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 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_GRAPH_TRACER
882 static void __trace_graph_entry(struct trace_array *tr,
883 struct trace_array_cpu *data,
884 struct ftrace_graph_ent *trace,
888 struct ring_buffer_event *event;
889 struct ftrace_graph_ent_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_GRAPH_ENT;
902 entry->graph_ent = *trace;
903 ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
906 static void __trace_graph_return(struct trace_array *tr,
907 struct trace_array_cpu *data,
908 struct ftrace_graph_ret *trace,
912 struct ring_buffer_event *event;
913 struct ftrace_graph_ret_entry *entry;
914 unsigned long irq_flags;
916 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
919 event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry),
923 entry = ring_buffer_event_data(event);
924 tracing_generic_entry_update(&entry->ent, flags, pc);
925 entry->ent.type = TRACE_GRAPH_RET;
927 ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
932 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
933 unsigned long ip, unsigned long parent_ip, unsigned long flags,
936 if (likely(!atomic_read(&data->disabled)))
937 trace_function(tr, data, ip, parent_ip, flags, pc);
940 static void ftrace_trace_stack(struct trace_array *tr,
941 struct trace_array_cpu *data,
945 #ifdef CONFIG_STACKTRACE
946 struct ring_buffer_event *event;
947 struct stack_entry *entry;
948 struct stack_trace trace;
949 unsigned long irq_flags;
951 if (!(trace_flags & TRACE_ITER_STACKTRACE))
954 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
958 entry = ring_buffer_event_data(event);
959 tracing_generic_entry_update(&entry->ent, flags, pc);
960 entry->ent.type = TRACE_STACK;
962 memset(&entry->caller, 0, sizeof(entry->caller));
964 trace.nr_entries = 0;
965 trace.max_entries = FTRACE_STACK_ENTRIES;
967 trace.entries = entry->caller;
969 save_stack_trace(&trace);
970 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
974 void __trace_stack(struct trace_array *tr,
975 struct trace_array_cpu *data,
979 ftrace_trace_stack(tr, data, flags, skip, preempt_count());
982 static void ftrace_trace_userstack(struct trace_array *tr,
983 struct trace_array_cpu *data,
984 unsigned long flags, int pc)
986 #ifdef CONFIG_STACKTRACE
987 struct ring_buffer_event *event;
988 struct userstack_entry *entry;
989 struct stack_trace trace;
990 unsigned long irq_flags;
992 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
995 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
999 entry = ring_buffer_event_data(event);
1000 tracing_generic_entry_update(&entry->ent, flags, pc);
1001 entry->ent.type = TRACE_USER_STACK;
1003 memset(&entry->caller, 0, sizeof(entry->caller));
1005 trace.nr_entries = 0;
1006 trace.max_entries = FTRACE_STACK_ENTRIES;
1008 trace.entries = entry->caller;
1010 save_stack_trace_user(&trace);
1011 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1015 void __trace_userstack(struct trace_array *tr,
1016 struct trace_array_cpu *data,
1017 unsigned long flags)
1019 ftrace_trace_userstack(tr, data, flags, preempt_count());
1023 ftrace_trace_special(void *__tr, void *__data,
1024 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1027 struct ring_buffer_event *event;
1028 struct trace_array_cpu *data = __data;
1029 struct trace_array *tr = __tr;
1030 struct special_entry *entry;
1031 unsigned long irq_flags;
1033 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1037 entry = ring_buffer_event_data(event);
1038 tracing_generic_entry_update(&entry->ent, 0, pc);
1039 entry->ent.type = TRACE_SPECIAL;
1043 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1044 ftrace_trace_stack(tr, data, irq_flags, 4, pc);
1045 ftrace_trace_userstack(tr, data, irq_flags, pc);
1051 __trace_special(void *__tr, void *__data,
1052 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1054 ftrace_trace_special(__tr, __data, arg1, arg2, arg3, preempt_count());
1058 tracing_sched_switch_trace(struct trace_array *tr,
1059 struct trace_array_cpu *data,
1060 struct task_struct *prev,
1061 struct task_struct *next,
1062 unsigned long flags, int pc)
1064 struct ring_buffer_event *event;
1065 struct ctx_switch_entry *entry;
1066 unsigned long irq_flags;
1068 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1072 entry = ring_buffer_event_data(event);
1073 tracing_generic_entry_update(&entry->ent, flags, pc);
1074 entry->ent.type = TRACE_CTX;
1075 entry->prev_pid = prev->pid;
1076 entry->prev_prio = prev->prio;
1077 entry->prev_state = prev->state;
1078 entry->next_pid = next->pid;
1079 entry->next_prio = next->prio;
1080 entry->next_state = next->state;
1081 entry->next_cpu = task_cpu(next);
1082 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1083 ftrace_trace_stack(tr, data, flags, 5, pc);
1084 ftrace_trace_userstack(tr, data, flags, pc);
1088 tracing_sched_wakeup_trace(struct trace_array *tr,
1089 struct trace_array_cpu *data,
1090 struct task_struct *wakee,
1091 struct task_struct *curr,
1092 unsigned long flags, int pc)
1094 struct ring_buffer_event *event;
1095 struct ctx_switch_entry *entry;
1096 unsigned long irq_flags;
1098 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1102 entry = ring_buffer_event_data(event);
1103 tracing_generic_entry_update(&entry->ent, flags, pc);
1104 entry->ent.type = TRACE_WAKE;
1105 entry->prev_pid = curr->pid;
1106 entry->prev_prio = curr->prio;
1107 entry->prev_state = curr->state;
1108 entry->next_pid = wakee->pid;
1109 entry->next_prio = wakee->prio;
1110 entry->next_state = wakee->state;
1111 entry->next_cpu = task_cpu(wakee);
1112 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1113 ftrace_trace_stack(tr, data, flags, 6, pc);
1114 ftrace_trace_userstack(tr, data, flags, pc);
1120 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1122 struct trace_array *tr = &global_trace;
1123 struct trace_array_cpu *data;
1124 unsigned long flags;
1128 if (tracing_disabled)
1131 pc = preempt_count();
1132 local_irq_save(flags);
1133 cpu = raw_smp_processor_id();
1134 data = tr->data[cpu];
1136 if (likely(atomic_inc_return(&data->disabled) == 1))
1137 ftrace_trace_special(tr, data, arg1, arg2, arg3, pc);
1139 atomic_dec(&data->disabled);
1140 local_irq_restore(flags);
1143 #ifdef CONFIG_FUNCTION_TRACER
1145 function_trace_call_preempt_only(unsigned long ip, unsigned long parent_ip)
1147 struct trace_array *tr = &global_trace;
1148 struct trace_array_cpu *data;
1149 unsigned long flags;
1154 if (unlikely(!ftrace_function_enabled))
1157 pc = preempt_count();
1158 resched = ftrace_preempt_disable();
1159 local_save_flags(flags);
1160 cpu = raw_smp_processor_id();
1161 data = tr->data[cpu];
1162 disabled = atomic_inc_return(&data->disabled);
1164 if (likely(disabled == 1))
1165 trace_function(tr, data, ip, parent_ip, flags, pc);
1167 atomic_dec(&data->disabled);
1168 ftrace_preempt_enable(resched);
1172 function_trace_call(unsigned long ip, unsigned long parent_ip)
1174 struct trace_array *tr = &global_trace;
1175 struct trace_array_cpu *data;
1176 unsigned long flags;
1181 if (unlikely(!ftrace_function_enabled))
1185 * Need to use raw, since this must be called before the
1186 * recursive protection is performed.
