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/ring_buffer.h>
15 #include <generated/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/notifier.h>
21 #include <linux/irqflags.h>
22 #include <linux/debugfs.h>
23 #include <linux/pagemap.h>
24 #include <linux/hardirq.h>
25 #include <linux/linkage.h>
26 #include <linux/uaccess.h>
27 #include <linux/kprobes.h>
28 #include <linux/ftrace.h>
29 #include <linux/module.h>
30 #include <linux/percpu.h>
31 #include <linux/splice.h>
32 #include <linux/kdebug.h>
33 #include <linux/string.h>
34 #include <linux/rwsem.h>
35 #include <linux/slab.h>
36 #include <linux/ctype.h>
37 #include <linux/init.h>
38 #include <linux/poll.h>
39 #include <linux/nmi.h>
43 #include "trace_output.h"
46 * On boot up, the ring buffer is set to the minimum size, so that
47 * we do not waste memory on systems that are not using tracing.
49 int ring_buffer_expanded;
52 * We need to change this state when a selftest is running.
53 * A selftest will lurk into the ring-buffer to count the
54 * entries inserted during the selftest although some concurrent
55 * insertions into the ring-buffer such as trace_printk could occurred
56 * at the same time, giving false positive or negative results.
58 static bool __read_mostly tracing_selftest_running;
61 * If a tracer is running, we do not want to run SELFTEST.
63 bool __read_mostly tracing_selftest_disabled;
65 /* For tracers that don't implement custom flags */
66 static struct tracer_opt dummy_tracer_opt[] = {
70 static struct tracer_flags dummy_tracer_flags = {
72 .opts = dummy_tracer_opt
75 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
81 * Kill all tracing for good (never come back).
82 * It is initialized to 1 but will turn to zero if the initialization
83 * of the tracer is successful. But that is the only place that sets
86 static int tracing_disabled = 1;
88 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
90 cpumask_var_t __read_mostly tracing_buffer_mask;
93 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
95 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
96 * is set, then ftrace_dump is called. This will output the contents
97 * of the ftrace buffers to the console. This is very useful for
98 * capturing traces that lead to crashes and outputing it to a
101 * It is default off, but you can enable it with either specifying
102 * "ftrace_dump_on_oops" in the kernel command line, or setting
103 * /proc/sys/kernel/ftrace_dump_on_oops
104 * Set 1 if you want to dump buffers of all CPUs
105 * Set 2 if you want to dump the buffer of the CPU that triggered oops
108 enum ftrace_dump_mode ftrace_dump_on_oops;
110 static int tracing_set_tracer(const char *buf);
112 #define MAX_TRACER_SIZE 100
113 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
114 static char *default_bootup_tracer;
116 static int __init set_cmdline_ftrace(char *str)
118 strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
119 default_bootup_tracer = bootup_tracer_buf;
120 /* We are using ftrace early, expand it */
121 ring_buffer_expanded = 1;
124 __setup("ftrace=", set_cmdline_ftrace);
126 static int __init set_ftrace_dump_on_oops(char *str)
128 if (*str++ != '=' || !*str) {
129 ftrace_dump_on_oops = DUMP_ALL;
133 if (!strcmp("orig_cpu", str)) {
134 ftrace_dump_on_oops = DUMP_ORIG;
140 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
142 unsigned long long ns2usecs(cycle_t nsec)
150 * The global_trace is the descriptor that holds the tracing
151 * buffers for the live tracing. For each CPU, it contains
152 * a link list of pages that will store trace entries. The
153 * page descriptor of the pages in the memory is used to hold
154 * the link list by linking the lru item in the page descriptor
155 * to each of the pages in the buffer per CPU.
157 * For each active CPU there is a data field that holds the
158 * pages for the buffer for that CPU. Each CPU has the same number
159 * of pages allocated for its buffer.
161 static struct trace_array global_trace;
163 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
165 int filter_current_check_discard(struct ring_buffer *buffer,
166 struct ftrace_event_call *call, void *rec,
167 struct ring_buffer_event *event)
169 return filter_check_discard(call, rec, buffer, event);
171 EXPORT_SYMBOL_GPL(filter_current_check_discard);
173 cycle_t ftrace_now(int cpu)
177 /* Early boot up does not have a buffer yet */
178 if (!global_trace.buffer)
179 return trace_clock_local();
181 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
182 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
188 * The max_tr is used to snapshot the global_trace when a maximum
189 * latency is reached. Some tracers will use this to store a maximum
190 * trace while it continues examining live traces.
192 * The buffers for the max_tr are set up the same as the global_trace.
193 * When a snapshot is taken, the link list of the max_tr is swapped
194 * with the link list of the global_trace and the buffers are reset for
195 * the global_trace so the tracing can continue.
197 static struct trace_array max_tr;
199 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
201 /* tracer_enabled is used to toggle activation of a tracer */
202 static int tracer_enabled = 1;
205 * tracing_is_enabled - return tracer_enabled status
207 * This function is used by other tracers to know the status
208 * of the tracer_enabled flag. Tracers may use this function
209 * to know if it should enable their features when starting
210 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
212 int tracing_is_enabled(void)
214 return tracer_enabled;
218 * trace_buf_size is the size in bytes that is allocated
219 * for a buffer. Note, the number of bytes is always rounded
222 * This number is purposely set to a low number of 16384.
223 * If the dump on oops happens, it will be much appreciated
224 * to not have to wait for all that output. Anyway this can be
225 * boot time and run time configurable.
227 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
229 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
231 /* trace_types holds a link list of available tracers. */
232 static struct tracer *trace_types __read_mostly;
234 /* current_trace points to the tracer that is currently active */
235 static struct tracer *current_trace __read_mostly;
238 * trace_types_lock is used to protect the trace_types list.
240 static DEFINE_MUTEX(trace_types_lock);
243 * serialize the access of the ring buffer
245 * ring buffer serializes readers, but it is low level protection.
246 * The validity of the events (which returns by ring_buffer_peek() ..etc)
247 * are not protected by ring buffer.
249 * The content of events may become garbage if we allow other process consumes
250 * these events concurrently:
251 * A) the page of the consumed events may become a normal page
252 * (not reader page) in ring buffer, and this page will be rewrited
253 * by events producer.
254 * B) The page of the consumed events may become a page for splice_read,
255 * and this page will be returned to system.
257 * These primitives allow multi process access to different cpu ring buffer
260 * These primitives don't distinguish read-only and read-consume access.
261 * Multi read-only access are also serialized.
265 static DECLARE_RWSEM(all_cpu_access_lock);
266 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
268 static inline void trace_access_lock(int cpu)
270 if (cpu == TRACE_PIPE_ALL_CPU) {
271 /* gain it for accessing the whole ring buffer. */
272 down_write(&all_cpu_access_lock);
274 /* gain it for accessing a cpu ring buffer. */
276 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
277 down_read(&all_cpu_access_lock);
279 /* Secondly block other access to this @cpu ring buffer. */
280 mutex_lock(&per_cpu(cpu_access_lock, cpu));
284 static inline void trace_access_unlock(int cpu)
286 if (cpu == TRACE_PIPE_ALL_CPU) {
287 up_write(&all_cpu_access_lock);
289 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
290 up_read(&all_cpu_access_lock);
294 static inline void trace_access_lock_init(void)
298 for_each_possible_cpu(cpu)
299 mutex_init(&per_cpu(cpu_access_lock, cpu));
304 static DEFINE_MUTEX(access_lock);
306 static inline void trace_access_lock(int cpu)
309 mutex_lock(&access_lock);
312 static inline void trace_access_unlock(int cpu)
315 mutex_unlock(&access_lock);
318 static inline void trace_access_lock_init(void)
324 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
325 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
327 /* trace_flags holds trace_options default values */
328 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
329 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
330 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE |
333 static int trace_stop_count;
334 static DEFINE_RAW_SPINLOCK(tracing_start_lock);
336 static void wakeup_work_handler(struct work_struct *work)
338 wake_up(&trace_wait);
341 static DECLARE_DELAYED_WORK(wakeup_work, wakeup_work_handler);
344 * tracing_on - enable tracing buffers
346 * This function enables tracing buffers that may have been
347 * disabled with tracing_off.
349 void tracing_on(void)
351 if (global_trace.buffer)
352 ring_buffer_record_on(global_trace.buffer);
354 * This flag is only looked at when buffers haven't been
355 * allocated yet. We don't really care about the race
356 * between setting this flag and actually turning
359 global_trace.buffer_disabled = 0;
361 EXPORT_SYMBOL_GPL(tracing_on);
364 * tracing_off - turn off tracing buffers
366 * This function stops the tracing buffers from recording data.
367 * It does not disable any overhead the tracers themselves may
368 * be causing. This function simply causes all recording to
369 * the ring buffers to fail.
371 void tracing_off(void)
373 if (global_trace.buffer)
374 ring_buffer_record_on(global_trace.buffer);
376 * This flag is only looked at when buffers haven't been
377 * allocated yet. We don't really care about the race
378 * between setting this flag and actually turning
381 global_trace.buffer_disabled = 1;
383 EXPORT_SYMBOL_GPL(tracing_off);
386 * tracing_is_on - show state of ring buffers enabled
388 int tracing_is_on(void)
390 if (global_trace.buffer)
391 return ring_buffer_record_is_on(global_trace.buffer);
392 return !global_trace.buffer_disabled;
394 EXPORT_SYMBOL_GPL(tracing_is_on);
397 * trace_wake_up - wake up tasks waiting for trace input
399 * Schedules a delayed work to wake up any task that is blocked on the
400 * trace_wait queue. These is used with trace_poll for tasks polling the
403 void trace_wake_up(void)
405 const unsigned long delay = msecs_to_jiffies(2);
407 if (trace_flags & TRACE_ITER_BLOCK)
409 schedule_delayed_work(&wakeup_work, delay);
412 static int __init set_buf_size(char *str)
414 unsigned long buf_size;
418 buf_size = memparse(str, &str);
419 /* nr_entries can not be zero */
422 trace_buf_size = buf_size;
425 __setup("trace_buf_size=", set_buf_size);
427 static int __init set_tracing_thresh(char *str)
429 unsigned long threshhold;
434 ret = strict_strtoul(str, 0, &threshhold);
437 tracing_thresh = threshhold * 1000;
440 __setup("tracing_thresh=", set_tracing_thresh);
442 unsigned long nsecs_to_usecs(unsigned long nsecs)
447 /* These must match the bit postions in trace_iterator_flags */
448 static const char *trace_options[] = {
480 { trace_clock_local, "local" },
481 { trace_clock_global, "global" },
482 { trace_clock_counter, "counter" },
488 * trace_parser_get_init - gets the buffer for trace parser
490 int trace_parser_get_init(struct trace_parser *parser, int size)
492 memset(parser, 0, sizeof(*parser));
494 parser->buffer = kmalloc(size, GFP_KERNEL);
503 * trace_parser_put - frees the buffer for trace parser
505 void trace_parser_put(struct trace_parser *parser)
507 kfree(parser->buffer);
511 * trace_get_user - reads the user input string separated by space
512 * (matched by isspace(ch))
514 * For each string found the 'struct trace_parser' is updated,
515 * and the function returns.
517 * Returns number of bytes read.
519 * See kernel/trace/trace.h for 'struct trace_parser' details.
521 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
522 size_t cnt, loff_t *ppos)
529 trace_parser_clear(parser);
531 ret = get_user(ch, ubuf++);
539 * The parser is not finished with the last write,
540 * continue reading the user input without skipping spaces.
543 /* skip white space */
544 while (cnt && isspace(ch)) {
545 ret = get_user(ch, ubuf++);
552 /* only spaces were written */
562 /* read the non-space input */
563 while (cnt && !isspace(ch)) {
564 if (parser->idx < parser->size - 1)
565 parser->buffer[parser->idx++] = ch;
570 ret = get_user(ch, ubuf++);
577 /* We either got finished input or we have to wait for another call. */
579 parser->buffer[parser->idx] = 0;
580 parser->cont = false;
583 parser->buffer[parser->idx++] = ch;
593 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
601 if (s->len <= s->readpos)
604 len = s->len - s->readpos;
607 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
617 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
621 if (s->len <= s->readpos)
624 len = s->len - s->readpos;
627 memcpy(buf, s->buffer + s->readpos, cnt);
634 * ftrace_max_lock is used to protect the swapping of buffers
635 * when taking a max snapshot. The buffers themselves are
636 * protected by per_cpu spinlocks. But the action of the swap
637 * needs its own lock.
639 * This is defined as a arch_spinlock_t in order to help
640 * with performance when lockdep debugging is enabled.
642 * It is also used in other places outside the update_max_tr
643 * so it needs to be defined outside of the
644 * CONFIG_TRACER_MAX_TRACE.
