock);
657 * ftrace_off_permanent - disable all ftrace code permanently
659 * This should only be called when a serious anomally has
660 * been detected. This will turn off the function tracing,
661 * ring buffers, and other tracing utilites. It takes no
662 * locks and can be called from any context.
664 void ftrace_off_permanent(void)
666 tracing_disabled = 1;
668 tracing_off_permanent();
672 * tracing_start - quick start of the tracer
674 * If tracing is enabled but was stopped by tracing_stop,
675 * this will start the tracer back up.
677 void tracing_start(void)
679 struct ring_buffer *buffer;
682 if (tracing_disabled)
685 spin_lock_irqsave(&tracing_start_lock, flags);
686 if (--trace_stop_count) {
687 if (trace_stop_count < 0) {
688 /* Someone screwed up their debugging */
690 trace_stop_count = 0;
696 buffer = global_trace.buffer;
698 ring_buffer_record_enable(buffer);
700 buffer = max_tr.buffer;
702 ring_buffer_record_enable(buffer);
706 spin_unlock_irqrestore(&tracing_start_lock, flags);
710 * tracing_stop - quick stop of the tracer
712 * Light weight way to stop tracing. Use in conjunction with
715 void tracing_stop(void)
717 struct ring_buffer *buffer;
721 spin_lock_irqsave(&tracing_start_lock, flags);
722 if (trace_stop_count++)
725 buffer = global_trace.buffer;
727 ring_buffer_record_disable(buffer);
729 buffer = max_tr.buffer;
731 ring_buffer_record_disable(buffer);
734 spin_unlock_irqrestore(&tracing_start_lock, flags);
737 void trace_stop_cmdline_recording(void);
739 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 == NO_CMDLINE_MAP) {
757 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
760 * Check whether the cmdline buffer at idx has a pid
761 * mapped. We are going to overwrite that entry so we
762 * need to clear the map_pid_to_cmdline. Otherwise we
763 * would read the new comm for the old pid.
765 pid = map_cmdline_to_pid[idx];
766 if (pid != NO_CMDLINE_MAP)
767 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
769 map_cmdline_to_pid[idx] = tsk->pid;
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 != NO_CMDLINE_MAP)
797 strcpy(comm, saved_cmdlines[map]);
799 strcpy(comm, "<...>");
801 __raw_spin_unlock(&trace_cmdline_lock);
804 void tracing_record_cmdline(struct task_struct *tsk)
806 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
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 static inline void __trace_buffer_unlock_commit(struct trace_array *tr,
856 struct ring_buffer_event *event,
857 unsigned long flags, int pc,
860 ring_buffer_unlock_commit(tr->buffer, event);
862 ftrace_trace_stack(tr, flags, 6, pc);
863 ftrace_trace_userstack(tr, flags, pc);
869 void trace_buffer_unlock_commit(struct trace_array *tr,
870 struct ring_buffer_event *event,
871 unsigned long flags, int pc)
873 __trace_buffer_unlock_commit(tr, event, flags, pc, 1);
876 struct ring_buffer_event *
877 trace_current_buffer_lock_reserve(unsigned char type, unsigned long len,
878 unsigned long flags, int pc)
880 return trace_buffer_lock_reserve(&global_trace,
881 type, len, flags, pc);
884 void trace_current_buffer_unlock_commit(struct ring_buffer_event *event,
885 unsigned long flags, int pc)
887 return __trace_buffer_unlock_commit(&global_trace, event, flags, pc, 1);
890 void trace_nowake_buffer_unlock_commit(struct ring_buffer_event *event,
891 unsigned long flags, int pc)
893 return __trace_buffer_unlock_commit(&global_trace, event, flags, pc, 0);
897 trace_function(struct trace_array *tr,
898 unsigned long ip, unsigned long parent_ip, unsigned long flags,
901 struct ring_buffer_event *event;
902 struct ftrace_entry *entry;
904 /* If we are reading the ring buffer, don't trace */
905 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
908 event = trace_buffer_lock_reserve(tr, TRACE_FN, sizeof(*entry),
912 entry = ring_buffer_event_data(event);
914 entry->parent_ip = parent_ip;
915 ring_buffer_unlock_commit(tr->buffer, event);
918 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
919 static int __trace_graph_entry(struct trace_array *tr,
920 struct ftrace_graph_ent *trace,
924 struct ring_buffer_event *event;
925 struct ftrace_graph_ent_entry *entry;
927 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
930 event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_ENT,
931 sizeof(*entry), flags, pc);
934 entry = ring_buffer_event_data(event);
935 entry->graph_ent = *trace;
936 ring_buffer_unlock_commit(global_trace.buffer, event);
941 static void __trace_graph_return(struct trace_array *tr,
942 struct ftrace_graph_ret *trace,
946 struct ring_buffer_event *event;
947 struct ftrace_graph_ret_entry *entry;
949 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
952 event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_RET,
953 sizeof(*entry), flags, pc);
956 entry = ring_buffer_event_data(event);
958 ring_buffer_unlock_commit(global_trace.buffer, event);
963 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
964 unsigned long ip, unsigned long parent_ip, unsigned long flags,
967 if (likely(!atomic_read(&data->disabled)))
968 trace_function(tr, ip, parent_ip, flags, pc);
971 static void __ftrace_trace_stack(struct trace_array *tr,
975 #ifdef CONFIG_STACKTRACE
976 struct ring_buffer_event *event;
977 struct stack_entry *entry;
978 struct stack_trace trace;
980 event = trace_buffer_lock_reserve(tr, TRACE_STACK,
981 sizeof(*entry), flags, pc);
984 entry = ring_buffer_event_data(event);
985 memset(&entry->caller, 0, sizeof(entry->caller));
987 trace.nr_entries = 0;
988 trace.max_entries = FTRACE_STACK_ENTRIES;
990 trace.entries = entry->caller;
992 save_stack_trace(&trace);
993 ring_buffer_unlock_commit(tr->buffer, event);
997 static void ftrace_trace_stack(struct trace_array *tr,
1001 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1004 __ftrace_trace_stack(tr, flags, skip, pc);
1007 void __trace_stack(struct trace_array *tr,
1008 unsigned long flags,
1011 __ftrace_trace_stack(tr, flags, skip, pc);
1014 static void ftrace_trace_userstack(struct trace_array *tr,
1015 unsigned long flags, int pc)
1017 #ifdef CONFIG_STACKTRACE
1018 struct ring_buffer_event *event;
1019 struct userstack_entry *entry;
1020 struct stack_trace trace;
1022 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1025 event = trace_buffer_lock_reserve(tr, TRACE_USER_STACK,
1026 sizeof(*entry), flags, pc);
1029 entry = ring_buffer_event_data(event);
1031 memset(&entry->caller, 0, sizeof(entry->caller));
1033 trace.nr_entries = 0;
1034 trace.max_entries = FTRACE_STACK_ENTRIES;
1036 trace.entries = entry->caller;
1038 save_stack_trace_user(&trace);
1039 ring_buffer_unlock_commit(tr->buffer, event);
1044 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1046 ftrace_trace_userstack(tr, flags, preempt_count());
1051 ftrace_trace_special(void *__tr,
1052 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1055 struct ring_buffer_event *event;
1056 struct trace_array *tr = __tr;
1057 struct special_entry *entry;
1059 event = trace_buffer_lock_reserve(tr, TRACE_SPECIAL,
1060 sizeof(*entry), 0, pc);
1063 entry = ring_buffer_event_data(event);
1067 trace_buffer_unlock_commit(tr, event, 0, pc);
1071 __trace_special(void *__tr, void *__data,
1072 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1074 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1078 tracing_sched_switch_trace(struct trace_array *tr,
1079 struct task_struct *prev,
1080 struct task_struct *next,
1081 unsigned long flags, int pc)
1083 struct ring_buffer_event *event;
1084 struct ctx_switch_entry *entry;
1086 event = trace_buffer_lock_reserve(tr, TRACE_CTX,
1087 sizeof(*entry), flags, pc);
1090 entry = ring_buffer_event_data(event);
1091 entry->prev_pid = prev->pid;
1092 entry->prev_prio = prev->prio;
1093 entry->prev_state = prev->state;
1094 entry->next_pid = next->pid;
1095 entry->next_prio = next->prio;
1096 entry->next_state = next->state;
1097 entry->next_cpu = task_cpu(next);
1098 trace_buffer_unlock_commit(tr, event, flags, pc);
1102 tracing_sched_wakeup_trace(struct trace_array *tr,
1103 struct task_struct *wakee,
1104 struct task_struct *curr,
