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>
37 #include <linux/irqflags.h>
41 #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
43 unsigned long __read_mostly tracing_max_latency = (cycle_t)ULONG_MAX;
44 unsigned long __read_mostly tracing_thresh;
46 /* For tracers that don't implement custom flags */
47 static struct tracer_opt dummy_tracer_opt[] = {
51 static struct tracer_flags dummy_tracer_flags = {
53 .opts = dummy_tracer_opt
56 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
62 * Kill all tracing for good (never come back).
63 * It is initialized to 1 but will turn to zero if the initialization
64 * of the tracer is successful. But that is the only place that sets
67 int tracing_disabled = 1;
69 static DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
71 static inline void ftrace_disable_cpu(void)
74 local_inc(&__get_cpu_var(ftrace_cpu_disabled));
77 static inline void ftrace_enable_cpu(void)
79 local_dec(&__get_cpu_var(ftrace_cpu_disabled));
83 static cpumask_t __read_mostly tracing_buffer_mask;
85 #define for_each_tracing_cpu(cpu) \
86 for_each_cpu_mask(cpu, tracing_buffer_mask)
89 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
91 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
92 * is set, then ftrace_dump is called. This will output the contents
93 * of the ftrace buffers to the console. This is very useful for
94 * capturing traces that lead to crashes and outputing it to a
97 * It is default off, but you can enable it with either specifying
98 * "ftrace_dump_on_oops" in the kernel command line, or setting
99 * /proc/sys/kernel/ftrace_dump_on_oops to true.
101 int ftrace_dump_on_oops;
103 static int tracing_set_tracer(char *buf);
105 static int __init set_ftrace(char *str)
107 tracing_set_tracer(str);
110 __setup("ftrace", set_ftrace);
112 static int __init set_ftrace_dump_on_oops(char *str)
114 ftrace_dump_on_oops = 1;
117 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
120 ns2usecs(cycle_t nsec)
127 cycle_t ftrace_now(int cpu)
129 u64 ts = ring_buffer_time_stamp(cpu);
130 ring_buffer_normalize_time_stamp(cpu, &ts);
135 * The global_trace is the descriptor that holds the tracing
136 * buffers for the live tracing. For each CPU, it contains
137 * a link list of pages that will store trace entries. The
138 * page descriptor of the pages in the memory is used to hold
139 * the link list by linking the lru item in the page descriptor
140 * to each of the pages in the buffer per CPU.
142 * For each active CPU there is a data field that holds the
143 * pages for the buffer for that CPU. Each CPU has the same number
144 * of pages allocated for its buffer.
146 static struct trace_array global_trace;
148 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
151 * The max_tr is used to snapshot the global_trace when a maximum
152 * latency is reached. Some tracers will use this to store a maximum
153 * trace while it continues examining live traces.
155 * The buffers for the max_tr are set up the same as the global_trace.
156 * When a snapshot is taken, the link list of the max_tr is swapped
157 * with the link list of the global_trace and the buffers are reset for
158 * the global_trace so the tracing can continue.
160 static struct trace_array max_tr;
162 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
164 /* tracer_enabled is used to toggle activation of a tracer */
165 static int tracer_enabled = 1;
168 * tracing_is_enabled - return tracer_enabled status
170 * This function is used by other tracers to know the status
171 * of the tracer_enabled flag. Tracers may use this function
172 * to know if it should enable their features when starting
173 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
175 int tracing_is_enabled(void)
177 return tracer_enabled;
180 /* function tracing enabled */
181 int ftrace_function_enabled;
184 * trace_buf_size is the size in bytes that is allocated
185 * for a buffer. Note, the number of bytes is always rounded
188 * This number is purposely set to a low number of 16384.
189 * If the dump on oops happens, it will be much appreciated
190 * to not have to wait for all that output. Anyway this can be
191 * boot time and run time configurable.
193 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
195 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
197 /* trace_types holds a link list of available tracers. */
198 static struct tracer *trace_types __read_mostly;
200 /* current_trace points to the tracer that is currently active */
201 static struct tracer *current_trace __read_mostly;
204 * max_tracer_type_len is used to simplify the allocating of
205 * buffers to read userspace tracer names. We keep track of
206 * the longest tracer name registered.
208 static int max_tracer_type_len;
211 * trace_types_lock is used to protect the trace_types list.
212 * This lock is also used to keep user access serialized.
213 * Accesses from userspace will grab this lock while userspace
214 * activities happen inside the kernel.
216 static DEFINE_MUTEX(trace_types_lock);
218 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
219 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
221 /* trace_flags holds trace_options default values */
222 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
226 * trace_wake_up - wake up tasks waiting for trace input
228 * Simply wakes up any task that is blocked on the trace_wait
229 * queue. These is used with trace_poll for tasks polling the trace.
231 void trace_wake_up(void)
234 * The runqueue_is_locked() can fail, but this is the best we
237 if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
238 wake_up(&trace_wait);
241 static int __init set_buf_size(char *str)
243 unsigned long buf_size;
248 ret = strict_strtoul(str, 0, &buf_size);
249 /* nr_entries can not be zero */
250 if (ret < 0 || buf_size == 0)
252 trace_buf_size = buf_size;
255 __setup("trace_buf_size=", set_buf_size);
257 unsigned long nsecs_to_usecs(unsigned long nsecs)
262 /* These must match the bit postions in trace_iterator_flags */
263 static const char *trace_options[] = {
282 * ftrace_max_lock is used to protect the swapping of buffers
283 * when taking a max snapshot. The buffers themselves are
284 * protected by per_cpu spinlocks. But the action of the swap
285 * needs its own lock.
287 * This is defined as a raw_spinlock_t in order to help
288 * with performance when lockdep debugging is enabled.
290 static raw_spinlock_t ftrace_max_lock =
291 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
294 * Copy the new maximum trace into the separate maximum-trace
295 * structure. (this way the maximum trace is permanently saved,
296 * for later retrieval via /debugfs/tracing/latency_trace)
299 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
301 struct trace_array_cpu *data = tr->data[cpu];
304 max_tr.time_start = data->preempt_timestamp;
306 data = max_tr.data[cpu];
307 data->saved_latency = tracing_max_latency;
309 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
310 data->pid = tsk->pid;
311 data->uid = tsk->uid;
312 data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
313 data->policy = tsk->policy;
314 data->rt_priority = tsk->rt_priority;
316 /* record this tasks comm */
317 tracing_record_cmdline(current);
321 * trace_seq_printf - sequence printing of trace information
322 * @s: trace sequence descriptor
323 * @fmt: printf format string
325 * The tracer may use either sequence operations or its own
326 * copy to user routines. To simplify formating of a trace
327 * trace_seq_printf is used to store strings into a special
328 * buffer (@s). Then the output may be either used by
329 * the sequencer or pulled into another buffer.
332 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
334 int len = (PAGE_SIZE - 1) - s->len;
342 ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
345 /* If we can't write it all, don't bother writing anything */
355 * trace_seq_puts - trace sequence printing of simple string
356 * @s: trace sequence descriptor
357 * @str: simple string to record
359 * The tracer may use either the sequence operations or its own
360 * copy to user routines. This function records a simple string
361 * into a special buffer (@s) for later retrieval by a sequencer
362 * or other mechanism.
365 trace_seq_puts(struct trace_seq *s, const char *str)
367 int len = strlen(str);
369 if (len > ((PAGE_SIZE - 1) - s->len))
372 memcpy(s->buffer + s->len, str, len);
379 trace_seq_putc(struct trace_seq *s, unsigned char c)
381 if (s->len >= (PAGE_SIZE - 1))
384 s->buffer[s->len++] = c;
390 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
392 if (len > ((PAGE_SIZE - 1) - s->len))
395 memcpy(s->buffer + s->len, mem, len);
401 #define MAX_MEMHEX_BYTES 8
402 #define HEX_CHARS (MAX_MEMHEX_BYTES*2 + 1)
405 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
407 unsigned char hex[HEX_CHARS];
408 unsigned char *data = mem;
412 for (i = 0, j = 0; i < len; i++) {
414 for (i = len-1, j = 0; i >= 0; i--) {
416 hex[j++] = hex_asc_hi(data[i]);
417 hex[j++] = hex_asc_lo(data[i]);
421 return trace_seq_putmem(s, hex, j);
425 trace_seq_reset(struct trace_seq *s)
431 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
436 if (s->len <= s->readpos)
439 len = s->len - s->readpos;
442 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
451 trace_print_seq(struct seq_file *m, struct trace_seq *s)
453 int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
456 seq_puts(m, s->buffer);
462 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
464 * @tsk: the task with the latency
465 * @cpu: The cpu that initiated the trace.
467 * Flip the buffers between the @tr and the max_tr and record information
468 * about which task was the cause of this latency.
471 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
473 struct ring_buffer *buf = tr->buffer;
475 WARN_ON_ONCE(!irqs_disabled());
476 __raw_spin_lock(&ftrace_max_lock);
478 tr->buffer = max_tr.buffer;
481 ftrace_disable_cpu();
482 ring_buffer_reset(tr->buffer);
485 __update_max_tr(tr, tsk, cpu);
486 __raw_spin_unlock(&ftrace_max_lock);
490 * update_max_tr_single - only copy one trace over, and reset the rest
492 * @tsk - task with the latency
493 * @cpu - the cpu of the buffer to copy.
