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/utsrelease.h>
15 #include <linux/kallsyms.h>
16 #include <linux/seq_file.h>
17 #include <linux/notifier.h>
18 #include <linux/debugfs.h>
19 #include <linux/pagemap.h>
20 #include <linux/hardirq.h>
21 #include <linux/linkage.h>
22 #include <linux/uaccess.h>
23 #include <linux/ftrace.h>
24 #include <linux/module.h>
25 #include <linux/percpu.h>
26 #include <linux/kdebug.h>
27 #include <linux/ctype.h>
28 #include <linux/init.h>
29 #include <linux/poll.h>
30 #include <linux/gfp.h>
32 #include <linux/kprobes.h>
33 #include <linux/writeback.h>
35 #include <linux/stacktrace.h>
36 #include <linux/ring_buffer.h>
40 #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
42 unsigned long __read_mostly tracing_max_latency = (cycle_t)ULONG_MAX;
43 unsigned long __read_mostly tracing_thresh;
45 static cpumask_t __read_mostly tracing_buffer_mask;
47 #define for_each_tracing_cpu(cpu) \
48 for_each_cpu_mask(cpu, tracing_buffer_mask)
50 static int tracing_disabled = 1;
53 ns2usecs(cycle_t nsec)
60 cycle_t ftrace_now(int cpu)
62 u64 ts = ring_buffer_time_stamp(cpu);
63 ring_buffer_normalize_time_stamp(cpu, &ts);
68 * The global_trace is the descriptor that holds the tracing
69 * buffers for the live tracing. For each CPU, it contains
70 * a link list of pages that will store trace entries. The
71 * page descriptor of the pages in the memory is used to hold
72 * the link list by linking the lru item in the page descriptor
73 * to each of the pages in the buffer per CPU.
75 * For each active CPU there is a data field that holds the
76 * pages for the buffer for that CPU. Each CPU has the same number
77 * of pages allocated for its buffer.
79 static struct trace_array global_trace;
81 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
84 * The max_tr is used to snapshot the global_trace when a maximum
85 * latency is reached. Some tracers will use this to store a maximum
86 * trace while it continues examining live traces.
88 * The buffers for the max_tr are set up the same as the global_trace.
89 * When a snapshot is taken, the link list of the max_tr is swapped
90 * with the link list of the global_trace and the buffers are reset for
91 * the global_trace so the tracing can continue.
93 static struct trace_array max_tr;
95 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
97 /* tracer_enabled is used to toggle activation of a tracer */
98 static int tracer_enabled = 1;
100 /* function tracing enabled */
101 int ftrace_function_enabled;
104 * trace_buf_size is the size in bytes that is allocated
105 * for a buffer. Note, the number of bytes is always rounded
108 * This number is purposely set to a low number of 16384.
109 * If the dump on oops happens, it will be much appreciated
110 * to not have to wait for all that output. Anyway this can be
111 * boot time and run time configurable.
113 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
115 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
117 /* trace_types holds a link list of available tracers. */
118 static struct tracer *trace_types __read_mostly;
120 /* current_trace points to the tracer that is currently active */
121 static struct tracer *current_trace __read_mostly;
124 * max_tracer_type_len is used to simplify the allocating of
125 * buffers to read userspace tracer names. We keep track of
126 * the longest tracer name registered.
128 static int max_tracer_type_len;
131 * trace_types_lock is used to protect the trace_types list.
132 * This lock is also used to keep user access serialized.
133 * Accesses from userspace will grab this lock while userspace
134 * activities happen inside the kernel.
136 static DEFINE_MUTEX(trace_types_lock);
138 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
139 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
141 /* trace_flags holds iter_ctrl options */
142 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT;
145 * trace_wake_up - wake up tasks waiting for trace input
147 * Simply wakes up any task that is blocked on the trace_wait
148 * queue. These is used with trace_poll for tasks polling the trace.
150 void trace_wake_up(void)
153 * The runqueue_is_locked() can fail, but this is the best we
156 if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
157 wake_up(&trace_wait);
160 static int __init set_buf_size(char *str)
162 unsigned long buf_size;
167 ret = strict_strtoul(str, 0, &buf_size);
168 /* nr_entries can not be zero */
169 if (ret < 0 || buf_size == 0)
171 trace_buf_size = buf_size;
174 __setup("trace_buf_size=", set_buf_size);
176 unsigned long nsecs_to_usecs(unsigned long nsecs)
182 * TRACE_ITER_SYM_MASK masks the options in trace_flags that
183 * control the output of kernel symbols.
185 #define TRACE_ITER_SYM_MASK \
186 (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
188 /* These must match the bit postions in trace_iterator_flags */
189 static const char *trace_options[] = {
205 * ftrace_max_lock is used to protect the swapping of buffers
206 * when taking a max snapshot. The buffers themselves are
207 * protected by per_cpu spinlocks. But the action of the swap
208 * needs its own lock.
210 * This is defined as a raw_spinlock_t in order to help
211 * with performance when lockdep debugging is enabled.
213 static raw_spinlock_t ftrace_max_lock =
214 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
217 * Copy the new maximum trace into the separate maximum-trace
218 * structure. (this way the maximum trace is permanently saved,
219 * for later retrieval via /debugfs/tracing/latency_trace)
222 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
224 struct trace_array_cpu *data = tr->data[cpu];
227 max_tr.time_start = data->preempt_timestamp;
229 data = max_tr.data[cpu];
230 data->saved_latency = tracing_max_latency;
232 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
233 data->pid = tsk->pid;
234 data->uid = tsk->uid;
235 data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
236 data->policy = tsk->policy;
237 data->rt_priority = tsk->rt_priority;
239 /* record this tasks comm */
240 tracing_record_cmdline(current);
244 * trace_seq_printf - sequence printing of trace information
245 * @s: trace sequence descriptor
246 * @fmt: printf format string
248 * The tracer may use either sequence operations or its own
249 * copy to user routines. To simplify formating of a trace
250 * trace_seq_printf is used to store strings into a special
251 * buffer (@s). Then the output may be either used by
252 * the sequencer or pulled into another buffer.
255 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
257 int len = (PAGE_SIZE - 1) - s->len;
265 ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
268 /* If we can't write it all, don't bother writing anything */
278 * trace_seq_puts - trace sequence printing of simple string
279 * @s: trace sequence descriptor
280 * @str: simple string to record
282 * The tracer may use either the sequence operations or its own
283 * copy to user routines. This function records a simple string
284 * into a special buffer (@s) for later retrieval by a sequencer
285 * or other mechanism.
288 trace_seq_puts(struct trace_seq *s, const char *str)
290 int len = strlen(str);
292 if (len > ((PAGE_SIZE - 1) - s->len))
295 memcpy(s->buffer + s->len, str, len);
302 trace_seq_putc(struct trace_seq *s, unsigned char c)
304 if (s->len >= (PAGE_SIZE - 1))
307 s->buffer[s->len++] = c;
313 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
315 if (len > ((PAGE_SIZE - 1) - s->len))
318 memcpy(s->buffer + s->len, mem, len);
325 static const char hex2asc[] = "0123456789abcdef";
328 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
330 unsigned char hex[HEX_CHARS];
331 unsigned char *data = mem;
335 BUG_ON(len >= HEX_CHARS);
338 for (i = 0, j = 0; i < len; i++) {
340 for (i = len-1, j = 0; i >= 0; i--) {
344 hex[j++] = hex2asc[byte & 0x0f];
345 hex[j++] = hex2asc[byte >> 4];
349 return trace_seq_putmem(s, hex, j);
353 trace_seq_reset(struct trace_seq *s)
359 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
364 if (s->len <= s->readpos)
367 len = s->len - s->readpos;
370 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
379 trace_print_seq(struct seq_file *m, struct trace_seq *s)
381 int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
384 seq_puts(m, s->buffer);
390 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
392 * @tsk: the task with the latency
393 * @cpu: The cpu that initiated the trace.
395 * Flip the buffers between the @tr and the max_tr and record information
396 * about which task was the cause of this latency.
399 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
401 struct ring_buffer *buf = tr->buffer;
403 WARN_ON_ONCE(!irqs_disabled());
404 __raw_spin_lock(&ftrace_max_lock);
406 tr->buffer = max_tr.buffer;
409 ring_buffer_reset(tr->buffer);
411 __update_max_tr(tr, tsk, cpu);
412 __raw_spin_unlock(&ftrace_max_lock);
416 * update_max_tr_single - only copy one trace over, and reset the rest
418 * @tsk - task with the latency
419 * @cpu - the cpu of the buffer to copy.
