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_off(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");
833 /* Add the warning after printing 'FAILED' */
837 /* Only reset on passing, to avoid touching corrupted buffers */
838 tracing_reset_online_cpus(tr);
840 /* Shrink the max buffer again */
841 if (ring_buffer_expanded && type->use_max_tr)
842 ring_buffer_resize(max_tr.buffer, 1,
843 RING_BUFFER_ALL_CPUS);
845 printk(KERN_CONT "PASSED\n");
849 type->next = trace_types;
853 tracing_selftest_running = false;
854 mutex_unlock(&trace_types_lock);
856 if (ret || !default_bootup_tracer)
859 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
862 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
863 /* Do we want this tracer to start on bootup? */
864 tracing_set_tracer(type->name);
865 default_bootup_tracer = NULL;
866 /* disable other selftests, since this will break it. */
867 tracing_selftest_disabled = 1;
868 #ifdef CONFIG_FTRACE_STARTUP_TEST
869 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
877 void unregister_tracer(struct tracer *type)
881 mutex_lock(&trace_types_lock);
882 for (t = &trace_types; *t; t = &(*t)->next) {
886 pr_info("Tracer %s not registered\n", type->name);
892 if (type == current_trace && tracer_enabled) {
895 if (current_trace->stop)
896 current_trace->stop(&global_trace);
897 current_trace = &nop_trace;
900 mutex_unlock(&trace_types_lock);
903 void tracing_reset(struct trace_array *tr, int cpu)
905 struct ring_buffer *buffer = tr->buffer;
907 ring_buffer_record_disable(buffer);
909 /* Make sure all commits have finished */
911 ring_buffer_reset_cpu(buffer, cpu);
913 ring_buffer_record_enable(buffer);
916 void tracing_reset_online_cpus(struct trace_array *tr)
918 struct ring_buffer *buffer = tr->buffer;
921 ring_buffer_record_disable(buffer);
923 /* Make sure all commits have finished */
926 tr->time_start = ftrace_now(tr->cpu);
928 for_each_online_cpu(cpu)
929 ring_buffer_reset_cpu(buffer, cpu);
931 ring_buffer_record_enable(buffer);
934 void tracing_reset_current(int cpu)
936 tracing_reset(&global_trace, cpu);
939 void tracing_reset_current_online_cpus(void)
941 tracing_reset_online_cpus(&global_trace);
944 #define SAVED_CMDLINES 128
945 #define NO_CMDLINE_MAP UINT_MAX
946 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
947 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
948 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
949 static int cmdline_idx;
950 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
952 /* temporary disable recording */
953 static atomic_t trace_record_cmdline_disabled __read_mostly;
955 static void trace_init_cmdlines(void)
957 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
958 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
962 int is_tracing_stopped(void)
964 return trace_stop_count;
968 * ftrace_off_permanent - disable all ftrace code permanently
970 * This should only be called when a serious anomally has
971 * been detected. This will turn off the function tracing,
972 * ring buffers, and other tracing utilites. It takes no
973 * locks and can be called from any context.
975 void ftrace_off_permanent(void)
977 tracing_disabled = 1;
979 tracing_off_permanent();
983 * tracing_start - quick start of the tracer
985 * If tracing is enabled but was stopped by tracing_stop,
986 * this will start the tracer back up.
988 void tracing_start(void)
990 struct ring_buffer *buffer;
993 if (tracing_disabled)
996 raw_spin_lock_irqsave(&tracing_start_lock, flags);
997 if (--trace_stop_count) {
998 if (trace_stop_count < 0) {
999 /* Someone screwed up their debugging */
1001 trace_stop_count = 0;
1006 /* Prevent the buffers from switching */
1007 arch_spin_lock(&ftrace_max_lock);
1009 buffer = global_trace.buffer;
1011 ring_buffer_record_enable(buffer);
1013 buffer = max_tr.buffer;
1015 ring_buffer_record_enable(buffer);
1017 arch_spin_unlock(&ftrace_max_lock);
1021 raw_spin_unlock_irqrestore(&tracing_start_lock, flags);
1025 * tracing_stop - quick stop of the tracer
1027 * Light weight way to stop tracing. Use in conjunction with
1030 void tracing_stop(void)
1032 struct ring_buffer *buffer;
1033 unsigned long flags;
1036 raw_spin_lock_irqsave(&tracing_start_lock, flags);
1037 if (trace_stop_count++)
1040 /* Prevent the buffers from switching */
1041 arch_spin_lock(&ftrace_max_lock);
1043 buffer = global_trace.buffer;
1045 ring_buffer_record_disable(buffer);
1047 buffer = max_tr.buffer;
1049 ring_buffer_record_disable(buffer);
1051 arch_spin_unlock(&ftrace_max_lock);
1054 raw_spin_unlock_irqrestore(&tracing_start_lock, flags);
1057 void trace_stop_cmdline_recording(void);
1059 static void trace_save_cmdline(struct task_struct *tsk)
1063 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1067 * It's not the end of the world if we don't get
1068 * the lock, but we also don't want to spin
1069 * nor do we want to disable interrupts,
1070 * so if we miss here, then better luck next time.
1072 if (!arch_spin_trylock(&trace_cmdline_lock))
1075 idx = map_pid_to_cmdline[tsk->pid];
1076 if (idx == NO_CMDLINE_MAP) {
1077 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1080 * Check whether the cmdline buffer at idx has a pid
1081 * mapped. We are going to overwrite that entry so we
1082 * need to clear the map_pid_to_cmdline. Otherwise we
1083 * would read the new comm for the old pid.