1188 local_irq_save(flags);
1189 cpu = raw_smp_processor_id();
1190 data = tr->data[cpu];
1191 disabled = atomic_inc_return(&data->disabled);
1193 if (likely(disabled == 1)) {
1194 pc = preempt_count();
1195 trace_function(tr, data, ip, parent_ip, flags, pc);
1198 atomic_dec(&data->disabled);
1199 local_irq_restore(flags);
1202 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1203 void trace_graph_entry(struct ftrace_graph_ent *trace)
1205 struct trace_array *tr = &global_trace;
1206 struct trace_array_cpu *data;
1207 unsigned long flags;
1212 raw_local_irq_save(flags);
1213 cpu = raw_smp_processor_id();
1214 data = tr->data[cpu];
1215 disabled = atomic_inc_return(&data->disabled);
1216 if (likely(disabled == 1)) {
1217 pc = preempt_count();
1218 __trace_graph_entry(tr, data, trace, flags, pc);
1220 atomic_dec(&data->disabled);
1221 raw_local_irq_restore(flags);
1224 void trace_graph_return(struct ftrace_graph_ret *trace)
1226 struct trace_array *tr = &global_trace;
1227 struct trace_array_cpu *data;
1228 unsigned long flags;
1233 raw_local_irq_save(flags);
1234 cpu = raw_smp_processor_id();
1235 data = tr->data[cpu];
1236 disabled = atomic_inc_return(&data->disabled);
1237 if (likely(disabled == 1)) {
1238 pc = preempt_count();
1239 __trace_graph_return(tr, data, trace, flags, pc);
1241 atomic_dec(&data->disabled);
1242 raw_local_irq_restore(flags);
1244 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1246 static struct ftrace_ops trace_ops __read_mostly =
1248 .func = function_trace_call,
1251 void tracing_start_function_trace(void)
1253 ftrace_function_enabled = 0;
1255 if (trace_flags & TRACE_ITER_PREEMPTONLY)
1256 trace_ops.func = function_trace_call_preempt_only;
1258 trace_ops.func = function_trace_call;
1260 register_ftrace_function(&trace_ops);
1261 ftrace_function_enabled = 1;
1264 void tracing_stop_function_trace(void)
1266 ftrace_function_enabled = 0;
1267 unregister_ftrace_function(&trace_ops);
1271 enum trace_file_type {
1272 TRACE_FILE_LAT_FMT = 1,
1273 TRACE_FILE_ANNOTATE = 2,
1276 static void trace_iterator_increment(struct trace_iterator *iter, int cpu)
1278 /* Don't allow ftrace to trace into the ring buffers */
1279 ftrace_disable_cpu();
1282 if (iter->buffer_iter[iter->cpu])
1283 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1285 ftrace_enable_cpu();
1288 static struct trace_entry *
1289 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1291 struct ring_buffer_event *event;
1292 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1294 /* Don't allow ftrace to trace into the ring buffers */
1295 ftrace_disable_cpu();
1298 event = ring_buffer_iter_peek(buf_iter, ts);
1300 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1302 ftrace_enable_cpu();
1304 return event ? ring_buffer_event_data(event) : NULL;
1307 static struct trace_entry *
1308 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1310 struct ring_buffer *buffer = iter->tr->buffer;
1311 struct trace_entry *ent, *next = NULL;
1312 u64 next_ts = 0, ts;
1316 for_each_tracing_cpu(cpu) {
1318 if (ring_buffer_empty_cpu(buffer, cpu))
1321 ent = peek_next_entry(iter, cpu, &ts);
1324 * Pick the entry with the smallest timestamp:
1326 if (ent && (!next || ts < next_ts)) {
1334 *ent_cpu = next_cpu;
1342 /* Find the next real entry, without updating the iterator itself */
1343 static struct trace_entry *
1344 find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1346 return __find_next_entry(iter, ent_cpu, ent_ts);
1349 /* Find the next real entry, and increment the iterator to the next entry */
1350 static void *find_next_entry_inc(struct trace_iterator *iter)
1352 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1355 trace_iterator_increment(iter, iter->cpu);
1357 return iter->ent ? iter : NULL;
1360 static void trace_consume(struct trace_iterator *iter)
1362 /* Don't allow ftrace to trace into the ring buffers */
1363 ftrace_disable_cpu();
1364 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1365 ftrace_enable_cpu();
1368 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1370 struct trace_iterator *iter = m->private;
1376 /* can't go backwards */
1381 ent = find_next_entry_inc(iter);
1385 while (ent && iter->idx < i)
1386 ent = find_next_entry_inc(iter);
1393 static void *s_start(struct seq_file *m, loff_t *pos)
1395 struct trace_iterator *iter = m->private;
1400 mutex_lock(&trace_types_lock);
1402 if (!current_trace || current_trace != iter->trace) {
1403 mutex_unlock(&trace_types_lock);
1407 atomic_inc(&trace_record_cmdline_disabled);
1409 if (*pos != iter->pos) {
1414 ftrace_disable_cpu();
1416 for_each_tracing_cpu(cpu) {
1417 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1420 ftrace_enable_cpu();
1422 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1427 p = s_next(m, p, &l);
1433 static void s_stop(struct seq_file *m, void *p)
1435 atomic_dec(&trace_record_cmdline_disabled);
1436 mutex_unlock(&trace_types_lock);
1439 #ifdef CONFIG_KRETPROBES
1440 static inline const char *kretprobed(const char *name)
1442 static const char tramp_name[] = "kretprobe_trampoline";
1443 int size = sizeof(tramp_name);
1445 if (strncmp(tramp_name, name, size) == 0)
1446 return "[unknown/kretprobe'd]";
1450 static inline const char *kretprobed(const char *name)
1454 #endif /* CONFIG_KRETPROBES */
1457 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1459 #ifdef CONFIG_KALLSYMS
1460 char str[KSYM_SYMBOL_LEN];
1463 kallsyms_lookup(address, NULL, NULL, NULL, str);
1465 name = kretprobed(str);
1467 return trace_seq_printf(s, fmt, name);
1473 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1474 unsigned long address)
1476 #ifdef CONFIG_KALLSYMS
1477 char str[KSYM_SYMBOL_LEN];
1480 sprint_symbol(str, address);
1481 name = kretprobed(str);
1483 return trace_seq_printf(s, fmt, name);
1488 #ifndef CONFIG_64BIT
1489 # define IP_FMT "%08lx"
1491 # define IP_FMT "%016lx"
1495 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1500 return trace_seq_printf(s, "0");
1502 if (sym_flags & TRACE_ITER_SYM_OFFSET)
1503 ret = seq_print_sym_offset(s, "%s", ip);
1505 ret = seq_print_sym_short(s, "%s", ip);
1510 if (sym_flags & TRACE_ITER_SYM_ADDR)
1511 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1515 static inline int seq_print_user_ip(struct trace_seq *s, struct mm_struct *mm,
1516 unsigned long ip, unsigned long sym_flags)
1518 struct file *file = NULL;
1519 unsigned long vmstart = 0;
1523 const struct vm_area_struct *vma;
1525 down_read(&mm->mmap_sem);
1526 vma = find_vma(mm, ip);
1528 file = vma->vm_file;
1529 vmstart = vma->vm_start;
1532 ret = trace_seq_path(s, &file->f_path);
1534 ret = trace_seq_printf(s, "[+0x%lx]", ip - vmstart);
1536 up_read(&mm->mmap_sem);
1538 if (ret && ((sym_flags & TRACE_ITER_SYM_ADDR) || !file))
1539 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1544 seq_print_userip_objs(const struct userstack_entry *entry, struct trace_seq *s,
1545 unsigned long sym_flags)
1547 struct mm_struct *mm = NULL;
1551 if (trace_flags & TRACE_ITER_SYM_USEROBJ) {
1552 struct task_struct *task;
1554 * we do the lookup on the thread group leader,
1555 * since individual threads might have already quit!