421 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
424 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
428 WARN_ON_ONCE(!irqs_disabled());
429 __raw_spin_lock(&ftrace_max_lock);
431 ring_buffer_reset(max_tr.buffer);
432 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
436 __update_max_tr(tr, tsk, cpu);
437 __raw_spin_unlock(&ftrace_max_lock);
441 * register_tracer - register a tracer with the ftrace system.
442 * @type - the plugin for the tracer
444 * Register a new plugin tracer.
446 int register_tracer(struct tracer *type)
453 pr_info("Tracer must have a name\n");
457 mutex_lock(&trace_types_lock);
458 for (t = trace_types; t; t = t->next) {
459 if (strcmp(type->name, t->name) == 0) {
461 pr_info("Trace %s already registered\n",
468 #ifdef CONFIG_FTRACE_STARTUP_TEST
469 if (type->selftest) {
470 struct tracer *saved_tracer = current_trace;
471 struct trace_array *tr = &global_trace;
472 int saved_ctrl = tr->ctrl;
475 * Run a selftest on this tracer.
476 * Here we reset the trace buffer, and set the current
477 * tracer to be this tracer. The tracer can then run some
478 * internal tracing to verify that everything is in order.
479 * If we fail, we do not register this tracer.
481 for_each_tracing_cpu(i) {
482 tracing_reset(tr, i);
484 current_trace = type;
486 /* the test is responsible for initializing and enabling */
487 pr_info("Testing tracer %s: ", type->name);
488 ret = type->selftest(type, tr);
489 /* the test is responsible for resetting too */
490 current_trace = saved_tracer;
491 tr->ctrl = saved_ctrl;
493 printk(KERN_CONT "FAILED!\n");
496 /* Only reset on passing, to avoid touching corrupted buffers */
497 for_each_tracing_cpu(i) {
498 tracing_reset(tr, i);
500 printk(KERN_CONT "PASSED\n");
504 type->next = trace_types;
506 len = strlen(type->name);
507 if (len > max_tracer_type_len)
508 max_tracer_type_len = len;
511 mutex_unlock(&trace_types_lock);
516 void unregister_tracer(struct tracer *type)
521 mutex_lock(&trace_types_lock);
522 for (t = &trace_types; *t; t = &(*t)->next) {
526 pr_info("Trace %s not registered\n", type->name);
531 if (strlen(type->name) != max_tracer_type_len)
534 max_tracer_type_len = 0;
535 for (t = &trace_types; *t; t = &(*t)->next) {
536 len = strlen((*t)->name);
537 if (len > max_tracer_type_len)
538 max_tracer_type_len = len;
541 mutex_unlock(&trace_types_lock);
544 void tracing_reset(struct trace_array *tr, int cpu)
546 ring_buffer_reset_cpu(tr->buffer, cpu);
549 #define SAVED_CMDLINES 128
550 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
551 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
552 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
553 static int cmdline_idx;
554 static DEFINE_SPINLOCK(trace_cmdline_lock);
556 /* temporary disable recording */
557 atomic_t trace_record_cmdline_disabled __read_mostly;
559 static void trace_init_cmdlines(void)
561 memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
562 memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
566 void trace_stop_cmdline_recording(void);
568 static void trace_save_cmdline(struct task_struct *tsk)
573 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
577 * It's not the end of the world if we don't get
578 * the lock, but we also don't want to spin
579 * nor do we want to disable interrupts,
580 * so if we miss here, then better luck next time.
582 if (!spin_trylock(&trace_cmdline_lock))
585 idx = map_pid_to_cmdline[tsk->pid];
586 if (idx >= SAVED_CMDLINES) {
587 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
589 map = map_cmdline_to_pid[idx];
590 if (map <= PID_MAX_DEFAULT)
591 map_pid_to_cmdline[map] = (unsigned)-1;
593 map_pid_to_cmdline[tsk->pid] = idx;
598 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
600 spin_unlock(&trace_cmdline_lock);
603 static char *trace_find_cmdline(int pid)
605 char *cmdline = "<...>";
611 if (pid > PID_MAX_DEFAULT)
614 map = map_pid_to_cmdline[pid];
615 if (map >= SAVED_CMDLINES)
618 cmdline = saved_cmdlines[map];
624 void tracing_record_cmdline(struct task_struct *tsk)
626 if (atomic_read(&trace_record_cmdline_disabled))
629 trace_save_cmdline(tsk);
633 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
635 struct task_struct *tsk = current;
638 pc = preempt_count();
640 entry->field.preempt_count = pc & 0xff;
641 entry->field.pid = (tsk) ? tsk->pid : 0;
643 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
644 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
645 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
646 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
650 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
651 unsigned long ip, unsigned long parent_ip, unsigned long flags)
653 struct ring_buffer_event *event;
654 struct trace_entry *entry;
655 unsigned long irq_flags;
657 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
661 entry = ring_buffer_event_data(event);
662 tracing_generic_entry_update(entry, flags);
663 entry->type = TRACE_FN;
664 entry->field.fn.ip = ip;
665 entry->field.fn.parent_ip = parent_ip;
666 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
670 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
671 unsigned long ip, unsigned long parent_ip, unsigned long flags)
673 if (likely(!atomic_read(&data->disabled)))
674 trace_function(tr, data, ip, parent_ip, flags);
677 void __trace_stack(struct trace_array *tr,
678 struct trace_array_cpu *data,
682 struct ring_buffer_event *event;
683 struct trace_entry *entry;
684 struct stack_trace trace;
685 unsigned long irq_flags;
687 if (!(trace_flags & TRACE_ITER_STACKTRACE))
690 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
694 entry = ring_buffer_event_data(event);
695 tracing_generic_entry_update(entry, flags);
696 entry->type = TRACE_STACK;
698 memset(&entry->field.stack, 0, sizeof(entry->field.stack));
700 trace.nr_entries = 0;
701 trace.max_entries = FTRACE_STACK_ENTRIES;
703 trace.entries = entry->field.stack.caller;
705 save_stack_trace(&trace);
706 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
710 __trace_special(void *__tr, void *__data,
711 unsigned long arg1, unsigned long arg2, unsigned long arg3)
713 struct ring_buffer_event *event;
714 struct trace_array_cpu *data = __data;
715 struct trace_array *tr = __tr;
716 struct trace_entry *entry;
717 unsigned long irq_flags;
719 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
723 entry = ring_buffer_event_data(event);
724 tracing_generic_entry_update(entry, 0);
725 entry->type = TRACE_SPECIAL;
726 entry->field.special.arg1 = arg1;
727 entry->field.special.arg2 = arg2;
728 entry->field.special.arg3 = arg3;
729 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
730 __trace_stack(tr, data, irq_flags, 4);
736 tracing_sched_switch_trace(struct trace_array *tr,
737 struct trace_array_cpu *data,
738 struct task_struct *prev,
739 struct task_struct *next,
742 struct ring_buffer_event *event;
743 struct trace_entry *entry;
744 unsigned long irq_flags;
746 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
750 entry = ring_buffer_event_data(event);
751 tracing_generic_entry_update(entry, flags);
752 entry->type = TRACE_CTX;
753 entry->field.ctx.prev_pid = prev->pid;
754 entry->field.ctx.prev_prio = prev->prio;
755 entry->field.ctx.prev_state = prev->state;
756 entry->field.ctx.next_pid = next->pid;
757 entry->field.ctx.next_prio = next->prio;
758 entry->field.ctx.next_state = next->state;
759 entry->field.ctx.next_cpu = task_cpu(next);
760 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
761 __trace_stack(tr, data, flags, 5);
765 tracing_sched_wakeup_trace(struct trace_array *tr,
766 struct trace_array_cpu *data,
767 struct task_struct *wakee,
768 struct task_struct *curr,
771 struct ring_buffer_event *event;
772 struct trace_entry *entry;
773 unsigned long irq_flags;
775 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
779 entry = ring_buffer_event_data(event);
780 tracing_generic_entry_update(entry, flags);
781 entry->type = TRACE_WAKE;
782 entry->field.ctx.prev_pid = curr->pid;
783 entry->field.ctx.prev_prio = curr->prio;
784 entry->field.ctx.prev_state = curr->state;
785 entry->field.ctx.next_pid = wakee->pid;
786 entry->field.ctx.next_prio = wakee->prio;
787 entry->field.ctx.next_state = wakee->state;
788 entry->field.ctx.next_cpu = task_cpu(wakee);
789 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
790 __trace_stack(tr, data, flags, 6);
796 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
798 struct trace_array *tr = &global_trace;
