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/debugfs.h>
18 #include <linux/pagemap.h>
19 #include <linux/hardirq.h>
20 #include <linux/linkage.h>
21 #include <linux/uaccess.h>
22 #include <linux/ftrace.h>
23 #include <linux/module.h>
24 #include <linux/percpu.h>
25 #include <linux/ctype.h>
26 #include <linux/init.h>
27 #include <linux/poll.h>
28 #include <linux/gfp.h>
31 #include <linux/stacktrace.h>
35 unsigned long __read_mostly tracing_max_latency = (cycle_t)ULONG_MAX;
36 unsigned long __read_mostly tracing_thresh;
38 static int tracing_disabled = 1;
41 ns2usecs(cycle_t nsec)
48 cycle_t ftrace_now(int cpu)
50 return cpu_clock(cpu);
53 static struct trace_array global_trace;
55 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
57 static struct trace_array max_tr;
59 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
61 static int tracer_enabled = 1;
62 static unsigned long trace_nr_entries = 65536UL;
64 static struct tracer *trace_types __read_mostly;
65 static struct tracer *current_trace __read_mostly;
66 static int max_tracer_type_len;
68 static DEFINE_MUTEX(trace_types_lock);
69 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
71 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT;
73 void trace_wake_up(void)
76 * The runqueue_is_locked() can fail, but this is the best we
79 if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
83 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
85 static int __init set_nr_entries(char *str)
89 trace_nr_entries = simple_strtoul(str, &str, 0);
92 __setup("trace_entries=", set_nr_entries);
94 unsigned long nsecs_to_usecs(unsigned long nsecs)
100 __TRACE_FIRST_TYPE = 0,
111 enum trace_flag_type {
112 TRACE_FLAG_IRQS_OFF = 0x01,
113 TRACE_FLAG_NEED_RESCHED = 0x02,
114 TRACE_FLAG_HARDIRQ = 0x04,
115 TRACE_FLAG_SOFTIRQ = 0x08,
118 #define TRACE_ITER_SYM_MASK \
119 (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
121 /* These must match the bit postions above */
122 static const char *trace_options[] = {
136 static DEFINE_SPINLOCK(ftrace_max_lock);
139 * Copy the new maximum trace into the separate maximum-trace
140 * structure. (this way the maximum trace is permanently saved,
141 * for later retrieval via /debugfs/tracing/latency_trace)
144 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
146 struct trace_array_cpu *data = tr->data[cpu];
149 max_tr.time_start = data->preempt_timestamp;
151 data = max_tr.data[cpu];
152 data->saved_latency = tracing_max_latency;
154 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
155 data->pid = tsk->pid;
156 data->uid = tsk->uid;
157 data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
158 data->policy = tsk->policy;
159 data->rt_priority = tsk->rt_priority;
161 /* record this tasks comm */
162 tracing_record_cmdline(current);
165 void check_pages(struct trace_array_cpu *data)
167 struct page *page, *tmp;
169 BUG_ON(data->trace_pages.next->prev != &data->trace_pages);
170 BUG_ON(data->trace_pages.prev->next != &data->trace_pages);
172 list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
173 BUG_ON(page->lru.next->prev != &page->lru);
174 BUG_ON(page->lru.prev->next != &page->lru);
178 void *head_page(struct trace_array_cpu *data)
183 if (list_empty(&data->trace_pages))
186 page = list_entry(data->trace_pages.next, struct page, lru);
187 BUG_ON(&page->lru == &data->trace_pages);
189 return page_address(page);
193 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
195 int len = (PAGE_SIZE - 1) - s->len;
203 ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
206 /* If we can't write it all, don't bother writing anything */
216 trace_seq_puts(struct trace_seq *s, const char *str)
218 int len = strlen(str);
220 if (len > ((PAGE_SIZE - 1) - s->len))
223 memcpy(s->buffer + s->len, str, len);
230 trace_seq_putc(struct trace_seq *s, unsigned char c)
232 if (s->len >= (PAGE_SIZE - 1))
235 s->buffer[s->len++] = c;
241 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
243 if (len > ((PAGE_SIZE - 1) - s->len))
246 memcpy(s->buffer + s->len, mem, len);
255 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
257 unsigned char hex[HEX_CHARS];
262 BUG_ON(len >= HEX_CHARS);
267 for (i = 0, j = 0; i < len; i++) {
269 for (i = len-1, j = 0; i >= 0; i--) {
273 hex[j] = byte & 0x0f;
290 return trace_seq_putmem(s, hex, j);
294 trace_seq_reset(struct trace_seq *s)
300 trace_print_seq(struct seq_file *m, struct trace_seq *s)
302 int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
305 seq_puts(m, s->buffer);
311 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
313 struct list_head flip_pages;
315 INIT_LIST_HEAD(&flip_pages);
317 memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
318 sizeof(struct trace_array_cpu) -
319 offsetof(struct trace_array_cpu, trace_head_idx));
323 list_splice_init(&tr1->trace_pages, &flip_pages);
324 list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
325 list_splice_init(&flip_pages, &tr2->trace_pages);
326 BUG_ON(!list_empty(&flip_pages));
332 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
334 struct trace_array_cpu *data;
337 WARN_ON_ONCE(!irqs_disabled());
338 spin_lock(&ftrace_max_lock);
339 /* clear out all the previous traces */
340 for_each_possible_cpu(i) {
342 flip_trace(max_tr.data[i], data);
346 __update_max_tr(tr, tsk, cpu);
347 spin_unlock(&ftrace_max_lock);
351 * update_max_tr_single - only copy one trace over, and reset the rest
353 * @tsk - task with the latency
354 * @cpu - the cpu of the buffer to copy.
357 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
359 struct trace_array_cpu *data = tr->data[cpu];
362 WARN_ON_ONCE(!irqs_disabled());
363 spin_lock(&ftrace_max_lock);
364 for_each_possible_cpu(i)
365 tracing_reset(max_tr.data[i]);
367 flip_trace(max_tr.data[cpu], data);
370 __update_max_tr(tr, tsk, cpu);
371 spin_unlock(&ftrace_max_lock);
374 int register_tracer(struct tracer *type)
381 pr_info("Tracer must have a name\n");
385 mutex_lock(&trace_types_lock);
386 for (t = trace_types; t; t = t->next) {
387 if (strcmp(type->name, t->name) == 0) {
389 pr_info("Trace %s already registered\n",
396 #ifdef CONFIG_FTRACE_STARTUP_TEST
397 if (type->selftest) {
398 struct tracer *saved_tracer = current_trace;
399 struct trace_array_cpu *data;
400 struct trace_array *tr = &global_trace;
401 int saved_ctrl = tr->ctrl;
404 * Run a selftest on this tracer.
405 * Here we reset the trace buffer, and set the current
406 * tracer to be this tracer. The tracer can then run some
407 * internal tracing to verify that everything is in order.