1558 task = find_task_by_vpid(entry->ent.tgid);
1560 mm = get_task_mm(task);
1564 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1565 unsigned long ip = entry->caller[i];
1567 if (ip == ULONG_MAX || !ret)
1570 ret = trace_seq_puts(s, " <- ");
1573 ret = trace_seq_puts(s, "??");
1579 ret = seq_print_user_ip(s, mm, ip, sym_flags);
1587 static void print_lat_help_header(struct seq_file *m)
1589 seq_puts(m, "# _------=> CPU# \n");
1590 seq_puts(m, "# / _-----=> irqs-off \n");
1591 seq_puts(m, "# | / _----=> need-resched \n");
1592 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1593 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1594 seq_puts(m, "# |||| / \n");
1595 seq_puts(m, "# ||||| delay \n");
1596 seq_puts(m, "# cmd pid ||||| time | caller \n");
1597 seq_puts(m, "# \\ / ||||| \\ | / \n");
1600 static void print_func_help_header(struct seq_file *m)
1602 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1603 seq_puts(m, "# | | | | |\n");
1608 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1610 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1611 struct trace_array *tr = iter->tr;
1612 struct trace_array_cpu *data = tr->data[tr->cpu];
1613 struct tracer *type = current_trace;
1614 unsigned long total;
1615 unsigned long entries;
1616 const char *name = "preemption";
1621 entries = ring_buffer_entries(iter->tr->buffer);
1623 ring_buffer_overruns(iter->tr->buffer);
1625 seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1627 seq_puts(m, "-----------------------------------"
1628 "---------------------------------\n");
1629 seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1630 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1631 nsecs_to_usecs(data->saved_latency),
1635 #if defined(CONFIG_PREEMPT_NONE)
1637 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1639 #elif defined(CONFIG_PREEMPT)
1644 /* These are reserved for later use */
1647 seq_printf(m, " #P:%d)\n", num_online_cpus());
1651 seq_puts(m, " -----------------\n");
1652 seq_printf(m, " | task: %.16s-%d "
1653 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1654 data->comm, data->pid, data->uid, data->nice,
1655 data->policy, data->rt_priority);
1656 seq_puts(m, " -----------------\n");
1658 if (data->critical_start) {
1659 seq_puts(m, " => started at: ");
1660 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1661 trace_print_seq(m, &iter->seq);
1662 seq_puts(m, "\n => ended at: ");
1663 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1664 trace_print_seq(m, &iter->seq);
1672 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1674 int hardirq, softirq;
1677 comm = trace_find_cmdline(entry->pid);
1679 trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1680 trace_seq_printf(s, "%3d", cpu);
1681 trace_seq_printf(s, "%c%c",
1682 (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' :
1683 (entry->flags & TRACE_FLAG_IRQS_NOSUPPORT) ? 'X' : '.',
1684 ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1686 hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1687 softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1688 if (hardirq && softirq) {
1689 trace_seq_putc(s, 'H');
1692 trace_seq_putc(s, 'h');
1695 trace_seq_putc(s, 's');
1697 trace_seq_putc(s, '.');
1701 if (entry->preempt_count)
1702 trace_seq_printf(s, "%x", entry->preempt_count);
1704 trace_seq_puts(s, ".");
1707 unsigned long preempt_mark_thresh = 100;
1710 lat_print_timestamp(struct trace_seq *s, u64 abs_usecs,
1711 unsigned long rel_usecs)
1713 trace_seq_printf(s, " %4lldus", abs_usecs);
1714 if (rel_usecs > preempt_mark_thresh)
1715 trace_seq_puts(s, "!: ");
1716 else if (rel_usecs > 1)
1717 trace_seq_puts(s, "+: ");
1719 trace_seq_puts(s, " : ");
1722 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1725 * The message is supposed to contain an ending newline.
1726 * If the printing stops prematurely, try to add a newline of our own.
1728 void trace_seq_print_cont(struct trace_seq *s, struct trace_iterator *iter)
1730 struct trace_entry *ent;
1731 struct trace_field_cont *cont;
1734 ent = peek_next_entry(iter, iter->cpu, NULL);
1735 if (!ent || ent->type != TRACE_CONT) {
1736 trace_seq_putc(s, '\n');
1741 cont = (struct trace_field_cont *)ent;
1743 ok = (trace_seq_printf(s, "%s", cont->buf) > 0);
1745 ftrace_disable_cpu();
1747 if (iter->buffer_iter[iter->cpu])
1748 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1750 ring_buffer_consume(iter->tr->buffer, iter->cpu, NULL);
1752 ftrace_enable_cpu();
1754 ent = peek_next_entry(iter, iter->cpu, NULL);
1755 } while (ent && ent->type == TRACE_CONT);
1758 trace_seq_putc(s, '\n');
1761 static void test_cpu_buff_start(struct trace_iterator *iter)
1763 struct trace_seq *s = &iter->seq;
1765 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1768 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1771 if (cpu_isset(iter->cpu, iter->started))
1774 cpu_set(iter->cpu, iter->started);
1775 trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1778 static enum print_line_t
1779 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1781 struct trace_seq *s = &iter->seq;
1782 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1783 struct trace_entry *next_entry;
1784 unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1785 struct trace_entry *entry = iter->ent;
1786 unsigned long abs_usecs;
1787 unsigned long rel_usecs;
1794 if (entry->type == TRACE_CONT)
1795 return TRACE_TYPE_HANDLED;
1797 test_cpu_buff_start(iter);
1799 next_entry = find_next_entry(iter, NULL, &next_ts);
1802 rel_usecs = ns2usecs(next_ts - iter->ts);
1803 abs_usecs = ns2usecs(iter->ts - iter->tr->time_start);
1806 comm = trace_find_cmdline(entry->pid);
1807 trace_seq_printf(s, "%16s %5d %3d %d %08x %08x [%08lx]"
1808 " %ld.%03ldms (+%ld.%03ldms): ",
1810 entry->pid, cpu, entry->flags,
1811 entry->preempt_count, trace_idx,
1814 abs_usecs % 1000, rel_usecs/1000,
1817 lat_print_generic(s, entry, cpu);
1818 lat_print_timestamp(s, abs_usecs, rel_usecs);
1820 switch (entry->type) {
1822 struct ftrace_entry *field;
1824 trace_assign_type(field, entry);
1826 seq_print_ip_sym(s, field->ip, sym_flags);
1827 trace_seq_puts(s, " (");
1828 seq_print_ip_sym(s, field->parent_ip, sym_flags);
1829 trace_seq_puts(s, ")\n");
1834 struct ctx_switch_entry *field;
1836 trace_assign_type(field, entry);
1838 T = field->next_state < sizeof(state_to_char) ?
1839 state_to_char[field->next_state] : 'X';
1841 state = field->prev_state ?
1842 __ffs(field->prev_state) + 1 : 0;
1843 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1844 comm = trace_find_cmdline(field->next_pid);
1845 trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c %s\n",
1848 S, entry->type == TRACE_CTX ? "==>" : " +",
1855 case TRACE_SPECIAL: {
1856 struct special_entry *field;
1858 trace_assign_type(field, entry);
1860 trace_seq_printf(s, "# %ld %ld %ld\n",
1867 struct stack_entry *field;
1869 trace_assign_type(field, entry);
1871 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1873 trace_seq_puts(s, " <= ");
1874 seq_print_ip_sym(s, field->caller[i], sym_flags);
1876 trace_seq_puts(s, "\n");
1880 struct print_entry *field;
1882 trace_assign_type(field, entry);
1884 seq_print_ip_sym(s, field->ip, sym_flags);
1885 trace_seq_printf(s, ": %s", field->buf);
1886 if (entry->flags & TRACE_FLAG_CONT)
1887 trace_seq_print_cont(s, iter);
1890 case TRACE_BRANCH: {
1891 struct trace_branch *field;
1893 trace_assign_type(field, entry);
1895 trace_seq_printf(s, "[%s] %s:%s:%d\n",
1896 field->correct ? " ok " : " MISS ",
1902 case TRACE_USER_STACK: {
1903 struct userstack_entry *field;
1905 trace_assign_type(field, entry);
1907 seq_print_userip_objs(field, s, sym_flags);
1908 trace_seq_putc(s, '\n');
1912 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1914 return TRACE_TYPE_HANDLED;
1917 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1919 struct trace_seq *s = &iter->seq;
1920 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1921 struct trace_entry *entry;
1922 unsigned long usec_rem;
1923 unsigned long long t;
1932 if (entry->type == TRACE_CONT)
1933 return TRACE_TYPE_HANDLED;
1935 test_cpu_buff_start(iter);
1937 comm = trace_find_cmdline(iter->ent->pid);
1939 t = ns2usecs(iter->ts);
1940 usec_rem = do_div(t, 1000000ULL);
1941 secs = (unsigned long)t;
1943 ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1945 return TRACE_TYPE_PARTIAL_LINE;
1946 ret = trace_seq_printf(s, "[%03d] ", iter->cpu);
1948 return TRACE_TYPE_PARTIAL_LINE;
1949 ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1951 return TRACE_TYPE_PARTIAL_LINE;
1953 switch (entry->type) {
1955 struct ftrace_entry *field;
1957 trace_assign_type(field, entry);
1959 ret = seq_print_ip_sym(s, field->ip, sym_flags);
1961 return TRACE_TYPE_PARTIAL_LINE;
1962 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1964 ret = trace_seq_printf(s, " <-");
1966 return TRACE_TYPE_PARTIAL_LINE;
1967 ret = seq_print_ip_sym(s,
1971 return TRACE_TYPE_PARTIAL_LINE;
1973 ret = trace_seq_printf(s, "\n");
1975 return TRACE_TYPE_PARTIAL_LINE;
1980 struct ctx_switch_entry *field;
1982 trace_assign_type(field, entry);