646 static arch_spinlock_t ftrace_max_lock =
647 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
649 unsigned long __read_mostly tracing_thresh;
651 #ifdef CONFIG_TRACER_MAX_TRACE
652 unsigned long __read_mostly tracing_max_latency;
655 * Copy the new maximum trace into the separate maximum-trace
656 * structure. (this way the maximum trace is permanently saved,
657 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
660 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
662 struct trace_array_cpu *data = tr->data[cpu];
663 struct trace_array_cpu *max_data;
666 max_tr.time_start = data->preempt_timestamp;
668 max_data = max_tr.data[cpu];
669 max_data->saved_latency = tracing_max_latency;
670 max_data->critical_start = data->critical_start;
671 max_data->critical_end = data->critical_end;
673 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
674 max_data->pid = tsk->pid;
675 max_data->uid = task_uid(tsk);
676 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
677 max_data->policy = tsk->policy;
678 max_data->rt_priority = tsk->rt_priority;
680 /* record this tasks comm */
681 tracing_record_cmdline(tsk);
685 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
687 * @tsk: the task with the latency
688 * @cpu: The cpu that initiated the trace.
690 * Flip the buffers between the @tr and the max_tr and record information
691 * about which task was the cause of this latency.
694 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
696 struct ring_buffer *buf = tr->buffer;
698 if (trace_stop_count)
701 WARN_ON_ONCE(!irqs_disabled());
702 if (!current_trace->use_max_tr) {
706 arch_spin_lock(&ftrace_max_lock);
708 tr->buffer = max_tr.buffer;
711 __update_max_tr(tr, tsk, cpu);
712 arch_spin_unlock(&ftrace_max_lock);
716 * update_max_tr_single - only copy one trace over, and reset the rest
718 * @tsk - task with the latency
719 * @cpu - the cpu of the buffer to copy.
721 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
724 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
728 if (trace_stop_count)
731 WARN_ON_ONCE(!irqs_disabled());
732 if (!current_trace->use_max_tr) {
737 arch_spin_lock(&ftrace_max_lock);
739 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
743 * We failed to swap the buffer due to a commit taking
744 * place on this CPU. We fail to record, but we reset
745 * the max trace buffer (no one writes directly to it)
746 * and flag that it failed.
748 trace_array_printk(&max_tr, _THIS_IP_,
749 "Failed to swap buffers due to commit in progress\n");
752 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
754 __update_max_tr(tr, tsk, cpu);
755 arch_spin_unlock(&ftrace_max_lock);
757 #endif /* CONFIG_TRACER_MAX_TRACE */
760 * register_tracer - register a tracer with the ftrace system.
761 * @type - the plugin for the tracer
763 * Register a new plugin tracer.
765 int register_tracer(struct tracer *type)
771 pr_info("Tracer must have a name\n");
775 if (strlen(type->name) >= MAX_TRACER_SIZE) {
776 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
780 mutex_lock(&trace_types_lock);
782 tracing_selftest_running = true;
784 for (t = trace_types; t; t = t->next) {
785 if (strcmp(type->name, t->name) == 0) {
787 pr_info("Tracer %s already registered\n",
795 type->set_flag = &dummy_set_flag;
797 type->flags = &dummy_tracer_flags;
799 if (!type->flags->opts)
800 type->flags->opts = dummy_tracer_opt;
801 if (!type->wait_pipe)
802 type->wait_pipe = default_wait_pipe;
805 #ifdef CONFIG_FTRACE_STARTUP_TEST
806 if (type->selftest && !tracing_selftest_disabled) {
807 struct tracer *saved_tracer = current_trace;
808 struct trace_array *tr = &global_trace;
811 * Run a selftest on this tracer.
812 * Here we reset the trace buffer, and set the current
813 * tracer to be this tracer. The tracer can then run some
814 * internal tracing to verify that everything is in order.
815 * If we fail, we do not register this tracer.
817 tracing_reset_online_cpus(tr);
819 current_trace = type;
821 /* If we expanded the buffers, make sure the max is expanded too */
822 if (ring_buffer_expanded && type->use_max_tr)
823 ring_buffer_resize(max_tr.buffer, trace_buf_size,
824 RING_BUFFER_ALL_CPUS);
826 /* the test is responsible for initializing and enabling */
827 pr_info("Testing tracer %s: ", type->name);
828 ret = type->selftest(type, tr);
829 /* the test is responsible for resetting too */
830 current_trace = saved_tracer;
832 printk(KERN_CONT "FAILED!\n");
835 /* Only reset on passing, to avoid touching corrupted buffers */
836 tracing_reset_online_cpus(tr);
838 /* Shrink the max buffer again */
839 if (ring_buffer_expanded && type->use_max_tr)
840 ring_buffer_resize(max_tr.buffer, 1,
841 RING_BUFFER_ALL_CPUS);
843 printk(KERN_CONT "PASSED\n");
847 type->next = trace_types;
851 tracing_selftest_running = false;
852 mutex_unlock(&trace_types_lock);
854 if (ret || !default_bootup_tracer)
857 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
860 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
861 /* Do we want this tracer to start on bootup? */
862 tracing_set_tracer(type->name);
863 default_bootup_tracer = NULL;
864 /* disable other selftests, since this will break it. */
865 tracing_selftest_disabled = 1;
866 #ifdef CONFIG_FTRACE_STARTUP_TEST
867 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
875 void unregister_tracer(struct tracer *type)
879 mutex_lock(&trace_types_lock);
880 for (t = &trace_types; *t; t = &(*t)->next) {
884 pr_info("Tracer %s not registered\n", type->name);
890 if (type == current_trace && tracer_enabled) {
893 if (current_trace->stop)
894 current_trace->stop(&global_trace);
895 current_trace = &nop_trace;
898 mutex_unlock(&trace_types_lock);
901 void tracing_reset(struct trace_array *tr, int cpu)
903 struct ring_buffer *buffer = tr->buffer;
905 ring_buffer_record_disable(buffer);
907 /* Make sure all commits have finished */
909 ring_buffer_reset_cpu(buffer, cpu);
911 ring_buffer_record_enable(buffer);
914 void tracing_reset_online_cpus(struct trace_array *tr)
916 struct ring_buffer *buffer = tr->buffer;
919 ring_buffer_record_disable(buffer);
921 /* Make sure all commits have finished */
924 tr->time_start = ftrace_now(tr->cpu);
926 for_each_online_cpu(cpu)
927 ring_buffer_reset_cpu(buffer, cpu);
929 ring_buffer_record_enable(buffer);
932 void tracing_reset_current(int cpu)
934 tracing_reset(&global_trace, cpu);
937 void tracing_reset_current_online_cpus(void)
939 tracing_reset_online_cpus(&global_trace);
942 #define SAVED_CMDLINES 128
943 #define NO_CMDLINE_MAP UINT_MAX
944 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
945 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
946 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
947 static int cmdline_idx;
948 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
950 /* temporary disable recording */
951 static atomic_t trace_record_cmdline_disabled __read_mostly;
953 static void trace_init_cmdlines(void)
955 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
956 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
960 int is_tracing_stopped(void)
962 return trace_stop_count;
966 * ftrace_off_permanent - disable all ftrace code permanently
968 * This should only be called when a serious anomally has
969 * been detected. This will turn off the function tracing,
970 * ring buffers, and other tracing utilites. It takes no
971 * locks and can be called from any context.
973 void ftrace_off_permanent(void)
975 tracing_disabled = 1;
977 tracing_off_permanent();
981 * tracing_start - quick start of the tracer
983 * If tracing is enabled but was stopped by tracing_stop,
984 * this will start the tracer back up.
986 void tracing_start(void)
988 struct ring_buffer *buffer;
991 if (tracing_disabled)
994 raw_spin_lock_irqsave(&tracing_start_lock, flags);
995 if (--trace_stop_count) {
996 if (trace_stop_count < 0) {
997 /* Someone screwed up their debugging */
999 trace_stop_count = 0;
1004 /* Prevent the buffers from switching */
1005 arch_spin_lock(&ftrace_max_lock);
1007 buffer = global_trace.buffer;
1009 ring_buffer_record_enable(buffer);
1011 buffer = max_tr.buffer;
1013 ring_buffer_record_enable(buffer);
1015 arch_spin_unlock(&ftrace_max_lock);
1019 raw_spin_unlock_irqrestore(&tracing_start_lock, flags);
1023 * tracing_stop - quick stop of the tracer
1025 * Light weight way to stop tracing. Use in conjunction with
1028 void tracing_stop(void)
1030 struct ring_buffer *buffer;
1031 unsigned long flags;
1034 raw_spin_lock_irqsave(&tracing_start_lock, flags);
1035 if (trace_stop_count++)
1038 /* Prevent the buffers from switching */
1039 arch_spin_lock(&ftrace_max_lock);
1041 buffer = global_trace.buffer;
1043 ring_buffer_record_disable(buffer);
1045 buffer = max_tr.buffer;
1047 ring_buffer_record_disable(buffer);
1049 arch_spin_unlock(&ftrace_max_lock);
1052 raw_spin_unlock_irqrestore(&tracing_start_lock, flags);
1055 void trace_stop_cmdline_recording(void);
1057 static void trace_save_cmdline(struct task_struct *tsk)
1061 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1065 * It's not the end of the world if we don't get
1066 * the lock, but we also don't want to spin
1067 * nor do we want to disable interrupts,
1068 * so if we miss here, then better luck next time.
1070 if (!arch_spin_trylock(&trace_cmdline_lock))
1073 idx = map_pid_to_cmdline[tsk->pid];
1074 if (idx == NO_CMDLINE_MAP) {
1075 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1078 * Check whether the cmdline buffer at idx has a pid
1079 * mapped. We are going to overwrite that entry so we
1080 * need to clear the map_pid_to_cmdline. Otherwise we
1081 * would read the new comm for the old pid.
1083 pid = map_cmdline_to_pid[idx];
1084 if (pid != NO_CMDLINE_MAP)
1085 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1087 map_cmdline_to_pid[idx] = tsk->pid;
1088 map_pid_to_cmdline[tsk->pid] = idx;
1093 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1095 arch_spin_unlock(&trace_cmdline_lock);
1098 void trace_find_cmdline(int pid, char comm[])
1103 strcpy(comm, "<idle>");
1107 if (WARN_ON_ONCE(pid < 0)) {
1108 strcpy(comm, "<XXX>");
1112 if (pid > PID_MAX_DEFAULT) {
1113 strcpy(comm, "<...>");
1118 arch_spin_lock(&trace_cmdline_lock);
1119 map = map_pid_to_cmdline[pid];
1120 if (map != NO_CMDLINE_MAP)
1121 strcpy(comm, saved_cmdlines[map]);
1123 strcpy(comm, "<...>");
1125 arch_spin_unlock(&trace_cmdline_lock);
1129 void tracing_record_cmdline(struct task_struct *tsk)
1131 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1135 trace_save_cmdline(tsk);
1139 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1142 struct task_struct *tsk = current;
1144 entry->preempt_count = pc & 0xff;
1145 entry->pid = (tsk) ? tsk->pid : 0;
1148 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1149 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1151 TRACE_FLAG_IRQS_NOSUPPORT |
1153 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1154 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1155 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1157 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1159 struct ring_buffer_event *
1160 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1163 unsigned long flags, int pc)
1165 struct ring_buffer_event *event;
1167 event = ring_buffer_lock_reserve(buffer, len);
1168 if (event != NULL) {
1169 struct trace_entry *ent = ring_buffer_event_data(event);
1171 tracing_generic_entry_update(ent, flags, pc);
1179 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1180 struct ring_buffer_event *event,
1181 unsigned long flags, int pc,
1184 ring_buffer_unlock_commit(buffer, event);
1186 ftrace_trace_stack(buffer, flags, 6, pc);
1187 ftrace_trace_userstack(buffer, flags, pc);
1193 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1194 struct ring_buffer_event *event,
1195 unsigned long flags, int pc)
1197 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1200 struct ring_buffer_event *
1201 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1202 int type, unsigned long len,
1203 unsigned long flags, int pc)
1205 *current_rb = global_trace.buffer;
1206 return trace_buffer_lock_reserve(*current_rb,
1207 type, len, flags, pc);
1209 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1211 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1212 struct ring_buffer_event *event,
1213 unsigned long flags, int pc)
1215 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1217 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1219 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1220 struct ring_buffer_event *event,
1221 unsigned long flags, int pc)
1223 __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1225 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1227 void trace_nowake_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1228 struct ring_buffer_event *event,
1229 unsigned long flags, int pc,
1230 struct pt_regs *regs)
1232 ring_buffer_unlock_commit(buffer, event);
1234 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1235 ftrace_trace_userstack(buffer, flags, pc);
1237 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit_regs);
1239 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1240 struct ring_buffer_event *event)
1242 ring_buffer_discard_commit(buffer, event);
1244 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1247 trace_function(struct trace_array *tr,
1248 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1251 struct ftrace_event_call *call = &event_function;
1252 struct ring_buffer *buffer = tr->buffer;
1253 struct ring_buffer_event *event;
1254 struct ftrace_entry *entry;
1256 /* If we are reading the ring buffer, don't trace */
1257 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1260 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1264 entry = ring_buffer_event_data(event);
1266 entry->parent_ip = parent_ip;
1268 if (!filter_check_discard(call, entry, buffer, event))
1269 ring_buffer_unlock_commit(buffer, event);
1273 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1274 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1277 if (likely(!atomic_read(&data->disabled)))
1278 trace_function(tr, ip, parent_ip, flags, pc);
1281 #ifdef CONFIG_STACKTRACE
1283 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1284 struct ftrace_stack {
1285 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1288 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1289 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1291 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1292 unsigned long flags,
1293 int skip, int pc, struct pt_regs *regs)
1295 struct ftrace_event_call *call = &event_kernel_stack;
1296 struct ring_buffer_event *event;
1297 struct stack_entry *entry;
1298 struct stack_trace trace;
1300 int size = FTRACE_STACK_ENTRIES;
1302 trace.nr_entries = 0;
1306 * Since events can happen in NMIs there's no safe way to
1307 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1308 * or NMI comes in, it will just have to use the default
1309 * FTRACE_STACK_SIZE.
