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;
48 * Kill all tracing for good (never come back).
49 * It is initialized to 1 but will turn to zero if the initialization
50 * of the tracer is successful. But that is the only place that sets
53 int tracing_disabled = 1;
55 static DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
57 static inline void ftrace_disable_cpu(void)
60 local_inc(&__get_cpu_var(ftrace_cpu_disabled));
63 static inline void ftrace_enable_cpu(void)
65 local_dec(&__get_cpu_var(ftrace_cpu_disabled));
69 static cpumask_t __read_mostly tracing_buffer_mask;
71 #define for_each_tracing_cpu(cpu) \
72 for_each_cpu_mask(cpu, tracing_buffer_mask)
75 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
77 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
78 * is set, then ftrace_dump is called. This will output the contents
79 * of the ftrace buffers to the console. This is very useful for
80 * capturing traces that lead to crashes and outputing it to a
83 * It is default off, but you can enable it with either specifying
84 * "ftrace_dump_on_oops" in the kernel command line, or setting
85 * /proc/sys/kernel/ftrace_dump_on_oops to true.
87 int ftrace_dump_on_oops;
89 static int tracing_set_tracer(char *buf);
91 static int __init set_ftrace(char *str)
93 tracing_set_tracer(str);
96 __setup("ftrace", set_ftrace);
98 static int __init set_ftrace_dump_on_oops(char *str)
100 ftrace_dump_on_oops = 1;
103 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
106 ns2usecs(cycle_t nsec)
113 cycle_t ftrace_now(int cpu)
115 u64 ts = ring_buffer_time_stamp(cpu);
116 ring_buffer_normalize_time_stamp(cpu, &ts);
121 * The global_trace is the descriptor that holds the tracing
122 * buffers for the live tracing. For each CPU, it contains
123 * a link list of pages that will store trace entries. The
124 * page descriptor of the pages in the memory is used to hold
125 * the link list by linking the lru item in the page descriptor
126 * to each of the pages in the buffer per CPU.
128 * For each active CPU there is a data field that holds the
129 * pages for the buffer for that CPU. Each CPU has the same number
130 * of pages allocated for its buffer.
132 static struct trace_array global_trace;
134 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
137 * The max_tr is used to snapshot the global_trace when a maximum
138 * latency is reached. Some tracers will use this to store a maximum
139 * trace while it continues examining live traces.
141 * The buffers for the max_tr are set up the same as the global_trace.
142 * When a snapshot is taken, the link list of the max_tr is swapped
143 * with the link list of the global_trace and the buffers are reset for
144 * the global_trace so the tracing can continue.
146 static struct trace_array max_tr;
148 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
150 /* tracer_enabled is used to toggle activation of a tracer */
151 static int tracer_enabled = 1;
154 * tracing_is_enabled - return tracer_enabled status
156 * This function is used by other tracers to know the status
157 * of the tracer_enabled flag. Tracers may use this function
158 * to know if it should enable their features when starting
159 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
161 int tracing_is_enabled(void)
163 return tracer_enabled;
166 /* function tracing enabled */
167 int ftrace_function_enabled;
170 * trace_buf_size is the size in bytes that is allocated
171 * for a buffer. Note, the number of bytes is always rounded
174 * This number is purposely set to a low number of 16384.
175 * If the dump on oops happens, it will be much appreciated
176 * to not have to wait for all that output. Anyway this can be
177 * boot time and run time configurable.
179 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
181 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
183 /* trace_types holds a link list of available tracers. */
184 static struct tracer *trace_types __read_mostly;
186 /* current_trace points to the tracer that is currently active */
187 static struct tracer *current_trace __read_mostly;
190 * max_tracer_type_len is used to simplify the allocating of
191 * buffers to read userspace tracer names. We keep track of
192 * the longest tracer name registered.
194 static int max_tracer_type_len;
197 * trace_types_lock is used to protect the trace_types list.
198 * This lock is also used to keep user access serialized.
199 * Accesses from userspace will grab this lock while userspace
200 * activities happen inside the kernel.
202 static DEFINE_MUTEX(trace_types_lock);
204 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
205 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
207 /* trace_flags holds iter_ctrl options */
208 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK;
211 * trace_wake_up - wake up tasks waiting for trace input
213 * Simply wakes up any task that is blocked on the trace_wait
214 * queue. These is used with trace_poll for tasks polling the trace.
216 void trace_wake_up(void)
219 * The runqueue_is_locked() can fail, but this is the best we
222 if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
223 wake_up(&trace_wait);
226 static int __init set_buf_size(char *str)
228 unsigned long buf_size;
233 ret = strict_strtoul(str, 0, &buf_size);
234 /* nr_entries can not be zero */
235 if (ret < 0 || buf_size == 0)
237 trace_buf_size = buf_size;
240 __setup("trace_buf_size=", set_buf_size);
242 unsigned long nsecs_to_usecs(unsigned long nsecs)
248 * TRACE_ITER_SYM_MASK masks the options in trace_flags that
249 * control the output of kernel symbols.
251 #define TRACE_ITER_SYM_MASK \
252 (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
254 /* These must match the bit postions in trace_iterator_flags */
255 static const char *trace_options[] = {
272 * ftrace_max_lock is used to protect the swapping of buffers
273 * when taking a max snapshot. The buffers themselves are
274 * protected by per_cpu spinlocks. But the action of the swap
275 * needs its own lock.
277 * This is defined as a raw_spinlock_t in order to help
278 * with performance when lockdep debugging is enabled.
280 static raw_spinlock_t ftrace_max_lock =
281 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
284 * Copy the new maximum trace into the separate maximum-trace
285 * structure. (this way the maximum trace is permanently saved,
286 * for later retrieval via /debugfs/tracing/latency_trace)
289 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
291 struct trace_array_cpu *data = tr->data[cpu];
294 max_tr.time_start = data->preempt_timestamp;
296 data = max_tr.data[cpu];
297 data->saved_latency = tracing_max_latency;
299 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
300 data->pid = tsk->pid;
301 data->uid = tsk->uid;
302 data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
303 data->policy = tsk->policy;
304 data->rt_priority = tsk->rt_priority;
306 /* record this tasks comm */
307 tracing_record_cmdline(current);
311 * trace_seq_printf - sequence printing of trace information
312 * @s: trace sequence descriptor
313 * @fmt: printf format string
315 * The tracer may use either sequence operations or its own
316 * copy to user routines. To simplify formating of a trace
317 * trace_seq_printf is used to store strings into a special
318 * buffer (@s). Then the output may be either used by
319 * the sequencer or pulled into another buffer.
322 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
324 int len = (PAGE_SIZE - 1) - s->len;
332 ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
335 /* If we can't write it all, don't bother writing anything */
345 * trace_seq_puts - trace sequence printing of simple string
346 * @s: trace sequence descriptor
347 * @str: simple string to record
349 * The tracer may use either the sequence operations or its own
350 * copy to user routines. This function records a simple string
351 * into a special buffer (@s) for later retrieval by a sequencer
352 * or other mechanism.
355 trace_seq_puts(struct trace_seq *s, const char *str)
357 int len = strlen(str);
359 if (len > ((PAGE_SIZE - 1) - s->len))
362 memcpy(s->buffer + s->len, str, len);
369 trace_seq_putc(struct trace_seq *s, unsigned char c)
371 if (s->len >= (PAGE_SIZE - 1))
374 s->buffer[s->len++] = c;
380 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
382 if (len > ((PAGE_SIZE - 1) - s->len))
385 memcpy(s->buffer + s->len, mem, len);
391 #define MAX_MEMHEX_BYTES 8
392 #define HEX_CHARS (MAX_MEMHEX_BYTES*2 + 1)
395 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
397 unsigned char hex[HEX_CHARS];
398 unsigned char *data = mem;
402 for (i = 0, j = 0; i < len; i++) {
404 for (i = len-1, j = 0; i >= 0; i--) {
406 hex[j++] = hex_asc_hi(data[i]);
407 hex[j++] = hex_asc_lo(data[i]);
411 return trace_seq_putmem(s, hex, j);
415 trace_seq_reset(struct trace_seq *s)
421 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
426 if (s->len <= s->readpos)
429 len = s->len - s->readpos;
432 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
441 trace_print_seq(struct seq_file *m, struct trace_seq *s)
443 int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
446 seq_puts(m, s->buffer);
452 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
454 * @tsk: the task with the latency
455 * @cpu: The cpu that initiated the trace.
457 * Flip the buffers between the @tr and the max_tr and record information
458 * about which task was the cause of this latency.
461 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
463 struct ring_buffer *buf = tr->buffer;
465 WARN_ON_ONCE(!irqs_disabled());
466 __raw_spin_lock(&ftrace_max_lock);
468 tr->buffer = max_tr.buffer;
471 ftrace_disable_cpu();
472 ring_buffer_reset(tr->buffer);
475 __update_max_tr(tr, tsk, cpu);
476 __raw_spin_unlock(&ftrace_max_lock);
480 * update_max_tr_single - only copy one trace over, and reset the rest
482 * @tsk - task with the latency
483 * @cpu - the cpu of the buffer to copy.