408 * If we fail, we do not register this tracer.
410 for_each_possible_cpu(i) {
412 if (!head_page(data))
416 current_trace = type;
418 /* the test is responsible for initializing and enabling */
419 pr_info("Testing tracer %s: ", type->name);
420 ret = type->selftest(type, tr);
421 /* the test is responsible for resetting too */
422 current_trace = saved_tracer;
423 tr->ctrl = saved_ctrl;
425 printk(KERN_CONT "FAILED!\n");
428 /* Only reset on passing, to avoid touching corrupted buffers */
429 for_each_possible_cpu(i) {
431 if (!head_page(data))
435 printk(KERN_CONT "PASSED\n");
439 type->next = trace_types;
441 len = strlen(type->name);
442 if (len > max_tracer_type_len)
443 max_tracer_type_len = len;
446 mutex_unlock(&trace_types_lock);
451 void unregister_tracer(struct tracer *type)
456 mutex_lock(&trace_types_lock);
457 for (t = &trace_types; *t; t = &(*t)->next) {
461 pr_info("Trace %s not registered\n", type->name);
466 if (strlen(type->name) != max_tracer_type_len)
469 max_tracer_type_len = 0;
470 for (t = &trace_types; *t; t = &(*t)->next) {
471 len = strlen((*t)->name);
472 if (len > max_tracer_type_len)
473 max_tracer_type_len = len;
476 mutex_unlock(&trace_types_lock);
479 void tracing_reset(struct trace_array_cpu *data)
482 data->trace_head = data->trace_tail = head_page(data);
483 data->trace_head_idx = 0;
484 data->trace_tail_idx = 0;
487 #define SAVED_CMDLINES 128
488 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
489 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
490 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
491 static int cmdline_idx;
492 static DEFINE_SPINLOCK(trace_cmdline_lock);
493 atomic_t trace_record_cmdline_disabled;
495 static void trace_init_cmdlines(void)
497 memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
498 memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
502 void trace_stop_cmdline_recording(void);
504 static void trace_save_cmdline(struct task_struct *tsk)
509 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
513 * It's not the end of the world if we don't get
514 * the lock, but we also don't want to spin
515 * nor do we want to disable interrupts,
516 * so if we miss here, then better luck next time.
518 if (!spin_trylock(&trace_cmdline_lock))
521 idx = map_pid_to_cmdline[tsk->pid];
522 if (idx >= SAVED_CMDLINES) {
523 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
525 map = map_cmdline_to_pid[idx];
526 if (map <= PID_MAX_DEFAULT)
527 map_pid_to_cmdline[map] = (unsigned)-1;
529 map_pid_to_cmdline[tsk->pid] = idx;
534 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
536 spin_unlock(&trace_cmdline_lock);
539 static char *trace_find_cmdline(int pid)
541 char *cmdline = "<...>";
547 if (pid > PID_MAX_DEFAULT)
550 map = map_pid_to_cmdline[pid];
551 if (map >= SAVED_CMDLINES)
554 cmdline = saved_cmdlines[map];
560 void tracing_record_cmdline(struct task_struct *tsk)
562 if (atomic_read(&trace_record_cmdline_disabled))
565 trace_save_cmdline(tsk);
568 static inline struct list_head *
569 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
572 * Roundrobin - but skip the head (which is not a real page):
575 if (unlikely(next == &data->trace_pages))
577 BUG_ON(next == &data->trace_pages);
583 trace_next_page(struct trace_array_cpu *data, void *addr)
585 struct list_head *next;
588 page = virt_to_page(addr);
590 next = trace_next_list(data, &page->lru);
591 page = list_entry(next, struct page, lru);
593 return page_address(page);
596 static inline struct trace_entry *
597 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
599 unsigned long idx, idx_next;
600 struct trace_entry *entry;
603 idx = data->trace_head_idx;
606 BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
608 entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
610 if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
611 data->trace_head = trace_next_page(data, data->trace_head);
615 if (data->trace_head == data->trace_tail &&
616 idx_next == data->trace_tail_idx) {
618 data->trace_tail_idx++;
619 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
621 trace_next_page(data, data->trace_tail);
622 data->trace_tail_idx = 0;
626 data->trace_head_idx = idx_next;
632 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
634 struct task_struct *tsk = current;
637 pc = preempt_count();
639 entry->preempt_count = pc & 0xff;
640 entry->pid = tsk->pid;
641 entry->t = ftrace_now(raw_smp_processor_id());
642 entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
643 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
644 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
645 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
649 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
650 unsigned long ip, unsigned long parent_ip, unsigned long flags)
652 struct trace_entry *entry;
653 unsigned long irq_flags;
655 spin_lock_irqsave(&data->lock, irq_flags);
656 entry = tracing_get_trace_entry(tr, data);
657 tracing_generic_entry_update(entry, flags);
658 entry->type = TRACE_FN;
660 entry->fn.parent_ip = parent_ip;
661 spin_unlock_irqrestore(&data->lock, irq_flags);
665 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
666 unsigned long ip, unsigned long parent_ip, unsigned long flags)
668 if (likely(!atomic_read(&data->disabled)))
669 trace_function(tr, data, ip, parent_ip, flags);
673 __trace_special(void *__tr, void *__data,
674 unsigned long arg1, unsigned long arg2, unsigned long arg3)
676 struct trace_array_cpu *data = __data;
677 struct trace_array *tr = __tr;
678 struct trace_entry *entry;
679 unsigned long irq_flags;
681 spin_lock_irqsave(&data->lock, irq_flags);
682 entry = tracing_get_trace_entry(tr, data);
683 tracing_generic_entry_update(entry, 0);
684 entry->type = TRACE_SPECIAL;
685 entry->special.arg1 = arg1;
686 entry->special.arg2 = arg2;
687 entry->special.arg3 = arg3;
688 spin_unlock_irqrestore(&data->lock, irq_flags);
693 void __trace_stack(struct trace_array *tr,
694 struct trace_array_cpu *data,
698 struct trace_entry *entry;
699 struct stack_trace trace;
701 if (!(trace_flags & TRACE_ITER_STACKTRACE))
704 entry = tracing_get_trace_entry(tr, data);
705 tracing_generic_entry_update(entry, flags);
706 entry->type = TRACE_STACK;
708 memset(&entry->stack, 0, sizeof(entry->stack));
710 trace.nr_entries = 0;
711 trace.max_entries = FTRACE_STACK_ENTRIES;
713 trace.entries = entry->stack.caller;
715 save_stack_trace(&trace);
719 tracing_sched_switch_trace(struct trace_array *tr,
720 struct trace_array_cpu *data,