1311 preempt_disable_notrace();
1313 use_stack = ++__get_cpu_var(ftrace_stack_reserve);
1315 * We don't need any atomic variables, just a barrier.
1316 * If an interrupt comes in, we don't care, because it would
1317 * have exited and put the counter back to what we want.
1318 * We just need a barrier to keep gcc from moving things
1322 if (use_stack == 1) {
1323 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1324 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1327 save_stack_trace_regs(regs, &trace);
1329 save_stack_trace(&trace);
1331 if (trace.nr_entries > size)
1332 size = trace.nr_entries;
1334 /* From now on, use_stack is a boolean */
1337 size *= sizeof(unsigned long);
1339 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1340 sizeof(*entry) + size, flags, pc);
1343 entry = ring_buffer_event_data(event);
1345 memset(&entry->caller, 0, size);
1348 memcpy(&entry->caller, trace.entries,
1349 trace.nr_entries * sizeof(unsigned long));
1351 trace.max_entries = FTRACE_STACK_ENTRIES;
1352 trace.entries = entry->caller;
1354 save_stack_trace_regs(regs, &trace);
1356 save_stack_trace(&trace);
1359 entry->size = trace.nr_entries;
1361 if (!filter_check_discard(call, entry, buffer, event))
1362 ring_buffer_unlock_commit(buffer, event);
1365 /* Again, don't let gcc optimize things here */
1367 __get_cpu_var(ftrace_stack_reserve)--;
1368 preempt_enable_notrace();
1372 void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1373 int skip, int pc, struct pt_regs *regs)
1375 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1378 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1381 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1384 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1387 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1390 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1393 __ftrace_trace_stack(tr->buffer, flags, skip, pc, NULL);
1397 * trace_dump_stack - record a stack back trace in the trace buffer
1399 void trace_dump_stack(void)
1401 unsigned long flags;
1403 if (tracing_disabled || tracing_selftest_running)
1406 local_save_flags(flags);
1408 /* skipping 3 traces, seems to get us at the caller of this function */
1409 __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count(), NULL);
1412 static DEFINE_PER_CPU(int, user_stack_count);
1415 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1417 struct ftrace_event_call *call = &event_user_stack;
1418 struct ring_buffer_event *event;
1419 struct userstack_entry *entry;
1420 struct stack_trace trace;
1422 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1426 * NMIs can not handle page faults, even with fix ups.
1427 * The save user stack can (and often does) fault.
1429 if (unlikely(in_nmi()))
1433 * prevent recursion, since the user stack tracing may
1434 * trigger other kernel events.
1437 if (__this_cpu_read(user_stack_count))
1440 __this_cpu_inc(user_stack_count);
1442 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1443 sizeof(*entry), flags, pc);
1445 goto out_drop_count;
1446 entry = ring_buffer_event_data(event);
1448 entry->tgid = current->tgid;
1449 memset(&entry->caller, 0, sizeof(entry->caller));
1451 trace.nr_entries = 0;
1452 trace.max_entries = FTRACE_STACK_ENTRIES;
1454 trace.entries = entry->caller;
1456 save_stack_trace_user(&trace);
1457 if (!filter_check_discard(call, entry, buffer, event))
1458 ring_buffer_unlock_commit(buffer, event);
1461 __this_cpu_dec(user_stack_count);
1467 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1469 ftrace_trace_userstack(tr, flags, preempt_count());
1473 #endif /* CONFIG_STACKTRACE */
1475 /* created for use with alloc_percpu */
1476 struct trace_buffer_struct {
1477 char buffer[TRACE_BUF_SIZE];
1480 static struct trace_buffer_struct *trace_percpu_buffer;
1481 static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1482 static struct trace_buffer_struct *trace_percpu_irq_buffer;
1483 static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1486 * The buffer used is dependent on the context. There is a per cpu
1487 * buffer for normal context, softirq contex, hard irq context and
1488 * for NMI context. Thise allows for lockless recording.
1490 * Note, if the buffers failed to be allocated, then this returns NULL
1492 static char *get_trace_buf(void)
1494 struct trace_buffer_struct *percpu_buffer;
1495 struct trace_buffer_struct *buffer;
1498 * If we have allocated per cpu buffers, then we do not
1499 * need to do any locking.
1502 percpu_buffer = trace_percpu_nmi_buffer;
1504 percpu_buffer = trace_percpu_irq_buffer;
1505 else if (in_softirq())
1506 percpu_buffer = trace_percpu_sirq_buffer;
1508 percpu_buffer = trace_percpu_buffer;
1513 buffer = per_cpu_ptr(percpu_buffer, smp_processor_id());
1515 return buffer->buffer;
1518 static int alloc_percpu_trace_buffer(void)
1520 struct trace_buffer_struct *buffers;
1521 struct trace_buffer_struct *sirq_buffers;
1522 struct trace_buffer_struct *irq_buffers;
1523 struct trace_buffer_struct *nmi_buffers;
1525 buffers = alloc_percpu(struct trace_buffer_struct);
1529 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
1533 irq_buffers = alloc_percpu(struct trace_buffer_struct);
1537 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
1541 trace_percpu_buffer = buffers;
1542 trace_percpu_sirq_buffer = sirq_buffers;
1543 trace_percpu_irq_buffer = irq_buffers;
1544 trace_percpu_nmi_buffer = nmi_buffers;
1549 free_percpu(irq_buffers);
1551 free_percpu(sirq_buffers);
1553 free_percpu(buffers);
1555 WARN(1, "Could not allocate percpu trace_printk buffer");
1559 void trace_printk_init_buffers(void)
1561 static int buffers_allocated;
1563 if (buffers_allocated)
1566 if (alloc_percpu_trace_buffer())
1569 pr_info("ftrace: Allocated trace_printk buffers\n");
1571 buffers_allocated = 1;
1575 * trace_vbprintk - write binary msg to tracing buffer
1578 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1580 struct ftrace_event_call *call = &event_bprint;
1581 struct ring_buffer_event *event;
1582 struct ring_buffer *buffer;
1583 struct trace_array *tr = &global_trace;
1584 struct bprint_entry *entry;
1585 unsigned long flags;
1587 int len = 0, size, pc;
1589 if (unlikely(tracing_selftest_running || tracing_disabled))
1592 /* Don't pollute graph traces with trace_vprintk internals */
1593 pause_graph_tracing();
1595 pc = preempt_count();
1596 preempt_disable_notrace();
1598 tbuffer = get_trace_buf();
1604 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
1606 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
1609 local_save_flags(flags);
1610 size = sizeof(*entry) + sizeof(u32) * len;
1611 buffer = tr->buffer;
1612 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1616 entry = ring_buffer_event_data(event);
1620 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
1621 if (!filter_check_discard(call, entry, buffer, event)) {
1622 ring_buffer_unlock_commit(buffer, event);
1623 ftrace_trace_stack(buffer, flags, 6, pc);
1627 preempt_enable_notrace();
1628 unpause_graph_tracing();
1632 EXPORT_SYMBOL_GPL(trace_vbprintk);
1634 int trace_array_printk(struct trace_array *tr,
1635 unsigned long ip, const char *fmt, ...)
1640 if (!(trace_flags & TRACE_ITER_PRINTK))
1644 ret = trace_array_vprintk(tr, ip, fmt, ap);
1649 int trace_array_vprintk(struct trace_array *tr,
1650 unsigned long ip, const char *fmt, va_list args)
1652 struct ftrace_event_call *call = &event_print;
1653 struct ring_buffer_event *event;
1654 struct ring_buffer *buffer;
1655 int len = 0, size, pc;
1656 struct print_entry *entry;
1657 unsigned long flags;
1660 if (tracing_disabled || tracing_selftest_running)
1663 /* Don't pollute graph traces with trace_vprintk internals */
1664 pause_graph_tracing();
1666 pc = preempt_count();
1667 preempt_disable_notrace();
1670 tbuffer = get_trace_buf();
1676 len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
1677 if (len > TRACE_BUF_SIZE)
1680 local_save_flags(flags);
1681 size = sizeof(*entry) + len + 1;
1682 buffer = tr->buffer;
1683 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1687 entry = ring_buffer_event_data(event);
1690 memcpy(&entry->buf, tbuffer, len);
1691 entry->buf[len] = '\0';
1692 if (!filter_check_discard(call, entry, buffer, event)) {
1693 ring_buffer_unlock_commit(buffer, event);
1694 ftrace_trace_stack(buffer, flags, 6, pc);
1697 preempt_enable_notrace();
1698 unpause_graph_tracing();
1703 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1705 return trace_array_vprintk(&global_trace, ip, fmt, args);
1707 EXPORT_SYMBOL_GPL(trace_vprintk);
1709 static void trace_iterator_increment(struct trace_iterator *iter)
1712 if (iter->buffer_iter[iter->cpu])
1713 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1716 static struct trace_entry *
1717 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1718 unsigned long *lost_events)
1720 struct ring_buffer_event *event;
1721 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1724 event = ring_buffer_iter_peek(buf_iter, ts);
1726 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1730 iter->ent_size = ring_buffer_event_length(event);
1731 return ring_buffer_event_data(event);
1737 static struct trace_entry *
1738 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1739 unsigned long *missing_events, u64 *ent_ts)
1741 struct ring_buffer *buffer = iter->tr->buffer;
1742 struct trace_entry *ent, *next = NULL;
1743 unsigned long lost_events = 0, next_lost = 0;
1744 int cpu_file = iter->cpu_file;
1745 u64 next_ts = 0, ts;
1751 * If we are in a per_cpu trace file, don't bother by iterating over
1752 * all cpu and peek directly.
1754 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1755 if (ring_buffer_empty_cpu(buffer, cpu_file))
1757 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1759 *ent_cpu = cpu_file;
1764 for_each_tracing_cpu(cpu) {
1766 if (ring_buffer_empty_cpu(buffer, cpu))
1769 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1772 * Pick the entry with the smallest timestamp:
1774 if (ent && (!next || ts < next_ts)) {
1778 next_lost = lost_events;
1779 next_size = iter->ent_size;
1783 iter->ent_size = next_size;
1786 *ent_cpu = next_cpu;
1792 *missing_events = next_lost;
1797 /* Find the next real entry, without updating the iterator itself */
1798 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1799 int *ent_cpu, u64 *ent_ts)
1801 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1804 /* Find the next real entry, and increment the iterator to the next entry */
1805 void *trace_find_next_entry_inc(struct trace_iterator *iter)
1807 iter->ent = __find_next_entry(iter, &iter->cpu,
1808 &iter->lost_events, &iter->ts);
1811 trace_iterator_increment(iter);
1813 return iter->ent ? iter : NULL;
1816 static void trace_consume(struct trace_iterator *iter)
1818 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1819 &iter->lost_events);
1822 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1824 struct trace_iterator *iter = m->private;
1828 WARN_ON_ONCE(iter->leftover);
1832 /* can't go backwards */
1837 ent = trace_find_next_entry_inc(iter);
1841 while (ent && iter->idx < i)
1842 ent = trace_find_next_entry_inc(iter);
1849 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1851 struct trace_array *tr = iter->tr;
1852 struct ring_buffer_event *event;
1853 struct ring_buffer_iter *buf_iter;
1854 unsigned long entries = 0;
1857 tr->data[cpu]->skipped_entries = 0;
1859 if (!iter->buffer_iter[cpu])
1862 buf_iter = iter->buffer_iter[cpu];
1863 ring_buffer_iter_reset(buf_iter);
1866 * We could have the case with the max latency tracers
1867 * that a reset never took place on a cpu. This is evident
1868 * by the timestamp being before the start of the buffer.
1870 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1871 if (ts >= iter->tr->time_start)
1874 ring_buffer_read(buf_iter, NULL);
1877 tr->data[cpu]->skipped_entries = entries;
1881 * The current tracer is copied to avoid a global locking
1884 static void *s_start(struct seq_file *m, loff_t *pos)
1886 struct trace_iterator *iter = m->private;
1887 static struct tracer *old_tracer;
1888 int cpu_file = iter->cpu_file;
1893 /* copy the tracer to avoid using a global lock all around */
1894 mutex_lock(&trace_types_lock);
1895 if (unlikely(old_tracer != current_trace && current_trace)) {
1896 old_tracer = current_trace;
1897 *iter->trace = *current_trace;
1899 mutex_unlock(&trace_types_lock);
1901 atomic_inc(&trace_record_cmdline_disabled);
1903 if (*pos != iter->pos) {
1908 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1909 for_each_tracing_cpu(cpu)
1910 tracing_iter_reset(iter, cpu);
1912 tracing_iter_reset(iter, cpu_file);
1915 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1920 * If we overflowed the seq_file before, then we want
1921 * to just reuse the trace_seq buffer again.
