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/seq_file.h>
34 #include <linux/writeback.h>
36 #include <linux/stacktrace.h>
37 #include <linux/ring_buffer.h>
38 #include <linux/irqflags.h>
42 #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
44 unsigned long __read_mostly tracing_max_latency = (cycle_t)ULONG_MAX;
45 unsigned long __read_mostly tracing_thresh;
47 /* For tracers that don't implement custom flags */
48 static struct tracer_opt dummy_tracer_opt[] = {
52 static struct tracer_flags dummy_tracer_flags = {
54 .opts = dummy_tracer_opt
57 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
63 * Kill all tracing for good (never come back).
64 * It is initialized to 1 but will turn to zero if the initialization
65 * of the tracer is successful. But that is the only place that sets
68 int tracing_disabled = 1;
70 static DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
72 static inline void ftrace_disable_cpu(void)
75 local_inc(&__get_cpu_var(ftrace_cpu_disabled));
78 static inline void ftrace_enable_cpu(void)
80 local_dec(&__get_cpu_var(ftrace_cpu_disabled));
84 static cpumask_t __read_mostly tracing_buffer_mask;
86 #define for_each_tracing_cpu(cpu) \
87 for_each_cpu_mask(cpu, tracing_buffer_mask)
90 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
92 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
93 * is set, then ftrace_dump is called. This will output the contents
94 * of the ftrace buffers to the console. This is very useful for
95 * capturing traces that lead to crashes and outputing it to a
98 * It is default off, but you can enable it with either specifying
99 * "ftrace_dump_on_oops" in the kernel command line, or setting
100 * /proc/sys/kernel/ftrace_dump_on_oops to true.
102 int ftrace_dump_on_oops;
104 static int tracing_set_tracer(char *buf);
106 static int __init set_ftrace(char *str)
108 tracing_set_tracer(str);
111 __setup("ftrace", set_ftrace);
113 static int __init set_ftrace_dump_on_oops(char *str)
115 ftrace_dump_on_oops = 1;
118 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
121 ns2usecs(cycle_t nsec)
128 cycle_t ftrace_now(int cpu)
130 u64 ts = ring_buffer_time_stamp(cpu);
131 ring_buffer_normalize_time_stamp(cpu, &ts);
136 * The global_trace is the descriptor that holds the tracing
137 * buffers for the live tracing. For each CPU, it contains
138 * a link list of pages that will store trace entries. The
139 * page descriptor of the pages in the memory is used to hold
140 * the link list by linking the lru item in the page descriptor
141 * to each of the pages in the buffer per CPU.
143 * For each active CPU there is a data field that holds the
144 * pages for the buffer for that CPU. Each CPU has the same number
145 * of pages allocated for its buffer.
147 static struct trace_array global_trace;
149 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
152 * The max_tr is used to snapshot the global_trace when a maximum
153 * latency is reached. Some tracers will use this to store a maximum
154 * trace while it continues examining live traces.
156 * The buffers for the max_tr are set up the same as the global_trace.
157 * When a snapshot is taken, the link list of the max_tr is swapped
158 * with the link list of the global_trace and the buffers are reset for
159 * the global_trace so the tracing can continue.
161 static struct trace_array max_tr;
163 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
165 /* tracer_enabled is used to toggle activation of a tracer */
166 static int tracer_enabled = 1;
169 * tracing_is_enabled - return tracer_enabled status
171 * This function is used by other tracers to know the status
172 * of the tracer_enabled flag. Tracers may use this function
173 * to know if it should enable their features when starting
174 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
176 int tracing_is_enabled(void)
178 return tracer_enabled;
181 /* function tracing enabled */
182 int ftrace_function_enabled;
185 * trace_buf_size is the size in bytes that is allocated
186 * for a buffer. Note, the number of bytes is always rounded
189 * This number is purposely set to a low number of 16384.
190 * If the dump on oops happens, it will be much appreciated
191 * to not have to wait for all that output. Anyway this can be
192 * boot time and run time configurable.
194 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
196 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
198 /* trace_types holds a link list of available tracers. */
199 static struct tracer *trace_types __read_mostly;
201 /* current_trace points to the tracer that is currently active */
202 static struct tracer *current_trace __read_mostly;
205 * max_tracer_type_len is used to simplify the allocating of
206 * buffers to read userspace tracer names. We keep track of
207 * the longest tracer name registered.
209 static int max_tracer_type_len;
212 * trace_types_lock is used to protect the trace_types list.
213 * This lock is also used to keep user access serialized.
214 * Accesses from userspace will grab this lock while userspace
215 * activities happen inside the kernel.
217 static DEFINE_MUTEX(trace_types_lock);
219 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
220 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
222 /* trace_flags holds trace_options default values */
223 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
227 * trace_wake_up - wake up tasks waiting for trace input
229 * Simply wakes up any task that is blocked on the trace_wait
230 * queue. These is used with trace_poll for tasks polling the trace.
232 void trace_wake_up(void)
235 * The runqueue_is_locked() can fail, but this is the best we
238 if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
239 wake_up(&trace_wait);
242 static int __init set_buf_size(char *str)
244 unsigned long buf_size;
249 ret = strict_strtoul(str, 0, &buf_size);
250 /* nr_entries can not be zero */
251 if (ret < 0 || buf_size == 0)
253 trace_buf_size = buf_size;
256 __setup("trace_buf_size=", set_buf_size);
258 unsigned long nsecs_to_usecs(unsigned long nsecs)
263 /* These must match the bit postions in trace_iterator_flags */
264 static const char *trace_options[] = {
285 * ftrace_max_lock is used to protect the swapping of buffers
286 * when taking a max snapshot. The buffers themselves are
287 * protected by per_cpu spinlocks. But the action of the swap
288 * needs its own lock.
290 * This is defined as a raw_spinlock_t in order to help
291 * with performance when lockdep debugging is enabled.
293 static raw_spinlock_t ftrace_max_lock =
294 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
297 * Copy the new maximum trace into the separate maximum-trace
298 * structure. (this way the maximum trace is permanently saved,
299 * for later retrieval via /debugfs/tracing/latency_trace)
302 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
304 struct trace_array_cpu *data = tr->data[cpu];
307 max_tr.time_start = data->preempt_timestamp;
309 data = max_tr.data[cpu];
310 data->saved_latency = tracing_max_latency;
312 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
313 data->pid = tsk->pid;
314 data->uid = tsk->uid;
315 data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
316 data->policy = tsk->policy;
317 data->rt_priority = tsk->rt_priority;
319 /* record this tasks comm */
320 tracing_record_cmdline(current);
324 * trace_seq_printf - sequence printing of trace information
325 * @s: trace sequence descriptor
326 * @fmt: printf format string
328 * The tracer may use either sequence operations or its own
329 * copy to user routines. To simplify formating of a trace
330 * trace_seq_printf is used to store strings into a special
331 * buffer (@s). Then the output may be either used by
332 * the sequencer or pulled into another buffer.
335 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
337 int len = (PAGE_SIZE - 1) - s->len;
345 ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
348 /* If we can't write it all, don't bother writing anything */
358 * trace_seq_puts - trace sequence printing of simple string
359 * @s: trace sequence descriptor
360 * @str: simple string to record
362 * The tracer may use either the sequence operations or its own
363 * copy to user routines. This function records a simple string
364 * into a special buffer (@s) for later retrieval by a sequencer
365 * or other mechanism.
368 trace_seq_puts(struct trace_seq *s, const char *str)
370 int len = strlen(str);
372 if (len > ((PAGE_SIZE - 1) - s->len))
375 memcpy(s->buffer + s->len, str, len);
382 trace_seq_putc(struct trace_seq *s, unsigned char c)
384 if (s->len >= (PAGE_SIZE - 1))
387 s->buffer[s->len++] = c;
393 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
395 if (len > ((PAGE_SIZE - 1) - s->len))
398 memcpy(s->buffer + s->len, mem, len);
404 #define MAX_MEMHEX_BYTES 8
405 #define HEX_CHARS (MAX_MEMHEX_BYTES*2 + 1)
408 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
410 unsigned char hex[HEX_CHARS];
411 unsigned char *data = mem;
415 for (i = 0, j = 0; i < len; i++) {
417 for (i = len-1, j = 0; i >= 0; i--) {
419 hex[j++] = hex_asc_hi(data[i]);
420 hex[j++] = hex_asc_lo(data[i]);
424 return trace_seq_putmem(s, hex, j);
428 trace_seq_path(struct trace_seq *s, struct path *path)
432 if (s->len >= (PAGE_SIZE - 1))
434 p = d_path(path, s->buffer + s->len, PAGE_SIZE - s->len);
436 p = mangle_path(s->buffer + s->len, p, "\n");
438 s->len = p - s->buffer;
442 s->buffer[s->len++] = '?';
450 trace_seq_reset(struct trace_seq *s)
456 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
461 if (s->len <= s->readpos)
464 len = s->len - s->readpos;
467 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
476 trace_print_seq(struct seq_file *m, struct trace_seq *s)
478 int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
481 seq_puts(m, s->buffer);
487 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
489 * @tsk: the task with the latency
490 * @cpu: The cpu that initiated the trace.
492 * Flip the buffers between the @tr and the max_tr and record information
493 * about which task was the cause of this latency.
496 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
498 struct ring_buffer *buf = tr->buffer;
500 WARN_ON_ONCE(!irqs_disabled());
501 __raw_spin_lock(&ftrace_max_lock);
503 tr->buffer = max_tr.buffer;
506 ftrace_disable_cpu();
507 ring_buffer_reset(tr->buffer);
510 __update_max_tr(tr, tsk, cpu);
511 __raw_spin_unlock(&ftrace_max_lock);
515 * update_max_tr_single - only copy one trace over, and reset the rest
517 * @tsk - task with the latency
518 * @cpu - the cpu of the buffer to copy.
520 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
523 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
527 WARN_ON_ONCE(!irqs_disabled());
528 __raw_spin_lock(&ftrace_max_lock);
530 ftrace_disable_cpu();
532 ring_buffer_reset(max_tr.buffer);
533 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
539 __update_max_tr(tr, tsk, cpu);
540 __raw_spin_unlock(&ftrace_max_lock);
544 * register_tracer - register a tracer with the ftrace system.