1984 S = field->prev_state < sizeof(state_to_char) ?
1985 state_to_char[field->prev_state] : 'X';
1986 T = field->next_state < sizeof(state_to_char) ?
1987 state_to_char[field->next_state] : 'X';
1988 ret = trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c\n",
1992 entry->type == TRACE_CTX ? "==>" : " +",
1998 return TRACE_TYPE_PARTIAL_LINE;
2001 case TRACE_SPECIAL: {
2002 struct special_entry *field;
2004 trace_assign_type(field, entry);
2006 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
2011 return TRACE_TYPE_PARTIAL_LINE;
2015 struct stack_entry *field;
2017 trace_assign_type(field, entry);
2019 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
2021 ret = trace_seq_puts(s, " <= ");
2023 return TRACE_TYPE_PARTIAL_LINE;
2025 ret = seq_print_ip_sym(s, field->caller[i],
2028 return TRACE_TYPE_PARTIAL_LINE;
2030 ret = trace_seq_puts(s, "\n");
2032 return TRACE_TYPE_PARTIAL_LINE;
2036 struct print_entry *field;
2038 trace_assign_type(field, entry);
2040 seq_print_ip_sym(s, field->ip, sym_flags);
2041 trace_seq_printf(s, ": %s", field->buf);
2042 if (entry->flags & TRACE_FLAG_CONT)
2043 trace_seq_print_cont(s, iter);
2046 case TRACE_GRAPH_RET: {
2047 return print_graph_function(iter);
2049 case TRACE_GRAPH_ENT: {
2050 return print_graph_function(iter);
2052 case TRACE_BRANCH: {
2053 struct trace_branch *field;
2055 trace_assign_type(field, entry);
2057 trace_seq_printf(s, "[%s] %s:%s:%d\n",
2058 field->correct ? " ok " : " MISS ",
2064 case TRACE_USER_STACK: {
2065 struct userstack_entry *field;
2067 trace_assign_type(field, entry);
2069 ret = seq_print_userip_objs(field, s, sym_flags);
2071 return TRACE_TYPE_PARTIAL_LINE;
2072 ret = trace_seq_putc(s, '\n');
2074 return TRACE_TYPE_PARTIAL_LINE;
2078 return TRACE_TYPE_HANDLED;
2081 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2083 struct trace_seq *s = &iter->seq;
2084 struct trace_entry *entry;
2090 if (entry->type == TRACE_CONT)
2091 return TRACE_TYPE_HANDLED;
2093 ret = trace_seq_printf(s, "%d %d %llu ",
2094 entry->pid, iter->cpu, iter->ts);
2096 return TRACE_TYPE_PARTIAL_LINE;
2098 switch (entry->type) {
2100 struct ftrace_entry *field;
2102 trace_assign_type(field, entry);
2104 ret = trace_seq_printf(s, "%x %x\n",
2108 return TRACE_TYPE_PARTIAL_LINE;
2113 struct ctx_switch_entry *field;
2115 trace_assign_type(field, entry);
2117 S = field->prev_state < sizeof(state_to_char) ?
2118 state_to_char[field->prev_state] : 'X';
2119 T = field->next_state < sizeof(state_to_char) ?
2120 state_to_char[field->next_state] : 'X';
2121 if (entry->type == TRACE_WAKE)
2123 ret = trace_seq_printf(s, "%d %d %c %d %d %d %c\n",
2132 return TRACE_TYPE_PARTIAL_LINE;
2136 case TRACE_USER_STACK:
2138 struct special_entry *field;
2140 trace_assign_type(field, entry);
2142 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
2147 return TRACE_TYPE_PARTIAL_LINE;
2151 struct print_entry *field;
2153 trace_assign_type(field, entry);
2155 trace_seq_printf(s, "# %lx %s", field->ip, field->buf);
2156 if (entry->flags & TRACE_FLAG_CONT)
2157 trace_seq_print_cont(s, iter);
2161 return TRACE_TYPE_HANDLED;
2164 #define SEQ_PUT_FIELD_RET(s, x) \
2166 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
2170 #define SEQ_PUT_HEX_FIELD_RET(s, x) \
2172 BUILD_BUG_ON(sizeof(x) > MAX_MEMHEX_BYTES); \
2173 if (!trace_seq_putmem_hex(s, &(x), sizeof(x))) \
2177 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2179 struct trace_seq *s = &iter->seq;
2180 unsigned char newline = '\n';
2181 struct trace_entry *entry;
2186 if (entry->type == TRACE_CONT)
2187 return TRACE_TYPE_HANDLED;
2189 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2190 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2191 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2193 switch (entry->type) {
2195 struct ftrace_entry *field;
2197 trace_assign_type(field, entry);
2199 SEQ_PUT_HEX_FIELD_RET(s, field->ip);
2200 SEQ_PUT_HEX_FIELD_RET(s, field->parent_ip);
2205 struct ctx_switch_entry *field;
2207 trace_assign_type(field, entry);
2209 S = field->prev_state < sizeof(state_to_char) ?
2210 state_to_char[field->prev_state] : 'X';