485 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
488 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
492 WARN_ON_ONCE(!irqs_disabled());
493 __raw_spin_lock(&ftrace_max_lock);
495 ftrace_disable_cpu();
497 ring_buffer_reset(max_tr.buffer);
498 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
504 __update_max_tr(tr, tsk, cpu);
505 __raw_spin_unlock(&ftrace_max_lock);
509 * register_tracer - register a tracer with the ftrace system.
510 * @type - the plugin for the tracer
512 * Register a new plugin tracer.
514 int register_tracer(struct tracer *type)
521 pr_info("Tracer must have a name\n");
525 mutex_lock(&trace_types_lock);
526 for (t = trace_types; t; t = t->next) {
527 if (strcmp(type->name, t->name) == 0) {
529 pr_info("Trace %s already registered\n",
536 #ifdef CONFIG_FTRACE_STARTUP_TEST
537 if (type->selftest) {
538 struct tracer *saved_tracer = current_trace;
539 struct trace_array *tr = &global_trace;
542 * Run a selftest on this tracer.
543 * Here we reset the trace buffer, and set the current
544 * tracer to be this tracer. The tracer can then run some
545 * internal tracing to verify that everything is in order.
546 * If we fail, we do not register this tracer.
548 for_each_tracing_cpu(i) {
549 tracing_reset(tr, i);
551 current_trace = type;
552 /* the test is responsible for initializing and enabling */
553 pr_info("Testing tracer %s: ", type->name);
554 ret = type->selftest(type, tr);
555 /* the test is responsible for resetting too */
556 current_trace = saved_tracer;
558 printk(KERN_CONT "FAILED!\n");
561 /* Only reset on passing, to avoid touching corrupted buffers */
562 for_each_tracing_cpu(i) {
563 tracing_reset(tr, i);
565 printk(KERN_CONT "PASSED\n");
569 type->next = trace_types;
571 len = strlen(type->name);
572 if (len > max_tracer_type_len)
573 max_tracer_type_len = len;
576 mutex_unlock(&trace_types_lock);
581 void unregister_tracer(struct tracer *type)
586 mutex_lock(&trace_types_lock);
587 for (t = &trace_types; *t; t = &(*t)->next) {
591 pr_info("Trace %s not registered\n", type->name);
596 if (strlen(type->name) != max_tracer_type_len)
599 max_tracer_type_len = 0;
600 for (t = &trace_types; *t; t = &(*t)->next) {
601 len = strlen((*t)->name);
602 if (len > max_tracer_type_len)
603 max_tracer_type_len = len;
606 mutex_unlock(&trace_types_lock);
609 void tracing_reset(struct trace_array *tr, int cpu)
611 ftrace_disable_cpu();
612 ring_buffer_reset_cpu(tr->buffer, cpu);
616 #define SAVED_CMDLINES 128
617 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
618 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
619 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
620 static int cmdline_idx;
621 static DEFINE_SPINLOCK(trace_cmdline_lock);
623 /* temporary disable recording */
624 atomic_t trace_record_cmdline_disabled __read_mostly;
626 static void trace_init_cmdlines(void)
628 memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
629 memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
633 static int trace_stop_count;
634 static DEFINE_SPINLOCK(tracing_start_lock);
637 * tracing_start - quick start of the tracer
639 * If tracing is enabled but was stopped by tracing_stop,
640 * this will start the tracer back up.
642 void tracing_start(void)
644 struct ring_buffer *buffer;
647 if (tracing_disabled)
650 spin_lock_irqsave(&tracing_start_lock, flags);
651 if (--trace_stop_count)
654 if (trace_stop_count < 0) {
655 /* Someone screwed up their debugging */
657 trace_stop_count = 0;
662 buffer = global_trace.buffer;
664 ring_buffer_record_enable(buffer);
666 buffer = max_tr.buffer;
668 ring_buffer_record_enable(buffer);
672 spin_unlock_irqrestore(&tracing_start_lock, flags);
676 * tracing_stop - quick stop of the tracer
678 * Light weight way to stop tracing. Use in conjunction with
681 void tracing_stop(void)
683 struct ring_buffer *buffer;
687 spin_lock_irqsave(&tracing_start_lock, flags);
688 if (trace_stop_count++)
691 buffer = global_trace.buffer;
693 ring_buffer_record_disable(buffer);
695 buffer = max_tr.buffer;
697 ring_buffer_record_disable(buffer);
700 spin_unlock_irqrestore(&tracing_start_lock, flags);
703 void trace_stop_cmdline_recording(void);
705 static void trace_save_cmdline(struct task_struct *tsk)
710 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
714 * It's not the end of the world if we don't get
715 * the lock, but we also don't want to spin
716 * nor do we want to disable interrupts,
717 * so if we miss here, then better luck next time.
719 if (!spin_trylock(&trace_cmdline_lock))
722 idx = map_pid_to_cmdline[tsk->pid];
723 if (idx >= SAVED_CMDLINES) {
724 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
726 map = map_cmdline_to_pid[idx];
727 if (map <= PID_MAX_DEFAULT)
728 map_pid_to_cmdline[map] = (unsigned)-1;
730 map_pid_to_cmdline[tsk->pid] = idx;
735 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
737 spin_unlock(&trace_cmdline_lock);
740 static char *trace_find_cmdline(int pid)
742 char *cmdline = "<...>";
748 if (pid > PID_MAX_DEFAULT)
751 map = map_pid_to_cmdline[pid];
752 if (map >= SAVED_CMDLINES)
755 cmdline = saved_cmdlines[map];
761 void tracing_record_cmdline(struct task_struct *tsk)
763 if (atomic_read(&trace_record_cmdline_disabled))
766 trace_save_cmdline(tsk);
770 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
773 struct task_struct *tsk = current;
775 entry->preempt_count = pc & 0xff;
776 entry->pid = (tsk) ? tsk->pid : 0;
778 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
779 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
781 TRACE_FLAG_IRQS_NOSUPPORT |
783 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
784 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
785 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
789 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
790 unsigned long ip, unsigned long parent_ip, unsigned long flags,
793 struct ring_buffer_event *event;
794 struct ftrace_entry *entry;
795 unsigned long irq_flags;
797 /* If we are reading the ring buffer, don't trace */
798 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
801 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
805 entry = ring_buffer_event_data(event);
806 tracing_generic_entry_update(&entry->ent, flags, pc);
807 entry->ent.type = TRACE_FN;
809 entry->parent_ip = parent_ip;
810 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
814 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
815 unsigned long ip, unsigned long parent_ip, unsigned long flags,
818 if (likely(!atomic_read(&data->disabled)))
819 trace_function(tr, data, ip, parent_ip, flags, pc);
822 static void ftrace_trace_stack(struct trace_array *tr,
823 struct trace_array_cpu *data,
827 #ifdef CONFIG_STACKTRACE
828 struct ring_buffer_event *event;
829 struct stack_entry *entry;
830 struct stack_trace trace;
831 unsigned long irq_flags;
833 if (!(trace_flags & TRACE_ITER_STACKTRACE))
836 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
840 entry = ring_buffer_event_data(event);
841 tracing_generic_entry_update(&entry->ent, flags, pc);
842 entry->ent.type = TRACE_STACK;
844 memset(&entry->caller, 0, sizeof(entry->caller));
846 trace.nr_entries = 0;
847 trace.max_entries = FTRACE_STACK_ENTRIES;
849 trace.entries = entry->caller;
851 save_stack_trace(&trace);
852 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
856 void __trace_stack(struct trace_array *tr,
857 struct trace_array_cpu *data,
861 ftrace_trace_stack(tr, data, flags, skip, preempt_count());
865 ftrace_trace_special(void *__tr, void *__data,
866 unsigned long arg1, unsigned long arg2, unsigned long arg3,
869 struct ring_buffer_event *event;
870 struct trace_array_cpu *data = __data;
871 struct trace_array *tr = __tr;
872 struct special_entry *entry;
873 unsigned long irq_flags;
875 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
879 entry = ring_buffer_event_data(event);
880 tracing_generic_entry_update(&entry->ent, 0, pc);
881 entry->ent.type = TRACE_SPECIAL;
885 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
886 ftrace_trace_stack(tr, data, irq_flags, 4, pc);
892 __trace_special(void *__tr, void *__data,
893 unsigned long arg1, unsigned long arg2, unsigned long arg3)
895 ftrace_trace_special(__tr, __data, arg1, arg2, arg3, preempt_count());