1085 pid = map_cmdline_to_pid[idx];
1086 if (pid != NO_CMDLINE_MAP)
1087 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1089 map_cmdline_to_pid[idx] = tsk->pid;
1090 map_pid_to_cmdline[tsk->pid] = idx;
1095 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1097 arch_spin_unlock(&trace_cmdline_lock);
1100 void trace_find_cmdline(int pid, char comm[])
1105 strcpy(comm, "<idle>");
1109 if (WARN_ON_ONCE(pid < 0)) {
1110 strcpy(comm, "<XXX>");
1114 if (pid > PID_MAX_DEFAULT) {
1115 strcpy(comm, "<...>");
1120 arch_spin_lock(&trace_cmdline_lock);
1121 map = map_pid_to_cmdline[pid];
1122 if (map != NO_CMDLINE_MAP)
1123 strcpy(comm, saved_cmdlines[map]);
1125 strcpy(comm, "<...>");
1127 arch_spin_unlock(&trace_cmdline_lock);
1131 void tracing_record_cmdline(struct task_struct *tsk)
1133 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1137 trace_save_cmdline(tsk);
1141 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1144 struct task_struct *tsk = current;
1146 entry->preempt_count = pc & 0xff;
1147 entry->pid = (tsk) ? tsk->pid : 0;
1150 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1151 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1153 TRACE_FLAG_IRQS_NOSUPPORT |
1155 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1156 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1157 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1159 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1161 struct ring_buffer_event *
1162 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1165 unsigned long flags, int pc)
1167 struct ring_buffer_event *event;
1169 event = ring_buffer_lock_reserve(buffer, len);
1170 if (event != NULL) {
1171 struct trace_entry *ent = ring_buffer_event_data(event);
1173 tracing_generic_entry_update(ent, flags, pc);
1181 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1182 struct ring_buffer_event *event,
1183 unsigned long flags, int pc,
1186 ring_buffer_unlock_commit(buffer, event);
1188 ftrace_trace_stack(buffer, flags, 6, pc);
1189 ftrace_trace_userstack(buffer, flags, pc);
1195 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1196 struct ring_buffer_event *event,
1197 unsigned long flags, int pc)
1199 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1202 struct ring_buffer_event *
1203 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1204 int type, unsigned long len,
1205 unsigned long flags, int pc)
1207 *current_rb = global_trace.buffer;
1208 return trace_buffer_lock_reserve(*current_rb,
1209 type, len, flags, pc);
1211 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1213 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1214 struct ring_buffer_event *event,
1215 unsigned long flags, int pc)
1217 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1219 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1221 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1222 struct ring_buffer_event *event,
1223 unsigned long flags, int pc)
1225 __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1227 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1229 void trace_nowake_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1230 struct ring_buffer_event *event,
1231 unsigned long flags, int pc,
1232 struct pt_regs *regs)
1234 ring_buffer_unlock_commit(buffer, event);
1236 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1237 ftrace_trace_userstack(buffer, flags, pc);
1239 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit_regs);
1241 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1242 struct ring_buffer_event *event)
1244 ring_buffer_discard_commit(buffer, event);
1246 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1249 trace_function(struct trace_array *tr,
1250 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1253 struct ftrace_event_call *call = &event_function;
1254 struct ring_buffer *buffer = tr->buffer;
1255 struct ring_buffer_event *event;
1256 struct ftrace_entry *entry;
1258 /* If we are reading the ring buffer, don't trace */
1259 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1262 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1266 entry = ring_buffer_event_data(event);
1268 entry->parent_ip = parent_ip;
1270 if (!filter_check_discard(call, entry, buffer, event))
1271 ring_buffer_unlock_commit(buffer, event);
1275 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1276 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1279 if (likely(!atomic_read(&data->disabled)))
1280 trace_function(tr, ip, parent_ip, flags, pc);
1283 #ifdef CONFIG_STACKTRACE
1285 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1286 struct ftrace_stack {
1287 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1290 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1291 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1293 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1294 unsigned long flags,
1295 int skip, int pc, struct pt_regs *regs)
1297 struct ftrace_event_call *call = &event_kernel_stack;
1298 struct ring_buffer_event *event;
1299 struct stack_entry *entry;
1300 struct stack_trace trace;
1302 int size = FTRACE_STACK_ENTRIES;
1304 trace.nr_entries = 0;
1308 * Since events can happen in NMIs there's no safe way to
1309 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1310 * or NMI comes in, it will just have to use the default
1311 * FTRACE_STACK_SIZE.
1313 preempt_disable_notrace();
1315 use_stack = ++__get_cpu_var(ftrace_stack_reserve);
1317 * We don't need any atomic variables, just a barrier.
1318 * If an interrupt comes in, we don't care, because it would
1319 * have exited and put the counter back to what we want.
1320 * We just need a barrier to keep gcc from moving things
1324 if (use_stack == 1) {
1325 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1326 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1329 save_stack_trace_regs(regs, &trace);
1331 save_stack_trace(&trace);
1333 if (trace.nr_entries > size)
1334 size = trace.nr_entries;
1336 /* From now on, use_stack is a boolean */
1339 size *= sizeof(unsigned long);
1341 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1342 sizeof(*entry) + size, flags, pc);
1345 entry = ring_buffer_event_data(event);
1347 memset(&entry->caller, 0, size);
1350 memcpy(&entry->caller, trace.entries,
1351 trace.nr_entries * sizeof(unsigned long));
1353 trace.max_entries = FTRACE_STACK_ENTRIES;
1354 trace.entries = entry->caller;
1356 save_stack_trace_regs(regs, &trace);
1358 save_stack_trace(&trace);
1361 entry->size = trace.nr_entries;
1363 if (!filter_check_discard(call, entry, buffer, event))
1364 ring_buffer_unlock_commit(buffer, event);
1367 /* Again, don't let gcc optimize things here */
1369 __get_cpu_var(ftrace_stack_reserve)--;
1370 preempt_enable_notrace();
1374 void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1375 int skip, int pc, struct pt_regs *regs)
1377 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1380 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1383 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1386 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1389 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1392 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1395 __ftrace_trace_stack(tr->buffer, flags, skip, pc, NULL);
1399 * trace_dump_stack - record a stack back trace in the trace buffer
1401 void trace_dump_stack(void)
1403 unsigned long flags;
1405 if (tracing_disabled || tracing_selftest_running)
1408 local_save_flags(flags);
1410 /* skipping 3 traces, seems to get us at the caller of this function */
1411 __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count(), NULL);
1414 static DEFINE_PER_CPU(int, user_stack_count);
1417 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1419 struct ftrace_event_call *call = &event_user_stack;
1420 struct ring_buffer_event *event;
1421 struct userstack_entry *entry;
1422 struct stack_trace trace;
1424 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1428 * NMIs can not handle page faults, even with fix ups.