721 struct task_struct *prev,
722 struct task_struct *next,
725 struct trace_entry *entry;
726 unsigned long irq_flags;
728 spin_lock_irqsave(&data->lock, irq_flags);
729 entry = tracing_get_trace_entry(tr, data);
730 tracing_generic_entry_update(entry, flags);
731 entry->type = TRACE_CTX;
732 entry->ctx.prev_pid = prev->pid;
733 entry->ctx.prev_prio = prev->prio;
734 entry->ctx.prev_state = prev->state;
735 entry->ctx.next_pid = next->pid;
736 entry->ctx.next_prio = next->prio;
737 entry->ctx.next_state = next->state;
738 __trace_stack(tr, data, flags, 4);
739 spin_unlock_irqrestore(&data->lock, irq_flags);
743 tracing_sched_wakeup_trace(struct trace_array *tr,
744 struct trace_array_cpu *data,
745 struct task_struct *wakee,
746 struct task_struct *curr,
749 struct trace_entry *entry;
750 unsigned long irq_flags;
752 spin_lock_irqsave(&data->lock, irq_flags);
753 entry = tracing_get_trace_entry(tr, data);
754 tracing_generic_entry_update(entry, flags);
755 entry->type = TRACE_WAKE;
756 entry->ctx.prev_pid = curr->pid;
757 entry->ctx.prev_prio = curr->prio;
758 entry->ctx.prev_state = curr->state;
759 entry->ctx.next_pid = wakee->pid;
760 entry->ctx.next_prio = wakee->prio;
761 entry->ctx.next_state = wakee->state;
762 __trace_stack(tr, data, flags, 5);
763 spin_unlock_irqrestore(&data->lock, irq_flags);
770 function_trace_call(unsigned long ip, unsigned long parent_ip)
772 struct trace_array *tr = &global_trace;
773 struct trace_array_cpu *data;
778 if (unlikely(!tracer_enabled))
781 local_irq_save(flags);
782 cpu = raw_smp_processor_id();
783 data = tr->data[cpu];
784 disabled = atomic_inc_return(&data->disabled);
786 if (likely(disabled == 1))
787 trace_function(tr, data, ip, parent_ip, flags);
789 atomic_dec(&data->disabled);
790 local_irq_restore(flags);
793 static struct ftrace_ops trace_ops __read_mostly =
795 .func = function_trace_call,
798 void tracing_start_function_trace(void)
800 register_ftrace_function(&trace_ops);
803 void tracing_stop_function_trace(void)
805 unregister_ftrace_function(&trace_ops);
809 enum trace_file_type {
810 TRACE_FILE_LAT_FMT = 1,
813 static struct trace_entry *
814 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
815 struct trace_iterator *iter, int cpu)
818 struct trace_entry *array;
820 if (iter->next_idx[cpu] >= tr->entries ||
821 iter->next_idx[cpu] >= data->trace_idx ||
822 (data->trace_head == data->trace_tail &&
823 data->trace_head_idx == data->trace_tail_idx))
826 if (!iter->next_page[cpu]) {
827 /* Initialize the iterator for this cpu trace buffer */
828 WARN_ON(!data->trace_tail);
829 page = virt_to_page(data->trace_tail);
830 iter->next_page[cpu] = &page->lru;
831 iter->next_page_idx[cpu] = data->trace_tail_idx;
834 page = list_entry(iter->next_page[cpu], struct page, lru);
835 BUG_ON(&data->trace_pages == &page->lru);
837 array = page_address(page);
839 WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
840 return &array[iter->next_page_idx[cpu]];
843 static struct trace_entry *
844 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
846 struct trace_array *tr = iter->tr;
847 struct trace_entry *ent, *next = NULL;
851 for_each_possible_cpu(cpu) {
852 if (!head_page(tr->data[cpu]))
854 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
856 * Pick the entry with the smallest timestamp:
858 if (ent && (!next || ent->t < next->t)) {
870 static void trace_iterator_increment(struct trace_iterator *iter)
873 iter->next_idx[iter->cpu]++;
874 iter->next_page_idx[iter->cpu]++;
876 if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
877 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
879 iter->next_page_idx[iter->cpu] = 0;
880 iter->next_page[iter->cpu] =
881 trace_next_list(data, iter->next_page[iter->cpu]);
885 static void trace_consume(struct trace_iterator *iter)
887 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
889 data->trace_tail_idx++;
890 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
891 data->trace_tail = trace_next_page(data, data->trace_tail);
892 data->trace_tail_idx = 0;
895 /* Check if we empty it, then reset the index */
896 if (data->trace_head == data->trace_tail &&
897 data->trace_head_idx == data->trace_tail_idx)
901 static void *find_next_entry_inc(struct trace_iterator *iter)
903 struct trace_entry *next;
906 next = find_next_entry(iter, &next_cpu);
908 iter->prev_ent = iter->ent;
909 iter->prev_cpu = iter->cpu;
912 iter->cpu = next_cpu;
915 trace_iterator_increment(iter);
917 return next ? iter : NULL;
920 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
922 struct trace_iterator *iter = m->private;
923 void *last_ent = iter->ent;
929 /* can't go backwards */
934 ent = find_next_entry_inc(iter);
938 while (ent && iter->idx < i)
939 ent = find_next_entry_inc(iter);
943 if (last_ent && !ent)
944 seq_puts(m, "\n\nvim:ft=help\n");
949 static void *s_start(struct seq_file *m, loff_t *pos)
951 struct trace_iterator *iter = m->private;
956 mutex_lock(&trace_types_lock);
958 if (!current_trace || current_trace != iter->trace)
961 atomic_inc(&trace_record_cmdline_disabled);
963 /* let the tracer grab locks here if needed */
964 if (current_trace->start)
965 current_trace->start(iter);
967 if (*pos != iter->pos) {
971 iter->prev_ent = NULL;
974 for_each_possible_cpu(i) {
975 iter->next_idx[i] = 0;
976 iter->next_page[i] = NULL;
979 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
984 p = s_next(m, p, &l);
990 static void s_stop(struct seq_file *m, void *p)
992 struct trace_iterator *iter = m->private;
994 atomic_dec(&trace_record_cmdline_disabled);
996 /* let the tracer release locks here if needed */
997 if (current_trace && current_trace == iter->trace && iter->trace->stop)
998 iter->trace->stop(iter);
1000 mutex_unlock(&trace_types_lock);
1004 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1006 #ifdef CONFIG_KALLSYMS
1007 char str[KSYM_SYMBOL_LEN];
1009 kallsyms_lookup(address, NULL, NULL, NULL, str);
1011 return trace_seq_printf(s, fmt, str);
1017 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1018 unsigned long address)
1020 #ifdef CONFIG_KALLSYMS
1021 char str[KSYM_SYMBOL_LEN];
1023 sprint_symbol(str, address);
1024 return trace_seq_printf(s, fmt, str);
1029 #ifndef CONFIG_64BIT
1030 # define IP_FMT "%08lx"
1032 # define IP_FMT "%016lx"
1036 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1041 return trace_seq_printf(s, "0");
1043 if (sym_flags & TRACE_ITER_SYM_OFFSET)
1044 ret = seq_print_sym_offset(s, "%s", ip);
1046 ret = seq_print_sym_short(s, "%s", ip);