899 tracing_sched_switch_trace(struct trace_array *tr,
900 struct trace_array_cpu *data,
901 struct task_struct *prev,
902 struct task_struct *next,
903 unsigned long flags, int pc)
905 struct ring_buffer_event *event;
906 struct ctx_switch_entry *entry;
907 unsigned long irq_flags;
909 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
913 entry = ring_buffer_event_data(event);
914 tracing_generic_entry_update(&entry->ent, flags, pc);
915 entry->ent.type = TRACE_CTX;
916 entry->prev_pid = prev->pid;
917 entry->prev_prio = prev->prio;
918 entry->prev_state = prev->state;
919 entry->next_pid = next->pid;
920 entry->next_prio = next->prio;
921 entry->next_state = next->state;
922 entry->next_cpu = task_cpu(next);
923 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
924 ftrace_trace_stack(tr, data, flags, 5, pc);
928 tracing_sched_wakeup_trace(struct trace_array *tr,
929 struct trace_array_cpu *data,
930 struct task_struct *wakee,
931 struct task_struct *curr,
932 unsigned long flags, int pc)
934 struct ring_buffer_event *event;
935 struct ctx_switch_entry *entry;
936 unsigned long irq_flags;
938 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
942 entry = ring_buffer_event_data(event);
943 tracing_generic_entry_update(&entry->ent, flags, pc);
944 entry->ent.type = TRACE_WAKE;
945 entry->prev_pid = curr->pid;
946 entry->prev_prio = curr->prio;
947 entry->prev_state = curr->state;
948 entry->next_pid = wakee->pid;
949 entry->next_prio = wakee->prio;
950 entry->next_state = wakee->state;
951 entry->next_cpu = task_cpu(wakee);
952 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
953 ftrace_trace_stack(tr, data, flags, 6, pc);
959 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
961 struct trace_array *tr = &global_trace;
962 struct trace_array_cpu *data;
966 if (tracing_disabled)
969 pc = preempt_count();
970 preempt_disable_notrace();
971 cpu = raw_smp_processor_id();
972 data = tr->data[cpu];
974 if (likely(!atomic_read(&data->disabled)))
975 ftrace_trace_special(tr, data, arg1, arg2, arg3, pc);
977 preempt_enable_notrace();
980 #ifdef CONFIG_FUNCTION_TRACER
982 function_trace_call_preempt_only(unsigned long ip, unsigned long parent_ip)
984 struct trace_array *tr = &global_trace;
985 struct trace_array_cpu *data;
991 if (unlikely(!ftrace_function_enabled))
994 pc = preempt_count();
995 resched = ftrace_preempt_disable();
996 local_save_flags(flags);
997 cpu = raw_smp_processor_id();
998 data = tr->data[cpu];
999 disabled = atomic_inc_return(&data->disabled);
1001 if (likely(disabled == 1))
1002 trace_function(tr, data, ip, parent_ip, flags, pc);
1004 atomic_dec(&data->disabled);
1005 ftrace_preempt_enable(resched);
1009 function_trace_call(unsigned long ip, unsigned long parent_ip)
1011 struct trace_array *tr = &global_trace;
1012 struct trace_array_cpu *data;
1013 unsigned long flags;
1018 if (unlikely(!ftrace_function_enabled))
1022 * Need to use raw, since this must be called before the
1023 * recursive protection is performed.
1025 raw_local_irq_save(flags);
1026 cpu = raw_smp_processor_id();
1027 data = tr->data[cpu];
1028 disabled = atomic_inc_return(&data->disabled);
1030 if (likely(disabled == 1)) {
1031 pc = preempt_count();
1032 trace_function(tr, data, ip, parent_ip, flags, pc);
1035 atomic_dec(&data->disabled);
1036 raw_local_irq_restore(flags);
1039 static struct ftrace_ops trace_ops __read_mostly =
1041 .func = function_trace_call,
1044 void tracing_start_function_trace(void)
1046 ftrace_function_enabled = 0;
1048 if (trace_flags & TRACE_ITER_PREEMPTONLY)
1049 trace_ops.func = function_trace_call_preempt_only;
1051 trace_ops.func = function_trace_call;
1053 register_ftrace_function(&trace_ops);
1054 ftrace_function_enabled = 1;
1057 void tracing_stop_function_trace(void)
1059 ftrace_function_enabled = 0;
1060 unregister_ftrace_function(&trace_ops);
1064 enum trace_file_type {
1065 TRACE_FILE_LAT_FMT = 1,
1068 static void trace_iterator_increment(struct trace_iterator *iter, int cpu)
1070 /* Don't allow ftrace to trace into the ring buffers */
1071 ftrace_disable_cpu();
1074 if (iter->buffer_iter[iter->cpu])
1075 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1077 ftrace_enable_cpu();
1080 static struct trace_entry *
1081 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1083 struct ring_buffer_event *event;
1084 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1086 /* Don't allow ftrace to trace into the ring buffers */
1087 ftrace_disable_cpu();
1090 event = ring_buffer_iter_peek(buf_iter, ts);
1092 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1094 ftrace_enable_cpu();
1096 return event ? ring_buffer_event_data(event) : NULL;
1099 static struct trace_entry *
1100 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1102 struct ring_buffer *buffer = iter->tr->buffer;
1103 struct trace_entry *ent, *next = NULL;
1104 u64 next_ts = 0, ts;
1108 for_each_tracing_cpu(cpu) {
1110 if (ring_buffer_empty_cpu(buffer, cpu))
1113 ent = peek_next_entry(iter, cpu, &ts);
1116 * Pick the entry with the smallest timestamp:
1118 if (ent && (!next || ts < next_ts)) {
1126 *ent_cpu = next_cpu;
1134 /* Find the next real entry, without updating the iterator itself */
1135 static struct trace_entry *
1136 find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1138 return __find_next_entry(iter, ent_cpu, ent_ts);
1141 /* Find the next real entry, and increment the iterator to the next entry */
1142 static void *find_next_entry_inc(struct trace_iterator *iter)
1144 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1147 trace_iterator_increment(iter, iter->cpu);
1149 return iter->ent ? iter : NULL;
1152 static void trace_consume(struct trace_iterator *iter)
1154 /* Don't allow ftrace to trace into the ring buffers */
1155 ftrace_disable_cpu();
1156 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1157 ftrace_enable_cpu();
1160 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1162 struct trace_iterator *iter = m->private;
1168 /* can't go backwards */
1173 ent = find_next_entry_inc(iter);
1177 while (ent && iter->idx < i)
1178 ent = find_next_entry_inc(iter);
1185 static void *s_start(struct seq_file *m, loff_t *pos)
1187 struct trace_iterator *iter = m->private;
1192 mutex_lock(&trace_types_lock);
1194 if (!current_trace || current_trace != iter->trace) {
1195 mutex_unlock(&trace_types_lock);
1199 atomic_inc(&trace_record_cmdline_disabled);
1201 if (*pos != iter->pos) {
1206 ftrace_disable_cpu();
1208 for_each_tracing_cpu(cpu) {
1209 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1212 ftrace_enable_cpu();
1214 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1219 p = s_next(m, p, &l);
1225 static void s_stop(struct seq_file *m, void *p)
1227 atomic_dec(&trace_record_cmdline_disabled);
1228 mutex_unlock(&trace_types_lock);
1231 #ifdef CONFIG_KRETPROBES
1232 static inline const char *kretprobed(const char *name)
1234 static const char tramp_name[] = "kretprobe_trampoline";
1235 int size = sizeof(tramp_name);
1237 if (strncmp(tramp_name, name, size) == 0)
1238 return "[unknown/kretprobe'd]";
1242 static inline const char *kretprobed(const char *name)
1246 #endif /* CONFIG_KRETPROBES */
1249 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1251 #ifdef CONFIG_KALLSYMS
1252 char str[KSYM_SYMBOL_LEN];
1255 kallsyms_lookup(address, NULL, NULL, NULL, str);
1257 name = kretprobed(str);
1259 return trace_seq_printf(s, fmt, name);
1265 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1266 unsigned long address)
1268 #ifdef CONFIG_KALLSYMS
1269 char str[KSYM_SYMBOL_LEN];
1272 sprint_symbol(str, address);
1273 name = kretprobed(str);
1275 return trace_seq_printf(s, fmt, name);
1280 #ifndef CONFIG_64BIT
1281 # define IP_FMT "%08lx"
1283 # define IP_FMT "%016lx"
1287 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1292 return trace_seq_printf(s, "0");
1294 if (sym_flags & TRACE_ITER_SYM_OFFSET)
1295 ret = seq_print_sym_offset(s, "%s", ip);
1297 ret = seq_print_sym_short(s, "%s", ip);
1302 if (sym_flags & TRACE_ITER_SYM_ADDR)
1303 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1307 static void print_lat_help_header(struct seq_file *m)
1309 seq_puts(m, "# _------=> CPU# \n");
1310 seq_puts(m, "# / _-----=> irqs-off \n");
