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 trace_options default values */
208 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
212 * trace_wake_up - wake up tasks waiting for trace input
214 * Simply wakes up any task that is blocked on the trace_wait
215 * queue. These is used with trace_poll for tasks polling the trace.
217 void trace_wake_up(void)
220 * The runqueue_is_locked() can fail, but this is the best we
223 if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
224 wake_up(&trace_wait);
227 static int __init set_buf_size(char *str)
229 unsigned long buf_size;
234 ret = strict_strtoul(str, 0, &buf_size);
235 /* nr_entries can not be zero */
236 if (ret < 0 || buf_size == 0)
238 trace_buf_size = buf_size;
241 __setup("trace_buf_size=", set_buf_size);
243 unsigned long nsecs_to_usecs(unsigned long nsecs)
248 /* These must match the bit postions in trace_iterator_flags */
249 static const char *trace_options[] = {
262 #ifdef CONFIG_BRANCH_TRACER
270 * ftrace_max_lock is used to protect the swapping of buffers
271 * when taking a max snapshot. The buffers themselves are
272 * protected by per_cpu spinlocks. But the action of the swap
273 * needs its own lock.
275 * This is defined as a raw_spinlock_t in order to help
276 * with performance when lockdep debugging is enabled.
278 static raw_spinlock_t ftrace_max_lock =
279 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
282 * Copy the new maximum trace into the separate maximum-trace
283 * structure. (this way the maximum trace is permanently saved,
284 * for later retrieval via /debugfs/tracing/latency_trace)
287 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
289 struct trace_array_cpu *data = tr->data[cpu];
292 max_tr.time_start = data->preempt_timestamp;
294 data = max_tr.data[cpu];
295 data->saved_latency = tracing_max_latency;
297 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
298 data->pid = tsk->pid;
299 data->uid = tsk->uid;
300 data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
301 data->policy = tsk->policy;
302 data->rt_priority = tsk->rt_priority;
304 /* record this tasks comm */
305 tracing_record_cmdline(current);
309 * trace_seq_printf - sequence printing of trace information
310 * @s: trace sequence descriptor
311 * @fmt: printf format string
313 * The tracer may use either sequence operations or its own
314 * copy to user routines. To simplify formating of a trace
315 * trace_seq_printf is used to store strings into a special
316 * buffer (@s). Then the output may be either used by
317 * the sequencer or pulled into another buffer.
320 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
322 int len = (PAGE_SIZE - 1) - s->len;
330 ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
333 /* If we can't write it all, don't bother writing anything */
343 * trace_seq_puts - trace sequence printing of simple string
344 * @s: trace sequence descriptor
345 * @str: simple string to record
347 * The tracer may use either the sequence operations or its own
348 * copy to user routines. This function records a simple string
349 * into a special buffer (@s) for later retrieval by a sequencer
350 * or other mechanism.
353 trace_seq_puts(struct trace_seq *s, const char *str)
355 int len = strlen(str);
357 if (len > ((PAGE_SIZE - 1) - s->len))
360 memcpy(s->buffer + s->len, str, len);
367 trace_seq_putc(struct trace_seq *s, unsigned char c)
369 if (s->len >= (PAGE_SIZE - 1))
372 s->buffer[s->len++] = c;
378 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
380 if (len > ((PAGE_SIZE - 1) - s->len))
383 memcpy(s->buffer + s->len, mem, len);
389 #define MAX_MEMHEX_BYTES 8
390 #define HEX_CHARS (MAX_MEMHEX_BYTES*2 + 1)
393 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
395 unsigned char hex[HEX_CHARS];
396 unsigned char *data = mem;
400 for (i = 0, j = 0; i < len; i++) {
402 for (i = len-1, j = 0; i >= 0; i--) {
404 hex[j++] = hex_asc_hi(data[i]);
405 hex[j++] = hex_asc_lo(data[i]);
409 return trace_seq_putmem(s, hex, j);
413 trace_seq_reset(struct trace_seq *s)
419 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
424 if (s->len <= s->readpos)
427 len = s->len - s->readpos;
430 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
439 trace_print_seq(struct seq_file *m, struct trace_seq *s)
441 int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
444 seq_puts(m, s->buffer);
450 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
452 * @tsk: the task with the latency
453 * @cpu: The cpu that initiated the trace.
455 * Flip the buffers between the @tr and the max_tr and record information
456 * about which task was the cause of this latency.
459 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
461 struct ring_buffer *buf = tr->buffer;
463 WARN_ON_ONCE(!irqs_disabled());
464 __raw_spin_lock(&ftrace_max_lock);
466 tr->buffer = max_tr.buffer;
469 ftrace_disable_cpu();
470 ring_buffer_reset(tr->buffer);
473 __update_max_tr(tr, tsk, cpu);
474 __raw_spin_unlock(&ftrace_max_lock);
478 * update_max_tr_single - only copy one trace over, and reset the rest
480 * @tsk - task with the latency
481 * @cpu - the cpu of the buffer to copy.
483 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
486 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
490 WARN_ON_ONCE(!irqs_disabled());
491 __raw_spin_lock(&ftrace_max_lock);
493 ftrace_disable_cpu();
495 ring_buffer_reset(max_tr.buffer);
496 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
502 __update_max_tr(tr, tsk, cpu);
503 __raw_spin_unlock(&ftrace_max_lock);
507 * register_tracer - register a tracer with the ftrace system.
508 * @type - the plugin for the tracer
510 * Register a new plugin tracer.
512 int register_tracer(struct tracer *type)
519 pr_info("Tracer must have a name\n");
523 mutex_lock(&trace_types_lock);
524 for (t = trace_types; t; t = t->next) {
525 if (strcmp(type->name, t->name) == 0) {
527 pr_info("Trace %s already registered\n",
534 #ifdef CONFIG_FTRACE_STARTUP_TEST
535 if (type->selftest) {
536 struct tracer *saved_tracer = current_trace;
537 struct trace_array *tr = &global_trace;
540 * Run a selftest on this tracer.
541 * Here we reset the trace buffer, and set the current
542 * tracer to be this tracer. The tracer can then run some
543 * internal tracing to verify that everything is in order.
544 * If we fail, we do not register this tracer.
546 for_each_tracing_cpu(i) {
547 tracing_reset(tr, i);
549 current_trace = type;
550 /* the test is responsible for initializing and enabling */
551 pr_info("Testing tracer %s: ", type->name);
552 ret = type->selftest(type, tr);
553 /* the test is responsible for resetting too */
554 current_trace = saved_tracer;
556 printk(KERN_CONT "FAILED!\n");
559 /* Only reset on passing, to avoid touching corrupted buffers */
560 for_each_tracing_cpu(i) {
561 tracing_reset(tr, i);
563 printk(KERN_CONT "PASSED\n");
567 type->next = trace_types;
569 len = strlen(type->name);
570 if (len > max_tracer_type_len)
571 max_tracer_type_len = len;
574 mutex_unlock(&trace_types_lock);
579 void unregister_tracer(struct tracer *type)
584 mutex_lock(&trace_types_lock);
585 for (t = &trace_types; *t; t = &(*t)->next) {
589 pr_info("Trace %s not registered\n", type->name);
594 if (strlen(type->name) != max_tracer_type_len)
597 max_tracer_type_len = 0;
598 for (t = &trace_types; *t; t = &(*t)->next) {
599 len = strlen((*t)->name);
600 if (len > max_tracer_type_len)
601 max_tracer_type_len = len;
604 mutex_unlock(&trace_types_lock);
607 void tracing_reset(struct trace_array *tr, int cpu)
609 ftrace_disable_cpu();
610 ring_buffer_reset_cpu(tr->buffer, cpu);
614 #define SAVED_CMDLINES 128
615 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
616 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
617 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
618 static int cmdline_idx;
619 static DEFINE_SPINLOCK(trace_cmdline_lock);
621 /* temporary disable recording */
622 atomic_t trace_record_cmdline_disabled __read_mostly;
624 static void trace_init_cmdlines(void)
626 memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
627 memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
631 static int trace_stop_count;
632 static DEFINE_SPINLOCK(tracing_start_lock);
635 * tracing_start - quick start of the tracer
637 * If tracing is enabled but was stopped by tracing_stop,
638 * this will start the tracer back up.