1429 * The save user stack can (and often does) fault.
1431 if (unlikely(in_nmi()))
1435 * prevent recursion, since the user stack tracing may
1436 * trigger other kernel events.
1439 if (__this_cpu_read(user_stack_count))
1442 __this_cpu_inc(user_stack_count);
1444 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1445 sizeof(*entry), flags, pc);
1447 goto out_drop_count;
1448 entry = ring_buffer_event_data(event);
1450 entry->tgid = current->tgid;
1451 memset(&entry->caller, 0, sizeof(entry->caller));
1453 trace.nr_entries = 0;
1454 trace.max_entries = FTRACE_STACK_ENTRIES;
1456 trace.entries = entry->caller;
1458 save_stack_trace_user(&trace);
1459 if (!filter_check_discard(call, entry, buffer, event))
1460 ring_buffer_unlock_commit(buffer, event);
1463 __this_cpu_dec(user_stack_count);
1469 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1471 ftrace_trace_userstack(tr, flags, preempt_count());
1475 #endif /* CONFIG_STACKTRACE */
1477 /* created for use with alloc_percpu */
1478 struct trace_buffer_struct {
1479 char buffer[TRACE_BUF_SIZE];
1482 static struct trace_buffer_struct *trace_percpu_buffer;
1483 static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1484 static struct trace_buffer_struct *trace_percpu_irq_buffer;
1485 static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1488 * The buffer used is dependent on the context. There is a per cpu
1489 * buffer for normal context, softirq contex, hard irq context and
1490 * for NMI context. Thise allows for lockless recording.
1492 * Note, if the buffers failed to be allocated, then this returns NULL
1494 static char *get_trace_buf(void)
1496 struct trace_buffer_struct *percpu_buffer;
1497 struct trace_buffer_struct *buffer;
1500 * If we have allocated per cpu buffers, then we do not
1501 * need to do any locking.
1504 percpu_buffer = trace_percpu_nmi_buffer;
1506 percpu_buffer = trace_percpu_irq_buffer;
1507 else if (in_softirq())
1508 percpu_buffer = trace_percpu_sirq_buffer;
1510 percpu_buffer = trace_percpu_buffer;
1515 buffer = per_cpu_ptr(percpu_buffer, smp_processor_id());
1517 return buffer->buffer;
1520 static int alloc_percpu_trace_buffer(void)
1522 struct trace_buffer_struct *buffers;
1523 struct trace_buffer_struct *sirq_buffers;
1524 struct trace_buffer_struct *irq_buffers;
1525 struct trace_buffer_struct *nmi_buffers;
1527 buffers = alloc_percpu(struct trace_buffer_struct);
1531 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
1535 irq_buffers = alloc_percpu(struct trace_buffer_struct);
1539 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
1543 trace_percpu_buffer = buffers;
1544 trace_percpu_sirq_buffer = sirq_buffers;
1545 trace_percpu_irq_buffer = irq_buffers;
1546 trace_percpu_nmi_buffer = nmi_buffers;
1551 free_percpu(irq_buffers);
1553 free_percpu(sirq_buffers);
1555 free_percpu(buffers);
1557 WARN(1, "Could not allocate percpu trace_printk buffer");
1561 void trace_printk_init_buffers(void)
1563 static int buffers_allocated;
1565 if (buffers_allocated)
1568 if (alloc_percpu_trace_buffer())
1571 pr_info("ftrace: Allocated trace_printk buffers\n");
1573 buffers_allocated = 1;
1577 * trace_vbprintk - write binary msg to tracing buffer
1580 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1582 struct ftrace_event_call *call = &event_bprint;
1583 struct ring_buffer_event *event;
1584 struct ring_buffer *buffer;
1585 struct trace_array *tr = &global_trace;
1586 struct bprint_entry *entry;
1587 unsigned long flags;
1589 int len = 0, size, pc;
1591 if (unlikely(tracing_selftest_running || tracing_disabled))
1594 /* Don't pollute graph traces with trace_vprintk internals */
1595 pause_graph_tracing();
1597 pc = preempt_count();
1598 preempt_disable_notrace();
1600 tbuffer = get_trace_buf();
1606 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
1608 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
1611 local_save_flags(flags);
1612 size = sizeof(*entry) + sizeof(u32) * len;
1613 buffer = tr->buffer;
1614 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1618 entry = ring_buffer_event_data(event);
1622 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
1623 if (!filter_check_discard(call, entry, buffer, event)) {
1624 ring_buffer_unlock_commit(buffer, event);
1625 ftrace_trace_stack(buffer, flags, 6, pc);
1629 preempt_enable_notrace();
1630 unpause_graph_tracing();
1634 EXPORT_SYMBOL_GPL(trace_vbprintk);
1636 int trace_array_printk(struct trace_array *tr,
1637 unsigned long ip, const char *fmt, ...)
1642 if (!(trace_flags & TRACE_ITER_PRINTK))
1646 ret = trace_array_vprintk(tr, ip, fmt, ap);
1651 int trace_array_vprintk(struct trace_array *tr,
1652 unsigned long ip, const char *fmt, va_list args)
1654 struct ftrace_event_call *call = &event_print;
1655 struct ring_buffer_event *event;
1656 struct ring_buffer *buffer;
1657 int len = 0, size, pc;
1658 struct print_entry *entry;
1659 unsigned long flags;
1662 if (tracing_disabled || tracing_selftest_running)
1665 /* Don't pollute graph traces with trace_vprintk internals */
1666 pause_graph_tracing();
1668 pc = preempt_count();
1669 preempt_disable_notrace();
1672 tbuffer = get_trace_buf();
1678 len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
1679 if (len > TRACE_BUF_SIZE)
1682 local_save_flags(flags);
1683 size = sizeof(*entry) + len + 1;
1684 buffer = tr->buffer;
1685 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1689 entry = ring_buffer_event_data(event);
1692 memcpy(&entry->buf, tbuffer, len);
1693 entry->buf[len] = '\0';
1694 if (!filter_check_discard(call, entry, buffer, event)) {
1695 ring_buffer_unlock_commit(buffer, event);
1696 ftrace_trace_stack(buffer, flags, 6, pc);
1699 preempt_enable_notrace();
1700 unpause_graph_tracing();
1705 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1707 return trace_array_vprintk(&global_trace, ip, fmt, args);
1709 EXPORT_SYMBOL_GPL(trace_vprintk);
1711 static void trace_iterator_increment(struct trace_iterator *iter)
1713 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
1717 ring_buffer_read(buf_iter, NULL);
1720 static struct trace_entry *
1721 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1722 unsigned long *lost_events)
1724 struct ring_buffer_event *event;
1725 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
1728 event = ring_buffer_iter_peek(buf_iter, ts);
1730 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1734 iter->ent_size = ring_buffer_event_length(event);
1735 return ring_buffer_event_data(event);
1741 static struct trace_entry *
1742 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1743 unsigned long *missing_events, u64 *ent_ts)
1745 struct ring_buffer *buffer = iter->tr->buffer;
1746 struct trace_entry *ent, *next = NULL;
1747 unsigned long lost_events = 0, next_lost = 0;
1748 int cpu_file = iter->cpu_file;
1749 u64 next_ts = 0, ts;
1755 * If we are in a per_cpu trace file, don't bother by iterating over
1756 * all cpu and peek directly.