1311 seq_puts(m, "# | / _----=> need-resched \n");
1312 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1313 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1314 seq_puts(m, "# |||| / \n");
1315 seq_puts(m, "# ||||| delay \n");
1316 seq_puts(m, "# cmd pid ||||| time | caller \n");
1317 seq_puts(m, "# \\ / ||||| \\ | / \n");
1320 static void print_func_help_header(struct seq_file *m)
1322 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1323 seq_puts(m, "# | | | | |\n");
1328 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1330 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1331 struct trace_array *tr = iter->tr;
1332 struct trace_array_cpu *data = tr->data[tr->cpu];
1333 struct tracer *type = current_trace;
1334 unsigned long total;
1335 unsigned long entries;
1336 const char *name = "preemption";
1341 entries = ring_buffer_entries(iter->tr->buffer);
1343 ring_buffer_overruns(iter->tr->buffer);
1345 seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1347 seq_puts(m, "-----------------------------------"
1348 "---------------------------------\n");
1349 seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1350 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1351 nsecs_to_usecs(data->saved_latency),
1355 #if defined(CONFIG_PREEMPT_NONE)
1357 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1359 #elif defined(CONFIG_PREEMPT)
1364 /* These are reserved for later use */
1367 seq_printf(m, " #P:%d)\n", num_online_cpus());
1371 seq_puts(m, " -----------------\n");
1372 seq_printf(m, " | task: %.16s-%d "
1373 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1374 data->comm, data->pid, data->uid, data->nice,
1375 data->policy, data->rt_priority);
1376 seq_puts(m, " -----------------\n");
1378 if (data->critical_start) {
1379 seq_puts(m, " => started at: ");
1380 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1381 trace_print_seq(m, &iter->seq);
1382 seq_puts(m, "\n => ended at: ");
1383 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1384 trace_print_seq(m, &iter->seq);
1392 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1394 int hardirq, softirq;
1397 comm = trace_find_cmdline(entry->pid);
1399 trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1400 trace_seq_printf(s, "%3d", cpu);
1401 trace_seq_printf(s, "%c%c",
1402 (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' :
1403 (entry->flags & TRACE_FLAG_IRQS_NOSUPPORT) ? 'X' : '.',
1404 ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1406 hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1407 softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1408 if (hardirq && softirq) {
1409 trace_seq_putc(s, 'H');
1412 trace_seq_putc(s, 'h');
1415 trace_seq_putc(s, 's');
1417 trace_seq_putc(s, '.');
1421 if (entry->preempt_count)
1422 trace_seq_printf(s, "%x", entry->preempt_count);
1424 trace_seq_puts(s, ".");
1427 unsigned long preempt_mark_thresh = 100;
1430 lat_print_timestamp(struct trace_seq *s, u64 abs_usecs,
1431 unsigned long rel_usecs)
1433 trace_seq_printf(s, " %4lldus", abs_usecs);
1434 if (rel_usecs > preempt_mark_thresh)
1435 trace_seq_puts(s, "!: ");
1436 else if (rel_usecs > 1)
1437 trace_seq_puts(s, "+: ");
1439 trace_seq_puts(s, " : ");
1442 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1445 * The message is supposed to contain an ending newline.
1446 * If the printing stops prematurely, try to add a newline of our own.
1448 void trace_seq_print_cont(struct trace_seq *s, struct trace_iterator *iter)
1450 struct trace_entry *ent;
1451 struct trace_field_cont *cont;
1454 ent = peek_next_entry(iter, iter->cpu, NULL);
1455 if (!ent || ent->type != TRACE_CONT) {
1456 trace_seq_putc(s, '\n');
1461 cont = (struct trace_field_cont *)ent;
1463 ok = (trace_seq_printf(s, "%s", cont->buf) > 0);
1465 ftrace_disable_cpu();
1467 if (iter->buffer_iter[iter->cpu])
1468 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1470 ring_buffer_consume(iter->tr->buffer, iter->cpu, NULL);
1472 ftrace_enable_cpu();
1474 ent = peek_next_entry(iter, iter->cpu, NULL);
1475 } while (ent && ent->type == TRACE_CONT);
1478 trace_seq_putc(s, '\n');
1481 static void test_cpu_buff_start(struct trace_iterator *iter)
1483 struct trace_seq *s = &iter->seq;
1485 if (cpu_isset(iter->cpu, iter->started))
1488 cpu_set(iter->cpu, iter->started);
1489 trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1492 static enum print_line_t
1493 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1495 struct trace_seq *s = &iter->seq;
1496 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1497 struct trace_entry *next_entry;
1498 unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1499 struct trace_entry *entry = iter->ent;
1500 unsigned long abs_usecs;
1501 unsigned long rel_usecs;
1508 if (entry->type == TRACE_CONT)
1509 return TRACE_TYPE_HANDLED;
1511 test_cpu_buff_start(iter);
1513 next_entry = find_next_entry(iter, NULL, &next_ts);
1516 rel_usecs = ns2usecs(next_ts - iter->ts);
1517 abs_usecs = ns2usecs(iter->ts - iter->tr->time_start);
1520 comm = trace_find_cmdline(entry->pid);
1521 trace_seq_printf(s, "%16s %5d %3d %d %08x %08x [%08lx]"
1522 " %ld.%03ldms (+%ld.%03ldms): ",
1524 entry->pid, cpu, entry->flags,
1525 entry->preempt_count, trace_idx,
1528 abs_usecs % 1000, rel_usecs/1000,
1531 lat_print_generic(s, entry, cpu);
1532 lat_print_timestamp(s, abs_usecs, rel_usecs);
1534 switch (entry->type) {
1536 struct ftrace_entry *field;
1538 trace_assign_type(field, entry);
1540 seq_print_ip_sym(s, field->ip, sym_flags);
1541 trace_seq_puts(s, " (");
1542 seq_print_ip_sym(s, field->parent_ip, sym_flags);
1543 trace_seq_puts(s, ")\n");
1548 struct ctx_switch_entry *field;
1550 trace_assign_type(field, entry);
1552 T = field->next_state < sizeof(state_to_char) ?
1553 state_to_char[field->next_state] : 'X';
1555 state = field->prev_state ?
1556 __ffs(field->prev_state) + 1 : 0;
1557 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1558 comm = trace_find_cmdline(field->next_pid);
1559 trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c %s\n",
1562 S, entry->type == TRACE_CTX ? "==>" : " +",
1569 case TRACE_SPECIAL: {
1570 struct special_entry *field;
1572 trace_assign_type(field, entry);
1574 trace_seq_printf(s, "# %ld %ld %ld\n",
1581 struct stack_entry *field;
1583 trace_assign_type(field, entry);
1585 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1587 trace_seq_puts(s, " <= ");
1588 seq_print_ip_sym(s, field->caller[i], sym_flags);
1590 trace_seq_puts(s, "\n");
1594 struct print_entry *field;
1596 trace_assign_type(field, entry);
1598 seq_print_ip_sym(s, field->ip, sym_flags);
1599 trace_seq_printf(s, ": %s", field->buf);
1600 if (entry->flags & TRACE_FLAG_CONT)
1601 trace_seq_print_cont(s, iter);
1605 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1607 return TRACE_TYPE_HANDLED;
1610 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1612 struct trace_seq *s = &iter->seq;
1613 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1614 struct trace_entry *entry;
1615 unsigned long usec_rem;
1616 unsigned long long t;
1625 if (entry->type == TRACE_CONT)
1626 return TRACE_TYPE_HANDLED;
1628 test_cpu_buff_start(iter);
1630 comm = trace_find_cmdline(iter->ent->pid);
1632 t = ns2usecs(iter->ts);
1633 usec_rem = do_div(t, 1000000ULL);
1634 secs = (unsigned long)t;
1636 ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1638 return TRACE_TYPE_PARTIAL_LINE;
1639 ret = trace_seq_printf(s, "[%03d] ", iter->cpu);
1641 return TRACE_TYPE_PARTIAL_LINE;
1642 ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1644 return TRACE_TYPE_PARTIAL_LINE;
1646 switch (entry->type) {
1648 struct ftrace_entry *field;
1650 trace_assign_type(field, entry);
1652 ret = seq_print_ip_sym(s, field->ip, sym_flags);
1654 return TRACE_TYPE_PARTIAL_LINE;
1655 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1657 ret = trace_seq_printf(s, " <-");
1659 return TRACE_TYPE_PARTIAL_LINE;
1660 ret = seq_print_ip_sym(s,
1664 return TRACE_TYPE_PARTIAL_LINE;
1666 ret = trace_seq_printf(s, "\n");
1668 return TRACE_TYPE_PARTIAL_LINE;
1673 struct ctx_switch_entry *field;
1675 trace_assign_type(field, entry);
1677 S = field->prev_state < sizeof(state_to_char) ?
1678 state_to_char[field->prev_state] : 'X';