495 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
498 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
502 WARN_ON_ONCE(!irqs_disabled());
503 __raw_spin_lock(&ftrace_max_lock);
505 ftrace_disable_cpu();
507 ring_buffer_reset(max_tr.buffer);
508 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
514 __update_max_tr(tr, tsk, cpu);
515 __raw_spin_unlock(&ftrace_max_lock);
519 * register_tracer - register a tracer with the ftrace system.
520 * @type - the plugin for the tracer
522 * Register a new plugin tracer.
524 int register_tracer(struct tracer *type)
531 pr_info("Tracer must have a name\n");
535 mutex_lock(&trace_types_lock);
536 for (t = trace_types; t; t = t->next) {
537 if (strcmp(type->name, t->name) == 0) {
539 pr_info("Trace %s already registered\n",
547 type->set_flag = &dummy_set_flag;
549 type->flags = &dummy_tracer_flags;
551 if (!type->flags->opts)
552 type->flags->opts = dummy_tracer_opt;
554 #ifdef CONFIG_FTRACE_STARTUP_TEST
555 if (type->selftest) {
556 struct tracer *saved_tracer = current_trace;
557 struct trace_array *tr = &global_trace;
560 * Run a selftest on this tracer.
561 * Here we reset the trace buffer, and set the current
562 * tracer to be this tracer. The tracer can then run some
563 * internal tracing to verify that everything is in order.
564 * If we fail, we do not register this tracer.
566 for_each_tracing_cpu(i) {
567 tracing_reset(tr, i);
569 current_trace = type;
570 /* the test is responsible for initializing and enabling */
571 pr_info("Testing tracer %s: ", type->name);
572 ret = type->selftest(type, tr);
573 /* the test is responsible for resetting too */
574 current_trace = saved_tracer;
576 printk(KERN_CONT "FAILED!\n");
579 /* Only reset on passing, to avoid touching corrupted buffers */
580 for_each_tracing_cpu(i) {
581 tracing_reset(tr, i);
583 printk(KERN_CONT "PASSED\n");
587 type->next = trace_types;
589 len = strlen(type->name);
590 if (len > max_tracer_type_len)
591 max_tracer_type_len = len;
594 mutex_unlock(&trace_types_lock);
599 void unregister_tracer(struct tracer *type)
604 mutex_lock(&trace_types_lock);
605 for (t = &trace_types; *t; t = &(*t)->next) {
609 pr_info("Trace %s not registered\n", type->name);
614 if (strlen(type->name) != max_tracer_type_len)
617 max_tracer_type_len = 0;
618 for (t = &trace_types; *t; t = &(*t)->next) {
619 len = strlen((*t)->name);
620 if (len > max_tracer_type_len)
621 max_tracer_type_len = len;
624 mutex_unlock(&trace_types_lock);
627 void tracing_reset(struct trace_array *tr, int cpu)
629 ftrace_disable_cpu();
630 ring_buffer_reset_cpu(tr->buffer, cpu);
634 #define SAVED_CMDLINES 128
635 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
636 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
637 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
638 static int cmdline_idx;
639 static DEFINE_SPINLOCK(trace_cmdline_lock);
641 /* temporary disable recording */
642 atomic_t trace_record_cmdline_disabled __read_mostly;
644 static void trace_init_cmdlines(void)
646 memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
647 memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
651 static int trace_stop_count;
652 static DEFINE_SPINLOCK(tracing_start_lock);
655 * tracing_start - quick start of the tracer
657 * If tracing is enabled but was stopped by tracing_stop,
658 * this will start the tracer back up.
660 void tracing_start(void)
662 struct ring_buffer *buffer;
665 if (tracing_disabled)
668 spin_lock_irqsave(&tracing_start_lock, flags);
669 if (--trace_stop_count)
672 if (trace_stop_count < 0) {
673 /* Someone screwed up their debugging */
675 trace_stop_count = 0;
680 buffer = global_trace.buffer;
682 ring_buffer_record_enable(buffer);
684 buffer = max_tr.buffer;
686 ring_buffer_record_enable(buffer);
690 spin_unlock_irqrestore(&tracing_start_lock, flags);
694 * tracing_stop - quick stop of the tracer
696 * Light weight way to stop tracing. Use in conjunction with
699 void tracing_stop(void)
701 struct ring_buffer *buffer;
705 spin_lock_irqsave(&tracing_start_lock, flags);
706 if (trace_stop_count++)
709 buffer = global_trace.buffer;
711 ring_buffer_record_disable(buffer);
713 buffer = max_tr.buffer;
715 ring_buffer_record_disable(buffer);
718 spin_unlock_irqrestore(&tracing_start_lock, flags);
721 void trace_stop_cmdline_recording(void);
723 static void trace_save_cmdline(struct task_struct *tsk)
728 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
732 * It's not the end of the world if we don't get
733 * the lock, but we also don't want to spin
734 * nor do we want to disable interrupts,
735 * so if we miss here, then better luck next time.
737 if (!spin_trylock(&trace_cmdline_lock))
740 idx = map_pid_to_cmdline[tsk->pid];
741 if (idx >= SAVED_CMDLINES) {
742 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
744 map = map_cmdline_to_pid[idx];
745 if (map <= PID_MAX_DEFAULT)
746 map_pid_to_cmdline[map] = (unsigned)-1;
748 map_pid_to_cmdline[tsk->pid] = idx;
753 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
755 spin_unlock(&trace_cmdline_lock);
758 static char *trace_find_cmdline(int pid)
760 char *cmdline = "<...>";
766 if (pid > PID_MAX_DEFAULT)
769 map = map_pid_to_cmdline[pid];
770 if (map >= SAVED_CMDLINES)
773 cmdline = saved_cmdlines[map];
779 void tracing_record_cmdline(struct task_struct *tsk)
781 if (atomic_read(&trace_record_cmdline_disabled))
784 trace_save_cmdline(tsk);
788 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
791 struct task_struct *tsk = current;
793 entry->preempt_count = pc & 0xff;
794 entry->pid = (tsk) ? tsk->pid : 0;
796 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
797 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
799 TRACE_FLAG_IRQS_NOSUPPORT |
801 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
802 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
803 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
807 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
808 unsigned long ip, unsigned long parent_ip, unsigned long flags,
811 struct ring_buffer_event *event;
812 struct ftrace_entry *entry;
813 unsigned long irq_flags;
815 /* If we are reading the ring buffer, don't trace */
816 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
819 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
823 entry = ring_buffer_event_data(event);
824 tracing_generic_entry_update(&entry->ent, flags, pc);
825 entry->ent.type = TRACE_FN;
827 entry->parent_ip = parent_ip;
828 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
831 #ifdef CONFIG_FUNCTION_RET_TRACER
832 static void __trace_function_return(struct trace_array *tr,
833 struct trace_array_cpu *data,
834 struct ftrace_retfunc *trace,
838 struct ring_buffer_event *event;
839 struct ftrace_ret_entry *entry;
840 unsigned long irq_flags;
842 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
845 event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry),
849 entry = ring_buffer_event_data(event);
850 tracing_generic_entry_update(&entry->ent, flags, pc);
851 entry->ent.type = TRACE_FN_RET;
852 entry->ip = trace->func;
853 entry->parent_ip = trace->ret;
854 entry->rettime = trace->rettime;
855 entry->calltime = trace->calltime;
856 ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
861 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
862 unsigned long ip, unsigned long parent_ip, unsigned long flags,
865 if (likely(!atomic_read(&data->disabled)))
866 trace_function(tr, data, ip, parent_ip, flags, pc);
869 static void ftrace_trace_stack(struct trace_array *tr,
870 struct trace_array_cpu *data,
874 #ifdef CONFIG_STACKTRACE
875 struct ring_buffer_event *event;
876 struct stack_entry *entry;
877 struct stack_trace trace;
878 unsigned long irq_flags;
880 if (!(trace_flags & TRACE_ITER_STACKTRACE))
883 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
887 entry = ring_buffer_event_data(event);
888 tracing_generic_entry_update(&entry->ent, flags, pc);
889 entry->ent.type = TRACE_STACK;
891 memset(&entry->caller, 0, sizeof(entry->caller));
893 trace.nr_entries = 0;
894 trace.max_entries = FTRACE_STACK_ENTRIES;
896 trace.entries = entry->caller;
898 save_stack_trace(&trace);
899 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
903 void __trace_stack(struct trace_array *tr,
904 struct trace_array_cpu *data,
908 ftrace_trace_stack(tr, data, flags, skip, preempt_count());
912 ftrace_trace_special(void *__tr, void *__data,
913 unsigned long arg1, unsigned long arg2, unsigned long arg3,
916 struct ring_buffer_event *event;
917 struct trace_array_cpu *data = __data;
918 struct trace_array *tr = __tr;
919 struct special_entry *entry;
920 unsigned long irq_flags;
922 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
926 entry = ring_buffer_event_data(event);
927 tracing_generic_entry_update(&entry->ent, 0, pc);
928 entry->ent.type = TRACE_SPECIAL;