2211 T = field->next_state < sizeof(state_to_char) ?
2212 state_to_char[field->next_state] : 'X';
2213 if (entry->type == TRACE_WAKE)
2215 SEQ_PUT_HEX_FIELD_RET(s, field->prev_pid);
2216 SEQ_PUT_HEX_FIELD_RET(s, field->prev_prio);
2217 SEQ_PUT_HEX_FIELD_RET(s, S);
2218 SEQ_PUT_HEX_FIELD_RET(s, field->next_cpu);
2219 SEQ_PUT_HEX_FIELD_RET(s, field->next_pid);
2220 SEQ_PUT_HEX_FIELD_RET(s, field->next_prio);
2221 SEQ_PUT_HEX_FIELD_RET(s, T);
2225 case TRACE_USER_STACK:
2227 struct special_entry *field;
2229 trace_assign_type(field, entry);
2231 SEQ_PUT_HEX_FIELD_RET(s, field->arg1);
2232 SEQ_PUT_HEX_FIELD_RET(s, field->arg2);
2233 SEQ_PUT_HEX_FIELD_RET(s, field->arg3);
2237 SEQ_PUT_FIELD_RET(s, newline);
2239 return TRACE_TYPE_HANDLED;
2242 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2244 struct trace_seq *s = &iter->seq;
2245 struct trace_entry *entry;
2249 if (entry->type == TRACE_CONT)
2250 return TRACE_TYPE_HANDLED;
2252 SEQ_PUT_FIELD_RET(s, entry->pid);
2253 SEQ_PUT_FIELD_RET(s, entry->cpu);
2254 SEQ_PUT_FIELD_RET(s, iter->ts);
2256 switch (entry->type) {
2258 struct ftrace_entry *field;
2260 trace_assign_type(field, entry);
2262 SEQ_PUT_FIELD_RET(s, field->ip);
2263 SEQ_PUT_FIELD_RET(s, field->parent_ip);
2267 struct ctx_switch_entry *field;
2269 trace_assign_type(field, entry);
2271 SEQ_PUT_FIELD_RET(s, field->prev_pid);
2272 SEQ_PUT_FIELD_RET(s, field->prev_prio);
2273 SEQ_PUT_FIELD_RET(s, field->prev_state);
2274 SEQ_PUT_FIELD_RET(s, field->next_pid);
2275 SEQ_PUT_FIELD_RET(s, field->next_prio);
2276 SEQ_PUT_FIELD_RET(s, field->next_state);
2280 case TRACE_USER_STACK:
2282 struct special_entry *field;
2284 trace_assign_type(field, entry);
2286 SEQ_PUT_FIELD_RET(s, field->arg1);
2287 SEQ_PUT_FIELD_RET(s, field->arg2);
2288 SEQ_PUT_FIELD_RET(s, field->arg3);
2295 static int trace_empty(struct trace_iterator *iter)
2299 for_each_tracing_cpu(cpu) {
2300 if (iter->buffer_iter[cpu]) {
2301 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2304 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2312 static enum print_line_t print_trace_line(struct trace_iterator *iter)
2314 enum print_line_t ret;
2316 if (iter->trace && iter->trace->print_line) {
2317 ret = iter->trace->print_line(iter);
2318 if (ret != TRACE_TYPE_UNHANDLED)
2322 if (trace_flags & TRACE_ITER_BIN)
2323 return print_bin_fmt(iter);
2325 if (trace_flags & TRACE_ITER_HEX)
2326 return print_hex_fmt(iter);
2328 if (trace_flags & TRACE_ITER_RAW)
2329 return print_raw_fmt(iter);
2331 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2332 return print_lat_fmt(iter, iter->idx, iter->cpu);
2334 return print_trace_fmt(iter);
2337 static int s_show(struct seq_file *m, void *v)
2339 struct trace_iterator *iter = v;
2341 if (iter->ent == NULL) {
2343 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2346 if (iter->trace && iter->trace->print_header)
2347 iter->trace->print_header(m);
2348 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2349 /* print nothing if the buffers are empty */
2350 if (trace_empty(iter))
2352 print_trace_header(m, iter);
2353 if (!(trace_flags & TRACE_ITER_VERBOSE))
2354 print_lat_help_header(m);
2356 if (!(trace_flags & TRACE_ITER_VERBOSE))
2357 print_func_help_header(m);
2360 print_trace_line(iter);
2361 trace_print_seq(m, &iter->seq);
2367 static struct seq_operations tracer_seq_ops = {
2374 static struct trace_iterator *
2375 __tracing_open(struct inode *inode, struct file *file, int *ret)
2377 struct trace_iterator *iter;
2381 if (tracing_disabled) {
2386 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2392 mutex_lock(&trace_types_lock);
2393 if (current_trace && current_trace->print_max)
2396 iter->tr = inode->i_private;
2397 iter->trace = current_trace;
2400 /* Notify the tracer early; before we stop tracing. */
2401 if (iter->trace && iter->trace->open)
2402 iter->trace->open(iter);
2404 /* Annotate start of buffers if we had overruns */
2405 if (ring_buffer_overruns(iter->tr->buffer))
2406 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2409 for_each_tracing_cpu(cpu) {
2411 iter->buffer_iter[cpu] =
2412 ring_buffer_read_start(iter->tr->buffer, cpu);
2414 if (!iter->buffer_iter[cpu])
2418 /* TODO stop tracer */
2419 *ret = seq_open(file, &tracer_seq_ops);
2423 m = file->private_data;
2426 /* stop the trace while dumping */
2429 mutex_unlock(&trace_types_lock);
2435 for_each_tracing_cpu(cpu) {
2436 if (iter->buffer_iter[cpu])
2437 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2439 mutex_unlock(&trace_types_lock);
2442 return ERR_PTR(-ENOMEM);
2445 int tracing_open_generic(struct inode *inode, struct file *filp)
2447 if (tracing_disabled)
2450 filp->private_data = inode->i_private;
2454 int tracing_release(struct inode *inode, struct file *file)
2456 struct seq_file *m = (struct seq_file *)file->private_data;
2457 struct trace_iterator *iter = m->private;
2460 mutex_lock(&trace_types_lock);
2461 for_each_tracing_cpu(cpu) {
2462 if (iter->buffer_iter[cpu])
2463 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2466 if (iter->trace && iter->trace->close)
2467 iter->trace->close(iter);
2469 /* reenable tracing if it was previously enabled */
2471 mutex_unlock(&trace_types_lock);
2473 seq_release(inode, file);
2478 static int tracing_open(struct inode *inode, struct file *file)
2482 __tracing_open(inode, file, &ret);
2487 static int tracing_lt_open(struct inode *inode, struct file *file)
2489 struct trace_iterator *iter;
2492 iter = __tracing_open(inode, file, &ret);
2495 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2502 t_next(struct seq_file *m, void *v, loff_t *pos)
2504 struct tracer *t = m->private;
2516 static void *t_start(struct seq_file *m, loff_t *pos)
2518 struct tracer *t = m->private;
2521 mutex_lock(&trace_types_lock);
2522 for (; t && l < *pos; t = t_next(m, t, &l))
2528 static void t_stop(struct seq_file *m, void *p)
2530 mutex_unlock(&trace_types_lock);
2533 static int t_show(struct seq_file *m, void *v)
2535 struct tracer *t = v;
2540 seq_printf(m, "%s", t->name);
2549 static struct seq_operations show_traces_seq_ops = {
2556 static int show_traces_open(struct inode *inode, struct file *file)
2560 if (tracing_disabled)
2563 ret = seq_open(file, &show_traces_seq_ops);
2565 struct seq_file *m = file->private_data;
2566 m->private = trace_types;
2572 static struct file_operations tracing_fops = {
2573 .open = tracing_open,
2575 .llseek = seq_lseek,
2576 .release = tracing_release,
2579 static struct file_operations tracing_lt_fops = {
2580 .open = tracing_lt_open,
2582 .llseek = seq_lseek,
2583 .release = tracing_release,
2586 static struct file_operations show_traces_fops = {
2587 .open = show_traces_open,
2589 .release = seq_release,
2593 * Only trace on a CPU if the bitmask is set:
2595 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
2598 * When tracing/tracing_cpu_mask is modified then this holds
2599 * the new bitmask we are about to install:
2601 static cpumask_t tracing_cpumask_new;
2604 * The tracer itself will not take this lock, but still we want
2605 * to provide a consistent cpumask to user-space:
2607 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2610 * Temporary storage for the character representation of the
2611 * CPU bitmask (and one more byte for the newline):
2613 static char mask_str[NR_CPUS + 1];
2616 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2617 size_t count, loff_t *ppos)
2621 mutex_lock(&tracing_cpumask_update_lock);
2623 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2624 if (count - len < 2) {
2628 len += sprintf(mask_str + len, "\n");
2629 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2632 mutex_unlock(&tracing_cpumask_update_lock);
2638 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2639 size_t count, loff_t *ppos)
2643 mutex_lock(&tracing_cpumask_update_lock);
2644 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2648 raw_local_irq_disable();
2649 __raw_spin_lock(&ftrace_max_lock);
2650 for_each_tracing_cpu(cpu) {
2652 * Increase/decrease the disabled counter if we are
2653 * about to flip a bit in the cpumask:
2655 if (cpu_isset(cpu, tracing_cpumask) &&
2656 !cpu_isset(cpu, tracing_cpumask_new)) {
2657 atomic_inc(&global_trace.data[cpu]->disabled);
2659 if (!cpu_isset(cpu, tracing_cpumask) &&
2660 cpu_isset(cpu, tracing_cpumask_new)) {
2661 atomic_dec(&global_trace.data[cpu]->disabled);
2664 __raw_spin_unlock(&ftrace_max_lock);
2665 raw_local_irq_enable();
2667 tracing_cpumask = tracing_cpumask_new;
2669 mutex_unlock(&tracing_cpumask_update_lock);
2674 mutex_unlock(&tracing_cpumask_update_lock);
2679 static struct file_operations tracing_cpumask_fops = {
2680 .open = tracing_open_generic,
2681 .read = tracing_cpumask_read,
2682 .write = tracing_cpumask_write,
2686 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2687 size_t cnt, loff_t *ppos)
2693 u32 tracer_flags = current_trace->flags->val;
2694 struct tracer_opt *trace_opts = current_trace->flags->opts;
2697 /* calulate max size */
2698 for (i = 0; trace_options[i]; i++) {
2699 len += strlen(trace_options[i]);
2700 len += 3; /* "no" and space */
2704 * Increase the size with names of options specific
2705 * of the current tracer.