1758 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1759 if (ring_buffer_empty_cpu(buffer, cpu_file))
1761 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1763 *ent_cpu = cpu_file;
1768 for_each_tracing_cpu(cpu) {
1770 if (ring_buffer_empty_cpu(buffer, cpu))
1773 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1776 * Pick the entry with the smallest timestamp:
1778 if (ent && (!next || ts < next_ts)) {
1782 next_lost = lost_events;
1783 next_size = iter->ent_size;
1787 iter->ent_size = next_size;
1790 *ent_cpu = next_cpu;
1796 *missing_events = next_lost;
1801 /* Find the next real entry, without updating the iterator itself */
1802 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1803 int *ent_cpu, u64 *ent_ts)
1805 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1808 /* Find the next real entry, and increment the iterator to the next entry */
1809 void *trace_find_next_entry_inc(struct trace_iterator *iter)
1811 iter->ent = __find_next_entry(iter, &iter->cpu,
1812 &iter->lost_events, &iter->ts);
1815 trace_iterator_increment(iter);
1817 return iter->ent ? iter : NULL;
1820 static void trace_consume(struct trace_iterator *iter)
1822 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1823 &iter->lost_events);
1826 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1828 struct trace_iterator *iter = m->private;
1832 WARN_ON_ONCE(iter->leftover);
1836 /* can't go backwards */
1841 ent = trace_find_next_entry_inc(iter);
1845 while (ent && iter->idx < i)
1846 ent = trace_find_next_entry_inc(iter);
1853 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1855 struct trace_array *tr = iter->tr;
1856 struct ring_buffer_event *event;
1857 struct ring_buffer_iter *buf_iter;
1858 unsigned long entries = 0;
1861 tr->data[cpu]->skipped_entries = 0;
1863 buf_iter = trace_buffer_iter(iter, cpu);
1867 ring_buffer_iter_reset(buf_iter);
1870 * We could have the case with the max latency tracers
1871 * that a reset never took place on a cpu. This is evident
1872 * by the timestamp being before the start of the buffer.
1874 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1875 if (ts >= iter->tr->time_start)
1878 ring_buffer_read(buf_iter, NULL);
1881 tr->data[cpu]->skipped_entries = entries;
1885 * The current tracer is copied to avoid a global locking
1888 static void *s_start(struct seq_file *m, loff_t *pos)
1890 struct trace_iterator *iter = m->private;
1891 static struct tracer *old_tracer;
1892 int cpu_file = iter->cpu_file;
1897 /* copy the tracer to avoid using a global lock all around */
1898 mutex_lock(&trace_types_lock);
1899 if (unlikely(old_tracer != current_trace && current_trace)) {
1900 old_tracer = current_trace;
1901 *iter->trace = *current_trace;
1903 mutex_unlock(&trace_types_lock);
1905 atomic_inc(&trace_record_cmdline_disabled);
1907 if (*pos != iter->pos) {
1912 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1913 for_each_tracing_cpu(cpu)
1914 tracing_iter_reset(iter, cpu);
1916 tracing_iter_reset(iter, cpu_file);
1919 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1924 * If we overflowed the seq_file before, then we want
1925 * to just reuse the trace_seq buffer again.
1931 p = s_next(m, p, &l);
1935 trace_event_read_lock();
1936 trace_access_lock(cpu_file);
1940 static void s_stop(struct seq_file *m, void *p)
1942 struct trace_iterator *iter = m->private;
1944 atomic_dec(&trace_record_cmdline_disabled);
1945 trace_access_unlock(iter->cpu_file);
1946 trace_event_read_unlock();
1950 get_total_entries(struct trace_array *tr, unsigned long *total, unsigned long *entries)
1952 unsigned long count;
1958 for_each_tracing_cpu(cpu) {
1959 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1961 * If this buffer has skipped entries, then we hold all
1962 * entries for the trace and we need to ignore the
1963 * ones before the time stamp.