799 struct trace_array_cpu *data;
804 if (tracing_disabled || !tr->ctrl)
807 local_irq_save(flags);
808 cpu = raw_smp_processor_id();
809 data = tr->data[cpu];
810 disabled = atomic_inc_return(&data->disabled);
812 if (likely(disabled == 1))
813 __trace_special(tr, data, arg1, arg2, arg3);
815 atomic_dec(&data->disabled);
816 local_irq_restore(flags);
821 function_trace_call(unsigned long ip, unsigned long parent_ip)
823 struct trace_array *tr = &global_trace;
824 struct trace_array_cpu *data;
829 if (unlikely(!ftrace_function_enabled))
835 local_irq_save(flags);
836 cpu = raw_smp_processor_id();
837 data = tr->data[cpu];
838 disabled = atomic_inc_return(&data->disabled);
840 if (likely(disabled == 1))
841 trace_function(tr, data, ip, parent_ip, flags);
843 atomic_dec(&data->disabled);
844 local_irq_restore(flags);
847 static struct ftrace_ops trace_ops __read_mostly =
849 .func = function_trace_call,
852 void tracing_start_function_trace(void)
854 ftrace_function_enabled = 0;
855 register_ftrace_function(&trace_ops);
857 ftrace_function_enabled = 1;
860 void tracing_stop_function_trace(void)
862 ftrace_function_enabled = 0;
863 unregister_ftrace_function(&trace_ops);
867 enum trace_file_type {
868 TRACE_FILE_LAT_FMT = 1,
871 static void trace_iterator_increment(struct trace_iterator *iter, int cpu)
874 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
877 static struct trace_entry *
878 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
880 struct ring_buffer_event *event;
881 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
883 event = ring_buffer_iter_peek(buf_iter, ts);
884 return event ? ring_buffer_event_data(event) : NULL;
886 static struct trace_entry *
887 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
889 struct ring_buffer *buffer = iter->tr->buffer;
890 struct trace_entry *ent, *next = NULL;
895 for_each_tracing_cpu(cpu) {
897 if (ring_buffer_empty_cpu(buffer, cpu))
900 ent = peek_next_entry(iter, cpu, &ts);
903 * Pick the entry with the smallest timestamp:
905 if (ent && (!next || ts < next_ts)) {
921 /* Find the next real entry, without updating the iterator itself */
922 static struct trace_entry *
923 find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
925 return __find_next_entry(iter, ent_cpu, ent_ts);
928 /* Find the next real entry, and increment the iterator to the next entry */
929 static void *find_next_entry_inc(struct trace_iterator *iter)
931 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
934 trace_iterator_increment(iter, iter->cpu);
936 return iter->ent ? iter : NULL;
939 static void trace_consume(struct trace_iterator *iter)
941 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
944 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
946 struct trace_iterator *iter = m->private;
952 /* can't go backwards */
957 ent = find_next_entry_inc(iter);
961 while (ent && iter->idx < i)
962 ent = find_next_entry_inc(iter);
969 static void *s_start(struct seq_file *m, loff_t *pos)
971 struct trace_iterator *iter = m->private;
976 mutex_lock(&trace_types_lock);
978 if (!current_trace || current_trace != iter->trace) {
979 mutex_unlock(&trace_types_lock);
983 atomic_inc(&trace_record_cmdline_disabled);
985 /* let the tracer grab locks here if needed */
986 if (current_trace->start)
987 current_trace->start(iter);
989 if (*pos != iter->pos) {
994 for_each_tracing_cpu(cpu) {
995 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
998 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1003 p = s_next(m, p, &l);
1009 static void s_stop(struct seq_file *m, void *p)
1011 struct trace_iterator *iter = m->private;
1013 atomic_dec(&trace_record_cmdline_disabled);
1015 /* let the tracer release locks here if needed */
1016 if (current_trace && current_trace == iter->trace && iter->trace->stop)
1017 iter->trace->stop(iter);
1019 mutex_unlock(&trace_types_lock);
1022 #define KRETPROBE_MSG "[unknown/kretprobe'd]"
1024 #ifdef CONFIG_KRETPROBES
1025 static inline int kretprobed(unsigned long addr)
1027 return addr == (unsigned long)kretprobe_trampoline;
1030 static inline int kretprobed(unsigned long addr)
1034 #endif /* CONFIG_KRETPROBES */
1037 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1039 #ifdef CONFIG_KALLSYMS
1040 char str[KSYM_SYMBOL_LEN];
1042 kallsyms_lookup(address, NULL, NULL, NULL, str);
1044 return trace_seq_printf(s, fmt, str);
1050 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1051 unsigned long address)
1053 #ifdef CONFIG_KALLSYMS
1054 char str[KSYM_SYMBOL_LEN];
1056 sprint_symbol(str, address);
1057 return trace_seq_printf(s, fmt, str);
1062 #ifndef CONFIG_64BIT
1063 # define IP_FMT "%08lx"
1065 # define IP_FMT "%016lx"
1069 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1074 return trace_seq_printf(s, "0");
1076 if (sym_flags & TRACE_ITER_SYM_OFFSET)
1077 ret = seq_print_sym_offset(s, "%s", ip);
1079 ret = seq_print_sym_short(s, "%s", ip);
1084 if (sym_flags & TRACE_ITER_SYM_ADDR)
1085 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1089 static void print_lat_help_header(struct seq_file *m)
1091 seq_puts(m, "# _------=> CPU# \n");
1092 seq_puts(m, "# / _-----=> irqs-off \n");
1093 seq_puts(m, "# | / _----=> need-resched \n");
1094 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1095 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1096 seq_puts(m, "# |||| / \n");
1097 seq_puts(m, "# ||||| delay \n");
1098 seq_puts(m, "# cmd pid ||||| time | caller \n");
1099 seq_puts(m, "# \\ / ||||| \\ | / \n");
1102 static void print_func_help_header(struct seq_file *m)
1104 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1105 seq_puts(m, "# | | | | |\n");
1110 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1112 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1113 struct trace_array *tr = iter->tr;
1114 struct trace_array_cpu *data = tr->data[tr->cpu];
1115 struct tracer *type = current_trace;
1116 unsigned long total;
1117 unsigned long entries;
1118 const char *name = "preemption";
1123 entries = ring_buffer_entries(iter->tr->buffer);
1125 ring_buffer_overruns(iter->tr->buffer);
1127 seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1129 seq_puts(m, "-----------------------------------"
1130 "---------------------------------\n");
1131 seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1132 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1133 nsecs_to_usecs(data->saved_latency),
1137 #if defined(CONFIG_PREEMPT_NONE)
1139 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1141 #elif defined(CONFIG_PREEMPT)
1146 /* These are reserved for later use */
1149 seq_printf(m, " #P:%d)\n", num_online_cpus());
1153 seq_puts(m, " -----------------\n");
1154 seq_printf(m, " | task: %.16s-%d "
1155 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1156 data->comm, data->pid, data->uid, data->nice,
1157 data->policy, data->rt_priority);
1158 seq_puts(m, " -----------------\n");
1160 if (data->critical_start) {
1161 seq_puts(m, " => started at: ");
1162 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1163 trace_print_seq(m, &iter->seq);
1164 seq_puts(m, "\n => ended at: ");
1165 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1166 trace_print_seq(m, &iter->seq);
1174 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1176 struct trace_field *field = &entry->field;
1177 int hardirq, softirq;
1180 comm = trace_find_cmdline(field->pid);
1182 trace_seq_printf(s, "%8.8s-%-5d ", comm, field->pid);
1183 trace_seq_printf(s, "%3d", cpu);
1184 trace_seq_printf(s, "%c%c",
1185 (field->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1186 ((field->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1188 hardirq = field->flags & TRACE_FLAG_HARDIRQ;
1189 softirq = field->flags & TRACE_FLAG_SOFTIRQ;
1190 if (hardirq && softirq) {
1191 trace_seq_putc(s, 'H');
1194 trace_seq_putc(s, 'h');
1197 trace_seq_putc(s, 's');
1199 trace_seq_putc(s, '.');
1203 if (field->preempt_count)
1204 trace_seq_printf(s, "%x", field->preempt_count);
1206 trace_seq_puts(s, ".");
1209 unsigned long preempt_mark_thresh = 100;
1212 lat_print_timestamp(struct trace_seq *s, u64 abs_usecs,
1213 unsigned long rel_usecs)
1215 trace_seq_printf(s, " %4lldus", abs_usecs);
1216 if (rel_usecs > preempt_mark_thresh)
1217 trace_seq_puts(s, "!: ");
1218 else if (rel_usecs > 1)
1219 trace_seq_puts(s, "+: ");
1221 trace_seq_puts(s, " : ");
1224 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1227 * The message is supposed to contain an ending newline.