1679 T = field->next_state < sizeof(state_to_char) ?
1680 state_to_char[field->next_state] : 'X';
1681 ret = trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c\n",
1685 entry->type == TRACE_CTX ? "==>" : " +",
1691 return TRACE_TYPE_PARTIAL_LINE;
1694 case TRACE_SPECIAL: {
1695 struct special_entry *field;
1697 trace_assign_type(field, entry);
1699 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1704 return TRACE_TYPE_PARTIAL_LINE;
1708 struct stack_entry *field;
1710 trace_assign_type(field, entry);
1712 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1714 ret = trace_seq_puts(s, " <= ");
1716 return TRACE_TYPE_PARTIAL_LINE;
1718 ret = seq_print_ip_sym(s, field->caller[i],
1721 return TRACE_TYPE_PARTIAL_LINE;
1723 ret = trace_seq_puts(s, "\n");
1725 return TRACE_TYPE_PARTIAL_LINE;
1729 struct print_entry *field;
1731 trace_assign_type(field, entry);
1733 seq_print_ip_sym(s, field->ip, sym_flags);
1734 trace_seq_printf(s, ": %s", field->buf);
1735 if (entry->flags & TRACE_FLAG_CONT)
1736 trace_seq_print_cont(s, iter);
1740 return TRACE_TYPE_HANDLED;
1743 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1745 struct trace_seq *s = &iter->seq;
1746 struct trace_entry *entry;
1752 if (entry->type == TRACE_CONT)
1753 return TRACE_TYPE_HANDLED;
1755 ret = trace_seq_printf(s, "%d %d %llu ",
1756 entry->pid, iter->cpu, iter->ts);
1758 return TRACE_TYPE_PARTIAL_LINE;
1760 switch (entry->type) {
1762 struct ftrace_entry *field;
1764 trace_assign_type(field, entry);
1766 ret = trace_seq_printf(s, "%x %x\n",
1770 return TRACE_TYPE_PARTIAL_LINE;
1775 struct ctx_switch_entry *field;
1777 trace_assign_type(field, entry);
1779 S = field->prev_state < sizeof(state_to_char) ?
1780 state_to_char[field->prev_state] : 'X';
1781 T = field->next_state < sizeof(state_to_char) ?
1782 state_to_char[field->next_state] : 'X';
1783 if (entry->type == TRACE_WAKE)
1785 ret = trace_seq_printf(s, "%d %d %c %d %d %d %c\n",
1794 return TRACE_TYPE_PARTIAL_LINE;
1799 struct special_entry *field;
1801 trace_assign_type(field, entry);
1803 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1808 return TRACE_TYPE_PARTIAL_LINE;
1812 struct print_entry *field;
1814 trace_assign_type(field, entry);
1816 trace_seq_printf(s, "# %lx %s", field->ip, field->buf);
1817 if (entry->flags & TRACE_FLAG_CONT)
1818 trace_seq_print_cont(s, iter);
1822 return TRACE_TYPE_HANDLED;
1825 #define SEQ_PUT_FIELD_RET(s, x) \
1827 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
1831 #define SEQ_PUT_HEX_FIELD_RET(s, x) \
1833 BUILD_BUG_ON(sizeof(x) > MAX_MEMHEX_BYTES); \
1834 if (!trace_seq_putmem_hex(s, &(x), sizeof(x))) \
1838 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1840 struct trace_seq *s = &iter->seq;
1841 unsigned char newline = '\n';
1842 struct trace_entry *entry;
1847 if (entry->type == TRACE_CONT)
1848 return TRACE_TYPE_HANDLED;
1850 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1851 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1852 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1854 switch (entry->type) {
1856 struct ftrace_entry *field;
1858 trace_assign_type(field, entry);
1860 SEQ_PUT_HEX_FIELD_RET(s, field->ip);
1861 SEQ_PUT_HEX_FIELD_RET(s, field->parent_ip);
1866 struct ctx_switch_entry *field;
1868 trace_assign_type(field, entry);
1870 S = field->prev_state < sizeof(state_to_char) ?
1871 state_to_char[field->prev_state] : 'X';
1872 T = field->next_state < sizeof(state_to_char) ?
1873 state_to_char[field->next_state] : 'X';
1874 if (entry->type == TRACE_WAKE)
1876 SEQ_PUT_HEX_FIELD_RET(s, field->prev_pid);
1877 SEQ_PUT_HEX_FIELD_RET(s, field->prev_prio);
1878 SEQ_PUT_HEX_FIELD_RET(s, S);
1879 SEQ_PUT_HEX_FIELD_RET(s, field->next_cpu);
1880 SEQ_PUT_HEX_FIELD_RET(s, field->next_pid);
1881 SEQ_PUT_HEX_FIELD_RET(s, field->next_prio);
1882 SEQ_PUT_HEX_FIELD_RET(s, T);
1887 struct special_entry *field;
1889 trace_assign_type(field, entry);
1891 SEQ_PUT_HEX_FIELD_RET(s, field->arg1);
1892 SEQ_PUT_HEX_FIELD_RET(s, field->arg2);
1893 SEQ_PUT_HEX_FIELD_RET(s, field->arg3);
1897 SEQ_PUT_FIELD_RET(s, newline);
1899 return TRACE_TYPE_HANDLED;
1902 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1904 struct trace_seq *s = &iter->seq;
1905 struct trace_entry *entry;
1909 if (entry->type == TRACE_CONT)
1910 return TRACE_TYPE_HANDLED;
1912 SEQ_PUT_FIELD_RET(s, entry->pid);
1913 SEQ_PUT_FIELD_RET(s, entry->cpu);
1914 SEQ_PUT_FIELD_RET(s, iter->ts);
1916 switch (entry->type) {
1918 struct ftrace_entry *field;
1920 trace_assign_type(field, entry);
1922 SEQ_PUT_FIELD_RET(s, field->ip);
1923 SEQ_PUT_FIELD_RET(s, field->parent_ip);
1927 struct ctx_switch_entry *field;
1929 trace_assign_type(field, entry);
1931 SEQ_PUT_FIELD_RET(s, field->prev_pid);
1932 SEQ_PUT_FIELD_RET(s, field->prev_prio);
1933 SEQ_PUT_FIELD_RET(s, field->prev_state);
1934 SEQ_PUT_FIELD_RET(s, field->next_pid);
1935 SEQ_PUT_FIELD_RET(s, field->next_prio);
1936 SEQ_PUT_FIELD_RET(s, field->next_state);
1941 struct special_entry *field;
1943 trace_assign_type(field, entry);
1945 SEQ_PUT_FIELD_RET(s, field->arg1);
1946 SEQ_PUT_FIELD_RET(s, field->arg2);
1947 SEQ_PUT_FIELD_RET(s, field->arg3);
1954 static int trace_empty(struct trace_iterator *iter)
1958 for_each_tracing_cpu(cpu) {
1959 if (iter->buffer_iter[cpu]) {
1960 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1963 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1971 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1973 enum print_line_t ret;
1975 if (iter->trace && iter->trace->print_line) {
1976 ret = iter->trace->print_line(iter);
1977 if (ret != TRACE_TYPE_UNHANDLED)
1981 if (trace_flags & TRACE_ITER_BIN)
1982 return print_bin_fmt(iter);
1984 if (trace_flags & TRACE_ITER_HEX)
1985 return print_hex_fmt(iter);
1987 if (trace_flags & TRACE_ITER_RAW)
1988 return print_raw_fmt(iter);
1990 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1991 return print_lat_fmt(iter, iter->idx, iter->cpu);
1993 return print_trace_fmt(iter);
1996 static int s_show(struct seq_file *m, void *v)
1998 struct trace_iterator *iter = v;
2000 if (iter->ent == NULL) {
2002 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2005 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2006 /* print nothing if the buffers are empty */
2007 if (trace_empty(iter))
2009 print_trace_header(m, iter);
2010 if (!(trace_flags & TRACE_ITER_VERBOSE))
2011 print_lat_help_header(m);
2013 if (!(trace_flags & TRACE_ITER_VERBOSE))
2014 print_func_help_header(m);
2017 print_trace_line(iter);
2018 trace_print_seq(m, &iter->seq);
2024 static struct seq_operations tracer_seq_ops = {
2031 static struct trace_iterator *
2032 __tracing_open(struct inode *inode, struct file *file, int *ret)
2034 struct trace_iterator *iter;
2038 if (tracing_disabled) {
2043 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2049 mutex_lock(&trace_types_lock);
2050 if (current_trace && current_trace->print_max)
2053 iter->tr = inode->i_private;
2054 iter->trace = current_trace;
2057 for_each_tracing_cpu(cpu) {
2059 iter->buffer_iter[cpu] =
2060 ring_buffer_read_start(iter->tr->buffer, cpu);
2062 if (!iter->buffer_iter[cpu])
2066 /* TODO stop tracer */
2067 *ret = seq_open(file, &tracer_seq_ops);
2071 m = file->private_data;
2074 /* stop the trace while dumping */
2077 if (iter->trace && iter->trace->open)
2078 iter->trace->open(iter);
2080 mutex_unlock(&trace_types_lock);
2086 for_each_tracing_cpu(cpu) {
2087 if (iter->buffer_iter[cpu])
2088 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2090 mutex_unlock(&trace_types_lock);
2092 return ERR_PTR(-ENOMEM);
2095 int tracing_open_generic(struct inode *inode, struct file *filp)
2097 if (tracing_disabled)
2100 filp->private_data = inode->i_private;
2104 int tracing_release(struct inode *inode, struct file *file)
2106 struct seq_file *m = (struct seq_file *)file->private_data;
2107 struct trace_iterator *iter = m->private;
2110 mutex_lock(&trace_types_lock);