1105 unsigned long flags, int pc)
1107 struct ring_buffer_event *event;
1108 struct ctx_switch_entry *entry;
1110 event = trace_buffer_lock_reserve(tr, TRACE_WAKE,
1111 sizeof(*entry), flags, pc);
1114 entry = ring_buffer_event_data(event);
1115 entry->prev_pid = curr->pid;
1116 entry->prev_prio = curr->prio;
1117 entry->prev_state = curr->state;
1118 entry->next_pid = wakee->pid;
1119 entry->next_prio = wakee->prio;
1120 entry->next_state = wakee->state;
1121 entry->next_cpu = task_cpu(wakee);
1123 ring_buffer_unlock_commit(tr->buffer, event);
1124 ftrace_trace_stack(tr, flags, 6, pc);
1125 ftrace_trace_userstack(tr, flags, pc);
1129 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1131 struct trace_array *tr = &global_trace;
1132 struct trace_array_cpu *data;
1133 unsigned long flags;
1137 if (tracing_disabled)
1140 pc = preempt_count();
1141 local_irq_save(flags);
1142 cpu = raw_smp_processor_id();
1143 data = tr->data[cpu];
1145 if (likely(atomic_inc_return(&data->disabled) == 1))
1146 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1148 atomic_dec(&data->disabled);
1149 local_irq_restore(flags);
1152 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1153 int trace_graph_entry(struct ftrace_graph_ent *trace)
1155 struct trace_array *tr = &global_trace;
1156 struct trace_array_cpu *data;
1157 unsigned long flags;
1163 if (!ftrace_trace_task(current))
1166 if (!ftrace_graph_addr(trace->func))
1169 local_irq_save(flags);
1170 cpu = raw_smp_processor_id();
1171 data = tr->data[cpu];
1172 disabled = atomic_inc_return(&data->disabled);
1173 if (likely(disabled == 1)) {
1174 pc = preempt_count();
1175 ret = __trace_graph_entry(tr, trace, flags, pc);
1179 /* Only do the atomic if it is not already set */
1180 if (!test_tsk_trace_graph(current))
1181 set_tsk_trace_graph(current);
1183 atomic_dec(&data->disabled);
1184 local_irq_restore(flags);
1189 void trace_graph_return(struct ftrace_graph_ret *trace)
1191 struct trace_array *tr = &global_trace;
1192 struct trace_array_cpu *data;
1193 unsigned long flags;
1198 local_irq_save(flags);
1199 cpu = raw_smp_processor_id();
1200 data = tr->data[cpu];
1201 disabled = atomic_inc_return(&data->disabled);
1202 if (likely(disabled == 1)) {
1203 pc = preempt_count();
1204 __trace_graph_return(tr, trace, flags, pc);
1207 clear_tsk_trace_graph(current);
1208 atomic_dec(&data->disabled);
1209 local_irq_restore(flags);
1211 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1215 * trace_vbprintk - write binary msg to tracing buffer
1218 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1220 static raw_spinlock_t trace_buf_lock =
1221 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
1222 static u32 trace_buf[TRACE_BUF_SIZE];
1224 struct ring_buffer_event *event;
1225 struct trace_array *tr = &global_trace;
1226 struct trace_array_cpu *data;
1227 struct bprint_entry *entry;
1228 unsigned long flags;
1230 int cpu, len = 0, size, pc;
1232 if (unlikely(tracing_selftest_running || tracing_disabled))
1235 /* Don't pollute graph traces with trace_vprintk internals */
1236 pause_graph_tracing();
1238 pc = preempt_count();
1239 resched = ftrace_preempt_disable();
1240 cpu = raw_smp_processor_id();
1241 data = tr->data[cpu];
1243 if (unlikely(atomic_read(&data->disabled)))
1246 /* Lockdep uses trace_printk for lock tracing */
1247 local_irq_save(flags);
1248 __raw_spin_lock(&trace_buf_lock);
1249 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1251 if (len > TRACE_BUF_SIZE || len < 0)
1254 size = sizeof(*entry) + sizeof(u32) * len;
1255 event = trace_buffer_lock_reserve(tr, TRACE_BPRINT, size, flags, pc);
1258 entry = ring_buffer_event_data(event);
1262 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1263 ring_buffer_unlock_commit(tr->buffer, event);
1266 __raw_spin_unlock(&trace_buf_lock);
1267 local_irq_restore(flags);
1270 ftrace_preempt_enable(resched);
1271 unpause_graph_tracing();
1275 EXPORT_SYMBOL_GPL(trace_vbprintk);
1277 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1279 static raw_spinlock_t trace_buf_lock = __RAW_SPIN_LOCK_UNLOCKED;
1280 static char trace_buf[TRACE_BUF_SIZE];
1282 struct ring_buffer_event *event;
1283 struct trace_array *tr = &global_trace;
1284 struct trace_array_cpu *data;
1285 int cpu, len = 0, size, pc;
1286 struct print_entry *entry;
1287 unsigned long irq_flags;
1289 if (tracing_disabled || tracing_selftest_running)
1292 pc = preempt_count();
1293 preempt_disable_notrace();
1294 cpu = raw_smp_processor_id();
1295 data = tr->data[cpu];
1297 if (unlikely(atomic_read(&data->disabled)))
1300 pause_graph_tracing();
1301 raw_local_irq_save(irq_flags);
1302 __raw_spin_lock(&trace_buf_lock);
1303 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1305 len = min(len, TRACE_BUF_SIZE-1);
1308 size = sizeof(*entry) + len + 1;
1309 event = trace_buffer_lock_reserve(tr, TRACE_PRINT, size, irq_flags, pc);
1312 entry = ring_buffer_event_data(event);
1315 memcpy(&entry->buf, trace_buf, len);
1316 entry->buf[len] = 0;
1317 ring_buffer_unlock_commit(tr->buffer, event);
1320 __raw_spin_unlock(&trace_buf_lock);
1321 raw_local_irq_restore(irq_flags);
1322 unpause_graph_tracing();
1324 preempt_enable_notrace();
1328 EXPORT_SYMBOL_GPL(trace_vprintk);
1330 enum trace_file_type {
1331 TRACE_FILE_LAT_FMT = 1,
1332 TRACE_FILE_ANNOTATE = 2,
1335 static void trace_iterator_increment(struct trace_iterator *iter)
1337 /* Don't allow ftrace to trace into the ring buffers */
1338 ftrace_disable_cpu();
1341 if (iter->buffer_iter[iter->cpu])
1342 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1344 ftrace_enable_cpu();
1347 static struct trace_entry *
1348 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1350 struct ring_buffer_event *event;
1351 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1353 /* Don't allow ftrace to trace into the ring buffers */
1354 ftrace_disable_cpu();
1357 event = ring_buffer_iter_peek(buf_iter, ts);
1359 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1361 ftrace_enable_cpu();
1363 return event ? ring_buffer_event_data(event) : NULL;
1366 static struct trace_entry *
1367 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1369 struct ring_buffer *buffer = iter->tr->buffer;
1370 struct trace_entry *ent, *next = NULL;
1371 int cpu_file = iter->cpu_file;
1372 u64 next_ts = 0, ts;
1377 * If we are in a per_cpu trace file, don't bother by iterating over
1378 * all cpu and peek directly.
1380 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1381 if (ring_buffer_empty_cpu(buffer, cpu_file))
1383 ent = peek_next_entry(iter, cpu_file, ent_ts);
1385 *ent_cpu = cpu_file;
1390 for_each_tracing_cpu(cpu) {
1392 if (ring_buffer_empty_cpu(buffer, cpu))
1395 ent = peek_next_entry(iter, cpu, &ts);
1398 * Pick the entry with the smallest timestamp:
1400 if (ent && (!next || ts < next_ts)) {
1408 *ent_cpu = next_cpu;
1416 /* Find the next real entry, without updating the iterator itself */
1417 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1418 int *ent_cpu, u64 *ent_ts)
1420 return __find_next_entry(iter, ent_cpu, ent_ts);
1423 /* Find the next real entry, and increment the iterator to the next entry */
1424 static void *find_next_entry_inc(struct trace_iterator *iter)
1426 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1429 trace_iterator_increment(iter);
1431 return iter->ent ? iter : NULL;
1434 static void trace_consume(struct trace_iterator *iter)
1436 /* Don't allow ftrace to trace into the ring buffers */
1437 ftrace_disable_cpu();
1438 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1439 ftrace_enable_cpu();
1442 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1444 struct trace_iterator *iter = m->private;
1450 /* can't go backwards */
1455 ent = find_next_entry_inc(iter);
1459 while (ent && iter->idx < i)
1460 ent = find_next_entry_inc(iter);
1468 * No necessary locking here. The worst thing which can
1469 * happen is loosing events consumed at the same time
1470 * by a trace_pipe reader.
1471 * Other than that, we don't risk to crash the ring buffer
1472 * because it serializes the readers.