1228 * If the printing stops prematurely, try to add a newline of our own.
1230 void trace_seq_print_cont(struct trace_seq *s, struct trace_iterator *iter)
1232 struct trace_entry *ent;
1235 ent = peek_next_entry(iter, iter->cpu, NULL);
1236 if (!ent || ent->type != TRACE_CONT) {
1237 trace_seq_putc(s, '\n');
1243 ok = (trace_seq_printf(s, "%s", ent->cont.buf) > 0);
1244 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1245 ent = peek_next_entry(iter, iter->cpu, NULL);
1246 } while (ent && ent->type == TRACE_CONT);
1249 trace_seq_putc(s, '\n');
1253 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1255 struct trace_seq *s = &iter->seq;
1256 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1257 struct trace_entry *next_entry;
1258 unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1259 struct trace_entry *entry = iter->ent;
1260 struct trace_field *field = &entry->field;
1261 unsigned long abs_usecs;
1262 unsigned long rel_usecs;
1269 if (entry->type == TRACE_CONT)
1272 next_entry = find_next_entry(iter, NULL, &next_ts);
1275 rel_usecs = ns2usecs(next_ts - iter->ts);
1276 abs_usecs = ns2usecs(iter->ts - iter->tr->time_start);
1279 comm = trace_find_cmdline(field->pid);
1280 trace_seq_printf(s, "%16s %5d %3d %d %08x %08x [%08lx]"
1281 " %ld.%03ldms (+%ld.%03ldms): ",
1283 field->pid, cpu, field->flags,
1284 field->preempt_count, trace_idx,
1287 abs_usecs % 1000, rel_usecs/1000,
1290 lat_print_generic(s, entry, cpu);
1291 lat_print_timestamp(s, abs_usecs, rel_usecs);
1293 switch (entry->type) {
1295 seq_print_ip_sym(s, field->fn.ip, sym_flags);
1296 trace_seq_puts(s, " (");
1297 if (kretprobed(field->fn.parent_ip))
1298 trace_seq_puts(s, KRETPROBE_MSG);
1300 seq_print_ip_sym(s, field->fn.parent_ip, sym_flags);
1301 trace_seq_puts(s, ")\n");
1305 T = field->ctx.next_state < sizeof(state_to_char) ?
1306 state_to_char[field->ctx.next_state] : 'X';
1308 state = field->ctx.prev_state ?
1309 __ffs(field->ctx.prev_state) + 1 : 0;
1310 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1311 comm = trace_find_cmdline(field->ctx.next_pid);
1312 trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c %s\n",
1313 field->ctx.prev_pid,
1314 field->ctx.prev_prio,
1315 S, entry->type == TRACE_CTX ? "==>" : " +",
1316 field->ctx.next_cpu,
1317 field->ctx.next_pid,
1318 field->ctx.next_prio,
1322 trace_seq_printf(s, "# %ld %ld %ld\n",
1323 field->special.arg1,
1324 field->special.arg2,
1325 field->special.arg3);
1328 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1330 trace_seq_puts(s, " <= ");
1331 seq_print_ip_sym(s, field->stack.caller[i], sym_flags);
1333 trace_seq_puts(s, "\n");
1336 seq_print_ip_sym(s, field->print.ip, sym_flags);
1337 trace_seq_printf(s, ": %s", field->print.buf);
1338 if (field->flags & TRACE_FLAG_CONT)
1339 trace_seq_print_cont(s, iter);
1342 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1347 static int print_trace_fmt(struct trace_iterator *iter)
1349 struct trace_seq *s = &iter->seq;
1350 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1351 struct trace_entry *entry;
1352 struct trace_field *field;
1353 unsigned long usec_rem;
1354 unsigned long long t;
1363 if (entry->type == TRACE_CONT)
1366 field = &entry->field;
1368 comm = trace_find_cmdline(iter->ent->field.pid);
1370 t = ns2usecs(iter->ts);
1371 usec_rem = do_div(t, 1000000ULL);
1372 secs = (unsigned long)t;
1374 ret = trace_seq_printf(s, "%16s-%-5d ", comm, field->pid);
1377 ret = trace_seq_printf(s, "[%03d] ", iter->cpu);
1380 ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1384 switch (entry->type) {
1386 ret = seq_print_ip_sym(s, field->fn.ip, sym_flags);
1389 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1390 field->fn.parent_ip) {
1391 ret = trace_seq_printf(s, " <-");
1394 if (kretprobed(field->fn.parent_ip))
1395 ret = trace_seq_puts(s, KRETPROBE_MSG);
1397 ret = seq_print_ip_sym(s,
1398 field->fn.parent_ip,
1403 ret = trace_seq_printf(s, "\n");
1409 S = field->ctx.prev_state < sizeof(state_to_char) ?
1410 state_to_char[field->ctx.prev_state] : 'X';
1411 T = field->ctx.next_state < sizeof(state_to_char) ?
1412 state_to_char[field->ctx.next_state] : 'X';
1413 ret = trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c\n",
1414 field->ctx.prev_pid,
1415 field->ctx.prev_prio,
1417 entry->type == TRACE_CTX ? "==>" : " +",
1418 field->ctx.next_cpu,
1419 field->ctx.next_pid,
1420 field->ctx.next_prio,
1426 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1427 field->special.arg1,
1428 field->special.arg2,
1429 field->special.arg3);
1434 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1436 ret = trace_seq_puts(s, " <= ");
1440 ret = seq_print_ip_sym(s, field->stack.caller[i],
1445 ret = trace_seq_puts(s, "\n");
1450 seq_print_ip_sym(s, field->print.ip, sym_flags);
1451 trace_seq_printf(s, ": %s", field->print.buf);
1452 if (field->flags & TRACE_FLAG_CONT)
1453 trace_seq_print_cont(s, iter);
1459 static int print_raw_fmt(struct trace_iterator *iter)
1461 struct trace_seq *s = &iter->seq;
1462 struct trace_entry *entry;
1463 struct trace_field *field;
1469 if (entry->type == TRACE_CONT)
1472 field = &entry->field;
1474 ret = trace_seq_printf(s, "%d %d %llu ",
1475 field->pid, iter->cpu, iter->ts);
1479 switch (entry->type) {
1481 ret = trace_seq_printf(s, "%x %x\n",
1483 field->fn.parent_ip);
1489 S = field->ctx.prev_state < sizeof(state_to_char) ?
1490 state_to_char[field->ctx.prev_state] : 'X';
1491 T = field->ctx.next_state < sizeof(state_to_char) ?
1492 state_to_char[field->ctx.next_state] : 'X';
1493 if (entry->type == TRACE_WAKE)
1495 ret = trace_seq_printf(s, "%d %d %c %d %d %d %c\n",
1496 field->ctx.prev_pid,
1497 field->ctx.prev_prio,
1499 field->ctx.next_cpu,
1500 field->ctx.next_pid,
1501 field->ctx.next_prio,
1508 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1509 field->special.arg1,
1510 field->special.arg2,
1511 field->special.arg3);
1516 trace_seq_printf(s, "# %lx %s",
1517 field->print.ip, field->print.buf);
1518 if (field->flags & TRACE_FLAG_CONT)
1519 trace_seq_print_cont(s, iter);
1525 #define SEQ_PUT_FIELD_RET(s, x) \
1527 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
1531 #define SEQ_PUT_HEX_FIELD_RET(s, x) \
1533 if (!trace_seq_putmem_hex(s, &(x), sizeof(x))) \
1537 static int print_hex_fmt(struct trace_iterator *iter)
1539 struct trace_seq *s = &iter->seq;
1540 unsigned char newline = '\n';
1541 struct trace_entry *entry;
1542 struct trace_field *field;
1547 if (entry->type == TRACE_CONT)
1550 field = &entry->field;
1552 SEQ_PUT_HEX_FIELD_RET(s, field->pid);
1553 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1554 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1556 switch (entry->type) {