2111 for_each_tracing_cpu(cpu) {
2112 if (iter->buffer_iter[cpu])
2113 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2116 if (iter->trace && iter->trace->close)
2117 iter->trace->close(iter);
2119 /* reenable tracing if it was previously enabled */
2121 mutex_unlock(&trace_types_lock);
2123 seq_release(inode, file);
2128 static int tracing_open(struct inode *inode, struct file *file)
2132 __tracing_open(inode, file, &ret);
2137 static int tracing_lt_open(struct inode *inode, struct file *file)
2139 struct trace_iterator *iter;
2142 iter = __tracing_open(inode, file, &ret);
2145 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2152 t_next(struct seq_file *m, void *v, loff_t *pos)
2154 struct tracer *t = m->private;
2166 static void *t_start(struct seq_file *m, loff_t *pos)
2168 struct tracer *t = m->private;
2171 mutex_lock(&trace_types_lock);
2172 for (; t && l < *pos; t = t_next(m, t, &l))
2178 static void t_stop(struct seq_file *m, void *p)
2180 mutex_unlock(&trace_types_lock);
2183 static int t_show(struct seq_file *m, void *v)
2185 struct tracer *t = v;
2190 seq_printf(m, "%s", t->name);
2199 static struct seq_operations show_traces_seq_ops = {
2206 static int show_traces_open(struct inode *inode, struct file *file)
2210 if (tracing_disabled)
2213 ret = seq_open(file, &show_traces_seq_ops);
2215 struct seq_file *m = file->private_data;
2216 m->private = trace_types;
2222 static struct file_operations tracing_fops = {
2223 .open = tracing_open,
2225 .llseek = seq_lseek,
2226 .release = tracing_release,
2229 static struct file_operations tracing_lt_fops = {
2230 .open = tracing_lt_open,
2232 .llseek = seq_lseek,
2233 .release = tracing_release,
2236 static struct file_operations show_traces_fops = {
2237 .open = show_traces_open,
2239 .release = seq_release,
2243 * Only trace on a CPU if the bitmask is set:
2245 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
2248 * When tracing/tracing_cpu_mask is modified then this holds
2249 * the new bitmask we are about to install:
2251 static cpumask_t tracing_cpumask_new;
2254 * The tracer itself will not take this lock, but still we want
2255 * to provide a consistent cpumask to user-space:
2257 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2260 * Temporary storage for the character representation of the
2261 * CPU bitmask (and one more byte for the newline):
2263 static char mask_str[NR_CPUS + 1];
2266 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2267 size_t count, loff_t *ppos)
2271 mutex_lock(&tracing_cpumask_update_lock);
2273 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2274 if (count - len < 2) {
2278 len += sprintf(mask_str + len, "\n");
2279 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2282 mutex_unlock(&tracing_cpumask_update_lock);
2288 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2289 size_t count, loff_t *ppos)
2293 mutex_lock(&tracing_cpumask_update_lock);
2294 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2298 raw_local_irq_disable();
2299 __raw_spin_lock(&ftrace_max_lock);
2300 for_each_tracing_cpu(cpu) {
2302 * Increase/decrease the disabled counter if we are
2303 * about to flip a bit in the cpumask:
2305 if (cpu_isset(cpu, tracing_cpumask) &&
2306 !cpu_isset(cpu, tracing_cpumask_new)) {
2307 atomic_inc(&global_trace.data[cpu]->disabled);
2309 if (!cpu_isset(cpu, tracing_cpumask) &&
2310 cpu_isset(cpu, tracing_cpumask_new)) {
2311 atomic_dec(&global_trace.data[cpu]->disabled);
2314 __raw_spin_unlock(&ftrace_max_lock);
2315 raw_local_irq_enable();
2317 tracing_cpumask = tracing_cpumask_new;
2319 mutex_unlock(&tracing_cpumask_update_lock);
2324 mutex_unlock(&tracing_cpumask_update_lock);
2329 static struct file_operations tracing_cpumask_fops = {
2330 .open = tracing_open_generic,
2331 .read = tracing_cpumask_read,
2332 .write = tracing_cpumask_write,
2336 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
2337 size_t cnt, loff_t *ppos)
2344 /* calulate max size */
2345 for (i = 0; trace_options[i]; i++) {
2346 len += strlen(trace_options[i]);
2347 len += 3; /* "no" and space */
2350 /* +2 for \n and \0 */
2351 buf = kmalloc(len + 2, GFP_KERNEL);
2355 for (i = 0; trace_options[i]; i++) {
2356 if (trace_flags & (1 << i))
2357 r += sprintf(buf + r, "%s ", trace_options[i]);
2359 r += sprintf(buf + r, "no%s ", trace_options[i]);
2362 r += sprintf(buf + r, "\n");
2363 WARN_ON(r >= len + 2);
2365 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2373 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
2374 size_t cnt, loff_t *ppos)
2381 if (cnt >= sizeof(buf))
2384 if (copy_from_user(&buf, ubuf, cnt))
2389 if (strncmp(buf, "no", 2) == 0) {
2394 for (i = 0; trace_options[i]; i++) {
2395 int len = strlen(trace_options[i]);
2397 if (strncmp(cmp, trace_options[i], len) == 0) {
2399 trace_flags &= ~(1 << i);
2401 trace_flags |= (1 << i);
2406 * If no option could be set, return an error:
2408 if (!trace_options[i])
2416 static struct file_operations tracing_iter_fops = {
2417 .open = tracing_open_generic,
2418 .read = tracing_iter_ctrl_read,
2419 .write = tracing_iter_ctrl_write,
2422 static const char readme_msg[] =
2423 "tracing mini-HOWTO:\n\n"
2425 "# mount -t debugfs nodev /debug\n\n"
2426 "# cat /debug/tracing/available_tracers\n"
2427 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2428 "# cat /debug/tracing/current_tracer\n"
2430 "# echo sched_switch > /debug/tracing/current_tracer\n"
2431 "# cat /debug/tracing/current_tracer\n"
2433 "# cat /debug/tracing/iter_ctrl\n"
2434 "noprint-parent nosym-offset nosym-addr noverbose\n"
2435 "# echo print-parent > /debug/tracing/iter_ctrl\n"
2436 "# echo 1 > /debug/tracing/tracing_enabled\n"
2437 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2438 "echo 0 > /debug/tracing/tracing_enabled\n"
2442 tracing_readme_read(struct file *filp, char __user *ubuf,
2443 size_t cnt, loff_t *ppos)
2445 return simple_read_from_buffer(ubuf, cnt, ppos,
2446 readme_msg, strlen(readme_msg));
2449 static struct file_operations tracing_readme_fops = {
2450 .open = tracing_open_generic,
2451 .read = tracing_readme_read,
2455 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2456 size_t cnt, loff_t *ppos)
2461 r = sprintf(buf, "%u\n", tracer_enabled);
2462 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2466 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2467 size_t cnt, loff_t *ppos)
2469 struct trace_array *tr = filp->private_data;
2474 if (cnt >= sizeof(buf))
2477 if (copy_from_user(&buf, ubuf, cnt))
2482 ret = strict_strtoul(buf, 10, &val);
2488 mutex_lock(&trace_types_lock);
2489 if (tracer_enabled ^ val) {
2492 if (current_trace->start)
2493 current_trace->start(tr);
2498 if (current_trace->stop)
2499 current_trace->stop(tr);
2502 mutex_unlock(&trace_types_lock);
2510 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2511 size_t cnt, loff_t *ppos)
2513 char buf[max_tracer_type_len+2];
2516 mutex_lock(&trace_types_lock);
2518 r = sprintf(buf, "%s\n", current_trace->name);
2520 r = sprintf(buf, "\n");
2521 mutex_unlock(&trace_types_lock);
2523 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2526 static int tracing_set_tracer(char *buf)
2528 struct trace_array *tr = &global_trace;
2532 mutex_lock(&trace_types_lock);
2533 for (t = trace_types; t; t = t->next) {
2534 if (strcmp(t->name, buf) == 0)
2541 if (t == current_trace)
2544 if (current_trace && current_trace->reset)
2545 current_trace->reset(tr);
2552 mutex_unlock(&trace_types_lock);
2558 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2559 size_t cnt, loff_t *ppos)
2561 char buf[max_tracer_type_len+1];
2565 if (cnt > max_tracer_type_len)
2566 cnt = max_tracer_type_len;
2568 if (copy_from_user(&buf, ubuf, cnt))
2573 /* strip ending whitespace. */
2574 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2577 ret = tracing_set_tracer(buf);
2588 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2589 size_t cnt, loff_t *ppos)