2707 for (i = 0; trace_opts[i].name; i++) {
2708 len += strlen(trace_opts[i].name);
2709 len += 3; /* "no" and space */
2712 /* +2 for \n and \0 */
2713 buf = kmalloc(len + 2, GFP_KERNEL);
2717 for (i = 0; trace_options[i]; i++) {
2718 if (trace_flags & (1 << i))
2719 r += sprintf(buf + r, "%s ", trace_options[i]);
2721 r += sprintf(buf + r, "no%s ", trace_options[i]);
2724 for (i = 0; trace_opts[i].name; i++) {
2725 if (tracer_flags & trace_opts[i].bit)
2726 r += sprintf(buf + r, "%s ",
2727 trace_opts[i].name);
2729 r += sprintf(buf + r, "no%s ",
2730 trace_opts[i].name);
2733 r += sprintf(buf + r, "\n");
2734 WARN_ON(r >= len + 2);
2736 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2743 /* Try to assign a tracer specific option */
2744 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2746 struct tracer_flags *trace_flags = trace->flags;
2747 struct tracer_opt *opts = NULL;
2751 for (i = 0; trace_flags->opts[i].name; i++) {
2752 opts = &trace_flags->opts[i];
2753 len = strlen(opts->name);
2755 if (strncmp(cmp, opts->name, len) == 0) {
2756 ret = trace->set_flag(trace_flags->val,
2762 if (!trace_flags->opts[i].name)
2765 /* Refused to handle */
2770 trace_flags->val &= ~opts->bit;
2772 trace_flags->val |= opts->bit;
2778 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2779 size_t cnt, loff_t *ppos)
2787 if (cnt >= sizeof(buf))
2790 if (copy_from_user(&buf, ubuf, cnt))
2795 if (strncmp(buf, "no", 2) == 0) {
2800 for (i = 0; trace_options[i]; i++) {
2801 int len = strlen(trace_options[i]);
2803 if (strncmp(cmp, trace_options[i], len) == 0) {
2805 trace_flags &= ~(1 << i);
2807 trace_flags |= (1 << i);
2812 /* If no option could be set, test the specific tracer options */
2813 if (!trace_options[i]) {
2814 ret = set_tracer_option(current_trace, cmp, neg);
2824 static struct file_operations tracing_iter_fops = {
2825 .open = tracing_open_generic,
2826 .read = tracing_trace_options_read,
2827 .write = tracing_trace_options_write,
2830 static const char readme_msg[] =
2831 "tracing mini-HOWTO:\n\n"
2833 "# mount -t debugfs nodev /debug\n\n"
2834 "# cat /debug/tracing/available_tracers\n"
2835 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2836 "# cat /debug/tracing/current_tracer\n"
2838 "# echo sched_switch > /debug/tracing/current_tracer\n"
2839 "# cat /debug/tracing/current_tracer\n"
2841 "# cat /debug/tracing/trace_options\n"
2842 "noprint-parent nosym-offset nosym-addr noverbose\n"
2843 "# echo print-parent > /debug/tracing/trace_options\n"
2844 "# echo 1 > /debug/tracing/tracing_enabled\n"
2845 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2846 "echo 0 > /debug/tracing/tracing_enabled\n"
2850 tracing_readme_read(struct file *filp, char __user *ubuf,
2851 size_t cnt, loff_t *ppos)
2853 return simple_read_from_buffer(ubuf, cnt, ppos,
2854 readme_msg, strlen(readme_msg));
2857 static struct file_operations tracing_readme_fops = {
2858 .open = tracing_open_generic,
2859 .read = tracing_readme_read,
2863 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2864 size_t cnt, loff_t *ppos)
2869 r = sprintf(buf, "%u\n", tracer_enabled);
2870 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2874 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2875 size_t cnt, loff_t *ppos)
2877 struct trace_array *tr = filp->private_data;
2882 if (cnt >= sizeof(buf))
2885 if (copy_from_user(&buf, ubuf, cnt))
2890 ret = strict_strtoul(buf, 10, &val);
2896 mutex_lock(&trace_types_lock);
2897 if (tracer_enabled ^ val) {
2900 if (current_trace->start)
2901 current_trace->start(tr);
2906 if (current_trace->stop)
2907 current_trace->stop(tr);
2910 mutex_unlock(&trace_types_lock);
2918 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2919 size_t cnt, loff_t *ppos)
2921 char buf[max_tracer_type_len+2];
2924 mutex_lock(&trace_types_lock);
2926 r = sprintf(buf, "%s\n", current_trace->name);
2928 r = sprintf(buf, "\n");
2929 mutex_unlock(&trace_types_lock);
2931 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2934 static int tracing_set_tracer(char *buf)
2936 struct trace_array *tr = &global_trace;
2940 mutex_lock(&trace_types_lock);
2941 for (t = trace_types; t; t = t->next) {
2942 if (strcmp(t->name, buf) == 0)
2949 if (t == current_trace)
2952 trace_branch_disable();
2953 if (current_trace && current_trace->reset)
2954 current_trace->reset(tr);
2963 trace_branch_enable(tr);
2965 mutex_unlock(&trace_types_lock);
2971 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2972 size_t cnt, loff_t *ppos)
2974 char buf[max_tracer_type_len+1];
2981 if (cnt > max_tracer_type_len)
2982 cnt = max_tracer_type_len;
2984 if (copy_from_user(&buf, ubuf, cnt))
2989 /* strip ending whitespace. */
2990 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2993 err = tracing_set_tracer(buf);
3003 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3004 size_t cnt, loff_t *ppos)
3006 unsigned long *ptr = filp->private_data;
3010 r = snprintf(buf, sizeof(buf), "%ld\n",
3011 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3012 if (r > sizeof(buf))
3014 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3018 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3019 size_t cnt, loff_t *ppos)
3021 long *ptr = filp->private_data;
3026 if (cnt >= sizeof(buf))
3029 if (copy_from_user(&buf, ubuf, cnt))
3034 ret = strict_strtoul(buf, 10, &val);
3043 static atomic_t tracing_reader;
3045 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3047 struct trace_iterator *iter;
3049 if (tracing_disabled)
3052 /* We only allow for reader of the pipe */
3053 if (atomic_inc_return(&tracing_reader) != 1) {
3054 atomic_dec(&tracing_reader);
3058 /* create a buffer to store the information to pass to userspace */
3059 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3063 mutex_lock(&trace_types_lock);
3065 /* trace pipe does not show start of buffer */
3066 cpus_setall(iter->started);
3068 iter->tr = &global_trace;
3069 iter->trace = current_trace;
3070 filp->private_data = iter;
3072 if (iter->trace->pipe_open)
3073 iter->trace->pipe_open(iter);
3074 mutex_unlock(&trace_types_lock);
3079 static int tracing_release_pipe(struct inode *inode, struct file *file)
3081 struct trace_iterator *iter = file->private_data;
3084 atomic_dec(&tracing_reader);
3090 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3092 struct trace_iterator *iter = filp->private_data;
3094 if (trace_flags & TRACE_ITER_BLOCK) {
3096 * Always select as readable when in blocking mode
3098 return POLLIN | POLLRDNORM;
3100 if (!trace_empty(iter))
3101 return POLLIN | POLLRDNORM;
3102 poll_wait(filp, &trace_wait, poll_table);
3103 if (!trace_empty(iter))
3104 return POLLIN | POLLRDNORM;
3114 tracing_read_pipe(struct file *filp, char __user *ubuf,
3115 size_t cnt, loff_t *ppos)
3117 struct trace_iterator *iter = filp->private_data;
3120 /* return any leftover data */
3121 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3125 trace_seq_reset(&iter->seq);
3127 mutex_lock(&trace_types_lock);
3128 if (iter->trace->read) {
3129 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3136 while (trace_empty(iter)) {
3138 if ((filp->f_flags & O_NONBLOCK)) {
3144 * This is a make-shift waitqueue. The reason we don't use
3145 * an actual wait queue is because:
3146 * 1) we only ever have one waiter
3147 * 2) the tracing, traces all functions, we don't want
3148 * the overhead of calling wake_up and friends
3149 * (and tracing them too)
3150 * Anyway, this is really very primitive wakeup.