1558 SEQ_PUT_HEX_FIELD_RET(s, field->fn.ip);
1559 SEQ_PUT_HEX_FIELD_RET(s, field->fn.parent_ip);
1563 S = field->ctx.prev_state < sizeof(state_to_char) ?
1564 state_to_char[field->ctx.prev_state] : 'X';
1565 T = field->ctx.next_state < sizeof(state_to_char) ?
1566 state_to_char[field->ctx.next_state] : 'X';
1567 if (entry->type == TRACE_WAKE)
1569 SEQ_PUT_HEX_FIELD_RET(s, field->ctx.prev_pid);
1570 SEQ_PUT_HEX_FIELD_RET(s, field->ctx.prev_prio);
1571 SEQ_PUT_HEX_FIELD_RET(s, S);
1572 SEQ_PUT_HEX_FIELD_RET(s, field->ctx.next_cpu);
1573 SEQ_PUT_HEX_FIELD_RET(s, field->ctx.next_pid);
1574 SEQ_PUT_HEX_FIELD_RET(s, field->ctx.next_prio);
1575 SEQ_PUT_HEX_FIELD_RET(s, T);
1579 SEQ_PUT_HEX_FIELD_RET(s, field->special.arg1);
1580 SEQ_PUT_HEX_FIELD_RET(s, field->special.arg2);
1581 SEQ_PUT_HEX_FIELD_RET(s, field->special.arg3);
1584 SEQ_PUT_FIELD_RET(s, newline);
1589 static int print_bin_fmt(struct trace_iterator *iter)
1591 struct trace_seq *s = &iter->seq;
1592 struct trace_entry *entry;
1593 struct trace_field *field;
1597 if (entry->type == TRACE_CONT)
1600 field = &entry->field;
1602 SEQ_PUT_FIELD_RET(s, field->pid);
1603 SEQ_PUT_FIELD_RET(s, field->cpu);
1604 SEQ_PUT_FIELD_RET(s, iter->ts);
1606 switch (entry->type) {
1608 SEQ_PUT_FIELD_RET(s, field->fn.ip);
1609 SEQ_PUT_FIELD_RET(s, field->fn.parent_ip);
1612 SEQ_PUT_FIELD_RET(s, field->ctx.prev_pid);
1613 SEQ_PUT_FIELD_RET(s, field->ctx.prev_prio);
1614 SEQ_PUT_FIELD_RET(s, field->ctx.prev_state);
1615 SEQ_PUT_FIELD_RET(s, field->ctx.next_pid);
1616 SEQ_PUT_FIELD_RET(s, field->ctx.next_prio);
1617 SEQ_PUT_FIELD_RET(s, field->ctx.next_state);
1621 SEQ_PUT_FIELD_RET(s, field->special.arg1);
1622 SEQ_PUT_FIELD_RET(s, field->special.arg2);
1623 SEQ_PUT_FIELD_RET(s, field->special.arg3);
1629 static int trace_empty(struct trace_iterator *iter)
1633 for_each_tracing_cpu(cpu) {
1634 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1640 static int print_trace_line(struct trace_iterator *iter)
1642 if (iter->trace && iter->trace->print_line)
1643 return iter->trace->print_line(iter);
1645 if (trace_flags & TRACE_ITER_BIN)
1646 return print_bin_fmt(iter);
1648 if (trace_flags & TRACE_ITER_HEX)
1649 return print_hex_fmt(iter);
1651 if (trace_flags & TRACE_ITER_RAW)
1652 return print_raw_fmt(iter);
1654 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1655 return print_lat_fmt(iter, iter->idx, iter->cpu);
1657 return print_trace_fmt(iter);
1660 static int s_show(struct seq_file *m, void *v)
1662 struct trace_iterator *iter = v;
1664 if (iter->ent == NULL) {
1666 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1669 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1670 /* print nothing if the buffers are empty */
1671 if (trace_empty(iter))
1673 print_trace_header(m, iter);
1674 if (!(trace_flags & TRACE_ITER_VERBOSE))
1675 print_lat_help_header(m);
1677 if (!(trace_flags & TRACE_ITER_VERBOSE))
1678 print_func_help_header(m);
1681 print_trace_line(iter);
1682 trace_print_seq(m, &iter->seq);
1688 static struct seq_operations tracer_seq_ops = {
1695 static struct trace_iterator *
1696 __tracing_open(struct inode *inode, struct file *file, int *ret)
1698 struct trace_iterator *iter;
1702 if (tracing_disabled) {
1707 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1713 mutex_lock(&trace_types_lock);
1714 if (current_trace && current_trace->print_max)
1717 iter->tr = inode->i_private;
1718 iter->trace = current_trace;
1721 for_each_tracing_cpu(cpu) {
1722 iter->buffer_iter[cpu] =
1723 ring_buffer_read_start(iter->tr->buffer, cpu);
1724 if (!iter->buffer_iter[cpu])
1728 /* TODO stop tracer */
1729 *ret = seq_open(file, &tracer_seq_ops);
1733 m = file->private_data;
1736 /* stop the trace while dumping */
1737 if (iter->tr->ctrl) {
1739 ftrace_function_enabled = 0;
1742 if (iter->trace && iter->trace->open)
1743 iter->trace->open(iter);
1745 mutex_unlock(&trace_types_lock);
1751 for_each_tracing_cpu(cpu) {
1752 if (iter->buffer_iter[cpu])
1753 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1755 mutex_unlock(&trace_types_lock);
1757 return ERR_PTR(-ENOMEM);
1760 int tracing_open_generic(struct inode *inode, struct file *filp)
1762 if (tracing_disabled)
1765 filp->private_data = inode->i_private;
1769 int tracing_release(struct inode *inode, struct file *file)
1771 struct seq_file *m = (struct seq_file *)file->private_data;
1772 struct trace_iterator *iter = m->private;
1775 mutex_lock(&trace_types_lock);
1776 for_each_tracing_cpu(cpu) {
1777 if (iter->buffer_iter[cpu])
1778 ring_buffer_read_finish(iter->buffer_iter[cpu]);
1781 if (iter->trace && iter->trace->close)
1782 iter->trace->close(iter);
1784 /* reenable tracing if it was previously enabled */
1785 if (iter->tr->ctrl) {
1788 * It is safe to enable function tracing even if it
1791 ftrace_function_enabled = 1;
1793 mutex_unlock(&trace_types_lock);
1795 seq_release(inode, file);
1800 static int tracing_open(struct inode *inode, struct file *file)
1804 __tracing_open(inode, file, &ret);
1809 static int tracing_lt_open(struct inode *inode, struct file *file)
1811 struct trace_iterator *iter;
1814 iter = __tracing_open(inode, file, &ret);
1817 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1824 t_next(struct seq_file *m, void *v, loff_t *pos)
1826 struct tracer *t = m->private;
1838 static void *t_start(struct seq_file *m, loff_t *pos)
1840 struct tracer *t = m->private;
1843 mutex_lock(&trace_types_lock);
1844 for (; t && l < *pos; t = t_next(m, t, &l))
1850 static void t_stop(struct seq_file *m, void *p)
1852 mutex_unlock(&trace_types_lock);
1855 static int t_show(struct seq_file *m, void *v)
1857 struct tracer *t = v;
1862 seq_printf(m, "%s", t->name);
1871 static struct seq_operations show_traces_seq_ops = {
1878 static int show_traces_open(struct inode *inode, struct file *file)
1882 if (tracing_disabled)
1885 ret = seq_open(file, &show_traces_seq_ops);
1887 struct seq_file *m = file->private_data;
1888 m->private = trace_types;
1894 static struct file_operations tracing_fops = {
1895 .open = tracing_open,
1897 .llseek = seq_lseek,
1898 .release = tracing_release,
1901 static struct file_operations tracing_lt_fops = {
1902 .open = tracing_lt_open,
1904 .llseek = seq_lseek,
1905 .release = tracing_release,
1908 static struct file_operations show_traces_fops = {
1909 .open = show_traces_open,
1911 .release = seq_release,
1915 * Only trace on a CPU if the bitmask is set:
1917 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
1920 * When tracing/tracing_cpu_mask is modified then this holds
1921 * the new bitmask we are about to install:
1923 static cpumask_t tracing_cpumask_new;
1926 * The tracer itself will not take this lock, but still we want
1927 * to provide a consistent cpumask to user-space:
1929 static DEFINE_MUTEX(tracing_cpumask_update_lock);
1932 * Temporary storage for the character representation of the
1933 * CPU bitmask (and one more byte for the newline):
1935 static char mask_str[NR_CPUS + 1];
1938 tracing_cpumask_read(struct file *filp, char __user *ubuf,
1939 size_t count, loff_t *ppos)
1943 mutex_lock(&tracing_cpumask_update_lock);
1945 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
1946 if (count - len < 2) {
1950 len += sprintf(mask_str + len, "\n");
1951 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
1954 mutex_unlock(&tracing_cpumask_update_lock);
1960 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
1961 size_t count, loff_t *ppos)
1965 mutex_lock(&tracing_cpumask_update_lock);
1966 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
1970 raw_local_irq_disable();
1971 __raw_spin_lock(&ftrace_max_lock);
1972 for_each_tracing_cpu(cpu) {
1974 * Increase/decrease the disabled counter if we are