1927 p = s_next(m, p, &l);
1931 trace_event_read_lock();
1932 trace_access_lock(cpu_file);
1936 static void s_stop(struct seq_file *m, void *p)
1938 struct trace_iterator *iter = m->private;
1940 atomic_dec(&trace_record_cmdline_disabled);
1941 trace_access_unlock(iter->cpu_file);
1942 trace_event_read_unlock();
1946 get_total_entries(struct trace_array *tr, unsigned long *total, unsigned long *entries)
1948 unsigned long count;
1954 for_each_tracing_cpu(cpu) {
1955 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1957 * If this buffer has skipped entries, then we hold all
1958 * entries for the trace and we need to ignore the
1959 * ones before the time stamp.
1961 if (tr->data[cpu]->skipped_entries) {
1962 count -= tr->data[cpu]->skipped_entries;
1963 /* total is the same as the entries */
1967 ring_buffer_overrun_cpu(tr->buffer, cpu);
1972 static void print_lat_help_header(struct seq_file *m)
1974 seq_puts(m, "# _------=> CPU# \n");
1975 seq_puts(m, "# / _-----=> irqs-off \n");
1976 seq_puts(m, "# | / _----=> need-resched \n");
1977 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1978 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1979 seq_puts(m, "# |||| / delay \n");
1980 seq_puts(m, "# cmd pid ||||| time | caller \n");
1981 seq_puts(m, "# \\ / ||||| \\ | / \n");
1984 static void print_event_info(struct trace_array *tr, struct seq_file *m)
1986 unsigned long total;
1987 unsigned long entries;
1989 get_total_entries(tr, &total, &entries);
1990 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
1991 entries, total, num_online_cpus());
1995 static void print_func_help_header(struct trace_array *tr, struct seq_file *m)
1997 print_event_info(tr, m);
1998 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1999 seq_puts(m, "# | | | | |\n");
2002 static void print_func_help_header_irq(struct trace_array *tr, struct seq_file *m)
2004 print_event_info(tr, m);
2005 seq_puts(m, "# _-----=> irqs-off\n");
2006 seq_puts(m, "# / _----=> need-resched\n");
2007 seq_puts(m, "# | / _---=> hardirq/softirq\n");
2008 seq_puts(m, "# || / _--=> preempt-depth\n");
2009 seq_puts(m, "# ||| / delay\n");
2010 seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2011 seq_puts(m, "# | | | |||| | |\n");
2015 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2017 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2018 struct trace_array *tr = iter->tr;
2019 struct trace_array_cpu *data = tr->data[tr->cpu];
2020 struct tracer *type = current_trace;
2021 unsigned long entries;
2022 unsigned long total;
2023 const char *name = "preemption";
2028 get_total_entries(tr, &total, &entries);
2030 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2032 seq_puts(m, "# -----------------------------------"
2033 "---------------------------------\n");
2034 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2035 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2036 nsecs_to_usecs(data->saved_latency),
2040 #if defined(CONFIG_PREEMPT_NONE)
2042 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2044 #elif defined(CONFIG_PREEMPT)
2049 /* These are reserved for later use */
2052 seq_printf(m, " #P:%d)\n", num_online_cpus());
2056 seq_puts(m, "# -----------------\n");
2057 seq_printf(m, "# | task: %.16s-%d "
2058 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2059 data->comm, data->pid, data->uid, data->nice,
2060 data->policy, data->rt_priority);
2061 seq_puts(m, "# -----------------\n");
2063 if (data->critical_start) {
2064 seq_puts(m, "# => started at: ");
2065 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2066 trace_print_seq(m, &iter->seq);
2067 seq_puts(m, "\n# => ended at: ");
2068 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2069 trace_print_seq(m, &iter->seq);
2070 seq_puts(m, "\n#\n");
2076 static void test_cpu_buff_start(struct trace_iterator *iter)
2078 struct trace_seq *s = &iter->seq;
2080 if (!(trace_flags & TRACE_ITER_ANNOTATE))
2083 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2086 if (cpumask_test_cpu(iter->cpu, iter->started))
2089 if (iter->tr->data[iter->cpu]->skipped_entries)
2092 cpumask_set_cpu(iter->cpu, iter->started);
2094 /* Don't print started cpu buffer for the first entry of the trace */
2096 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2100 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2102 struct trace_seq *s = &iter->seq;
2103 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2104 struct trace_entry *entry;
2105 struct trace_event *event;
2109 test_cpu_buff_start(iter);
2111 event = ftrace_find_event(entry->type);
2113 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2114 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2115 if (!trace_print_lat_context(iter))
2118 if (!trace_print_context(iter))
2124 return event->funcs->trace(iter, sym_flags, event);
2126 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2129 return TRACE_TYPE_HANDLED;
2131 return TRACE_TYPE_PARTIAL_LINE;
2134 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2136 struct trace_seq *s = &iter->seq;
2137 struct trace_entry *entry;
2138 struct trace_event *event;
2142 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2143 if (!trace_seq_printf(s, "%d %d %llu ",
2144 entry->pid, iter->cpu, iter->ts))
2148 event = ftrace_find_event(entry->type);
2150 return event->funcs->raw(iter, 0, event);
2152 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2155 return TRACE_TYPE_HANDLED;
2157 return TRACE_TYPE_PARTIAL_LINE;
2160 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2162 struct trace_seq *s = &iter->seq;
2163 unsigned char newline = '\n';
2164 struct trace_entry *entry;
2165 struct trace_event *event;
2169 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2170 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2171 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2172 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2175 event = ftrace_find_event(entry->type);
2177 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2178 if (ret != TRACE_TYPE_HANDLED)
2182 SEQ_PUT_FIELD_RET(s, newline);
2184 return TRACE_TYPE_HANDLED;
2187 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2189 struct trace_seq *s = &iter->seq;
2190 struct trace_entry *entry;
2191 struct trace_event *event;
2195 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2196 SEQ_PUT_FIELD_RET(s, entry->pid);
2197 SEQ_PUT_FIELD_RET(s, iter->cpu);
2198 SEQ_PUT_FIELD_RET(s, iter->ts);
2201 event = ftrace_find_event(entry->type);
2202 return event ? event->funcs->binary(iter, 0, event) :
2206 int trace_empty(struct trace_iterator *iter)
2210 /* If we are looking at one CPU buffer, only check that one */
2211 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
2212 cpu = iter->cpu_file;
2213 if (iter->buffer_iter[cpu]) {
2214 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2217 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2223 for_each_tracing_cpu(cpu) {
2224 if (iter->buffer_iter[cpu]) {
2225 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2228 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2236 /* Called with trace_event_read_lock() held. */
2237 enum print_line_t print_trace_line(struct trace_iterator *iter)
2239 enum print_line_t ret;
2241 if (iter->lost_events &&
2242 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2243 iter->cpu, iter->lost_events))
2244 return TRACE_TYPE_PARTIAL_LINE;
2246 if (iter->trace && iter->trace->print_line) {
2247 ret = iter->trace->print_line(iter);
2248 if (ret != TRACE_TYPE_UNHANDLED)
2252 if (iter->ent->type == TRACE_BPRINT &&
2253 trace_flags & TRACE_ITER_PRINTK &&
2254 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2255 return trace_print_bprintk_msg_only(iter);
2257 if (iter->ent->type == TRACE_PRINT &&
2258 trace_flags & TRACE_ITER_PRINTK &&
2259 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2260 return trace_print_printk_msg_only(iter);
2262 if (trace_flags & TRACE_ITER_BIN)
2263 return print_bin_fmt(iter);
2265 if (trace_flags & TRACE_ITER_HEX)
2266 return print_hex_fmt(iter);
2268 if (trace_flags & TRACE_ITER_RAW)
2269 return print_raw_fmt(iter);
2271 return print_trace_fmt(iter);
2274 void trace_latency_header(struct seq_file *m)
2276 struct trace_iterator *iter = m->private;
2278 /* print nothing if the buffers are empty */
2279 if (trace_empty(iter))
2282 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2283 print_trace_header(m, iter);
2285 if (!(trace_flags & TRACE_ITER_VERBOSE))
2286 print_lat_help_header(m);
2289 void trace_default_header(struct seq_file *m)
2291 struct trace_iterator *iter = m->private;
2293 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2296 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2297 /* print nothing if the buffers are empty */
2298 if (trace_empty(iter))
2300 print_trace_header(m, iter);
2301 if (!(trace_flags & TRACE_ITER_VERBOSE))
2302 print_lat_help_header(m);
2304 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2305 if (trace_flags & TRACE_ITER_IRQ_INFO)
2306 print_func_help_header_irq(iter->tr, m);
2308 print_func_help_header(iter->tr, m);
2313 static void test_ftrace_alive(struct seq_file *m)
2315 if (!ftrace_is_dead())
2317 seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2318 seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2321 static int s_show(struct seq_file *m, void *v)
2323 struct trace_iterator *iter = v;
2326 if (iter->ent == NULL) {
2328 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2330 test_ftrace_alive(m);
2332 if (iter->trace && iter->trace->print_header)
2333 iter->trace->print_header(m);
2335 trace_default_header(m);
2337 } else if (iter->leftover) {
2339 * If we filled the seq_file buffer earlier, we
2340 * want to just show it now.
2342 ret = trace_print_seq(m, &iter->seq);
2344 /* ret should this time be zero, but you never know */
2345 iter->leftover = ret;
2348 print_trace_line(iter);
2349 ret = trace_print_seq(m, &iter->seq);
2351 * If we overflow the seq_file buffer, then it will
2352 * ask us for this data again at start up.
2354 * ret is 0 if seq_file write succeeded.
2357 iter->leftover = ret;
2363 static const struct seq_operations tracer_seq_ops = {
2370 static struct trace_iterator *
2371 __tracing_open(struct inode *inode, struct file *file)
2373 long cpu_file = (long) inode->i_private;
2374 struct trace_iterator *iter;
2377 if (tracing_disabled)
2378 return ERR_PTR(-ENODEV);
2380 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
2382 return ERR_PTR(-ENOMEM);
2385 * We make a copy of the current tracer to avoid concurrent
2386 * changes on it while we are reading.