1051 if (sym_flags & TRACE_ITER_SYM_ADDR)
1052 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1056 static void print_lat_help_header(struct seq_file *m)
1058 seq_puts(m, "# _------=> CPU# \n");
1059 seq_puts(m, "# / _-----=> irqs-off \n");
1060 seq_puts(m, "# | / _----=> need-resched \n");
1061 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1062 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1063 seq_puts(m, "# |||| / \n");
1064 seq_puts(m, "# ||||| delay \n");
1065 seq_puts(m, "# cmd pid ||||| time | caller \n");
1066 seq_puts(m, "# \\ / ||||| \\ | / \n");
1069 static void print_func_help_header(struct seq_file *m)
1071 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1072 seq_puts(m, "# | | | | |\n");
1077 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1079 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1080 struct trace_array *tr = iter->tr;
1081 struct trace_array_cpu *data = tr->data[tr->cpu];
1082 struct tracer *type = current_trace;
1083 unsigned long total = 0;
1084 unsigned long entries = 0;
1086 const char *name = "preemption";
1091 for_each_possible_cpu(cpu) {
1092 if (head_page(tr->data[cpu])) {
1093 total += tr->data[cpu]->trace_idx;
1094 if (tr->data[cpu]->trace_idx > tr->entries)
1095 entries += tr->entries;
1097 entries += tr->data[cpu]->trace_idx;
1101 seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1103 seq_puts(m, "-----------------------------------"
1104 "---------------------------------\n");
1105 seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1106 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1107 nsecs_to_usecs(data->saved_latency),
1111 #if defined(CONFIG_PREEMPT_NONE)
1113 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1115 #elif defined(CONFIG_PREEMPT_DESKTOP)
1120 /* These are reserved for later use */
1123 seq_printf(m, " #P:%d)\n", num_online_cpus());
1127 seq_puts(m, " -----------------\n");
1128 seq_printf(m, " | task: %.16s-%d "
1129 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1130 data->comm, data->pid, data->uid, data->nice,
1131 data->policy, data->rt_priority);
1132 seq_puts(m, " -----------------\n");
1134 if (data->critical_start) {
1135 seq_puts(m, " => started at: ");
1136 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1137 trace_print_seq(m, &iter->seq);
1138 seq_puts(m, "\n => ended at: ");
1139 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1140 trace_print_seq(m, &iter->seq);
1148 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1150 int hardirq, softirq;
1153 comm = trace_find_cmdline(entry->pid);
1155 trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1156 trace_seq_printf(s, "%d", cpu);
1157 trace_seq_printf(s, "%c%c",
1158 (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1159 ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1161 hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1162 softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1163 if (hardirq && softirq)
1164 trace_seq_putc(s, 'H');
1167 trace_seq_putc(s, 'h');
1170 trace_seq_putc(s, 's');
1172 trace_seq_putc(s, '.');
1176 if (entry->preempt_count)
1177 trace_seq_printf(s, "%x", entry->preempt_count);
1179 trace_seq_puts(s, ".");
1182 unsigned long preempt_mark_thresh = 100;
1185 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1186 unsigned long rel_usecs)
1188 trace_seq_printf(s, " %4lldus", abs_usecs);
1189 if (rel_usecs > preempt_mark_thresh)
1190 trace_seq_puts(s, "!: ");
1191 else if (rel_usecs > 1)
1192 trace_seq_puts(s, "+: ");
1194 trace_seq_puts(s, " : ");
1197 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1200 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1202 struct trace_seq *s = &iter->seq;
1203 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1204 struct trace_entry *next_entry = find_next_entry(iter, NULL);
1205 unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1206 struct trace_entry *entry = iter->ent;
1207 unsigned long abs_usecs;
1208 unsigned long rel_usecs;
1215 rel_usecs = ns2usecs(next_entry->t - entry->t);
1216 abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1219 comm = trace_find_cmdline(entry->pid);
1220 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1221 " %ld.%03ldms (+%ld.%03ldms): ",
1223 entry->pid, cpu, entry->flags,
1224 entry->preempt_count, trace_idx,
1227 abs_usecs % 1000, rel_usecs/1000,
1230 lat_print_generic(s, entry, cpu);
1231 lat_print_timestamp(s, abs_usecs, rel_usecs);
1233 switch (entry->type) {
1235 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1236 trace_seq_puts(s, " (");
1237 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1238 trace_seq_puts(s, ")\n");
1242 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1243 state_to_char[entry->ctx.prev_state] : 'X';
1244 T = entry->ctx.next_state < sizeof(state_to_char) ?
1245 state_to_char[entry->ctx.next_state] : 'X';
1247 comm = trace_find_cmdline(entry->ctx.next_pid);
1248 trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c %s\n",
1249 entry->ctx.prev_pid,
1250 entry->ctx.prev_prio,
1251 S, entry->type == TRACE_CTX ? "==>" : " +",
1252 entry->ctx.next_pid,
1253 entry->ctx.next_prio,
1257 trace_seq_printf(s, "# %ld %ld %ld\n",
1258 entry->special.arg1,
1259 entry->special.arg2,
1260 entry->special.arg3);
1263 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1265 trace_seq_puts(s, " <= ");
1266 seq_print_ip_sym(s, entry->stack.caller[i], sym_flags);
1268 trace_seq_puts(s, "\n");
1271 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1276 static int print_trace_fmt(struct trace_iterator *iter)
1278 struct trace_seq *s = &iter->seq;
1279 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1280 struct trace_entry *entry;
1281 unsigned long usec_rem;
1282 unsigned long long t;
1291 comm = trace_find_cmdline(iter->ent->pid);
1293 t = ns2usecs(entry->t);
1294 usec_rem = do_div(t, 1000000ULL);
1295 secs = (unsigned long)t;
1297 ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1300 ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1303 ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1307 switch (entry->type) {
1309 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1312 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1313 entry->fn.parent_ip) {
1314 ret = trace_seq_printf(s, " <-");
1317 ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1322 ret = trace_seq_printf(s, "\n");
1328 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1329 state_to_char[entry->ctx.prev_state] : 'X';
