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;
48 * We need to change this state when a selftest is running.
49 * A selftest will lurk into the ring-buffer to count the
50 * entries inserted during the selftest although some concurrent
51 * insertions into the ring-buffer such as ftrace_printk could occurred
52 * at the same time, giving false positive or negative results.
54 static bool __read_mostly tracing_selftest_running;
56 /* For tracers that don't implement custom flags */
57 static struct tracer_opt dummy_tracer_opt[] = {
61 static struct tracer_flags dummy_tracer_flags = {
63 .opts = dummy_tracer_opt
66 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
72 * Kill all tracing for good (never come back).
73 * It is initialized to 1 but will turn to zero if the initialization
74 * of the tracer is successful. But that is the only place that sets
77 int tracing_disabled = 1;
79 static DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
81 static inline void ftrace_disable_cpu(void)
84 local_inc(&__get_cpu_var(ftrace_cpu_disabled));
87 static inline void ftrace_enable_cpu(void)
89 local_dec(&__get_cpu_var(ftrace_cpu_disabled));
93 static cpumask_t __read_mostly tracing_buffer_mask;
95 #define for_each_tracing_cpu(cpu) \
96 for_each_cpu_mask(cpu, tracing_buffer_mask)
99 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
101 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
102 * is set, then ftrace_dump is called. This will output the contents
103 * of the ftrace buffers to the console. This is very useful for
104 * capturing traces that lead to crashes and outputing it to a
107 * It is default off, but you can enable it with either specifying
108 * "ftrace_dump_on_oops" in the kernel command line, or setting
109 * /proc/sys/kernel/ftrace_dump_on_oops to true.
111 int ftrace_dump_on_oops;
113 static int tracing_set_tracer(char *buf);
115 static int __init set_ftrace(char *str)
117 tracing_set_tracer(str);
120 __setup("ftrace", set_ftrace);
122 static int __init set_ftrace_dump_on_oops(char *str)
124 ftrace_dump_on_oops = 1;
127 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
130 ns2usecs(cycle_t nsec)
137 cycle_t ftrace_now(int cpu)
139 u64 ts = ring_buffer_time_stamp(cpu);
140 ring_buffer_normalize_time_stamp(cpu, &ts);
145 * The global_trace is the descriptor that holds the tracing
146 * buffers for the live tracing. For each CPU, it contains
147 * a link list of pages that will store trace entries. The
148 * page descriptor of the pages in the memory is used to hold
149 * the link list by linking the lru item in the page descriptor
150 * to each of the pages in the buffer per CPU.
152 * For each active CPU there is a data field that holds the
153 * pages for the buffer for that CPU. Each CPU has the same number
154 * of pages allocated for its buffer.
156 static struct trace_array global_trace;
158 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
161 * The max_tr is used to snapshot the global_trace when a maximum
162 * latency is reached. Some tracers will use this to store a maximum
163 * trace while it continues examining live traces.
165 * The buffers for the max_tr are set up the same as the global_trace.
166 * When a snapshot is taken, the link list of the max_tr is swapped
167 * with the link list of the global_trace and the buffers are reset for
168 * the global_trace so the tracing can continue.
170 static struct trace_array max_tr;
172 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
174 /* tracer_enabled is used to toggle activation of a tracer */
175 static int tracer_enabled = 1;
178 * tracing_is_enabled - return tracer_enabled status
180 * This function is used by other tracers to know the status
181 * of the tracer_enabled flag. Tracers may use this function
182 * to know if it should enable their features when starting
183 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
185 int tracing_is_enabled(void)
187 return tracer_enabled;
190 /* function tracing enabled */
191 int ftrace_function_enabled;
194 * trace_buf_size is the size in bytes that is allocated
195 * for a buffer. Note, the number of bytes is always rounded
198 * This number is purposely set to a low number of 16384.
199 * If the dump on oops happens, it will be much appreciated
200 * to not have to wait for all that output. Anyway this can be
201 * boot time and run time configurable.
203 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
205 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
207 /* trace_types holds a link list of available tracers. */
208 static struct tracer *trace_types __read_mostly;
210 /* current_trace points to the tracer that is currently active */
211 static struct tracer *current_trace __read_mostly;
214 * max_tracer_type_len is used to simplify the allocating of
215 * buffers to read userspace tracer names. We keep track of
216 * the longest tracer name registered.
218 static int max_tracer_type_len;
221 * trace_types_lock is used to protect the trace_types list.
222 * This lock is also used to keep user access serialized.
223 * Accesses from userspace will grab this lock while userspace
224 * activities happen inside the kernel.
226 static DEFINE_MUTEX(trace_types_lock);
228 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
229 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
231 /* trace_flags holds trace_options default values */
232 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
236 * trace_wake_up - wake up tasks waiting for trace input
238 * Simply wakes up any task that is blocked on the trace_wait
239 * queue. These is used with trace_poll for tasks polling the trace.
241 void trace_wake_up(void)
244 * The runqueue_is_locked() can fail, but this is the best we
247 if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
248 wake_up(&trace_wait);
251 static int __init set_buf_size(char *str)
253 unsigned long buf_size;
258 ret = strict_strtoul(str, 0, &buf_size);
259 /* nr_entries can not be zero */
260 if (ret < 0 || buf_size == 0)
262 trace_buf_size = buf_size;
265 __setup("trace_buf_size=", set_buf_size);
267 unsigned long nsecs_to_usecs(unsigned long nsecs)
272 /* These must match the bit postions in trace_iterator_flags */
273 static const char *trace_options[] = {
295 * ftrace_max_lock is used to protect the swapping of buffers
296 * when taking a max snapshot. The buffers themselves are
297 * protected by per_cpu spinlocks. But the action of the swap
298 * needs its own lock.
300 * This is defined as a raw_spinlock_t in order to help
301 * with performance when lockdep debugging is enabled.
303 static raw_spinlock_t ftrace_max_lock =
304 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
307 * Copy the new maximum trace into the separate maximum-trace
308 * structure. (this way the maximum trace is permanently saved,
309 * for later retrieval via /debugfs/tracing/latency_trace)
312 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
314 struct trace_array_cpu *data = tr->data[cpu];
317 max_tr.time_start = data->preempt_timestamp;
319 data = max_tr.data[cpu];
320 data->saved_latency = tracing_max_latency;
322 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
323 data->pid = tsk->pid;
324 data->uid = tsk->uid;
325 data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
326 data->policy = tsk->policy;
327 data->rt_priority = tsk->rt_priority;
329 /* record this tasks comm */
330 tracing_record_cmdline(current);
334 * trace_seq_printf - sequence printing of trace information
335 * @s: trace sequence descriptor
336 * @fmt: printf format string
338 * The tracer may use either sequence operations or its own
339 * copy to user routines. To simplify formating of a trace
340 * trace_seq_printf is used to store strings into a special
341 * buffer (@s). Then the output may be either used by
342 * the sequencer or pulled into another buffer.
345 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
347 int len = (PAGE_SIZE - 1) - s->len;
355 ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
358 /* If we can't write it all, don't bother writing anything */
368 * trace_seq_puts - trace sequence printing of simple string
369 * @s: trace sequence descriptor
370 * @str: simple string to record
372 * The tracer may use either the sequence operations or its own
373 * copy to user routines. This function records a simple string
374 * into a special buffer (@s) for later retrieval by a sequencer
375 * or other mechanism.
378 trace_seq_puts(struct trace_seq *s, const char *str)
380 int len = strlen(str);
382 if (len > ((PAGE_SIZE - 1) - s->len))
385 memcpy(s->buffer + s->len, str, len);
392 trace_seq_putc(struct trace_seq *s, unsigned char c)
394 if (s->len >= (PAGE_SIZE - 1))
397 s->buffer[s->len++] = c;
403 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
405 if (len > ((PAGE_SIZE - 1) - s->len))
408 memcpy(s->buffer + s->len, mem, len);
414 #define MAX_MEMHEX_BYTES 8
415 #define HEX_CHARS (MAX_MEMHEX_BYTES*2 + 1)
418 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
420 unsigned char hex[HEX_CHARS];
421 unsigned char *data = mem;
425 for (i = 0, j = 0; i < len; i++) {
427 for (i = len-1, j = 0; i >= 0; i--) {
429 hex[j++] = hex_asc_hi(data[i]);
430 hex[j++] = hex_asc_lo(data[i]);
434 return trace_seq_putmem(s, hex, j);
438 trace_seq_path(struct trace_seq *s, struct path *path)
442 if (s->len >= (PAGE_SIZE - 1))
444 p = d_path(path, s->buffer + s->len, PAGE_SIZE - s->len);
446 p = mangle_path(s->buffer + s->len, p, "\n");
448 s->len = p - s->buffer;
452 s->buffer[s->len++] = '?';
460 trace_seq_reset(struct trace_seq *s)
466 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
471 if (s->len <= s->readpos)
474 len = s->len - s->readpos;
477 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
486 trace_print_seq(struct seq_file *m, struct trace_seq *s)
488 int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
491 seq_puts(m, s->buffer);
497 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
499 * @tsk: the task with the latency
500 * @cpu: The cpu that initiated the trace.
502 * Flip the buffers between the @tr and the max_tr and record information
503 * about which task was the cause of this latency.
506 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
508 struct ring_buffer *buf = tr->buffer;
510 WARN_ON_ONCE(!irqs_disabled());
511 __raw_spin_lock(&ftrace_max_lock);
513 tr->buffer = max_tr.buffer;
516 ftrace_disable_cpu();
517 ring_buffer_reset(tr->buffer);
520 __update_max_tr(tr, tsk, cpu);
521 __raw_spin_unlock(&ftrace_max_lock);
525 * update_max_tr_single - only copy one trace over, and reset the rest
527 * @tsk - task with the latency
528 * @cpu - the cpu of the buffer to copy.
530 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
533 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
537 WARN_ON_ONCE(!irqs_disabled());
538 __raw_spin_lock(&ftrace_max_lock);
540 ftrace_disable_cpu();
542 ring_buffer_reset(max_tr.buffer);
543 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
549 __update_max_tr(tr, tsk, cpu);
550 __raw_spin_unlock(&ftrace_max_lock);
554 * register_tracer - register a tracer with the ftrace system.
555 * @type - the plugin for the tracer
557 * Register a new plugin tracer.