932 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
933 ftrace_trace_stack(tr, data, irq_flags, 4, pc);
939 __trace_special(void *__tr, void *__data,
940 unsigned long arg1, unsigned long arg2, unsigned long arg3)
942 ftrace_trace_special(__tr, __data, arg1, arg2, arg3, preempt_count());
946 tracing_sched_switch_trace(struct trace_array *tr,
947 struct trace_array_cpu *data,
948 struct task_struct *prev,
949 struct task_struct *next,
950 unsigned long flags, int pc)
952 struct ring_buffer_event *event;
953 struct ctx_switch_entry *entry;
954 unsigned long irq_flags;
956 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
960 entry = ring_buffer_event_data(event);
961 tracing_generic_entry_update(&entry->ent, flags, pc);
962 entry->ent.type = TRACE_CTX;
963 entry->prev_pid = prev->pid;
964 entry->prev_prio = prev->prio;
965 entry->prev_state = prev->state;
966 entry->next_pid = next->pid;
967 entry->next_prio = next->prio;
968 entry->next_state = next->state;
969 entry->next_cpu = task_cpu(next);
970 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
971 ftrace_trace_stack(tr, data, flags, 5, pc);
975 tracing_sched_wakeup_trace(struct trace_array *tr,
976 struct trace_array_cpu *data,
977 struct task_struct *wakee,
978 struct task_struct *curr,
979 unsigned long flags, int pc)
981 struct ring_buffer_event *event;
982 struct ctx_switch_entry *entry;
983 unsigned long irq_flags;
985 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
989 entry = ring_buffer_event_data(event);
990 tracing_generic_entry_update(&entry->ent, flags, pc);
991 entry->ent.type = TRACE_WAKE;
992 entry->prev_pid = curr->pid;
993 entry->prev_prio = curr->prio;
994 entry->prev_state = curr->state;
995 entry->next_pid = wakee->pid;
996 entry->next_prio = wakee->prio;
997 entry->next_state = wakee->state;
998 entry->next_cpu = task_cpu(wakee);
999 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1000 ftrace_trace_stack(tr, data, flags, 6, pc);
1006 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1008 struct trace_array *tr = &global_trace;
1009 struct trace_array_cpu *data;
1010 unsigned long flags;
1014 if (tracing_disabled)
1017 pc = preempt_count();
1018 local_irq_save(flags);
1019 cpu = raw_smp_processor_id();
1020 data = tr->data[cpu];
1022 if (likely(atomic_inc_return(&data->disabled) == 1))
1023 ftrace_trace_special(tr, data, arg1, arg2, arg3, pc);
1025 atomic_dec(&data->disabled);
1026 local_irq_restore(flags);
1029 #ifdef CONFIG_FUNCTION_TRACER
1031 function_trace_call_preempt_only(unsigned long ip, unsigned long parent_ip)
1033 struct trace_array *tr = &global_trace;
1034 struct trace_array_cpu *data;
1035 unsigned long flags;
1040 if (unlikely(!ftrace_function_enabled))
1043 pc = preempt_count();
1044 resched = ftrace_preempt_disable();
1045 local_save_flags(flags);
1046 cpu = raw_smp_processor_id();
1047 data = tr->data[cpu];
1048 disabled = atomic_inc_return(&data->disabled);
1050 if (likely(disabled == 1))
1051 trace_function(tr, data, ip, parent_ip, flags, pc);
1053 atomic_dec(&data->disabled);
1054 ftrace_preempt_enable(resched);
1058 function_trace_call(unsigned long ip, unsigned long parent_ip)
1060 struct trace_array *tr = &global_trace;
1061 struct trace_array_cpu *data;
1062 unsigned long flags;
1067 if (unlikely(!ftrace_function_enabled))
1071 * Need to use raw, since this must be called before the
1072 * recursive protection is performed.
1074 local_irq_save(flags);
1075 cpu = raw_smp_processor_id();
1076 data = tr->data[cpu];
1077 disabled = atomic_inc_return(&data->disabled);
1079 if (likely(disabled == 1)) {
1080 pc = preempt_count();
1081 trace_function(tr, data, ip, parent_ip, flags, pc);
1084 atomic_dec(&data->disabled);
1085 local_irq_restore(flags);
1088 #ifdef CONFIG_FUNCTION_RET_TRACER
1089 void trace_function_return(struct ftrace_retfunc *trace)
1091 struct trace_array *tr = &global_trace;
1092 struct trace_array_cpu *data;
1093 unsigned long flags;
1098 raw_local_irq_save(flags);
1099 cpu = raw_smp_processor_id();
1100 data = tr->data[cpu];
1101 disabled = atomic_inc_return(&data->disabled);
1102 if (likely(disabled == 1)) {
1103 pc = preempt_count();
1104 __trace_function_return(tr, data, trace, flags, pc);
1106 atomic_dec(&data->disabled);
1107 raw_local_irq_restore(flags);
1109 #endif /* CONFIG_FUNCTION_RET_TRACER */
1111 static struct ftrace_ops trace_ops __read_mostly =
1113 .func = function_trace_call,
1116 void tracing_start_function_trace(void)
1118 ftrace_function_enabled = 0;
1120 if (trace_flags & TRACE_ITER_PREEMPTONLY)
1121 trace_ops.func = function_trace_call_preempt_only;
1123 trace_ops.func = function_trace_call;
1125 register_ftrace_function(&trace_ops);
1126 ftrace_function_enabled = 1;
1129 void tracing_stop_function_trace(void)
1131 ftrace_function_enabled = 0;
1132 unregister_ftrace_function(&trace_ops);
1136 enum trace_file_type {
1137 TRACE_FILE_LAT_FMT = 1,
1138 TRACE_FILE_ANNOTATE = 2,
1141 static void trace_iterator_increment(struct trace_iterator *iter, int cpu)
1143 /* Don't allow ftrace to trace into the ring buffers */
1144 ftrace_disable_cpu();
1147 if (iter->buffer_iter[iter->cpu])
1148 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1150 ftrace_enable_cpu();
1153 static struct trace_entry *
1154 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1156 struct ring_buffer_event *event;
1157 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1159 /* Don't allow ftrace to trace into the ring buffers */
1160 ftrace_disable_cpu();
1163 event = ring_buffer_iter_peek(buf_iter, ts);
1165 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1167 ftrace_enable_cpu();
1169 return event ? ring_buffer_event_data(event) : NULL;
1172 static struct trace_entry *
1173 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1175 struct ring_buffer *buffer = iter->tr->buffer;
1176 struct trace_entry *ent, *next = NULL;
1177 u64 next_ts = 0, ts;
1181 for_each_tracing_cpu(cpu) {
1183 if (ring_buffer_empty_cpu(buffer, cpu))
1186 ent = peek_next_entry(iter, cpu, &ts);
1189 * Pick the entry with the smallest timestamp:
1191 if (ent && (!next || ts < next_ts)) {
1199 *ent_cpu = next_cpu;
1207 /* Find the next real entry, without updating the iterator itself */
1208 static struct trace_entry *
1209 find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1211 return __find_next_entry(iter, ent_cpu, ent_ts);
1214 /* Find the next real entry, and increment the iterator to the next entry */
1215 static void *find_next_entry_inc(struct trace_iterator *iter)
1217 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1220 trace_iterator_increment(iter, iter->cpu);
1222 return iter->ent ? iter : NULL;
1225 static void trace_consume(struct trace_iterator *iter)
1227 /* Don't allow ftrace to trace into the ring buffers */
1228 ftrace_disable_cpu();
1229 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1230 ftrace_enable_cpu();
1233 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1235 struct trace_iterator *iter = m->private;
1241 /* can't go backwards */
1246 ent = find_next_entry_inc(iter);
1250 while (ent && iter->idx < i)
1251 ent = find_next_entry_inc(iter);
1258 static void *s_start(struct seq_file *m, loff_t *pos)
1260 struct trace_iterator *iter = m->private;
1265 mutex_lock(&trace_types_lock);
1267 if (!current_trace || current_trace != iter->trace) {
1268 mutex_unlock(&trace_types_lock);
1272 atomic_inc(&trace_record_cmdline_disabled);
1274 if (*pos != iter->pos) {
1279 ftrace_disable_cpu();
1281 for_each_tracing_cpu(cpu) {
1282 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1285 ftrace_enable_cpu();
1287 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1292 p = s_next(m, p, &l);
1298 static void s_stop(struct seq_file *m, void *p)
1300 atomic_dec(&trace_record_cmdline_disabled);
1301 mutex_unlock(&trace_types_lock);
1304 #ifdef CONFIG_KRETPROBES
1305 static inline const char *kretprobed(const char *name)
1307 static const char tramp_name[] = "kretprobe_trampoline";
1308 int size = sizeof(tramp_name);
1310 if (strncmp(tramp_name, name, size) == 0)
1311 return "[unknown/kretprobe'd]";
1315 static inline const char *kretprobed(const char *name)
1319 #endif /* CONFIG_KRETPROBES */
1322 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1324 #ifdef CONFIG_KALLSYMS
1325 char str[KSYM_SYMBOL_LEN];
1328 kallsyms_lookup(address, NULL, NULL, NULL, str);
1330 name = kretprobed(str);
1332 return trace_seq_printf(s, fmt, name);
1338 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1339 unsigned long address)
1341 #ifdef CONFIG_KALLSYMS
1342 char str[KSYM_SYMBOL_LEN];
1345 sprint_symbol(str, address);
1346 name = kretprobed(str);
1348 return trace_seq_printf(s, fmt, name);
1353 #ifndef CONFIG_64BIT
1354 # define IP_FMT "%08lx"
1356 # define IP_FMT "%016lx"
1360 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1365 return trace_seq_printf(s, "0");
1367 if (sym_flags & TRACE_ITER_SYM_OFFSET)