2388 mutex_lock(&trace_types_lock);
2389 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2394 *iter->trace = *current_trace;
2396 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2399 if (current_trace && current_trace->print_max)
2402 iter->tr = &global_trace;
2404 mutex_init(&iter->mutex);
2405 iter->cpu_file = cpu_file;
2407 /* Notify the tracer early; before we stop tracing. */
2408 if (iter->trace && iter->trace->open)
2409 iter->trace->open(iter);
2411 /* Annotate start of buffers if we had overruns */
2412 if (ring_buffer_overruns(iter->tr->buffer))
2413 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2415 /* stop the trace while dumping */
2418 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2419 for_each_tracing_cpu(cpu) {
2420 iter->buffer_iter[cpu] =
2421 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2423 ring_buffer_read_prepare_sync();
2424 for_each_tracing_cpu(cpu) {
2425 ring_buffer_read_start(iter->buffer_iter[cpu]);
2426 tracing_iter_reset(iter, cpu);
2429 cpu = iter->cpu_file;
2430 iter->buffer_iter[cpu] =
2431 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2432 ring_buffer_read_prepare_sync();
2433 ring_buffer_read_start(iter->buffer_iter[cpu]);
2434 tracing_iter_reset(iter, cpu);
2437 mutex_unlock(&trace_types_lock);
2442 mutex_unlock(&trace_types_lock);
2444 seq_release_private(inode, file);
2445 return ERR_PTR(-ENOMEM);
2448 int tracing_open_generic(struct inode *inode, struct file *filp)
2450 if (tracing_disabled)
2453 filp->private_data = inode->i_private;
2457 static int tracing_release(struct inode *inode, struct file *file)
2459 struct seq_file *m = file->private_data;
2460 struct trace_iterator *iter;
2463 if (!(file->f_mode & FMODE_READ))
2468 mutex_lock(&trace_types_lock);
2469 for_each_tracing_cpu(cpu) {
2470 if (iter->buffer_iter[cpu])
2471 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2474 if (iter->trace && iter->trace->close)
2475 iter->trace->close(iter);
2477 /* reenable tracing if it was previously enabled */
2479 mutex_unlock(&trace_types_lock);
2481 mutex_destroy(&iter->mutex);
2482 free_cpumask_var(iter->started);
2484 seq_release_private(inode, file);
2488 static int tracing_open(struct inode *inode, struct file *file)
2490 struct trace_iterator *iter;
2493 /* If this file was open for write, then erase contents */
2494 if ((file->f_mode & FMODE_WRITE) &&
2495 (file->f_flags & O_TRUNC)) {
2496 long cpu = (long) inode->i_private;
2498 if (cpu == TRACE_PIPE_ALL_CPU)
2499 tracing_reset_online_cpus(&global_trace);
2501 tracing_reset(&global_trace, cpu);
2504 if (file->f_mode & FMODE_READ) {
2505 iter = __tracing_open(inode, file);
2507 ret = PTR_ERR(iter);
2508 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2509 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2515 t_next(struct seq_file *m, void *v, loff_t *pos)
2517 struct tracer *t = v;
2527 static void *t_start(struct seq_file *m, loff_t *pos)
2532 mutex_lock(&trace_types_lock);
2533 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2539 static void t_stop(struct seq_file *m, void *p)
2541 mutex_unlock(&trace_types_lock);
2544 static int t_show(struct seq_file *m, void *v)
2546 struct tracer *t = v;
2551 seq_printf(m, "%s", t->name);
2560 static const struct seq_operations show_traces_seq_ops = {
2567 static int show_traces_open(struct inode *inode, struct file *file)
2569 if (tracing_disabled)
2572 return seq_open(file, &show_traces_seq_ops);
2576 tracing_write_stub(struct file *filp, const char __user *ubuf,
2577 size_t count, loff_t *ppos)
2582 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
2584 if (file->f_mode & FMODE_READ)
2585 return seq_lseek(file, offset, origin);
2590 static const struct file_operations tracing_fops = {
2591 .open = tracing_open,
2593 .write = tracing_write_stub,
2594 .llseek = tracing_seek,
2595 .release = tracing_release,
2598 static const struct file_operations show_traces_fops = {
2599 .open = show_traces_open,
2601 .release = seq_release,
2602 .llseek = seq_lseek,
2606 * Only trace on a CPU if the bitmask is set:
2608 static cpumask_var_t tracing_cpumask;
2611 * The tracer itself will not take this lock, but still we want
2612 * to provide a consistent cpumask to user-space:
2614 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2617 * Temporary storage for the character representation of the
2618 * CPU bitmask (and one more byte for the newline):
2620 static char mask_str[NR_CPUS + 1];
2623 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2624 size_t count, loff_t *ppos)
2628 mutex_lock(&tracing_cpumask_update_lock);
2630 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2631 if (count - len < 2) {
2635 len += sprintf(mask_str + len, "\n");
2636 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2639 mutex_unlock(&tracing_cpumask_update_lock);
2645 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2646 size_t count, loff_t *ppos)
2649 cpumask_var_t tracing_cpumask_new;
2651 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2654 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2658 mutex_lock(&tracing_cpumask_update_lock);
2660 local_irq_disable();
2661 arch_spin_lock(&ftrace_max_lock);
2662 for_each_tracing_cpu(cpu) {
2664 * Increase/decrease the disabled counter if we are
2665 * about to flip a bit in the cpumask:
2667 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2668 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2669 atomic_inc(&global_trace.data[cpu]->disabled);
2670 ring_buffer_record_disable_cpu(global_trace.buffer, cpu);
2672 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2673 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2674 atomic_dec(&global_trace.data[cpu]->disabled);
2675 ring_buffer_record_enable_cpu(global_trace.buffer, cpu);
2678 arch_spin_unlock(&ftrace_max_lock);
2681 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2683 mutex_unlock(&tracing_cpumask_update_lock);
2684 free_cpumask_var(tracing_cpumask_new);
2689 free_cpumask_var(tracing_cpumask_new);
2694 static const struct file_operations tracing_cpumask_fops = {
2695 .open = tracing_open_generic,
2696 .read = tracing_cpumask_read,
2697 .write = tracing_cpumask_write,
2698 .llseek = generic_file_llseek,
2701 static int tracing_trace_options_show(struct seq_file *m, void *v)
2703 struct tracer_opt *trace_opts;
2707 mutex_lock(&trace_types_lock);
2708 tracer_flags = current_trace->flags->val;
2709 trace_opts = current_trace->flags->opts;
2711 for (i = 0; trace_options[i]; i++) {
2712 if (trace_flags & (1 << i))
2713 seq_printf(m, "%s\n", trace_options[i]);
2715 seq_printf(m, "no%s\n", trace_options[i]);
2718 for (i = 0; trace_opts[i].name; i++) {
2719 if (tracer_flags & trace_opts[i].bit)
2720 seq_printf(m, "%s\n", trace_opts[i].name);
2722 seq_printf(m, "no%s\n", trace_opts[i].name);
2724 mutex_unlock(&trace_types_lock);
2729 static int __set_tracer_option(struct tracer *trace,
2730 struct tracer_flags *tracer_flags,
2731 struct tracer_opt *opts, int neg)
2735 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2740 tracer_flags->val &= ~opts->bit;
2742 tracer_flags->val |= opts->bit;
2746 /* Try to assign a tracer specific option */
2747 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2749 struct tracer_flags *tracer_flags = trace->flags;
2750 struct tracer_opt *opts = NULL;
2753 for (i = 0; tracer_flags->opts[i].name; i++) {
2754 opts = &tracer_flags->opts[i];
2756 if (strcmp(cmp, opts->name) == 0)
2757 return __set_tracer_option(trace, trace->flags,
2764 static void set_tracer_flags(unsigned int mask, int enabled)
2766 /* do nothing if flag is already set */
2767 if (!!(trace_flags & mask) == !!enabled)
2771 trace_flags |= mask;
2773 trace_flags &= ~mask;
2775 if (mask == TRACE_ITER_RECORD_CMD)
2776 trace_event_enable_cmd_record(enabled);
2778 if (mask == TRACE_ITER_OVERWRITE)
2779 ring_buffer_change_overwrite(global_trace.buffer, enabled);
2783 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2784 size_t cnt, loff_t *ppos)
2792 if (cnt >= sizeof(buf))
2795 if (copy_from_user(&buf, ubuf, cnt))
2799 cmp = strstrip(buf);
2801 if (strncmp(cmp, "no", 2) == 0) {
2806 for (i = 0; trace_options[i]; i++) {
2807 if (strcmp(cmp, trace_options[i]) == 0) {
2808 set_tracer_flags(1 << i, !neg);
2813 /* If no option could be set, test the specific tracer options */
2814 if (!trace_options[i]) {
2815 mutex_lock(&trace_types_lock);
2816 ret = set_tracer_option(current_trace, cmp, neg);
2817 mutex_unlock(&trace_types_lock);
2827 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2829 if (tracing_disabled)
2831 return single_open(file, tracing_trace_options_show, NULL);
2834 static const struct file_operations tracing_iter_fops = {
2835 .open = tracing_trace_options_open,
2837 .llseek = seq_lseek,
2838 .release = single_release,
2839 .write = tracing_trace_options_write,
2842 static const char readme_msg[] =
2843 "tracing mini-HOWTO:\n\n"
2844 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2845 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2846 "wakeup wakeup_rt preemptirqsoff preemptoff irqsoff function nop\n\n"
2847 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2849 "# echo wakeup > /sys/kernel/debug/tracing/current_tracer\n"
2850 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2852 "# cat /sys/kernel/debug/tracing/trace_options\n"
2853 "noprint-parent nosym-offset nosym-addr noverbose\n"
2854 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2855 "# echo 1 > /sys/kernel/debug/tracing/tracing_on\n"
2856 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2857 "# echo 0 > /sys/kernel/debug/tracing/tracing_on\n"
2861 tracing_readme_read(struct file *filp, char __user *ubuf,
2862 size_t cnt, loff_t *ppos)
2864 return simple_read_from_buffer(ubuf, cnt, ppos,
2865 readme_msg, strlen(readme_msg));
2868 static const struct file_operations tracing_readme_fops = {
2869 .open = tracing_open_generic,
2870 .read = tracing_readme_read,
2871 .llseek = generic_file_llseek,
2875 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2876 size_t cnt, loff_t *ppos)
2885 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2889 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2897 for (i = 0; i < SAVED_CMDLINES; i++) {
2900 pid = map_cmdline_to_pid[i];
2901 if (pid == -1 || pid == NO_CMDLINE_MAP)
2904 trace_find_cmdline(pid, buf_comm);
2905 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2910 len = simple_read_from_buffer(ubuf, cnt, ppos,
2919 static const struct file_operations tracing_saved_cmdlines_fops = {
2920 .open = tracing_open_generic,
2921 .read = tracing_saved_cmdlines_read,
2922 .llseek = generic_file_llseek,
2926 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2927 size_t cnt, loff_t *ppos)
2932 r = sprintf(buf, "%u\n", tracer_enabled);
2933 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2937 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2938 size_t cnt, loff_t *ppos)
2940 struct trace_array *tr = filp->private_data;
2944 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
2950 mutex_lock(&trace_types_lock);
2951 if (tracer_enabled ^ val) {
2953 /* Only need to warn if this is used to change the state */
2954 WARN_ONCE(1, "tracing_enabled is deprecated. Use tracing_on");
2958 if (current_trace->start)
2959 current_trace->start(tr);
2964 if (current_trace->stop)
2965 current_trace->stop(tr);
2968 mutex_unlock(&trace_types_lock);
2976 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2977 size_t cnt, loff_t *ppos)
2979 char buf[MAX_TRACER_SIZE+2];
2982 mutex_lock(&trace_types_lock);
2984 r = sprintf(buf, "%s\n", current_trace->name);
2986 r = sprintf(buf, "\n");
2987 mutex_unlock(&trace_types_lock);
2989 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2992 int tracer_init(struct tracer *t, struct trace_array *tr)
2994 tracing_reset_online_cpus(tr);
2998 static void set_buffer_entries(struct trace_array *tr, unsigned long val)
3001 for_each_tracing_cpu(cpu)
3002 tr->data[cpu]->entries = val;
3005 static int __tracing_resize_ring_buffer(unsigned long size, int cpu)
3010 * If kernel or user changes the size of the ring buffer
3011 * we use the size that was given, and we can forget about
3012 * expanding it later.
3014 ring_buffer_expanded = 1;
3016 ret = ring_buffer_resize(global_trace.buffer, size, cpu);
3020 if (!current_trace->use_max_tr)
3023 ret = ring_buffer_resize(max_tr.buffer, size, cpu);
3027 if (cpu == RING_BUFFER_ALL_CPUS) {
3029 for_each_tracing_cpu(i) {
3030 r = ring_buffer_resize(global_trace.buffer,
3031 global_trace.data[i]->entries,
3037 r = ring_buffer_resize(global_trace.buffer,
3038 global_trace.data[cpu]->entries,
3044 * AARGH! We are left with different
3045 * size max buffer!!!!
3046 * The max buffer is our "snapshot" buffer.
3047 * When a tracer needs a snapshot (one of the
3048 * latency tracers), it swaps the max buffer
3049 * with the saved snap shot. We succeeded to
3050 * update the size of the main buffer, but failed to
3051 * update the size of the max buffer. But when we tried
3052 * to reset the main buffer to the original size, we
3053 * failed there too. This is very unlikely to
3054 * happen, but if it does, warn and kill all
3058 tracing_disabled = 1;
3063 if (cpu == RING_BUFFER_ALL_CPUS)
3064 set_buffer_entries(&max_tr, size);
3066 max_tr.data[cpu]->entries = size;
3069 if (cpu == RING_BUFFER_ALL_CPUS)
3070 set_buffer_entries(&global_trace, size);
3072 global_trace.data[cpu]->entries = size;
3077 static ssize_t tracing_resize_ring_buffer(unsigned long size, int cpu_id)
3081 mutex_lock(&trace_types_lock);
3083 if (cpu_id != RING_BUFFER_ALL_CPUS) {
3084 /* make sure, this cpu is enabled in the mask */
3085 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
3091 ret = __tracing_resize_ring_buffer(size, cpu_id);
3096 mutex_unlock(&trace_types_lock);
3103 * tracing_update_buffers - used by tracing facility to expand ring buffers
3105 * To save on memory when the tracing is never used on a system with it
3106 * configured in. The ring buffers are set to a minimum size. But once
3107 * a user starts to use the tracing facility, then they need to grow
3108 * to their default size.
3110 * This function is to be called when a tracer is about to be used.
3112 int tracing_update_buffers(void)
3116 mutex_lock(&trace_types_lock);
3117 if (!ring_buffer_expanded)
3118 ret = __tracing_resize_ring_buffer(trace_buf_size,
3119 RING_BUFFER_ALL_CPUS);
3120 mutex_unlock(&trace_types_lock);
3125 struct trace_option_dentry;
3127 static struct trace_option_dentry *
3128 create_trace_option_files(struct tracer *tracer);
3131 destroy_trace_option_files(struct trace_option_dentry *topts);
3133 static int tracing_set_tracer(const char *buf)
3135 static struct trace_option_dentry *topts;
3136 struct trace_array *tr = &global_trace;
3140 mutex_lock(&trace_types_lock);
3142 if (!ring_buffer_expanded) {
3143 ret = __tracing_resize_ring_buffer(trace_buf_size,
3144 RING_BUFFER_ALL_CPUS);
3150 for (t = trace_types; t; t = t->next) {
3151 if (strcmp(t->name, buf) == 0)
3158 if (t == current_trace)
3161 trace_branch_disable();
3162 if (current_trace && current_trace->reset)
3163 current_trace->reset(tr);
3164 if (current_trace && current_trace->use_max_tr) {
3166 * We don't free the ring buffer. instead, resize it because
3167 * The max_tr ring buffer has some state (e.g. ring->clock) and
3168 * we want preserve it.