1330 T = entry->ctx.next_state < sizeof(state_to_char) ?
1331 state_to_char[entry->ctx.next_state] : 'X';
1332 ret = trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c\n",
1333 entry->ctx.prev_pid,
1334 entry->ctx.prev_prio,
1336 entry->type == TRACE_CTX ? "==>" : " +",
1337 entry->ctx.next_pid,
1338 entry->ctx.next_prio,
1344 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1345 entry->special.arg1,
1346 entry->special.arg2,
1347 entry->special.arg3);
1352 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1354 ret = trace_seq_puts(s, " <= ");
1358 ret = seq_print_ip_sym(s, entry->stack.caller[i],
1363 ret = trace_seq_puts(s, "\n");
1371 static int print_raw_fmt(struct trace_iterator *iter)
1373 struct trace_seq *s = &iter->seq;
1374 struct trace_entry *entry;
1380 ret = trace_seq_printf(s, "%d %d %llu ",
1381 entry->pid, iter->cpu, entry->t);
1385 switch (entry->type) {
1387 ret = trace_seq_printf(s, "%x %x\n",
1388 entry->fn.ip, entry->fn.parent_ip);
1394 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1395 state_to_char[entry->ctx.prev_state] : 'X';
1396 T = entry->ctx.next_state < sizeof(state_to_char) ?
1397 state_to_char[entry->ctx.next_state] : 'X';
1398 if (entry->type == TRACE_WAKE)
1400 ret = trace_seq_printf(s, "%d %d %c %d %d %c\n",
1401 entry->ctx.prev_pid,
1402 entry->ctx.prev_prio,
1404 entry->ctx.next_pid,
1405 entry->ctx.next_prio,
1412 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1413 entry->special.arg1,
1414 entry->special.arg2,
1415 entry->special.arg3);
1423 #define SEQ_PUT_FIELD_RET(s, x) \
1425 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
1429 #define SEQ_PUT_HEX_FIELD_RET(s, x) \
1431 if (!trace_seq_putmem_hex(s, &(x), sizeof(x))) \
1435 static int print_hex_fmt(struct trace_iterator *iter)
1437 struct trace_seq *s = &iter->seq;
1438 unsigned char newline = '\n';
1439 struct trace_entry *entry;
1444 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1445 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1446 SEQ_PUT_HEX_FIELD_RET(s, entry->t);
1448 switch (entry->type) {
1450 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.ip);
1451 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1455 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1456 state_to_char[entry->ctx.prev_state] : 'X';
1457 T = entry->ctx.next_state < sizeof(state_to_char) ?
1458 state_to_char[entry->ctx.next_state] : 'X';
1459 if (entry->type == TRACE_WAKE)
1461 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_pid);
1462 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_prio);
1463 SEQ_PUT_HEX_FIELD_RET(s, S);
1464 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_pid);
1465 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_prio);
1466 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1467 SEQ_PUT_HEX_FIELD_RET(s, T);
1471 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg1);
1472 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg2);
1473 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg3);
1476 SEQ_PUT_FIELD_RET(s, newline);
1481 static int print_bin_fmt(struct trace_iterator *iter)
1483 struct trace_seq *s = &iter->seq;
1484 struct trace_entry *entry;
1488 SEQ_PUT_FIELD_RET(s, entry->pid);
1489 SEQ_PUT_FIELD_RET(s, entry->cpu);
1490 SEQ_PUT_FIELD_RET(s, entry->t);
1492 switch (entry->type) {
1494 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1495 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1498 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1499 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1500 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1501 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1502 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1503 SEQ_PUT_FIELD_RET(s, entry->ctx.next_state);
1507 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1508 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1509 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1515 static int trace_empty(struct trace_iterator *iter)
1517 struct trace_array_cpu *data;
1520 for_each_possible_cpu(cpu) {
1521 data = iter->tr->data[cpu];
1523 if (head_page(data) && data->trace_idx &&
1524 (data->trace_tail != data->trace_head ||
1525 data->trace_tail_idx != data->trace_head_idx))
1531 static int print_trace_line(struct trace_iterator *iter)
1533 if (trace_flags & TRACE_ITER_BIN)
1534 return print_bin_fmt(iter);
1536 if (trace_flags & TRACE_ITER_HEX)
1537 return print_hex_fmt(iter);
1539 if (trace_flags & TRACE_ITER_RAW)
1540 return print_raw_fmt(iter);
1542 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1543 return print_lat_fmt(iter, iter->idx, iter->cpu);
1545 return print_trace_fmt(iter);
1548 static int s_show(struct seq_file *m, void *v)
1550 struct trace_iterator *iter = v;
1552 if (iter->ent == NULL) {
1554 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1557 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1558 /* print nothing if the buffers are empty */
1559 if (trace_empty(iter))
1561 print_trace_header(m, iter);
1562 if (!(trace_flags & TRACE_ITER_VERBOSE))
1563 print_lat_help_header(m);
1565 if (!(trace_flags & TRACE_ITER_VERBOSE))
1566 print_func_help_header(m);
1569 print_trace_line(iter);
1570 trace_print_seq(m, &iter->seq);
1576 static struct seq_operations tracer_seq_ops = {
1583 static struct trace_iterator *
1584 __tracing_open(struct inode *inode, struct file *file, int *ret)
1586 struct trace_iterator *iter;
1588 if (tracing_disabled) {
1593 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1599 mutex_lock(&trace_types_lock);
1600 if (current_trace && current_trace->print_max)
1603 iter->tr = inode->i_private;
1604 iter->trace = current_trace;
1607 /* TODO stop tracer */
1608 *ret = seq_open(file, &tracer_seq_ops);
1610 struct seq_file *m = file->private_data;
1613 /* stop the trace while dumping */
1617 if (iter->trace && iter->trace->open)
1618 iter->trace->open(iter);
1623 mutex_unlock(&trace_types_lock);
1629 int tracing_open_generic(struct inode *inode, struct file *filp)
1631 if (tracing_disabled)
1634 filp->private_data = inode->i_private;
1638 int tracing_release(struct inode *inode, struct file *file)
1640 struct seq_file *m = (struct seq_file *)file->private_data;
1641 struct trace_iterator *iter = m->private;
1643 mutex_lock(&trace_types_lock);
1644 if (iter->trace && iter->trace->close)
1645 iter->trace->close(iter);
1647 /* reenable tracing if it was previously enabled */