545 * @type - the plugin for the tracer
547 * Register a new plugin tracer.
549 int register_tracer(struct tracer *type)
556 pr_info("Tracer must have a name\n");
561 * When this gets called we hold the BKL which means that
562 * preemption is disabled. Various trace selftests however
563 * need to disable and enable preemption for successful tests.
564 * So we drop the BKL here and grab it after the tests again.
567 mutex_lock(&trace_types_lock);
569 for (t = trace_types; t; t = t->next) {
570 if (strcmp(type->name, t->name) == 0) {
572 pr_info("Trace %s already registered\n",
580 type->set_flag = &dummy_set_flag;
582 type->flags = &dummy_tracer_flags;
584 if (!type->flags->opts)
585 type->flags->opts = dummy_tracer_opt;
587 #ifdef CONFIG_FTRACE_STARTUP_TEST
588 if (type->selftest) {
589 struct tracer *saved_tracer = current_trace;
590 struct trace_array *tr = &global_trace;
593 * Run a selftest on this tracer.
594 * Here we reset the trace buffer, and set the current
595 * tracer to be this tracer. The tracer can then run some
596 * internal tracing to verify that everything is in order.
597 * If we fail, we do not register this tracer.
599 for_each_tracing_cpu(i)
600 tracing_reset(tr, i);
602 current_trace = type;
603 /* the test is responsible for initializing and enabling */
604 pr_info("Testing tracer %s: ", type->name);
605 ret = type->selftest(type, tr);
606 /* the test is responsible for resetting too */
607 current_trace = saved_tracer;
609 printk(KERN_CONT "FAILED!\n");
612 /* Only reset on passing, to avoid touching corrupted buffers */
613 for_each_tracing_cpu(i)
614 tracing_reset(tr, i);
616 printk(KERN_CONT "PASSED\n");
620 type->next = trace_types;
622 len = strlen(type->name);
623 if (len > max_tracer_type_len)
624 max_tracer_type_len = len;
627 mutex_unlock(&trace_types_lock);
633 void unregister_tracer(struct tracer *type)
638 mutex_lock(&trace_types_lock);
639 for (t = &trace_types; *t; t = &(*t)->next) {
643 pr_info("Trace %s not registered\n", type->name);
648 if (strlen(type->name) != max_tracer_type_len)
651 max_tracer_type_len = 0;
652 for (t = &trace_types; *t; t = &(*t)->next) {
653 len = strlen((*t)->name);
654 if (len > max_tracer_type_len)
655 max_tracer_type_len = len;
658 mutex_unlock(&trace_types_lock);
661 void tracing_reset(struct trace_array *tr, int cpu)
663 ftrace_disable_cpu();
664 ring_buffer_reset_cpu(tr->buffer, cpu);
668 #define SAVED_CMDLINES 128
669 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
670 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
671 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
672 static int cmdline_idx;
673 static DEFINE_SPINLOCK(trace_cmdline_lock);
675 /* temporary disable recording */
676 atomic_t trace_record_cmdline_disabled __read_mostly;
678 static void trace_init_cmdlines(void)
680 memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
681 memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
685 static int trace_stop_count;
686 static DEFINE_SPINLOCK(tracing_start_lock);
689 * ftrace_off_permanent - disable all ftrace code permanently
691 * This should only be called when a serious anomally has
692 * been detected. This will turn off the function tracing,
693 * ring buffers, and other tracing utilites. It takes no
694 * locks and can be called from any context.
696 void ftrace_off_permanent(void)
698 tracing_disabled = 1;
700 tracing_off_permanent();
704 * tracing_start - quick start of the tracer
706 * If tracing is enabled but was stopped by tracing_stop,
707 * this will start the tracer back up.
709 void tracing_start(void)
711 struct ring_buffer *buffer;
714 if (tracing_disabled)
717 spin_lock_irqsave(&tracing_start_lock, flags);
718 if (--trace_stop_count)
721 if (trace_stop_count < 0) {
722 /* Someone screwed up their debugging */
724 trace_stop_count = 0;
729 buffer = global_trace.buffer;
731 ring_buffer_record_enable(buffer);
733 buffer = max_tr.buffer;
735 ring_buffer_record_enable(buffer);
739 spin_unlock_irqrestore(&tracing_start_lock, flags);
743 * tracing_stop - quick stop of the tracer
745 * Light weight way to stop tracing. Use in conjunction with
748 void tracing_stop(void)
750 struct ring_buffer *buffer;
754 spin_lock_irqsave(&tracing_start_lock, flags);
755 if (trace_stop_count++)
758 buffer = global_trace.buffer;
760 ring_buffer_record_disable(buffer);
762 buffer = max_tr.buffer;
764 ring_buffer_record_disable(buffer);
767 spin_unlock_irqrestore(&tracing_start_lock, flags);
770 void trace_stop_cmdline_recording(void);
772 static void trace_save_cmdline(struct task_struct *tsk)
777 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
781 * It's not the end of the world if we don't get
782 * the lock, but we also don't want to spin
783 * nor do we want to disable interrupts,
784 * so if we miss here, then better luck next time.
786 if (!spin_trylock(&trace_cmdline_lock))
789 idx = map_pid_to_cmdline[tsk->pid];
790 if (idx >= SAVED_CMDLINES) {
791 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
793 map = map_cmdline_to_pid[idx];
794 if (map <= PID_MAX_DEFAULT)
795 map_pid_to_cmdline[map] = (unsigned)-1;
797 map_pid_to_cmdline[tsk->pid] = idx;
802 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
804 spin_unlock(&trace_cmdline_lock);
807 static char *trace_find_cmdline(int pid)
809 char *cmdline = "<...>";
815 if (pid > PID_MAX_DEFAULT)
818 map = map_pid_to_cmdline[pid];
819 if (map >= SAVED_CMDLINES)
822 cmdline = saved_cmdlines[map];
828 void tracing_record_cmdline(struct task_struct *tsk)
830 if (atomic_read(&trace_record_cmdline_disabled))
833 trace_save_cmdline(tsk);
837 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
840 struct task_struct *tsk = current;
842 entry->preempt_count = pc & 0xff;
843 entry->pid = (tsk) ? tsk->pid : 0;
844 entry->tgid = (tsk) ? tsk->tgid : 0;
846 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
847 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
849 TRACE_FLAG_IRQS_NOSUPPORT |
851 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
852 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
853 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
857 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
858 unsigned long ip, unsigned long parent_ip, unsigned long flags,
861 struct ring_buffer_event *event;
862 struct ftrace_entry *entry;
863 unsigned long irq_flags;
865 /* If we are reading the ring buffer, don't trace */
866 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
869 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
873 entry = ring_buffer_event_data(event);
874 tracing_generic_entry_update(&entry->ent, flags, pc);
875 entry->ent.type = TRACE_FN;
877 entry->parent_ip = parent_ip;
878 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
881 #ifdef CONFIG_FUNCTION_RET_TRACER
882 static void __trace_function_return(struct trace_array *tr,
883 struct trace_array_cpu *data,
884 struct ftrace_retfunc *trace,
888 struct ring_buffer_event *event;
889 struct ftrace_ret_entry *entry;
890 unsigned long irq_flags;
892 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
895 event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry),
899 entry = ring_buffer_event_data(event);
900 tracing_generic_entry_update(&entry->ent, flags, pc);
901 entry->ent.type = TRACE_FN_RET;
902 entry->ip = trace->func;
903 entry->parent_ip = trace->ret;
904 entry->rettime = trace->rettime;
905 entry->calltime = trace->calltime;
906 entry->overrun = trace->overrun;
907 ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
912 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
913 unsigned long ip, unsigned long parent_ip, unsigned long flags,
916 if (likely(!atomic_read(&data->disabled)))
917 trace_function(tr, data, ip, parent_ip, flags, pc);
920 static void ftrace_trace_stack(struct trace_array *tr,
921 struct trace_array_cpu *data,
925 #ifdef CONFIG_STACKTRACE
926 struct ring_buffer_event *event;
927 struct stack_entry *entry;
928 struct stack_trace trace;
929 unsigned long irq_flags;
931 if (!(trace_flags & TRACE_ITER_STACKTRACE))
934 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
938 entry = ring_buffer_event_data(event);
939 tracing_generic_entry_update(&entry->ent, flags, pc);
940 entry->ent.type = TRACE_STACK;
942 memset(&entry->caller, 0, sizeof(entry->caller));
944 trace.nr_entries = 0;
945 trace.max_entries = FTRACE_STACK_ENTRIES;
947 trace.entries = entry->caller;
949 save_stack_trace(&trace);
950 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
954 void __trace_stack(struct trace_array *tr,
955 struct trace_array_cpu *data,
959 ftrace_trace_stack(tr, data, flags, skip, preempt_count());
962 static void ftrace_trace_userstack(struct trace_array *tr,
963 struct trace_array_cpu *data,
964 unsigned long flags, int pc)
966 struct ring_buffer_event *event;
967 struct userstack_entry *entry;
968 struct stack_trace trace;
969 unsigned long irq_flags;
971 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
974 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
978 entry = ring_buffer_event_data(event);
979 tracing_generic_entry_update(&entry->ent, flags, pc);
980 entry->ent.type = TRACE_USER_STACK;
982 memset(&entry->caller, 0, sizeof(entry->caller));
984 trace.nr_entries = 0;
985 trace.max_entries = FTRACE_STACK_ENTRIES;
987 trace.entries = entry->caller;
989 save_stack_trace_user(&trace);
990 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
993 void __trace_userstack(struct trace_array *tr,
994 struct trace_array_cpu *data,
997 ftrace_trace_userstack(tr, data, flags, preempt_count());
1001 ftrace_trace_special(void *__tr, void *__data,
1002 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1005 struct ring_buffer_event *event;
1006 struct trace_array_cpu *data = __data;
1007 struct trace_array *tr = __tr;
1008 struct special_entry *entry;
1009 unsigned long irq_flags;
1011 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1015 entry = ring_buffer_event_data(event);
1016 tracing_generic_entry_update(&entry->ent, 0, pc);
1017 entry->ent.type = TRACE_SPECIAL;
1021 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1022 ftrace_trace_stack(tr, data, irq_flags, 4, pc);