1474 * The current tracer is copied to avoid a global locking
1477 static void *s_start(struct seq_file *m, loff_t *pos)
1479 struct trace_iterator *iter = m->private;
1480 static struct tracer *old_tracer;
1481 int cpu_file = iter->cpu_file;
1486 /* copy the tracer to avoid using a global lock all around */
1487 mutex_lock(&trace_types_lock);
1488 if (unlikely(old_tracer != current_trace && current_trace)) {
1489 old_tracer = current_trace;
1490 *iter->trace = *current_trace;
1492 mutex_unlock(&trace_types_lock);
1494 atomic_inc(&trace_record_cmdline_disabled);
1496 if (*pos != iter->pos) {
1501 ftrace_disable_cpu();
1503 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1504 for_each_tracing_cpu(cpu)
1505 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1507 ring_buffer_iter_reset(iter->buffer_iter[cpu_file]);
1510 ftrace_enable_cpu();
1512 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1517 p = s_next(m, p, &l);
1523 static void s_stop(struct seq_file *m, void *p)
1525 atomic_dec(&trace_record_cmdline_disabled);
1528 static void print_lat_help_header(struct seq_file *m)
1530 seq_puts(m, "# _------=> CPU# \n");
1531 seq_puts(m, "# / _-----=> irqs-off \n");
1532 seq_puts(m, "# | / _----=> need-resched \n");
1533 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1534 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1535 seq_puts(m, "# |||| / \n");
1536 seq_puts(m, "# ||||| delay \n");
1537 seq_puts(m, "# cmd pid ||||| time | caller \n");
1538 seq_puts(m, "# \\ / ||||| \\ | / \n");
1541 static void print_func_help_header(struct seq_file *m)
1543 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1544 seq_puts(m, "# | | | | |\n");
1549 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1551 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1552 struct trace_array *tr = iter->tr;
1553 struct trace_array_cpu *data = tr->data[tr->cpu];
1554 struct tracer *type = current_trace;
1555 unsigned long total;
1556 unsigned long entries;
1557 const char *name = "preemption";
1562 entries = ring_buffer_entries(iter->tr->buffer);
1564 ring_buffer_overruns(iter->tr->buffer);
1566 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1568 seq_puts(m, "# -----------------------------------"
1569 "---------------------------------\n");
1570 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1571 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1572 nsecs_to_usecs(data->saved_latency),
1576 #if defined(CONFIG_PREEMPT_NONE)
1578 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1580 #elif defined(CONFIG_PREEMPT)
1585 /* These are reserved for later use */
1588 seq_printf(m, " #P:%d)\n", num_online_cpus());
1592 seq_puts(m, "# -----------------\n");
1593 seq_printf(m, "# | task: %.16s-%d "
1594 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1595 data->comm, data->pid, data->uid, data->nice,
1596 data->policy, data->rt_priority);
1597 seq_puts(m, "# -----------------\n");
1599 if (data->critical_start) {
1600 seq_puts(m, "# => started at: ");
1601 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1602 trace_print_seq(m, &iter->seq);
1603 seq_puts(m, "\n# => ended at: ");
1604 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1605 trace_print_seq(m, &iter->seq);
1612 static void test_cpu_buff_start(struct trace_iterator *iter)
1614 struct trace_seq *s = &iter->seq;
1616 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1619 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1622 if (cpumask_test_cpu(iter->cpu, iter->started))
1625 cpumask_set_cpu(iter->cpu, iter->started);
1626 trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1629 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1631 struct trace_seq *s = &iter->seq;
1632 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1633 struct trace_entry *entry;
1634 struct trace_event *event;
1638 test_cpu_buff_start(iter);
1640 event = ftrace_find_event(entry->type);
1642 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1643 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1644 if (!trace_print_lat_context(iter))
1647 if (!trace_print_context(iter))
1653 return event->trace(iter, sym_flags);
1655 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1658 return TRACE_TYPE_HANDLED;
1660 return TRACE_TYPE_PARTIAL_LINE;
1663 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1665 struct trace_seq *s = &iter->seq;
1666 struct trace_entry *entry;
1667 struct trace_event *event;
1671 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1672 if (!trace_seq_printf(s, "%d %d %llu ",
1673 entry->pid, iter->cpu, iter->ts))
1677 event = ftrace_find_event(entry->type);
1679 return event->raw(iter, 0);
1681 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1684 return TRACE_TYPE_HANDLED;
1686 return TRACE_TYPE_PARTIAL_LINE;
1689 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1691 struct trace_seq *s = &iter->seq;
1692 unsigned char newline = '\n';
1693 struct trace_entry *entry;
1694 struct trace_event *event;
1698 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1699 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1700 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1701 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1704 event = ftrace_find_event(entry->type);
1706 enum print_line_t ret = event->hex(iter, 0);
1707 if (ret != TRACE_TYPE_HANDLED)
1711 SEQ_PUT_FIELD_RET(s, newline);
1713 return TRACE_TYPE_HANDLED;
1716 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1718 struct trace_seq *s = &iter->seq;
1719 struct trace_entry *entry;
1720 struct trace_event *event;
1724 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1725 SEQ_PUT_FIELD_RET(s, entry->pid);
1726 SEQ_PUT_FIELD_RET(s, iter->cpu);
1727 SEQ_PUT_FIELD_RET(s, iter->ts);
1730 event = ftrace_find_event(entry->type);
1731 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1734 static int trace_empty(struct trace_iterator *iter)
1738 /* If we are looking at one CPU buffer, only check that one */
1739 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1740 cpu = iter->cpu_file;
1741 if (iter->buffer_iter[cpu]) {
1742 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1745 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1751 for_each_tracing_cpu(cpu) {
1752 if (iter->buffer_iter[cpu]) {
1753 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1756 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1764 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1766 enum print_line_t ret;
1768 if (iter->trace && iter->trace->print_line) {
1769 ret = iter->trace->print_line(iter);
1770 if (ret != TRACE_TYPE_UNHANDLED)
1774 if (iter->ent->type == TRACE_BPRINT &&
1775 trace_flags & TRACE_ITER_PRINTK &&
1776 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1777 return trace_print_bprintk_msg_only(iter);
1779 if (iter->ent->type == TRACE_PRINT &&
1780 trace_flags & TRACE_ITER_PRINTK &&
1781 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1782 return trace_print_printk_msg_only(iter);
1784 if (trace_flags & TRACE_ITER_BIN)
1785 return print_bin_fmt(iter);
1787 if (trace_flags & TRACE_ITER_HEX)
1788 return print_hex_fmt(iter);
1790 if (trace_flags & TRACE_ITER_RAW)
1791 return print_raw_fmt(iter);
1793 return print_trace_fmt(iter);
1796 static int s_show(struct seq_file *m, void *v)
1798 struct trace_iterator *iter = v;
1800 if (iter->ent == NULL) {
1802 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1805 if (iter->trace && iter->trace->print_header)
1806 iter->trace->print_header(m);
1807 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1808 /* print nothing if the buffers are empty */
1809 if (trace_empty(iter))
1811 print_trace_header(m, iter);
1812 if (!(trace_flags & TRACE_ITER_VERBOSE))
1813 print_lat_help_header(m);
1815 if (!(trace_flags & TRACE_ITER_VERBOSE))
1816 print_func_help_header(m);
1819 print_trace_line(iter);
1820 trace_print_seq(m, &iter->seq);
1826 static struct seq_operations tracer_seq_ops = {
1833 static struct trace_iterator *
1834 __tracing_open(struct inode *inode, struct file *file)
1836 long cpu_file = (long) inode->i_private;
1837 void *fail_ret = ERR_PTR(-ENOMEM);
1838 struct trace_iterator *iter;
1842 if (tracing_disabled)
1843 return ERR_PTR(-ENODEV);
1845 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1847 return ERR_PTR(-ENOMEM);
1850 * We make a copy of the current tracer to avoid concurrent
1851 * changes on it while we are reading.