1368 ret = seq_print_sym_offset(s, "%s", ip);
1370 ret = seq_print_sym_short(s, "%s", ip);
1375 if (sym_flags & TRACE_ITER_SYM_ADDR)
1376 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1380 static void print_lat_help_header(struct seq_file *m)
1382 seq_puts(m, "# _------=> CPU# \n");
1383 seq_puts(m, "# / _-----=> irqs-off \n");
1384 seq_puts(m, "# | / _----=> need-resched \n");
1385 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1386 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1387 seq_puts(m, "# |||| / \n");
1388 seq_puts(m, "# ||||| delay \n");
1389 seq_puts(m, "# cmd pid ||||| time | caller \n");
1390 seq_puts(m, "# \\ / ||||| \\ | / \n");
1393 static void print_func_help_header(struct seq_file *m)
1395 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1396 seq_puts(m, "# | | | | |\n");
1401 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1403 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1404 struct trace_array *tr = iter->tr;
1405 struct trace_array_cpu *data = tr->data[tr->cpu];
1406 struct tracer *type = current_trace;
1407 unsigned long total;
1408 unsigned long entries;
1409 const char *name = "preemption";
1414 entries = ring_buffer_entries(iter->tr->buffer);
1416 ring_buffer_overruns(iter->tr->buffer);
1418 seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1420 seq_puts(m, "-----------------------------------"
1421 "---------------------------------\n");
1422 seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1423 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1424 nsecs_to_usecs(data->saved_latency),
1428 #if defined(CONFIG_PREEMPT_NONE)
1430 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1432 #elif defined(CONFIG_PREEMPT)
1437 /* These are reserved for later use */
1440 seq_printf(m, " #P:%d)\n", num_online_cpus());
1444 seq_puts(m, " -----------------\n");
1445 seq_printf(m, " | task: %.16s-%d "
1446 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1447 data->comm, data->pid, data->uid, data->nice,
1448 data->policy, data->rt_priority);
1449 seq_puts(m, " -----------------\n");
1451 if (data->critical_start) {
1452 seq_puts(m, " => started at: ");
1453 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1454 trace_print_seq(m, &iter->seq);
1455 seq_puts(m, "\n => ended at: ");
1456 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1457 trace_print_seq(m, &iter->seq);
1465 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1467 int hardirq, softirq;
1470 comm = trace_find_cmdline(entry->pid);
1472 trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1473 trace_seq_printf(s, "%3d", cpu);
1474 trace_seq_printf(s, "%c%c",
1475 (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' :
1476 (entry->flags & TRACE_FLAG_IRQS_NOSUPPORT) ? 'X' : '.',
1477 ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1479 hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1480 softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1481 if (hardirq && softirq) {
1482 trace_seq_putc(s, 'H');
1485 trace_seq_putc(s, 'h');
1488 trace_seq_putc(s, 's');
1490 trace_seq_putc(s, '.');
1494 if (entry->preempt_count)
1495 trace_seq_printf(s, "%x", entry->preempt_count);
1497 trace_seq_puts(s, ".");
1500 unsigned long preempt_mark_thresh = 100;
1503 lat_print_timestamp(struct trace_seq *s, u64 abs_usecs,
1504 unsigned long rel_usecs)
1506 trace_seq_printf(s, " %4lldus", abs_usecs);
1507 if (rel_usecs > preempt_mark_thresh)
1508 trace_seq_puts(s, "!: ");
1509 else if (rel_usecs > 1)
1510 trace_seq_puts(s, "+: ");
1512 trace_seq_puts(s, " : ");
1515 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1518 * The message is supposed to contain an ending newline.
1519 * If the printing stops prematurely, try to add a newline of our own.
1521 void trace_seq_print_cont(struct trace_seq *s, struct trace_iterator *iter)
1523 struct trace_entry *ent;
1524 struct trace_field_cont *cont;
1527 ent = peek_next_entry(iter, iter->cpu, NULL);
1528 if (!ent || ent->type != TRACE_CONT) {
1529 trace_seq_putc(s, '\n');
1534 cont = (struct trace_field_cont *)ent;
1536 ok = (trace_seq_printf(s, "%s", cont->buf) > 0);
1538 ftrace_disable_cpu();
1540 if (iter->buffer_iter[iter->cpu])
1541 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1543 ring_buffer_consume(iter->tr->buffer, iter->cpu, NULL);
1545 ftrace_enable_cpu();
1547 ent = peek_next_entry(iter, iter->cpu, NULL);
1548 } while (ent && ent->type == TRACE_CONT);
1551 trace_seq_putc(s, '\n');
1554 static void test_cpu_buff_start(struct trace_iterator *iter)
1556 struct trace_seq *s = &iter->seq;
1558 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1561 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1564 if (cpu_isset(iter->cpu, iter->started))
1567 cpu_set(iter->cpu, iter->started);
1568 trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1571 static enum print_line_t
1572 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1574 struct trace_seq *s = &iter->seq;
1575 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1576 struct trace_entry *next_entry;
1577 unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1578 struct trace_entry *entry = iter->ent;
1579 unsigned long abs_usecs;
1580 unsigned long rel_usecs;
1587 if (entry->type == TRACE_CONT)
1588 return TRACE_TYPE_HANDLED;
1590 test_cpu_buff_start(iter);
1592 next_entry = find_next_entry(iter, NULL, &next_ts);
1595 rel_usecs = ns2usecs(next_ts - iter->ts);
1596 abs_usecs = ns2usecs(iter->ts - iter->tr->time_start);
1599 comm = trace_find_cmdline(entry->pid);
1600 trace_seq_printf(s, "%16s %5d %3d %d %08x %08x [%08lx]"
1601 " %ld.%03ldms (+%ld.%03ldms): ",
1603 entry->pid, cpu, entry->flags,
1604 entry->preempt_count, trace_idx,
1607 abs_usecs % 1000, rel_usecs/1000,
1610 lat_print_generic(s, entry, cpu);
1611 lat_print_timestamp(s, abs_usecs, rel_usecs);
1613 switch (entry->type) {
1615 struct ftrace_entry *field;
1617 trace_assign_type(field, entry);
1619 seq_print_ip_sym(s, field->ip, sym_flags);
1620 trace_seq_puts(s, " (");
1621 seq_print_ip_sym(s, field->parent_ip, sym_flags);
1622 trace_seq_puts(s, ")\n");
1627 struct ctx_switch_entry *field;
1629 trace_assign_type(field, entry);
1631 T = field->next_state < sizeof(state_to_char) ?
1632 state_to_char[field->next_state] : 'X';
1634 state = field->prev_state ?
1635 __ffs(field->prev_state) + 1 : 0;
1636 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1637 comm = trace_find_cmdline(field->next_pid);
1638 trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c %s\n",
1641 S, entry->type == TRACE_CTX ? "==>" : " +",
1648 case TRACE_SPECIAL: {
1649 struct special_entry *field;
1651 trace_assign_type(field, entry);
1653 trace_seq_printf(s, "# %ld %ld %ld\n",
1660 struct stack_entry *field;
1662 trace_assign_type(field, entry);
1664 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1666 trace_seq_puts(s, " <= ");
1667 seq_print_ip_sym(s, field->caller[i], sym_flags);
1669 trace_seq_puts(s, "\n");
1673 struct print_entry *field;
1675 trace_assign_type(field, entry);
1677 seq_print_ip_sym(s, field->ip, sym_flags);
1678 trace_seq_printf(s, ": %s", field->buf);
1679 if (entry->flags & TRACE_FLAG_CONT)
1680 trace_seq_print_cont(s, iter);
1683 case TRACE_BRANCH: {
1684 struct trace_branch *field;
1686 trace_assign_type(field, entry);
1688 trace_seq_printf(s, "[%s] %s:%s:%d\n",
1689 field->correct ? " ok " : " MISS ",
1696 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1698 return TRACE_TYPE_HANDLED;
1701 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1703 struct trace_seq *s = &iter->seq;
1704 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1705 struct trace_entry *entry;
1706 unsigned long usec_rem;
1707 unsigned long long t;
1716 if (entry->type == TRACE_CONT)
1717 return TRACE_TYPE_HANDLED;
1719 test_cpu_buff_start(iter);
1721 comm = trace_find_cmdline(iter->ent->pid);
1723 t = ns2usecs(iter->ts);
1724 usec_rem = do_div(t, 1000000ULL);
1725 secs = (unsigned long)t;
1727 ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1729 return TRACE_TYPE_PARTIAL_LINE;
1730 ret = trace_seq_printf(s, "[%03d] ", iter->cpu);
1732 return TRACE_TYPE_PARTIAL_LINE;
1733 ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1735 return TRACE_TYPE_PARTIAL_LINE;
1737 switch (entry->type) {
1739 struct ftrace_entry *field;
1741 trace_assign_type(field, entry);
1743 ret = seq_print_ip_sym(s, field->ip, sym_flags);
1745 return TRACE_TYPE_PARTIAL_LINE;
1746 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1748 ret = trace_seq_printf(s, " <-");
1750 return TRACE_TYPE_PARTIAL_LINE;
1751 ret = seq_print_ip_sym(s,
1755 return TRACE_TYPE_PARTIAL_LINE;
1757 ret = trace_seq_printf(s, "\n");
1759 return TRACE_TYPE_PARTIAL_LINE;
1764 struct ctx_switch_entry *field;
1766 trace_assign_type(field, entry);
1768 S = field->prev_state < sizeof(state_to_char) ?
1769 state_to_char[field->prev_state] : 'X';