1023 ftrace_trace_userstack(tr, data, irq_flags, pc);
1029 __trace_special(void *__tr, void *__data,
1030 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1032 ftrace_trace_special(__tr, __data, arg1, arg2, arg3, preempt_count());
1036 tracing_sched_switch_trace(struct trace_array *tr,
1037 struct trace_array_cpu *data,
1038 struct task_struct *prev,
1039 struct task_struct *next,
1040 unsigned long flags, int pc)
1042 struct ring_buffer_event *event;
1043 struct ctx_switch_entry *entry;
1044 unsigned long irq_flags;
1046 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1050 entry = ring_buffer_event_data(event);
1051 tracing_generic_entry_update(&entry->ent, flags, pc);
1052 entry->ent.type = TRACE_CTX;
1053 entry->prev_pid = prev->pid;
1054 entry->prev_prio = prev->prio;
1055 entry->prev_state = prev->state;
1056 entry->next_pid = next->pid;
1057 entry->next_prio = next->prio;
1058 entry->next_state = next->state;
1059 entry->next_cpu = task_cpu(next);
1060 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1061 ftrace_trace_stack(tr, data, flags, 5, pc);
1062 ftrace_trace_userstack(tr, data, flags, pc);
1066 tracing_sched_wakeup_trace(struct trace_array *tr,
1067 struct trace_array_cpu *data,
1068 struct task_struct *wakee,
1069 struct task_struct *curr,
1070 unsigned long flags, int pc)
1072 struct ring_buffer_event *event;
1073 struct ctx_switch_entry *entry;
1074 unsigned long irq_flags;
1076 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1080 entry = ring_buffer_event_data(event);
1081 tracing_generic_entry_update(&entry->ent, flags, pc);
1082 entry->ent.type = TRACE_WAKE;
1083 entry->prev_pid = curr->pid;
1084 entry->prev_prio = curr->prio;
1085 entry->prev_state = curr->state;
1086 entry->next_pid = wakee->pid;
1087 entry->next_prio = wakee->prio;
1088 entry->next_state = wakee->state;
1089 entry->next_cpu = task_cpu(wakee);
1090 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1091 ftrace_trace_stack(tr, data, flags, 6, pc);
1092 ftrace_trace_userstack(tr, data, flags, pc);
1098 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1100 struct trace_array *tr = &global_trace;
1101 struct trace_array_cpu *data;
1102 unsigned long flags;
1106 if (tracing_disabled)
1109 pc = preempt_count();
1110 local_irq_save(flags);
1111 cpu = raw_smp_processor_id();
1112 data = tr->data[cpu];
1114 if (likely(atomic_inc_return(&data->disabled) == 1))
1115 ftrace_trace_special(tr, data, arg1, arg2, arg3, pc);
1117 atomic_dec(&data->disabled);
1118 local_irq_restore(flags);
1121 #ifdef CONFIG_FUNCTION_TRACER
1123 function_trace_call_preempt_only(unsigned long ip, unsigned long parent_ip)
1125 struct trace_array *tr = &global_trace;
1126 struct trace_array_cpu *data;
1127 unsigned long flags;
1132 if (unlikely(!ftrace_function_enabled))
1135 pc = preempt_count();
1136 resched = ftrace_preempt_disable();
1137 local_save_flags(flags);
1138 cpu = raw_smp_processor_id();
1139 data = tr->data[cpu];
1140 disabled = atomic_inc_return(&data->disabled);
1142 if (likely(disabled == 1))
1143 trace_function(tr, data, ip, parent_ip, flags, pc);
1145 atomic_dec(&data->disabled);
1146 ftrace_preempt_enable(resched);
1150 function_trace_call(unsigned long ip, unsigned long parent_ip)
1152 struct trace_array *tr = &global_trace;
1153 struct trace_array_cpu *data;
1154 unsigned long flags;
1159 if (unlikely(!ftrace_function_enabled))
1163 * Need to use raw, since this must be called before the
1164 * recursive protection is performed.
1166 local_irq_save(flags);
1167 cpu = raw_smp_processor_id();
1168 data = tr->data[cpu];
1169 disabled = atomic_inc_return(&data->disabled);
1171 if (likely(disabled == 1)) {
1172 pc = preempt_count();
1173 trace_function(tr, data, ip, parent_ip, flags, pc);
1176 atomic_dec(&data->disabled);
1177 local_irq_restore(flags);
1180 #ifdef CONFIG_FUNCTION_RET_TRACER
1181 void trace_function_return(struct ftrace_retfunc *trace)
1183 struct trace_array *tr = &global_trace;
1184 struct trace_array_cpu *data;
1185 unsigned long flags;
1190 raw_local_irq_save(flags);
1191 cpu = raw_smp_processor_id();
1192 data = tr->data[cpu];
1193 disabled = atomic_inc_return(&data->disabled);
1194 if (likely(disabled == 1)) {
1195 pc = preempt_count();
1196 __trace_function_return(tr, data, trace, flags, pc);
1198 atomic_dec(&data->disabled);
1199 raw_local_irq_restore(flags);
1201 #endif /* CONFIG_FUNCTION_RET_TRACER */
1203 static struct ftrace_ops trace_ops __read_mostly =
1205 .func = function_trace_call,
1208 void tracing_start_function_trace(void)
1210 ftrace_function_enabled = 0;
1212 if (trace_flags & TRACE_ITER_PREEMPTONLY)
1213 trace_ops.func = function_trace_call_preempt_only;
1215 trace_ops.func = function_trace_call;
1217 register_ftrace_function(&trace_ops);
1218 ftrace_function_enabled = 1;
1221 void tracing_stop_function_trace(void)
1223 ftrace_function_enabled = 0;
1224 unregister_ftrace_function(&trace_ops);
1228 enum trace_file_type {
1229 TRACE_FILE_LAT_FMT = 1,
1230 TRACE_FILE_ANNOTATE = 2,
1233 static void trace_iterator_increment(struct trace_iterator *iter, int cpu)
1235 /* Don't allow ftrace to trace into the ring buffers */
1236 ftrace_disable_cpu();
1239 if (iter->buffer_iter[iter->cpu])
1240 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1242 ftrace_enable_cpu();
1245 static struct trace_entry *
1246 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1248 struct ring_buffer_event *event;
1249 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1251 /* Don't allow ftrace to trace into the ring buffers */
1252 ftrace_disable_cpu();
1255 event = ring_buffer_iter_peek(buf_iter, ts);
1257 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1259 ftrace_enable_cpu();
1261 return event ? ring_buffer_event_data(event) : NULL;
1264 static struct trace_entry *
1265 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1267 struct ring_buffer *buffer = iter->tr->buffer;
1268 struct trace_entry *ent, *next = NULL;
1269 u64 next_ts = 0, ts;
1273 for_each_tracing_cpu(cpu) {
1275 if (ring_buffer_empty_cpu(buffer, cpu))
1278 ent = peek_next_entry(iter, cpu, &ts);
1281 * Pick the entry with the smallest timestamp:
1283 if (ent && (!next || ts < next_ts)) {
1291 *ent_cpu = next_cpu;
1299 /* Find the next real entry, without updating the iterator itself */
1300 static struct trace_entry *
1301 find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1303 return __find_next_entry(iter, ent_cpu, ent_ts);
1306 /* Find the next real entry, and increment the iterator to the next entry */
1307 static void *find_next_entry_inc(struct trace_iterator *iter)
1309 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1312 trace_iterator_increment(iter, iter->cpu);
1314 return iter->ent ? iter : NULL;
1317 static void trace_consume(struct trace_iterator *iter)
1319 /* Don't allow ftrace to trace into the ring buffers */
1320 ftrace_disable_cpu();
1321 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1322 ftrace_enable_cpu();
1325 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1327 struct trace_iterator *iter = m->private;
1333 /* can't go backwards */
1338 ent = find_next_entry_inc(iter);
1342 while (ent && iter->idx < i)
1343 ent = find_next_entry_inc(iter);
1350 static void *s_start(struct seq_file *m, loff_t *pos)
1352 struct trace_iterator *iter = m->private;
1357 mutex_lock(&trace_types_lock);
1359 if (!current_trace || current_trace != iter->trace) {
1360 mutex_unlock(&trace_types_lock);
1364 atomic_inc(&trace_record_cmdline_disabled);
1366 if (*pos != iter->pos) {
1371 ftrace_disable_cpu();
1373 for_each_tracing_cpu(cpu) {
1374 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1377 ftrace_enable_cpu();
1379 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1384 p = s_next(m, p, &l);
1390 static void s_stop(struct seq_file *m, void *p)
1392 atomic_dec(&trace_record_cmdline_disabled);
1393 mutex_unlock(&trace_types_lock);
1396 #ifdef CONFIG_KRETPROBES
1397 static inline const char *kretprobed(const char *name)
1399 static const char tramp_name[] = "kretprobe_trampoline";
1400 int size = sizeof(tramp_name);
1402 if (strncmp(tramp_name, name, size) == 0)
1403 return "[unknown/kretprobe'd]";
1407 static inline const char *kretprobed(const char *name)
1411 #endif /* CONFIG_KRETPROBES */
1414 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1416 #ifdef CONFIG_KALLSYMS
1417 char str[KSYM_SYMBOL_LEN];
1420 kallsyms_lookup(address, NULL, NULL, NULL, str);
1422 name = kretprobed(str);
1424 return trace_seq_printf(s, fmt, name);
1430 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1431 unsigned long address)
1433 #ifdef CONFIG_KALLSYMS
1434 char str[KSYM_SYMBOL_LEN];
1437 sprint_symbol(str, address);
1438 name = kretprobed(str);
1440 return trace_seq_printf(s, fmt, name);
1445 #ifndef CONFIG_64BIT
1446 # define IP_FMT "%08lx"
1448 # define IP_FMT "%016lx"
1452 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1457 return trace_seq_printf(s, "0");
1459 if (sym_flags & TRACE_ITER_SYM_OFFSET)
1460 ret = seq_print_sym_offset(s, "%s", ip);
1462 ret = seq_print_sym_short(s, "%s", ip);
1467 if (sym_flags & TRACE_ITER_SYM_ADDR)
1468 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1472 static inline int seq_print_user_ip(struct trace_seq *s, struct mm_struct *mm,
1473 unsigned long ip, unsigned long sym_flags)
1475 struct file *file = NULL;
1476 unsigned long vmstart = 0;
1480 const struct vm_area_struct *vma;
1482 down_read(&mm->mmap_sem);
1483 vma = find_vma(mm, ip);
1485 file = vma->vm_file;
1486 vmstart = vma->vm_start;
1489 ret = trace_seq_path(s, &file->f_path);
1491 ret = trace_seq_printf(s, "[+0x%lx]", ip - vmstart);
1493 up_read(&mm->mmap_sem);
1495 if (ret && ((sym_flags & TRACE_ITER_SYM_ADDR) || !file))
1496 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1501 seq_print_userip_objs(const struct userstack_entry *entry, struct trace_seq *s,
1502 unsigned long sym_flags)
1504 struct mm_struct *mm = NULL;
1508 if (trace_flags & TRACE_ITER_SYM_USEROBJ) {
1509 struct task_struct *task;
1511 * we do the lookup on the thread group leader,
1512 * since individual threads might have already quit!