3170 ring_buffer_resize(max_tr.buffer, 1, RING_BUFFER_ALL_CPUS);
3171 set_buffer_entries(&max_tr, 1);
3173 destroy_trace_option_files(topts);
3177 topts = create_trace_option_files(current_trace);
3178 if (current_trace->use_max_tr) {
3180 /* we need to make per cpu buffer sizes equivalent */
3181 for_each_tracing_cpu(cpu) {
3182 ret = ring_buffer_resize(max_tr.buffer,
3183 global_trace.data[cpu]->entries,
3187 max_tr.data[cpu]->entries =
3188 global_trace.data[cpu]->entries;
3193 ret = tracer_init(t, tr);
3198 trace_branch_enable(tr);
3200 mutex_unlock(&trace_types_lock);
3206 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3207 size_t cnt, loff_t *ppos)
3209 char buf[MAX_TRACER_SIZE+1];
3216 if (cnt > MAX_TRACER_SIZE)
3217 cnt = MAX_TRACER_SIZE;
3219 if (copy_from_user(&buf, ubuf, cnt))
3224 /* strip ending whitespace. */
3225 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3228 err = tracing_set_tracer(buf);
3238 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3239 size_t cnt, loff_t *ppos)
3241 unsigned long *ptr = filp->private_data;
3245 r = snprintf(buf, sizeof(buf), "%ld\n",
3246 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3247 if (r > sizeof(buf))
3249 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3253 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3254 size_t cnt, loff_t *ppos)
3256 unsigned long *ptr = filp->private_data;
3260 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3269 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3271 long cpu_file = (long) inode->i_private;
3272 struct trace_iterator *iter;
3275 if (tracing_disabled)
3278 mutex_lock(&trace_types_lock);
3280 /* create a buffer to store the information to pass to userspace */
3281 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3288 * We make a copy of the current tracer to avoid concurrent
3289 * changes on it while we are reading.
3291 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3297 *iter->trace = *current_trace;
3299 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3304 /* trace pipe does not show start of buffer */
3305 cpumask_setall(iter->started);
3307 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3308 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3310 iter->cpu_file = cpu_file;
3311 iter->tr = &global_trace;
3312 mutex_init(&iter->mutex);
3313 filp->private_data = iter;
3315 if (iter->trace->pipe_open)
3316 iter->trace->pipe_open(iter);
3318 nonseekable_open(inode, filp);
3320 mutex_unlock(&trace_types_lock);
3326 mutex_unlock(&trace_types_lock);
3330 static int tracing_release_pipe(struct inode *inode, struct file *file)
3332 struct trace_iterator *iter = file->private_data;
3334 mutex_lock(&trace_types_lock);
3336 if (iter->trace->pipe_close)
3337 iter->trace->pipe_close(iter);
3339 mutex_unlock(&trace_types_lock);
3341 free_cpumask_var(iter->started);
3342 mutex_destroy(&iter->mutex);
3350 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3352 struct trace_iterator *iter = filp->private_data;
3354 if (trace_flags & TRACE_ITER_BLOCK) {
3356 * Always select as readable when in blocking mode
3358 return POLLIN | POLLRDNORM;
3360 if (!trace_empty(iter))
3361 return POLLIN | POLLRDNORM;
3362 poll_wait(filp, &trace_wait, poll_table);
3363 if (!trace_empty(iter))
3364 return POLLIN | POLLRDNORM;
3371 void default_wait_pipe(struct trace_iterator *iter)
3375 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3377 if (trace_empty(iter))
3380 finish_wait(&trace_wait, &wait);
3384 * This is a make-shift waitqueue.
3385 * A tracer might use this callback on some rare cases:
3387 * 1) the current tracer might hold the runqueue lock when it wakes up
3388 * a reader, hence a deadlock (sched, function, and function graph tracers)
3389 * 2) the function tracers, trace all functions, we don't want
3390 * the overhead of calling wake_up and friends
3391 * (and tracing them too)
3393 * Anyway, this is really very primitive wakeup.
3395 void poll_wait_pipe(struct trace_iterator *iter)
3397 set_current_state(TASK_INTERRUPTIBLE);
3398 /* sleep for 100 msecs, and try again. */
3399 schedule_timeout(HZ / 10);
3402 /* Must be called with trace_types_lock mutex held. */
3403 static int tracing_wait_pipe(struct file *filp)
3405 struct trace_iterator *iter = filp->private_data;
3407 while (trace_empty(iter)) {
3409 if ((filp->f_flags & O_NONBLOCK)) {
3413 mutex_unlock(&iter->mutex);
3415 iter->trace->wait_pipe(iter);
3417 mutex_lock(&iter->mutex);
3419 if (signal_pending(current))
3423 * We block until we read something and tracing is disabled.
3424 * We still block if tracing is disabled, but we have never
3425 * read anything. This allows a user to cat this file, and
3426 * then enable tracing. But after we have read something,
3427 * we give an EOF when tracing is again disabled.
3429 * iter->pos will be 0 if we haven't read anything.
3431 if (!tracer_enabled && iter->pos)
3442 tracing_read_pipe(struct file *filp, char __user *ubuf,
3443 size_t cnt, loff_t *ppos)
3445 struct trace_iterator *iter = filp->private_data;
3446 static struct tracer *old_tracer;
3449 /* return any leftover data */
3450 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3454 trace_seq_init(&iter->seq);
3456 /* copy the tracer to avoid using a global lock all around */
3457 mutex_lock(&trace_types_lock);
3458 if (unlikely(old_tracer != current_trace && current_trace)) {
3459 old_tracer = current_trace;
3460 *iter->trace = *current_trace;
3462 mutex_unlock(&trace_types_lock);
3465 * Avoid more than one consumer on a single file descriptor
3466 * This is just a matter of traces coherency, the ring buffer itself
3469 mutex_lock(&iter->mutex);
3470 if (iter->trace->read) {
3471 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3477 sret = tracing_wait_pipe(filp);
3481 /* stop when tracing is finished */
3482 if (trace_empty(iter)) {
3487 if (cnt >= PAGE_SIZE)
3488 cnt = PAGE_SIZE - 1;
3490 /* reset all but tr, trace, and overruns */
3491 memset(&iter->seq, 0,
3492 sizeof(struct trace_iterator) -
3493 offsetof(struct trace_iterator, seq));
3496 trace_event_read_lock();
3497 trace_access_lock(iter->cpu_file);
3498 while (trace_find_next_entry_inc(iter) != NULL) {
3499 enum print_line_t ret;
3500 int len = iter->seq.len;
3502 ret = print_trace_line(iter);
3503 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3504 /* don't print partial lines */
3505 iter->seq.len = len;
3508 if (ret != TRACE_TYPE_NO_CONSUME)
3509 trace_consume(iter);
3511 if (iter->seq.len >= cnt)
3515 * Setting the full flag means we reached the trace_seq buffer
3516 * size and we should leave by partial output condition above.
3517 * One of the trace_seq_* functions is not used properly.
3519 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
3522 trace_access_unlock(iter->cpu_file);
3523 trace_event_read_unlock();
3525 /* Now copy what we have to the user */
3526 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3527 if (iter->seq.readpos >= iter->seq.len)
3528 trace_seq_init(&iter->seq);
3531 * If there was nothing to send to user, in spite of consuming trace
3532 * entries, go back to wait for more entries.
3538 mutex_unlock(&iter->mutex);
3543 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3544 struct pipe_buffer *buf)
3546 __free_page(buf->page);
3549 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3552 __free_page(spd->pages[idx]);
3555 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3557 .map = generic_pipe_buf_map,
3558 .unmap = generic_pipe_buf_unmap,
3559 .confirm = generic_pipe_buf_confirm,
3560 .release = tracing_pipe_buf_release,
3561 .steal = generic_pipe_buf_steal,
3562 .get = generic_pipe_buf_get,
3566 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3571 /* Seq buffer is page-sized, exactly what we need. */
3573 count = iter->seq.len;
3574 ret = print_trace_line(iter);
3575 count = iter->seq.len - count;
3578 iter->seq.len -= count;
3581 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3582 iter->seq.len -= count;
3586 if (ret != TRACE_TYPE_NO_CONSUME)
3587 trace_consume(iter);
3589 if (!trace_find_next_entry_inc(iter)) {
3599 static ssize_t tracing_splice_read_pipe(struct file *filp,
3601 struct pipe_inode_info *pipe,
3605 struct page *pages_def[PIPE_DEF_BUFFERS];
3606 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3607 struct trace_iterator *iter = filp->private_data;
3608 struct splice_pipe_desc spd = {
3610 .partial = partial_def,
3611 .nr_pages = 0, /* This gets updated below. */
3613 .ops = &tracing_pipe_buf_ops,
3614 .spd_release = tracing_spd_release_pipe,
3616 static struct tracer *old_tracer;
3621 if (splice_grow_spd(pipe, &spd))
3624 /* copy the tracer to avoid using a global lock all around */
3625 mutex_lock(&trace_types_lock);
3626 if (unlikely(old_tracer != current_trace && current_trace)) {
3627 old_tracer = current_trace;
3628 *iter->trace = *current_trace;
3630 mutex_unlock(&trace_types_lock);
3632 mutex_lock(&iter->mutex);
3634 if (iter->trace->splice_read) {
3635 ret = iter->trace->splice_read(iter, filp,
3636 ppos, pipe, len, flags);
3641 ret = tracing_wait_pipe(filp);
3645 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
3650 trace_event_read_lock();
3651 trace_access_lock(iter->cpu_file);
3653 /* Fill as many pages as possible. */
3654 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3655 spd.pages[i] = alloc_page(GFP_KERNEL);
3659 rem = tracing_fill_pipe_page(rem, iter);
3661 /* Copy the data into the page, so we can start over. */
3662 ret = trace_seq_to_buffer(&iter->seq,
3663 page_address(spd.pages[i]),
3666 __free_page(spd.pages[i]);
3669 spd.partial[i].offset = 0;
3670 spd.partial[i].len = iter->seq.len;
3672 trace_seq_init(&iter->seq);
3675 trace_access_unlock(iter->cpu_file);
3676 trace_event_read_unlock();
3677 mutex_unlock(&iter->mutex);
3681 ret = splice_to_pipe(pipe, &spd);
3683 splice_shrink_spd(pipe, &spd);
3687 mutex_unlock(&iter->mutex);
3691 struct ftrace_entries_info {
3692 struct trace_array *tr;
3696 static int tracing_entries_open(struct inode *inode, struct file *filp)
3698 struct ftrace_entries_info *info;
3700 if (tracing_disabled)
3703 info = kzalloc(sizeof(*info), GFP_KERNEL);
3707 info->tr = &global_trace;
3708 info->cpu = (unsigned long)inode->i_private;
3710 filp->private_data = info;
3716 tracing_entries_read(struct file *filp, char __user *ubuf,
3717 size_t cnt, loff_t *ppos)
3719 struct ftrace_entries_info *info = filp->private_data;
3720 struct trace_array *tr = info->tr;
3725 mutex_lock(&trace_types_lock);
3727 if (info->cpu == RING_BUFFER_ALL_CPUS) {
3728 int cpu, buf_size_same;
3733 /* check if all cpu sizes are same */
3734 for_each_tracing_cpu(cpu) {
3735 /* fill in the size from first enabled cpu */
3737 size = tr->data[cpu]->entries;
3738 if (size != tr->data[cpu]->entries) {
3744 if (buf_size_same) {
3745 if (!ring_buffer_expanded)
3746 r = sprintf(buf, "%lu (expanded: %lu)\n",
3748 trace_buf_size >> 10);
3750 r = sprintf(buf, "%lu\n", size >> 10);
3752 r = sprintf(buf, "X\n");
3754 r = sprintf(buf, "%lu\n", tr->data[info->cpu]->entries >> 10);
3756 mutex_unlock(&trace_types_lock);
3758 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3763 tracing_entries_write(struct file *filp, const char __user *ubuf,
3764 size_t cnt, loff_t *ppos)
3766 struct ftrace_entries_info *info = filp->private_data;
3770 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3774 /* must have at least 1 entry */
3778 /* value is in KB */
3781 ret = tracing_resize_ring_buffer(val, info->cpu);
3791 tracing_entries_release(struct inode *inode, struct file *filp)
3793 struct ftrace_entries_info *info = filp->private_data;
3801 tracing_total_entries_read(struct file *filp, char __user *ubuf,
3802 size_t cnt, loff_t *ppos)
3804 struct trace_array *tr = filp->private_data;
3807 unsigned long size = 0, expanded_size = 0;
3809 mutex_lock(&trace_types_lock);
3810 for_each_tracing_cpu(cpu) {
3811 size += tr->data[cpu]->entries >> 10;
3812 if (!ring_buffer_expanded)
3813 expanded_size += trace_buf_size >> 10;
3815 if (ring_buffer_expanded)
3816 r = sprintf(buf, "%lu\n", size);
3818 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
3819 mutex_unlock(&trace_types_lock);
3821 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3825 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
3826 size_t cnt, loff_t *ppos)
3829 * There is no need to read what the user has written, this function
3830 * is just to make sure that there is no error when "echo" is used
3839 tracing_free_buffer_release(struct inode *inode, struct file *filp)
3841 /* disable tracing ? */
3842 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
3844 /* resize the ring buffer to 0 */
3845 tracing_resize_ring_buffer(0, RING_BUFFER_ALL_CPUS);
3851 tracing_mark_write(struct file *filp, const char __user *ubuf,
3852 size_t cnt, loff_t *fpos)
3854 unsigned long addr = (unsigned long)ubuf;
3855 struct ring_buffer_event *event;
3856 struct ring_buffer *buffer;
3857 struct print_entry *entry;
3858 unsigned long irq_flags;
3859 struct page *pages[2];
3869 if (tracing_disabled)
3872 if (cnt > TRACE_BUF_SIZE)
3873 cnt = TRACE_BUF_SIZE;
3876 * Userspace is injecting traces into the kernel trace buffer.