1853 mutex_lock(&trace_types_lock);
1854 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
1859 *iter->trace = *current_trace;
1861 if (current_trace && current_trace->print_max)
1864 iter->tr = &global_trace;
1866 mutex_init(&iter->mutex);
1867 iter->cpu_file = cpu_file;
1869 /* Notify the tracer early; before we stop tracing. */
1870 if (iter->trace && iter->trace->open)
1871 iter->trace->open(iter);
1873 /* Annotate start of buffers if we had overruns */
1874 if (ring_buffer_overruns(iter->tr->buffer))
1875 iter->iter_flags |= TRACE_FILE_ANNOTATE;
1877 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
1878 for_each_tracing_cpu(cpu) {
1880 iter->buffer_iter[cpu] =
1881 ring_buffer_read_start(iter->tr->buffer, cpu);
1884 cpu = iter->cpu_file;
1885 iter->buffer_iter[cpu] =
1886 ring_buffer_read_start(iter->tr->buffer, cpu);
1889 /* TODO stop tracer */
1890 ret = seq_open(file, &tracer_seq_ops);
1892 fail_ret = ERR_PTR(ret);
1896 m = file->private_data;
1899 /* stop the trace while dumping */
1902 mutex_unlock(&trace_types_lock);
1907 for_each_tracing_cpu(cpu) {
1908 if (iter->buffer_iter[cpu])
1909 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1912 mutex_unlock(&trace_types_lock);
1919 int tracing_open_generic(struct inode *inode, struct file *filp)
1921 if (tracing_disabled)
1924 filp->private_data = inode->i_private;
1928 static int tracing_release(struct inode *inode, struct file *file)
1930 struct seq_file *m = (struct seq_file *)file->private_data;
1931 struct trace_iterator *iter;
1934 if (!(file->f_mode & FMODE_READ))
1939 mutex_lock(&trace_types_lock);
1940 for_each_tracing_cpu(cpu) {
1941 if (iter->buffer_iter[cpu])
1942 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1945 if (iter->trace && iter->trace->close)
1946 iter->trace->close(iter);
1948 /* reenable tracing if it was previously enabled */
1950 mutex_unlock(&trace_types_lock);
1952 seq_release(inode, file);
1953 mutex_destroy(&iter->mutex);
1959 static int tracing_open(struct inode *inode, struct file *file)
1961 struct trace_iterator *iter;
1964 /* If this file was open for write, then erase contents */
1965 if ((file->f_mode & FMODE_WRITE) &&
1966 !(file->f_flags & O_APPEND)) {
1967 long cpu = (long) inode->i_private;
1969 if (cpu == TRACE_PIPE_ALL_CPU)
1970 tracing_reset_online_cpus(&global_trace);
1972 tracing_reset(&global_trace, cpu);
1975 if (file->f_mode & FMODE_READ) {
1976 iter = __tracing_open(inode, file);
1978 ret = PTR_ERR(iter);
1979 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
1980 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1986 t_next(struct seq_file *m, void *v, loff_t *pos)
1988 struct tracer *t = m->private;
2000 static void *t_start(struct seq_file *m, loff_t *pos)
2002 struct tracer *t = m->private;
2005 mutex_lock(&trace_types_lock);
2006 for (; t && l < *pos; t = t_next(m, t, &l))
2012 static void t_stop(struct seq_file *m, void *p)
2014 mutex_unlock(&trace_types_lock);
2017 static int t_show(struct seq_file *m, void *v)
2019 struct tracer *t = v;
2024 seq_printf(m, "%s", t->name);
2033 static struct seq_operations show_traces_seq_ops = {
2040 static int show_traces_open(struct inode *inode, struct file *file)
2044 if (tracing_disabled)
2047 ret = seq_open(file, &show_traces_seq_ops);
2049 struct seq_file *m = file->private_data;
2050 m->private = trace_types;
2057 tracing_write_stub(struct file *filp, const char __user *ubuf,
2058 size_t count, loff_t *ppos)
2063 static const struct file_operations tracing_fops = {
2064 .open = tracing_open,
2066 .write = tracing_write_stub,
2067 .llseek = seq_lseek,
2068 .release = tracing_release,
2071 static const struct file_operations show_traces_fops = {
2072 .open = show_traces_open,
2074 .release = seq_release,
2078 * Only trace on a CPU if the bitmask is set:
2080 static cpumask_var_t tracing_cpumask;
2083 * The tracer itself will not take this lock, but still we want
2084 * to provide a consistent cpumask to user-space:
2086 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2089 * Temporary storage for the character representation of the
2090 * CPU bitmask (and one more byte for the newline):
2092 static char mask_str[NR_CPUS + 1];
2095 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2096 size_t count, loff_t *ppos)
2100 mutex_lock(&tracing_cpumask_update_lock);
2102 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2103 if (count - len < 2) {
2107 len += sprintf(mask_str + len, "\n");
2108 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2111 mutex_unlock(&tracing_cpumask_update_lock);
2117 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2118 size_t count, loff_t *ppos)
2121 cpumask_var_t tracing_cpumask_new;
2123 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2126 mutex_lock(&tracing_cpumask_update_lock);
2127 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2131 local_irq_disable();
2132 __raw_spin_lock(&ftrace_max_lock);
2133 for_each_tracing_cpu(cpu) {
2135 * Increase/decrease the disabled counter if we are
2136 * about to flip a bit in the cpumask:
2138 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2139 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2140 atomic_inc(&global_trace.data[cpu]->disabled);
2142 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2143 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2144 atomic_dec(&global_trace.data[cpu]->disabled);
2147 __raw_spin_unlock(&ftrace_max_lock);
2150 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2152 mutex_unlock(&tracing_cpumask_update_lock);
2153 free_cpumask_var(tracing_cpumask_new);
2158 mutex_unlock(&tracing_cpumask_update_lock);
2159 free_cpumask_var(tracing_cpumask);
2164 static const struct file_operations tracing_cpumask_fops = {
2165 .open = tracing_open_generic,
2166 .read = tracing_cpumask_read,
2167 .write = tracing_cpumask_write,
2171 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2172 size_t cnt, loff_t *ppos)
2174 struct tracer_opt *trace_opts;
2182 /* calculate max size */
2183 for (i = 0; trace_options[i]; i++) {
2184 len += strlen(trace_options[i]);
2185 len += 3; /* "no" and newline */
2188 mutex_lock(&trace_types_lock);
2189 tracer_flags = current_trace->flags->val;
2190 trace_opts = current_trace->flags->opts;
2193 * Increase the size with names of options specific
2194 * of the current tracer.
2196 for (i = 0; trace_opts[i].name; i++) {
2197 len += strlen(trace_opts[i].name);
2198 len += 3; /* "no" and newline */
2201 /* +2 for \n and \0 */
2202 buf = kmalloc(len + 2, GFP_KERNEL);
2204 mutex_unlock(&trace_types_lock);
2208 for (i = 0; trace_options[i]; i++) {
2209 if (trace_flags & (1 << i))
2210 r += sprintf(buf + r, "%s\n", trace_options[i]);
2212 r += sprintf(buf + r, "no%s\n", trace_options[i]);
2215 for (i = 0; trace_opts[i].name; i++) {
2216 if (tracer_flags & trace_opts[i].bit)
2217 r += sprintf(buf + r, "%s\n",
2218 trace_opts[i].name);
2220 r += sprintf(buf + r, "no%s\n",
2221 trace_opts[i].name);
2223 mutex_unlock(&trace_types_lock);
2225 WARN_ON(r >= len + 2);
2227 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2233 /* Try to assign a tracer specific option */
2234 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2236 struct tracer_flags *trace_flags = trace->flags;
2237 struct tracer_opt *opts = NULL;
2241 for (i = 0; trace_flags->opts[i].name; i++) {
2242 opts = &trace_flags->opts[i];
2243 len = strlen(opts->name);
2245 if (strncmp(cmp, opts->name, len) == 0) {
2246 ret = trace->set_flag(trace_flags->val,
2252 if (!trace_flags->opts[i].name)
2255 /* Refused to handle */
2260 trace_flags->val &= ~opts->bit;
2262 trace_flags->val |= opts->bit;
2267 static void set_tracer_flags(unsigned int mask, int enabled)
2269 /* do nothing if flag is already set */
2270 if (!!(trace_flags & mask) == !!enabled)
2274 trace_flags |= mask;
2276 trace_flags &= ~mask;
2278 if (mask == TRACE_ITER_GLOBAL_CLK) {
2282 func = trace_clock_global;
2284 func = trace_clock_local;
2286 mutex_lock(&trace_types_lock);
2287 ring_buffer_set_clock(global_trace.buffer, func);
2290 ring_buffer_set_clock(max_tr.buffer, func);
2291 mutex_unlock(&trace_types_lock);
2296 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2297 size_t cnt, loff_t *ppos)
2305 if (cnt >= sizeof(buf))
2308 if (copy_from_user(&buf, ubuf, cnt))
2313 if (strncmp(buf, "no", 2) == 0) {
2318 for (i = 0; trace_options[i]; i++) {
2319 int len = strlen(trace_options[i]);
2321 if (strncmp(cmp, trace_options[i], len) == 0) {
2322 set_tracer_flags(1 << i, !neg);
2327 /* If no option could be set, test the specific tracer options */
2328 if (!trace_options[i]) {
2329 mutex_lock(&trace_types_lock);
2330 ret = set_tracer_option(current_trace, cmp, neg);
2331 mutex_unlock(&trace_types_lock);
2341 static const struct file_operations tracing_iter_fops = {
2342 .open = tracing_open_generic,
2343 .read = tracing_trace_options_read,
2344 .write = tracing_trace_options_write,
2347 static const char readme_msg[] =
2348 "tracing mini-HOWTO:\n\n"
2350 "# mount -t debugfs nodev /debug\n\n"
2351 "# cat /debug/tracing/available_tracers\n"
2352 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2353 "# cat /debug/tracing/current_tracer\n"
2355 "# echo sched_switch > /debug/tracing/current_tracer\n"
2356 "# cat /debug/tracing/current_tracer\n"
2358 "# cat /debug/tracing/trace_options\n"
2359 "noprint-parent nosym-offset nosym-addr noverbose\n"
2360 "# echo print-parent > /debug/tracing/trace_options\n"
2361 "# echo 1 > /debug/tracing/tracing_enabled\n"
2362 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2363 "echo 0 > /debug/tracing/tracing_enabled\n"
2367 tracing_readme_read(struct file *filp, char __user *ubuf,
2368 size_t cnt, loff_t *ppos)
2370 return simple_read_from_buffer(ubuf, cnt, ppos,
2371 readme_msg, strlen(readme_msg));
2374 static const struct file_operations tracing_readme_fops = {
2375 .open = tracing_open_generic,
2376 .read = tracing_readme_read,
2380 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2381 size_t cnt, loff_t *ppos)
2386 r = sprintf(buf, "%u\n", tracer_enabled);
2387 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2391 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2392 size_t cnt, loff_t *ppos)
2394 struct trace_array *tr = filp->private_data;
2399 if (cnt >= sizeof(buf))
2402 if (copy_from_user(&buf, ubuf, cnt))
2407 ret = strict_strtoul(buf, 10, &val);
2413 mutex_lock(&trace_types_lock);
2414 if (tracer_enabled ^ val) {
2417 if (current_trace->start)
2418 current_trace->start(tr);
2423 if (current_trace->stop)
2424 current_trace->stop(tr);
2427 mutex_unlock(&trace_types_lock);
2435 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2436 size_t cnt, loff_t *ppos)
2438 char buf[max_tracer_type_len+2];
2441 mutex_lock(&trace_types_lock);
2443 r = sprintf(buf, "%s\n", current_trace->name);
2445 r = sprintf(buf, "\n");
2446 mutex_unlock(&trace_types_lock);
2448 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2451 int tracer_init(struct tracer *t, struct trace_array *tr)
2453 tracing_reset_online_cpus(tr);
2457 static int tracing_resize_ring_buffer(unsigned long size)
2462 * If kernel or user changes the size of the ring buffer
2463 * we use the size that was given, and we can forget about
2464 * expanding it later.