1515 task = find_task_by_vpid(entry->ent.tgid);
1517 mm = get_task_mm(task);
1521 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1522 unsigned long ip = entry->caller[i];
1524 if (ip == ULONG_MAX || !ret)
1527 ret = trace_seq_puts(s, " <- ");
1530 ret = trace_seq_puts(s, "??");
1536 ret = seq_print_user_ip(s, mm, ip, sym_flags);
1544 static void print_lat_help_header(struct seq_file *m)
1546 seq_puts(m, "# _------=> CPU# \n");
1547 seq_puts(m, "# / _-----=> irqs-off \n");
1548 seq_puts(m, "# | / _----=> need-resched \n");
1549 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1550 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1551 seq_puts(m, "# |||| / \n");
1552 seq_puts(m, "# ||||| delay \n");
1553 seq_puts(m, "# cmd pid ||||| time | caller \n");
1554 seq_puts(m, "# \\ / ||||| \\ | / \n");
1557 static void print_func_help_header(struct seq_file *m)
1559 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1560 seq_puts(m, "# | | | | |\n");
1565 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1567 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1568 struct trace_array *tr = iter->tr;
1569 struct trace_array_cpu *data = tr->data[tr->cpu];
1570 struct tracer *type = current_trace;
1571 unsigned long total;
1572 unsigned long entries;
1573 const char *name = "preemption";
1578 entries = ring_buffer_entries(iter->tr->buffer);
1580 ring_buffer_overruns(iter->tr->buffer);
1582 seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1584 seq_puts(m, "-----------------------------------"
1585 "---------------------------------\n");
1586 seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1587 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1588 nsecs_to_usecs(data->saved_latency),
1592 #if defined(CONFIG_PREEMPT_NONE)
1594 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1596 #elif defined(CONFIG_PREEMPT)
1601 /* These are reserved for later use */
1604 seq_printf(m, " #P:%d)\n", num_online_cpus());
1608 seq_puts(m, " -----------------\n");
1609 seq_printf(m, " | task: %.16s-%d "
1610 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1611 data->comm, data->pid, data->uid, data->nice,
1612 data->policy, data->rt_priority);
1613 seq_puts(m, " -----------------\n");
1615 if (data->critical_start) {
1616 seq_puts(m, " => started at: ");
1617 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1618 trace_print_seq(m, &iter->seq);
1619 seq_puts(m, "\n => ended at: ");
1620 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1621 trace_print_seq(m, &iter->seq);
1629 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1631 int hardirq, softirq;
1634 comm = trace_find_cmdline(entry->pid);
1636 trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1637 trace_seq_printf(s, "%3d", cpu);
1638 trace_seq_printf(s, "%c%c",
1639 (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' :
1640 (entry->flags & TRACE_FLAG_IRQS_NOSUPPORT) ? 'X' : '.',
1641 ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1643 hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1644 softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1645 if (hardirq && softirq) {
1646 trace_seq_putc(s, 'H');
1649 trace_seq_putc(s, 'h');
1652 trace_seq_putc(s, 's');
1654 trace_seq_putc(s, '.');
1658 if (entry->preempt_count)
1659 trace_seq_printf(s, "%x", entry->preempt_count);
1661 trace_seq_puts(s, ".");
1664 unsigned long preempt_mark_thresh = 100;
1667 lat_print_timestamp(struct trace_seq *s, u64 abs_usecs,
1668 unsigned long rel_usecs)
1670 trace_seq_printf(s, " %4lldus", abs_usecs);
1671 if (rel_usecs > preempt_mark_thresh)
1672 trace_seq_puts(s, "!: ");
1673 else if (rel_usecs > 1)
1674 trace_seq_puts(s, "+: ");
1676 trace_seq_puts(s, " : ");
1679 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1682 * The message is supposed to contain an ending newline.
1683 * If the printing stops prematurely, try to add a newline of our own.
1685 void trace_seq_print_cont(struct trace_seq *s, struct trace_iterator *iter)
1687 struct trace_entry *ent;
1688 struct trace_field_cont *cont;
1691 ent = peek_next_entry(iter, iter->cpu, NULL);
1692 if (!ent || ent->type != TRACE_CONT) {
1693 trace_seq_putc(s, '\n');
1698 cont = (struct trace_field_cont *)ent;
1700 ok = (trace_seq_printf(s, "%s", cont->buf) > 0);
1702 ftrace_disable_cpu();
1704 if (iter->buffer_iter[iter->cpu])
1705 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1707 ring_buffer_consume(iter->tr->buffer, iter->cpu, NULL);
1709 ftrace_enable_cpu();
1711 ent = peek_next_entry(iter, iter->cpu, NULL);
1712 } while (ent && ent->type == TRACE_CONT);
1715 trace_seq_putc(s, '\n');
1718 static void test_cpu_buff_start(struct trace_iterator *iter)
1720 struct trace_seq *s = &iter->seq;
1722 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1725 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1728 if (cpu_isset(iter->cpu, iter->started))
1731 cpu_set(iter->cpu, iter->started);
1732 trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1735 static enum print_line_t
1736 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1738 struct trace_seq *s = &iter->seq;
1739 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1740 struct trace_entry *next_entry;
1741 unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1742 struct trace_entry *entry = iter->ent;
1743 unsigned long abs_usecs;
1744 unsigned long rel_usecs;
1751 if (entry->type == TRACE_CONT)
1752 return TRACE_TYPE_HANDLED;
1754 test_cpu_buff_start(iter);
1756 next_entry = find_next_entry(iter, NULL, &next_ts);
1759 rel_usecs = ns2usecs(next_ts - iter->ts);
1760 abs_usecs = ns2usecs(iter->ts - iter->tr->time_start);
1763 comm = trace_find_cmdline(entry->pid);
1764 trace_seq_printf(s, "%16s %5d %3d %d %08x %08x [%08lx]"
1765 " %ld.%03ldms (+%ld.%03ldms): ",
1767 entry->pid, cpu, entry->flags,
1768 entry->preempt_count, trace_idx,
1771 abs_usecs % 1000, rel_usecs/1000,
1774 lat_print_generic(s, entry, cpu);
1775 lat_print_timestamp(s, abs_usecs, rel_usecs);
1777 switch (entry->type) {
1779 struct ftrace_entry *field;
1781 trace_assign_type(field, entry);
1783 seq_print_ip_sym(s, field->ip, sym_flags);
1784 trace_seq_puts(s, " (");
1785 seq_print_ip_sym(s, field->parent_ip, sym_flags);
1786 trace_seq_puts(s, ")\n");
1791 struct ctx_switch_entry *field;
1793 trace_assign_type(field, entry);
1795 T = field->next_state < sizeof(state_to_char) ?
1796 state_to_char[field->next_state] : 'X';
1798 state = field->prev_state ?
1799 __ffs(field->prev_state) + 1 : 0;
1800 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1801 comm = trace_find_cmdline(field->next_pid);
1802 trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c %s\n",
1805 S, entry->type == TRACE_CTX ? "==>" : " +",
1812 case TRACE_SPECIAL: {
1813 struct special_entry *field;
1815 trace_assign_type(field, entry);
1817 trace_seq_printf(s, "# %ld %ld %ld\n",
1824 struct stack_entry *field;
1826 trace_assign_type(field, entry);
1828 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1830 trace_seq_puts(s, " <= ");
1831 seq_print_ip_sym(s, field->caller[i], sym_flags);
1833 trace_seq_puts(s, "\n");
1837 struct print_entry *field;
1839 trace_assign_type(field, entry);
1841 seq_print_ip_sym(s, field->ip, sym_flags);
1842 trace_seq_printf(s, ": %s", field->buf);
1843 if (entry->flags & TRACE_FLAG_CONT)
1844 trace_seq_print_cont(s, iter);
1847 case TRACE_BRANCH: {
1848 struct trace_branch *field;
1850 trace_assign_type(field, entry);
1852 trace_seq_printf(s, "[%s] %s:%s:%d\n",
1853 field->correct ? " ok " : " MISS ",
1859 case TRACE_USER_STACK: {
1860 struct userstack_entry *field;
1862 trace_assign_type(field, entry);
1864 seq_print_userip_objs(field, s, sym_flags);
1865 trace_seq_putc(s, '\n');
1869 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1871 return TRACE_TYPE_HANDLED;
1874 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1876 struct trace_seq *s = &iter->seq;
1877 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1878 struct trace_entry *entry;
1879 unsigned long usec_rem;
1880 unsigned long long t;
1889 if (entry->type == TRACE_CONT)
1890 return TRACE_TYPE_HANDLED;
1892 test_cpu_buff_start(iter);
1894 comm = trace_find_cmdline(iter->ent->pid);
1896 t = ns2usecs(iter->ts);
1897 usec_rem = do_div(t, 1000000ULL);
1898 secs = (unsigned long)t;
1900 ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1902 return TRACE_TYPE_PARTIAL_LINE;
1903 ret = trace_seq_printf(s, "[%03d] ", iter->cpu);
1905 return TRACE_TYPE_PARTIAL_LINE;
1906 ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1908 return TRACE_TYPE_PARTIAL_LINE;
1910 switch (entry->type) {
1912 struct ftrace_entry *field;
1914 trace_assign_type(field, entry);
1916 ret = seq_print_ip_sym(s, field->ip, sym_flags);
1918 return TRACE_TYPE_PARTIAL_LINE;
1919 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1921 ret = trace_seq_printf(s, " <-");
1923 return TRACE_TYPE_PARTIAL_LINE;
1924 ret = seq_print_ip_sym(s,
1928 return TRACE_TYPE_PARTIAL_LINE;
1930 ret = trace_seq_printf(s, "\n");
1932 return TRACE_TYPE_PARTIAL_LINE;
1937 struct ctx_switch_entry *field;
1939 trace_assign_type(field, entry);