1770 T = field->next_state < sizeof(state_to_char) ?
1771 state_to_char[field->next_state] : 'X';
1772 ret = trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c\n",
1776 entry->type == TRACE_CTX ? "==>" : " +",
1782 return TRACE_TYPE_PARTIAL_LINE;
1785 case TRACE_SPECIAL: {
1786 struct special_entry *field;
1788 trace_assign_type(field, entry);
1790 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1795 return TRACE_TYPE_PARTIAL_LINE;
1799 struct stack_entry *field;
1801 trace_assign_type(field, entry);
1803 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1805 ret = trace_seq_puts(s, " <= ");
1807 return TRACE_TYPE_PARTIAL_LINE;
1809 ret = seq_print_ip_sym(s, field->caller[i],
1812 return TRACE_TYPE_PARTIAL_LINE;
1814 ret = trace_seq_puts(s, "\n");
1816 return TRACE_TYPE_PARTIAL_LINE;
1820 struct print_entry *field;
1822 trace_assign_type(field, entry);
1824 seq_print_ip_sym(s, field->ip, sym_flags);
1825 trace_seq_printf(s, ": %s", field->buf);
1826 if (entry->flags & TRACE_FLAG_CONT)
1827 trace_seq_print_cont(s, iter);
1830 case TRACE_FN_RET: {
1831 return print_return_function(iter);
1834 case TRACE_BRANCH: {
1835 struct trace_branch *field;
1837 trace_assign_type(field, entry);
1839 trace_seq_printf(s, "[%s] %s:%s:%d\n",
1840 field->correct ? " ok " : " MISS ",
1847 return TRACE_TYPE_HANDLED;
1850 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1852 struct trace_seq *s = &iter->seq;
1853 struct trace_entry *entry;
1859 if (entry->type == TRACE_CONT)
1860 return TRACE_TYPE_HANDLED;
1862 ret = trace_seq_printf(s, "%d %d %llu ",
1863 entry->pid, iter->cpu, iter->ts);
1865 return TRACE_TYPE_PARTIAL_LINE;
1867 switch (entry->type) {
1869 struct ftrace_entry *field;
1871 trace_assign_type(field, entry);
1873 ret = trace_seq_printf(s, "%x %x\n",
1877 return TRACE_TYPE_PARTIAL_LINE;
1882 struct ctx_switch_entry *field;
1884 trace_assign_type(field, entry);
1886 S = field->prev_state < sizeof(state_to_char) ?
1887 state_to_char[field->prev_state] : 'X';
1888 T = field->next_state < sizeof(state_to_char) ?
1889 state_to_char[field->next_state] : 'X';
1890 if (entry->type == TRACE_WAKE)
1892 ret = trace_seq_printf(s, "%d %d %c %d %d %d %c\n",
1901 return TRACE_TYPE_PARTIAL_LINE;
1906 struct special_entry *field;
1908 trace_assign_type(field, entry);
1910 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1915 return TRACE_TYPE_PARTIAL_LINE;
1919 struct print_entry *field;
1921 trace_assign_type(field, entry);
1923 trace_seq_printf(s, "# %lx %s", field->ip, field->buf);
1924 if (entry->flags & TRACE_FLAG_CONT)
1925 trace_seq_print_cont(s, iter);
1929 return TRACE_TYPE_HANDLED;
1932 #define SEQ_PUT_FIELD_RET(s, x) \
1934 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
1938 #define SEQ_PUT_HEX_FIELD_RET(s, x) \
1940 BUILD_BUG_ON(sizeof(x) > MAX_MEMHEX_BYTES); \
1941 if (!trace_seq_putmem_hex(s, &(x), sizeof(x))) \
1945 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1947 struct trace_seq *s = &iter->seq;
1948 unsigned char newline = '\n';
1949 struct trace_entry *entry;
1954 if (entry->type == TRACE_CONT)
1955 return TRACE_TYPE_HANDLED;
1957 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1958 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1959 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1961 switch (entry->type) {
1963 struct ftrace_entry *field;
1965 trace_assign_type(field, entry);
1967 SEQ_PUT_HEX_FIELD_RET(s, field->ip);
1968 SEQ_PUT_HEX_FIELD_RET(s, field->parent_ip);
1973 struct ctx_switch_entry *field;
1975 trace_assign_type(field, entry);
1977 S = field->prev_state < sizeof(state_to_char) ?
1978 state_to_char[field->prev_state] : 'X';
1979 T = field->next_state < sizeof(state_to_char) ?
1980 state_to_char[field->next_state] : 'X';
1981 if (entry->type == TRACE_WAKE)
1983 SEQ_PUT_HEX_FIELD_RET(s, field->prev_pid);
1984 SEQ_PUT_HEX_FIELD_RET(s, field->prev_prio);
1985 SEQ_PUT_HEX_FIELD_RET(s, S);
1986 SEQ_PUT_HEX_FIELD_RET(s, field->next_cpu);
1987 SEQ_PUT_HEX_FIELD_RET(s, field->next_pid);
1988 SEQ_PUT_HEX_FIELD_RET(s, field->next_prio);
1989 SEQ_PUT_HEX_FIELD_RET(s, T);
1994 struct special_entry *field;
1996 trace_assign_type(field, entry);
1998 SEQ_PUT_HEX_FIELD_RET(s, field->arg1);
1999 SEQ_PUT_HEX_FIELD_RET(s, field->arg2);
2000 SEQ_PUT_HEX_FIELD_RET(s, field->arg3);
2004 SEQ_PUT_FIELD_RET(s, newline);
2006 return TRACE_TYPE_HANDLED;
2009 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2011 struct trace_seq *s = &iter->seq;
2012 struct trace_entry *entry;
2016 if (entry->type == TRACE_CONT)
2017 return TRACE_TYPE_HANDLED;
2019 SEQ_PUT_FIELD_RET(s, entry->pid);
2020 SEQ_PUT_FIELD_RET(s, entry->cpu);
2021 SEQ_PUT_FIELD_RET(s, iter->ts);
2023 switch (entry->type) {
2025 struct ftrace_entry *field;
2027 trace_assign_type(field, entry);
2029 SEQ_PUT_FIELD_RET(s, field->ip);
2030 SEQ_PUT_FIELD_RET(s, field->parent_ip);
2034 struct ctx_switch_entry *field;
2036 trace_assign_type(field, entry);
2038 SEQ_PUT_FIELD_RET(s, field->prev_pid);
2039 SEQ_PUT_FIELD_RET(s, field->prev_prio);
2040 SEQ_PUT_FIELD_RET(s, field->prev_state);
2041 SEQ_PUT_FIELD_RET(s, field->next_pid);
2042 SEQ_PUT_FIELD_RET(s, field->next_prio);
2043 SEQ_PUT_FIELD_RET(s, field->next_state);
2048 struct special_entry *field;
2050 trace_assign_type(field, entry);
2052 SEQ_PUT_FIELD_RET(s, field->arg1);
2053 SEQ_PUT_FIELD_RET(s, field->arg2);
2054 SEQ_PUT_FIELD_RET(s, field->arg3);
2061 static int trace_empty(struct trace_iterator *iter)
2065 for_each_tracing_cpu(cpu) {
2066 if (iter->buffer_iter[cpu]) {
2067 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2070 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2078 static enum print_line_t print_trace_line(struct trace_iterator *iter)
2080 enum print_line_t ret;
2082 if (iter->trace && iter->trace->print_line) {
2083 ret = iter->trace->print_line(iter);
2084 if (ret != TRACE_TYPE_UNHANDLED)
2088 if (trace_flags & TRACE_ITER_BIN)
2089 return print_bin_fmt(iter);
2091 if (trace_flags & TRACE_ITER_HEX)
2092 return print_hex_fmt(iter);
2094 if (trace_flags & TRACE_ITER_RAW)
2095 return print_raw_fmt(iter);
2097 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2098 return print_lat_fmt(iter, iter->idx, iter->cpu);
2100 return print_trace_fmt(iter);
2103 static int s_show(struct seq_file *m, void *v)
2105 struct trace_iterator *iter = v;
2107 if (iter->ent == NULL) {
2109 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2112 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2113 /* print nothing if the buffers are empty */
2114 if (trace_empty(iter))
2116 print_trace_header(m, iter);
2117 if (!(trace_flags & TRACE_ITER_VERBOSE))
2118 print_lat_help_header(m);
2120 if (!(trace_flags & TRACE_ITER_VERBOSE))
2121 print_func_help_header(m);
2124 print_trace_line(iter);
2125 trace_print_seq(m, &iter->seq);
2131 static struct seq_operations tracer_seq_ops = {
2138 static struct trace_iterator *
2139 __tracing_open(struct inode *inode, struct file *file, int *ret)
2141 struct trace_iterator *iter;
2145 if (tracing_disabled) {
2150 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2156 mutex_lock(&trace_types_lock);
2157 if (current_trace && current_trace->print_max)
2160 iter->tr = inode->i_private;
2161 iter->trace = current_trace;
2164 /* Annotate start of buffers if we had overruns */
2165 if (ring_buffer_overruns(iter->tr->buffer))
2166 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2169 for_each_tracing_cpu(cpu) {
2171 iter->buffer_iter[cpu] =
2172 ring_buffer_read_start(iter->tr->buffer, cpu);
2174 if (!iter->buffer_iter[cpu])
2178 /* TODO stop tracer */
2179 *ret = seq_open(file, &tracer_seq_ops);
2183 m = file->private_data;
2186 /* stop the trace while dumping */
2189 if (iter->trace && iter->trace->open)
2190 iter->trace->open(iter);