2466 ring_buffer_expanded = 1;
2468 ret = ring_buffer_resize(global_trace.buffer, size);
2472 ret = ring_buffer_resize(max_tr.buffer, size);
2476 r = ring_buffer_resize(global_trace.buffer,
2477 global_trace.entries);
2480 * AARGH! We are left with different
2481 * size max buffer!!!!
2482 * The max buffer is our "snapshot" buffer.
2483 * When a tracer needs a snapshot (one of the
2484 * latency tracers), it swaps the max buffer
2485 * with the saved snap shot. We succeeded to
2486 * update the size of the main buffer, but failed to
2487 * update the size of the max buffer. But when we tried
2488 * to reset the main buffer to the original size, we
2489 * failed there too. This is very unlikely to
2490 * happen, but if it does, warn and kill all
2494 tracing_disabled = 1;
2499 global_trace.entries = size;
2505 * tracing_update_buffers - used by tracing facility to expand ring buffers
2507 * To save on memory when the tracing is never used on a system with it
2508 * configured in. The ring buffers are set to a minimum size. But once
2509 * a user starts to use the tracing facility, then they need to grow
2510 * to their default size.
2512 * This function is to be called when a tracer is about to be used.
2514 int tracing_update_buffers(void)
2518 mutex_lock(&trace_types_lock);
2519 if (!ring_buffer_expanded)
2520 ret = tracing_resize_ring_buffer(trace_buf_size);
2521 mutex_unlock(&trace_types_lock);
2526 struct trace_option_dentry;
2528 static struct trace_option_dentry *
2529 create_trace_option_files(struct tracer *tracer);
2532 destroy_trace_option_files(struct trace_option_dentry *topts);
2534 static int tracing_set_tracer(const char *buf)
2536 static struct trace_option_dentry *topts;
2537 struct trace_array *tr = &global_trace;
2541 mutex_lock(&trace_types_lock);
2543 if (!ring_buffer_expanded) {
2544 ret = tracing_resize_ring_buffer(trace_buf_size);
2550 for (t = trace_types; t; t = t->next) {
2551 if (strcmp(t->name, buf) == 0)
2558 if (t == current_trace)
2561 trace_branch_disable();
2562 if (current_trace && current_trace->reset)
2563 current_trace->reset(tr);
2565 destroy_trace_option_files(topts);
2569 topts = create_trace_option_files(current_trace);
2572 ret = tracer_init(t, tr);
2577 trace_branch_enable(tr);
2579 mutex_unlock(&trace_types_lock);
2585 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2586 size_t cnt, loff_t *ppos)
2588 char buf[max_tracer_type_len+1];
2595 if (cnt > max_tracer_type_len)
2596 cnt = max_tracer_type_len;
2598 if (copy_from_user(&buf, ubuf, cnt))
2603 /* strip ending whitespace. */
2604 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2607 err = tracing_set_tracer(buf);
2617 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2618 size_t cnt, loff_t *ppos)
2620 unsigned long *ptr = filp->private_data;
2624 r = snprintf(buf, sizeof(buf), "%ld\n",
2625 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2626 if (r > sizeof(buf))
2628 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2632 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2633 size_t cnt, loff_t *ppos)
2635 unsigned long *ptr = filp->private_data;
2640 if (cnt >= sizeof(buf))
2643 if (copy_from_user(&buf, ubuf, cnt))
2648 ret = strict_strtoul(buf, 10, &val);
2657 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2659 long cpu_file = (long) inode->i_private;
2660 struct trace_iterator *iter;
2663 if (tracing_disabled)
2666 mutex_lock(&trace_types_lock);
2668 /* We only allow one reader per cpu */
2669 if (cpu_file == TRACE_PIPE_ALL_CPU) {
2670 if (!cpumask_empty(tracing_reader_cpumask)) {
2674 cpumask_setall(tracing_reader_cpumask);
2676 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2677 cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2684 /* create a buffer to store the information to pass to userspace */
2685 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2692 * We make a copy of the current tracer to avoid concurrent
2693 * changes on it while we are reading.
2695 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2701 *iter->trace = *current_trace;
2703 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2708 /* trace pipe does not show start of buffer */
2709 cpumask_setall(iter->started);
2711 iter->cpu_file = cpu_file;
2712 iter->tr = &global_trace;
2713 mutex_init(&iter->mutex);
2714 filp->private_data = iter;
2716 if (iter->trace->pipe_open)
2717 iter->trace->pipe_open(iter);
2720 mutex_unlock(&trace_types_lock);
2726 mutex_unlock(&trace_types_lock);
2730 static int tracing_release_pipe(struct inode *inode, struct file *file)
2732 struct trace_iterator *iter = file->private_data;
2734 mutex_lock(&trace_types_lock);
2736 if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2737 cpumask_clear(tracing_reader_cpumask);
2739 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2741 mutex_unlock(&trace_types_lock);
2743 free_cpumask_var(iter->started);
2744 mutex_destroy(&iter->mutex);
2752 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2754 struct trace_iterator *iter = filp->private_data;
2756 if (trace_flags & TRACE_ITER_BLOCK) {
2758 * Always select as readable when in blocking mode
2760 return POLLIN | POLLRDNORM;
2762 if (!trace_empty(iter))
2763 return POLLIN | POLLRDNORM;
2764 poll_wait(filp, &trace_wait, poll_table);
2765 if (!trace_empty(iter))
2766 return POLLIN | POLLRDNORM;
2773 void default_wait_pipe(struct trace_iterator *iter)
2777 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2779 if (trace_empty(iter))
2782 finish_wait(&trace_wait, &wait);
2786 * This is a make-shift waitqueue.
2787 * A tracer might use this callback on some rare cases:
2789 * 1) the current tracer might hold the runqueue lock when it wakes up
2790 * a reader, hence a deadlock (sched, function, and function graph tracers)
2791 * 2) the function tracers, trace all functions, we don't want
2792 * the overhead of calling wake_up and friends
2793 * (and tracing them too)
2795 * Anyway, this is really very primitive wakeup.
2797 void poll_wait_pipe(struct trace_iterator *iter)
2799 set_current_state(TASK_INTERRUPTIBLE);
2800 /* sleep for 100 msecs, and try again. */
2801 schedule_timeout(HZ / 10);
2804 /* Must be called with trace_types_lock mutex held. */
2805 static int tracing_wait_pipe(struct file *filp)
2807 struct trace_iterator *iter = filp->private_data;
2809 while (trace_empty(iter)) {
2811 if ((filp->f_flags & O_NONBLOCK)) {
2815 mutex_unlock(&iter->mutex);
2817 iter->trace->wait_pipe(iter);
2819 mutex_lock(&iter->mutex);
2821 if (signal_pending(current))
2825 * We block until we read something and tracing is disabled.
2826 * We still block if tracing is disabled, but we have never
2827 * read anything. This allows a user to cat this file, and
2828 * then enable tracing. But after we have read something,
2829 * we give an EOF when tracing is again disabled.
2831 * iter->pos will be 0 if we haven't read anything.