1965 if (tr->data[cpu]->skipped_entries) {
1966 count -= tr->data[cpu]->skipped_entries;
1967 /* total is the same as the entries */
1971 ring_buffer_overrun_cpu(tr->buffer, cpu);
1976 static void print_lat_help_header(struct seq_file *m)
1978 seq_puts(m, "# _------=> CPU# \n");
1979 seq_puts(m, "# / _-----=> irqs-off \n");
1980 seq_puts(m, "# | / _----=> need-resched \n");
1981 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1982 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1983 seq_puts(m, "# |||| / delay \n");
1984 seq_puts(m, "# cmd pid ||||| time | caller \n");
1985 seq_puts(m, "# \\ / ||||| \\ | / \n");
1988 static void print_event_info(struct trace_array *tr, struct seq_file *m)
1990 unsigned long total;
1991 unsigned long entries;
1993 get_total_entries(tr, &total, &entries);
1994 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
1995 entries, total, num_online_cpus());
1999 static void print_func_help_header(struct trace_array *tr, struct seq_file *m)
2001 print_event_info(tr, m);
2002 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2003 seq_puts(m, "# | | | | |\n");
2006 static void print_func_help_header_irq(struct trace_array *tr, struct seq_file *m)
2008 print_event_info(tr, m);
2009 seq_puts(m, "# _-----=> irqs-off\n");
2010 seq_puts(m, "# / _----=> need-resched\n");
2011 seq_puts(m, "# | / _---=> hardirq/softirq\n");
2012 seq_puts(m, "# || / _--=> preempt-depth\n");
2013 seq_puts(m, "# ||| / delay\n");
2014 seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2015 seq_puts(m, "# | | | |||| | |\n");
2019 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2021 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2022 struct trace_array *tr = iter->tr;
2023 struct trace_array_cpu *data = tr->data[tr->cpu];
2024 struct tracer *type = current_trace;
2025 unsigned long entries;
2026 unsigned long total;
2027 const char *name = "preemption";
2032 get_total_entries(tr, &total, &entries);
2034 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2036 seq_puts(m, "# -----------------------------------"
2037 "---------------------------------\n");
2038 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2039 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2040 nsecs_to_usecs(data->saved_latency),
2044 #if defined(CONFIG_PREEMPT_NONE)
2046 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2048 #elif defined(CONFIG_PREEMPT)
2053 /* These are reserved for later use */
2056 seq_printf(m, " #P:%d)\n", num_online_cpus());
2060 seq_puts(m, "# -----------------\n");
2061 seq_printf(m, "# | task: %.16s-%d "
2062 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2063 data->comm, data->pid, data->uid, data->nice,
2064 data->policy, data->rt_priority);
2065 seq_puts(m, "# -----------------\n");
2067 if (data->critical_start) {
2068 seq_puts(m, "# => started at: ");
2069 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2070 trace_print_seq(m, &iter->seq);
2071 seq_puts(m, "\n# => ended at: ");
2072 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2073 trace_print_seq(m, &iter->seq);
2074 seq_puts(m, "\n#\n");
2080 static void test_cpu_buff_start(struct trace_iterator *iter)
2082 struct trace_seq *s = &iter->seq;
2084 if (!(trace_flags & TRACE_ITER_ANNOTATE))
2087 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2090 if (cpumask_test_cpu(iter->cpu, iter->started))
2093 if (iter->tr->data[iter->cpu]->skipped_entries)
2096 cpumask_set_cpu(iter->cpu, iter->started);
2098 /* Don't print started cpu buffer for the first entry of the trace */
2100 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2104 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2106 struct trace_seq *s = &iter->seq;
2107 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2108 struct trace_entry *entry;
2109 struct trace_event *event;
2113 test_cpu_buff_start(iter);
2115 event = ftrace_find_event(entry->type);
2117 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2118 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2119 if (!trace_print_lat_context(iter))
2122 if (!trace_print_context(iter))
2128 return event->funcs->trace(iter, sym_flags, event);
2130 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2133 return TRACE_TYPE_HANDLED;
2135 return TRACE_TYPE_PARTIAL_LINE;
2138 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2140 struct trace_seq *s = &iter->seq;
2141 struct trace_entry *entry;
2142 struct trace_event *event;
2146 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2147 if (!trace_seq_printf(s, "%d %d %llu ",
2148 entry->pid, iter->cpu, iter->ts))
2152 event = ftrace_find_event(entry->type);
2154 return event->funcs->raw(iter, 0, event);
2156 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2159 return TRACE_TYPE_HANDLED;
2161 return TRACE_TYPE_PARTIAL_LINE;
2164 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2166 struct trace_seq *s = &iter->seq;
2167 unsigned char newline = '\n';
2168 struct trace_entry *entry;
2169 struct trace_event *event;
2173 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2174 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2175 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2176 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2179 event = ftrace_find_event(entry->type);
2181 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2182 if (ret != TRACE_TYPE_HANDLED)
2186 SEQ_PUT_FIELD_RET(s, newline);
2188 return TRACE_TYPE_HANDLED;
2191 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2193 struct trace_seq *s = &iter->seq;
2194 struct trace_entry *entry;
2195 struct trace_event *event;
2199 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2200 SEQ_PUT_FIELD_RET(s, entry->pid);
2201 SEQ_PUT_FIELD_RET(s, iter->cpu);
2202 SEQ_PUT_FIELD_RET(s, iter->ts);
2205 event = ftrace_find_event(entry->type);
2206 return event ? event->funcs->binary(iter, 0, event) :
2210 int trace_empty(struct trace_iterator *iter)
2212 struct ring_buffer_iter *buf_iter;
2215 /* If we are looking at one CPU buffer, only check that one */
2216 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
2217 cpu = iter->cpu_file;
2218 buf_iter = trace_buffer_iter(iter, cpu);
2220 if (!ring_buffer_iter_empty(buf_iter))
2223 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2229 for_each_tracing_cpu(cpu) {
2230 buf_iter = trace_buffer_iter(iter, cpu);
2232 if (!ring_buffer_iter_empty(buf_iter))
2235 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2243 /* Called with trace_event_read_lock() held. */
2244 enum print_line_t print_trace_line(struct trace_iterator *iter)
2246 enum print_line_t ret;
2248 if (iter->lost_events &&
2249 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2250 iter->cpu, iter->lost_events))
2251 return TRACE_TYPE_PARTIAL_LINE;
2253 if (iter->trace && iter->trace->print_line) {
2254 ret = iter->trace->print_line(iter);
2255 if (ret != TRACE_TYPE_UNHANDLED)
2259 if (iter->ent->type == TRACE_BPRINT &&
2260 trace_flags & TRACE_ITER_PRINTK &&
2261 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2262 return trace_print_bprintk_msg_only(iter);
2264 if (iter->ent->type == TRACE_PRINT &&
2265 trace_flags & TRACE_ITER_PRINTK &&
2266 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2267 return trace_print_printk_msg_only(iter);
2269 if (trace_flags & TRACE_ITER_BIN)
2270 return print_bin_fmt(iter);
2272 if (trace_flags & TRACE_ITER_HEX)
2273 return print_hex_fmt(iter);
2275 if (trace_flags & TRACE_ITER_RAW)
2276 return print_raw_fmt(iter);
2278 return print_trace_fmt(iter);
2281 void trace_latency_header(struct seq_file *m)
2283 struct trace_iterator *iter = m->private;
2285 /* print nothing if the buffers are empty */
2286 if (trace_empty(iter))
2289 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2290 print_trace_header(m, iter);
2292 if (!(trace_flags & TRACE_ITER_VERBOSE))
2293 print_lat_help_header(m);
2296 void trace_default_header(struct seq_file *m)
2298 struct trace_iterator *iter = m->private;
2300 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2303 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2304 /* print nothing if the buffers are empty */
2305 if (trace_empty(iter))
2307 print_trace_header(m, iter);
2308 if (!(trace_flags & TRACE_ITER_VERBOSE))
2309 print_lat_help_header(m);
2311 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2312 if (trace_flags & TRACE_ITER_IRQ_INFO)
2313 print_func_help_header_irq(iter->tr, m);
2315 print_func_help_header(iter->tr, m);
2320 static void test_ftrace_alive(struct seq_file *m)
2322 if (!ftrace_is_dead())
2324 seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2325 seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2328 static int s_show(struct seq_file *m, void *v)
2330 struct trace_iterator *iter = v;
2333 if (iter->ent == NULL) {
2335 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2337 test_ftrace_alive(m);
2339 if (iter->trace && iter->trace->print_header)
2340 iter->trace->print_header(m);
2342 trace_default_header(m);
2344 } else if (iter->leftover) {
2346 * If we filled the seq_file buffer earlier, we
2347 * want to just show it now.