1941 S = field->prev_state < sizeof(state_to_char) ?
1942 state_to_char[field->prev_state] : 'X';
1943 T = field->next_state < sizeof(state_to_char) ?
1944 state_to_char[field->next_state] : 'X';
1945 ret = trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c\n",
1949 entry->type == TRACE_CTX ? "==>" : " +",
1955 return TRACE_TYPE_PARTIAL_LINE;
1958 case TRACE_SPECIAL: {
1959 struct special_entry *field;
1961 trace_assign_type(field, entry);
1963 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1968 return TRACE_TYPE_PARTIAL_LINE;
1972 struct stack_entry *field;
1974 trace_assign_type(field, entry);
1976 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1978 ret = trace_seq_puts(s, " <= ");
1980 return TRACE_TYPE_PARTIAL_LINE;
1982 ret = seq_print_ip_sym(s, field->caller[i],
1985 return TRACE_TYPE_PARTIAL_LINE;
1987 ret = trace_seq_puts(s, "\n");
1989 return TRACE_TYPE_PARTIAL_LINE;
1993 struct print_entry *field;
1995 trace_assign_type(field, entry);
1997 seq_print_ip_sym(s, field->ip, sym_flags);
1998 trace_seq_printf(s, ": %s", field->buf);
1999 if (entry->flags & TRACE_FLAG_CONT)
2000 trace_seq_print_cont(s, iter);
2003 case TRACE_FN_RET: {
2004 return print_return_function(iter);
2007 case TRACE_BRANCH: {
2008 struct trace_branch *field;
2010 trace_assign_type(field, entry);
2012 trace_seq_printf(s, "[%s] %s:%s:%d\n",
2013 field->correct ? " ok " : " MISS ",
2019 case TRACE_USER_STACK: {
2020 struct userstack_entry *field;
2022 trace_assign_type(field, entry);
2024 ret = seq_print_userip_objs(field, s, sym_flags);
2026 return TRACE_TYPE_PARTIAL_LINE;
2027 ret = trace_seq_putc(s, '\n');
2029 return TRACE_TYPE_PARTIAL_LINE;
2033 return TRACE_TYPE_HANDLED;
2036 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2038 struct trace_seq *s = &iter->seq;
2039 struct trace_entry *entry;
2045 if (entry->type == TRACE_CONT)
2046 return TRACE_TYPE_HANDLED;
2048 ret = trace_seq_printf(s, "%d %d %llu ",
2049 entry->pid, iter->cpu, iter->ts);
2051 return TRACE_TYPE_PARTIAL_LINE;
2053 switch (entry->type) {
2055 struct ftrace_entry *field;
2057 trace_assign_type(field, entry);
2059 ret = trace_seq_printf(s, "%x %x\n",
2063 return TRACE_TYPE_PARTIAL_LINE;
2068 struct ctx_switch_entry *field;
2070 trace_assign_type(field, entry);
2072 S = field->prev_state < sizeof(state_to_char) ?
2073 state_to_char[field->prev_state] : 'X';
2074 T = field->next_state < sizeof(state_to_char) ?
2075 state_to_char[field->next_state] : 'X';
2076 if (entry->type == TRACE_WAKE)
2078 ret = trace_seq_printf(s, "%d %d %c %d %d %d %c\n",
2087 return TRACE_TYPE_PARTIAL_LINE;
2091 case TRACE_USER_STACK:
2093 struct special_entry *field;
2095 trace_assign_type(field, entry);
2097 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
2102 return TRACE_TYPE_PARTIAL_LINE;
2106 struct print_entry *field;
2108 trace_assign_type(field, entry);
2110 trace_seq_printf(s, "# %lx %s", field->ip, field->buf);
2111 if (entry->flags & TRACE_FLAG_CONT)
2112 trace_seq_print_cont(s, iter);
2116 return TRACE_TYPE_HANDLED;
2119 #define SEQ_PUT_FIELD_RET(s, x) \
2121 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
2125 #define SEQ_PUT_HEX_FIELD_RET(s, x) \
2127 BUILD_BUG_ON(sizeof(x) > MAX_MEMHEX_BYTES); \
2128 if (!trace_seq_putmem_hex(s, &(x), sizeof(x))) \
2132 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2134 struct trace_seq *s = &iter->seq;
2135 unsigned char newline = '\n';
2136 struct trace_entry *entry;
2141 if (entry->type == TRACE_CONT)
2142 return TRACE_TYPE_HANDLED;
2144 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2145 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2146 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2148 switch (entry->type) {
2150 struct ftrace_entry *field;
2152 trace_assign_type(field, entry);
2154 SEQ_PUT_HEX_FIELD_RET(s, field->ip);
2155 SEQ_PUT_HEX_FIELD_RET(s, field->parent_ip);
2160 struct ctx_switch_entry *field;
2162 trace_assign_type(field, entry);
2164 S = field->prev_state < sizeof(state_to_char) ?
2165 state_to_char[field->prev_state] : 'X';
2166 T = field->next_state < sizeof(state_to_char) ?
2167 state_to_char[field->next_state] : 'X';
2168 if (entry->type == TRACE_WAKE)
2170 SEQ_PUT_HEX_FIELD_RET(s, field->prev_pid);
2171 SEQ_PUT_HEX_FIELD_RET(s, field->prev_prio);
2172 SEQ_PUT_HEX_FIELD_RET(s, S);
2173 SEQ_PUT_HEX_FIELD_RET(s, field->next_cpu);
2174 SEQ_PUT_HEX_FIELD_RET(s, field->next_pid);
2175 SEQ_PUT_HEX_FIELD_RET(s, field->next_prio);
2176 SEQ_PUT_HEX_FIELD_RET(s, T);
2180 case TRACE_USER_STACK:
2182 struct special_entry *field;
2184 trace_assign_type(field, entry);
2186 SEQ_PUT_HEX_FIELD_RET(s, field->arg1);
2187 SEQ_PUT_HEX_FIELD_RET(s, field->arg2);
2188 SEQ_PUT_HEX_FIELD_RET(s, field->arg3);
2192 SEQ_PUT_FIELD_RET(s, newline);
2194 return TRACE_TYPE_HANDLED;
2197 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2199 struct trace_seq *s = &iter->seq;
2200 struct trace_entry *entry;
2204 if (entry->type == TRACE_CONT)
2205 return TRACE_TYPE_HANDLED;
2207 SEQ_PUT_FIELD_RET(s, entry->pid);
2208 SEQ_PUT_FIELD_RET(s, entry->cpu);
2209 SEQ_PUT_FIELD_RET(s, iter->ts);
2211 switch (entry->type) {
2213 struct ftrace_entry *field;
2215 trace_assign_type(field, entry);
2217 SEQ_PUT_FIELD_RET(s, field->ip);
2218 SEQ_PUT_FIELD_RET(s, field->parent_ip);
2222 struct ctx_switch_entry *field;
2224 trace_assign_type(field, entry);
2226 SEQ_PUT_FIELD_RET(s, field->prev_pid);
2227 SEQ_PUT_FIELD_RET(s, field->prev_prio);
2228 SEQ_PUT_FIELD_RET(s, field->prev_state);
2229 SEQ_PUT_FIELD_RET(s, field->next_pid);
2230 SEQ_PUT_FIELD_RET(s, field->next_prio);
2231 SEQ_PUT_FIELD_RET(s, field->next_state);
2235 case TRACE_USER_STACK:
2237 struct special_entry *field;
2239 trace_assign_type(field, entry);
2241 SEQ_PUT_FIELD_RET(s, field->arg1);
2242 SEQ_PUT_FIELD_RET(s, field->arg2);
2243 SEQ_PUT_FIELD_RET(s, field->arg3);
2250 static int trace_empty(struct trace_iterator *iter)
2254 for_each_tracing_cpu(cpu) {
2255 if (iter->buffer_iter[cpu]) {
2256 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2259 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2267 static enum print_line_t print_trace_line(struct trace_iterator *iter)
2269 enum print_line_t ret;
2271 if (iter->trace && iter->trace->print_line) {
2272 ret = iter->trace->print_line(iter);
2273 if (ret != TRACE_TYPE_UNHANDLED)
2277 if (trace_flags & TRACE_ITER_BIN)
2278 return print_bin_fmt(iter);
2280 if (trace_flags & TRACE_ITER_HEX)
2281 return print_hex_fmt(iter);
2283 if (trace_flags & TRACE_ITER_RAW)
2284 return print_raw_fmt(iter);
2286 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2287 return print_lat_fmt(iter, iter->idx, iter->cpu);
2289 return print_trace_fmt(iter);
2292 static int s_show(struct seq_file *m, void *v)
2294 struct trace_iterator *iter = v;
2296 if (iter->ent == NULL) {
2298 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2301 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2302 /* print nothing if the buffers are empty */
2303 if (trace_empty(iter))
2305 print_trace_header(m, iter);
2306 if (!(trace_flags & TRACE_ITER_VERBOSE))
2307 print_lat_help_header(m);
2309 if (!(trace_flags & TRACE_ITER_VERBOSE))
2310 print_func_help_header(m);
2313 print_trace_line(iter);
2314 trace_print_seq(m, &iter->seq);
2320 static struct seq_operations tracer_seq_ops = {
2327 static struct trace_iterator *
2328 __tracing_open(struct inode *inode, struct file *file, int *ret)
2330 struct trace_iterator *iter;
2334 if (tracing_disabled) {
2339 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2345 mutex_lock(&trace_types_lock);
2346 if (current_trace && current_trace->print_max)
2349 iter->tr = inode->i_private;
2350 iter->trace = current_trace;
2353 /* Annotate start of buffers if we had overruns */
2354 if (ring_buffer_overruns(iter->tr->buffer))
2355 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2358 for_each_tracing_cpu(cpu) {
2360 iter->buffer_iter[cpu] =