559 int register_tracer(struct tracer *type)
566 pr_info("Tracer must have a name\n");
571 * When this gets called we hold the BKL which means that
572 * preemption is disabled. Various trace selftests however
573 * need to disable and enable preemption for successful tests.
574 * So we drop the BKL here and grab it after the tests again.
577 mutex_lock(&trace_types_lock);
579 tracing_selftest_running = true;
581 for (t = trace_types; t; t = t->next) {
582 if (strcmp(type->name, t->name) == 0) {
584 pr_info("Trace %s already registered\n",
592 type->set_flag = &dummy_set_flag;
594 type->flags = &dummy_tracer_flags;
596 if (!type->flags->opts)
597 type->flags->opts = dummy_tracer_opt;
599 #ifdef CONFIG_FTRACE_STARTUP_TEST
600 if (type->selftest) {
601 struct tracer *saved_tracer = current_trace;
602 struct trace_array *tr = &global_trace;
606 * Run a selftest on this tracer.
607 * Here we reset the trace buffer, and set the current
608 * tracer to be this tracer. The tracer can then run some
609 * internal tracing to verify that everything is in order.
610 * If we fail, we do not register this tracer.
612 for_each_tracing_cpu(i)
613 tracing_reset(tr, i);
615 current_trace = type;
616 /* the test is responsible for initializing and enabling */
617 pr_info("Testing tracer %s: ", type->name);
618 ret = type->selftest(type, tr);
619 /* the test is responsible for resetting too */
620 current_trace = saved_tracer;
622 printk(KERN_CONT "FAILED!\n");
625 /* Only reset on passing, to avoid touching corrupted buffers */
626 for_each_tracing_cpu(i)
627 tracing_reset(tr, i);
629 printk(KERN_CONT "PASSED\n");
633 type->next = trace_types;
635 len = strlen(type->name);
636 if (len > max_tracer_type_len)
637 max_tracer_type_len = len;
640 tracing_selftest_running = false;
641 mutex_unlock(&trace_types_lock);
647 void unregister_tracer(struct tracer *type)
652 mutex_lock(&trace_types_lock);
653 for (t = &trace_types; *t; t = &(*t)->next) {
657 pr_info("Trace %s not registered\n", type->name);
662 if (strlen(type->name) != max_tracer_type_len)
665 max_tracer_type_len = 0;
666 for (t = &trace_types; *t; t = &(*t)->next) {
667 len = strlen((*t)->name);
668 if (len > max_tracer_type_len)
669 max_tracer_type_len = len;
672 mutex_unlock(&trace_types_lock);
675 void tracing_reset(struct trace_array *tr, int cpu)
677 ftrace_disable_cpu();
678 ring_buffer_reset_cpu(tr->buffer, cpu);
682 #define SAVED_CMDLINES 128
683 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
684 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
685 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
686 static int cmdline_idx;
687 static DEFINE_SPINLOCK(trace_cmdline_lock);
689 /* temporary disable recording */
690 atomic_t trace_record_cmdline_disabled __read_mostly;
692 static void trace_init_cmdlines(void)
694 memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
695 memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
699 static int trace_stop_count;
700 static DEFINE_SPINLOCK(tracing_start_lock);
703 * ftrace_off_permanent - disable all ftrace code permanently
705 * This should only be called when a serious anomally has
706 * been detected. This will turn off the function tracing,
707 * ring buffers, and other tracing utilites. It takes no
708 * locks and can be called from any context.
710 void ftrace_off_permanent(void)
712 tracing_disabled = 1;
714 tracing_off_permanent();
718 * tracing_start - quick start of the tracer
720 * If tracing is enabled but was stopped by tracing_stop,
721 * this will start the tracer back up.
723 void tracing_start(void)
725 struct ring_buffer *buffer;
728 if (tracing_disabled)
731 spin_lock_irqsave(&tracing_start_lock, flags);
732 if (--trace_stop_count)
735 if (trace_stop_count < 0) {
736 /* Someone screwed up their debugging */
738 trace_stop_count = 0;
743 buffer = global_trace.buffer;
745 ring_buffer_record_enable(buffer);
747 buffer = max_tr.buffer;
749 ring_buffer_record_enable(buffer);
753 spin_unlock_irqrestore(&tracing_start_lock, flags);
757 * tracing_stop - quick stop of the tracer
759 * Light weight way to stop tracing. Use in conjunction with
762 void tracing_stop(void)
764 struct ring_buffer *buffer;
768 spin_lock_irqsave(&tracing_start_lock, flags);
769 if (trace_stop_count++)
772 buffer = global_trace.buffer;
774 ring_buffer_record_disable(buffer);
776 buffer = max_tr.buffer;
778 ring_buffer_record_disable(buffer);
781 spin_unlock_irqrestore(&tracing_start_lock, flags);
784 void trace_stop_cmdline_recording(void);
786 static void trace_save_cmdline(struct task_struct *tsk)
791 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
795 * It's not the end of the world if we don't get
796 * the lock, but we also don't want to spin
797 * nor do we want to disable interrupts,
798 * so if we miss here, then better luck next time.
800 if (!spin_trylock(&trace_cmdline_lock))
803 idx = map_pid_to_cmdline[tsk->pid];
804 if (idx >= SAVED_CMDLINES) {
805 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
807 map = map_cmdline_to_pid[idx];
808 if (map <= PID_MAX_DEFAULT)
809 map_pid_to_cmdline[map] = (unsigned)-1;
811 map_pid_to_cmdline[tsk->pid] = idx;
816 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
818 spin_unlock(&trace_cmdline_lock);
821 char *trace_find_cmdline(int pid)
823 char *cmdline = "<...>";
829 if (pid > PID_MAX_DEFAULT)
832 map = map_pid_to_cmdline[pid];
833 if (map >= SAVED_CMDLINES)
836 cmdline = saved_cmdlines[map];
842 void tracing_record_cmdline(struct task_struct *tsk)
844 if (atomic_read(&trace_record_cmdline_disabled))
847 trace_save_cmdline(tsk);
851 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
854 struct task_struct *tsk = current;
856 entry->preempt_count = pc & 0xff;
857 entry->pid = (tsk) ? tsk->pid : 0;
858 entry->tgid = (tsk) ? tsk->tgid : 0;
860 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
861 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
863 TRACE_FLAG_IRQS_NOSUPPORT |
865 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
866 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
867 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
871 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
872 unsigned long ip, unsigned long parent_ip, unsigned long flags,
875 struct ring_buffer_event *event;
876 struct ftrace_entry *entry;
877 unsigned long irq_flags;
879 /* If we are reading the ring buffer, don't trace */
880 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
883 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
887 entry = ring_buffer_event_data(event);
888 tracing_generic_entry_update(&entry->ent, flags, pc);
889 entry->ent.type = TRACE_FN;
891 entry->parent_ip = parent_ip;
892 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
895 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
896 static void __trace_graph_entry(struct trace_array *tr,
897 struct trace_array_cpu *data,
898 struct ftrace_graph_ent *trace,
902 struct ring_buffer_event *event;
903 struct ftrace_graph_ent_entry *entry;
904 unsigned long irq_flags;
906 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
909 event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry),
913 entry = ring_buffer_event_data(event);
914 tracing_generic_entry_update(&entry->ent, flags, pc);
915 entry->ent.type = TRACE_GRAPH_ENT;
916 entry->graph_ent = *trace;
917 ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
920 static void __trace_graph_return(struct trace_array *tr,
921 struct trace_array_cpu *data,
922 struct ftrace_graph_ret *trace,
926 struct ring_buffer_event *event;
927 struct ftrace_graph_ret_entry *entry;
928 unsigned long irq_flags;
930 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
933 event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry),
937 entry = ring_buffer_event_data(event);
938 tracing_generic_entry_update(&entry->ent, flags, pc);
939 entry->ent.type = TRACE_GRAPH_RET;
941 ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
946 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
947 unsigned long ip, unsigned long parent_ip, unsigned long flags,
950 if (likely(!atomic_read(&data->disabled)))
951 trace_function(tr, data, ip, parent_ip, flags, pc);
954 static void ftrace_trace_stack(struct trace_array *tr,
955 struct trace_array_cpu *data,
959 #ifdef CONFIG_STACKTRACE
960 struct ring_buffer_event *event;
961 struct stack_entry *entry;
962 struct stack_trace trace;
963 unsigned long irq_flags;
965 if (!(trace_flags & TRACE_ITER_STACKTRACE))
968 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
972 entry = ring_buffer_event_data(event);
973 tracing_generic_entry_update(&entry->ent, flags, pc);
974 entry->ent.type = TRACE_STACK;
976 memset(&entry->caller, 0, sizeof(entry->caller));
978 trace.nr_entries = 0;
979 trace.max_entries = FTRACE_STACK_ENTRIES;
981 trace.entries = entry->caller;
983 save_stack_trace(&trace);
984 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
988 void __trace_stack(struct trace_array *tr,
989 struct trace_array_cpu *data,
993 ftrace_trace_stack(tr, data, flags, skip, preempt_count());
996 static void ftrace_trace_userstack(struct trace_array *tr,
997 struct trace_array_cpu *data,
998 unsigned long flags, int pc)
1000 #ifdef CONFIG_STACKTRACE
1001 struct ring_buffer_event *event;
1002 struct userstack_entry *entry;
1003 struct stack_trace trace;
1004 unsigned long irq_flags;
1006 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1009 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1013 entry = ring_buffer_event_data(event);
1014 tracing_generic_entry_update(&entry->ent, flags, pc);
1015 entry->ent.type = TRACE_USER_STACK;
1017 memset(&entry->caller, 0, sizeof(entry->caller));
1019 trace.nr_entries = 0;
1020 trace.max_entries = FTRACE_STACK_ENTRIES;
1022 trace.entries = entry->caller;
1024 save_stack_trace_user(&trace);
1025 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1029 void __trace_userstack(struct trace_array *tr,
1030 struct trace_array_cpu *data,
1031 unsigned long flags)
1033 ftrace_trace_userstack(tr, data, flags, preempt_count());
1037 ftrace_trace_special(void *__tr, void *__data,
1038 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1041 struct ring_buffer_event *event;
1042 struct trace_array_cpu *data = __data;
1043 struct trace_array *tr = __tr;
1044 struct special_entry *entry;
1045 unsigned long irq_flags;
1047 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1051 entry = ring_buffer_event_data(event);
1052 tracing_generic_entry_update(&entry->ent, 0, pc);
1053 entry->ent.type = TRACE_SPECIAL;
1057 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1058 ftrace_trace_stack(tr, data, irq_flags, 4, pc);
1059 ftrace_trace_userstack(tr, data, irq_flags, pc);
1065 __trace_special(void *__tr, void *__data,