2349 ret = trace_print_seq(m, &iter->seq);
2351 /* ret should this time be zero, but you never know */
2352 iter->leftover = ret;
2355 print_trace_line(iter);
2356 ret = trace_print_seq(m, &iter->seq);
2358 * If we overflow the seq_file buffer, then it will
2359 * ask us for this data again at start up.
2361 * ret is 0 if seq_file write succeeded.
2364 iter->leftover = ret;
2370 static const struct seq_operations tracer_seq_ops = {
2377 static struct trace_iterator *
2378 __tracing_open(struct inode *inode, struct file *file)
2380 long cpu_file = (long) inode->i_private;
2381 struct trace_iterator *iter;
2384 if (tracing_disabled)
2385 return ERR_PTR(-ENODEV);
2387 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
2389 return ERR_PTR(-ENOMEM);
2391 iter->buffer_iter = kzalloc(sizeof(*iter->buffer_iter) * num_possible_cpus(),
2394 * We make a copy of the current tracer to avoid concurrent
2395 * changes on it while we are reading.
2397 mutex_lock(&trace_types_lock);
2398 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2403 *iter->trace = *current_trace;
2405 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2408 if (current_trace && current_trace->print_max)
2411 iter->tr = &global_trace;
2413 mutex_init(&iter->mutex);
2414 iter->cpu_file = cpu_file;
2416 /* Notify the tracer early; before we stop tracing. */
2417 if (iter->trace && iter->trace->open)
2418 iter->trace->open(iter);
2420 /* Annotate start of buffers if we had overruns */
2421 if (ring_buffer_overruns(iter->tr->buffer))
2422 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2424 /* stop the trace while dumping */
2427 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2428 for_each_tracing_cpu(cpu) {
2429 iter->buffer_iter[cpu] =
2430 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2432 ring_buffer_read_prepare_sync();
2433 for_each_tracing_cpu(cpu) {
2434 ring_buffer_read_start(iter->buffer_iter[cpu]);
2435 tracing_iter_reset(iter, cpu);
2438 cpu = iter->cpu_file;
2439 iter->buffer_iter[cpu] =
2440 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2441 ring_buffer_read_prepare_sync();
2442 ring_buffer_read_start(iter->buffer_iter[cpu]);
2443 tracing_iter_reset(iter, cpu);
2446 mutex_unlock(&trace_types_lock);
2451 mutex_unlock(&trace_types_lock);
2453 kfree(iter->buffer_iter);
2454 seq_release_private(inode, file);
2455 return ERR_PTR(-ENOMEM);
2458 int tracing_open_generic(struct inode *inode, struct file *filp)
2460 if (tracing_disabled)
2463 filp->private_data = inode->i_private;
2467 static int tracing_release(struct inode *inode, struct file *file)
2469 struct seq_file *m = file->private_data;
2470 struct trace_iterator *iter;
2473 if (!(file->f_mode & FMODE_READ))
2478 mutex_lock(&trace_types_lock);
2479 for_each_tracing_cpu(cpu) {
2480 if (iter->buffer_iter[cpu])
2481 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2484 if (iter->trace && iter->trace->close)
2485 iter->trace->close(iter);
2487 /* reenable tracing if it was previously enabled */
2489 mutex_unlock(&trace_types_lock);
2491 mutex_destroy(&iter->mutex);
2492 free_cpumask_var(iter->started);
2494 kfree(iter->buffer_iter);
2495 seq_release_private(inode, file);
2499 static int tracing_open(struct inode *inode, struct file *file)
2501 struct trace_iterator *iter;
2504 /* If this file was open for write, then erase contents */
2505 if ((file->f_mode & FMODE_WRITE) &&
2506 (file->f_flags & O_TRUNC)) {
2507 long cpu = (long) inode->i_private;
2509 if (cpu == TRACE_PIPE_ALL_CPU)
2510 tracing_reset_online_cpus(&global_trace);
2512 tracing_reset(&global_trace, cpu);
2515 if (file->f_mode & FMODE_READ) {
2516 iter = __tracing_open(inode, file);
2518 ret = PTR_ERR(iter);
2519 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2520 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2526 t_next(struct seq_file *m, void *v, loff_t *pos)
2528 struct tracer *t = v;
2538 static void *t_start(struct seq_file *m, loff_t *pos)
2543 mutex_lock(&trace_types_lock);
2544 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2550 static void t_stop(struct seq_file *m, void *p)
2552 mutex_unlock(&trace_types_lock);
2555 static int t_show(struct seq_file *m, void *v)
2557 struct tracer *t = v;
2562 seq_printf(m, "%s", t->name);
2571 static const struct seq_operations show_traces_seq_ops = {
2578 static int show_traces_open(struct inode *inode, struct file *file)
2580 if (tracing_disabled)
2583 return seq_open(file, &show_traces_seq_ops);
2587 tracing_write_stub(struct file *filp, const char __user *ubuf,
2588 size_t count, loff_t *ppos)
2593 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
2595 if (file->f_mode & FMODE_READ)
2596 return seq_lseek(file, offset, origin);
2601 static const struct file_operations tracing_fops = {
2602 .open = tracing_open,
2604 .write = tracing_write_stub,
2605 .llseek = tracing_seek,
2606 .release = tracing_release,
2609 static const struct file_operations show_traces_fops = {
2610 .open = show_traces_open,
2612 .release = seq_release,
2613 .llseek = seq_lseek,
2617 * Only trace on a CPU if the bitmask is set:
2619 static cpumask_var_t tracing_cpumask;
2622 * The tracer itself will not take this lock, but still we want
2623 * to provide a consistent cpumask to user-space:
2625 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2628 * Temporary storage for the character representation of the