2833 if (!tracer_enabled && iter->pos)
2844 tracing_read_pipe(struct file *filp, char __user *ubuf,
2845 size_t cnt, loff_t *ppos)
2847 struct trace_iterator *iter = filp->private_data;
2848 static struct tracer *old_tracer;
2851 /* return any leftover data */
2852 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2856 trace_seq_init(&iter->seq);
2858 /* copy the tracer to avoid using a global lock all around */
2859 mutex_lock(&trace_types_lock);
2860 if (unlikely(old_tracer != current_trace && current_trace)) {
2861 old_tracer = current_trace;
2862 *iter->trace = *current_trace;
2864 mutex_unlock(&trace_types_lock);
2867 * Avoid more than one consumer on a single file descriptor
2868 * This is just a matter of traces coherency, the ring buffer itself
2871 mutex_lock(&iter->mutex);
2872 if (iter->trace->read) {
2873 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2879 sret = tracing_wait_pipe(filp);
2883 /* stop when tracing is finished */
2884 if (trace_empty(iter)) {
2889 if (cnt >= PAGE_SIZE)
2890 cnt = PAGE_SIZE - 1;
2892 /* reset all but tr, trace, and overruns */
2893 memset(&iter->seq, 0,
2894 sizeof(struct trace_iterator) -
2895 offsetof(struct trace_iterator, seq));
2898 while (find_next_entry_inc(iter) != NULL) {
2899 enum print_line_t ret;
2900 int len = iter->seq.len;
2902 ret = print_trace_line(iter);
2903 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2904 /* don't print partial lines */
2905 iter->seq.len = len;
2908 if (ret != TRACE_TYPE_NO_CONSUME)
2909 trace_consume(iter);
2911 if (iter->seq.len >= cnt)
2915 /* Now copy what we have to the user */
2916 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2917 if (iter->seq.readpos >= iter->seq.len)
2918 trace_seq_init(&iter->seq);
2921 * If there was nothing to send to user, inspite of consuming trace
2922 * entries, go back to wait for more entries.
2928 mutex_unlock(&iter->mutex);
2933 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
2934 struct pipe_buffer *buf)
2936 __free_page(buf->page);
2939 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
2942 __free_page(spd->pages[idx]);
2945 static struct pipe_buf_operations tracing_pipe_buf_ops = {
2947 .map = generic_pipe_buf_map,
2948 .unmap = generic_pipe_buf_unmap,
2949 .confirm = generic_pipe_buf_confirm,
2950 .release = tracing_pipe_buf_release,
2951 .steal = generic_pipe_buf_steal,
2952 .get = generic_pipe_buf_get,
2956 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
2961 /* Seq buffer is page-sized, exactly what we need. */
2963 count = iter->seq.len;
2964 ret = print_trace_line(iter);
2965 count = iter->seq.len - count;
2968 iter->seq.len -= count;
2971 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2972 iter->seq.len -= count;
2976 trace_consume(iter);
2978 if (!find_next_entry_inc(iter)) {
2988 static ssize_t tracing_splice_read_pipe(struct file *filp,
2990 struct pipe_inode_info *pipe,
2994 struct page *pages[PIPE_BUFFERS];
2995 struct partial_page partial[PIPE_BUFFERS];
2996 struct trace_iterator *iter = filp->private_data;
2997 struct splice_pipe_desc spd = {
3000 .nr_pages = 0, /* This gets updated below. */
3002 .ops = &tracing_pipe_buf_ops,
3003 .spd_release = tracing_spd_release_pipe,
3005 static struct tracer *old_tracer;
3010 /* copy the tracer to avoid using a global lock all around */
3011 mutex_lock(&trace_types_lock);
3012 if (unlikely(old_tracer != current_trace && current_trace)) {
3013 old_tracer = current_trace;
3014 *iter->trace = *current_trace;
3016 mutex_unlock(&trace_types_lock);
3018 mutex_lock(&iter->mutex);
3020 if (iter->trace->splice_read) {
3021 ret = iter->trace->splice_read(iter, filp,
3022 ppos, pipe, len, flags);
3027 ret = tracing_wait_pipe(filp);
3031 if (!iter->ent && !find_next_entry_inc(iter)) {
3036 /* Fill as many pages as possible. */
3037 for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3038 pages[i] = alloc_page(GFP_KERNEL);
3042 rem = tracing_fill_pipe_page(rem, iter);
3044 /* Copy the data into the page, so we can start over. */
3045 ret = trace_seq_to_buffer(&iter->seq,
3046 page_address(pages[i]),
3049 __free_page(pages[i]);
3052 partial[i].offset = 0;
3053 partial[i].len = iter->seq.len;
3055 trace_seq_init(&iter->seq);
3058 mutex_unlock(&iter->mutex);
3062 return splice_to_pipe(pipe, &spd);
3065 mutex_unlock(&iter->mutex);
3071 tracing_entries_read(struct file *filp, char __user *ubuf,
3072 size_t cnt, loff_t *ppos)
3074 struct trace_array *tr = filp->private_data;
3078 mutex_lock(&trace_types_lock);
3079 if (!ring_buffer_expanded)
3080 r = sprintf(buf, "%lu (expanded: %lu)\n",
3082 trace_buf_size >> 10);
3084 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3085 mutex_unlock(&trace_types_lock);
3087 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3091 tracing_entries_write(struct file *filp, const char __user *ubuf,
3092 size_t cnt, loff_t *ppos)
3098 if (cnt >= sizeof(buf))
3101 if (copy_from_user(&buf, ubuf, cnt))
3106 ret = strict_strtoul(buf, 10, &val);
3110 /* must have at least 1 entry */
3114 mutex_lock(&trace_types_lock);
3118 /* disable all cpu buffers */
3119 for_each_tracing_cpu(cpu) {
3120 if (global_trace.data[cpu])
3121 atomic_inc(&global_trace.data[cpu]->disabled);
3122 if (max_tr.data[cpu])
3123 atomic_inc(&max_tr.data[cpu]->disabled);
3126 /* value is in KB */
3129 if (val != global_trace.entries) {
3130 ret = tracing_resize_ring_buffer(val);
3139 /* If check pages failed, return ENOMEM */
3140 if (tracing_disabled)
3143 for_each_tracing_cpu(cpu) {
3144 if (global_trace.data[cpu])
3145 atomic_dec(&global_trace.data[cpu]->disabled);
3146 if (max_tr.data[cpu])
3147 atomic_dec(&max_tr.data[cpu]->disabled);
3151 max_tr.entries = global_trace.entries;
3152 mutex_unlock(&trace_types_lock);
3157 static int mark_printk(const char *fmt, ...)
3161 va_start(args, fmt);
3162 ret = trace_vprintk(0, fmt, args);
3168 tracing_mark_write(struct file *filp, const char __user *ubuf,
3169 size_t cnt, loff_t *fpos)
3174 if (tracing_disabled)
3177 if (cnt > TRACE_BUF_SIZE)
3178 cnt = TRACE_BUF_SIZE;
3180 buf = kmalloc(cnt + 1, GFP_KERNEL);
3184 if (copy_from_user(buf, ubuf, cnt)) {
3189 /* Cut from the first nil or newline. */
3191 end = strchr(buf, '\n');
3195 cnt = mark_printk("%s\n", buf);
3202 static const struct file_operations tracing_max_lat_fops = {
3203 .open = tracing_open_generic,
3204 .read = tracing_max_lat_read,
3205 .write = tracing_max_lat_write,
3208 static const struct file_operations tracing_ctrl_fops = {
3209 .open = tracing_open_generic,
3210 .read = tracing_ctrl_read,
3211 .write = tracing_ctrl_write,
3214 static const struct file_operations set_tracer_fops = {
3215 .open = tracing_open_generic,
3216 .read = tracing_set_trace_read,
3217 .write = tracing_set_trace_write,
3220 static const struct file_operations tracing_pipe_fops = {
3221 .open = tracing_open_pipe,
3222 .poll = tracing_poll_pipe,
3223 .read = tracing_read_pipe,
3224 .splice_read = tracing_splice_read_pipe,
3225 .release = tracing_release_pipe,
3228 static const struct file_operations tracing_entries_fops = {
3229 .open = tracing_open_generic,
3230 .read = tracing_entries_read,
3231 .write = tracing_entries_write,
3234 static const struct file_operations tracing_mark_fops = {
3235 .open = tracing_open_generic,
3236 .write = tracing_mark_write,
3239 struct ftrace_buffer_info {
3240 struct trace_array *tr;
3246 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3248 int cpu = (int)(long)inode->i_private;
3249 struct ftrace_buffer_info *info;
3251 if (tracing_disabled)
3254 info = kzalloc(sizeof(*info), GFP_KERNEL);
3258 info->tr = &global_trace;
3260 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3261 /* Force reading ring buffer for first read */
3262 info->read = (unsigned int)-1;
3266 filp->private_data = info;
3276 tracing_buffers_read(struct file *filp, char __user *ubuf,
3277 size_t count, loff_t *ppos)
3279 struct ftrace_buffer_info *info = filp->private_data;
3287 /* Do we have previous read data to read? */
3288 if (info->read < PAGE_SIZE)
3293 ret = ring_buffer_read_page(info->tr->buffer,
3300 pos = ring_buffer_page_len(info->spare);
3302 if (pos < PAGE_SIZE)
3303 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3306 size = PAGE_SIZE - info->read;
3310 ret = copy_to_user(ubuf, info->spare + info->read, size);
3321 static int tracing_buffers_release(struct inode *inode, struct file *file)
3323 struct ftrace_buffer_info *info = file->private_data;
3325 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3332 struct ring_buffer *buffer;
3337 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3338 struct pipe_buffer *buf)
3340 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3345 ring_buffer_free_read_page(ref->buffer, ref->page);
3350 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3351 struct pipe_buffer *buf)
3356 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3357 struct pipe_buffer *buf)
3359 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3364 /* Pipe buffer operations for a buffer. */
3365 static struct pipe_buf_operations buffer_pipe_buf_ops = {
3367 .map = generic_pipe_buf_map,
3368 .unmap = generic_pipe_buf_unmap,
3369 .confirm = generic_pipe_buf_confirm,
3370 .release = buffer_pipe_buf_release,
3371 .steal = buffer_pipe_buf_steal,
3372 .get = buffer_pipe_buf_get,
3376 * Callback from splice_to_pipe(), if we need to release some pages
3377 * at the end of the spd in case we error'ed out in filling the pipe.