640 void tracing_start(void)
642 struct ring_buffer *buffer;
645 if (tracing_disabled)
648 spin_lock_irqsave(&tracing_start_lock, flags);
649 if (--trace_stop_count)
652 if (trace_stop_count < 0) {
653 /* Someone screwed up their debugging */
655 trace_stop_count = 0;
660 buffer = global_trace.buffer;
662 ring_buffer_record_enable(buffer);
664 buffer = max_tr.buffer;
666 ring_buffer_record_enable(buffer);
670 spin_unlock_irqrestore(&tracing_start_lock, flags);
674 * tracing_stop - quick stop of the tracer
676 * Light weight way to stop tracing. Use in conjunction with
679 void tracing_stop(void)
681 struct ring_buffer *buffer;
685 spin_lock_irqsave(&tracing_start_lock, flags);
686 if (trace_stop_count++)
689 buffer = global_trace.buffer;
691 ring_buffer_record_disable(buffer);
693 buffer = max_tr.buffer;
695 ring_buffer_record_disable(buffer);
698 spin_unlock_irqrestore(&tracing_start_lock, flags);
701 void trace_stop_cmdline_recording(void);
703 static void trace_save_cmdline(struct task_struct *tsk)
708 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
712 * It's not the end of the world if we don't get
713 * the lock, but we also don't want to spin
714 * nor do we want to disable interrupts,
715 * so if we miss here, then better luck next time.
717 if (!spin_trylock(&trace_cmdline_lock))
720 idx = map_pid_to_cmdline[tsk->pid];
721 if (idx >= SAVED_CMDLINES) {
722 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
724 map = map_cmdline_to_pid[idx];
725 if (map <= PID_MAX_DEFAULT)
726 map_pid_to_cmdline[map] = (unsigned)-1;
728 map_pid_to_cmdline[tsk->pid] = idx;
733 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
735 spin_unlock(&trace_cmdline_lock);
738 static char *trace_find_cmdline(int pid)
740 char *cmdline = "<...>";
746 if (pid > PID_MAX_DEFAULT)
749 map = map_pid_to_cmdline[pid];
750 if (map >= SAVED_CMDLINES)
753 cmdline = saved_cmdlines[map];
759 void tracing_record_cmdline(struct task_struct *tsk)
761 if (atomic_read(&trace_record_cmdline_disabled))
764 trace_save_cmdline(tsk);
768 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
771 struct task_struct *tsk = current;
773 entry->preempt_count = pc & 0xff;
774 entry->pid = (tsk) ? tsk->pid : 0;
776 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
777 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
779 TRACE_FLAG_IRQS_NOSUPPORT |
781 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
782 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
783 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
787 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
788 unsigned long ip, unsigned long parent_ip, unsigned long flags,
791 struct ring_buffer_event *event;
792 struct ftrace_entry *entry;
793 unsigned long irq_flags;
795 /* If we are reading the ring buffer, don't trace */
796 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
799 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
803 entry = ring_buffer_event_data(event);
804 tracing_generic_entry_update(&entry->ent, flags, pc);
805 entry->ent.type = TRACE_FN;
807 entry->parent_ip = parent_ip;
808 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
811 #ifdef CONFIG_FUNCTION_RET_TRACER
812 static void __trace_function_return(struct trace_array *tr,
813 struct trace_array_cpu *data,
814 struct ftrace_retfunc *trace,
818 struct ring_buffer_event *event;
819 struct ftrace_ret_entry *entry;
820 unsigned long irq_flags;
822 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
825 event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry),
829 entry = ring_buffer_event_data(event);
830 tracing_generic_entry_update(&entry->ent, flags, pc);
831 entry->ent.type = TRACE_FN_RET;
832 entry->ip = trace->func;
833 entry->parent_ip = trace->ret;
834 entry->rettime = trace->rettime;
835 entry->calltime = trace->calltime;
836 ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
841 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
842 unsigned long ip, unsigned long parent_ip, unsigned long flags,
845 if (likely(!atomic_read(&data->disabled)))
846 trace_function(tr, data, ip, parent_ip, flags, pc);
849 static void ftrace_trace_stack(struct trace_array *tr,
850 struct trace_array_cpu *data,
854 #ifdef CONFIG_STACKTRACE
855 struct ring_buffer_event *event;
856 struct stack_entry *entry;
857 struct stack_trace trace;
858 unsigned long irq_flags;
860 if (!(trace_flags & TRACE_ITER_STACKTRACE))
863 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
867 entry = ring_buffer_event_data(event);
868 tracing_generic_entry_update(&entry->ent, flags, pc);
869 entry->ent.type = TRACE_STACK;
871 memset(&entry->caller, 0, sizeof(entry->caller));
873 trace.nr_entries = 0;
874 trace.max_entries = FTRACE_STACK_ENTRIES;
876 trace.entries = entry->caller;
878 save_stack_trace(&trace);
879 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
883 void __trace_stack(struct trace_array *tr,
884 struct trace_array_cpu *data,
888 ftrace_trace_stack(tr, data, flags, skip, preempt_count());
892 ftrace_trace_special(void *__tr, void *__data,
893 unsigned long arg1, unsigned long arg2, unsigned long arg3,
896 struct ring_buffer_event *event;
897 struct trace_array_cpu *data = __data;
898 struct trace_array *tr = __tr;
899 struct special_entry *entry;
900 unsigned long irq_flags;
902 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
906 entry = ring_buffer_event_data(event);
907 tracing_generic_entry_update(&entry->ent, 0, pc);
908 entry->ent.type = TRACE_SPECIAL;
912 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
913 ftrace_trace_stack(tr, data, irq_flags, 4, pc);
919 __trace_special(void *__tr, void *__data,
920 unsigned long arg1, unsigned long arg2, unsigned long arg3)
922 ftrace_trace_special(__tr, __data, arg1, arg2, arg3, preempt_count());
926 tracing_sched_switch_trace(struct trace_array *tr,
927 struct trace_array_cpu *data,
928 struct task_struct *prev,
929 struct task_struct *next,
930 unsigned long flags, int pc)
932 struct ring_buffer_event *event;
933 struct ctx_switch_entry *entry;
934 unsigned long irq_flags;
936 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
940 entry = ring_buffer_event_data(event);
941 tracing_generic_entry_update(&entry->ent, flags, pc);
942 entry->ent.type = TRACE_CTX;
943 entry->prev_pid = prev->pid;
944 entry->prev_prio = prev->prio;
945 entry->prev_state = prev->state;
946 entry->next_pid = next->pid;
947 entry->next_prio = next->prio;
948 entry->next_state = next->state;
949 entry->next_cpu = task_cpu(next);
950 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
951 ftrace_trace_stack(tr, data, flags, 5, pc);
955 tracing_sched_wakeup_trace(struct trace_array *tr,
956 struct trace_array_cpu *data,
957 struct task_struct *wakee,
958 struct task_struct *curr,
959 unsigned long flags, int pc)
961 struct ring_buffer_event *event;
962 struct ctx_switch_entry *entry;
963 unsigned long irq_flags;
965 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
969 entry = ring_buffer_event_data(event);
970 tracing_generic_entry_update(&entry->ent, flags, pc);
971 entry->ent.type = TRACE_WAKE;
972 entry->prev_pid = curr->pid;
973 entry->prev_prio = curr->prio;
974 entry->prev_state = curr->state;
975 entry->next_pid = wakee->pid;
976 entry->next_prio = wakee->prio;
977 entry->next_state = wakee->state;
978 entry->next_cpu = task_cpu(wakee);
979 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
980 ftrace_trace_stack(tr, data, flags, 6, pc);
986 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
988 struct trace_array *tr = &global_trace;
989 struct trace_array_cpu *data;
994 if (tracing_disabled)
997 pc = preempt_count();
998 local_irq_save(flags);
999 cpu = raw_smp_processor_id();
1000 data = tr->data[cpu];
1002 if (likely(atomic_inc_return(&data->disabled) == 1))
1003 ftrace_trace_special(tr, data, arg1, arg2, arg3, pc);
1005 atomic_dec(&data->disabled);
1006 local_irq_restore(flags);
1009 #ifdef CONFIG_FUNCTION_TRACER
1011 function_trace_call_preempt_only(unsigned long ip, unsigned long parent_ip)
1013 struct trace_array *tr = &global_trace;
1014 struct trace_array_cpu *data;
1015 unsigned long flags;
1020 if (unlikely(!ftrace_function_enabled))
1023 pc = preempt_count();
1024 resched = ftrace_preempt_disable();
1025 local_save_flags(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 trace_function(tr, data, ip, parent_ip, flags, pc);
1033 atomic_dec(&data->disabled);
1034 ftrace_preempt_enable(resched);
1038 function_trace_call(unsigned long ip, unsigned long parent_ip)
1040 struct trace_array *tr = &global_trace;
1041 struct trace_array_cpu *data;
1042 unsigned long flags;
1047 if (unlikely(!ftrace_function_enabled))
1051 * Need to use raw, since this must be called before the
1052 * recursive protection is performed.