1650 mutex_unlock(&trace_types_lock);
1652 seq_release(inode, file);
1657 static int tracing_open(struct inode *inode, struct file *file)
1661 __tracing_open(inode, file, &ret);
1666 static int tracing_lt_open(struct inode *inode, struct file *file)
1668 struct trace_iterator *iter;
1671 iter = __tracing_open(inode, file, &ret);
1674 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1681 t_next(struct seq_file *m, void *v, loff_t *pos)
1683 struct tracer *t = m->private;
1695 static void *t_start(struct seq_file *m, loff_t *pos)
1697 struct tracer *t = m->private;
1700 mutex_lock(&trace_types_lock);
1701 for (; t && l < *pos; t = t_next(m, t, &l))
1707 static void t_stop(struct seq_file *m, void *p)
1709 mutex_unlock(&trace_types_lock);
1712 static int t_show(struct seq_file *m, void *v)
1714 struct tracer *t = v;
1719 seq_printf(m, "%s", t->name);
1728 static struct seq_operations show_traces_seq_ops = {
1735 static int show_traces_open(struct inode *inode, struct file *file)
1739 if (tracing_disabled)
1742 ret = seq_open(file, &show_traces_seq_ops);
1744 struct seq_file *m = file->private_data;
1745 m->private = trace_types;
1751 static struct file_operations tracing_fops = {
1752 .open = tracing_open,
1754 .llseek = seq_lseek,
1755 .release = tracing_release,
1758 static struct file_operations tracing_lt_fops = {
1759 .open = tracing_lt_open,
1761 .llseek = seq_lseek,
1762 .release = tracing_release,
1765 static struct file_operations show_traces_fops = {
1766 .open = show_traces_open,
1768 .release = seq_release,
1772 * Only trace on a CPU if the bitmask is set:
1774 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
1777 * When tracing/tracing_cpu_mask is modified then this holds
1778 * the new bitmask we are about to install:
1780 static cpumask_t tracing_cpumask_new;
1783 * The tracer itself will not take this lock, but still we want
1784 * to provide a consistent cpumask to user-space:
1786 static DEFINE_MUTEX(tracing_cpumask_update_lock);
1789 * Temporary storage for the character representation of the
1790 * CPU bitmask (and one more byte for the newline):
1792 static char mask_str[NR_CPUS + 1];
1795 tracing_cpumask_read(struct file *filp, char __user *ubuf,
1796 size_t count, loff_t *ppos)
1800 mutex_lock(&tracing_cpumask_update_lock);
1802 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
1803 if (count - len < 2) {
1807 len += sprintf(mask_str + len, "\n");
1808 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
1811 mutex_unlock(&tracing_cpumask_update_lock);
1817 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
1818 size_t count, loff_t *ppos)
1822 mutex_lock(&tracing_cpumask_update_lock);
1823 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
1827 spin_lock_irq(&ftrace_max_lock);
1828 for_each_possible_cpu(cpu) {
1830 * Increase/decrease the disabled counter if we are
1831 * about to flip a bit in the cpumask:
1833 if (cpu_isset(cpu, tracing_cpumask) &&
1834 !cpu_isset(cpu, tracing_cpumask_new)) {
1835 atomic_inc(&global_trace.data[cpu]->disabled);
1837 if (!cpu_isset(cpu, tracing_cpumask) &&
1838 cpu_isset(cpu, tracing_cpumask_new)) {
1839 atomic_dec(&global_trace.data[cpu]->disabled);
1842 spin_unlock_irq(&ftrace_max_lock);
1844 tracing_cpumask = tracing_cpumask_new;
1846 mutex_unlock(&tracing_cpumask_update_lock);
1851 mutex_unlock(&tracing_cpumask_update_lock);
1856 static struct file_operations tracing_cpumask_fops = {
1857 .open = tracing_open_generic,
1858 .read = tracing_cpumask_read,
1859 .write = tracing_cpumask_write,
1863 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
1864 size_t cnt, loff_t *ppos)
1871 /* calulate max size */
1872 for (i = 0; trace_options[i]; i++) {
1873 len += strlen(trace_options[i]);
1874 len += 3; /* "no" and space */
1877 /* +2 for \n and \0 */
1878 buf = kmalloc(len + 2, GFP_KERNEL);
1882 for (i = 0; trace_options[i]; i++) {
1883 if (trace_flags & (1 << i))
1884 r += sprintf(buf + r, "%s ", trace_options[i]);
1886 r += sprintf(buf + r, "no%s ", trace_options[i]);
1889 r += sprintf(buf + r, "\n");
1890 WARN_ON(r >= len + 2);
1892 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1900 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
1901 size_t cnt, loff_t *ppos)
1911 if (copy_from_user(&buf, ubuf, cnt))
1916 if (strncmp(buf, "no", 2) == 0) {
1921 for (i = 0; trace_options[i]; i++) {
1922 int len = strlen(trace_options[i]);
1924 if (strncmp(cmp, trace_options[i], len) == 0) {
1926 trace_flags &= ~(1 << i);
1928 trace_flags |= (1 << i);
1933 * If no option could be set, return an error:
1935 if (!trace_options[i])
1943 static struct file_operations tracing_iter_fops = {
1944 .open = tracing_open_generic,
1945 .read = tracing_iter_ctrl_read,
1946 .write = tracing_iter_ctrl_write,
1949 static const char readme_msg[] =
1950 "tracing mini-HOWTO:\n\n"
1952 "# mount -t debugfs nodev /debug\n\n"
1953 "# cat /debug/tracing/available_tracers\n"
1954 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1955 "# cat /debug/tracing/current_tracer\n"
1957 "# echo sched_switch > /debug/tracing/current_tracer\n"
1958 "# cat /debug/tracing/current_tracer\n"
1960 "# cat /debug/tracing/iter_ctrl\n"
1961 "noprint-parent nosym-offset nosym-addr noverbose\n"
1962 "# echo print-parent > /debug/tracing/iter_ctrl\n"
1963 "# echo 1 > /debug/tracing/tracing_enabled\n"
1964 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1965 "echo 0 > /debug/tracing/tracing_enabled\n"
1969 tracing_readme_read(struct file *filp, char __user *ubuf,
1970 size_t cnt, loff_t *ppos)
1972 return simple_read_from_buffer(ubuf, cnt, ppos,
1973 readme_msg, strlen(readme_msg));
1976 static struct file_operations tracing_readme_fops = {
1977 .open = tracing_open_generic,
1978 .read = tracing_readme_read,
1982 tracing_ctrl_read(struct file *filp, char __user *ubuf,
1983 size_t cnt, loff_t *ppos)
1985 struct trace_array *tr = filp->private_data;
1989 r = sprintf(buf, "%ld\n", tr->ctrl);
1990 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1994 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
1995 size_t cnt, loff_t *ppos)
1997 struct trace_array *tr = filp->private_data;
2004 if (copy_from_user(&buf, ubuf, cnt))
2009 val = simple_strtoul(buf, NULL, 10);
2013 mutex_lock(&trace_types_lock);
2014 if (tr->ctrl ^ val) {
2022 if (current_trace && current_trace->ctrl_update)
2023 current_trace->ctrl_update(tr);
2025 mutex_unlock(&trace_types_lock);