2361 ring_buffer_read_start(iter->tr->buffer, cpu);
2363 if (!iter->buffer_iter[cpu])
2367 /* TODO stop tracer */
2368 *ret = seq_open(file, &tracer_seq_ops);
2372 m = file->private_data;
2375 /* stop the trace while dumping */
2378 if (iter->trace && iter->trace->open)
2379 iter->trace->open(iter);
2381 mutex_unlock(&trace_types_lock);
2387 for_each_tracing_cpu(cpu) {
2388 if (iter->buffer_iter[cpu])
2389 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2391 mutex_unlock(&trace_types_lock);
2394 return ERR_PTR(-ENOMEM);
2397 int tracing_open_generic(struct inode *inode, struct file *filp)
2399 if (tracing_disabled)
2402 filp->private_data = inode->i_private;
2406 int tracing_release(struct inode *inode, struct file *file)
2408 struct seq_file *m = (struct seq_file *)file->private_data;
2409 struct trace_iterator *iter = m->private;
2412 mutex_lock(&trace_types_lock);
2413 for_each_tracing_cpu(cpu) {
2414 if (iter->buffer_iter[cpu])
2415 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2418 if (iter->trace && iter->trace->close)
2419 iter->trace->close(iter);
2421 /* reenable tracing if it was previously enabled */
2423 mutex_unlock(&trace_types_lock);
2425 seq_release(inode, file);
2430 static int tracing_open(struct inode *inode, struct file *file)
2434 __tracing_open(inode, file, &ret);
2439 static int tracing_lt_open(struct inode *inode, struct file *file)
2441 struct trace_iterator *iter;
2444 iter = __tracing_open(inode, file, &ret);
2447 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2454 t_next(struct seq_file *m, void *v, loff_t *pos)
2456 struct tracer *t = m->private;
2468 static void *t_start(struct seq_file *m, loff_t *pos)
2470 struct tracer *t = m->private;
2473 mutex_lock(&trace_types_lock);
2474 for (; t && l < *pos; t = t_next(m, t, &l))
2480 static void t_stop(struct seq_file *m, void *p)
2482 mutex_unlock(&trace_types_lock);
2485 static int t_show(struct seq_file *m, void *v)
2487 struct tracer *t = v;
2492 seq_printf(m, "%s", t->name);
2501 static struct seq_operations show_traces_seq_ops = {
2508 static int show_traces_open(struct inode *inode, struct file *file)
2512 if (tracing_disabled)
2515 ret = seq_open(file, &show_traces_seq_ops);
2517 struct seq_file *m = file->private_data;
2518 m->private = trace_types;
2524 static struct file_operations tracing_fops = {
2525 .open = tracing_open,
2527 .llseek = seq_lseek,
2528 .release = tracing_release,
2531 static struct file_operations tracing_lt_fops = {
2532 .open = tracing_lt_open,
2534 .llseek = seq_lseek,
2535 .release = tracing_release,
2538 static struct file_operations show_traces_fops = {
2539 .open = show_traces_open,
2541 .release = seq_release,
2545 * Only trace on a CPU if the bitmask is set:
2547 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
2550 * When tracing/tracing_cpu_mask is modified then this holds
2551 * the new bitmask we are about to install:
2553 static cpumask_t tracing_cpumask_new;
2556 * The tracer itself will not take this lock, but still we want
2557 * to provide a consistent cpumask to user-space:
2559 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2562 * Temporary storage for the character representation of the
2563 * CPU bitmask (and one more byte for the newline):
2565 static char mask_str[NR_CPUS + 1];
2568 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2569 size_t count, loff_t *ppos)
2573 mutex_lock(&tracing_cpumask_update_lock);
2575 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2576 if (count - len < 2) {
2580 len += sprintf(mask_str + len, "\n");
2581 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2584 mutex_unlock(&tracing_cpumask_update_lock);
2590 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2591 size_t count, loff_t *ppos)
2595 mutex_lock(&tracing_cpumask_update_lock);
2596 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2600 raw_local_irq_disable();
2601 __raw_spin_lock(&ftrace_max_lock);
2602 for_each_tracing_cpu(cpu) {
2604 * Increase/decrease the disabled counter if we are
2605 * about to flip a bit in the cpumask:
2607 if (cpu_isset(cpu, tracing_cpumask) &&
2608 !cpu_isset(cpu, tracing_cpumask_new)) {
2609 atomic_inc(&global_trace.data[cpu]->disabled);
2611 if (!cpu_isset(cpu, tracing_cpumask) &&
2612 cpu_isset(cpu, tracing_cpumask_new)) {
2613 atomic_dec(&global_trace.data[cpu]->disabled);
2616 __raw_spin_unlock(&ftrace_max_lock);
2617 raw_local_irq_enable();
2619 tracing_cpumask = tracing_cpumask_new;
2621 mutex_unlock(&tracing_cpumask_update_lock);
2626 mutex_unlock(&tracing_cpumask_update_lock);
2631 static struct file_operations tracing_cpumask_fops = {
2632 .open = tracing_open_generic,
2633 .read = tracing_cpumask_read,
2634 .write = tracing_cpumask_write,
2638 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2639 size_t cnt, loff_t *ppos)
2645 u32 tracer_flags = current_trace->flags->val;
2646 struct tracer_opt *trace_opts = current_trace->flags->opts;
2649 /* calulate max size */
2650 for (i = 0; trace_options[i]; i++) {
2651 len += strlen(trace_options[i]);
2652 len += 3; /* "no" and space */
2656 * Increase the size with names of options specific
2657 * of the current tracer.
2659 for (i = 0; trace_opts[i].name; i++) {
2660 len += strlen(trace_opts[i].name);
2661 len += 3; /* "no" and space */
2664 /* +2 for \n and \0 */
2665 buf = kmalloc(len + 2, GFP_KERNEL);
2669 for (i = 0; trace_options[i]; i++) {
2670 if (trace_flags & (1 << i))
2671 r += sprintf(buf + r, "%s ", trace_options[i]);
2673 r += sprintf(buf + r, "no%s ", trace_options[i]);
2676 for (i = 0; trace_opts[i].name; i++) {
2677 if (tracer_flags & trace_opts[i].bit)
2678 r += sprintf(buf + r, "%s ",
2679 trace_opts[i].name);
2681 r += sprintf(buf + r, "no%s ",
2682 trace_opts[i].name);
2685 r += sprintf(buf + r, "\n");
2686 WARN_ON(r >= len + 2);
2688 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2695 /* Try to assign a tracer specific option */
2696 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2698 struct tracer_flags *trace_flags = trace->flags;
2699 struct tracer_opt *opts = NULL;
2703 for (i = 0; trace_flags->opts[i].name; i++) {
2704 opts = &trace_flags->opts[i];
2705 len = strlen(opts->name);
2707 if (strncmp(cmp, opts->name, len) == 0) {
2708 ret = trace->set_flag(trace_flags->val,
2714 if (!trace_flags->opts[i].name)
2717 /* Refused to handle */
2722 trace_flags->val &= ~opts->bit;
2724 trace_flags->val |= opts->bit;
2730 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2731 size_t cnt, loff_t *ppos)
2739 if (cnt >= sizeof(buf))
2742 if (copy_from_user(&buf, ubuf, cnt))
2747 if (strncmp(buf, "no", 2) == 0) {
2752 for (i = 0; trace_options[i]; i++) {
2753 int len = strlen(trace_options[i]);
2755 if (strncmp(cmp, trace_options[i], len) == 0) {
2757 trace_flags &= ~(1 << i);
2759 trace_flags |= (1 << i);
2764 /* If no option could be set, test the specific tracer options */
2765 if (!trace_options[i]) {
2766 ret = set_tracer_option(current_trace, cmp, neg);
2776 static struct file_operations tracing_iter_fops = {
2777 .open = tracing_open_generic,
2778 .read = tracing_trace_options_read,
2779 .write = tracing_trace_options_write,
2782 static const char readme_msg[] =
2783 "tracing mini-HOWTO:\n\n"
2785 "# mount -t debugfs nodev /debug\n\n"
2786 "# cat /debug/tracing/available_tracers\n"
2787 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2788 "# cat /debug/tracing/current_tracer\n"
2790 "# echo sched_switch > /debug/tracing/current_tracer\n"
2791 "# cat /debug/tracing/current_tracer\n"
2793 "# cat /debug/tracing/trace_options\n"
2794 "noprint-parent nosym-offset nosym-addr noverbose\n"
2795 "# echo print-parent > /debug/tracing/trace_options\n"
2796 "# echo 1 > /debug/tracing/tracing_enabled\n"
2797 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2798 "echo 0 > /debug/tracing/tracing_enabled\n"
2802 tracing_readme_read(struct file *filp, char __user *ubuf,
2803 size_t cnt, loff_t *ppos)