1066 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1068 ftrace_trace_special(__tr, __data, arg1, arg2, arg3, preempt_count());
1072 tracing_sched_switch_trace(struct trace_array *tr,
1073 struct trace_array_cpu *data,
1074 struct task_struct *prev,
1075 struct task_struct *next,
1076 unsigned long flags, int pc)
1078 struct ring_buffer_event *event;
1079 struct ctx_switch_entry *entry;
1080 unsigned long irq_flags;
1082 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1086 entry = ring_buffer_event_data(event);
1087 tracing_generic_entry_update(&entry->ent, flags, pc);
1088 entry->ent.type = TRACE_CTX;
1089 entry->prev_pid = prev->pid;
1090 entry->prev_prio = prev->prio;
1091 entry->prev_state = prev->state;
1092 entry->next_pid = next->pid;
1093 entry->next_prio = next->prio;
1094 entry->next_state = next->state;
1095 entry->next_cpu = task_cpu(next);
1096 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1097 ftrace_trace_stack(tr, data, flags, 5, pc);
1098 ftrace_trace_userstack(tr, data, flags, pc);
1102 tracing_sched_wakeup_trace(struct trace_array *tr,
1103 struct trace_array_cpu *data,
1104 struct task_struct *wakee,
1105 struct task_struct *curr,
1106 unsigned long flags, int pc)
1108 struct ring_buffer_event *event;
1109 struct ctx_switch_entry *entry;
1110 unsigned long irq_flags;
1112 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1116 entry = ring_buffer_event_data(event);
1117 tracing_generic_entry_update(&entry->ent, flags, pc);
1118 entry->ent.type = TRACE_WAKE;
1119 entry->prev_pid = curr->pid;
1120 entry->prev_prio = curr->prio;
1121 entry->prev_state = curr->state;
1122 entry->next_pid = wakee->pid;
1123 entry->next_prio = wakee->prio;
1124 entry->next_state = wakee->state;
1125 entry->next_cpu = task_cpu(wakee);
1126 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1127 ftrace_trace_stack(tr, data, flags, 6, pc);
1128 ftrace_trace_userstack(tr, data, flags, pc);
1134 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1136 struct trace_array *tr = &global_trace;
1137 struct trace_array_cpu *data;
1138 unsigned long flags;
1142 if (tracing_disabled)
1145 pc = preempt_count();
1146 local_irq_save(flags);
1147 cpu = raw_smp_processor_id();
1148 data = tr->data[cpu];
1150 if (likely(atomic_inc_return(&data->disabled) == 1))
1151 ftrace_trace_special(tr, data, arg1, arg2, arg3, pc);
1153 atomic_dec(&data->disabled);
1154 local_irq_restore(flags);
1157 #ifdef CONFIG_FUNCTION_TRACER
1159 function_trace_call_preempt_only(unsigned long ip, unsigned long parent_ip)
1161 struct trace_array *tr = &global_trace;
1162 struct trace_array_cpu *data;
1163 unsigned long flags;
1168 if (unlikely(!ftrace_function_enabled))
1171 pc = preempt_count();
1172 resched = ftrace_preempt_disable();
1173 local_save_flags(flags);
1174 cpu = raw_smp_processor_id();
1175 data = tr->data[cpu];
1176 disabled = atomic_inc_return(&data->disabled);
1178 if (likely(disabled == 1))
1179 trace_function(tr, data, ip, parent_ip, flags, pc);
1181 atomic_dec(&data->disabled);
1182 ftrace_preempt_enable(resched);
1186 function_trace_call(unsigned long ip, unsigned long parent_ip)
1188 struct trace_array *tr = &global_trace;
1189 struct trace_array_cpu *data;
1190 unsigned long flags;
1195 if (unlikely(!ftrace_function_enabled))
1199 * Need to use raw, since this must be called before the
1200 * recursive protection is performed.
1202 local_irq_save(flags);
1203 cpu = raw_smp_processor_id();
1204 data = tr->data[cpu];
1205 disabled = atomic_inc_return(&data->disabled);
1207 if (likely(disabled == 1)) {
1208 pc = preempt_count();
1209 trace_function(tr, data, ip, parent_ip, flags, pc);
1212 atomic_dec(&data->disabled);
1213 local_irq_restore(flags);
1216 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1217 int trace_graph_entry(struct ftrace_graph_ent *trace)
1219 struct trace_array *tr = &global_trace;
1220 struct trace_array_cpu *data;
1221 unsigned long flags;
1226 if (!ftrace_trace_task(current))
1229 if (!ftrace_graph_addr(trace->func))
1232 local_irq_save(flags);
1233 cpu = raw_smp_processor_id();
1234 data = tr->data[cpu];
1235 disabled = atomic_inc_return(&data->disabled);
1236 if (likely(disabled == 1)) {
1237 pc = preempt_count();
1238 __trace_graph_entry(tr, data, trace, flags, pc);
1240 /* Only do the atomic if it is not already set */
1241 if (!test_tsk_trace_graph(current))
1242 set_tsk_trace_graph(current);
1243 atomic_dec(&data->disabled);
1244 local_irq_restore(flags);
1249 void trace_graph_return(struct ftrace_graph_ret *trace)
1251 struct trace_array *tr = &global_trace;
1252 struct trace_array_cpu *data;
1253 unsigned long flags;
1258 local_irq_save(flags);
1259 cpu = raw_smp_processor_id();
1260 data = tr->data[cpu];
1261 disabled = atomic_inc_return(&data->disabled);
1262 if (likely(disabled == 1)) {
1263 pc = preempt_count();
1264 __trace_graph_return(tr, data, trace, flags, pc);
1267 clear_tsk_trace_graph(current);
1268 atomic_dec(&data->disabled);
1269 local_irq_restore(flags);
1271 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1273 static struct ftrace_ops trace_ops __read_mostly =
1275 .func = function_trace_call,
1278 void tracing_start_function_trace(void)
1280 ftrace_function_enabled = 0;
1282 if (trace_flags & TRACE_ITER_PREEMPTONLY)
1283 trace_ops.func = function_trace_call_preempt_only;
1285 trace_ops.func = function_trace_call;
1287 register_ftrace_function(&trace_ops);
1288 ftrace_function_enabled = 1;
1291 void tracing_stop_function_trace(void)
1293 ftrace_function_enabled = 0;
1294 unregister_ftrace_function(&trace_ops);
1298 enum trace_file_type {
1299 TRACE_FILE_LAT_FMT = 1,
1300 TRACE_FILE_ANNOTATE = 2,
1303 static void trace_iterator_increment(struct trace_iterator *iter, int cpu)
1305 /* Don't allow ftrace to trace into the ring buffers */
1306 ftrace_disable_cpu();
1309 if (iter->buffer_iter[iter->cpu])
1310 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1312 ftrace_enable_cpu();
1315 static struct trace_entry *
1316 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1318 struct ring_buffer_event *event;
1319 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1321 /* Don't allow ftrace to trace into the ring buffers */
1322 ftrace_disable_cpu();
1325 event = ring_buffer_iter_peek(buf_iter, ts);
1327 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1329 ftrace_enable_cpu();
1331 return event ? ring_buffer_event_data(event) : NULL;
1334 static struct trace_entry *
1335 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1337 struct ring_buffer *buffer = iter->tr->buffer;
1338 struct trace_entry *ent, *next = NULL;
1339 u64 next_ts = 0, ts;
1343 for_each_tracing_cpu(cpu) {
1345 if (ring_buffer_empty_cpu(buffer, cpu))
1348 ent = peek_next_entry(iter, cpu, &ts);
1351 * Pick the entry with the smallest timestamp:
1353 if (ent && (!next || ts < next_ts)) {
1361 *ent_cpu = next_cpu;
1369 /* Find the next real entry, without updating the iterator itself */
1370 static struct trace_entry *
1371 find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1373 return __find_next_entry(iter, ent_cpu, ent_ts);
1376 /* Find the next real entry, and increment the iterator to the next entry */
1377 static void *find_next_entry_inc(struct trace_iterator *iter)
1379 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1382 trace_iterator_increment(iter, iter->cpu);
1384 return iter->ent ? iter : NULL;
1387 static void trace_consume(struct trace_iterator *iter)
1389 /* Don't allow ftrace to trace into the ring buffers */
1390 ftrace_disable_cpu();
1391 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1392 ftrace_enable_cpu();
1395 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1397 struct trace_iterator *iter = m->private;
1403 /* can't go backwards */
1408 ent = find_next_entry_inc(iter);
1412 while (ent && iter->idx < i)
1413 ent = find_next_entry_inc(iter);
1420 static void *s_start(struct seq_file *m, loff_t *pos)
1422 struct trace_iterator *iter = m->private;
1427 mutex_lock(&trace_types_lock);
1429 if (!current_trace || current_trace != iter->trace) {
1430 mutex_unlock(&trace_types_lock);
1434 atomic_inc(&trace_record_cmdline_disabled);
1436 if (*pos != iter->pos) {
1441 ftrace_disable_cpu();
1443 for_each_tracing_cpu(cpu) {
1444 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1447 ftrace_enable_cpu();
1449 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1454 p = s_next(m, p, &l);
1460 static void s_stop(struct seq_file *m, void *p)
1462 atomic_dec(&trace_record_cmdline_disabled);
1463 mutex_unlock(&trace_types_lock);
1466 #ifdef CONFIG_KRETPROBES
1467 static inline const char *kretprobed(const char *name)
1469 static const char tramp_name[] = "kretprobe_trampoline";
1470 int size = sizeof(tramp_name);
1472 if (strncmp(tramp_name, name, size) == 0)
1473 return "[unknown/kretprobe'd]";
1477 static inline const char *kretprobed(const char *name)
1481 #endif /* CONFIG_KRETPROBES */
1484 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1486 #ifdef CONFIG_KALLSYMS
1487 char str[KSYM_SYMBOL_LEN];
1490 kallsyms_lookup(address, NULL, NULL, NULL, str);
1492 name = kretprobed(str);
1494 return trace_seq_printf(s, fmt, name);
1500 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1501 unsigned long address)
1503 #ifdef CONFIG_KALLSYMS
1504 char str[KSYM_SYMBOL_LEN];
1507 sprint_symbol(str, address);
1508 name = kretprobed(str);
1510 return trace_seq_printf(s, fmt, name);
1515 #ifndef CONFIG_64BIT
1516 # define IP_FMT "%08lx"
1518 # define IP_FMT "%016lx"
1522 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1527 return trace_seq_printf(s, "0");
1529 if (sym_flags & TRACE_ITER_SYM_OFFSET)
1530 ret = seq_print_sym_offset(s, "%s", ip);
1532 ret = seq_print_sym_short(s, "%s", ip);
1537 if (sym_flags & TRACE_ITER_SYM_ADDR)
1538 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1542 static inline int seq_print_user_ip(struct trace_seq *s, struct mm_struct *mm,
1543 unsigned long ip, unsigned long sym_flags)
1545 struct file *file = NULL;
1546 unsigned long vmstart = 0;
1550 const struct vm_area_struct *vma;
1552 down_read(&mm->mmap_sem);
1553 vma = find_vma(mm, ip);
1555 file = vma->vm_file;
1556 vmstart = vma->vm_start;
1559 ret = trace_seq_path(s, &file->f_path);
1561 ret = trace_seq_printf(s, "[+0x%lx]", ip - vmstart);
1563 up_read(&mm->mmap_sem);
1565 if (ret && ((sym_flags & TRACE_ITER_SYM_ADDR) || !file))
1566 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1571 seq_print_userip_objs(const struct userstack_entry *entry, struct trace_seq *s,
1572 unsigned long sym_flags)
1574 struct mm_struct *mm = NULL;
1578 if (trace_flags & TRACE_ITER_SYM_USEROBJ) {
1579 struct task_struct *task;
1581 * we do the lookup on the thread group leader,
1582 * since individual threads might have already quit!