2033 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2034 size_t cnt, loff_t *ppos)
2036 char buf[max_tracer_type_len+2];
2039 mutex_lock(&trace_types_lock);
2041 r = sprintf(buf, "%s\n", current_trace->name);
2043 r = sprintf(buf, "\n");
2044 mutex_unlock(&trace_types_lock);
2046 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2050 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2051 size_t cnt, loff_t *ppos)
2053 struct trace_array *tr = &global_trace;
2055 char buf[max_tracer_type_len+1];
2058 if (cnt > max_tracer_type_len)
2059 cnt = max_tracer_type_len;
2061 if (copy_from_user(&buf, ubuf, cnt))
2066 /* strip ending whitespace. */
2067 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2070 mutex_lock(&trace_types_lock);
2071 for (t = trace_types; t; t = t->next) {
2072 if (strcmp(t->name, buf) == 0)
2075 if (!t || t == current_trace)
2078 if (current_trace && current_trace->reset)
2079 current_trace->reset(tr);
2086 mutex_unlock(&trace_types_lock);
2094 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2095 size_t cnt, loff_t *ppos)
2097 unsigned long *ptr = filp->private_data;
2101 r = snprintf(buf, 64, "%ld\n",
2102 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2105 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2109 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2110 size_t cnt, loff_t *ppos)
2112 long *ptr = filp->private_data;
2119 if (copy_from_user(&buf, ubuf, cnt))
2124 val = simple_strtoul(buf, NULL, 10);
2131 static atomic_t tracing_reader;
2133 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2135 struct trace_iterator *iter;
2137 if (tracing_disabled)
2140 /* We only allow for reader of the pipe */
2141 if (atomic_inc_return(&tracing_reader) != 1) {
2142 atomic_dec(&tracing_reader);
2146 /* create a buffer to store the information to pass to userspace */
2147 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2151 iter->tr = &global_trace;
2153 filp->private_data = iter;
2158 static int tracing_release_pipe(struct inode *inode, struct file *file)
2160 struct trace_iterator *iter = file->private_data;
2163 atomic_dec(&tracing_reader);
2169 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2171 struct trace_iterator *iter = filp->private_data;
2173 if (trace_flags & TRACE_ITER_BLOCK) {
2175 * Always select as readable when in blocking mode
2177 return POLLIN | POLLRDNORM;
2180 if (!trace_empty(iter))
2181 return POLLIN | POLLRDNORM;
2182 poll_wait(filp, &trace_wait, poll_table);
2183 if (!trace_empty(iter))
2184 return POLLIN | POLLRDNORM;
2194 tracing_read_pipe(struct file *filp, char __user *ubuf,
2195 size_t cnt, loff_t *ppos)
2197 struct trace_iterator *iter = filp->private_data;
2198 struct trace_array_cpu *data;
2199 static cpumask_t mask;
2201 unsigned long flags;
2202 #ifdef CONFIG_FTRACE
2210 /* return any leftover data */
2211 if (iter->seq.len > start) {
2212 len = iter->seq.len - start;
2215 ret = copy_to_user(ubuf, iter->seq.buffer + start, cnt);
2224 trace_seq_reset(&iter->seq);
2227 while (trace_empty(iter)) {
2228 if (!(trace_flags & TRACE_ITER_BLOCK))
2229 return -EWOULDBLOCK;
2231 * This is a make-shift waitqueue. The reason we don't use
2232 * an actual wait queue is because:
2233 * 1) we only ever have one waiter
2234 * 2) the tracing, traces all functions, we don't want
2235 * the overhead of calling wake_up and friends
2236 * (and tracing them too)
2237 * Anyway, this is really very primitive wakeup.
2239 set_current_state(TASK_INTERRUPTIBLE);
2240 iter->tr->waiter = current;
2242 /* sleep for one second, and try again. */
2243 schedule_timeout(HZ);
2245 iter->tr->waiter = NULL;
2247 if (signal_pending(current))
2251 * We block until we read something and tracing is disabled.
2252 * We still block if tracing is disabled, but we have never
2253 * read anything. This allows a user to cat this file, and
2254 * then enable tracing. But after we have read something,
2255 * we give an EOF when tracing is again disabled.
2257 * iter->pos will be 0 if we haven't read anything.
2259 if (!tracer_enabled && iter->pos)
2265 /* stop when tracing is finished */
2266 if (trace_empty(iter))
2269 if (cnt >= PAGE_SIZE)
2270 cnt = PAGE_SIZE - 1;
2272 memset(iter, 0, sizeof(*iter));
2273 iter->tr = &global_trace;
2277 * We need to stop all tracing on all CPUS to read the
2278 * the next buffer. This is a bit expensive, but is
2279 * not done often. We fill all what we can read,
2280 * and then release the locks again.
2284 local_irq_save(flags);
2285 #ifdef CONFIG_FTRACE
2286 ftrace_save = ftrace_enabled;
2290 for_each_possible_cpu(cpu) {
2291 data = iter->tr->data[cpu];
2293 if (!head_page(data) || !data->trace_idx)
2296 atomic_inc(&data->disabled);
2300 for_each_cpu_mask(cpu, mask) {
2301 data = iter->tr->data[cpu];
2302 spin_lock(&data->lock);
2305 while (find_next_entry_inc(iter) != NULL) {
2306 int len = iter->seq.len;
2308 ret = print_trace_line(iter);
2310 /* don't print partial lines */
2311 iter->seq.len = len;
2315 trace_consume(iter);
2317 if (iter->seq.len >= cnt)
2321 for_each_cpu_mask(cpu, mask) {
2322 data = iter->tr->data[cpu];
2323 spin_unlock(&data->lock);
2326 for_each_cpu_mask(cpu, mask) {
2327 data = iter->tr->data[cpu];
2328 atomic_dec(&data->disabled);
2330 #ifdef CONFIG_FTRACE
2331 ftrace_enabled = ftrace_save;
2333 local_irq_restore(flags);
2335 /* Now copy what we have to the user */
2336 read = iter->seq.len;
2340 ret = copy_to_user(ubuf, iter->seq.buffer, read);
2342 if (read < iter->seq.len)
2345 trace_seq_reset(&iter->seq);
2353 static struct file_operations tracing_max_lat_fops = {
2354 .open = tracing_open_generic,
2355 .read = tracing_max_lat_read,
2356 .write = tracing_max_lat_write,
2359 static struct file_operations tracing_ctrl_fops = {
2360 .open = tracing_open_generic,
2361 .read = tracing_ctrl_read,
2362 .write = tracing_ctrl_write,
2365 static struct file_operations set_tracer_fops = {
2366 .open = tracing_open_generic,
2367 .read = tracing_set_trace_read,
2368 .write = tracing_set_trace_write,
2371 static struct file_operations tracing_pipe_fops = {
2372 .open = tracing_open_pipe,
2373 .poll = tracing_poll_pipe,
2374 .read = tracing_read_pipe,
2375 .release = tracing_release_pipe,
2378 #ifdef CONFIG_DYNAMIC_FTRACE
2381 tracing_read_long(struct file *filp, char __user *ubuf,
2382 size_t cnt, loff_t *ppos)
2384 unsigned long *p = filp->private_data;
2388 r = sprintf(buf, "%ld\n", *p);
2390 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2393 static struct file_operations tracing_read_long_fops = {