3379 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3381 struct buffer_ref *ref =
3382 (struct buffer_ref *)spd->partial[i].private;
3387 ring_buffer_free_read_page(ref->buffer, ref->page);
3389 spd->partial[i].private = 0;
3393 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3394 struct pipe_inode_info *pipe, size_t len,
3397 struct ftrace_buffer_info *info = file->private_data;
3398 struct partial_page partial[PIPE_BUFFERS];
3399 struct page *pages[PIPE_BUFFERS];
3400 struct splice_pipe_desc spd = {
3404 .ops = &buffer_pipe_buf_ops,
3405 .spd_release = buffer_spd_release,
3407 struct buffer_ref *ref;
3412 * We can't seek on a buffer input
3414 if (unlikely(*ppos))
3418 for (i = 0; i < PIPE_BUFFERS && len; i++, len -= size) {
3422 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3426 ref->buffer = info->tr->buffer;
3427 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3433 r = ring_buffer_read_page(ref->buffer, &ref->page,
3436 ring_buffer_free_read_page(ref->buffer,
3443 * zero out any left over data, this is going to
3446 size = ring_buffer_page_len(ref->page);
3447 if (size < PAGE_SIZE)
3448 memset(ref->page + size, 0, PAGE_SIZE - size);
3450 page = virt_to_page(ref->page);
3452 spd.pages[i] = page;
3453 spd.partial[i].len = PAGE_SIZE;
3454 spd.partial[i].offset = 0;
3455 spd.partial[i].private = (unsigned long)ref;
3461 /* did we read anything? */
3462 if (!spd.nr_pages) {
3463 if (flags & SPLICE_F_NONBLOCK)
3471 ret = splice_to_pipe(pipe, &spd);
3476 static const struct file_operations tracing_buffers_fops = {
3477 .open = tracing_buffers_open,
3478 .read = tracing_buffers_read,
3479 .release = tracing_buffers_release,
3480 .splice_read = tracing_buffers_splice_read,
3481 .llseek = no_llseek,
3484 #ifdef CONFIG_DYNAMIC_FTRACE
3486 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3492 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3493 size_t cnt, loff_t *ppos)
3495 static char ftrace_dyn_info_buffer[1024];
3496 static DEFINE_MUTEX(dyn_info_mutex);
3497 unsigned long *p = filp->private_data;
3498 char *buf = ftrace_dyn_info_buffer;
3499 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3502 mutex_lock(&dyn_info_mutex);
3503 r = sprintf(buf, "%ld ", *p);
3505 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3508 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3510 mutex_unlock(&dyn_info_mutex);
3515 static const struct file_operations tracing_dyn_info_fops = {
3516 .open = tracing_open_generic,
3517 .read = tracing_read_dyn_info,
3521 static struct dentry *d_tracer;
3523 struct dentry *tracing_init_dentry(void)
3530 if (!debugfs_initialized())
3533 d_tracer = debugfs_create_dir("tracing", NULL);
3535 if (!d_tracer && !once) {
3537 pr_warning("Could not create debugfs directory 'tracing'\n");
3544 static struct dentry *d_percpu;
3546 struct dentry *tracing_dentry_percpu(void)
3549 struct dentry *d_tracer;
3554 d_tracer = tracing_init_dentry();
3559 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3561 if (!d_percpu && !once) {
3563 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3570 static void tracing_init_debugfs_percpu(long cpu)
3572 struct dentry *d_percpu = tracing_dentry_percpu();
3573 struct dentry *entry, *d_cpu;
3574 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3577 if (cpu > 999 || cpu < 0)
3580 sprintf(cpu_dir, "cpu%ld", cpu);
3581 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3583 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3587 /* per cpu trace_pipe */
3588 entry = debugfs_create_file("trace_pipe", 0444, d_cpu,
3589 (void *) cpu, &tracing_pipe_fops);
3591 pr_warning("Could not create debugfs 'trace_pipe' entry\n");
3594 entry = debugfs_create_file("trace", 0644, d_cpu,
3595 (void *) cpu, &tracing_fops);
3597 pr_warning("Could not create debugfs 'trace' entry\n");
3599 entry = debugfs_create_file("trace_pipe_raw", 0444, d_cpu,
3600 (void *) cpu, &tracing_buffers_fops);
3602 pr_warning("Could not create debugfs 'trace_pipe_raw' entry\n");
3605 #ifdef CONFIG_FTRACE_SELFTEST
3606 /* Let selftest have access to static functions in this file */
3607 #include "trace_selftest.c"
3610 struct trace_option_dentry {
3611 struct tracer_opt *opt;
3612 struct tracer_flags *flags;
3613 struct dentry *entry;
3617 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3620 struct trace_option_dentry *topt = filp->private_data;
3623 if (topt->flags->val & topt->opt->bit)
3628 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3632 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3635 struct trace_option_dentry *topt = filp->private_data;
3640 if (cnt >= sizeof(buf))
3643 if (copy_from_user(&buf, ubuf, cnt))
3648 ret = strict_strtoul(buf, 10, &val);
3655 /* do nothing if already cleared */
3656 if (!(topt->flags->val & topt->opt->bit))
3659 mutex_lock(&trace_types_lock);
3660 if (current_trace->set_flag)
3661 ret = current_trace->set_flag(topt->flags->val,
3663 mutex_unlock(&trace_types_lock);
3666 topt->flags->val &= ~topt->opt->bit;
3669 /* do nothing if already set */
3670 if (topt->flags->val & topt->opt->bit)
3673 mutex_lock(&trace_types_lock);
3674 if (current_trace->set_flag)
3675 ret = current_trace->set_flag(topt->flags->val,
3677 mutex_unlock(&trace_types_lock);
3680 topt->flags->val |= topt->opt->bit;
3693 static const struct file_operations trace_options_fops = {
3694 .open = tracing_open_generic,
3695 .read = trace_options_read,
3696 .write = trace_options_write,
3700 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
3703 long index = (long)filp->private_data;
3706 if (trace_flags & (1 << index))
3711 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3715 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
3718 long index = (long)filp->private_data;
3723 if (cnt >= sizeof(buf))
3726 if (copy_from_user(&buf, ubuf, cnt))
3731 ret = strict_strtoul(buf, 10, &val);
3737 trace_flags &= ~(1 << index);
3740 trace_flags |= 1 << index;
3752 static const struct file_operations trace_options_core_fops = {
3753 .open = tracing_open_generic,
3754 .read = trace_options_core_read,
3755 .write = trace_options_core_write,
3758 static struct dentry *trace_options_init_dentry(void)
3760 struct dentry *d_tracer;
3761 static struct dentry *t_options;
3766 d_tracer = tracing_init_dentry();
3770 t_options = debugfs_create_dir("options", d_tracer);
3772 pr_warning("Could not create debugfs directory 'options'\n");
3780 create_trace_option_file(struct trace_option_dentry *topt,
3781 struct tracer_flags *flags,
3782 struct tracer_opt *opt)
3784 struct dentry *t_options;
3785 struct dentry *entry;
3787 t_options = trace_options_init_dentry();
3791 topt->flags = flags;
3794 entry = debugfs_create_file(opt->name, 0644, t_options, topt,
3795 &trace_options_fops);
3797 topt->entry = entry;
3801 static struct trace_option_dentry *
3802 create_trace_option_files(struct tracer *tracer)
3804 struct trace_option_dentry *topts;
3805 struct tracer_flags *flags;
3806 struct tracer_opt *opts;
3812 flags = tracer->flags;
3814 if (!flags || !flags->opts)
3819 for (cnt = 0; opts[cnt].name; cnt++)
3822 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
3826 for (cnt = 0; opts[cnt].name; cnt++)
3827 create_trace_option_file(&topts[cnt], flags,
3834 destroy_trace_option_files(struct trace_option_dentry *topts)
3841 for (cnt = 0; topts[cnt].opt; cnt++) {
3842 if (topts[cnt].entry)
3843 debugfs_remove(topts[cnt].entry);
3849 static struct dentry *
3850 create_trace_option_core_file(const char *option, long index)
3852 struct dentry *t_options;
3853 struct dentry *entry;
3855 t_options = trace_options_init_dentry();
3859 entry = debugfs_create_file(option, 0644, t_options, (void *)index,
3860 &trace_options_core_fops);
3865 static __init void create_trace_options_dir(void)
3867 struct dentry *t_options;
3868 struct dentry *entry;
3871 t_options = trace_options_init_dentry();
3875 for (i = 0; trace_options[i]; i++) {
3876 entry = create_trace_option_core_file(trace_options[i], i);
3878 pr_warning("Could not create debugfs %s entry\n",
3883 static __init int tracer_init_debugfs(void)
3885 struct dentry *d_tracer;
3886 struct dentry *entry;