2591 unsigned long *ptr = filp->private_data;
2595 r = snprintf(buf, sizeof(buf), "%ld\n",
2596 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2597 if (r > sizeof(buf))
2599 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2603 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2604 size_t cnt, loff_t *ppos)
2606 long *ptr = filp->private_data;
2611 if (cnt >= sizeof(buf))
2614 if (copy_from_user(&buf, ubuf, cnt))
2619 ret = strict_strtoul(buf, 10, &val);
2628 static atomic_t tracing_reader;
2630 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2632 struct trace_iterator *iter;
2634 if (tracing_disabled)
2637 /* We only allow for reader of the pipe */
2638 if (atomic_inc_return(&tracing_reader) != 1) {
2639 atomic_dec(&tracing_reader);
2643 /* create a buffer to store the information to pass to userspace */
2644 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2648 mutex_lock(&trace_types_lock);
2650 /* trace pipe does not show start of buffer */
2651 cpus_setall(iter->started);
2653 iter->tr = &global_trace;
2654 iter->trace = current_trace;
2655 filp->private_data = iter;
2657 if (iter->trace->pipe_open)
2658 iter->trace->pipe_open(iter);
2659 mutex_unlock(&trace_types_lock);
2664 static int tracing_release_pipe(struct inode *inode, struct file *file)
2666 struct trace_iterator *iter = file->private_data;
2669 atomic_dec(&tracing_reader);
2675 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2677 struct trace_iterator *iter = filp->private_data;
2679 if (trace_flags & TRACE_ITER_BLOCK) {
2681 * Always select as readable when in blocking mode
2683 return POLLIN | POLLRDNORM;
2685 if (!trace_empty(iter))
2686 return POLLIN | POLLRDNORM;
2687 poll_wait(filp, &trace_wait, poll_table);
2688 if (!trace_empty(iter))
2689 return POLLIN | POLLRDNORM;
2699 tracing_read_pipe(struct file *filp, char __user *ubuf,
2700 size_t cnt, loff_t *ppos)
2702 struct trace_iterator *iter = filp->private_data;
2705 /* return any leftover data */
2706 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2710 trace_seq_reset(&iter->seq);
2712 mutex_lock(&trace_types_lock);
2713 if (iter->trace->read) {
2714 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2721 while (trace_empty(iter)) {
2723 if ((filp->f_flags & O_NONBLOCK)) {
2729 * This is a make-shift waitqueue. The reason we don't use
2730 * an actual wait queue is because:
2731 * 1) we only ever have one waiter
2732 * 2) the tracing, traces all functions, we don't want
2733 * the overhead of calling wake_up and friends
2734 * (and tracing them too)
2735 * Anyway, this is really very primitive wakeup.
2737 set_current_state(TASK_INTERRUPTIBLE);
2738 iter->tr->waiter = current;
2740 mutex_unlock(&trace_types_lock);
2742 /* sleep for 100 msecs, and try again. */
2743 schedule_timeout(HZ/10);
2745 mutex_lock(&trace_types_lock);
2747 iter->tr->waiter = NULL;
2749 if (signal_pending(current)) {
2754 if (iter->trace != current_trace)
2758 * We block until we read something and tracing is disabled.
2759 * We still block if tracing is disabled, but we have never
2760 * read anything. This allows a user to cat this file, and
2761 * then enable tracing. But after we have read something,
2762 * we give an EOF when tracing is again disabled.
2764 * iter->pos will be 0 if we haven't read anything.
2766 if (!tracer_enabled && iter->pos)
2772 /* stop when tracing is finished */
2773 if (trace_empty(iter))
2776 if (cnt >= PAGE_SIZE)
2777 cnt = PAGE_SIZE - 1;
2779 /* reset all but tr, trace, and overruns */
2780 memset(&iter->seq, 0,
2781 sizeof(struct trace_iterator) -
2782 offsetof(struct trace_iterator, seq));
2785 while (find_next_entry_inc(iter) != NULL) {
2786 enum print_line_t ret;
2787 int len = iter->seq.len;
2789 ret = print_trace_line(iter);
2790 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2791 /* don't print partial lines */
2792 iter->seq.len = len;
2796 trace_consume(iter);
2798 if (iter->seq.len >= cnt)
2802 /* Now copy what we have to the user */
2803 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2804 if (iter->seq.readpos >= iter->seq.len)
2805 trace_seq_reset(&iter->seq);
2808 * If there was nothing to send to user, inspite of consuming trace
2809 * entries, go back to wait for more entries.
2815 mutex_unlock(&trace_types_lock);
2821 tracing_entries_read(struct file *filp, char __user *ubuf,
2822 size_t cnt, loff_t *ppos)
2824 struct trace_array *tr = filp->private_data;
2828 r = sprintf(buf, "%lu\n", tr->entries);
2829 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2833 tracing_entries_write(struct file *filp, const char __user *ubuf,
2834 size_t cnt, loff_t *ppos)
2840 if (cnt >= sizeof(buf))
2843 if (copy_from_user(&buf, ubuf, cnt))
2848 ret = strict_strtoul(buf, 10, &val);
2852 /* must have at least 1 entry */
2856 mutex_lock(&trace_types_lock);
2860 if (val != global_trace.entries) {
2861 ret = ring_buffer_resize(global_trace.buffer, val);
2867 ret = ring_buffer_resize(max_tr.buffer, val);
2871 r = ring_buffer_resize(global_trace.buffer,
2872 global_trace.entries);
2874 /* AARGH! We are left with different
2875 * size max buffer!!!! */
2877 tracing_disabled = 1;
2882 global_trace.entries = val;
2887 /* If check pages failed, return ENOMEM */
2888 if (tracing_disabled)
2892 max_tr.entries = global_trace.entries;
2893 mutex_unlock(&trace_types_lock);
2898 static int mark_printk(const char *fmt, ...)
2902 va_start(args, fmt);
2903 ret = trace_vprintk(0, fmt, args);
2909 tracing_mark_write(struct file *filp, const char __user *ubuf,
2910 size_t cnt, loff_t *fpos)
2915 if (tracing_disabled)
2918 if (cnt > TRACE_BUF_SIZE)
2919 cnt = TRACE_BUF_SIZE;
2921 buf = kmalloc(cnt + 1, GFP_KERNEL);
2925 if (copy_from_user(buf, ubuf, cnt)) {
2930 /* Cut from the first nil or newline. */
2932 end = strchr(buf, '\n');
2936 cnt = mark_printk("%s\n", buf);
2943 static struct file_operations tracing_max_lat_fops = {
2944 .open = tracing_open_generic,
2945 .read = tracing_max_lat_read,
2946 .write = tracing_max_lat_write,
2949 static struct file_operations tracing_ctrl_fops = {
2950 .open = tracing_open_generic,
2951 .read = tracing_ctrl_read,
2952 .write = tracing_ctrl_write,
2955 static struct file_operations set_tracer_fops = {
2956 .open = tracing_open_generic,
2957 .read = tracing_set_trace_read,
2958 .write = tracing_set_trace_write,
2961 static struct file_operations tracing_pipe_fops = {
2962 .open = tracing_open_pipe,
2963 .poll = tracing_poll_pipe,
2964 .read = tracing_read_pipe,
2965 .release = tracing_release_pipe,
2968 static struct file_operations tracing_entries_fops = {
2969 .open = tracing_open_generic,
2970 .read = tracing_entries_read,
2971 .write = tracing_entries_write,
2974 static struct file_operations tracing_mark_fops = {
2975 .open = tracing_open_generic,
2976 .write = tracing_mark_write,
2979 #ifdef CONFIG_DYNAMIC_FTRACE
2981 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
2987 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
2988 size_t cnt, loff_t *ppos)
2990 static char ftrace_dyn_info_buffer[1024];
2991 static DEFINE_MUTEX(dyn_info_mutex);
2992 unsigned long *p = filp->private_data;
2993 char *buf = ftrace_dyn_info_buffer;
2994 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
2997 mutex_lock(&dyn_info_mutex);
2998 r = sprintf(buf, "%ld ", *p);
3000 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3003 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3005 mutex_unlock(&dyn_info_mutex);
3010 static struct file_operations tracing_dyn_info_fops = {
3011 .open = tracing_open_generic,
3012 .read = tracing_read_dyn_info,
3016 static struct dentry *d_tracer;
3018 struct dentry *tracing_init_dentry(void)
3025 d_tracer = debugfs_create_dir("tracing", NULL);
3027 if (!d_tracer && !once) {
3029 pr_warning("Could not create debugfs directory 'tracing'\n");
3036 #ifdef CONFIG_FTRACE_SELFTEST
3037 /* Let selftest have access to static functions in this file */