3877 * We want to be as non intrusive as possible.
3878 * To do so, we do not want to allocate any special buffers
3879 * or take any locks, but instead write the userspace data
3880 * straight into the ring buffer.
3882 * First we need to pin the userspace buffer into memory,
3883 * which, most likely it is, because it just referenced it.
3884 * But there's no guarantee that it is. By using get_user_pages_fast()
3885 * and kmap_atomic/kunmap_atomic() we can get access to the
3886 * pages directly. We then write the data directly into the
3889 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
3891 /* check if we cross pages */
3892 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
3895 offset = addr & (PAGE_SIZE - 1);
3898 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
3899 if (ret < nr_pages) {
3901 put_page(pages[ret]);
3906 for (i = 0; i < nr_pages; i++)
3907 map_page[i] = kmap_atomic(pages[i]);
3909 local_save_flags(irq_flags);
3910 size = sizeof(*entry) + cnt + 2; /* possible \n added */
3911 buffer = global_trace.buffer;
3912 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3913 irq_flags, preempt_count());
3915 /* Ring buffer disabled, return as if not open for write */
3920 entry = ring_buffer_event_data(event);
3921 entry->ip = _THIS_IP_;
3923 if (nr_pages == 2) {
3924 len = PAGE_SIZE - offset;
3925 memcpy(&entry->buf, map_page[0] + offset, len);
3926 memcpy(&entry->buf[len], map_page[1], cnt - len);
3928 memcpy(&entry->buf, map_page[0] + offset, cnt);
3930 if (entry->buf[cnt - 1] != '\n') {
3931 entry->buf[cnt] = '\n';
3932 entry->buf[cnt + 1] = '\0';
3934 entry->buf[cnt] = '\0';
3936 ring_buffer_unlock_commit(buffer, event);
3943 for (i = 0; i < nr_pages; i++){
3944 kunmap_atomic(map_page[i]);
3951 static int tracing_clock_show(struct seq_file *m, void *v)
3955 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3957 "%s%s%s%s", i ? " " : "",
3958 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3959 i == trace_clock_id ? "]" : "");
3965 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3966 size_t cnt, loff_t *fpos)
3969 const char *clockstr;
3972 if (cnt >= sizeof(buf))
3975 if (copy_from_user(&buf, ubuf, cnt))
3980 clockstr = strstrip(buf);
3982 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3983 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3986 if (i == ARRAY_SIZE(trace_clocks))
3991 mutex_lock(&trace_types_lock);
3993 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3995 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3997 mutex_unlock(&trace_types_lock);
4004 static int tracing_clock_open(struct inode *inode, struct file *file)
4006 if (tracing_disabled)
4008 return single_open(file, tracing_clock_show, NULL);
4011 static const struct file_operations tracing_max_lat_fops = {
4012 .open = tracing_open_generic,
4013 .read = tracing_max_lat_read,
4014 .write = tracing_max_lat_write,
4015 .llseek = generic_file_llseek,
4018 static const struct file_operations tracing_ctrl_fops = {
4019 .open = tracing_open_generic,
4020 .read = tracing_ctrl_read,
4021 .write = tracing_ctrl_write,
4022 .llseek = generic_file_llseek,
4025 static const struct file_operations set_tracer_fops = {
4026 .open = tracing_open_generic,
4027 .read = tracing_set_trace_read,
4028 .write = tracing_set_trace_write,
4029 .llseek = generic_file_llseek,
4032 static const struct file_operations tracing_pipe_fops = {
4033 .open = tracing_open_pipe,
4034 .poll = tracing_poll_pipe,
4035 .read = tracing_read_pipe,
4036 .splice_read = tracing_splice_read_pipe,
4037 .release = tracing_release_pipe,
4038 .llseek = no_llseek,
4041 static const struct file_operations tracing_entries_fops = {
4042 .open = tracing_entries_open,
4043 .read = tracing_entries_read,
4044 .write = tracing_entries_write,
4045 .release = tracing_entries_release,
4046 .llseek = generic_file_llseek,
4049 static const struct file_operations tracing_total_entries_fops = {
4050 .open = tracing_open_generic,
4051 .read = tracing_total_entries_read,
4052 .llseek = generic_file_llseek,
4055 static const struct file_operations tracing_free_buffer_fops = {
4056 .write = tracing_free_buffer_write,
4057 .release = tracing_free_buffer_release,
4060 static const struct file_operations tracing_mark_fops = {
4061 .open = tracing_open_generic,
4062 .write = tracing_mark_write,
4063 .llseek = generic_file_llseek,
4066 static const struct file_operations trace_clock_fops = {
4067 .open = tracing_clock_open,
4069 .llseek = seq_lseek,
4070 .release = single_release,
4071 .write = tracing_clock_write,
4074 struct ftrace_buffer_info {
4075 struct trace_array *tr;
4081 static int tracing_buffers_open(struct inode *inode, struct file *filp)
4083 int cpu = (int)(long)inode->i_private;
4084 struct ftrace_buffer_info *info;
4086 if (tracing_disabled)
4089 info = kzalloc(sizeof(*info), GFP_KERNEL);
4093 info->tr = &global_trace;
4096 /* Force reading ring buffer for first read */
4097 info->read = (unsigned int)-1;
4099 filp->private_data = info;
4101 return nonseekable_open(inode, filp);
4105 tracing_buffers_read(struct file *filp, char __user *ubuf,
4106 size_t count, loff_t *ppos)
4108 struct ftrace_buffer_info *info = filp->private_data;
4116 info->spare = ring_buffer_alloc_read_page(info->tr->buffer, info->cpu);
4120 /* Do we have previous read data to read? */
4121 if (info->read < PAGE_SIZE)
4124 trace_access_lock(info->cpu);
4125 ret = ring_buffer_read_page(info->tr->buffer,
4129 trace_access_unlock(info->cpu);
4136 size = PAGE_SIZE - info->read;
4140 ret = copy_to_user(ubuf, info->spare + info->read, size);
4151 static int tracing_buffers_release(struct inode *inode, struct file *file)
4153 struct ftrace_buffer_info *info = file->private_data;
4156 ring_buffer_free_read_page(info->tr->buffer, info->spare);
4163 struct ring_buffer *buffer;
4168 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
4169 struct pipe_buffer *buf)
4171 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4176 ring_buffer_free_read_page(ref->buffer, ref->page);
4181 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
4182 struct pipe_buffer *buf)
4187 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
4188 struct pipe_buffer *buf)
4190 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4195 /* Pipe buffer operations for a buffer. */
4196 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
4198 .map = generic_pipe_buf_map,
4199 .unmap = generic_pipe_buf_unmap,
4200 .confirm = generic_pipe_buf_confirm,
4201 .release = buffer_pipe_buf_release,
4202 .steal = buffer_pipe_buf_steal,
4203 .get = buffer_pipe_buf_get,
4207 * Callback from splice_to_pipe(), if we need to release some pages
4208 * at the end of the spd in case we error'ed out in filling the pipe.
4210 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
4212 struct buffer_ref *ref =
4213 (struct buffer_ref *)spd->partial[i].private;
4218 ring_buffer_free_read_page(ref->buffer, ref->page);
4220 spd->partial[i].private = 0;
4224 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4225 struct pipe_inode_info *pipe, size_t len,
4228 struct ftrace_buffer_info *info = file->private_data;
4229 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4230 struct page *pages_def[PIPE_DEF_BUFFERS];
4231 struct splice_pipe_desc spd = {
4233 .partial = partial_def,
4235 .ops = &buffer_pipe_buf_ops,
4236 .spd_release = buffer_spd_release,
4238 struct buffer_ref *ref;
4239 int entries, size, i;
4242 if (splice_grow_spd(pipe, &spd))
4245 if (*ppos & (PAGE_SIZE - 1)) {
4246 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
4251 if (len & (PAGE_SIZE - 1)) {
4252 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
4253 if (len < PAGE_SIZE) {
4260 trace_access_lock(info->cpu);
4261 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
4263 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
4267 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
4272 ref->buffer = info->tr->buffer;
4273 ref->page = ring_buffer_alloc_read_page(ref->buffer, info->cpu);
4279 r = ring_buffer_read_page(ref->buffer, &ref->page,
4282 ring_buffer_free_read_page(ref->buffer, ref->page);
4288 * zero out any left over data, this is going to
4291 size = ring_buffer_page_len(ref->page);
4292 if (size < PAGE_SIZE)
4293 memset(ref->page + size, 0, PAGE_SIZE - size);
4295 page = virt_to_page(ref->page);
4297 spd.pages[i] = page;
4298 spd.partial[i].len = PAGE_SIZE;
4299 spd.partial[i].offset = 0;
4300 spd.partial[i].private = (unsigned long)ref;
4304 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
4307 trace_access_unlock(info->cpu);
4310 /* did we read anything? */
4311 if (!spd.nr_pages) {
4312 if (flags & SPLICE_F_NONBLOCK)
4320 ret = splice_to_pipe(pipe, &spd);
4321 splice_shrink_spd(pipe, &spd);
4326 static const struct file_operations tracing_buffers_fops = {
4327 .open = tracing_buffers_open,
4328 .read = tracing_buffers_read,
4329 .release = tracing_buffers_release,
4330 .splice_read = tracing_buffers_splice_read,
4331 .llseek = no_llseek,
4335 tracing_stats_read(struct file *filp, char __user *ubuf,
4336 size_t count, loff_t *ppos)
4338 unsigned long cpu = (unsigned long)filp->private_data;
4339 struct trace_array *tr = &global_trace;
4340 struct trace_seq *s;
4342 unsigned long long t;
4343 unsigned long usec_rem;
4345 s = kmalloc(sizeof(*s), GFP_KERNEL);
4351 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
4352 trace_seq_printf(s, "entries: %ld\n", cnt);
4354 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
4355 trace_seq_printf(s, "overrun: %ld\n", cnt);
4357 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
4358 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
4360 cnt = ring_buffer_bytes_cpu(tr->buffer, cpu);
4361 trace_seq_printf(s, "bytes: %ld\n", cnt);
4363 t = ns2usecs(ring_buffer_oldest_event_ts(tr->buffer, cpu));
4364 usec_rem = do_div(t, USEC_PER_SEC);
4365 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n", t, usec_rem);
4367 t = ns2usecs(ring_buffer_time_stamp(tr->buffer, cpu));
4368 usec_rem = do_div(t, USEC_PER_SEC);
4369 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
4371 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
4378 static const struct file_operations tracing_stats_fops = {
4379 .open = tracing_open_generic,
4380 .read = tracing_stats_read,
4381 .llseek = generic_file_llseek,
4384 #ifdef CONFIG_DYNAMIC_FTRACE
4386 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
4392 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
4393 size_t cnt, loff_t *ppos)
4395 static char ftrace_dyn_info_buffer[1024];
4396 static DEFINE_MUTEX(dyn_info_mutex);
4397 unsigned long *p = filp->private_data;
4398 char *buf = ftrace_dyn_info_buffer;
4399 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
4402 mutex_lock(&dyn_info_mutex);
4403 r = sprintf(buf, "%ld ", *p);
4405 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
4408 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4410 mutex_unlock(&dyn_info_mutex);
4415 static const struct file_operations tracing_dyn_info_fops = {
4416 .open = tracing_open_generic,
4417 .read = tracing_read_dyn_info,
4418 .llseek = generic_file_llseek,
4422 static struct dentry *d_tracer;
4424 struct dentry *tracing_init_dentry(void)
4431 if (!debugfs_initialized())
4434 d_tracer = debugfs_create_dir("tracing", NULL);
4436 if (!d_tracer && !once) {
4438 pr_warning("Could not create debugfs directory 'tracing'\n");
4445 static struct dentry *d_percpu;
4447 struct dentry *tracing_dentry_percpu(void)
4450 struct dentry *d_tracer;
4455 d_tracer = tracing_init_dentry();
4460 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
4462 if (!d_percpu && !once) {
4464 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4471 static void tracing_init_debugfs_percpu(long cpu)
4473 struct dentry *d_percpu = tracing_dentry_percpu();
4474 struct dentry *d_cpu;
4475 char cpu_dir[30]; /* 30 characters should be more than enough */
4480 snprintf(cpu_dir, 30, "cpu%ld", cpu);
4481 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4483 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4487 /* per cpu trace_pipe */
4488 trace_create_file("trace_pipe", 0444, d_cpu,
4489 (void *) cpu, &tracing_pipe_fops);
4492 trace_create_file("trace", 0644, d_cpu,
4493 (void *) cpu, &tracing_fops);
4495 trace_create_file("trace_pipe_raw", 0444, d_cpu,
4496 (void *) cpu, &tracing_buffers_fops);
4498 trace_create_file("stats", 0444, d_cpu,
4499 (void *) cpu, &tracing_stats_fops);
4501 trace_create_file("buffer_size_kb", 0444, d_cpu,
4502 (void *) cpu, &tracing_entries_fops);
4505 #ifdef CONFIG_FTRACE_SELFTEST
4506 /* Let selftest have access to static functions in this file */
4507 #include "trace_selftest.c"
4510 struct trace_option_dentry {
4511 struct tracer_opt *opt;
4512 struct tracer_flags *flags;
4513 struct dentry *entry;