3889 d_tracer = tracing_init_dentry();
3891 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3892 &global_trace, &tracing_ctrl_fops);
3894 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3896 entry = debugfs_create_file("trace_options", 0644, d_tracer,
3897 NULL, &tracing_iter_fops);
3899 pr_warning("Could not create debugfs 'trace_options' entry\n");
3901 create_trace_options_dir();
3903 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3904 NULL, &tracing_cpumask_fops);
3906 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3908 entry = debugfs_create_file("trace", 0644, d_tracer,
3909 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
3911 pr_warning("Could not create debugfs 'trace' entry\n");
3913 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3914 &global_trace, &show_traces_fops);
3916 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3918 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3919 &global_trace, &set_tracer_fops);
3921 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3923 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3924 &tracing_max_latency,
3925 &tracing_max_lat_fops);
3927 pr_warning("Could not create debugfs "
3928 "'tracing_max_latency' entry\n");
3930 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3931 &tracing_thresh, &tracing_max_lat_fops);
3933 pr_warning("Could not create debugfs "
3934 "'tracing_thresh' entry\n");
3935 entry = debugfs_create_file("README", 0644, d_tracer,
3936 NULL, &tracing_readme_fops);
3938 pr_warning("Could not create debugfs 'README' entry\n");
3940 entry = debugfs_create_file("trace_pipe", 0444, d_tracer,
3941 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
3943 pr_warning("Could not create debugfs "
3944 "'trace_pipe' entry\n");
3946 entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
3947 &global_trace, &tracing_entries_fops);
3949 pr_warning("Could not create debugfs "
3950 "'buffer_size_kb' entry\n");
3952 entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3953 NULL, &tracing_mark_fops);
3955 pr_warning("Could not create debugfs "
3956 "'trace_marker' entry\n");
3958 #ifdef CONFIG_DYNAMIC_FTRACE
3959 entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3960 &ftrace_update_tot_cnt,
3961 &tracing_dyn_info_fops);
3963 pr_warning("Could not create debugfs "
3964 "'dyn_ftrace_total_info' entry\n");
3966 #ifdef CONFIG_SYSPROF_TRACER
3967 init_tracer_sysprof_debugfs(d_tracer);
3970 for_each_tracing_cpu(cpu)
3971 tracing_init_debugfs_percpu(cpu);
3976 static int trace_panic_handler(struct notifier_block *this,
3977 unsigned long event, void *unused)
3979 if (ftrace_dump_on_oops)
3984 static struct notifier_block trace_panic_notifier = {
3985 .notifier_call = trace_panic_handler,
3987 .priority = 150 /* priority: INT_MAX >= x >= 0 */
3990 static int trace_die_handler(struct notifier_block *self,
3996 if (ftrace_dump_on_oops)
4005 static struct notifier_block trace_die_notifier = {
4006 .notifier_call = trace_die_handler,
4011 * printk is set to max of 1024, we really don't need it that big.
4012 * Nothing should be printing 1000 characters anyway.
4014 #define TRACE_MAX_PRINT 1000
4017 * Define here KERN_TRACE so that we have one place to modify
4018 * it if we decide to change what log level the ftrace dump
4021 #define KERN_TRACE KERN_EMERG
4024 trace_printk_seq(struct trace_seq *s)
4026 /* Probably should print a warning here. */
4030 /* should be zero ended, but we are paranoid. */
4031 s->buffer[s->len] = 0;
4033 printk(KERN_TRACE "%s", s->buffer);
4038 static void __ftrace_dump(bool disable_tracing)
4040 static DEFINE_SPINLOCK(ftrace_dump_lock);
4041 /* use static because iter can be a bit big for the stack */
4042 static struct trace_iterator iter;
4043 unsigned int old_userobj;
4044 static int dump_ran;
4045 unsigned long flags;
4049 spin_lock_irqsave(&ftrace_dump_lock, flags);
4057 if (disable_tracing)
4060 for_each_tracing_cpu(cpu) {
4061 atomic_inc(&global_trace.data[cpu]->disabled);
4064 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4066 /* don't look at user memory in panic mode */
4067 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4069 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4071 /* Simulate the iterator */
4072 iter.tr = &global_trace;
4073 iter.trace = current_trace;
4074 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4077 * We need to stop all tracing on all CPUS to read the
4078 * the next buffer. This is a bit expensive, but is
4079 * not done often. We fill all what we can read,
4080 * and then release the locks again.
4083 while (!trace_empty(&iter)) {
4086 printk(KERN_TRACE "---------------------------------\n");
4090 /* reset all but tr, trace, and overruns */
4091 memset(&iter.seq, 0,
4092 sizeof(struct trace_iterator) -
4093 offsetof(struct trace_iterator, seq));
4094 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4097 if (find_next_entry_inc(&iter) != NULL) {
4098 print_trace_line(&iter);
4099 trace_consume(&iter);
4102 trace_printk_seq(&iter.seq);
4106 printk(KERN_TRACE " (ftrace buffer empty)\n");
4108 printk(KERN_TRACE "---------------------------------\n");
4110 /* Re-enable tracing if requested */
4111 if (!disable_tracing) {
4112 trace_flags |= old_userobj;
4114 for_each_tracing_cpu(cpu) {
4115 atomic_dec(&global_trace.data[cpu]->disabled);
4121 spin_unlock_irqrestore(&ftrace_dump_lock, flags);
4124 /* By default: disable tracing after the dump */
4125 void ftrace_dump(void)
4127 __ftrace_dump(true);
4130 __init static int tracer_alloc_buffers(void)
4132 struct trace_array_cpu *data;
4137 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4140 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4141 goto out_free_buffer_mask;
4143 if (!alloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
4144 goto out_free_tracing_cpumask;
4146 /* To save memory, keep the ring buffer size to its minimum */
4147 if (ring_buffer_expanded)
4148 ring_buf_size = trace_buf_size;
4152 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4153 cpumask_copy(tracing_cpumask, cpu_all_mask);
4154 cpumask_clear(tracing_reader_cpumask);
4156 /* TODO: make the number of buffers hot pluggable with CPUS */
4157 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4158 TRACE_BUFFER_FLAGS);
4159 if (!global_trace.buffer) {
4160 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4162 goto out_free_cpumask;
4164 global_trace.entries = ring_buffer_size(global_trace.buffer);
4167 #ifdef CONFIG_TRACER_MAX_TRACE
4168 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4169 TRACE_BUFFER_FLAGS);
4170 if (!max_tr.buffer) {
4171 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4173 ring_buffer_free(global_trace.buffer);
4174 goto out_free_cpumask;
4176 max_tr.entries = ring_buffer_size(max_tr.buffer);
4177 WARN_ON(max_tr.entries != global_trace.entries);
4180 /* Allocate the first page for all buffers */
4181 for_each_tracing_cpu(i) {
4182 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4183 max_tr.data[i] = &per_cpu(max_data, i);
4186 trace_init_cmdlines();
4188 register_tracer(&nop_trace);
4189 current_trace = &nop_trace;
4190 #ifdef CONFIG_BOOT_TRACER
4191 register_tracer(&boot_tracer);
4193 /* All seems OK, enable tracing */
4194 tracing_disabled = 0;
4196 atomic_notifier_chain_register(&panic_notifier_list,
4197 &trace_panic_notifier);
4199 register_die_notifier(&trace_die_notifier);
4204 free_cpumask_var(tracing_reader_cpumask);
4205 out_free_tracing_cpumask:
4206 free_cpumask_var(tracing_cpumask);
4207 out_free_buffer_mask:
4208 free_cpumask_var(tracing_buffer_mask);
4213 __init static int clear_boot_tracer(void)
4216 * The default tracer at boot buffer is an init section.
4217 * This function is called in lateinit. If we did not
4218 * find the boot tracer, then clear it out, to prevent
4219 * later registration from accessing the buffer that is
4220 * about to be freed.
4222 if (!default_bootup_tracer)
4225 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4226 default_bootup_tracer);
4227 default_bootup_tracer = NULL;
4232 early_initcall(tracer_alloc_buffers);
4233 fs_initcall(tracer_init_debugfs);
4234 late_initcall(clear_boot_tracer);