1054 local_irq_save(flags);
1055 cpu = raw_smp_processor_id();
1056 data = tr->data[cpu];
1057 disabled = atomic_inc_return(&data->disabled);
1059 if (likely(disabled == 1)) {
1060 pc = preempt_count();
1061 trace_function(tr, data, ip, parent_ip, flags, pc);
1064 atomic_dec(&data->disabled);
1065 local_irq_restore(flags);
1068 #ifdef CONFIG_FUNCTION_RET_TRACER
1069 void trace_function_return(struct ftrace_retfunc *trace)
1071 struct trace_array *tr = &global_trace;
1072 struct trace_array_cpu *data;
1073 unsigned long flags;
1078 raw_local_irq_save(flags);
1079 cpu = raw_smp_processor_id();
1080 data = tr->data[cpu];
1081 disabled = atomic_inc_return(&data->disabled);
1082 if (likely(disabled == 1)) {
1083 pc = preempt_count();
1084 __trace_function_return(tr, data, trace, flags, pc);
1086 atomic_dec(&data->disabled);
1087 raw_local_irq_restore(flags);
1089 #endif /* CONFIG_FUNCTION_RET_TRACER */
1091 static struct ftrace_ops trace_ops __read_mostly =
1093 .func = function_trace_call,
1096 void tracing_start_function_trace(void)
1098 ftrace_function_enabled = 0;
1100 if (trace_flags & TRACE_ITER_PREEMPTONLY)
1101 trace_ops.func = function_trace_call_preempt_only;
1103 trace_ops.func = function_trace_call;
1105 register_ftrace_function(&trace_ops);
1106 ftrace_function_enabled = 1;
1109 void tracing_stop_function_trace(void)
1111 ftrace_function_enabled = 0;
1112 unregister_ftrace_function(&trace_ops);
1116 enum trace_file_type {
1117 TRACE_FILE_LAT_FMT = 1,
1118 TRACE_FILE_ANNOTATE = 2,
1121 static void trace_iterator_increment(struct trace_iterator *iter, int cpu)
1123 /* Don't allow ftrace to trace into the ring buffers */
1124 ftrace_disable_cpu();
1127 if (iter->buffer_iter[iter->cpu])
1128 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1130 ftrace_enable_cpu();
1133 static struct trace_entry *
1134 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1136 struct ring_buffer_event *event;
1137 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1139 /* Don't allow ftrace to trace into the ring buffers */
1140 ftrace_disable_cpu();
1143 event = ring_buffer_iter_peek(buf_iter, ts);
1145 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1147 ftrace_enable_cpu();
1149 return event ? ring_buffer_event_data(event) : NULL;
1152 static struct trace_entry *
1153 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1155 struct ring_buffer *buffer = iter->tr->buffer;
1156 struct trace_entry *ent, *next = NULL;
1157 u64 next_ts = 0, ts;
1161 for_each_tracing_cpu(cpu) {
1163 if (ring_buffer_empty_cpu(buffer, cpu))
1166 ent = peek_next_entry(iter, cpu, &ts);
1169 * Pick the entry with the smallest timestamp:
1171 if (ent && (!next || ts < next_ts)) {
1179 *ent_cpu = next_cpu;
1187 /* Find the next real entry, without updating the iterator itself */
1188 static struct trace_entry *
1189 find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1191 return __find_next_entry(iter, ent_cpu, ent_ts);
1194 /* Find the next real entry, and increment the iterator to the next entry */
1195 static void *find_next_entry_inc(struct trace_iterator *iter)
1197 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1200 trace_iterator_increment(iter, iter->cpu);
1202 return iter->ent ? iter : NULL;
1205 static void trace_consume(struct trace_iterator *iter)
1207 /* Don't allow ftrace to trace into the ring buffers */
1208 ftrace_disable_cpu();
1209 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1210 ftrace_enable_cpu();
1213 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1215 struct trace_iterator *iter = m->private;
1221 /* can't go backwards */
1226 ent = find_next_entry_inc(iter);
1230 while (ent && iter->idx < i)
1231 ent = find_next_entry_inc(iter);
1238 static void *s_start(struct seq_file *m, loff_t *pos)
1240 struct trace_iterator *iter = m->private;
1245 mutex_lock(&trace_types_lock);
1247 if (!current_trace || current_trace != iter->trace) {
1248 mutex_unlock(&trace_types_lock);
1252 atomic_inc(&trace_record_cmdline_disabled);
1254 if (*pos != iter->pos) {
1259 ftrace_disable_cpu();
1261 for_each_tracing_cpu(cpu) {
1262 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1265 ftrace_enable_cpu();
1267 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1272 p = s_next(m, p, &l);
1278 static void s_stop(struct seq_file *m, void *p)
1280 atomic_dec(&trace_record_cmdline_disabled);
1281 mutex_unlock(&trace_types_lock);
1284 #ifdef CONFIG_KRETPROBES
1285 static inline const char *kretprobed(const char *name)
1287 static const char tramp_name[] = "kretprobe_trampoline";
1288 int size = sizeof(tramp_name);
1290 if (strncmp(tramp_name, name, size) == 0)
1291 return "[unknown/kretprobe'd]";
1295 static inline const char *kretprobed(const char *name)
1299 #endif /* CONFIG_KRETPROBES */
1302 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1304 #ifdef CONFIG_KALLSYMS
1305 char str[KSYM_SYMBOL_LEN];
1308 kallsyms_lookup(address, NULL, NULL, NULL, str);
1310 name = kretprobed(str);
1312 return trace_seq_printf(s, fmt, name);
1318 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1319 unsigned long address)
1321 #ifdef CONFIG_KALLSYMS
1322 char str[KSYM_SYMBOL_LEN];
1325 sprint_symbol(str, address);
1326 name = kretprobed(str);
1328 return trace_seq_printf(s, fmt, name);
1333 #ifndef CONFIG_64BIT
1334 # define IP_FMT "%08lx"
1336 # define IP_FMT "%016lx"
1340 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1345 return trace_seq_printf(s, "0");
1347 if (sym_flags & TRACE_ITER_SYM_OFFSET)
1348 ret = seq_print_sym_offset(s, "%s", ip);
1350 ret = seq_print_sym_short(s, "%s", ip);
1355 if (sym_flags & TRACE_ITER_SYM_ADDR)
1356 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1360 static void print_lat_help_header(struct seq_file *m)
1362 seq_puts(m, "# _------=> CPU# \n");
1363 seq_puts(m, "# / _-----=> irqs-off \n");
1364 seq_puts(m, "# | / _----=> need-resched \n");
1365 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1366 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1367 seq_puts(m, "# |||| / \n");
1368 seq_puts(m, "# ||||| delay \n");
1369 seq_puts(m, "# cmd pid ||||| time | caller \n");
1370 seq_puts(m, "# \\ / ||||| \\ | / \n");
1373 static void print_func_help_header(struct seq_file *m)
1375 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1376 seq_puts(m, "# | | | | |\n");
1381 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1383 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1384 struct trace_array *tr = iter->tr;
1385 struct trace_array_cpu *data = tr->data[tr->cpu];
1386 struct tracer *type = current_trace;
1387 unsigned long total;
1388 unsigned long entries;
1389 const char *name = "preemption";
1394 entries = ring_buffer_entries(iter->tr->buffer);
1396 ring_buffer_overruns(iter->tr->buffer);
1398 seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1400 seq_puts(m, "-----------------------------------"
1401 "---------------------------------\n");
1402 seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1403 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1404 nsecs_to_usecs(data->saved_latency),
1408 #if defined(CONFIG_PREEMPT_NONE)
1410 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1412 #elif defined(CONFIG_PREEMPT)
1417 /* These are reserved for later use */
1420 seq_printf(m, " #P:%d)\n", num_online_cpus());
1424 seq_puts(m, " -----------------\n");
1425 seq_printf(m, " | task: %.16s-%d "
1426 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1427 data->comm, data->pid, data->uid, data->nice,
1428 data->policy, data->rt_priority);
1429 seq_puts(m, " -----------------\n");
1431 if (data->critical_start) {
1432 seq_puts(m, " => started at: ");
1433 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1434 trace_print_seq(m, &iter->seq);
1435 seq_puts(m, "\n => ended at: ");
1436 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1437 trace_print_seq(m, &iter->seq);
1445 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1447 int hardirq, softirq;
1450 comm = trace_find_cmdline(entry->pid);
1452 trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1453 trace_seq_printf(s, "%3d", cpu);
1454 trace_seq_printf(s, "%c%c",
1455 (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' :
1456 (entry->flags & TRACE_FLAG_IRQS_NOSUPPORT) ? 'X' : '.',
1457 ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1459 hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1460 softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1461 if (hardirq && softirq) {
1462 trace_seq_putc(s, 'H');
1465 trace_seq_putc(s, 'h');
1468 trace_seq_putc(s, 's');
1470 trace_seq_putc(s, '.');
1474 if (entry->preempt_count)
1475 trace_seq_printf(s, "%x", entry->preempt_count);
1477 trace_seq_puts(s, ".");
1480 unsigned long preempt_mark_thresh = 100;
1483 lat_print_timestamp(struct trace_seq *s, u64 abs_usecs,
1484 unsigned long rel_usecs)
1486 trace_seq_printf(s, " %4lldus", abs_usecs);
1487 if (rel_usecs > preempt_mark_thresh)
1488 trace_seq_puts(s, "!: ");
1489 else if (rel_usecs > 1)
1490 trace_seq_puts(s, "+: ");
1492 trace_seq_puts(s, " : ");
1495 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1498 * The message is supposed to contain an ending newline.