1585 task = find_task_by_vpid(entry->ent.tgid);
1587 mm = get_task_mm(task);
1591 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1592 unsigned long ip = entry->caller[i];
1594 if (ip == ULONG_MAX || !ret)
1597 ret = trace_seq_puts(s, " <- ");
1600 ret = trace_seq_puts(s, "??");
1606 ret = seq_print_user_ip(s, mm, ip, sym_flags);
1614 static void print_lat_help_header(struct seq_file *m)
1616 seq_puts(m, "# _------=> CPU# \n");
1617 seq_puts(m, "# / _-----=> irqs-off \n");
1618 seq_puts(m, "# | / _----=> need-resched \n");
1619 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1620 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1621 seq_puts(m, "# |||| / \n");
1622 seq_puts(m, "# ||||| delay \n");
1623 seq_puts(m, "# cmd pid ||||| time | caller \n");
1624 seq_puts(m, "# \\ / ||||| \\ | / \n");
1627 static void print_func_help_header(struct seq_file *m)
1629 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1630 seq_puts(m, "# | | | | |\n");
1635 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1637 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1638 struct trace_array *tr = iter->tr;
1639 struct trace_array_cpu *data = tr->data[tr->cpu];
1640 struct tracer *type = current_trace;
1641 unsigned long total;
1642 unsigned long entries;
1643 const char *name = "preemption";
1648 entries = ring_buffer_entries(iter->tr->buffer);
1650 ring_buffer_overruns(iter->tr->buffer);
1652 seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1654 seq_puts(m, "-----------------------------------"
1655 "---------------------------------\n");
1656 seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1657 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1658 nsecs_to_usecs(data->saved_latency),
1662 #if defined(CONFIG_PREEMPT_NONE)
1664 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1666 #elif defined(CONFIG_PREEMPT)
1671 /* These are reserved for later use */
1674 seq_printf(m, " #P:%d)\n", num_online_cpus());
1678 seq_puts(m, " -----------------\n");
1679 seq_printf(m, " | task: %.16s-%d "
1680 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1681 data->comm, data->pid, data->uid, data->nice,
1682 data->policy, data->rt_priority);
1683 seq_puts(m, " -----------------\n");
1685 if (data->critical_start) {
1686 seq_puts(m, " => started at: ");
1687 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1688 trace_print_seq(m, &iter->seq);
1689 seq_puts(m, "\n => ended at: ");
1690 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1691 trace_print_seq(m, &iter->seq);
1699 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1701 int hardirq, softirq;
1704 comm = trace_find_cmdline(entry->pid);
1706 trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1707 trace_seq_printf(s, "%3d", cpu);
1708 trace_seq_printf(s, "%c%c",
1709 (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' :
1710 (entry->flags & TRACE_FLAG_IRQS_NOSUPPORT) ? 'X' : '.',
1711 ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1713 hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1714 softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1715 if (hardirq && softirq) {
1716 trace_seq_putc(s, 'H');
1719 trace_seq_putc(s, 'h');
1722 trace_seq_putc(s, 's');
1724 trace_seq_putc(s, '.');
1728 if (entry->preempt_count)
1729 trace_seq_printf(s, "%x", entry->preempt_count);
1731 trace_seq_puts(s, ".");
1734 unsigned long preempt_mark_thresh = 100;
1737 lat_print_timestamp(struct trace_seq *s, u64 abs_usecs,
1738 unsigned long rel_usecs)
1740 trace_seq_printf(s, " %4lldus", abs_usecs);
1741 if (rel_usecs > preempt_mark_thresh)
1742 trace_seq_puts(s, "!: ");
1743 else if (rel_usecs > 1)
1744 trace_seq_puts(s, "+: ");
1746 trace_seq_puts(s, " : ");
1749 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1752 * The message is supposed to contain an ending newline.
1753 * If the printing stops prematurely, try to add a newline of our own.
1755 void trace_seq_print_cont(struct trace_seq *s, struct trace_iterator *iter)
1757 struct trace_entry *ent;
1758 struct trace_field_cont *cont;
1761 ent = peek_next_entry(iter, iter->cpu, NULL);
1762 if (!ent || ent->type != TRACE_CONT) {
1763 trace_seq_putc(s, '\n');
1768 cont = (struct trace_field_cont *)ent;
1770 ok = (trace_seq_printf(s, "%s", cont->buf) > 0);
1772 ftrace_disable_cpu();
1774 if (iter->buffer_iter[iter->cpu])
1775 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1777 ring_buffer_consume(iter->tr->buffer, iter->cpu, NULL);
1779 ftrace_enable_cpu();
1781 ent = peek_next_entry(iter, iter->cpu, NULL);
1782 } while (ent && ent->type == TRACE_CONT);
1785 trace_seq_putc(s, '\n');
1788 static void test_cpu_buff_start(struct trace_iterator *iter)
1790 struct trace_seq *s = &iter->seq;
1792 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1795 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1798 if (cpu_isset(iter->cpu, iter->started))
1801 cpu_set(iter->cpu, iter->started);
1802 trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1805 static enum print_line_t
1806 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1808 struct trace_seq *s = &iter->seq;
1809 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1810 struct trace_entry *next_entry;
1811 unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1812 struct trace_entry *entry = iter->ent;
1813 unsigned long abs_usecs;
1814 unsigned long rel_usecs;
1820 if (entry->type == TRACE_CONT)
1821 return TRACE_TYPE_HANDLED;
1823 test_cpu_buff_start(iter);
1825 next_entry = find_next_entry(iter, NULL, &next_ts);
1828 rel_usecs = ns2usecs(next_ts - iter->ts);
1829 abs_usecs = ns2usecs(iter->ts - iter->tr->time_start);
1832 comm = trace_find_cmdline(entry->pid);
1833 trace_seq_printf(s, "%16s %5d %3d %d %08x %08x [%08lx]"
1834 " %ld.%03ldms (+%ld.%03ldms): ",
1836 entry->pid, cpu, entry->flags,
1837 entry->preempt_count, trace_idx,
1840 abs_usecs % 1000, rel_usecs/1000,
1843 lat_print_generic(s, entry, cpu);
1844 lat_print_timestamp(s, abs_usecs, rel_usecs);
1846 switch (entry->type) {
1848 struct ftrace_entry *field;
1850 trace_assign_type(field, entry);
1852 seq_print_ip_sym(s, field->ip, sym_flags);
1853 trace_seq_puts(s, " (");
1854 seq_print_ip_sym(s, field->parent_ip, sym_flags);
1855 trace_seq_puts(s, ")\n");
1860 struct ctx_switch_entry *field;
1862 trace_assign_type(field, entry);
1864 T = field->next_state < sizeof(state_to_char) ?
1865 state_to_char[field->next_state] : 'X';
1867 state = field->prev_state ?
1868 __ffs(field->prev_state) + 1 : 0;
1869 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1870 comm = trace_find_cmdline(field->next_pid);
1871 trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c %s\n",
1874 S, entry->type == TRACE_CTX ? "==>" : " +",
1881 case TRACE_SPECIAL: {
1882 struct special_entry *field;
1884 trace_assign_type(field, entry);
1886 trace_seq_printf(s, "# %ld %ld %ld\n",
1893 struct stack_entry *field;
1895 trace_assign_type(field, entry);
1897 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1899 trace_seq_puts(s, " <= ");
1900 seq_print_ip_sym(s, field->caller[i], sym_flags);
1902 trace_seq_puts(s, "\n");
1906 struct print_entry *field;
1908 trace_assign_type(field, entry);
1910 seq_print_ip_sym(s, field->ip, sym_flags);
1911 trace_seq_printf(s, ": %s", field->buf);
1912 if (entry->flags & TRACE_FLAG_CONT)
1913 trace_seq_print_cont(s, iter);
1916 case TRACE_BRANCH: {
1917 struct trace_branch *field;
1919 trace_assign_type(field, entry);
1921 trace_seq_printf(s, "[%s] %s:%s:%d\n",
1922 field->correct ? " ok " : " MISS ",
1928 case TRACE_USER_STACK: {
1929 struct userstack_entry *field;
1931 trace_assign_type(field, entry);
1933 seq_print_userip_objs(field, s, sym_flags);
1934 trace_seq_putc(s, '\n');
1938 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1940 return TRACE_TYPE_HANDLED;
1943 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1945 struct trace_seq *s = &iter->seq;
1946 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1947 struct trace_entry *entry;
1948 unsigned long usec_rem;
1949 unsigned long long t;
1958 if (entry->type == TRACE_CONT)
1959 return TRACE_TYPE_HANDLED;
1961 test_cpu_buff_start(iter);
1963 comm = trace_find_cmdline(iter->ent->pid);
1965 t = ns2usecs(iter->ts);
1966 usec_rem = do_div(t, 1000000ULL);
1967 secs = (unsigned long)t;
1969 ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1971 return TRACE_TYPE_PARTIAL_LINE;
1972 ret = trace_seq_printf(s, "[%03d] ", iter->cpu);
1974 return TRACE_TYPE_PARTIAL_LINE;
1975 ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1977 return TRACE_TYPE_PARTIAL_LINE;
1979 switch (entry->type) {
1981 struct ftrace_entry *field;
1983 trace_assign_type(field, entry);
1985 ret = seq_print_ip_sym(s, field->ip, sym_flags);
1987 return TRACE_TYPE_PARTIAL_LINE;
1988 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1990 ret = trace_seq_printf(s, " <-");
1992 return TRACE_TYPE_PARTIAL_LINE;
1993 ret = seq_print_ip_sym(s,
1997 return TRACE_TYPE_PARTIAL_LINE;
1999 ret = trace_seq_printf(s, "\n");
2001 return TRACE_TYPE_PARTIAL_LINE;
2006 struct ctx_switch_entry *field;
2008 trace_assign_type(field, entry);