2629 * CPU bitmask (and one more byte for the newline):
2631 static char mask_str[NR_CPUS + 1];
2634 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2635 size_t count, loff_t *ppos)
2639 mutex_lock(&tracing_cpumask_update_lock);
2641 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2642 if (count - len < 2) {
2646 len += sprintf(mask_str + len, "\n");
2647 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2650 mutex_unlock(&tracing_cpumask_update_lock);
2656 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2657 size_t count, loff_t *ppos)
2660 cpumask_var_t tracing_cpumask_new;
2662 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2665 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2669 mutex_lock(&tracing_cpumask_update_lock);
2671 local_irq_disable();
2672 arch_spin_lock(&ftrace_max_lock);
2673 for_each_tracing_cpu(cpu) {
2675 * Increase/decrease the disabled counter if we are
2676 * about to flip a bit in the cpumask:
2678 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2679 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2680 atomic_inc(&global_trace.data[cpu]->disabled);
2681 ring_buffer_record_disable_cpu(global_trace.buffer, cpu);
2683 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2684 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2685 atomic_dec(&global_trace.data[cpu]->disabled);
2686 ring_buffer_record_enable_cpu(global_trace.buffer, cpu);
2689 arch_spin_unlock(&ftrace_max_lock);
2692 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2694 mutex_unlock(&tracing_cpumask_update_lock);
2695 free_cpumask_var(tracing_cpumask_new);
2700 free_cpumask_var(tracing_cpumask_new);
2705 static const struct file_operations tracing_cpumask_fops = {
2706 .open = tracing_open_generic,
2707 .read = tracing_cpumask_read,
2708 .write = tracing_cpumask_write,
2709 .llseek = generic_file_llseek,
2712 static int tracing_trace_options_show(struct seq_file *m, void *v)
2714 struct tracer_opt *trace_opts;
2718 mutex_lock(&trace_types_lock);
2719 tracer_flags = current_trace->flags->val;
2720 trace_opts = current_trace->flags->opts;
2722 for (i = 0; trace_options[i]; i++) {
2723 if (trace_flags & (1 << i))
2724 seq_printf(m, "%s\n", trace_options[i]);
2726 seq_printf(m, "no%s\n", trace_options[i]);
2729 for (i = 0; trace_opts[i].name; i++) {
2730 if (tracer_flags & trace_opts[i].bit)
2731 seq_printf(m, "%s\n", trace_opts[i].name);
2733 seq_printf(m, "no%s\n", trace_opts[i].name);
2735 mutex_unlock(&trace_types_lock);
2740 static int __set_tracer_option(struct tracer *trace,
2741 struct tracer_flags *tracer_flags,
2742 struct tracer_opt *opts, int neg)
2746 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2751 tracer_flags->val &= ~opts->bit;
2753 tracer_flags->val |= opts->bit;
2757 /* Try to assign a tracer specific option */
2758 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2760 struct tracer_flags *tracer_flags = trace->flags;
2761 struct tracer_opt *opts = NULL;
2764 for (i = 0; tracer_flags->opts[i].name; i++) {
2765 opts = &tracer_flags->opts[i];
2767 if (strcmp(cmp, opts->name) == 0)
2768 return __set_tracer_option(trace, trace->flags,
2775 static void set_tracer_flags(unsigned int mask, int enabled)
2777 /* do nothing if flag is already set */
2778 if (!!(trace_flags & mask) == !!enabled)
2782 trace_flags |= mask;
2784 trace_flags &= ~mask;
2786 if (mask == TRACE_ITER_RECORD_CMD)
2787 trace_event_enable_cmd_record(enabled);
2789 if (mask == TRACE_ITER_OVERWRITE)
2790 ring_buffer_change_overwrite(global_trace.buffer, enabled);
2794 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2795 size_t cnt, loff_t *ppos)
2803 if (cnt >= sizeof(buf))
2806 if (copy_from_user(&buf, ubuf, cnt))
2810 cmp = strstrip(buf);
2812 if (strncmp(cmp, "no", 2) == 0) {
2817 for (i = 0; trace_options[i]; i++) {
2818 if (strcmp(cmp, trace_options[i]) == 0) {
2819 set_tracer_flags(1 << i, !neg);
2824 /* If no option could be set, test the specific tracer options */
2825 if (!trace_options[i]) {
2826 mutex_lock(&trace_types_lock);
2827 ret = set_tracer_option(current_trace, cmp, neg);
2828 mutex_unlock(&trace_types_lock);
2838 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2840 if (tracing_disabled)
2842 return single_open(file, tracing_trace_options_show, NULL);
2845 static const struct file_operations tracing_iter_fops = {
2846 .open = tracing_trace_options_open,
2848 .llseek = seq_lseek,
2849 .release = single_release,
2850 .write = tracing_trace_options_write,
2853 static const char readme_msg[] =
2854 "tracing mini-HOWTO:\n\n"
2855 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2856 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2857 "wakeup wakeup_rt preemptirqsoff preemptoff irqsoff function nop\n\n"
2858 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2860 "# echo wakeup > /sys/kernel/debug/tracing/current_tracer\n"
2861 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2863 "# cat /sys/kernel/debug/tracing/trace_options\n"
2864 "noprint-parent nosym-offset nosym-addr noverbose\n"
2865 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2866 "# echo 1 > /sys/kernel/debug/tracing/tracing_on\n"
2867 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2868 "# echo 0 > /sys/kernel/debug/tracing/tracing_on\n"
2872 tracing_readme_read(struct file *filp, char __user *ubuf,
2873 size_t cnt, loff_t *ppos)
2875 return simple_read_from_buffer(ubuf, cnt, ppos,