1499 * If the printing stops prematurely, try to add a newline of our own.
1501 void trace_seq_print_cont(struct trace_seq *s, struct trace_iterator *iter)
1503 struct trace_entry *ent;
1504 struct trace_field_cont *cont;
1507 ent = peek_next_entry(iter, iter->cpu, NULL);
1508 if (!ent || ent->type != TRACE_CONT) {
1509 trace_seq_putc(s, '\n');
1514 cont = (struct trace_field_cont *)ent;
1516 ok = (trace_seq_printf(s, "%s", cont->buf) > 0);
1518 ftrace_disable_cpu();
1520 if (iter->buffer_iter[iter->cpu])
1521 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1523 ring_buffer_consume(iter->tr->buffer, iter->cpu, NULL);
1525 ftrace_enable_cpu();
1527 ent = peek_next_entry(iter, iter->cpu, NULL);
1528 } while (ent && ent->type == TRACE_CONT);
1531 trace_seq_putc(s, '\n');
1534 static void test_cpu_buff_start(struct trace_iterator *iter)
1536 struct trace_seq *s = &iter->seq;
1538 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1541 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1544 if (cpu_isset(iter->cpu, iter->started))
1547 cpu_set(iter->cpu, iter->started);
1548 trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1551 static enum print_line_t
1552 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1554 struct trace_seq *s = &iter->seq;
1555 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1556 struct trace_entry *next_entry;
1557 unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1558 struct trace_entry *entry = iter->ent;
1559 unsigned long abs_usecs;
1560 unsigned long rel_usecs;
1567 if (entry->type == TRACE_CONT)
1568 return TRACE_TYPE_HANDLED;
1570 test_cpu_buff_start(iter);
1572 next_entry = find_next_entry(iter, NULL, &next_ts);
1575 rel_usecs = ns2usecs(next_ts - iter->ts);
1576 abs_usecs = ns2usecs(iter->ts - iter->tr->time_start);
1579 comm = trace_find_cmdline(entry->pid);
1580 trace_seq_printf(s, "%16s %5d %3d %d %08x %08x [%08lx]"
1581 " %ld.%03ldms (+%ld.%03ldms): ",
1583 entry->pid, cpu, entry->flags,
1584 entry->preempt_count, trace_idx,
1587 abs_usecs % 1000, rel_usecs/1000,
1590 lat_print_generic(s, entry, cpu);
1591 lat_print_timestamp(s, abs_usecs, rel_usecs);
1593 switch (entry->type) {
1595 struct ftrace_entry *field;
1597 trace_assign_type(field, entry);
1599 seq_print_ip_sym(s, field->ip, sym_flags);
1600 trace_seq_puts(s, " (");
1601 seq_print_ip_sym(s, field->parent_ip, sym_flags);
1602 trace_seq_puts(s, ")\n");
1607 struct ctx_switch_entry *field;
1609 trace_assign_type(field, entry);
1611 T = field->next_state < sizeof(state_to_char) ?
1612 state_to_char[field->next_state] : 'X';
1614 state = field->prev_state ?
1615 __ffs(field->prev_state) + 1 : 0;
1616 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1617 comm = trace_find_cmdline(field->next_pid);
1618 trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c %s\n",
1621 S, entry->type == TRACE_CTX ? "==>" : " +",
1628 case TRACE_SPECIAL: {
1629 struct special_entry *field;
1631 trace_assign_type(field, entry);
1633 trace_seq_printf(s, "# %ld %ld %ld\n",
1640 struct stack_entry *field;
1642 trace_assign_type(field, entry);
1644 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1646 trace_seq_puts(s, " <= ");
1647 seq_print_ip_sym(s, field->caller[i], sym_flags);
1649 trace_seq_puts(s, "\n");
1653 struct print_entry *field;
1655 trace_assign_type(field, entry);
1657 seq_print_ip_sym(s, field->ip, sym_flags);
1658 trace_seq_printf(s, ": %s", field->buf);
1659 if (entry->flags & TRACE_FLAG_CONT)
1660 trace_seq_print_cont(s, iter);
1663 case TRACE_BRANCH: {
1664 struct trace_branch *field;
1666 trace_assign_type(field, entry);
1668 trace_seq_printf(s, "[%s] %s:%s:%d\n",
1669 field->correct ? " ok " : " MISS ",
1676 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1678 return TRACE_TYPE_HANDLED;
1681 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1683 struct trace_seq *s = &iter->seq;
1684 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1685 struct trace_entry *entry;
1686 unsigned long usec_rem;
1687 unsigned long long t;
1696 if (entry->type == TRACE_CONT)
1697 return TRACE_TYPE_HANDLED;
1699 test_cpu_buff_start(iter);
1701 comm = trace_find_cmdline(iter->ent->pid);
1703 t = ns2usecs(iter->ts);
1704 usec_rem = do_div(t, 1000000ULL);
1705 secs = (unsigned long)t;
1707 ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1709 return TRACE_TYPE_PARTIAL_LINE;
1710 ret = trace_seq_printf(s, "[%03d] ", iter->cpu);
1712 return TRACE_TYPE_PARTIAL_LINE;
1713 ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1715 return TRACE_TYPE_PARTIAL_LINE;
1717 switch (entry->type) {
1719 struct ftrace_entry *field;
1721 trace_assign_type(field, entry);
1723 ret = seq_print_ip_sym(s, field->ip, sym_flags);
1725 return TRACE_TYPE_PARTIAL_LINE;
1726 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1728 ret = trace_seq_printf(s, " <-");
1730 return TRACE_TYPE_PARTIAL_LINE;
1731 ret = seq_print_ip_sym(s,
1735 return TRACE_TYPE_PARTIAL_LINE;
1737 ret = trace_seq_printf(s, "\n");
1739 return TRACE_TYPE_PARTIAL_LINE;
1744 struct ctx_switch_entry *field;
1746 trace_assign_type(field, entry);
1748 S = field->prev_state < sizeof(state_to_char) ?
1749 state_to_char[field->prev_state] : 'X';
1750 T = field->next_state < sizeof(state_to_char) ?
1751 state_to_char[field->next_state] : 'X';
1752 ret = trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c\n",
1756 entry->type == TRACE_CTX ? "==>" : " +",
1762 return TRACE_TYPE_PARTIAL_LINE;
1765 case TRACE_SPECIAL: {
1766 struct special_entry *field;
1768 trace_assign_type(field, entry);
1770 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1775 return TRACE_TYPE_PARTIAL_LINE;
1779 struct stack_entry *field;
1781 trace_assign_type(field, entry);
1783 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1785 ret = trace_seq_puts(s, " <= ");
1787 return TRACE_TYPE_PARTIAL_LINE;
1789 ret = seq_print_ip_sym(s, field->caller[i],
1792 return TRACE_TYPE_PARTIAL_LINE;
1794 ret = trace_seq_puts(s, "\n");
1796 return TRACE_TYPE_PARTIAL_LINE;
1800 struct print_entry *field;
1802 trace_assign_type(field, entry);
1804 seq_print_ip_sym(s, field->ip, sym_flags);
1805 trace_seq_printf(s, ": %s", field->buf);
1806 if (entry->flags & TRACE_FLAG_CONT)
1807 trace_seq_print_cont(s, iter);
1810 case TRACE_FN_RET: {
1811 return print_return_function(iter);
1814 case TRACE_BRANCH: {
1815 struct trace_branch *field;
1817 trace_assign_type(field, entry);
1819 trace_seq_printf(s, "[%s] %s:%s:%d\n",
1820 field->correct ? " ok " : " MISS ",
1827 return TRACE_TYPE_HANDLED;
1830 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1832 struct trace_seq *s = &iter->seq;
1833 struct trace_entry *entry;
1839 if (entry->type == TRACE_CONT)
1840 return TRACE_TYPE_HANDLED;
1842 ret = trace_seq_printf(s, "%d %d %llu ",
1843 entry->pid, iter->cpu, iter->ts);
1845 return TRACE_TYPE_PARTIAL_LINE;
1847 switch (entry->type) {
1849 struct ftrace_entry *field;
1851 trace_assign_type(field, entry);
1853 ret = trace_seq_printf(s, "%x %x\n",
1857 return TRACE_TYPE_PARTIAL_LINE;
1862 struct ctx_switch_entry *field;
1864 trace_assign_type(field, entry);