2010 S = field->prev_state < sizeof(state_to_char) ?
2011 state_to_char[field->prev_state] : 'X';
2012 T = field->next_state < sizeof(state_to_char) ?
2013 state_to_char[field->next_state] : 'X';
2014 ret = trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c\n",
2018 entry->type == TRACE_CTX ? "==>" : " +",
2024 return TRACE_TYPE_PARTIAL_LINE;
2027 case TRACE_SPECIAL: {
2028 struct special_entry *field;
2030 trace_assign_type(field, entry);
2032 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
2037 return TRACE_TYPE_PARTIAL_LINE;
2041 struct stack_entry *field;
2043 trace_assign_type(field, entry);
2045 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
2047 ret = trace_seq_puts(s, " <= ");
2049 return TRACE_TYPE_PARTIAL_LINE;
2051 ret = seq_print_ip_sym(s, field->caller[i],
2054 return TRACE_TYPE_PARTIAL_LINE;
2056 ret = trace_seq_puts(s, "\n");
2058 return TRACE_TYPE_PARTIAL_LINE;
2062 struct print_entry *field;
2064 trace_assign_type(field, entry);
2066 seq_print_ip_sym(s, field->ip, sym_flags);
2067 trace_seq_printf(s, ": %s", field->buf);
2068 if (entry->flags & TRACE_FLAG_CONT)
2069 trace_seq_print_cont(s, iter);
2072 case TRACE_GRAPH_RET: {
2073 return print_graph_function(iter);
2075 case TRACE_GRAPH_ENT: {
2076 return print_graph_function(iter);
2078 case TRACE_BRANCH: {
2079 struct trace_branch *field;
2081 trace_assign_type(field, entry);
2083 trace_seq_printf(s, "[%s] %s:%s:%d\n",
2084 field->correct ? " ok " : " MISS ",
2090 case TRACE_USER_STACK: {
2091 struct userstack_entry *field;
2093 trace_assign_type(field, entry);
2095 ret = seq_print_userip_objs(field, s, sym_flags);
2097 return TRACE_TYPE_PARTIAL_LINE;
2098 ret = trace_seq_putc(s, '\n');
2100 return TRACE_TYPE_PARTIAL_LINE;
2104 return TRACE_TYPE_HANDLED;
2107 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2109 struct trace_seq *s = &iter->seq;
2110 struct trace_entry *entry;
2116 if (entry->type == TRACE_CONT)
2117 return TRACE_TYPE_HANDLED;
2119 ret = trace_seq_printf(s, "%d %d %llu ",
2120 entry->pid, iter->cpu, iter->ts);
2122 return TRACE_TYPE_PARTIAL_LINE;
2124 switch (entry->type) {
2126 struct ftrace_entry *field;
2128 trace_assign_type(field, entry);
2130 ret = trace_seq_printf(s, "%x %x\n",
2134 return TRACE_TYPE_PARTIAL_LINE;
2139 struct ctx_switch_entry *field;
2141 trace_assign_type(field, entry);
2143 S = field->prev_state < sizeof(state_to_char) ?
2144 state_to_char[field->prev_state] : 'X';
2145 T = field->next_state < sizeof(state_to_char) ?
2146 state_to_char[field->next_state] : 'X';
2147 if (entry->type == TRACE_WAKE)
2149 ret = trace_seq_printf(s, "%d %d %c %d %d %d %c\n",
2158 return TRACE_TYPE_PARTIAL_LINE;
2162 case TRACE_USER_STACK:
2164 struct special_entry *field;
2166 trace_assign_type(field, entry);
2168 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
2173 return TRACE_TYPE_PARTIAL_LINE;
2177 struct print_entry *field;
2179 trace_assign_type(field, entry);
2181 trace_seq_printf(s, "# %lx %s", field->ip, field->buf);
2182 if (entry->flags & TRACE_FLAG_CONT)
2183 trace_seq_print_cont(s, iter);
2187 return TRACE_TYPE_HANDLED;
2190 #define SEQ_PUT_FIELD_RET(s, x) \
2192 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
2196 #define SEQ_PUT_HEX_FIELD_RET(s, x) \
2198 BUILD_BUG_ON(sizeof(x) > MAX_MEMHEX_BYTES); \
2199 if (!trace_seq_putmem_hex(s, &(x), sizeof(x))) \
2203 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2205 struct trace_seq *s = &iter->seq;
2206 unsigned char newline = '\n';
2207 struct trace_entry *entry;
2212 if (entry->type == TRACE_CONT)
2213 return TRACE_TYPE_HANDLED;
2215 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2216 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2217 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2219 switch (entry->type) {
2221 struct ftrace_entry *field;
2223 trace_assign_type(field, entry);
2225 SEQ_PUT_HEX_FIELD_RET(s, field->ip);
2226 SEQ_PUT_HEX_FIELD_RET(s, field->parent_ip);
2231 struct ctx_switch_entry *field;
2233 trace_assign_type(field, entry);
2235 S = field->prev_state < sizeof(state_to_char) ?
2236 state_to_char[field->prev_state] : 'X';
2237 T = field->next_state < sizeof(state_to_char) ?
2238 state_to_char[field->next_state] : 'X';
2239 if (entry->type == TRACE_WAKE)
2241 SEQ_PUT_HEX_FIELD_RET(s, field->prev_pid);
2242 SEQ_PUT_HEX_FIELD_RET(s, field->prev_prio);
2243 SEQ_PUT_HEX_FIELD_RET(s, S);
2244 SEQ_PUT_HEX_FIELD_RET(s, field->next_cpu);
2245 SEQ_PUT_HEX_FIELD_RET(s, field->next_pid);
2246 SEQ_PUT_HEX_FIELD_RET(s, field->next_prio);
2247 SEQ_PUT_HEX_FIELD_RET(s, T);
2251 case TRACE_USER_STACK:
2253 struct special_entry *field;
2255 trace_assign_type(field, entry);
2257 SEQ_PUT_HEX_FIELD_RET(s, field->arg1);
2258 SEQ_PUT_HEX_FIELD_RET(s, field->arg2);
2259 SEQ_PUT_HEX_FIELD_RET(s, field->arg3);
2263 SEQ_PUT_FIELD_RET(s, newline);
2265 return TRACE_TYPE_HANDLED;
2268 static enum print_line_t print_printk_msg_only(struct trace_iterator *iter)
2270 struct trace_seq *s = &iter->seq;
2271 struct trace_entry *entry = iter->ent;
2272 struct print_entry *field;
2275 trace_assign_type(field, entry);
2277 ret = trace_seq_printf(s, field->buf);
2279 return TRACE_TYPE_PARTIAL_LINE;
2281 if (entry->flags & TRACE_FLAG_CONT)
2282 trace_seq_print_cont(s, iter);
2284 return TRACE_TYPE_HANDLED;
2287 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2289 struct trace_seq *s = &iter->seq;
2290 struct trace_entry *entry;
2294 if (entry->type == TRACE_CONT)
2295 return TRACE_TYPE_HANDLED;
2297 SEQ_PUT_FIELD_RET(s, entry->pid);
2298 SEQ_PUT_FIELD_RET(s, entry->cpu);
2299 SEQ_PUT_FIELD_RET(s, iter->ts);
2301 switch (entry->type) {
2303 struct ftrace_entry *field;
2305 trace_assign_type(field, entry);
2307 SEQ_PUT_FIELD_RET(s, field->ip);
2308 SEQ_PUT_FIELD_RET(s, field->parent_ip);
2312 struct ctx_switch_entry *field;
2314 trace_assign_type(field, entry);
2316 SEQ_PUT_FIELD_RET(s, field->prev_pid);
2317 SEQ_PUT_FIELD_RET(s, field->prev_prio);
2318 SEQ_PUT_FIELD_RET(s, field->prev_state);
2319 SEQ_PUT_FIELD_RET(s, field->next_pid);
2320 SEQ_PUT_FIELD_RET(s, field->next_prio);
2321 SEQ_PUT_FIELD_RET(s, field->next_state);
2325 case TRACE_USER_STACK:
2327 struct special_entry *field;
2329 trace_assign_type(field, entry);
2331 SEQ_PUT_FIELD_RET(s, field->arg1);
2332 SEQ_PUT_FIELD_RET(s, field->arg2);
2333 SEQ_PUT_FIELD_RET(s, field->arg3);
2340 static int trace_empty(struct trace_iterator *iter)
2344 for_each_tracing_cpu(cpu) {
2345 if (iter->buffer_iter[cpu]) {
2346 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2349 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2357 static enum print_line_t print_trace_line(struct trace_iterator *iter)
2359 enum print_line_t ret;
2361 if (iter->trace && iter->trace->print_line) {
2362 ret = iter->trace->print_line(iter);
2363 if (ret != TRACE_TYPE_UNHANDLED)
2367 if (iter->ent->type == TRACE_PRINT &&
2368 trace_flags & TRACE_ITER_PRINTK &&