2805 return simple_read_from_buffer(ubuf, cnt, ppos,
2806 readme_msg, strlen(readme_msg));
2809 static struct file_operations tracing_readme_fops = {
2810 .open = tracing_open_generic,
2811 .read = tracing_readme_read,
2815 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2816 size_t cnt, loff_t *ppos)
2821 r = sprintf(buf, "%u\n", tracer_enabled);
2822 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2826 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2827 size_t cnt, loff_t *ppos)
2829 struct trace_array *tr = filp->private_data;
2834 if (cnt >= sizeof(buf))
2837 if (copy_from_user(&buf, ubuf, cnt))
2842 ret = strict_strtoul(buf, 10, &val);
2848 mutex_lock(&trace_types_lock);
2849 if (tracer_enabled ^ val) {
2852 if (current_trace->start)
2853 current_trace->start(tr);
2858 if (current_trace->stop)
2859 current_trace->stop(tr);
2862 mutex_unlock(&trace_types_lock);
2870 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2871 size_t cnt, loff_t *ppos)
2873 char buf[max_tracer_type_len+2];
2876 mutex_lock(&trace_types_lock);
2878 r = sprintf(buf, "%s\n", current_trace->name);
2880 r = sprintf(buf, "\n");
2881 mutex_unlock(&trace_types_lock);
2883 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2886 static int tracing_set_tracer(char *buf)
2888 struct trace_array *tr = &global_trace;
2892 mutex_lock(&trace_types_lock);
2893 for (t = trace_types; t; t = t->next) {
2894 if (strcmp(t->name, buf) == 0)
2901 if (t == current_trace)
2904 trace_branch_disable();
2905 if (current_trace && current_trace->reset)
2906 current_trace->reset(tr);
2915 trace_branch_enable(tr);
2917 mutex_unlock(&trace_types_lock);
2923 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2924 size_t cnt, loff_t *ppos)
2926 char buf[max_tracer_type_len+1];
2933 if (cnt > max_tracer_type_len)
2934 cnt = max_tracer_type_len;
2936 if (copy_from_user(&buf, ubuf, cnt))
2941 /* strip ending whitespace. */
2942 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2945 err = tracing_set_tracer(buf);
2955 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2956 size_t cnt, loff_t *ppos)
2958 unsigned long *ptr = filp->private_data;
2962 r = snprintf(buf, sizeof(buf), "%ld\n",
2963 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2964 if (r > sizeof(buf))
2966 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2970 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2971 size_t cnt, loff_t *ppos)
2973 long *ptr = filp->private_data;
2978 if (cnt >= sizeof(buf))
2981 if (copy_from_user(&buf, ubuf, cnt))
2986 ret = strict_strtoul(buf, 10, &val);
2995 static atomic_t tracing_reader;
2997 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2999 struct trace_iterator *iter;
3001 if (tracing_disabled)
3004 /* We only allow for reader of the pipe */
3005 if (atomic_inc_return(&tracing_reader) != 1) {
3006 atomic_dec(&tracing_reader);
3010 /* create a buffer to store the information to pass to userspace */
3011 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3015 mutex_lock(&trace_types_lock);
3017 /* trace pipe does not show start of buffer */
3018 cpus_setall(iter->started);
3020 iter->tr = &global_trace;
3021 iter->trace = current_trace;
3022 filp->private_data = iter;
3024 if (iter->trace->pipe_open)
3025 iter->trace->pipe_open(iter);
3026 mutex_unlock(&trace_types_lock);
3031 static int tracing_release_pipe(struct inode *inode, struct file *file)
3033 struct trace_iterator *iter = file->private_data;
3036 atomic_dec(&tracing_reader);
3042 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3044 struct trace_iterator *iter = filp->private_data;
3046 if (trace_flags & TRACE_ITER_BLOCK) {
3048 * Always select as readable when in blocking mode
3050 return POLLIN | POLLRDNORM;
3052 if (!trace_empty(iter))
3053 return POLLIN | POLLRDNORM;
3054 poll_wait(filp, &trace_wait, poll_table);
3055 if (!trace_empty(iter))
3056 return POLLIN | POLLRDNORM;
3066 tracing_read_pipe(struct file *filp, char __user *ubuf,
3067 size_t cnt, loff_t *ppos)
3069 struct trace_iterator *iter = filp->private_data;
3072 /* return any leftover data */
3073 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3077 trace_seq_reset(&iter->seq);
3079 mutex_lock(&trace_types_lock);
3080 if (iter->trace->read) {
3081 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3088 while (trace_empty(iter)) {
3090 if ((filp->f_flags & O_NONBLOCK)) {
3096 * This is a make-shift waitqueue. The reason we don't use
3097 * an actual wait queue is because:
3098 * 1) we only ever have one waiter
3099 * 2) the tracing, traces all functions, we don't want
3100 * the overhead of calling wake_up and friends
3101 * (and tracing them too)
3102 * Anyway, this is really very primitive wakeup.
3104 set_current_state(TASK_INTERRUPTIBLE);
3105 iter->tr->waiter = current;
3107 mutex_unlock(&trace_types_lock);
3109 /* sleep for 100 msecs, and try again. */
3110 schedule_timeout(HZ/10);
3112 mutex_lock(&trace_types_lock);
3114 iter->tr->waiter = NULL;
3116 if (signal_pending(current)) {
3121 if (iter->trace != current_trace)
3125 * We block until we read something and tracing is disabled.
3126 * We still block if tracing is disabled, but we have never
3127 * read anything. This allows a user to cat this file, and
3128 * then enable tracing. But after we have read something,
3129 * we give an EOF when tracing is again disabled.
3131 * iter->pos will be 0 if we haven't read anything.
3133 if (!tracer_enabled && iter->pos)
3139 /* stop when tracing is finished */
3140 if (trace_empty(iter))
3143 if (cnt >= PAGE_SIZE)
3144 cnt = PAGE_SIZE - 1;
3146 /* reset all but tr, trace, and overruns */
3147 memset(&iter->seq, 0,
3148 sizeof(struct trace_iterator) -
3149 offsetof(struct trace_iterator, seq));
3152 while (find_next_entry_inc(iter) != NULL) {
3153 enum print_line_t ret;
3154 int len = iter->seq.len;
3156 ret = print_trace_line(iter);
3157 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3158 /* don't print partial lines */
3159 iter->seq.len = len;
3163 trace_consume(iter);
3165 if (iter->seq.len >= cnt)
3169 /* Now copy what we have to the user */
3170 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3171 if (iter->seq.readpos >= iter->seq.len)
3172 trace_seq_reset(&iter->seq);
3175 * If there was nothing to send to user, inspite of consuming trace
3176 * entries, go back to wait for more entries.
3182 mutex_unlock(&trace_types_lock);
3188 tracing_entries_read(struct file *filp, char __user *ubuf,
3189 size_t cnt, loff_t *ppos)
3191 struct trace_array *tr = filp->private_data;
3195 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3196 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3200 tracing_entries_write(struct file *filp, const char __user *ubuf,
3201 size_t cnt, loff_t *ppos)
3207 if (cnt >= sizeof(buf))
3210 if (copy_from_user(&buf, ubuf, cnt))
3215 ret = strict_strtoul(buf, 10, &val);
3219 /* must have at least 1 entry */
3223 mutex_lock(&trace_types_lock);
3227 /* disable all cpu buffers */
3228 for_each_tracing_cpu(cpu) {
3229 if (global_trace.data[cpu])
3230 atomic_inc(&global_trace.data[cpu]->disabled);
3231 if (max_tr.data[cpu])
3232 atomic_inc(&max_tr.data[cpu]->disabled);
3235 /* value is in KB */
3238 if (val != global_trace.entries) {
3239 ret = ring_buffer_resize(global_trace.buffer, val);
3245 ret = ring_buffer_resize(max_tr.buffer, val);
3249 r = ring_buffer_resize(global_trace.buffer,
3250 global_trace.entries);
3252 /* AARGH! We are left with different
3253 * size max buffer!!!! */
3255 tracing_disabled = 1;
3260 global_trace.entries = val;
3265 /* If check pages failed, return ENOMEM */
3266 if (tracing_disabled)
3269 for_each_tracing_cpu(cpu) {
3270 if (global_trace.data[cpu])
3271 atomic_dec(&global_trace.data[cpu]->disabled);
3272 if (max_tr.data[cpu])
3273 atomic_dec(&max_tr.data[cpu]->disabled);
3277 max_tr.entries = global_trace.entries;
3278 mutex_unlock(&trace_types_lock);
3283 static int mark_printk(const char *fmt, ...)