3152 set_current_state(TASK_INTERRUPTIBLE);
3153 iter->tr->waiter = current;
3155 mutex_unlock(&trace_types_lock);
3157 /* sleep for 100 msecs, and try again. */
3158 schedule_timeout(HZ/10);
3160 mutex_lock(&trace_types_lock);
3162 iter->tr->waiter = NULL;
3164 if (signal_pending(current)) {
3169 if (iter->trace != current_trace)
3173 * We block until we read something and tracing is disabled.
3174 * We still block if tracing is disabled, but we have never
3175 * read anything. This allows a user to cat this file, and
3176 * then enable tracing. But after we have read something,
3177 * we give an EOF when tracing is again disabled.
3179 * iter->pos will be 0 if we haven't read anything.
3181 if (!tracer_enabled && iter->pos)
3187 /* stop when tracing is finished */
3188 if (trace_empty(iter))
3191 if (cnt >= PAGE_SIZE)
3192 cnt = PAGE_SIZE - 1;
3194 /* reset all but tr, trace, and overruns */
3195 memset(&iter->seq, 0,
3196 sizeof(struct trace_iterator) -
3197 offsetof(struct trace_iterator, seq));
3200 while (find_next_entry_inc(iter) != NULL) {
3201 enum print_line_t ret;
3202 int len = iter->seq.len;
3204 ret = print_trace_line(iter);
3205 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3206 /* don't print partial lines */
3207 iter->seq.len = len;
3211 trace_consume(iter);
3213 if (iter->seq.len >= cnt)
3217 /* Now copy what we have to the user */
3218 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3219 if (iter->seq.readpos >= iter->seq.len)
3220 trace_seq_reset(&iter->seq);
3223 * If there was nothing to send to user, inspite of consuming trace
3224 * entries, go back to wait for more entries.
3230 mutex_unlock(&trace_types_lock);
3236 tracing_entries_read(struct file *filp, char __user *ubuf,
3237 size_t cnt, loff_t *ppos)
3239 struct trace_array *tr = filp->private_data;
3243 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3244 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3248 tracing_entries_write(struct file *filp, const char __user *ubuf,
3249 size_t cnt, loff_t *ppos)
3255 if (cnt >= sizeof(buf))
3258 if (copy_from_user(&buf, ubuf, cnt))
3263 ret = strict_strtoul(buf, 10, &val);
3267 /* must have at least 1 entry */
3271 mutex_lock(&trace_types_lock);
3275 /* disable all cpu buffers */
3276 for_each_tracing_cpu(cpu) {
3277 if (global_trace.data[cpu])
3278 atomic_inc(&global_trace.data[cpu]->disabled);
3279 if (max_tr.data[cpu])
3280 atomic_inc(&max_tr.data[cpu]->disabled);
3283 /* value is in KB */
3286 if (val != global_trace.entries) {
3287 ret = ring_buffer_resize(global_trace.buffer, val);
3293 ret = ring_buffer_resize(max_tr.buffer, val);
3297 r = ring_buffer_resize(global_trace.buffer,
3298 global_trace.entries);
3300 /* AARGH! We are left with different
3301 * size max buffer!!!! */
3303 tracing_disabled = 1;
3308 global_trace.entries = val;
3313 /* If check pages failed, return ENOMEM */
3314 if (tracing_disabled)
3317 for_each_tracing_cpu(cpu) {
3318 if (global_trace.data[cpu])
3319 atomic_dec(&global_trace.data[cpu]->disabled);
3320 if (max_tr.data[cpu])
3321 atomic_dec(&max_tr.data[cpu]->disabled);
3325 max_tr.entries = global_trace.entries;
3326 mutex_unlock(&trace_types_lock);
3331 static int mark_printk(const char *fmt, ...)
3335 va_start(args, fmt);
3336 ret = trace_vprintk(0, fmt, args);
3342 tracing_mark_write(struct file *filp, const char __user *ubuf,
3343 size_t cnt, loff_t *fpos)
3348 if (tracing_disabled)
3351 if (cnt > TRACE_BUF_SIZE)
3352 cnt = TRACE_BUF_SIZE;
3354 buf = kmalloc(cnt + 1, GFP_KERNEL);
3358 if (copy_from_user(buf, ubuf, cnt)) {
3363 /* Cut from the first nil or newline. */
3365 end = strchr(buf, '\n');
3369 cnt = mark_printk("%s\n", buf);
3376 static struct file_operations tracing_max_lat_fops = {
3377 .open = tracing_open_generic,
3378 .read = tracing_max_lat_read,
3379 .write = tracing_max_lat_write,
3382 static struct file_operations tracing_ctrl_fops = {
3383 .open = tracing_open_generic,
3384 .read = tracing_ctrl_read,
3385 .write = tracing_ctrl_write,
3388 static struct file_operations set_tracer_fops = {
3389 .open = tracing_open_generic,
3390 .read = tracing_set_trace_read,
3391 .write = tracing_set_trace_write,
3394 static struct file_operations tracing_pipe_fops = {
3395 .open = tracing_open_pipe,
3396 .poll = tracing_poll_pipe,
3397 .read = tracing_read_pipe,
3398 .release = tracing_release_pipe,
3401 static struct file_operations tracing_entries_fops = {
3402 .open = tracing_open_generic,
3403 .read = tracing_entries_read,
3404 .write = tracing_entries_write,
3407 static struct file_operations tracing_mark_fops = {
3408 .open = tracing_open_generic,
3409 .write = tracing_mark_write,
3412 #ifdef CONFIG_DYNAMIC_FTRACE
3414 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3420 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3421 size_t cnt, loff_t *ppos)
3423 static char ftrace_dyn_info_buffer[1024];
3424 static DEFINE_MUTEX(dyn_info_mutex);
3425 unsigned long *p = filp->private_data;
3426 char *buf = ftrace_dyn_info_buffer;
3427 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3430 mutex_lock(&dyn_info_mutex);
3431 r = sprintf(buf, "%ld ", *p);
3433 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3436 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3438 mutex_unlock(&dyn_info_mutex);
3443 static struct file_operations tracing_dyn_info_fops = {
3444 .open = tracing_open_generic,
3445 .read = tracing_read_dyn_info,
3449 static struct dentry *d_tracer;
3451 struct dentry *tracing_init_dentry(void)
3458 d_tracer = debugfs_create_dir("tracing", NULL);
3460 if (!d_tracer && !once) {
3462 pr_warning("Could not create debugfs directory 'tracing'\n");
3469 #ifdef CONFIG_FTRACE_SELFTEST
3470 /* Let selftest have access to static functions in this file */
3471 #include "trace_selftest.c"
3474 static __init int tracer_init_debugfs(void)
3476 struct dentry *d_tracer;
3477 struct dentry *entry;
3479 d_tracer = tracing_init_dentry();
3481 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3482 &global_trace, &tracing_ctrl_fops);
3484 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3486 entry = debugfs_create_file("trace_options", 0644, d_tracer,
3487 NULL, &tracing_iter_fops);
3489 pr_warning("Could not create debugfs 'trace_options' entry\n");
3491 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3492 NULL, &tracing_cpumask_fops);
3494 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3496 entry = debugfs_create_file("latency_trace", 0444, d_tracer,
3497 &global_trace, &tracing_lt_fops);
3499 pr_warning("Could not create debugfs 'latency_trace' entry\n");
3501 entry = debugfs_create_file("trace", 0444, d_tracer,
3502 &global_trace, &tracing_fops);
3504 pr_warning("Could not create debugfs 'trace' entry\n");
3506 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3507 &global_trace, &show_traces_fops);