2369 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2370 return print_printk_msg_only(iter);
2372 if (trace_flags & TRACE_ITER_BIN)
2373 return print_bin_fmt(iter);
2375 if (trace_flags & TRACE_ITER_HEX)
2376 return print_hex_fmt(iter);
2378 if (trace_flags & TRACE_ITER_RAW)
2379 return print_raw_fmt(iter);
2381 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2382 return print_lat_fmt(iter, iter->idx, iter->cpu);
2384 return print_trace_fmt(iter);
2387 static int s_show(struct seq_file *m, void *v)
2389 struct trace_iterator *iter = v;
2391 if (iter->ent == NULL) {
2393 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2396 if (iter->trace && iter->trace->print_header)
2397 iter->trace->print_header(m);
2398 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2399 /* print nothing if the buffers are empty */
2400 if (trace_empty(iter))
2402 print_trace_header(m, iter);
2403 if (!(trace_flags & TRACE_ITER_VERBOSE))
2404 print_lat_help_header(m);
2406 if (!(trace_flags & TRACE_ITER_VERBOSE))
2407 print_func_help_header(m);
2410 print_trace_line(iter);
2411 trace_print_seq(m, &iter->seq);
2417 static struct seq_operations tracer_seq_ops = {
2424 static struct trace_iterator *
2425 __tracing_open(struct inode *inode, struct file *file, int *ret)
2427 struct trace_iterator *iter;
2431 if (tracing_disabled) {
2436 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2442 mutex_lock(&trace_types_lock);
2443 if (current_trace && current_trace->print_max)
2446 iter->tr = inode->i_private;
2447 iter->trace = current_trace;
2450 /* Notify the tracer early; before we stop tracing. */
2451 if (iter->trace && iter->trace->open)
2452 iter->trace->open(iter);
2454 /* Annotate start of buffers if we had overruns */
2455 if (ring_buffer_overruns(iter->tr->buffer))
2456 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2459 for_each_tracing_cpu(cpu) {
2461 iter->buffer_iter[cpu] =
2462 ring_buffer_read_start(iter->tr->buffer, cpu);
2464 if (!iter->buffer_iter[cpu])
2468 /* TODO stop tracer */
2469 *ret = seq_open(file, &tracer_seq_ops);
2473 m = file->private_data;
2476 /* stop the trace while dumping */
2479 mutex_unlock(&trace_types_lock);
2485 for_each_tracing_cpu(cpu) {
2486 if (iter->buffer_iter[cpu])
2487 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2489 mutex_unlock(&trace_types_lock);
2492 return ERR_PTR(-ENOMEM);
2495 int tracing_open_generic(struct inode *inode, struct file *filp)
2497 if (tracing_disabled)
2500 filp->private_data = inode->i_private;
2504 int tracing_release(struct inode *inode, struct file *file)
2506 struct seq_file *m = (struct seq_file *)file->private_data;
2507 struct trace_iterator *iter = m->private;
2510 mutex_lock(&trace_types_lock);
2511 for_each_tracing_cpu(cpu) {
2512 if (iter->buffer_iter[cpu])
2513 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2516 if (iter->trace && iter->trace->close)
2517 iter->trace->close(iter);
2519 /* reenable tracing if it was previously enabled */
2521 mutex_unlock(&trace_types_lock);
2523 seq_release(inode, file);
2528 static int tracing_open(struct inode *inode, struct file *file)
2532 __tracing_open(inode, file, &ret);
2537 static int tracing_lt_open(struct inode *inode, struct file *file)
2539 struct trace_iterator *iter;
2542 iter = __tracing_open(inode, file, &ret);
2545 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2552 t_next(struct seq_file *m, void *v, loff_t *pos)
2554 struct tracer *t = m->private;
2566 static void *t_start(struct seq_file *m, loff_t *pos)
2568 struct tracer *t = m->private;
2571 mutex_lock(&trace_types_lock);
2572 for (; t && l < *pos; t = t_next(m, t, &l))
2578 static void t_stop(struct seq_file *m, void *p)
2580 mutex_unlock(&trace_types_lock);
2583 static int t_show(struct seq_file *m, void *v)
2585 struct tracer *t = v;
2590 seq_printf(m, "%s", t->name);
2599 static struct seq_operations show_traces_seq_ops = {
2606 static int show_traces_open(struct inode *inode, struct file *file)
2610 if (tracing_disabled)
2613 ret = seq_open(file, &show_traces_seq_ops);
2615 struct seq_file *m = file->private_data;
2616 m->private = trace_types;
2622 static struct file_operations tracing_fops = {
2623 .open = tracing_open,
2625 .llseek = seq_lseek,
2626 .release = tracing_release,
2629 static struct file_operations tracing_lt_fops = {
2630 .open = tracing_lt_open,
2632 .llseek = seq_lseek,
2633 .release = tracing_release,
2636 static struct file_operations show_traces_fops = {
2637 .open = show_traces_open,
2639 .release = seq_release,
2643 * Only trace on a CPU if the bitmask is set:
2645 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
2648 * When tracing/tracing_cpu_mask is modified then this holds
2649 * the new bitmask we are about to install:
2651 static cpumask_t tracing_cpumask_new;
2654 * The tracer itself will not take this lock, but still we want
2655 * to provide a consistent cpumask to user-space:
2657 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2660 * Temporary storage for the character representation of the
2661 * CPU bitmask (and one more byte for the newline):
2663 static char mask_str[NR_CPUS + 1];
2666 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2667 size_t count, loff_t *ppos)
2671 mutex_lock(&tracing_cpumask_update_lock);
2673 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2674 if (count - len < 2) {
2678 len += sprintf(mask_str + len, "\n");
2679 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2682 mutex_unlock(&tracing_cpumask_update_lock);
2688 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2689 size_t count, loff_t *ppos)
2693 mutex_lock(&tracing_cpumask_update_lock);
2694 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2698 local_irq_disable();
2699 __raw_spin_lock(&ftrace_max_lock);
2700 for_each_tracing_cpu(cpu) {
2702 * Increase/decrease the disabled counter if we are
2703 * about to flip a bit in the cpumask:
2705 if (cpu_isset(cpu, tracing_cpumask) &&
2706 !cpu_isset(cpu, tracing_cpumask_new)) {
2707 atomic_inc(&global_trace.data[cpu]->disabled);
2709 if (!cpu_isset(cpu, tracing_cpumask) &&
2710 cpu_isset(cpu, tracing_cpumask_new)) {
2711 atomic_dec(&global_trace.data[cpu]->disabled);
2714 __raw_spin_unlock(&ftrace_max_lock);
2717 tracing_cpumask = tracing_cpumask_new;
2719 mutex_unlock(&tracing_cpumask_update_lock);
2724 mutex_unlock(&tracing_cpumask_update_lock);
2729 static struct file_operations tracing_cpumask_fops = {
2730 .open = tracing_open_generic,
2731 .read = tracing_cpumask_read,
2732 .write = tracing_cpumask_write,
2736 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2737 size_t cnt, loff_t *ppos)
2743 u32 tracer_flags = current_trace->flags->val;
2744 struct tracer_opt *trace_opts = current_trace->flags->opts;
2747 /* calulate max size */
2748 for (i = 0; trace_options[i]; i++) {
2749 len += strlen(trace_options[i]);
2750 len += 3; /* "no" and space */
2754 * Increase the size with names of options specific
2755 * of the current tracer.