1975 * about to flip a bit in the cpumask:
1977 if (cpu_isset(cpu, tracing_cpumask) &&
1978 !cpu_isset(cpu, tracing_cpumask_new)) {
1979 atomic_inc(&global_trace.data[cpu]->disabled);
1981 if (!cpu_isset(cpu, tracing_cpumask) &&
1982 cpu_isset(cpu, tracing_cpumask_new)) {
1983 atomic_dec(&global_trace.data[cpu]->disabled);
1986 __raw_spin_unlock(&ftrace_max_lock);
1987 raw_local_irq_enable();
1989 tracing_cpumask = tracing_cpumask_new;
1991 mutex_unlock(&tracing_cpumask_update_lock);
1996 mutex_unlock(&tracing_cpumask_update_lock);
2001 static struct file_operations tracing_cpumask_fops = {
2002 .open = tracing_open_generic,
2003 .read = tracing_cpumask_read,
2004 .write = tracing_cpumask_write,
2008 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
2009 size_t cnt, loff_t *ppos)
2016 /* calulate max size */
2017 for (i = 0; trace_options[i]; i++) {
2018 len += strlen(trace_options[i]);
2019 len += 3; /* "no" and space */
2022 /* +2 for \n and \0 */
2023 buf = kmalloc(len + 2, GFP_KERNEL);
2027 for (i = 0; trace_options[i]; i++) {
2028 if (trace_flags & (1 << i))
2029 r += sprintf(buf + r, "%s ", trace_options[i]);
2031 r += sprintf(buf + r, "no%s ", trace_options[i]);
2034 r += sprintf(buf + r, "\n");
2035 WARN_ON(r >= len + 2);
2037 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2045 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
2046 size_t cnt, loff_t *ppos)
2053 if (cnt >= sizeof(buf))
2056 if (copy_from_user(&buf, ubuf, cnt))
2061 if (strncmp(buf, "no", 2) == 0) {
2066 for (i = 0; trace_options[i]; i++) {
2067 int len = strlen(trace_options[i]);
2069 if (strncmp(cmp, trace_options[i], len) == 0) {
2071 trace_flags &= ~(1 << i);
2073 trace_flags |= (1 << i);
2078 * If no option could be set, return an error:
2080 if (!trace_options[i])
2088 static struct file_operations tracing_iter_fops = {
2089 .open = tracing_open_generic,
2090 .read = tracing_iter_ctrl_read,
2091 .write = tracing_iter_ctrl_write,
2094 static const char readme_msg[] =
2095 "tracing mini-HOWTO:\n\n"
2097 "# mount -t debugfs nodev /debug\n\n"
2098 "# cat /debug/tracing/available_tracers\n"
2099 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2100 "# cat /debug/tracing/current_tracer\n"
2102 "# echo sched_switch > /debug/tracing/current_tracer\n"
2103 "# cat /debug/tracing/current_tracer\n"
2105 "# cat /debug/tracing/iter_ctrl\n"
2106 "noprint-parent nosym-offset nosym-addr noverbose\n"
2107 "# echo print-parent > /debug/tracing/iter_ctrl\n"
2108 "# echo 1 > /debug/tracing/tracing_enabled\n"
2109 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2110 "echo 0 > /debug/tracing/tracing_enabled\n"
2114 tracing_readme_read(struct file *filp, char __user *ubuf,
2115 size_t cnt, loff_t *ppos)
2117 return simple_read_from_buffer(ubuf, cnt, ppos,
2118 readme_msg, strlen(readme_msg));
2121 static struct file_operations tracing_readme_fops = {
2122 .open = tracing_open_generic,
2123 .read = tracing_readme_read,
2127 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2128 size_t cnt, loff_t *ppos)
2130 struct trace_array *tr = filp->private_data;
2134 r = sprintf(buf, "%ld\n", tr->ctrl);
2135 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2139 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2140 size_t cnt, loff_t *ppos)
2142 struct trace_array *tr = filp->private_data;
2147 if (cnt >= sizeof(buf))
2150 if (copy_from_user(&buf, ubuf, cnt))
2155 ret = strict_strtoul(buf, 10, &val);
2161 mutex_lock(&trace_types_lock);
2162 if (tr->ctrl ^ val) {
2170 if (current_trace && current_trace->ctrl_update)
2171 current_trace->ctrl_update(tr);
2173 mutex_unlock(&trace_types_lock);
2181 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2182 size_t cnt, loff_t *ppos)
2184 char buf[max_tracer_type_len+2];
2187 mutex_lock(&trace_types_lock);
2189 r = sprintf(buf, "%s\n", current_trace->name);
2191 r = sprintf(buf, "\n");
2192 mutex_unlock(&trace_types_lock);
2194 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2198 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2199 size_t cnt, loff_t *ppos)
2201 struct trace_array *tr = &global_trace;
2203 char buf[max_tracer_type_len+1];
2206 if (cnt > max_tracer_type_len)
2207 cnt = max_tracer_type_len;
2209 if (copy_from_user(&buf, ubuf, cnt))
2214 /* strip ending whitespace. */
2215 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2218 mutex_lock(&trace_types_lock);
2219 for (t = trace_types; t; t = t->next) {
2220 if (strcmp(t->name, buf) == 0)
2223 if (!t || t == current_trace)
2226 if (current_trace && current_trace->reset)
2227 current_trace->reset(tr);
2234 mutex_unlock(&trace_types_lock);
2242 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2243 size_t cnt, loff_t *ppos)
2245 unsigned long *ptr = filp->private_data;
2249 r = snprintf(buf, sizeof(buf), "%ld\n",
2250 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2251 if (r > sizeof(buf))
2253 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2257 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2258 size_t cnt, loff_t *ppos)
2260 long *ptr = filp->private_data;
2265 if (cnt >= sizeof(buf))
2268 if (copy_from_user(&buf, ubuf, cnt))
2273 ret = strict_strtoul(buf, 10, &val);
2282 static atomic_t tracing_reader;
2284 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2286 struct trace_iterator *iter;
2289 if (tracing_disabled)
2292 /* We only allow for reader of the pipe */
2293 if (atomic_inc_return(&tracing_reader) != 1) {
2294 atomic_dec(&tracing_reader);
2298 /* create a buffer to store the information to pass to userspace */
2299 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2303 mutex_lock(&trace_types_lock);
2304 iter->tr = &global_trace;
2305 iter->trace = current_trace;
2306 filp->private_data = iter;
2308 for_each_tracing_cpu(cpu) {
2309 iter->buffer_iter[cpu] =
2310 ring_buffer_read_start(iter->tr->buffer, cpu);
2311 if (!iter->buffer_iter[cpu])
2315 if (iter->trace->pipe_open)
2316 iter->trace->pipe_open(iter);
2317 mutex_unlock(&trace_types_lock);
2322 for_each_tracing_cpu(cpu) {
2323 if (iter->buffer_iter[cpu])
2324 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2326 mutex_unlock(&trace_types_lock);
2331 static int tracing_release_pipe(struct inode *inode, struct file *file)
2333 struct trace_iterator *iter = file->private_data;
2336 for_each_tracing_cpu(cpu) {
2337 if (iter->buffer_iter[cpu])
2338 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2341 atomic_dec(&tracing_reader);
2347 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2349 struct trace_iterator *iter = filp->private_data;
2351 if (trace_flags & TRACE_ITER_BLOCK) {
2353 * Always select as readable when in blocking mode
2355 return POLLIN | POLLRDNORM;
2357 if (!trace_empty(iter))
2358 return POLLIN | POLLRDNORM;
2359 poll_wait(filp, &trace_wait, poll_table);
2360 if (!trace_empty(iter))
2361 return POLLIN | POLLRDNORM;
2371 tracing_read_pipe(struct file *filp, char __user *ubuf,
2372 size_t cnt, loff_t *ppos)
2374 struct trace_iterator *iter = filp->private_data;
2375 unsigned long flags;
2376 #ifdef CONFIG_FTRACE
2381 /* return any leftover data */
2382 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2387 trace_seq_reset(&iter->seq);
2389 mutex_lock(&trace_types_lock);
2390 if (iter->trace->read) {
2391 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2396 while (trace_empty(iter)) {
2398 if ((filp->f_flags & O_NONBLOCK)) {
2404 * This is a make-shift waitqueue. The reason we don't use
2405 * an actual wait queue is because:
2406 * 1) we only ever have one waiter
2407 * 2) the tracing, traces all functions, we don't want
2408 * the overhead of calling wake_up and friends
2409 * (and tracing them too)
2410 * Anyway, this is really very primitive wakeup.