3509 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3511 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3512 &global_trace, &set_tracer_fops);
3514 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3516 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3517 &tracing_max_latency,
3518 &tracing_max_lat_fops);
3520 pr_warning("Could not create debugfs "
3521 "'tracing_max_latency' entry\n");
3523 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3524 &tracing_thresh, &tracing_max_lat_fops);
3526 pr_warning("Could not create debugfs "
3527 "'tracing_thresh' entry\n");
3528 entry = debugfs_create_file("README", 0644, d_tracer,
3529 NULL, &tracing_readme_fops);
3531 pr_warning("Could not create debugfs 'README' entry\n");
3533 entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
3534 NULL, &tracing_pipe_fops);
3536 pr_warning("Could not create debugfs "
3537 "'trace_pipe' entry\n");
3539 entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
3540 &global_trace, &tracing_entries_fops);
3542 pr_warning("Could not create debugfs "
3543 "'buffer_size_kb' entry\n");
3545 entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3546 NULL, &tracing_mark_fops);
3548 pr_warning("Could not create debugfs "
3549 "'trace_marker' entry\n");
3551 #ifdef CONFIG_DYNAMIC_FTRACE
3552 entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3553 &ftrace_update_tot_cnt,
3554 &tracing_dyn_info_fops);
3556 pr_warning("Could not create debugfs "
3557 "'dyn_ftrace_total_info' entry\n");
3559 #ifdef CONFIG_SYSPROF_TRACER
3560 init_tracer_sysprof_debugfs(d_tracer);
3565 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3567 static DEFINE_SPINLOCK(trace_buf_lock);
3568 static char trace_buf[TRACE_BUF_SIZE];
3570 struct ring_buffer_event *event;
3571 struct trace_array *tr = &global_trace;
3572 struct trace_array_cpu *data;
3573 struct print_entry *entry;
3574 unsigned long flags, irq_flags;
3575 int cpu, len = 0, size, pc;
3577 if (tracing_disabled)
3580 pc = preempt_count();
3581 preempt_disable_notrace();
3582 cpu = raw_smp_processor_id();
3583 data = tr->data[cpu];
3585 if (unlikely(atomic_read(&data->disabled)))
3588 spin_lock_irqsave(&trace_buf_lock, flags);
3589 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
3591 len = min(len, TRACE_BUF_SIZE-1);
3594 size = sizeof(*entry) + len + 1;
3595 event = ring_buffer_lock_reserve(tr->buffer, size, &irq_flags);
3598 entry = ring_buffer_event_data(event);
3599 tracing_generic_entry_update(&entry->ent, flags, pc);
3600 entry->ent.type = TRACE_PRINT;
3603 memcpy(&entry->buf, trace_buf, len);
3604 entry->buf[len] = 0;
3605 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
3608 spin_unlock_irqrestore(&trace_buf_lock, flags);
3611 preempt_enable_notrace();
3615 EXPORT_SYMBOL_GPL(trace_vprintk);
3617 int __ftrace_printk(unsigned long ip, const char *fmt, ...)
3622 if (!(trace_flags & TRACE_ITER_PRINTK))
3626 ret = trace_vprintk(ip, fmt, ap);
3630 EXPORT_SYMBOL_GPL(__ftrace_printk);
3632 static int trace_panic_handler(struct notifier_block *this,
3633 unsigned long event, void *unused)
3635 if (ftrace_dump_on_oops)
3640 static struct notifier_block trace_panic_notifier = {
3641 .notifier_call = trace_panic_handler,
3643 .priority = 150 /* priority: INT_MAX >= x >= 0 */
3646 static int trace_die_handler(struct notifier_block *self,
3652 if (ftrace_dump_on_oops)
3661 static struct notifier_block trace_die_notifier = {
3662 .notifier_call = trace_die_handler,
3667 * printk is set to max of 1024, we really don't need it that big.
3668 * Nothing should be printing 1000 characters anyway.
3670 #define TRACE_MAX_PRINT 1000
3673 * Define here KERN_TRACE so that we have one place to modify
3674 * it if we decide to change what log level the ftrace dump
3677 #define KERN_TRACE KERN_INFO
3680 trace_printk_seq(struct trace_seq *s)
3682 /* Probably should print a warning here. */
3686 /* should be zero ended, but we are paranoid. */
3687 s->buffer[s->len] = 0;
3689 printk(KERN_TRACE "%s", s->buffer);
3694 void ftrace_dump(void)
3696 static DEFINE_SPINLOCK(ftrace_dump_lock);
3697 /* use static because iter can be a bit big for the stack */
3698 static struct trace_iterator iter;
3699 static cpumask_t mask;
3700 static int dump_ran;
3701 unsigned long flags;
3705 spin_lock_irqsave(&ftrace_dump_lock, flags);
3711 /* No turning back! */
3714 for_each_tracing_cpu(cpu) {
3715 atomic_inc(&global_trace.data[cpu]->disabled);
3718 /* don't look at user memory in panic mode */
3719 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
3721 printk(KERN_TRACE "Dumping ftrace buffer:\n");
3723 iter.tr = &global_trace;
3724 iter.trace = current_trace;
3727 * We need to stop all tracing on all CPUS to read the
3728 * the next buffer. This is a bit expensive, but is
3729 * not done often. We fill all what we can read,
3730 * and then release the locks again.
3735 while (!trace_empty(&iter)) {
3738 printk(KERN_TRACE "---------------------------------\n");
3742 /* reset all but tr, trace, and overruns */
3743 memset(&iter.seq, 0,
3744 sizeof(struct trace_iterator) -
3745 offsetof(struct trace_iterator, seq));
3746 iter.iter_flags |= TRACE_FILE_LAT_FMT;
3749 if (find_next_entry_inc(&iter) != NULL) {
3750 print_trace_line(&iter);
3751 trace_consume(&iter);
3754 trace_printk_seq(&iter.seq);
3758 printk(KERN_TRACE " (ftrace buffer empty)\n");
3760 printk(KERN_TRACE "---------------------------------\n");
3763 spin_unlock_irqrestore(&ftrace_dump_lock, flags);
3766 __init static int tracer_alloc_buffers(void)
3768 struct trace_array_cpu *data;
3771 /* TODO: make the number of buffers hot pluggable with CPUS */
3772 tracing_buffer_mask = cpu_possible_map;
3774 global_trace.buffer = ring_buffer_alloc(trace_buf_size,
3775 TRACE_BUFFER_FLAGS);
3776 if (!global_trace.buffer) {
3777 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
3781 global_trace.entries = ring_buffer_size(global_trace.buffer);
3783 #ifdef CONFIG_TRACER_MAX_TRACE
3784 max_tr.buffer = ring_buffer_alloc(trace_buf_size,
3785 TRACE_BUFFER_FLAGS);
3786 if (!max_tr.buffer) {
3787 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
3789 ring_buffer_free(global_trace.buffer);
3792 max_tr.entries = ring_buffer_size(max_tr.buffer);
3793 WARN_ON(max_tr.entries != global_trace.entries);
3796 /* Allocate the first page for all buffers */
3797 for_each_tracing_cpu(i) {
3798 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
3799 max_tr.data[i] = &per_cpu(max_data, i);
3802 trace_init_cmdlines();
3804 register_tracer(&nop_trace);
3805 #ifdef CONFIG_BOOT_TRACER
3806 register_tracer(&boot_tracer);
3807 current_trace = &boot_tracer;
3808 current_trace->init(&global_trace);
3810 current_trace = &nop_trace;
3813 /* All seems OK, enable tracing */
3814 tracing_disabled = 0;
3816 atomic_notifier_chain_register(&panic_notifier_list,
3817 &trace_panic_notifier);
3819 register_die_notifier(&trace_die_notifier);
3823 early_initcall(tracer_alloc_buffers);
3824 fs_initcall(tracer_init_debugfs);