4517 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4520 struct trace_option_dentry *topt = filp->private_data;
4523 if (topt->flags->val & topt->opt->bit)
4528 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4532 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4535 struct trace_option_dentry *topt = filp->private_data;
4539 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4543 if (val != 0 && val != 1)
4546 if (!!(topt->flags->val & topt->opt->bit) != val) {
4547 mutex_lock(&trace_types_lock);
4548 ret = __set_tracer_option(current_trace, topt->flags,
4550 mutex_unlock(&trace_types_lock);
4561 static const struct file_operations trace_options_fops = {
4562 .open = tracing_open_generic,
4563 .read = trace_options_read,
4564 .write = trace_options_write,
4565 .llseek = generic_file_llseek,
4569 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4572 long index = (long)filp->private_data;
4575 if (trace_flags & (1 << index))
4580 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4584 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4587 long index = (long)filp->private_data;
4591 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4595 if (val != 0 && val != 1)
4597 set_tracer_flags(1 << index, val);
4604 static const struct file_operations trace_options_core_fops = {
4605 .open = tracing_open_generic,
4606 .read = trace_options_core_read,
4607 .write = trace_options_core_write,
4608 .llseek = generic_file_llseek,
4611 struct dentry *trace_create_file(const char *name,
4613 struct dentry *parent,
4615 const struct file_operations *fops)
4619 ret = debugfs_create_file(name, mode, parent, data, fops);
4621 pr_warning("Could not create debugfs '%s' entry\n", name);
4627 static struct dentry *trace_options_init_dentry(void)
4629 struct dentry *d_tracer;
4630 static struct dentry *t_options;
4635 d_tracer = tracing_init_dentry();
4639 t_options = debugfs_create_dir("options", d_tracer);
4641 pr_warning("Could not create debugfs directory 'options'\n");
4649 create_trace_option_file(struct trace_option_dentry *topt,
4650 struct tracer_flags *flags,
4651 struct tracer_opt *opt)
4653 struct dentry *t_options;
4655 t_options = trace_options_init_dentry();
4659 topt->flags = flags;
4662 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4663 &trace_options_fops);
4667 static struct trace_option_dentry *
4668 create_trace_option_files(struct tracer *tracer)
4670 struct trace_option_dentry *topts;
4671 struct tracer_flags *flags;
4672 struct tracer_opt *opts;
4678 flags = tracer->flags;
4680 if (!flags || !flags->opts)
4685 for (cnt = 0; opts[cnt].name; cnt++)
4688 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4692 for (cnt = 0; opts[cnt].name; cnt++)
4693 create_trace_option_file(&topts[cnt], flags,
4700 destroy_trace_option_files(struct trace_option_dentry *topts)
4707 for (cnt = 0; topts[cnt].opt; cnt++) {
4708 if (topts[cnt].entry)
4709 debugfs_remove(topts[cnt].entry);
4715 static struct dentry *
4716 create_trace_option_core_file(const char *option, long index)
4718 struct dentry *t_options;
4720 t_options = trace_options_init_dentry();
4724 return trace_create_file(option, 0644, t_options, (void *)index,
4725 &trace_options_core_fops);
4728 static __init void create_trace_options_dir(void)
4730 struct dentry *t_options;
4733 t_options = trace_options_init_dentry();
4737 for (i = 0; trace_options[i]; i++)
4738 create_trace_option_core_file(trace_options[i], i);
4742 rb_simple_read(struct file *filp, char __user *ubuf,
4743 size_t cnt, loff_t *ppos)
4745 struct trace_array *tr = filp->private_data;
4746 struct ring_buffer *buffer = tr->buffer;
4751 r = ring_buffer_record_is_on(buffer);
4755 r = sprintf(buf, "%d\n", r);
4757 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4761 rb_simple_write(struct file *filp, const char __user *ubuf,
4762 size_t cnt, loff_t *ppos)
4764 struct trace_array *tr = filp->private_data;
4765 struct ring_buffer *buffer = tr->buffer;
4769 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4775 ring_buffer_record_on(buffer);
4777 ring_buffer_record_off(buffer);
4785 static const struct file_operations rb_simple_fops = {
4786 .open = tracing_open_generic,
4787 .read = rb_simple_read,
4788 .write = rb_simple_write,
4789 .llseek = default_llseek,
4792 static __init int tracer_init_debugfs(void)
4794 struct dentry *d_tracer;
4797 trace_access_lock_init();
4799 d_tracer = tracing_init_dentry();
4801 trace_create_file("tracing_enabled", 0644, d_tracer,
4802 &global_trace, &tracing_ctrl_fops);
4804 trace_create_file("trace_options", 0644, d_tracer,
4805 NULL, &tracing_iter_fops);
4807 trace_create_file("tracing_cpumask", 0644, d_tracer,
4808 NULL, &tracing_cpumask_fops);
4810 trace_create_file("trace", 0644, d_tracer,
4811 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4813 trace_create_file("available_tracers", 0444, d_tracer,
4814 &global_trace, &show_traces_fops);
4816 trace_create_file("current_tracer", 0644, d_tracer,
4817 &global_trace, &set_tracer_fops);
4819 #ifdef CONFIG_TRACER_MAX_TRACE
4820 trace_create_file("tracing_max_latency", 0644, d_tracer,
4821 &tracing_max_latency, &tracing_max_lat_fops);
4824 trace_create_file("tracing_thresh", 0644, d_tracer,
4825 &tracing_thresh, &tracing_max_lat_fops);
4827 trace_create_file("README", 0444, d_tracer,
4828 NULL, &tracing_readme_fops);
4830 trace_create_file("trace_pipe", 0444, d_tracer,
4831 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4833 trace_create_file("buffer_size_kb", 0644, d_tracer,
4834 (void *) RING_BUFFER_ALL_CPUS, &tracing_entries_fops);
4836 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
4837 &global_trace, &tracing_total_entries_fops);
4839 trace_create_file("free_buffer", 0644, d_tracer,
4840 &global_trace, &tracing_free_buffer_fops);
4842 trace_create_file("trace_marker", 0220, d_tracer,
4843 NULL, &tracing_mark_fops);
4845 trace_create_file("saved_cmdlines", 0444, d_tracer,
4846 NULL, &tracing_saved_cmdlines_fops);
4848 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4851 trace_create_file("tracing_on", 0644, d_tracer,
4852 &global_trace, &rb_simple_fops);
4854 #ifdef CONFIG_DYNAMIC_FTRACE
4855 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4856 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4859 create_trace_options_dir();
4861 for_each_tracing_cpu(cpu)
4862 tracing_init_debugfs_percpu(cpu);
4867 static int trace_panic_handler(struct notifier_block *this,
4868 unsigned long event, void *unused)
4870 if (ftrace_dump_on_oops)
4871 ftrace_dump(ftrace_dump_on_oops);
4875 static struct notifier_block trace_panic_notifier = {
4876 .notifier_call = trace_panic_handler,
4878 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4881 static int trace_die_handler(struct notifier_block *self,
4887 if (ftrace_dump_on_oops)
4888 ftrace_dump(ftrace_dump_on_oops);
4896 static struct notifier_block trace_die_notifier = {
4897 .notifier_call = trace_die_handler,
4902 * printk is set to max of 1024, we really don't need it that big.
4903 * Nothing should be printing 1000 characters anyway.
4905 #define TRACE_MAX_PRINT 1000
4908 * Define here KERN_TRACE so that we have one place to modify
4909 * it if we decide to change what log level the ftrace dump
4912 #define KERN_TRACE KERN_EMERG
4915 trace_printk_seq(struct trace_seq *s)
4917 /* Probably should print a warning here. */
4921 /* should be zero ended, but we are paranoid. */
4922 s->buffer[s->len] = 0;
4924 printk(KERN_TRACE "%s", s->buffer);
4929 void trace_init_global_iter(struct trace_iterator *iter)
4931 iter->tr = &global_trace;
4932 iter->trace = current_trace;
4933 iter->cpu_file = TRACE_PIPE_ALL_CPU;
4937 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4939 static arch_spinlock_t ftrace_dump_lock =
4940 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4941 /* use static because iter can be a bit big for the stack */
4942 static struct trace_iterator iter;
4943 unsigned int old_userobj;
4944 static int dump_ran;
4945 unsigned long flags;
4949 local_irq_save(flags);
4950 arch_spin_lock(&ftrace_dump_lock);
4958 /* Did function tracer already get disabled? */
4959 if (ftrace_is_dead()) {
4960 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
4961 printk("# MAY BE MISSING FUNCTION EVENTS\n");
4964 if (disable_tracing)
4967 trace_init_global_iter(&iter);
4969 for_each_tracing_cpu(cpu) {
4970 atomic_inc(&iter.tr->data[cpu]->disabled);
4973 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4975 /* don't look at user memory in panic mode */
4976 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4978 /* Simulate the iterator */
4979 iter.tr = &global_trace;
4980 iter.trace = current_trace;
4982 switch (oops_dump_mode) {
4984 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4987 iter.cpu_file = raw_smp_processor_id();
4992 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
4993 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4996 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4999 * We need to stop all tracing on all CPUS to read the
5000 * the next buffer. This is a bit expensive, but is
5001 * not done often. We fill all what we can read,
5002 * and then release the locks again.
5005 while (!trace_empty(&iter)) {
5008 printk(KERN_TRACE "---------------------------------\n");
5012 /* reset all but tr, trace, and overruns */
5013 memset(&iter.seq, 0,
5014 sizeof(struct trace_iterator) -
5015 offsetof(struct trace_iterator, seq));
5016 iter.iter_flags |= TRACE_FILE_LAT_FMT;
5019 if (trace_find_next_entry_inc(&iter) != NULL) {
5022 ret = print_trace_line(&iter);
5023 if (ret != TRACE_TYPE_NO_CONSUME)
5024 trace_consume(&iter);
5026 touch_nmi_watchdog();
5028 trace_printk_seq(&iter.seq);
5032 printk(KERN_TRACE " (ftrace buffer empty)\n");
5034 printk(KERN_TRACE "---------------------------------\n");
5037 /* Re-enable tracing if requested */
5038 if (!disable_tracing) {
5039 trace_flags |= old_userobj;
5041 for_each_tracing_cpu(cpu) {
5042 atomic_dec(&iter.tr->data[cpu]->disabled);
5048 arch_spin_unlock(&ftrace_dump_lock);
5049 local_irq_restore(flags);
5052 /* By default: disable tracing after the dump */
5053 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
5055 __ftrace_dump(true, oops_dump_mode);
5057 EXPORT_SYMBOL_GPL(ftrace_dump);
5059 __init static int tracer_alloc_buffers(void)
5062 enum ring_buffer_flags rb_flags;
5067 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
5070 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
5071 goto out_free_buffer_mask;
5073 /* Only allocate trace_printk buffers if a trace_printk exists */
5074 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
5075 trace_printk_init_buffers();
5077 /* To save memory, keep the ring buffer size to its minimum */
5078 if (ring_buffer_expanded)
5079 ring_buf_size = trace_buf_size;
5083 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
5085 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
5086 cpumask_copy(tracing_cpumask, cpu_all_mask);
5088 /* TODO: make the number of buffers hot pluggable with CPUS */
5089 global_trace.buffer = ring_buffer_alloc(ring_buf_size, rb_flags);
5090 if (!global_trace.buffer) {
5091 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
5093 goto out_free_cpumask;
5095 if (global_trace.buffer_disabled)
5099 #ifdef CONFIG_TRACER_MAX_TRACE
5100 max_tr.buffer = ring_buffer_alloc(1, rb_flags);
5101 if (!max_tr.buffer) {
5102 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
5104 ring_buffer_free(global_trace.buffer);
5105 goto out_free_cpumask;
5109 /* Allocate the first page for all buffers */
5110 for_each_tracing_cpu(i) {
5111 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
5112 max_tr.data[i] = &per_cpu(max_tr_data, i);
5115 set_buffer_entries(&global_trace,
5116 ring_buffer_size(global_trace.buffer, 0));
5117 #ifdef CONFIG_TRACER_MAX_TRACE
5118 set_buffer_entries(&max_tr, 1);
5121 trace_init_cmdlines();
5123 register_tracer(&nop_trace);
5124 current_trace = &nop_trace;
5125 /* All seems OK, enable tracing */
5126 tracing_disabled = 0;
5128 atomic_notifier_chain_register(&panic_notifier_list,
5129 &trace_panic_notifier);
5131 register_die_notifier(&trace_die_notifier);
5136 free_cpumask_var(tracing_cpumask);
5137 out_free_buffer_mask:
5138 free_cpumask_var(tracing_buffer_mask);
5143 __init static int clear_boot_tracer(void)
5146 * The default tracer at boot buffer is an init section.
5147 * This function is called in lateinit. If we did not
5148 * find the boot tracer, then clear it out, to prevent
5149 * later registration from accessing the buffer that is
5150 * about to be freed.
5152 if (!default_bootup_tracer)
5155 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
5156 default_bootup_tracer);
5157 default_bootup_tracer = NULL;
5162 early_initcall(tracer_alloc_buffers);
5163 fs_initcall(tracer_init_debugfs);
5164 late_initcall(clear_boot_tracer);