1866 S = field->prev_state < sizeof(state_to_char) ?
1867 state_to_char[field->prev_state] : 'X';
1868 T = field->next_state < sizeof(state_to_char) ?
1869 state_to_char[field->next_state] : 'X';
1870 if (entry->type == TRACE_WAKE)
1872 ret = trace_seq_printf(s, "%d %d %c %d %d %d %c\n",
1881 return TRACE_TYPE_PARTIAL_LINE;
1886 struct special_entry *field;
1888 trace_assign_type(field, entry);
1890 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1895 return TRACE_TYPE_PARTIAL_LINE;
1899 struct print_entry *field;
1901 trace_assign_type(field, entry);
1903 trace_seq_printf(s, "# %lx %s", field->ip, field->buf);
1904 if (entry->flags & TRACE_FLAG_CONT)
1905 trace_seq_print_cont(s, iter);
1909 return TRACE_TYPE_HANDLED;
1912 #define SEQ_PUT_FIELD_RET(s, x) \
1914 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
1918 #define SEQ_PUT_HEX_FIELD_RET(s, x) \
1920 BUILD_BUG_ON(sizeof(x) > MAX_MEMHEX_BYTES); \
1921 if (!trace_seq_putmem_hex(s, &(x), sizeof(x))) \
1925 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1927 struct trace_seq *s = &iter->seq;
1928 unsigned char newline = '\n';
1929 struct trace_entry *entry;
1934 if (entry->type == TRACE_CONT)
1935 return TRACE_TYPE_HANDLED;
1937 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1938 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1939 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1941 switch (entry->type) {
1943 struct ftrace_entry *field;
1945 trace_assign_type(field, entry);
1947 SEQ_PUT_HEX_FIELD_RET(s, field->ip);
1948 SEQ_PUT_HEX_FIELD_RET(s, field->parent_ip);
1953 struct ctx_switch_entry *field;
1955 trace_assign_type(field, entry);
1957 S = field->prev_state < sizeof(state_to_char) ?
1958 state_to_char[field->prev_state] : 'X';
1959 T = field->next_state < sizeof(state_to_char) ?
1960 state_to_char[field->next_state] : 'X';
1961 if (entry->type == TRACE_WAKE)
1963 SEQ_PUT_HEX_FIELD_RET(s, field->prev_pid);
1964 SEQ_PUT_HEX_FIELD_RET(s, field->prev_prio);
1965 SEQ_PUT_HEX_FIELD_RET(s, S);
1966 SEQ_PUT_HEX_FIELD_RET(s, field->next_cpu);
1967 SEQ_PUT_HEX_FIELD_RET(s, field->next_pid);
1968 SEQ_PUT_HEX_FIELD_RET(s, field->next_prio);
1969 SEQ_PUT_HEX_FIELD_RET(s, T);
1974 struct special_entry *field;
1976 trace_assign_type(field, entry);
1978 SEQ_PUT_HEX_FIELD_RET(s, field->arg1);
1979 SEQ_PUT_HEX_FIELD_RET(s, field->arg2);
1980 SEQ_PUT_HEX_FIELD_RET(s, field->arg3);
1984 SEQ_PUT_FIELD_RET(s, newline);
1986 return TRACE_TYPE_HANDLED;
1989 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1991 struct trace_seq *s = &iter->seq;
1992 struct trace_entry *entry;
1996 if (entry->type == TRACE_CONT)
1997 return TRACE_TYPE_HANDLED;
1999 SEQ_PUT_FIELD_RET(s, entry->pid);
2000 SEQ_PUT_FIELD_RET(s, entry->cpu);
2001 SEQ_PUT_FIELD_RET(s, iter->ts);
2003 switch (entry->type) {
2005 struct ftrace_entry *field;
2007 trace_assign_type(field, entry);
2009 SEQ_PUT_FIELD_RET(s, field->ip);
2010 SEQ_PUT_FIELD_RET(s, field->parent_ip);
2014 struct ctx_switch_entry *field;
2016 trace_assign_type(field, entry);
2018 SEQ_PUT_FIELD_RET(s, field->prev_pid);
2019 SEQ_PUT_FIELD_RET(s, field->prev_prio);
2020 SEQ_PUT_FIELD_RET(s, field->prev_state);
2021 SEQ_PUT_FIELD_RET(s, field->next_pid);
2022 SEQ_PUT_FIELD_RET(s, field->next_prio);
2023 SEQ_PUT_FIELD_RET(s, field->next_state);
2028 struct special_entry *field;
2030 trace_assign_type(field, entry);
2032 SEQ_PUT_FIELD_RET(s, field->arg1);
2033 SEQ_PUT_FIELD_RET(s, field->arg2);
2034 SEQ_PUT_FIELD_RET(s, field->arg3);
2041 static int trace_empty(struct trace_iterator *iter)
2045 for_each_tracing_cpu(cpu) {
2046 if (iter->buffer_iter[cpu]) {
2047 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2050 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2058 static enum print_line_t print_trace_line(struct trace_iterator *iter)
2060 enum print_line_t ret;
2062 if (iter->trace && iter->trace->print_line) {
2063 ret = iter->trace->print_line(iter);
2064 if (ret != TRACE_TYPE_UNHANDLED)
2068 if (trace_flags & TRACE_ITER_BIN)
2069 return print_bin_fmt(iter);
2071 if (trace_flags & TRACE_ITER_HEX)
2072 return print_hex_fmt(iter);
2074 if (trace_flags & TRACE_ITER_RAW)
2075 return print_raw_fmt(iter);
2077 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2078 return print_lat_fmt(iter, iter->idx, iter->cpu);
2080 return print_trace_fmt(iter);
2083 static int s_show(struct seq_file *m, void *v)
2085 struct trace_iterator *iter = v;
2087 if (iter->ent == NULL) {
2089 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2092 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2093 /* print nothing if the buffers are empty */
2094 if (trace_empty(iter))
2096 print_trace_header(m, iter);
2097 if (!(trace_flags & TRACE_ITER_VERBOSE))
2098 print_lat_help_header(m);
2100 if (!(trace_flags & TRACE_ITER_VERBOSE))
2101 print_func_help_header(m);
2104 print_trace_line(iter);
2105 trace_print_seq(m, &iter->seq);
2111 static struct seq_operations tracer_seq_ops = {
2118 static struct trace_iterator *
2119 __tracing_open(struct inode *inode, struct file *file, int *ret)
2121 struct trace_iterator *iter;
2125 if (tracing_disabled) {
2130 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2136 mutex_lock(&trace_types_lock);
2137 if (current_trace && current_trace->print_max)
2140 iter->tr = inode->i_private;
2141 iter->trace = current_trace;
2144 /* Annotate start of buffers if we had overruns */
2145 if (ring_buffer_overruns(iter->tr->buffer))
2146 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2149 for_each_tracing_cpu(cpu) {
2151 iter->buffer_iter[cpu] =
2152 ring_buffer_read_start(iter->tr->buffer, cpu);
2154 if (!iter->buffer_iter[cpu])
2158 /* TODO stop tracer */
2159 *ret = seq_open(file, &tracer_seq_ops);
2163 m = file->private_data;
2166 /* stop the trace while dumping */
2169 if (iter->trace && iter->trace->open)
2170 iter->trace->open(iter);
2172 mutex_unlock(&trace_types_lock);
2178 for_each_tracing_cpu(cpu) {
2179 if (iter->buffer_iter[cpu])
2180 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2182 mutex_unlock(&trace_types_lock);
2184 return ERR_PTR(-ENOMEM);
2187 int tracing_open_generic(struct inode *inode, struct file *filp)
2189 if (tracing_disabled)
2192 filp->private_data = inode->i_private;
2196 int tracing_release(struct inode *inode, struct file *file)
2198 struct seq_file *m = (struct seq_file *)file->private_data;
2199 struct trace_iterator *iter = m->private;
2202 mutex_lock(&trace_types_lock);
2203 for_each_tracing_cpu(cpu) {
2204 if (iter->buffer_iter[cpu])
2205 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2208 if (iter->trace && iter->trace->close)
2209 iter->trace->close(iter);
2211 /* reenable tracing if it was previously enabled */
2213 mutex_unlock(&trace_types_lock);
2215 seq_release(inode, file);
2220 static int tracing_open(struct inode *inode, struct file *file)
2224 __tracing_open(inode, file, &ret);
2229 static int tracing_lt_open(struct inode *inode, struct file *file)
2231 struct trace_iterator *iter;
2234 iter = __tracing_open(inode, file, &ret);
2237 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2244 t_next(struct seq_file *m, void *v, loff_t *pos)
2246 struct tracer *t = m->private;
2258 static void *t_start(struct seq_file *m, loff_t *pos)
2260 struct tracer *t = m->private;
2263 mutex_lock(&trace_types_lock);
2264 for (; t && l < *pos; t = t_next(m, t, &l))
2270 static void t_stop(struct seq_file *m, void *p)
2272 mutex_unlock(&trace_types_lock);
2275 static int t_show(struct seq_file *m, void *v)
2277 struct tracer *t = v;
2282 seq_printf(m, "%s", t->name);
2291 static struct seq_operations show_traces_seq_ops = {
2298 static int show_traces_open(struct inode *inode, struct file *file)
2302 if (tracing_disabled)
2305 ret = seq_open(file, &show_traces_seq_ops);
2307 struct seq_file *m = file->private_data;
2308 m->private = trace_types;
2314 static struct file_operations tracing_fops = {
2315 .open = tracing_open,
2317 .llseek = seq_lseek,
2318 .release = tracing_release,
2321 static struct file_operations tracing_lt_fops = {
2322 .open = tracing_lt_open,
2324 .llseek = seq_lseek,
2325 .release = tracing_release,
2328 static struct file_operations show_traces_fops = {
2329 .open = show_traces_open,
2331 .release = seq_release,
2335 * Only trace on a CPU if the bitmask is set:
2337 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
2340 * When tracing/tracing_cpu_mask is modified then this holds
2341 * the new bitmask we are about to install:
2343 static cpumask_t tracing_cpumask_new;