2757 for (i = 0; trace_opts[i].name; i++) {
2758 len += strlen(trace_opts[i].name);
2759 len += 3; /* "no" and space */
2762 /* +2 for \n and \0 */
2763 buf = kmalloc(len + 2, GFP_KERNEL);
2767 for (i = 0; trace_options[i]; i++) {
2768 if (trace_flags & (1 << i))
2769 r += sprintf(buf + r, "%s ", trace_options[i]);
2771 r += sprintf(buf + r, "no%s ", trace_options[i]);
2774 for (i = 0; trace_opts[i].name; i++) {
2775 if (tracer_flags & trace_opts[i].bit)
2776 r += sprintf(buf + r, "%s ",
2777 trace_opts[i].name);
2779 r += sprintf(buf + r, "no%s ",
2780 trace_opts[i].name);
2783 r += sprintf(buf + r, "\n");
2784 WARN_ON(r >= len + 2);
2786 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2793 /* Try to assign a tracer specific option */
2794 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2796 struct tracer_flags *trace_flags = trace->flags;
2797 struct tracer_opt *opts = NULL;
2801 for (i = 0; trace_flags->opts[i].name; i++) {
2802 opts = &trace_flags->opts[i];
2803 len = strlen(opts->name);
2805 if (strncmp(cmp, opts->name, len) == 0) {
2806 ret = trace->set_flag(trace_flags->val,
2812 if (!trace_flags->opts[i].name)
2815 /* Refused to handle */
2820 trace_flags->val &= ~opts->bit;
2822 trace_flags->val |= opts->bit;
2828 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2829 size_t cnt, loff_t *ppos)
2837 if (cnt >= sizeof(buf))
2840 if (copy_from_user(&buf, ubuf, cnt))
2845 if (strncmp(buf, "no", 2) == 0) {
2850 for (i = 0; trace_options[i]; i++) {
2851 int len = strlen(trace_options[i]);
2853 if (strncmp(cmp, trace_options[i], len) == 0) {
2855 trace_flags &= ~(1 << i);
2857 trace_flags |= (1 << i);
2862 /* If no option could be set, test the specific tracer options */
2863 if (!trace_options[i]) {
2864 ret = set_tracer_option(current_trace, cmp, neg);
2874 static struct file_operations tracing_iter_fops = {
2875 .open = tracing_open_generic,
2876 .read = tracing_trace_options_read,
2877 .write = tracing_trace_options_write,
2880 static const char readme_msg[] =
2881 "tracing mini-HOWTO:\n\n"
2883 "# mount -t debugfs nodev /debug\n\n"
2884 "# cat /debug/tracing/available_tracers\n"
2885 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2886 "# cat /debug/tracing/current_tracer\n"
2888 "# echo sched_switch > /debug/tracing/current_tracer\n"
2889 "# cat /debug/tracing/current_tracer\n"
2891 "# cat /debug/tracing/trace_options\n"
2892 "noprint-parent nosym-offset nosym-addr noverbose\n"
2893 "# echo print-parent > /debug/tracing/trace_options\n"
2894 "# echo 1 > /debug/tracing/tracing_enabled\n"
2895 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2896 "echo 0 > /debug/tracing/tracing_enabled\n"
2900 tracing_readme_read(struct file *filp, char __user *ubuf,
2901 size_t cnt, loff_t *ppos)
2903 return simple_read_from_buffer(ubuf, cnt, ppos,
2904 readme_msg, strlen(readme_msg));
2907 static struct file_operations tracing_readme_fops = {
2908 .open = tracing_open_generic,
2909 .read = tracing_readme_read,
2913 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2914 size_t cnt, loff_t *ppos)
2919 r = sprintf(buf, "%u\n", tracer_enabled);
2920 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2924 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2925 size_t cnt, loff_t *ppos)
2927 struct trace_array *tr = filp->private_data;
2932 if (cnt >= sizeof(buf))
2935 if (copy_from_user(&buf, ubuf, cnt))
2940 ret = strict_strtoul(buf, 10, &val);
2946 mutex_lock(&trace_types_lock);
2947 if (tracer_enabled ^ val) {
2950 if (current_trace->start)
2951 current_trace->start(tr);
2956 if (current_trace->stop)
2957 current_trace->stop(tr);
2960 mutex_unlock(&trace_types_lock);
2968 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2969 size_t cnt, loff_t *ppos)
2971 char buf[max_tracer_type_len+2];
2974 mutex_lock(&trace_types_lock);
2976 r = sprintf(buf, "%s\n", current_trace->name);
2978 r = sprintf(buf, "\n");
2979 mutex_unlock(&trace_types_lock);
2981 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2984 static int tracing_set_tracer(char *buf)
2986 struct trace_array *tr = &global_trace;
2990 mutex_lock(&trace_types_lock);
2991 for (t = trace_types; t; t = t->next) {
2992 if (strcmp(t->name, buf) == 0)
2999 if (t == current_trace)
3002 trace_branch_disable();
3003 if (current_trace && current_trace->reset)
3004 current_trace->reset(tr);
3013 trace_branch_enable(tr);
3015 mutex_unlock(&trace_types_lock);
3021 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3022 size_t cnt, loff_t *ppos)
3024 char buf[max_tracer_type_len+1];
3031 if (cnt > max_tracer_type_len)
3032 cnt = max_tracer_type_len;
3034 if (copy_from_user(&buf, ubuf, cnt))
3039 /* strip ending whitespace. */
3040 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3043 err = tracing_set_tracer(buf);
3053 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3054 size_t cnt, loff_t *ppos)
3056 unsigned long *ptr = filp->private_data;
3060 r = snprintf(buf, sizeof(buf), "%ld\n",
3061 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3062 if (r > sizeof(buf))
3064 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3068 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3069 size_t cnt, loff_t *ppos)
3071 long *ptr = filp->private_data;
3076 if (cnt >= sizeof(buf))
3079 if (copy_from_user(&buf, ubuf, cnt))
3084 ret = strict_strtoul(buf, 10, &val);
3093 static atomic_t tracing_reader;
3095 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3097 struct trace_iterator *iter;
3099 if (tracing_disabled)
3102 /* We only allow for reader of the pipe */
3103 if (atomic_inc_return(&tracing_reader) != 1) {
3104 atomic_dec(&tracing_reader);
3108 /* create a buffer to store the information to pass to userspace */
3109 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3113 mutex_lock(&trace_types_lock);
3115 /* trace pipe does not show start of buffer */
3116 cpus_setall(iter->started);
3118 iter->tr = &global_trace;
3119 iter->trace = current_trace;
3120 filp->private_data = iter;
3122 if (iter->trace->pipe_open)
3123 iter->trace->pipe_open(iter);
3124 mutex_unlock(&trace_types_lock);
3129 static int tracing_release_pipe(struct inode *inode, struct file *file)
3131 struct trace_iterator *iter = file->private_data;
3134 atomic_dec(&tracing_reader);
3140 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3142 struct trace_iterator *iter = filp->private_data;
3144 if (trace_flags & TRACE_ITER_BLOCK) {
3146 * Always select as readable when in blocking mode
3148 return POLLIN | POLLRDNORM;
3150 if (!trace_empty(iter))
3151 return POLLIN | POLLRDNORM;
3152 poll_wait(filp, &trace_wait, poll_table);
3153 if (!trace_empty(iter))
3154 return POLLIN | POLLRDNORM;
3164 tracing_read_pipe(struct file *filp, char __user *ubuf,
3165 size_t cnt, loff_t *ppos)
3167 struct trace_iterator *iter = filp->private_data;
3170 /* return any leftover data */
3171 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3175 trace_seq_reset(&iter->seq);
3177 mutex_lock(&trace_types_lock);
3178 if (iter->trace->read) {
3179 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3186 while (trace_empty(iter)) {
3188 if ((filp->f_flags & O_NONBLOCK)) {
3194 * This is a make-shift waitqueue. The reason we don't use
3195 * an actual wait queue is because:
3196 * 1) we only ever have one waiter
3197 * 2) the tracing, traces all functions, we don't want
3198 * the overhead of calling wake_up and friends
3199 * (and tracing them too)
3200 * Anyway, this is really very primitive wakeup.
3202 set_current_state(TASK_INTERRUPTIBLE);
3203 iter->tr->waiter = current;
3205 mutex_unlock(&trace_types_lock);
3207 /* sleep for 100 msecs, and try again. */
3208 schedule_timeout(HZ/10);
3210 mutex_lock(&trace_types_lock);
3212 iter->tr->waiter = NULL;
3214 if (signal_pending(current)) {
3219 if (iter->trace != current_trace)
3223 * We block until we read something and tracing is disabled.
3224 * We still block if tracing is disabled, but we have never
3225 * read anything. This allows a user to cat this file, and
3226 * then enable tracing. But after we have read something,
3227 * we give an EOF when tracing is again disabled.
3229 * iter->pos will be 0 if we haven't read anything.
3231 if (!tracer_enabled && iter->pos)
3237 /* stop when tracing is finished */
3238 if (trace_empty(iter))
3241 if (cnt >= PAGE_SIZE)
3242 cnt = PAGE_SIZE - 1;
3244 /* reset all but tr, trace, and overruns */
3245 memset(&iter->seq, 0,
3246 sizeof(struct trace_iterator) -
3247 offsetof(struct trace_iterator, seq));
3250 while (find_next_entry_inc(iter) != NULL) {
3251 enum print_line_t ret;
3252 int len = iter->seq.len;
3254 ret = print_trace_line(iter);
3255 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3256 /* don't print partial lines */
3257 iter->seq.len = len;
3261 trace_consume(iter);
3263 if (iter->seq.len >= cnt)
3267 /* Now copy what we have to the user */
3268 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3269 if (iter->seq.readpos >= iter->seq.len)
3270 trace_seq_reset(&iter->seq);
3273 * If there was nothing to send to user, inspite of consuming trace
3274 * entries, go back to wait for more entries.
3280 mutex_unlock(&trace_types_lock);
3286 tracing_entries_read(struct file *filp, char __user *ubuf,
3287 size_t cnt, loff_t *ppos)
3289 struct trace_array *tr = filp->private_data;
3293 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3294 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3298 tracing_entries_write(struct file *filp, const char __user *ubuf,
3299 size_t cnt, loff_t *ppos)
3305 if (cnt >= sizeof(buf))
3308 if (copy_from_user(&buf, ubuf, cnt))
3313 ret = strict_strtoul(buf, 10, &val);
3317 /* must have at least 1 entry */
3321 mutex_lock(&trace_types_lock);
3325 /* disable all cpu buffers */
3326 for_each_tracing_cpu(cpu) {
3327 if (global_trace.data[cpu])
3328 atomic_inc(&global_trace.data[cpu]->disabled);
3329 if (max_tr.data[cpu])
3330 atomic_inc(&max_tr.data[cpu]->disabled);
3333 /* value is in KB */
3336 if (val != global_trace.entries) {
3337 ret = ring_buffer_resize(global_trace.buffer, val);
3343 ret = ring_buffer_resize(max_tr.buffer, val);
3347 r = ring_buffer_resize(global_trace.buffer,
3348 global_trace.entries);
3350 /* AARGH! We are left with different
3351 * size max buffer!!!! */
3353 tracing_disabled = 1;
3358 global_trace.entries = val;
3363 /* If check pages failed, return ENOMEM */
3364 if (tracing_disabled)
3367 for_each_tracing_cpu(cpu) {
3368 if (global_trace.data[cpu])
3369 atomic_dec(&global_trace.data[cpu]->disabled);
3370 if (max_tr.data[cpu])
3371 atomic_dec(&max_tr.data[cpu]->disabled);
3375 max_tr.entries = global_trace.entries;
3376 mutex_unlock(&trace_types_lock);
3381 static int mark_printk(const char *fmt, ...)