3287 va_start(args, fmt);
3288 ret = trace_vprintk(0, fmt, args);
3294 tracing_mark_write(struct file *filp, const char __user *ubuf,
3295 size_t cnt, loff_t *fpos)
3300 if (tracing_disabled)
3303 if (cnt > TRACE_BUF_SIZE)
3304 cnt = TRACE_BUF_SIZE;
3306 buf = kmalloc(cnt + 1, GFP_KERNEL);
3310 if (copy_from_user(buf, ubuf, cnt)) {
3315 /* Cut from the first nil or newline. */
3317 end = strchr(buf, '\n');
3321 cnt = mark_printk("%s\n", buf);
3328 static struct file_operations tracing_max_lat_fops = {
3329 .open = tracing_open_generic,
3330 .read = tracing_max_lat_read,
3331 .write = tracing_max_lat_write,
3334 static struct file_operations tracing_ctrl_fops = {
3335 .open = tracing_open_generic,
3336 .read = tracing_ctrl_read,
3337 .write = tracing_ctrl_write,
3340 static struct file_operations set_tracer_fops = {
3341 .open = tracing_open_generic,
3342 .read = tracing_set_trace_read,
3343 .write = tracing_set_trace_write,
3346 static struct file_operations tracing_pipe_fops = {
3347 .open = tracing_open_pipe,
3348 .poll = tracing_poll_pipe,
3349 .read = tracing_read_pipe,
3350 .release = tracing_release_pipe,
3353 static struct file_operations tracing_entries_fops = {
3354 .open = tracing_open_generic,
3355 .read = tracing_entries_read,
3356 .write = tracing_entries_write,
3359 static struct file_operations tracing_mark_fops = {
3360 .open = tracing_open_generic,
3361 .write = tracing_mark_write,
3364 #ifdef CONFIG_DYNAMIC_FTRACE
3366 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3372 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3373 size_t cnt, loff_t *ppos)
3375 static char ftrace_dyn_info_buffer[1024];
3376 static DEFINE_MUTEX(dyn_info_mutex);
3377 unsigned long *p = filp->private_data;
3378 char *buf = ftrace_dyn_info_buffer;
3379 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3382 mutex_lock(&dyn_info_mutex);
3383 r = sprintf(buf, "%ld ", *p);
3385 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3388 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3390 mutex_unlock(&dyn_info_mutex);
3395 static struct file_operations tracing_dyn_info_fops = {
3396 .open = tracing_open_generic,
3397 .read = tracing_read_dyn_info,
3401 static struct dentry *d_tracer;
3403 struct dentry *tracing_init_dentry(void)
3410 d_tracer = debugfs_create_dir("tracing", NULL);
3412 if (!d_tracer && !once) {
3414 pr_warning("Could not create debugfs directory 'tracing'\n");
3421 #ifdef CONFIG_FTRACE_SELFTEST
3422 /* Let selftest have access to static functions in this file */
3423 #include "trace_selftest.c"
3426 static __init int tracer_init_debugfs(void)
3428 struct dentry *d_tracer;
3429 struct dentry *entry;
3431 d_tracer = tracing_init_dentry();
3433 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3434 &global_trace, &tracing_ctrl_fops);
3436 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3438 entry = debugfs_create_file("trace_options", 0644, d_tracer,
3439 NULL, &tracing_iter_fops);
3441 pr_warning("Could not create debugfs 'trace_options' entry\n");
3443 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3444 NULL, &tracing_cpumask_fops);
3446 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3448 entry = debugfs_create_file("latency_trace", 0444, d_tracer,
3449 &global_trace, &tracing_lt_fops);
3451 pr_warning("Could not create debugfs 'latency_trace' entry\n");
3453 entry = debugfs_create_file("trace", 0444, d_tracer,
3454 &global_trace, &tracing_fops);
3456 pr_warning("Could not create debugfs 'trace' entry\n");
3458 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3459 &global_trace, &show_traces_fops);
3461 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3463 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3464 &global_trace, &set_tracer_fops);
3466 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3468 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3469 &tracing_max_latency,
3470 &tracing_max_lat_fops);
3472 pr_warning("Could not create debugfs "
3473 "'tracing_max_latency' entry\n");
3475 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3476 &tracing_thresh, &tracing_max_lat_fops);
3478 pr_warning("Could not create debugfs "
3479 "'tracing_thresh' entry\n");
3480 entry = debugfs_create_file("README", 0644, d_tracer,
3481 NULL, &tracing_readme_fops);
3483 pr_warning("Could not create debugfs 'README' entry\n");
3485 entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
3486 NULL, &tracing_pipe_fops);
3488 pr_warning("Could not create debugfs "
3489 "'trace_pipe' entry\n");
3491 entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
3492 &global_trace, &tracing_entries_fops);
3494 pr_warning("Could not create debugfs "
3495 "'buffer_size_kb' entry\n");
3497 entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3498 NULL, &tracing_mark_fops);
3500 pr_warning("Could not create debugfs "
3501 "'trace_marker' entry\n");
3503 #ifdef CONFIG_DYNAMIC_FTRACE
3504 entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3505 &ftrace_update_tot_cnt,
3506 &tracing_dyn_info_fops);
3508 pr_warning("Could not create debugfs "
3509 "'dyn_ftrace_total_info' entry\n");
3511 #ifdef CONFIG_SYSPROF_TRACER
3512 init_tracer_sysprof_debugfs(d_tracer);
3517 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3519 static DEFINE_SPINLOCK(trace_buf_lock);
3520 static char trace_buf[TRACE_BUF_SIZE];
3522 struct ring_buffer_event *event;
3523 struct trace_array *tr = &global_trace;
3524 struct trace_array_cpu *data;
3525 struct print_entry *entry;
3526 unsigned long flags, irq_flags;
3527 int cpu, len = 0, size, pc;
3529 if (tracing_disabled)
3532 pc = preempt_count();
3533 preempt_disable_notrace();
3534 cpu = raw_smp_processor_id();
3535 data = tr->data[cpu];
3537 if (unlikely(atomic_read(&data->disabled)))
3540 spin_lock_irqsave(&trace_buf_lock, flags);
3541 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
3543 len = min(len, TRACE_BUF_SIZE-1);
3546 size = sizeof(*entry) + len + 1;
3547 event = ring_buffer_lock_reserve(tr->buffer, size, &irq_flags);
3550 entry = ring_buffer_event_data(event);
3551 tracing_generic_entry_update(&entry->ent, flags, pc);
3552 entry->ent.type = TRACE_PRINT;
3555 memcpy(&entry->buf, trace_buf, len);
3556 entry->buf[len] = 0;
3557 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
3560 spin_unlock_irqrestore(&trace_buf_lock, flags);
3563 preempt_enable_notrace();
3567 EXPORT_SYMBOL_GPL(trace_vprintk);
3569 int __ftrace_printk(unsigned long ip, const char *fmt, ...)
3574 if (!(trace_flags & TRACE_ITER_PRINTK))
3578 ret = trace_vprintk(ip, fmt, ap);
3582 EXPORT_SYMBOL_GPL(__ftrace_printk);
3584 static int trace_panic_handler(struct notifier_block *this,
3585 unsigned long event, void *unused)
3587 if (ftrace_dump_on_oops)
3592 static struct notifier_block trace_panic_notifier = {
3593 .notifier_call = trace_panic_handler,
3595 .priority = 150 /* priority: INT_MAX >= x >= 0 */
3598 static int trace_die_handler(struct notifier_block *self,
3604 if (ftrace_dump_on_oops)
3613 static struct notifier_block trace_die_notifier = {
3614 .notifier_call = trace_die_handler,
3619 * printk is set to max of 1024, we really don't need it that big.
3620 * Nothing should be printing 1000 characters anyway.
3622 #define TRACE_MAX_PRINT 1000
3625 * Define here KERN_TRACE so that we have one place to modify
3626 * it if we decide to change what log level the ftrace dump
3629 #define KERN_TRACE KERN_INFO
3632 trace_printk_seq(struct trace_seq *s)
3634 /* Probably should print a warning here. */
3638 /* should be zero ended, but we are paranoid. */
3639 s->buffer[s->len] = 0;
3641 printk(KERN_TRACE "%s", s->buffer);
3646 void ftrace_dump(void)
3648 static DEFINE_SPINLOCK(ftrace_dump_lock);
3649 /* use static because iter can be a bit big for the stack */
3650 static struct trace_iterator iter;
3651 static cpumask_t mask;
3652 static int dump_ran;
3653 unsigned long flags;
3657 spin_lock_irqsave(&ftrace_dump_lock, flags);
3663 /* No turning back! */
3666 for_each_tracing_cpu(cpu) {
3667 atomic_inc(&global_trace.data[cpu]->disabled);
3670 /* don't look at user memory in panic mode */
3671 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
3673 printk(KERN_TRACE "Dumping ftrace buffer:\n");
3675 iter.tr = &global_trace;
3676 iter.trace = current_trace;
3679 * We need to stop all tracing on all CPUS to read the
3680 * the next buffer. This is a bit expensive, but is
3681 * not done often. We fill all what we can read,
3682 * and then release the locks again.
3687 while (!trace_empty(&iter)) {
3690 printk(KERN_TRACE "---------------------------------\n");
3694 /* reset all but tr, trace, and overruns */
3695 memset(&iter.seq, 0,
3696 sizeof(struct trace_iterator) -
3697 offsetof(struct trace_iterator, seq));
3698 iter.iter_flags |= TRACE_FILE_LAT_FMT;
3701 if (find_next_entry_inc(&iter) != NULL) {
3702 print_trace_line(&iter);
3703 trace_consume(&iter);
3706 trace_printk_seq(&iter.seq);
3710 printk(KERN_TRACE " (ftrace buffer empty)\n");
3712 printk(KERN_TRACE "---------------------------------\n");
3715 spin_unlock_irqrestore(&ftrace_dump_lock, flags);
3718 __init static int tracer_alloc_buffers(void)
3720 struct trace_array_cpu *data;
3723 /* TODO: make the number of buffers hot pluggable with CPUS */
3724 tracing_buffer_mask = cpu_possible_map;
3726 global_trace.buffer = ring_buffer_alloc(trace_buf_size,
3727 TRACE_BUFFER_FLAGS);
3728 if (!global_trace.buffer) {
3729 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
3733 global_trace.entries = ring_buffer_size(global_trace.buffer);
3735 #ifdef CONFIG_TRACER_MAX_TRACE
3736 max_tr.buffer = ring_buffer_alloc(trace_buf_size,
3737 TRACE_BUFFER_FLAGS);
3738 if (!max_tr.buffer) {
3739 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
3741 ring_buffer_free(global_trace.buffer);
3744 max_tr.entries = ring_buffer_size(max_tr.buffer);
3745 WARN_ON(max_tr.entries != global_trace.entries);
3748 /* Allocate the first page for all buffers */
3749 for_each_tracing_cpu(i) {
3750 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
3751 max_tr.data[i] = &per_cpu(max_data, i);
3754 trace_init_cmdlines();
3756 register_tracer(&nop_trace);
3757 #ifdef CONFIG_BOOT_TRACER
3758 register_tracer(&boot_tracer);
3759 current_trace = &boot_tracer;
3760 current_trace->init(&global_trace);
3762 current_trace = &nop_trace;
3765 /* All seems OK, enable tracing */
3766 tracing_disabled = 0;
3768 atomic_notifier_chain_register(&panic_notifier_list,
3769 &trace_panic_notifier);
3771 register_die_notifier(&trace_die_notifier);
3775 early_initcall(tracer_alloc_buffers);
3776 fs_initcall(tracer_init_debugfs);