2192 mutex_unlock(&trace_types_lock);
2198 for_each_tracing_cpu(cpu) {
2199 if (iter->buffer_iter[cpu])
2200 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2202 mutex_unlock(&trace_types_lock);
2204 return ERR_PTR(-ENOMEM);
2207 int tracing_open_generic(struct inode *inode, struct file *filp)
2209 if (tracing_disabled)
2212 filp->private_data = inode->i_private;
2216 int tracing_release(struct inode *inode, struct file *file)
2218 struct seq_file *m = (struct seq_file *)file->private_data;
2219 struct trace_iterator *iter = m->private;
2222 mutex_lock(&trace_types_lock);
2223 for_each_tracing_cpu(cpu) {
2224 if (iter->buffer_iter[cpu])
2225 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2228 if (iter->trace && iter->trace->close)
2229 iter->trace->close(iter);
2231 /* reenable tracing if it was previously enabled */
2233 mutex_unlock(&trace_types_lock);
2235 seq_release(inode, file);
2240 static int tracing_open(struct inode *inode, struct file *file)
2244 __tracing_open(inode, file, &ret);
2249 static int tracing_lt_open(struct inode *inode, struct file *file)
2251 struct trace_iterator *iter;
2254 iter = __tracing_open(inode, file, &ret);
2257 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2264 t_next(struct seq_file *m, void *v, loff_t *pos)
2266 struct tracer *t = m->private;
2278 static void *t_start(struct seq_file *m, loff_t *pos)
2280 struct tracer *t = m->private;
2283 mutex_lock(&trace_types_lock);
2284 for (; t && l < *pos; t = t_next(m, t, &l))
2290 static void t_stop(struct seq_file *m, void *p)
2292 mutex_unlock(&trace_types_lock);
2295 static int t_show(struct seq_file *m, void *v)
2297 struct tracer *t = v;
2302 seq_printf(m, "%s", t->name);
2311 static struct seq_operations show_traces_seq_ops = {
2318 static int show_traces_open(struct inode *inode, struct file *file)
2322 if (tracing_disabled)
2325 ret = seq_open(file, &show_traces_seq_ops);
2327 struct seq_file *m = file->private_data;
2328 m->private = trace_types;
2334 static struct file_operations tracing_fops = {
2335 .open = tracing_open,
2337 .llseek = seq_lseek,
2338 .release = tracing_release,
2341 static struct file_operations tracing_lt_fops = {
2342 .open = tracing_lt_open,
2344 .llseek = seq_lseek,
2345 .release = tracing_release,
2348 static struct file_operations show_traces_fops = {
2349 .open = show_traces_open,
2351 .release = seq_release,
2355 * Only trace on a CPU if the bitmask is set:
2357 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
2360 * When tracing/tracing_cpu_mask is modified then this holds
2361 * the new bitmask we are about to install:
2363 static cpumask_t tracing_cpumask_new;
2366 * The tracer itself will not take this lock, but still we want
2367 * to provide a consistent cpumask to user-space:
2369 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2372 * Temporary storage for the character representation of the
2373 * CPU bitmask (and one more byte for the newline):
2375 static char mask_str[NR_CPUS + 1];
2378 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2379 size_t count, loff_t *ppos)
2383 mutex_lock(&tracing_cpumask_update_lock);
2385 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2386 if (count - len < 2) {
2390 len += sprintf(mask_str + len, "\n");
2391 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2394 mutex_unlock(&tracing_cpumask_update_lock);
2400 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2401 size_t count, loff_t *ppos)
2405 mutex_lock(&tracing_cpumask_update_lock);
2406 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2410 raw_local_irq_disable();
2411 __raw_spin_lock(&ftrace_max_lock);
2412 for_each_tracing_cpu(cpu) {
2414 * Increase/decrease the disabled counter if we are
2415 * about to flip a bit in the cpumask:
2417 if (cpu_isset(cpu, tracing_cpumask) &&
2418 !cpu_isset(cpu, tracing_cpumask_new)) {
2419 atomic_inc(&global_trace.data[cpu]->disabled);
2421 if (!cpu_isset(cpu, tracing_cpumask) &&
2422 cpu_isset(cpu, tracing_cpumask_new)) {
2423 atomic_dec(&global_trace.data[cpu]->disabled);
2426 __raw_spin_unlock(&ftrace_max_lock);
2427 raw_local_irq_enable();
2429 tracing_cpumask = tracing_cpumask_new;
2431 mutex_unlock(&tracing_cpumask_update_lock);
2436 mutex_unlock(&tracing_cpumask_update_lock);
2441 static struct file_operations tracing_cpumask_fops = {
2442 .open = tracing_open_generic,
2443 .read = tracing_cpumask_read,
2444 .write = tracing_cpumask_write,
2448 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2449 size_t cnt, loff_t *ppos)
2455 u32 tracer_flags = current_trace->flags->val;
2456 struct tracer_opt *trace_opts = current_trace->flags->opts;
2459 /* calulate max size */
2460 for (i = 0; trace_options[i]; i++) {
2461 len += strlen(trace_options[i]);
2462 len += 3; /* "no" and space */
2466 * Increase the size with names of options specific
2467 * of the current tracer.
2469 for (i = 0; trace_opts[i].name; i++) {
2470 len += strlen(trace_opts[i].name);
2471 len += 3; /* "no" and space */
2474 /* +2 for \n and \0 */
2475 buf = kmalloc(len + 2, GFP_KERNEL);
2479 for (i = 0; trace_options[i]; i++) {
2480 if (trace_flags & (1 << i))
2481 r += sprintf(buf + r, "%s ", trace_options[i]);
2483 r += sprintf(buf + r, "no%s ", trace_options[i]);
2486 for (i = 0; trace_opts[i].name; i++) {
2487 if (tracer_flags & trace_opts[i].bit)
2488 r += sprintf(buf + r, "%s ",
2489 trace_opts[i].name);
2491 r += sprintf(buf + r, "no%s ",
2492 trace_opts[i].name);
2495 r += sprintf(buf + r, "\n");
2496 WARN_ON(r >= len + 2);
2498 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2505 /* Try to assign a tracer specific option */
2506 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2508 struct tracer_flags *trace_flags = trace->flags;
2509 struct tracer_opt *opts = NULL;
2513 for (i = 0; trace_flags->opts[i].name; i++) {
2514 opts = &trace_flags->opts[i];
2515 len = strlen(opts->name);
2517 if (strncmp(cmp, opts->name, len) == 0) {
2518 ret = trace->set_flag(trace_flags->val,
2524 if (!trace_flags->opts[i].name)
2527 /* Refused to handle */
2532 trace_flags->val &= ~opts->bit;
2534 trace_flags->val |= opts->bit;
2540 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2541 size_t cnt, loff_t *ppos)
2549 if (cnt >= sizeof(buf))
2552 if (copy_from_user(&buf, ubuf, cnt))
2557 if (strncmp(buf, "no", 2) == 0) {
2562 for (i = 0; trace_options[i]; i++) {
2563 int len = strlen(trace_options[i]);
2565 if (strncmp(cmp, trace_options[i], len) == 0) {
2567 trace_flags &= ~(1 << i);
2569 trace_flags |= (1 << i);
2574 /* If no option could be set, test the specific tracer options */
2575 if (!trace_options[i]) {
2576 ret = set_tracer_option(current_trace, cmp, neg);
2586 static struct file_operations tracing_iter_fops = {
2587 .open = tracing_open_generic,
2588 .read = tracing_trace_options_read,
2589 .write = tracing_trace_options_write,
2592 static const char readme_msg[] =
2593 "tracing mini-HOWTO:\n\n"
2595 "# mount -t debugfs nodev /debug\n\n"
2596 "# cat /debug/tracing/available_tracers\n"
2597 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2598 "# cat /debug/tracing/current_tracer\n"
2600 "# echo sched_switch > /debug/tracing/current_tracer\n"
2601 "# cat /debug/tracing/current_tracer\n"
2603 "# cat /debug/tracing/trace_options\n"
2604 "noprint-parent nosym-offset nosym-addr noverbose\n"
2605 "# echo print-parent > /debug/tracing/trace_options\n"
2606 "# echo 1 > /debug/tracing/tracing_enabled\n"
2607 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2608 "echo 0 > /debug/tracing/tracing_enabled\n"
2612 tracing_readme_read(struct file *filp, char __user *ubuf,
2613 size_t cnt, loff_t *ppos)
2615 return simple_read_from_buffer(ubuf, cnt, ppos,
2616 readme_msg, strlen(readme_msg));
2619 static struct file_operations tracing_readme_fops = {