2394 .open = tracing_open_generic,
2395 .read = tracing_read_long,
2399 static struct dentry *d_tracer;
2401 struct dentry *tracing_init_dentry(void)
2408 d_tracer = debugfs_create_dir("tracing", NULL);
2410 if (!d_tracer && !once) {
2412 pr_warning("Could not create debugfs directory 'tracing'\n");
2419 #ifdef CONFIG_FTRACE_SELFTEST
2420 /* Let selftest have access to static functions in this file */
2421 #include "trace_selftest.c"
2424 static __init void tracer_init_debugfs(void)
2426 struct dentry *d_tracer;
2427 struct dentry *entry;
2429 d_tracer = tracing_init_dentry();
2431 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2432 &global_trace, &tracing_ctrl_fops);
2434 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2436 entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2437 NULL, &tracing_iter_fops);
2439 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2441 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
2442 NULL, &tracing_cpumask_fops);
2444 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2446 entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2447 &global_trace, &tracing_lt_fops);
2449 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2451 entry = debugfs_create_file("trace", 0444, d_tracer,
2452 &global_trace, &tracing_fops);
2454 pr_warning("Could not create debugfs 'trace' entry\n");
2456 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2457 &global_trace, &show_traces_fops);
2459 pr_warning("Could not create debugfs 'trace' entry\n");
2461 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2462 &global_trace, &set_tracer_fops);
2464 pr_warning("Could not create debugfs 'trace' entry\n");
2466 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2467 &tracing_max_latency,
2468 &tracing_max_lat_fops);
2470 pr_warning("Could not create debugfs "
2471 "'tracing_max_latency' entry\n");
2473 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2474 &tracing_thresh, &tracing_max_lat_fops);
2476 pr_warning("Could not create debugfs "
2477 "'tracing_threash' entry\n");
2478 entry = debugfs_create_file("README", 0644, d_tracer,
2479 NULL, &tracing_readme_fops);
2481 pr_warning("Could not create debugfs 'README' entry\n");
2483 entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2484 NULL, &tracing_pipe_fops);
2486 pr_warning("Could not create debugfs "
2487 "'tracing_threash' entry\n");
2489 #ifdef CONFIG_DYNAMIC_FTRACE
2490 entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2491 &ftrace_update_tot_cnt,
2492 &tracing_read_long_fops);
2494 pr_warning("Could not create debugfs "
2495 "'dyn_ftrace_total_info' entry\n");
2499 /* dummy trace to disable tracing */
2500 static struct tracer no_tracer __read_mostly =
2505 static int trace_alloc_page(void)
2507 struct trace_array_cpu *data;
2508 struct page *page, *tmp;
2513 /* first allocate a page for each CPU */
2514 for_each_possible_cpu(i) {
2515 array = (void *)__get_free_page(GFP_KERNEL);
2516 if (array == NULL) {
2517 printk(KERN_ERR "tracer: failed to allocate page"
2518 "for trace buffer!\n");
2522 page = virt_to_page(array);
2523 list_add(&page->lru, &pages);
2525 /* Only allocate if we are actually using the max trace */
2526 #ifdef CONFIG_TRACER_MAX_TRACE
2527 array = (void *)__get_free_page(GFP_KERNEL);
2528 if (array == NULL) {
2529 printk(KERN_ERR "tracer: failed to allocate page"
2530 "for trace buffer!\n");
2533 page = virt_to_page(array);
2534 list_add(&page->lru, &pages);
2538 /* Now that we successfully allocate a page per CPU, add them */
2539 for_each_possible_cpu(i) {
2540 data = global_trace.data[i];
2541 spin_lock_init(&data->lock);
2542 lockdep_set_class(&data->lock, &data->lock_key);
2543 page = list_entry(pages.next, struct page, lru);
2544 list_del_init(&page->lru);
2545 list_add_tail(&page->lru, &data->trace_pages);
2548 #ifdef CONFIG_TRACER_MAX_TRACE
2549 data = max_tr.data[i];
2550 spin_lock_init(&data->lock);
2551 lockdep_set_class(&data->lock, &data->lock_key);
2552 page = list_entry(pages.next, struct page, lru);
2553 list_del_init(&page->lru);
2554 list_add_tail(&page->lru, &data->trace_pages);
2558 global_trace.entries += ENTRIES_PER_PAGE;
2563 list_for_each_entry_safe(page, tmp, &pages, lru) {
2564 list_del_init(&page->lru);
2570 __init static int tracer_alloc_buffers(void)
2572 struct trace_array_cpu *data;
2579 global_trace.ctrl = tracer_enabled;
2581 /* Allocate the first page for all buffers */
2582 for_each_possible_cpu(i) {
2583 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2584 max_tr.data[i] = &per_cpu(max_data, i);
2586 array = (void *)__get_free_page(GFP_KERNEL);
2587 if (array == NULL) {
2588 printk(KERN_ERR "tracer: failed to allocate page"
2589 "for trace buffer!\n");
2593 /* set the array to the list */
2594 INIT_LIST_HEAD(&data->trace_pages);
2595 page = virt_to_page(array);
2596 list_add(&page->lru, &data->trace_pages);
2597 /* use the LRU flag to differentiate the two buffers */
2600 /* Only allocate if we are actually using the max trace */
2601 #ifdef CONFIG_TRACER_MAX_TRACE
2602 array = (void *)__get_free_page(GFP_KERNEL);
2603 if (array == NULL) {
2604 printk(KERN_ERR "tracer: failed to allocate page"
2605 "for trace buffer!\n");
2609 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
2610 page = virt_to_page(array);
2611 list_add(&page->lru, &max_tr.data[i]->trace_pages);
2617 * Since we allocate by orders of pages, we may be able to
2620 global_trace.entries = ENTRIES_PER_PAGE;
2623 while (global_trace.entries < trace_nr_entries) {
2624 if (trace_alloc_page())
2628 max_tr.entries = global_trace.entries;
2630 pr_info("tracer: %d pages allocated for %ld",
2631 pages, trace_nr_entries);
2632 pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
2633 pr_info(" actual entries %ld\n", global_trace.entries);
2635 tracer_init_debugfs();
2637 trace_init_cmdlines();
2639 register_tracer(&no_tracer);
2640 current_trace = &no_tracer;
2642 /* All seems OK, enable tracing */
2643 tracing_disabled = 0;
2648 for (i-- ; i >= 0; i--) {
2649 struct page *page, *tmp;
2650 struct trace_array_cpu *data = global_trace.data[i];
2653 list_for_each_entry_safe(page, tmp,
2654 &data->trace_pages, lru) {
2655 list_del_init(&page->lru);
2660 #ifdef CONFIG_TRACER_MAX_TRACE
2661 data = max_tr.data[i];
2663 list_for_each_entry_safe(page, tmp,
2664 &data->trace_pages, lru) {
2665 list_del_init(&page->lru);
2673 fs_initcall(tracer_alloc_buffers);