2412 set_current_state(TASK_INTERRUPTIBLE);
2413 iter->tr->waiter = current;
2415 mutex_unlock(&trace_types_lock);
2417 /* sleep for 100 msecs, and try again. */
2418 schedule_timeout(HZ/10);
2420 mutex_lock(&trace_types_lock);
2422 iter->tr->waiter = NULL;
2424 if (signal_pending(current)) {
2429 if (iter->trace != current_trace)
2433 * We block until we read something and tracing is disabled.
2434 * We still block if tracing is disabled, but we have never
2435 * read anything. This allows a user to cat this file, and
2436 * then enable tracing. But after we have read something,
2437 * we give an EOF when tracing is again disabled.
2439 * iter->pos will be 0 if we haven't read anything.
2441 if (!tracer_enabled && iter->pos)
2447 /* stop when tracing is finished */
2448 if (trace_empty(iter))
2451 if (cnt >= PAGE_SIZE)
2452 cnt = PAGE_SIZE - 1;
2454 /* reset all but tr, trace, and overruns */
2455 memset(&iter->seq, 0,
2456 sizeof(struct trace_iterator) -
2457 offsetof(struct trace_iterator, seq));
2461 * We need to stop all tracing on all CPUS to read the
2462 * the next buffer. This is a bit expensive, but is
2463 * not done often. We fill all what we can read,
2464 * and then release the locks again.
2467 local_irq_disable();
2468 #ifdef CONFIG_FTRACE
2469 ftrace_save = ftrace_enabled;
2473 ring_buffer_lock(iter->tr->buffer, &flags);
2475 while (find_next_entry_inc(iter) != NULL) {
2477 int len = iter->seq.len;
2479 ret = print_trace_line(iter);
2481 /* don't print partial lines */
2482 iter->seq.len = len;
2486 trace_consume(iter);
2488 if (iter->seq.len >= cnt)
2492 ring_buffer_unlock(iter->tr->buffer, flags);
2493 #ifdef CONFIG_FTRACE
2494 ftrace_enabled = ftrace_save;
2498 /* Now copy what we have to the user */
2499 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2500 if (iter->seq.readpos >= iter->seq.len)
2501 trace_seq_reset(&iter->seq);
2506 mutex_unlock(&trace_types_lock);
2512 tracing_entries_read(struct file *filp, char __user *ubuf,
2513 size_t cnt, loff_t *ppos)
2515 struct trace_array *tr = filp->private_data;
2519 r = sprintf(buf, "%lu\n", tr->entries);
2520 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2524 tracing_entries_write(struct file *filp, const char __user *ubuf,
2525 size_t cnt, loff_t *ppos)
2530 struct trace_array *tr = filp->private_data;
2532 if (cnt >= sizeof(buf))
2535 if (copy_from_user(&buf, ubuf, cnt))
2540 ret = strict_strtoul(buf, 10, &val);
2544 /* must have at least 1 entry */
2548 mutex_lock(&trace_types_lock);
2552 pr_info("ftrace: please disable tracing"
2553 " before modifying buffer size\n");
2557 if (val != global_trace.entries) {
2558 ret = ring_buffer_resize(global_trace.buffer, val);
2564 ret = ring_buffer_resize(max_tr.buffer, val);
2568 r = ring_buffer_resize(global_trace.buffer,
2569 global_trace.entries);
2571 /* AARGH! We are left with different
2572 * size max buffer!!!! */
2574 tracing_disabled = 1;
2579 global_trace.entries = val;
2584 /* If check pages failed, return ENOMEM */
2585 if (tracing_disabled)
2588 max_tr.entries = global_trace.entries;
2589 mutex_unlock(&trace_types_lock);
2594 static int mark_printk(const char *fmt, ...)
2598 va_start(args, fmt);
2599 ret = trace_vprintk(0, fmt, args);
2605 tracing_mark_write(struct file *filp, const char __user *ubuf,
2606 size_t cnt, loff_t *fpos)
2610 struct trace_array *tr = &global_trace;
2612 if (!tr->ctrl || tracing_disabled)
2615 if (cnt > TRACE_BUF_SIZE)
2616 cnt = TRACE_BUF_SIZE;
2618 buf = kmalloc(cnt + 1, GFP_KERNEL);
2622 if (copy_from_user(buf, ubuf, cnt)) {
2627 /* Cut from the first nil or newline. */
2629 end = strchr(buf, '\n');
2633 cnt = mark_printk("%s\n", buf);
2640 static struct file_operations tracing_max_lat_fops = {
2641 .open = tracing_open_generic,
2642 .read = tracing_max_lat_read,
2643 .write = tracing_max_lat_write,
2646 static struct file_operations tracing_ctrl_fops = {
2647 .open = tracing_open_generic,
2648 .read = tracing_ctrl_read,
2649 .write = tracing_ctrl_write,
2652 static struct file_operations set_tracer_fops = {
2653 .open = tracing_open_generic,
2654 .read = tracing_set_trace_read,
2655 .write = tracing_set_trace_write,
2658 static struct file_operations tracing_pipe_fops = {
2659 .open = tracing_open_pipe,
2660 .poll = tracing_poll_pipe,
2661 .read = tracing_read_pipe,
2662 .release = tracing_release_pipe,
2665 static struct file_operations tracing_entries_fops = {
2666 .open = tracing_open_generic,
2667 .read = tracing_entries_read,
2668 .write = tracing_entries_write,
2671 static struct file_operations tracing_mark_fops = {
2672 .open = tracing_open_generic,
2673 .write = tracing_mark_write,
2676 #ifdef CONFIG_DYNAMIC_FTRACE
2679 tracing_read_long(struct file *filp, char __user *ubuf,
2680 size_t cnt, loff_t *ppos)
2682 unsigned long *p = filp->private_data;
2686 r = sprintf(buf, "%ld\n", *p);
2688 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2691 static struct file_operations tracing_read_long_fops = {
2692 .open = tracing_open_generic,
2693 .read = tracing_read_long,
2697 static struct dentry *d_tracer;
2699 struct dentry *tracing_init_dentry(void)
2706 d_tracer = debugfs_create_dir("tracing", NULL);
2708 if (!d_tracer && !once) {
2710 pr_warning("Could not create debugfs directory 'tracing'\n");
2717 #ifdef CONFIG_FTRACE_SELFTEST
2718 /* Let selftest have access to static functions in this file */
2719 #include "trace_selftest.c"
2722 static __init int tracer_init_debugfs(void)
2724 struct dentry *d_tracer;
2725 struct dentry *entry;
2727 d_tracer = tracing_init_dentry();
2729 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2730 &global_trace, &tracing_ctrl_fops);
2732 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2734 entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2735 NULL, &tracing_iter_fops);
2737 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2739 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
2740 NULL, &tracing_cpumask_fops);
2742 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2744 entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2745 &global_trace, &tracing_lt_fops);
2747 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2749 entry = debugfs_create_file("trace", 0444, d_tracer,
2750 &global_trace, &tracing_fops);
2752 pr_warning("Could not create debugfs 'trace' entry\n");
2754 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2755 &global_trace, &show_traces_fops);
2757 pr_warning("Could not create debugfs 'available_tracers' entry\n");
2759 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2760 &global_trace, &set_tracer_fops);
2762 pr_warning("Could not create debugfs 'current_tracer' entry\n");
2764 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2765 &tracing_max_latency,
2766 &tracing_max_lat_fops);
2768 pr_warning("Could not create debugfs "
2769 "'tracing_max_latency' entry\n");
2771 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2772 &tracing_thresh, &tracing_max_lat_fops);