3038 #include "trace_selftest.c"
3041 static __init int tracer_init_debugfs(void)
3043 struct dentry *d_tracer;
3044 struct dentry *entry;
3046 d_tracer = tracing_init_dentry();
3048 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3049 &global_trace, &tracing_ctrl_fops);
3051 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3053 entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
3054 NULL, &tracing_iter_fops);
3056 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
3058 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3059 NULL, &tracing_cpumask_fops);
3061 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3063 entry = debugfs_create_file("latency_trace", 0444, d_tracer,
3064 &global_trace, &tracing_lt_fops);
3066 pr_warning("Could not create debugfs 'latency_trace' entry\n");
3068 entry = debugfs_create_file("trace", 0444, d_tracer,
3069 &global_trace, &tracing_fops);
3071 pr_warning("Could not create debugfs 'trace' entry\n");
3073 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3074 &global_trace, &show_traces_fops);
3076 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3078 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3079 &global_trace, &set_tracer_fops);
3081 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3083 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3084 &tracing_max_latency,
3085 &tracing_max_lat_fops);
3087 pr_warning("Could not create debugfs "
3088 "'tracing_max_latency' entry\n");
3090 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3091 &tracing_thresh, &tracing_max_lat_fops);
3093 pr_warning("Could not create debugfs "
3094 "'tracing_thresh' entry\n");
3095 entry = debugfs_create_file("README", 0644, d_tracer,
3096 NULL, &tracing_readme_fops);
3098 pr_warning("Could not create debugfs 'README' entry\n");
3100 entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
3101 NULL, &tracing_pipe_fops);
3103 pr_warning("Could not create debugfs "
3104 "'trace_pipe' entry\n");
3106 entry = debugfs_create_file("trace_entries", 0644, d_tracer,
3107 &global_trace, &tracing_entries_fops);
3109 pr_warning("Could not create debugfs "
3110 "'trace_entries' entry\n");
3112 entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3113 NULL, &tracing_mark_fops);
3115 pr_warning("Could not create debugfs "
3116 "'trace_marker' entry\n");
3118 #ifdef CONFIG_DYNAMIC_FTRACE
3119 entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3120 &ftrace_update_tot_cnt,
3121 &tracing_dyn_info_fops);
3123 pr_warning("Could not create debugfs "
3124 "'dyn_ftrace_total_info' entry\n");
3126 #ifdef CONFIG_SYSPROF_TRACER
3127 init_tracer_sysprof_debugfs(d_tracer);
3132 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3134 static DEFINE_SPINLOCK(trace_buf_lock);
3135 static char trace_buf[TRACE_BUF_SIZE];
3137 struct ring_buffer_event *event;
3138 struct trace_array *tr = &global_trace;
3139 struct trace_array_cpu *data;
3140 struct print_entry *entry;
3141 unsigned long flags, irq_flags;
3142 int cpu, len = 0, size, pc;
3144 if (tracing_disabled)
3147 pc = preempt_count();
3148 preempt_disable_notrace();
3149 cpu = raw_smp_processor_id();
3150 data = tr->data[cpu];
3152 if (unlikely(atomic_read(&data->disabled)))
3155 spin_lock_irqsave(&trace_buf_lock, flags);
3156 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
3158 len = min(len, TRACE_BUF_SIZE-1);
3161 size = sizeof(*entry) + len + 1;
3162 event = ring_buffer_lock_reserve(tr->buffer, size, &irq_flags);
3165 entry = ring_buffer_event_data(event);
3166 tracing_generic_entry_update(&entry->ent, flags, pc);
3167 entry->ent.type = TRACE_PRINT;
3170 memcpy(&entry->buf, trace_buf, len);
3171 entry->buf[len] = 0;
3172 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
3175 spin_unlock_irqrestore(&trace_buf_lock, flags);
3178 preempt_enable_notrace();
3182 EXPORT_SYMBOL_GPL(trace_vprintk);
3184 int __ftrace_printk(unsigned long ip, const char *fmt, ...)
3189 if (!(trace_flags & TRACE_ITER_PRINTK))
3193 ret = trace_vprintk(ip, fmt, ap);
3197 EXPORT_SYMBOL_GPL(__ftrace_printk);
3199 static int trace_panic_handler(struct notifier_block *this,
3200 unsigned long event, void *unused)
3202 if (ftrace_dump_on_oops)
3207 static struct notifier_block trace_panic_notifier = {
3208 .notifier_call = trace_panic_handler,
3210 .priority = 150 /* priority: INT_MAX >= x >= 0 */
3213 static int trace_die_handler(struct notifier_block *self,
3219 if (ftrace_dump_on_oops)
3228 static struct notifier_block trace_die_notifier = {
3229 .notifier_call = trace_die_handler,
3234 * printk is set to max of 1024, we really don't need it that big.
3235 * Nothing should be printing 1000 characters anyway.
3237 #define TRACE_MAX_PRINT 1000
3240 * Define here KERN_TRACE so that we have one place to modify
3241 * it if we decide to change what log level the ftrace dump
3244 #define KERN_TRACE KERN_INFO
3247 trace_printk_seq(struct trace_seq *s)
3249 /* Probably should print a warning here. */
3253 /* should be zero ended, but we are paranoid. */
3254 s->buffer[s->len] = 0;
3256 printk(KERN_TRACE "%s", s->buffer);
3261 void ftrace_dump(void)
3263 static DEFINE_SPINLOCK(ftrace_dump_lock);
3264 /* use static because iter can be a bit big for the stack */
3265 static struct trace_iterator iter;
3266 static cpumask_t mask;
3267 static int dump_ran;
3268 unsigned long flags;
3272 spin_lock_irqsave(&ftrace_dump_lock, flags);
3278 /* No turning back! */
3281 for_each_tracing_cpu(cpu) {
3282 atomic_inc(&global_trace.data[cpu]->disabled);
3285 printk(KERN_TRACE "Dumping ftrace buffer:\n");
3287 iter.tr = &global_trace;
3288 iter.trace = current_trace;
3291 * We need to stop all tracing on all CPUS to read the
3292 * the next buffer. This is a bit expensive, but is
3293 * not done often. We fill all what we can read,
3294 * and then release the locks again.
3299 while (!trace_empty(&iter)) {
3302 printk(KERN_TRACE "---------------------------------\n");
3306 /* reset all but tr, trace, and overruns */
3307 memset(&iter.seq, 0,
3308 sizeof(struct trace_iterator) -
3309 offsetof(struct trace_iterator, seq));
3310 iter.iter_flags |= TRACE_FILE_LAT_FMT;
3313 if (find_next_entry_inc(&iter) != NULL) {
3314 print_trace_line(&iter);
3315 trace_consume(&iter);
3318 trace_printk_seq(&iter.seq);
3322 printk(KERN_TRACE " (ftrace buffer empty)\n");
3324 printk(KERN_TRACE "---------------------------------\n");
3327 spin_unlock_irqrestore(&ftrace_dump_lock, flags);
3330 __init static int tracer_alloc_buffers(void)
3332 struct trace_array_cpu *data;
3335 /* TODO: make the number of buffers hot pluggable with CPUS */
3336 tracing_buffer_mask = cpu_possible_map;
3338 global_trace.buffer = ring_buffer_alloc(trace_buf_size,
3339 TRACE_BUFFER_FLAGS);
3340 if (!global_trace.buffer) {
3341 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
3345 global_trace.entries = ring_buffer_size(global_trace.buffer);
3347 #ifdef CONFIG_TRACER_MAX_TRACE
3348 max_tr.buffer = ring_buffer_alloc(trace_buf_size,
3349 TRACE_BUFFER_FLAGS);
3350 if (!max_tr.buffer) {
3351 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
3353 ring_buffer_free(global_trace.buffer);
3356 max_tr.entries = ring_buffer_size(max_tr.buffer);
3357 WARN_ON(max_tr.entries != global_trace.entries);
3360 /* Allocate the first page for all buffers */
3361 for_each_tracing_cpu(i) {
3362 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
3363 max_tr.data[i] = &per_cpu(max_data, i);
3366 trace_init_cmdlines();
3368 register_tracer(&nop_trace);
3369 #ifdef CONFIG_BOOT_TRACER
3370 register_tracer(&boot_tracer);
3371 current_trace = &boot_tracer;
3372 current_trace->init(&global_trace);
3374 current_trace = &nop_trace;
3377 /* All seems OK, enable tracing */
3378 tracing_disabled = 0;
3380 atomic_notifier_chain_register(&panic_notifier_list,
3381 &trace_panic_notifier);
3383 register_die_notifier(&trace_die_notifier);
3387 early_initcall(tracer_alloc_buffers);
3388 fs_initcall(tracer_init_debugfs);