2346 * The tracer itself will not take this lock, but still we want
2347 * to provide a consistent cpumask to user-space:
2349 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2352 * Temporary storage for the character representation of the
2353 * CPU bitmask (and one more byte for the newline):
2355 static char mask_str[NR_CPUS + 1];
2358 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2359 size_t count, loff_t *ppos)
2363 mutex_lock(&tracing_cpumask_update_lock);
2365 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2366 if (count - len < 2) {
2370 len += sprintf(mask_str + len, "\n");
2371 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2374 mutex_unlock(&tracing_cpumask_update_lock);
2380 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2381 size_t count, loff_t *ppos)
2385 mutex_lock(&tracing_cpumask_update_lock);
2386 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2390 raw_local_irq_disable();
2391 __raw_spin_lock(&ftrace_max_lock);
2392 for_each_tracing_cpu(cpu) {
2394 * Increase/decrease the disabled counter if we are
2395 * about to flip a bit in the cpumask:
2397 if (cpu_isset(cpu, tracing_cpumask) &&
2398 !cpu_isset(cpu, tracing_cpumask_new)) {
2399 atomic_inc(&global_trace.data[cpu]->disabled);
2401 if (!cpu_isset(cpu, tracing_cpumask) &&
2402 cpu_isset(cpu, tracing_cpumask_new)) {
2403 atomic_dec(&global_trace.data[cpu]->disabled);
2406 __raw_spin_unlock(&ftrace_max_lock);
2407 raw_local_irq_enable();
2409 tracing_cpumask = tracing_cpumask_new;
2411 mutex_unlock(&tracing_cpumask_update_lock);
2416 mutex_unlock(&tracing_cpumask_update_lock);
2421 static struct file_operations tracing_cpumask_fops = {
2422 .open = tracing_open_generic,
2423 .read = tracing_cpumask_read,
2424 .write = tracing_cpumask_write,
2428 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2429 size_t cnt, loff_t *ppos)
2436 /* calulate max size */
2437 for (i = 0; trace_options[i]; i++) {
2438 len += strlen(trace_options[i]);
2439 len += 3; /* "no" and space */
2442 /* +2 for \n and \0 */
2443 buf = kmalloc(len + 2, GFP_KERNEL);
2447 for (i = 0; trace_options[i]; i++) {
2448 if (trace_flags & (1 << i))
2449 r += sprintf(buf + r, "%s ", trace_options[i]);
2451 r += sprintf(buf + r, "no%s ", trace_options[i]);
2454 r += sprintf(buf + r, "\n");
2455 WARN_ON(r >= len + 2);
2457 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2465 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2466 size_t cnt, loff_t *ppos)
2473 if (cnt >= sizeof(buf))
2476 if (copy_from_user(&buf, ubuf, cnt))
2481 if (strncmp(buf, "no", 2) == 0) {
2486 for (i = 0; trace_options[i]; i++) {
2487 int len = strlen(trace_options[i]);
2489 if (strncmp(cmp, trace_options[i], len) == 0) {
2491 trace_flags &= ~(1 << i);
2493 trace_flags |= (1 << i);
2498 * If no option could be set, return an error:
2500 if (!trace_options[i])
2508 static struct file_operations tracing_iter_fops = {
2509 .open = tracing_open_generic,
2510 .read = tracing_trace_options_read,
2511 .write = tracing_trace_options_write,
2514 static const char readme_msg[] =
2515 "tracing mini-HOWTO:\n\n"
2517 "# mount -t debugfs nodev /debug\n\n"
2518 "# cat /debug/tracing/available_tracers\n"
2519 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2520 "# cat /debug/tracing/current_tracer\n"
2522 "# echo sched_switch > /debug/tracing/current_tracer\n"
2523 "# cat /debug/tracing/current_tracer\n"
2525 "# cat /debug/tracing/trace_options\n"
2526 "noprint-parent nosym-offset nosym-addr noverbose\n"
2527 "# echo print-parent > /debug/tracing/trace_options\n"
2528 "# echo 1 > /debug/tracing/tracing_enabled\n"
2529 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2530 "echo 0 > /debug/tracing/tracing_enabled\n"
2534 tracing_readme_read(struct file *filp, char __user *ubuf,
2535 size_t cnt, loff_t *ppos)
2537 return simple_read_from_buffer(ubuf, cnt, ppos,
2538 readme_msg, strlen(readme_msg));
2541 static struct file_operations tracing_readme_fops = {
2542 .open = tracing_open_generic,
2543 .read = tracing_readme_read,
2547 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2548 size_t cnt, loff_t *ppos)
2553 r = sprintf(buf, "%u\n", tracer_enabled);
2554 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2558 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2559 size_t cnt, loff_t *ppos)
2561 struct trace_array *tr = filp->private_data;
2566 if (cnt >= sizeof(buf))
2569 if (copy_from_user(&buf, ubuf, cnt))
2574 ret = strict_strtoul(buf, 10, &val);
2580 mutex_lock(&trace_types_lock);
2581 if (tracer_enabled ^ val) {
2584 if (current_trace->start)
2585 current_trace->start(tr);
2590 if (current_trace->stop)
2591 current_trace->stop(tr);
2594 mutex_unlock(&trace_types_lock);
2602 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2603 size_t cnt, loff_t *ppos)
2605 char buf[max_tracer_type_len+2];
2608 mutex_lock(&trace_types_lock);
2610 r = sprintf(buf, "%s\n", current_trace->name);
2612 r = sprintf(buf, "\n");
2613 mutex_unlock(&trace_types_lock);
2615 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2618 static int tracing_set_tracer(char *buf)
2620 struct trace_array *tr = &global_trace;
2624 mutex_lock(&trace_types_lock);
2625 for (t = trace_types; t; t = t->next) {
2626 if (strcmp(t->name, buf) == 0)
2633 if (t == current_trace)
2636 trace_branch_disable();
2637 if (current_trace && current_trace->reset)
2638 current_trace->reset(tr);
2647 trace_branch_enable(tr);
2649 mutex_unlock(&trace_types_lock);
2655 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2656 size_t cnt, loff_t *ppos)
2658 char buf[max_tracer_type_len+1];
2665 if (cnt > max_tracer_type_len)
2666 cnt = max_tracer_type_len;
2668 if (copy_from_user(&buf, ubuf, cnt))
2673 /* strip ending whitespace. */
2674 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2677 err = tracing_set_tracer(buf);
2687 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2688 size_t cnt, loff_t *ppos)
2690 unsigned long *ptr = filp->private_data;
2694 r = snprintf(buf, sizeof(buf), "%ld\n",
2695 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2696 if (r > sizeof(buf))
2698 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2702 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2703 size_t cnt, loff_t *ppos)
2705 long *ptr = filp->private_data;
2710 if (cnt >= sizeof(buf))
2713 if (copy_from_user(&buf, ubuf, cnt))
2718 ret = strict_strtoul(buf, 10, &val);
2727 static atomic_t tracing_reader;
2729 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2731 struct trace_iterator *iter;
2733 if (tracing_disabled)
2736 /* We only allow for reader of the pipe */
2737 if (atomic_inc_return(&tracing_reader) != 1) {
2738 atomic_dec(&tracing_reader);
2742 /* create a buffer to store the information to pass to userspace */
2743 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2747 mutex_lock(&trace_types_lock);
2749 /* trace pipe does not show start of buffer */
2750 cpus_setall(iter->started);
2752 iter->tr = &global_trace;
2753 iter->trace = current_trace;
2754 filp->private_data = iter;
2756 if (iter->trace->pipe_open)
2757 iter->trace->pipe_open(iter);
2758 mutex_unlock(&trace_types_lock);
2763 static int tracing_release_pipe(struct inode *inode, struct file *file)
2765 struct trace_iterator *iter = file->private_data;
2768 atomic_dec(&tracing_reader);
2774 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2776 struct trace_iterator *iter = filp->private_data;
2778 if (trace_flags & TRACE_ITER_BLOCK) {
2780 * Always select as readable when in blocking mode
2782 return POLLIN | POLLRDNORM;
2784 if (!trace_empty(iter))
2785 return POLLIN | POLLRDNORM;
2786 poll_wait(filp, &trace_wait, poll_table);
2787 if (!trace_empty(iter))
2788 return POLLIN | POLLRDNORM;
2798 tracing_read_pipe(struct file *filp, char __user *ubuf,
2799 size_t cnt, loff_t *ppos)
2801 struct trace_iterator *iter = filp->private_data;
2804 /* return any leftover data */
2805 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2809 trace_seq_reset(&iter->seq);
2811 mutex_lock(&trace_types_lock);
2812 if (iter->trace->read) {
2813 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2820 while (trace_empty(iter)) {
2822 if ((filp->f_flags & O_NONBLOCK)) {
2828 * This is a make-shift waitqueue. The reason we don't use
2829 * an actual wait queue is because:
2830 * 1) we only ever have one waiter
2831 * 2) the tracing, traces all functions, we don't want
2832 * the overhead of calling wake_up and friends
2833 * (and tracing them too)
2834 * Anyway, this is really very primitive wakeup.