3385 va_start(args, fmt);
3386 ret = trace_vprintk(0, -1, fmt, args);
3392 tracing_mark_write(struct file *filp, const char __user *ubuf,
3393 size_t cnt, loff_t *fpos)
3398 if (tracing_disabled)
3401 if (cnt > TRACE_BUF_SIZE)
3402 cnt = TRACE_BUF_SIZE;
3404 buf = kmalloc(cnt + 1, GFP_KERNEL);
3408 if (copy_from_user(buf, ubuf, cnt)) {
3413 /* Cut from the first nil or newline. */
3415 end = strchr(buf, '\n');
3419 cnt = mark_printk("%s\n", buf);
3426 static struct file_operations tracing_max_lat_fops = {
3427 .open = tracing_open_generic,
3428 .read = tracing_max_lat_read,
3429 .write = tracing_max_lat_write,
3432 static struct file_operations tracing_ctrl_fops = {
3433 .open = tracing_open_generic,
3434 .read = tracing_ctrl_read,
3435 .write = tracing_ctrl_write,
3438 static struct file_operations set_tracer_fops = {
3439 .open = tracing_open_generic,
3440 .read = tracing_set_trace_read,
3441 .write = tracing_set_trace_write,
3444 static struct file_operations tracing_pipe_fops = {
3445 .open = tracing_open_pipe,
3446 .poll = tracing_poll_pipe,
3447 .read = tracing_read_pipe,
3448 .release = tracing_release_pipe,
3451 static struct file_operations tracing_entries_fops = {
3452 .open = tracing_open_generic,
3453 .read = tracing_entries_read,
3454 .write = tracing_entries_write,
3457 static struct file_operations tracing_mark_fops = {
3458 .open = tracing_open_generic,
3459 .write = tracing_mark_write,
3462 #ifdef CONFIG_DYNAMIC_FTRACE
3464 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3470 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3471 size_t cnt, loff_t *ppos)
3473 static char ftrace_dyn_info_buffer[1024];
3474 static DEFINE_MUTEX(dyn_info_mutex);
3475 unsigned long *p = filp->private_data;
3476 char *buf = ftrace_dyn_info_buffer;
3477 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3480 mutex_lock(&dyn_info_mutex);
3481 r = sprintf(buf, "%ld ", *p);
3483 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3486 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3488 mutex_unlock(&dyn_info_mutex);
3493 static struct file_operations tracing_dyn_info_fops = {
3494 .open = tracing_open_generic,
3495 .read = tracing_read_dyn_info,
3499 static struct dentry *d_tracer;
3501 struct dentry *tracing_init_dentry(void)
3508 d_tracer = debugfs_create_dir("tracing", NULL);
3510 if (!d_tracer && !once) {
3512 pr_warning("Could not create debugfs directory 'tracing'\n");
3519 #ifdef CONFIG_FTRACE_SELFTEST
3520 /* Let selftest have access to static functions in this file */
3521 #include "trace_selftest.c"
3524 static __init int tracer_init_debugfs(void)
3526 struct dentry *d_tracer;
3527 struct dentry *entry;
3529 d_tracer = tracing_init_dentry();
3531 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3532 &global_trace, &tracing_ctrl_fops);
3534 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3536 entry = debugfs_create_file("trace_options", 0644, d_tracer,
3537 NULL, &tracing_iter_fops);
3539 pr_warning("Could not create debugfs 'trace_options' entry\n");
3541 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3542 NULL, &tracing_cpumask_fops);
3544 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3546 entry = debugfs_create_file("latency_trace", 0444, d_tracer,
3547 &global_trace, &tracing_lt_fops);
3549 pr_warning("Could not create debugfs 'latency_trace' entry\n");
3551 entry = debugfs_create_file("trace", 0444, d_tracer,
3552 &global_trace, &tracing_fops);
3554 pr_warning("Could not create debugfs 'trace' entry\n");
3556 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3557 &global_trace, &show_traces_fops);
3559 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3561 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3562 &global_trace, &set_tracer_fops);
3564 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3566 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3567 &tracing_max_latency,
3568 &tracing_max_lat_fops);
3570 pr_warning("Could not create debugfs "
3571 "'tracing_max_latency' entry\n");
3573 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3574 &tracing_thresh, &tracing_max_lat_fops);
3576 pr_warning("Could not create debugfs "
3577 "'tracing_thresh' entry\n");
3578 entry = debugfs_create_file("README", 0644, d_tracer,
3579 NULL, &tracing_readme_fops);
3581 pr_warning("Could not create debugfs 'README' entry\n");
3583 entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
3584 NULL, &tracing_pipe_fops);
3586 pr_warning("Could not create debugfs "
3587 "'trace_pipe' entry\n");
3589 entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
3590 &global_trace, &tracing_entries_fops);
3592 pr_warning("Could not create debugfs "
3593 "'buffer_size_kb' entry\n");
3595 entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3596 NULL, &tracing_mark_fops);
3598 pr_warning("Could not create debugfs "
3599 "'trace_marker' entry\n");
3601 #ifdef CONFIG_DYNAMIC_FTRACE
3602 entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3603 &ftrace_update_tot_cnt,
3604 &tracing_dyn_info_fops);
3606 pr_warning("Could not create debugfs "
3607 "'dyn_ftrace_total_info' entry\n");
3609 #ifdef CONFIG_SYSPROF_TRACER
3610 init_tracer_sysprof_debugfs(d_tracer);
3615 int trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args)
3617 static DEFINE_SPINLOCK(trace_buf_lock);
3618 static char trace_buf[TRACE_BUF_SIZE];
3620 struct ring_buffer_event *event;
3621 struct trace_array *tr = &global_trace;
3622 struct trace_array_cpu *data;
3623 int cpu, len = 0, size, pc;
3624 struct print_entry *entry;
3625 unsigned long irq_flags;
3627 if (tracing_disabled || tracing_selftest_running)
3630 pc = preempt_count();
3631 preempt_disable_notrace();
3632 cpu = raw_smp_processor_id();
3633 data = tr->data[cpu];
3635 if (unlikely(atomic_read(&data->disabled)))
3638 pause_graph_tracing();
3639 spin_lock_irqsave(&trace_buf_lock, irq_flags);
3640 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
3642 len = min(len, TRACE_BUF_SIZE-1);
3645 size = sizeof(*entry) + len + 1;
3646 event = ring_buffer_lock_reserve(tr->buffer, size, &irq_flags);
3649 entry = ring_buffer_event_data(event);
3650 tracing_generic_entry_update(&entry->ent, irq_flags, pc);
3651 entry->ent.type = TRACE_PRINT;
3653 entry->depth = depth;
3655 memcpy(&entry->buf, trace_buf, len);
3656 entry->buf[len] = 0;
3657 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
3660 spin_unlock_irqrestore(&trace_buf_lock, irq_flags);
3661 unpause_graph_tracing();
3663 preempt_enable_notrace();
3667 EXPORT_SYMBOL_GPL(trace_vprintk);
3669 int __ftrace_printk(unsigned long ip, const char *fmt, ...)
3674 if (!(trace_flags & TRACE_ITER_PRINTK))
3678 ret = trace_vprintk(ip, task_curr_ret_stack(current), fmt, ap);
3682 EXPORT_SYMBOL_GPL(__ftrace_printk);
3684 static int trace_panic_handler(struct notifier_block *this,
3685 unsigned long event, void *unused)
3687 if (ftrace_dump_on_oops)
3692 static struct notifier_block trace_panic_notifier = {
3693 .notifier_call = trace_panic_handler,
3695 .priority = 150 /* priority: INT_MAX >= x >= 0 */
3698 static int trace_die_handler(struct notifier_block *self,
3704 if (ftrace_dump_on_oops)
3713 static struct notifier_block trace_die_notifier = {
3714 .notifier_call = trace_die_handler,
3719 * printk is set to max of 1024, we really don't need it that big.
3720 * Nothing should be printing 1000 characters anyway.
3722 #define TRACE_MAX_PRINT 1000
3725 * Define here KERN_TRACE so that we have one place to modify
3726 * it if we decide to change what log level the ftrace dump
3729 #define KERN_TRACE KERN_INFO
3732 trace_printk_seq(struct trace_seq *s)
3734 /* Probably should print a warning here. */
3738 /* should be zero ended, but we are paranoid. */
3739 s->buffer[s->len] = 0;
3741 printk(KERN_TRACE "%s", s->buffer);
3746 void ftrace_dump(void)
3748 static DEFINE_SPINLOCK(ftrace_dump_lock);
3749 /* use static because iter can be a bit big for the stack */
3750 static struct trace_iterator iter;
3751 static cpumask_t mask;
3752 static int dump_ran;
3753 unsigned long flags;
3757 spin_lock_irqsave(&ftrace_dump_lock, flags);
3763 /* No turning back! */
3766 for_each_tracing_cpu(cpu) {
3767 atomic_inc(&global_trace.data[cpu]->disabled);
3770 /* don't look at user memory in panic mode */
3771 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
3773 printk(KERN_TRACE "Dumping ftrace buffer:\n");
3775 iter.tr = &global_trace;
3776 iter.trace = current_trace;
3779 * We need to stop all tracing on all CPUS to read the
3780 * the next buffer. This is a bit expensive, but is
3781 * not done often. We fill all what we can read,
3782 * and then release the locks again.
3787 while (!trace_empty(&iter)) {
3790 printk(KERN_TRACE "---------------------------------\n");
3794 /* reset all but tr, trace, and overruns */
3795 memset(&iter.seq, 0,
3796 sizeof(struct trace_iterator) -
3797 offsetof(struct trace_iterator, seq));
3798 iter.iter_flags |= TRACE_FILE_LAT_FMT;
3801 if (find_next_entry_inc(&iter) != NULL) {
3802 print_trace_line(&iter);
3803 trace_consume(&iter);
3806 trace_printk_seq(&iter.seq);
3810 printk(KERN_TRACE " (ftrace buffer empty)\n");
3812 printk(KERN_TRACE "---------------------------------\n");
3815 spin_unlock_irqrestore(&ftrace_dump_lock, flags);
3818 __init static int tracer_alloc_buffers(void)
3820 struct trace_array_cpu *data;
3823 /* TODO: make the number of buffers hot pluggable with CPUS */
3824 tracing_buffer_mask = cpu_possible_map;
3826 global_trace.buffer = ring_buffer_alloc(trace_buf_size,
3827 TRACE_BUFFER_FLAGS);
3828 if (!global_trace.buffer) {
3829 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
3833 global_trace.entries = ring_buffer_size(global_trace.buffer);
3835 #ifdef CONFIG_TRACER_MAX_TRACE
3836 max_tr.buffer = ring_buffer_alloc(trace_buf_size,
3837 TRACE_BUFFER_FLAGS);
3838 if (!max_tr.buffer) {
3839 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
3841 ring_buffer_free(global_trace.buffer);
3844 max_tr.entries = ring_buffer_size(max_tr.buffer);
3845 WARN_ON(max_tr.entries != global_trace.entries);
3848 /* Allocate the first page for all buffers */
3849 for_each_tracing_cpu(i) {
3850 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
3851 max_tr.data[i] = &per_cpu(max_data, i);
3854 trace_init_cmdlines();
3856 register_tracer(&nop_trace);
3857 #ifdef CONFIG_BOOT_TRACER
3858 register_tracer(&boot_tracer);
3859 current_trace = &boot_tracer;
3860 current_trace->init(&global_trace);
3862 current_trace = &nop_trace;
3865 /* All seems OK, enable tracing */
3866 tracing_disabled = 0;
3868 atomic_notifier_chain_register(&panic_notifier_list,
3869 &trace_panic_notifier);
3871 register_die_notifier(&trace_die_notifier);
3875 early_initcall(tracer_alloc_buffers);
3876 fs_initcall(tracer_init_debugfs);