2620 .open = tracing_open_generic,
2621 .read = tracing_readme_read,
2625 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2626 size_t cnt, loff_t *ppos)
2631 r = sprintf(buf, "%u\n", tracer_enabled);
2632 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2636 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2637 size_t cnt, loff_t *ppos)
2639 struct trace_array *tr = filp->private_data;
2644 if (cnt >= sizeof(buf))
2647 if (copy_from_user(&buf, ubuf, cnt))
2652 ret = strict_strtoul(buf, 10, &val);
2658 mutex_lock(&trace_types_lock);
2659 if (tracer_enabled ^ val) {
2662 if (current_trace->start)
2663 current_trace->start(tr);
2668 if (current_trace->stop)
2669 current_trace->stop(tr);
2672 mutex_unlock(&trace_types_lock);
2680 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2681 size_t cnt, loff_t *ppos)
2683 char buf[max_tracer_type_len+2];
2686 mutex_lock(&trace_types_lock);
2688 r = sprintf(buf, "%s\n", current_trace->name);
2690 r = sprintf(buf, "\n");
2691 mutex_unlock(&trace_types_lock);
2693 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2696 static int tracing_set_tracer(char *buf)
2698 struct trace_array *tr = &global_trace;
2702 mutex_lock(&trace_types_lock);
2703 for (t = trace_types; t; t = t->next) {
2704 if (strcmp(t->name, buf) == 0)
2711 if (t == current_trace)
2714 trace_branch_disable();
2715 if (current_trace && current_trace->reset)
2716 current_trace->reset(tr);
2725 trace_branch_enable(tr);
2727 mutex_unlock(&trace_types_lock);
2733 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2734 size_t cnt, loff_t *ppos)
2736 char buf[max_tracer_type_len+1];
2743 if (cnt > max_tracer_type_len)
2744 cnt = max_tracer_type_len;
2746 if (copy_from_user(&buf, ubuf, cnt))
2751 /* strip ending whitespace. */
2752 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2755 err = tracing_set_tracer(buf);
2765 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2766 size_t cnt, loff_t *ppos)
2768 unsigned long *ptr = filp->private_data;
2772 r = snprintf(buf, sizeof(buf), "%ld\n",
2773 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2774 if (r > sizeof(buf))
2776 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2780 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2781 size_t cnt, loff_t *ppos)
2783 long *ptr = filp->private_data;
2788 if (cnt >= sizeof(buf))
2791 if (copy_from_user(&buf, ubuf, cnt))
2796 ret = strict_strtoul(buf, 10, &val);
2805 static atomic_t tracing_reader;
2807 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2809 struct trace_iterator *iter;
2811 if (tracing_disabled)
2814 /* We only allow for reader of the pipe */
2815 if (atomic_inc_return(&tracing_reader) != 1) {
2816 atomic_dec(&tracing_reader);
2820 /* create a buffer to store the information to pass to userspace */
2821 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2825 mutex_lock(&trace_types_lock);
2827 /* trace pipe does not show start of buffer */
2828 cpus_setall(iter->started);
2830 iter->tr = &global_trace;
2831 iter->trace = current_trace;
2832 filp->private_data = iter;
2834 if (iter->trace->pipe_open)
2835 iter->trace->pipe_open(iter);
2836 mutex_unlock(&trace_types_lock);
2841 static int tracing_release_pipe(struct inode *inode, struct file *file)
2843 struct trace_iterator *iter = file->private_data;
2846 atomic_dec(&tracing_reader);
2852 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2854 struct trace_iterator *iter = filp->private_data;
2856 if (trace_flags & TRACE_ITER_BLOCK) {
2858 * Always select as readable when in blocking mode
2860 return POLLIN | POLLRDNORM;
2862 if (!trace_empty(iter))
2863 return POLLIN | POLLRDNORM;
2864 poll_wait(filp, &trace_wait, poll_table);
2865 if (!trace_empty(iter))
2866 return POLLIN | POLLRDNORM;
2876 tracing_read_pipe(struct file *filp, char __user *ubuf,
2877 size_t cnt, loff_t *ppos)
2879 struct trace_iterator *iter = filp->private_data;
2882 /* return any leftover data */
2883 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2887 trace_seq_reset(&iter->seq);
2889 mutex_lock(&trace_types_lock);
2890 if (iter->trace->read) {
2891 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2898 while (trace_empty(iter)) {
2900 if ((filp->f_flags & O_NONBLOCK)) {
2906 * This is a make-shift waitqueue. The reason we don't use
2907 * an actual wait queue is because:
2908 * 1) we only ever have one waiter
2909 * 2) the tracing, traces all functions, we don't want
2910 * the overhead of calling wake_up and friends
2911 * (and tracing them too)
2912 * Anyway, this is really very primitive wakeup.
2914 set_current_state(TASK_INTERRUPTIBLE);
2915 iter->tr->waiter = current;
2917 mutex_unlock(&trace_types_lock);
2919 /* sleep for 100 msecs, and try again. */
2920 schedule_timeout(HZ/10);
2922 mutex_lock(&trace_types_lock);
2924 iter->tr->waiter = NULL;
2926 if (signal_pending(current)) {
2931 if (iter->trace != current_trace)
2935 * We block until we read something and tracing is disabled.
2936 * We still block if tracing is disabled, but we have never
2937 * read anything. This allows a user to cat this file, and
2938 * then enable tracing. But after we have read something,
2939 * we give an EOF when tracing is again disabled.
2941 * iter->pos will be 0 if we haven't read anything.
2943 if (!tracer_enabled && iter->pos)
2949 /* stop when tracing is finished */
2950 if (trace_empty(iter))
2953 if (cnt >= PAGE_SIZE)
2954 cnt = PAGE_SIZE - 1;
2956 /* reset all but tr, trace, and overruns */
2957 memset(&iter->seq, 0,
2958 sizeof(struct trace_iterator) -
2959 offsetof(struct trace_iterator, seq));
2962 while (find_next_entry_inc(iter) != NULL) {
2963 enum print_line_t ret;
2964 int len = iter->seq.len;
2966 ret = print_trace_line(iter);
2967 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2968 /* don't print partial lines */
2969 iter->seq.len = len;
2973 trace_consume(iter);
2975 if (iter->seq.len >= cnt)
2979 /* Now copy what we have to the user */
2980 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2981 if (iter->seq.readpos >= iter->seq.len)
2982 trace_seq_reset(&iter->seq);
2985 * If there was nothing to send to user, inspite of consuming trace
2986 * entries, go back to wait for more entries.
2992 mutex_unlock(&trace_types_lock);
2998 tracing_entries_read(struct file *filp, char __user *ubuf,
2999 size_t cnt, loff_t *ppos)
3001 struct trace_array *tr = filp->private_data;
3005 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3006 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3010 tracing_entries_write(struct file *filp, const char __user *ubuf,
3011 size_t cnt, loff_t *ppos)
3017 if (cnt >= sizeof(buf))
3020 if (copy_from_user(&buf, ubuf, cnt))
3025 ret = strict_strtoul(buf, 10, &val);
3029 /* must have at least 1 entry */
3033 mutex_lock(&trace_types_lock);
3037 /* disable all cpu buffers */
3038 for_each_tracing_cpu(cpu) {
3039 if (global_trace.data[cpu])
3040 atomic_inc(&global_trace.data[cpu]->disabled);
3041 if (max_tr.data[cpu])
3042 atomic_inc(&max_tr.data[cpu]->disabled);
3045 /* value is in KB */
3048 if (val != global_trace.entries) {
3049 ret = ring_buffer_resize(global_trace.buffer, val);
3055 ret = ring_buffer_resize(max_tr.buffer, val);
3059 r = ring_buffer_resize(global_trace.buffer,
3060 global_trace.entries);
3062 /* AARGH! We are left with different
3063 * size max buffer!!!! */
3065 tracing_disabled = 1;
3070 global_trace.entries = val;
3075 /* If check pages failed, return ENOMEM */
3076 if (tracing_disabled)
3079 for_each_tracing_cpu(cpu) {
3080 if (global_trace.data[cpu])
3081 atomic_dec(&global_trace.data[cpu]->disabled);
3082 if (max_tr.data[cpu])
3083 atomic_dec(&max_tr.data[cpu]->disabled);
3087 max_tr.entries = global_trace.entries;
3088 mutex_unlock(&trace_types_lock);
3093 static int mark_printk(const char *fmt, ...)