2836 set_current_state(TASK_INTERRUPTIBLE);
2837 iter->tr->waiter = current;
2839 mutex_unlock(&trace_types_lock);
2841 /* sleep for 100 msecs, and try again. */
2842 schedule_timeout(HZ/10);
2844 mutex_lock(&trace_types_lock);
2846 iter->tr->waiter = NULL;
2848 if (signal_pending(current)) {
2853 if (iter->trace != current_trace)
2857 * We block until we read something and tracing is disabled.
2858 * We still block if tracing is disabled, but we have never
2859 * read anything. This allows a user to cat this file, and
2860 * then enable tracing. But after we have read something,
2861 * we give an EOF when tracing is again disabled.
2863 * iter->pos will be 0 if we haven't read anything.
2865 if (!tracer_enabled && iter->pos)
2871 /* stop when tracing is finished */
2872 if (trace_empty(iter))
2875 if (cnt >= PAGE_SIZE)
2876 cnt = PAGE_SIZE - 1;
2878 /* reset all but tr, trace, and overruns */
2879 memset(&iter->seq, 0,
2880 sizeof(struct trace_iterator) -
2881 offsetof(struct trace_iterator, seq));
2884 while (find_next_entry_inc(iter) != NULL) {
2885 enum print_line_t ret;
2886 int len = iter->seq.len;
2888 ret = print_trace_line(iter);
2889 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2890 /* don't print partial lines */
2891 iter->seq.len = len;
2895 trace_consume(iter);
2897 if (iter->seq.len >= cnt)
2901 /* Now copy what we have to the user */
2902 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2903 if (iter->seq.readpos >= iter->seq.len)
2904 trace_seq_reset(&iter->seq);
2907 * If there was nothing to send to user, inspite of consuming trace
2908 * entries, go back to wait for more entries.
2914 mutex_unlock(&trace_types_lock);
2920 tracing_entries_read(struct file *filp, char __user *ubuf,
2921 size_t cnt, loff_t *ppos)
2923 struct trace_array *tr = filp->private_data;
2927 r = sprintf(buf, "%lu\n", tr->entries >> 10);
2928 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2932 tracing_entries_write(struct file *filp, const char __user *ubuf,
2933 size_t cnt, loff_t *ppos)
2939 if (cnt >= sizeof(buf))
2942 if (copy_from_user(&buf, ubuf, cnt))
2947 ret = strict_strtoul(buf, 10, &val);
2951 /* must have at least 1 entry */
2955 mutex_lock(&trace_types_lock);
2959 /* disable all cpu buffers */
2960 for_each_tracing_cpu(cpu) {
2961 if (global_trace.data[cpu])
2962 atomic_inc(&global_trace.data[cpu]->disabled);
2963 if (max_tr.data[cpu])
2964 atomic_inc(&max_tr.data[cpu]->disabled);
2967 /* value is in KB */
2970 if (val != global_trace.entries) {
2971 ret = ring_buffer_resize(global_trace.buffer, val);
2977 ret = ring_buffer_resize(max_tr.buffer, val);
2981 r = ring_buffer_resize(global_trace.buffer,
2982 global_trace.entries);
2984 /* AARGH! We are left with different
2985 * size max buffer!!!! */
2987 tracing_disabled = 1;
2992 global_trace.entries = val;
2997 /* If check pages failed, return ENOMEM */
2998 if (tracing_disabled)
3001 for_each_tracing_cpu(cpu) {
3002 if (global_trace.data[cpu])
3003 atomic_dec(&global_trace.data[cpu]->disabled);
3004 if (max_tr.data[cpu])
3005 atomic_dec(&max_tr.data[cpu]->disabled);
3009 max_tr.entries = global_trace.entries;
3010 mutex_unlock(&trace_types_lock);
3015 static int mark_printk(const char *fmt, ...)
3019 va_start(args, fmt);
3020 ret = trace_vprintk(0, fmt, args);
3026 tracing_mark_write(struct file *filp, const char __user *ubuf,
3027 size_t cnt, loff_t *fpos)
3032 if (tracing_disabled)
3035 if (cnt > TRACE_BUF_SIZE)
3036 cnt = TRACE_BUF_SIZE;
3038 buf = kmalloc(cnt + 1, GFP_KERNEL);
3042 if (copy_from_user(buf, ubuf, cnt)) {
3047 /* Cut from the first nil or newline. */
3049 end = strchr(buf, '\n');
3053 cnt = mark_printk("%s\n", buf);
3060 static struct file_operations tracing_max_lat_fops = {
3061 .open = tracing_open_generic,
3062 .read = tracing_max_lat_read,
3063 .write = tracing_max_lat_write,
3066 static struct file_operations tracing_ctrl_fops = {
3067 .open = tracing_open_generic,
3068 .read = tracing_ctrl_read,
3069 .write = tracing_ctrl_write,
3072 static struct file_operations set_tracer_fops = {
3073 .open = tracing_open_generic,
3074 .read = tracing_set_trace_read,
3075 .write = tracing_set_trace_write,
3078 static struct file_operations tracing_pipe_fops = {
3079 .open = tracing_open_pipe,
3080 .poll = tracing_poll_pipe,
3081 .read = tracing_read_pipe,
3082 .release = tracing_release_pipe,
3085 static struct file_operations tracing_entries_fops = {
3086 .open = tracing_open_generic,
3087 .read = tracing_entries_read,
3088 .write = tracing_entries_write,
3091 static struct file_operations tracing_mark_fops = {
3092 .open = tracing_open_generic,
3093 .write = tracing_mark_write,
3096 #ifdef CONFIG_DYNAMIC_FTRACE
3098 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3104 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3105 size_t cnt, loff_t *ppos)
3107 static char ftrace_dyn_info_buffer[1024];
3108 static DEFINE_MUTEX(dyn_info_mutex);
3109 unsigned long *p = filp->private_data;
3110 char *buf = ftrace_dyn_info_buffer;
3111 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3114 mutex_lock(&dyn_info_mutex);
3115 r = sprintf(buf, "%ld ", *p);
3117 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3120 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3122 mutex_unlock(&dyn_info_mutex);
3127 static struct file_operations tracing_dyn_info_fops = {
3128 .open = tracing_open_generic,
3129 .read = tracing_read_dyn_info,
3133 static struct dentry *d_tracer;
3135 struct dentry *tracing_init_dentry(void)
3142 d_tracer = debugfs_create_dir("tracing", NULL);
3144 if (!d_tracer && !once) {
3146 pr_warning("Could not create debugfs directory 'tracing'\n");
3153 #ifdef CONFIG_FTRACE_SELFTEST
3154 /* Let selftest have access to static functions in this file */
3155 #include "trace_selftest.c"
3158 static __init int tracer_init_debugfs(void)
3160 struct dentry *d_tracer;
3161 struct dentry *entry;
3163 d_tracer = tracing_init_dentry();
3165 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3166 &global_trace, &tracing_ctrl_fops);
3168 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3170 entry = debugfs_create_file("trace_options", 0644, d_tracer,
3171 NULL, &tracing_iter_fops);
3173 pr_warning("Could not create debugfs 'trace_options' entry\n");
3175 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3176 NULL, &tracing_cpumask_fops);
3178 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3180 entry = debugfs_create_file("latency_trace", 0444, d_tracer,
3181 &global_trace, &tracing_lt_fops);
3183 pr_warning("Could not create debugfs 'latency_trace' entry\n");
3185 entry = debugfs_create_file("trace", 0444, d_tracer,
3186 &global_trace, &tracing_fops);
3188 pr_warning("Could not create debugfs 'trace' entry\n");
3190 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3191 &global_trace, &show_traces_fops);