2774 pr_warning("Could not create debugfs "
2775 "'tracing_thresh' entry\n");
2776 entry = debugfs_create_file("README", 0644, d_tracer,
2777 NULL, &tracing_readme_fops);
2779 pr_warning("Could not create debugfs 'README' entry\n");
2781 entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2782 NULL, &tracing_pipe_fops);
2784 pr_warning("Could not create debugfs "
2785 "'trace_pipe' entry\n");
2787 entry = debugfs_create_file("trace_entries", 0644, d_tracer,
2788 &global_trace, &tracing_entries_fops);
2790 pr_warning("Could not create debugfs "
2791 "'trace_entries' entry\n");
2793 entry = debugfs_create_file("trace_marker", 0220, d_tracer,
2794 NULL, &tracing_mark_fops);
2796 pr_warning("Could not create debugfs "
2797 "'trace_marker' entry\n");
2799 #ifdef CONFIG_DYNAMIC_FTRACE
2800 entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2801 &ftrace_update_tot_cnt,
2802 &tracing_read_long_fops);
2804 pr_warning("Could not create debugfs "
2805 "'dyn_ftrace_total_info' entry\n");
2807 #ifdef CONFIG_SYSPROF_TRACER
2808 init_tracer_sysprof_debugfs(d_tracer);
2813 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
2815 static DEFINE_SPINLOCK(trace_buf_lock);
2816 static char trace_buf[TRACE_BUF_SIZE];
2818 struct ring_buffer_event *event;
2819 struct trace_array *tr = &global_trace;
2820 struct trace_array_cpu *data;
2821 struct trace_entry *entry;
2822 unsigned long flags, irq_flags;
2824 int cpu, len = 0, write, written = 0;
2826 if (!tr->ctrl || tracing_disabled)
2829 local_irq_save(flags);
2830 cpu = raw_smp_processor_id();
2831 data = tr->data[cpu];
2832 disabled = atomic_inc_return(&data->disabled);
2834 if (unlikely(disabled != 1))
2837 spin_lock(&trace_buf_lock);
2838 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
2840 len = min(len, TRACE_BUF_SIZE-1);
2843 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
2847 entry = ring_buffer_event_data(event);
2848 tracing_generic_entry_update(entry, flags);
2849 entry->type = TRACE_PRINT;
2850 entry->field.print.ip = ip;
2852 write = min(len, (int)(TRACE_PRINT_BUF_SIZE-1));
2854 memcpy(&entry->field.print.buf, trace_buf, write);
2855 entry->field.print.buf[write] = 0;
2857 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
2860 entry->field.flags |= TRACE_FLAG_CONT;
2862 while (written != len) {
2863 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
2867 entry = ring_buffer_event_data(event);
2869 entry->type = TRACE_CONT;
2870 write = min(len - written, (int)(TRACE_CONT_BUF_SIZE-1));
2871 memcpy(&entry->cont.buf, trace_buf+written, write);
2872 entry->cont.buf[write] = 0;
2874 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
2878 spin_unlock(&trace_buf_lock);
2881 atomic_dec(&data->disabled);
2882 local_irq_restore(flags);
2886 EXPORT_SYMBOL_GPL(trace_vprintk);
2888 int __ftrace_printk(unsigned long ip, const char *fmt, ...)
2893 if (!(trace_flags & TRACE_ITER_PRINTK))
2897 ret = trace_vprintk(ip, fmt, ap);
2901 EXPORT_SYMBOL_GPL(__ftrace_printk);
2903 static int trace_panic_handler(struct notifier_block *this,
2904 unsigned long event, void *unused)
2910 static struct notifier_block trace_panic_notifier = {
2911 .notifier_call = trace_panic_handler,
2913 .priority = 150 /* priority: INT_MAX >= x >= 0 */
2916 static int trace_die_handler(struct notifier_block *self,
2930 static struct notifier_block trace_die_notifier = {
2931 .notifier_call = trace_die_handler,
2936 * printk is set to max of 1024, we really don't need it that big.
2937 * Nothing should be printing 1000 characters anyway.
2939 #define TRACE_MAX_PRINT 1000
2942 * Define here KERN_TRACE so that we have one place to modify
2943 * it if we decide to change what log level the ftrace dump
2946 #define KERN_TRACE KERN_INFO
2949 trace_printk_seq(struct trace_seq *s)
2951 /* Probably should print a warning here. */
2955 /* should be zero ended, but we are paranoid. */
2956 s->buffer[s->len] = 0;
2958 printk(KERN_TRACE "%s", s->buffer);
2964 void ftrace_dump(void)
2966 static DEFINE_SPINLOCK(ftrace_dump_lock);
2967 /* use static because iter can be a bit big for the stack */
2968 static struct trace_iterator iter;
2969 static cpumask_t mask;
2970 static int dump_ran;
2971 unsigned long flags, irq_flags;
2975 spin_lock_irqsave(&ftrace_dump_lock, flags);
2981 /* No turning back! */
2982 ftrace_kill_atomic();
2984 printk(KERN_TRACE "Dumping ftrace buffer:\n");
2986 iter.tr = &global_trace;
2987 iter.trace = current_trace;
2990 * We need to stop all tracing on all CPUS to read the
2991 * the next buffer. This is a bit expensive, but is
2992 * not done often. We fill all what we can read,
2993 * and then release the locks again.
2998 ring_buffer_lock(iter.tr->buffer, &irq_flags);
3000 while (!trace_empty(&iter)) {
3003 printk(KERN_TRACE "---------------------------------\n");
3007 /* reset all but tr, trace, and overruns */
3008 memset(&iter.seq, 0,
3009 sizeof(struct trace_iterator) -
3010 offsetof(struct trace_iterator, seq));
3011 iter.iter_flags |= TRACE_FILE_LAT_FMT;
3014 if (find_next_entry_inc(&iter) != NULL) {
3015 print_trace_line(&iter);
3016 trace_consume(&iter);
3019 trace_printk_seq(&iter.seq);
3023 printk(KERN_TRACE " (ftrace buffer empty)\n");
3025 printk(KERN_TRACE "---------------------------------\n");
3027 ring_buffer_unlock(iter.tr->buffer, irq_flags);
3030 spin_unlock_irqrestore(&ftrace_dump_lock, flags);
3033 __init static int tracer_alloc_buffers(void)
3035 struct trace_array_cpu *data;
3038 /* TODO: make the number of buffers hot pluggable with CPUS */
3039 tracing_buffer_mask = cpu_possible_map;
3041 global_trace.buffer = ring_buffer_alloc(trace_buf_size,
3042 TRACE_BUFFER_FLAGS);
3043 if (!global_trace.buffer) {
3044 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
3048 global_trace.entries = ring_buffer_size(global_trace.buffer);
3050 #ifdef CONFIG_TRACER_MAX_TRACE
3051 max_tr.buffer = ring_buffer_alloc(trace_buf_size,
3052 TRACE_BUFFER_FLAGS);
3053 if (!max_tr.buffer) {
3054 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
3056 ring_buffer_free(global_trace.buffer);
3059 max_tr.entries = ring_buffer_size(max_tr.buffer);
3060 WARN_ON(max_tr.entries != global_trace.entries);
3063 /* Allocate the first page for all buffers */
3064 for_each_tracing_cpu(i) {
3065 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
3066 max_tr.data[i] = &per_cpu(max_data, i);
3069 trace_init_cmdlines();
3071 register_tracer(&nop_trace);
3072 #ifdef CONFIG_BOOT_TRACER
3073 register_tracer(&boot_tracer);
3074 current_trace = &boot_tracer;
3075 current_trace->init(&global_trace);
3077 current_trace = &nop_trace;
3080 /* All seems OK, enable tracing */
3081 global_trace.ctrl = tracer_enabled;
3082 tracing_disabled = 0;
3084 atomic_notifier_chain_register(&panic_notifier_list,
3085 &trace_panic_notifier);
3087 register_die_notifier(&trace_die_notifier);
3091 early_initcall(tracer_alloc_buffers);
3092 fs_initcall(tracer_init_debugfs);