3097 va_start(args, fmt);
3098 ret = trace_vprintk(0, fmt, args);
3104 tracing_mark_write(struct file *filp, const char __user *ubuf,
3105 size_t cnt, loff_t *fpos)
3110 if (tracing_disabled)
3113 if (cnt > TRACE_BUF_SIZE)
3114 cnt = TRACE_BUF_SIZE;
3116 buf = kmalloc(cnt + 1, GFP_KERNEL);
3120 if (copy_from_user(buf, ubuf, cnt)) {
3125 /* Cut from the first nil or newline. */
3127 end = strchr(buf, '\n');
3131 cnt = mark_printk("%s\n", buf);
3138 static struct file_operations tracing_max_lat_fops = {
3139 .open = tracing_open_generic,
3140 .read = tracing_max_lat_read,
3141 .write = tracing_max_lat_write,
3144 static struct file_operations tracing_ctrl_fops = {
3145 .open = tracing_open_generic,
3146 .read = tracing_ctrl_read,
3147 .write = tracing_ctrl_write,
3150 static struct file_operations set_tracer_fops = {
3151 .open = tracing_open_generic,
3152 .read = tracing_set_trace_read,
3153 .write = tracing_set_trace_write,
3156 static struct file_operations tracing_pipe_fops = {
3157 .open = tracing_open_pipe,
3158 .poll = tracing_poll_pipe,
3159 .read = tracing_read_pipe,
3160 .release = tracing_release_pipe,
3163 static struct file_operations tracing_entries_fops = {
3164 .open = tracing_open_generic,
3165 .read = tracing_entries_read,
3166 .write = tracing_entries_write,
3169 static struct file_operations tracing_mark_fops = {
3170 .open = tracing_open_generic,
3171 .write = tracing_mark_write,
3174 #ifdef CONFIG_DYNAMIC_FTRACE
3176 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3182 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3183 size_t cnt, loff_t *ppos)
3185 static char ftrace_dyn_info_buffer[1024];
3186 static DEFINE_MUTEX(dyn_info_mutex);
3187 unsigned long *p = filp->private_data;
3188 char *buf = ftrace_dyn_info_buffer;
3189 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3192 mutex_lock(&dyn_info_mutex);
3193 r = sprintf(buf, "%ld ", *p);
3195 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3198 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3200 mutex_unlock(&dyn_info_mutex);
3205 static struct file_operations tracing_dyn_info_fops = {
3206 .open = tracing_open_generic,
3207 .read = tracing_read_dyn_info,
3211 static struct dentry *d_tracer;
3213 struct dentry *tracing_init_dentry(void)
3220 d_tracer = debugfs_create_dir("tracing", NULL);
3222 if (!d_tracer && !once) {
3224 pr_warning("Could not create debugfs directory 'tracing'\n");
3231 #ifdef CONFIG_FTRACE_SELFTEST
3232 /* Let selftest have access to static functions in this file */
3233 #include "trace_selftest.c"
3236 static __init int tracer_init_debugfs(void)
3238 struct dentry *d_tracer;
3239 struct dentry *entry;
3241 d_tracer = tracing_init_dentry();
3243 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3244 &global_trace, &tracing_ctrl_fops);
3246 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3248 entry = debugfs_create_file("trace_options", 0644, d_tracer,
3249 NULL, &tracing_iter_fops);
3251 pr_warning("Could not create debugfs 'trace_options' entry\n");
3253 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3254 NULL, &tracing_cpumask_fops);
3256 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3258 entry = debugfs_create_file("latency_trace", 0444, d_tracer,
3259 &global_trace, &tracing_lt_fops);
3261 pr_warning("Could not create debugfs 'latency_trace' entry\n");
3263 entry = debugfs_create_file("trace", 0444, d_tracer,
3264 &global_trace, &tracing_fops);
3266 pr_warning("Could not create debugfs 'trace' entry\n");
3268 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3269 &global_trace, &show_traces_fops);
3271 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3273 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3274 &global_trace, &set_tracer_fops);
3276 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3278 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3279 &tracing_max_latency,
3280 &tracing_max_lat_fops);
3282 pr_warning("Could not create debugfs "
3283 "'tracing_max_latency' entry\n");
3285 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3286 &tracing_thresh, &tracing_max_lat_fops);
3288 pr_warning("Could not create debugfs "
3289 "'tracing_thresh' entry\n");
3290 entry = debugfs_create_file("README", 0644, d_tracer,
3291 NULL, &tracing_readme_fops);
3293 pr_warning("Could not create debugfs 'README' entry\n");
3295 entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
3296 NULL, &tracing_pipe_fops);
3298 pr_warning("Could not create debugfs "
3299 "'trace_pipe' entry\n");
3301 entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
3302 &global_trace, &tracing_entries_fops);
3304 pr_warning("Could not create debugfs "
3305 "'buffer_size_kb' entry\n");
3307 entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3308 NULL, &tracing_mark_fops);
3310 pr_warning("Could not create debugfs "
3311 "'trace_marker' entry\n");
3313 #ifdef CONFIG_DYNAMIC_FTRACE
3314 entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3315 &ftrace_update_tot_cnt,
3316 &tracing_dyn_info_fops);
3318 pr_warning("Could not create debugfs "
3319 "'dyn_ftrace_total_info' entry\n");
3321 #ifdef CONFIG_SYSPROF_TRACER
3322 init_tracer_sysprof_debugfs(d_tracer);
3327 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3329 static DEFINE_SPINLOCK(trace_buf_lock);
3330 static char trace_buf[TRACE_BUF_SIZE];
3332 struct ring_buffer_event *event;
3333 struct trace_array *tr = &global_trace;
3334 struct trace_array_cpu *data;
3335 struct print_entry *entry;
3336 unsigned long flags, irq_flags;
3337 int cpu, len = 0, size, pc;
3339 if (tracing_disabled)
3342 pc = preempt_count();
3343 preempt_disable_notrace();
3344 cpu = raw_smp_processor_id();
3345 data = tr->data[cpu];
3347 if (unlikely(atomic_read(&data->disabled)))
3350 spin_lock_irqsave(&trace_buf_lock, flags);
3351 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
3353 len = min(len, TRACE_BUF_SIZE-1);
3356 size = sizeof(*entry) + len + 1;
3357 event = ring_buffer_lock_reserve(tr->buffer, size, &irq_flags);
3360 entry = ring_buffer_event_data(event);
3361 tracing_generic_entry_update(&entry->ent, flags, pc);
3362 entry->ent.type = TRACE_PRINT;
3365 memcpy(&entry->buf, trace_buf, len);
3366 entry->buf[len] = 0;
3367 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
3370 spin_unlock_irqrestore(&trace_buf_lock, flags);
3373 preempt_enable_notrace();
3377 EXPORT_SYMBOL_GPL(trace_vprintk);
3379 int __ftrace_printk(unsigned long ip, const char *fmt, ...)
3384 if (!(trace_flags & TRACE_ITER_PRINTK))
3388 ret = trace_vprintk(ip, fmt, ap);
3392 EXPORT_SYMBOL_GPL(__ftrace_printk);
3394 static int trace_panic_handler(struct notifier_block *this,
3395 unsigned long event, void *unused)
3397 if (ftrace_dump_on_oops)
3402 static struct notifier_block trace_panic_notifier = {
3403 .notifier_call = trace_panic_handler,
3405 .priority = 150 /* priority: INT_MAX >= x >= 0 */
3408 static int trace_die_handler(struct notifier_block *self,
3414 if (ftrace_dump_on_oops)
3423 static struct notifier_block trace_die_notifier = {
3424 .notifier_call = trace_die_handler,
3429 * printk is set to max of 1024, we really don't need it that big.
3430 * Nothing should be printing 1000 characters anyway.
3432 #define TRACE_MAX_PRINT 1000
3435 * Define here KERN_TRACE so that we have one place to modify
3436 * it if we decide to change what log level the ftrace dump
3439 #define KERN_TRACE KERN_INFO
3442 trace_printk_seq(struct trace_seq *s)
3444 /* Probably should print a warning here. */
3448 /* should be zero ended, but we are paranoid. */
3449 s->buffer[s->len] = 0;
3451 printk(KERN_TRACE "%s", s->buffer);
3456 void ftrace_dump(void)
3458 static DEFINE_SPINLOCK(ftrace_dump_lock);
3459 /* use static because iter can be a bit big for the stack */
3460 static struct trace_iterator iter;
3461 static cpumask_t mask;
3462 static int dump_ran;
3463 unsigned long flags;
3467 spin_lock_irqsave(&ftrace_dump_lock, flags);
3473 /* No turning back! */
3476 for_each_tracing_cpu(cpu) {
3477 atomic_inc(&global_trace.data[cpu]->disabled);
3480 printk(KERN_TRACE "Dumping ftrace buffer:\n");
3482 iter.tr = &global_trace;
3483 iter.trace = current_trace;
3486 * We need to stop all tracing on all CPUS to read the
3487 * the next buffer. This is a bit expensive, but is
3488 * not done often. We fill all what we can read,
3489 * and then release the locks again.
3494 while (!trace_empty(&iter)) {
3497 printk(KERN_TRACE "---------------------------------\n");
3501 /* reset all but tr, trace, and overruns */
3502 memset(&iter.seq, 0,
3503 sizeof(struct trace_iterator) -
3504 offsetof(struct trace_iterator, seq));
3505 iter.iter_flags |= TRACE_FILE_LAT_FMT;
3508 if (find_next_entry_inc(&iter) != NULL) {
3509 print_trace_line(&iter);
3510 trace_consume(&iter);
3513 trace_printk_seq(&iter.seq);
3517 printk(KERN_TRACE " (ftrace buffer empty)\n");
3519 printk(KERN_TRACE "---------------------------------\n");
3522 spin_unlock_irqrestore(&ftrace_dump_lock, flags);
3525 __init static int tracer_alloc_buffers(void)
3527 struct trace_array_cpu *data;
3530 /* TODO: make the number of buffers hot pluggable with CPUS */
3531 tracing_buffer_mask = cpu_possible_map;
3533 global_trace.buffer = ring_buffer_alloc(trace_buf_size,
3534 TRACE_BUFFER_FLAGS);
3535 if (!global_trace.buffer) {
3536 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
3540 global_trace.entries = ring_buffer_size(global_trace.buffer);
3542 #ifdef CONFIG_TRACER_MAX_TRACE
3543 max_tr.buffer = ring_buffer_alloc(trace_buf_size,
3544 TRACE_BUFFER_FLAGS);
3545 if (!max_tr.buffer) {
3546 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
3548 ring_buffer_free(global_trace.buffer);
3551 max_tr.entries = ring_buffer_size(max_tr.buffer);
3552 WARN_ON(max_tr.entries != global_trace.entries);
3555 /* Allocate the first page for all buffers */
3556 for_each_tracing_cpu(i) {
3557 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
3558 max_tr.data[i] = &per_cpu(max_data, i);
3561 trace_init_cmdlines();
3563 register_tracer(&nop_trace);
3564 #ifdef CONFIG_BOOT_TRACER
3565 register_tracer(&boot_tracer);
3566 current_trace = &boot_tracer;
3567 current_trace->init(&global_trace);
3569 current_trace = &nop_trace;
3572 /* All seems OK, enable tracing */
3573 tracing_disabled = 0;
3575 atomic_notifier_chain_register(&panic_notifier_list,
3576 &trace_panic_notifier);
3578 register_die_notifier(&trace_die_notifier);
3582 early_initcall(tracer_alloc_buffers);
3583 fs_initcall(tracer_init_debugfs);