3193 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3195 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3196 &global_trace, &set_tracer_fops);
3198 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3200 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3201 &tracing_max_latency,
3202 &tracing_max_lat_fops);
3204 pr_warning("Could not create debugfs "
3205 "'tracing_max_latency' entry\n");
3207 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3208 &tracing_thresh, &tracing_max_lat_fops);
3210 pr_warning("Could not create debugfs "
3211 "'tracing_thresh' entry\n");
3212 entry = debugfs_create_file("README", 0644, d_tracer,
3213 NULL, &tracing_readme_fops);
3215 pr_warning("Could not create debugfs 'README' entry\n");
3217 entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
3218 NULL, &tracing_pipe_fops);
3220 pr_warning("Could not create debugfs "
3221 "'trace_pipe' entry\n");
3223 entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
3224 &global_trace, &tracing_entries_fops);
3226 pr_warning("Could not create debugfs "
3227 "'buffer_size_kb' entry\n");
3229 entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3230 NULL, &tracing_mark_fops);
3232 pr_warning("Could not create debugfs "
3233 "'trace_marker' entry\n");
3235 #ifdef CONFIG_DYNAMIC_FTRACE
3236 entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3237 &ftrace_update_tot_cnt,
3238 &tracing_dyn_info_fops);
3240 pr_warning("Could not create debugfs "
3241 "'dyn_ftrace_total_info' entry\n");
3243 #ifdef CONFIG_SYSPROF_TRACER
3244 init_tracer_sysprof_debugfs(d_tracer);
3249 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3251 static DEFINE_SPINLOCK(trace_buf_lock);
3252 static char trace_buf[TRACE_BUF_SIZE];
3254 struct ring_buffer_event *event;
3255 struct trace_array *tr = &global_trace;
3256 struct trace_array_cpu *data;
3257 struct print_entry *entry;
3258 unsigned long flags, irq_flags;
3259 int cpu, len = 0, size, pc;
3261 if (tracing_disabled)
3264 pc = preempt_count();
3265 preempt_disable_notrace();
3266 cpu = raw_smp_processor_id();
3267 data = tr->data[cpu];
3269 if (unlikely(atomic_read(&data->disabled)))
3272 spin_lock_irqsave(&trace_buf_lock, flags);
3273 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
3275 len = min(len, TRACE_BUF_SIZE-1);
3278 size = sizeof(*entry) + len + 1;
3279 event = ring_buffer_lock_reserve(tr->buffer, size, &irq_flags);
3282 entry = ring_buffer_event_data(event);
3283 tracing_generic_entry_update(&entry->ent, flags, pc);
3284 entry->ent.type = TRACE_PRINT;
3287 memcpy(&entry->buf, trace_buf, len);
3288 entry->buf[len] = 0;
3289 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
3292 spin_unlock_irqrestore(&trace_buf_lock, flags);
3295 preempt_enable_notrace();
3299 EXPORT_SYMBOL_GPL(trace_vprintk);
3301 int __ftrace_printk(unsigned long ip, const char *fmt, ...)
3306 if (!(trace_flags & TRACE_ITER_PRINTK))
3310 ret = trace_vprintk(ip, fmt, ap);
3314 EXPORT_SYMBOL_GPL(__ftrace_printk);
3316 static int trace_panic_handler(struct notifier_block *this,
3317 unsigned long event, void *unused)
3319 if (ftrace_dump_on_oops)
3324 static struct notifier_block trace_panic_notifier = {
3325 .notifier_call = trace_panic_handler,
3327 .priority = 150 /* priority: INT_MAX >= x >= 0 */
3330 static int trace_die_handler(struct notifier_block *self,
3336 if (ftrace_dump_on_oops)
3345 static struct notifier_block trace_die_notifier = {
3346 .notifier_call = trace_die_handler,
3351 * printk is set to max of 1024, we really don't need it that big.
3352 * Nothing should be printing 1000 characters anyway.
3354 #define TRACE_MAX_PRINT 1000
3357 * Define here KERN_TRACE so that we have one place to modify
3358 * it if we decide to change what log level the ftrace dump
3361 #define KERN_TRACE KERN_INFO
3364 trace_printk_seq(struct trace_seq *s)
3366 /* Probably should print a warning here. */
3370 /* should be zero ended, but we are paranoid. */
3371 s->buffer[s->len] = 0;
3373 printk(KERN_TRACE "%s", s->buffer);
3378 void ftrace_dump(void)
3380 static DEFINE_SPINLOCK(ftrace_dump_lock);
3381 /* use static because iter can be a bit big for the stack */
3382 static struct trace_iterator iter;
3383 static cpumask_t mask;
3384 static int dump_ran;
3385 unsigned long flags;
3389 spin_lock_irqsave(&ftrace_dump_lock, flags);
3395 /* No turning back! */
3398 for_each_tracing_cpu(cpu) {
3399 atomic_inc(&global_trace.data[cpu]->disabled);
3402 printk(KERN_TRACE "Dumping ftrace buffer:\n");
3404 iter.tr = &global_trace;
3405 iter.trace = current_trace;
3408 * We need to stop all tracing on all CPUS to read the
3409 * the next buffer. This is a bit expensive, but is
3410 * not done often. We fill all what we can read,
3411 * and then release the locks again.
3416 while (!trace_empty(&iter)) {
3419 printk(KERN_TRACE "---------------------------------\n");
3423 /* reset all but tr, trace, and overruns */
3424 memset(&iter.seq, 0,
3425 sizeof(struct trace_iterator) -
3426 offsetof(struct trace_iterator, seq));
3427 iter.iter_flags |= TRACE_FILE_LAT_FMT;
3430 if (find_next_entry_inc(&iter) != NULL) {
3431 print_trace_line(&iter);
3432 trace_consume(&iter);
3435 trace_printk_seq(&iter.seq);
3439 printk(KERN_TRACE " (ftrace buffer empty)\n");
3441 printk(KERN_TRACE "---------------------------------\n");
3444 spin_unlock_irqrestore(&ftrace_dump_lock, flags);
3447 __init static int tracer_alloc_buffers(void)
3449 struct trace_array_cpu *data;
3452 /* TODO: make the number of buffers hot pluggable with CPUS */
3453 tracing_buffer_mask = cpu_possible_map;
3455 global_trace.buffer = ring_buffer_alloc(trace_buf_size,
3456 TRACE_BUFFER_FLAGS);
3457 if (!global_trace.buffer) {
3458 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
3462 global_trace.entries = ring_buffer_size(global_trace.buffer);
3464 #ifdef CONFIG_TRACER_MAX_TRACE
3465 max_tr.buffer = ring_buffer_alloc(trace_buf_size,
3466 TRACE_BUFFER_FLAGS);
3467 if (!max_tr.buffer) {
3468 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
3470 ring_buffer_free(global_trace.buffer);
3473 max_tr.entries = ring_buffer_size(max_tr.buffer);
3474 WARN_ON(max_tr.entries != global_trace.entries);
3477 /* Allocate the first page for all buffers */
3478 for_each_tracing_cpu(i) {
3479 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
3480 max_tr.data[i] = &per_cpu(max_data, i);
3483 trace_init_cmdlines();
3485 register_tracer(&nop_trace);
3486 #ifdef CONFIG_BOOT_TRACER
3487 register_tracer(&boot_tracer);
3488 current_trace = &boot_tracer;
3489 current_trace->init(&global_trace);
3491 current_trace = &nop_trace;
3494 /* All seems OK, enable tracing */
3495 tracing_disabled = 0;
3497 atomic_notifier_chain_register(&panic_notifier_list,
3498 &trace_panic_notifier);
3500 register_die_notifier(&trace_die_notifier);
3504 early_initcall(tracer_alloc_buffers);
3505 fs_initcall(tracer_init_debugfs);