2876 readme_msg, strlen(readme_msg));
2879 static const struct file_operations tracing_readme_fops = {
2880 .open = tracing_open_generic,
2881 .read = tracing_readme_read,
2882 .llseek = generic_file_llseek,
2886 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2887 size_t cnt, loff_t *ppos)
2896 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2900 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2908 for (i = 0; i < SAVED_CMDLINES; i++) {
2911 pid = map_cmdline_to_pid[i];
2912 if (pid == -1 || pid == NO_CMDLINE_MAP)
2915 trace_find_cmdline(pid, buf_comm);
2916 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2921 len = simple_read_from_buffer(ubuf, cnt, ppos,
2930 static const struct file_operations tracing_saved_cmdlines_fops = {
2931 .open = tracing_open_generic,
2932 .read = tracing_saved_cmdlines_read,
2933 .llseek = generic_file_llseek,
2937 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2938 size_t cnt, loff_t *ppos)
2943 r = sprintf(buf, "%u\n", tracer_enabled);
2944 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2948 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2949 size_t cnt, loff_t *ppos)
2951 struct trace_array *tr = filp->private_data;
2955 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
2961 mutex_lock(&trace_types_lock);
2962 if (tracer_enabled ^ val) {
2964 /* Only need to warn if this is used to change the state */
2965 WARN_ONCE(1, "tracing_enabled is deprecated. Use tracing_on");
2969 if (current_trace->start)
2970 current_trace->start(tr);
2975 if (current_trace->stop)
2976 current_trace->stop(tr);
2979 mutex_unlock(&trace_types_lock);
2987 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2988 size_t cnt, loff_t *ppos)
2990 char buf[MAX_TRACER_SIZE+2];
2993 mutex_lock(&trace_types_lock);
2995 r = sprintf(buf, "%s\n", current_trace->name);
2997 r = sprintf(buf, "\n");
2998 mutex_unlock(&trace_types_lock);
3000 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3003 int tracer_init(struct tracer *t, struct trace_array *tr)
3005 tracing_reset_online_cpus(tr);
3009 static void set_buffer_entries(struct trace_array *tr, unsigned long val)
3012 for_each_tracing_cpu(cpu)
3013 tr->data[cpu]->entries = val;
3016 static int __tracing_resize_ring_buffer(unsigned long size, int cpu)
3021 * If kernel or user changes the size of the ring buffer
3022 * we use the size that was given, and we can forget about
3023 * expanding it later.
3025 ring_buffer_expanded = 1;
3027 ret = ring_buffer_resize(global_trace.buffer, size, cpu);
3031 if (!current_trace->use_max_tr)
3034 ret = ring_buffer_resize(max_tr.buffer, size, cpu);
3038 if (cpu == RING_BUFFER_ALL_CPUS) {
3040 for_each_tracing_cpu(i) {
3041 r = ring_buffer_resize(global_trace.buffer,
3042 global_trace.data[i]->entries,
3048 r = ring_buffer_resize(global_trace.buffer,
3049 global_trace.data[cpu]->entries,
3055 * AARGH! We are left with different
3056 * size max buffer!!!!
3057 * The max buffer is our "snapshot" buffer.
3058 * When a tracer needs a snapshot (one of the
3059 * latency tracers), it swaps the max buffer
3060 * with the saved snap shot. We succeeded to
3061 * update the size of the main buffer, but failed to
3062 * update the size of the max buffer. But when we tried
3063 * to reset the main buffer to the original size, we
3064 * failed there too. This is very unlikely to
3065 * happen, but if it does, warn and kill all
3069 tracing_disabled = 1;
3074 if (cpu == RING_BUFFER_ALL_CPUS)
3075 set_buffer_entries(&max_tr, size);
3077 max_tr.data[cpu]->entries = size;
3080 if (cpu == RING_BUFFER_ALL_CPUS)
3081 set_buffer_entries(&global_trace, size);
3083 global_trace.data[cpu]->entries = size;
3088 static ssize_t tracing_resize_ring_buffer(unsigned long size, int cpu_id)
3092 mutex_lock(&trace_types_lock);
3094 if (cpu_id != RING_BUFFER_ALL_CPUS) {
3095 /* make sure, this cpu is enabled in the mask */
3096 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
3102 ret = __tracing_resize_ring_buffer(size, cpu_id);
3107 mutex_unlock(&trace_types_lock);
3114 * tracing_update_buffers - used by tracing facility to expand ring buffers
3116 * To save on memory when the tracing is never used on a system with it
3117 * configured in. The ring buffers are set to a minimum size. But once
3118 * a user starts to use the tracing facility, then they need to grow
3119 * to their default size.
3121 * This function is to be called when a tracer is about to be used.
3123 int tracing_update_buffers(void)
3127 mutex_lock(&trace_types_lock);
3128 if (!ring_buffer_expanded)
3129 ret = __tracing_resize_ring_buffer(trace_buf_size,
3130 RING_BUFFER_ALL_CPUS);
3131 mutex_unlock(&trace_types_lock);
3136 struct trace_option_dentry;
3138 static struct trace_option_dentry *
3139 create_trace_option_files(struct tracer *tracer);
3142 destroy_trace_option_files(struct trace_option_dentry *topts);
3144 static int tracing_set_tracer(const char *buf)
3146 static struct trace_option_dentry *topts;
3147 struct trace_array *tr = &global_trace;
3151 mutex_lock(&trace_types_lock);
3153 if (!ring_buffer_expanded) {
3154 ret = __tracing_resize_ring_buffer(trace_buf_size,
3155 RING_BUFFER_ALL_CPUS);
3161 for (t = trace_types; t; t = t->next) {
3162 if (strcmp(t->name, buf) == 0)
3169 if (t == current_trace)
3172 trace_branch_disable();
3173 if (current_trace && current_trace->reset)
3174 current_trace->reset(tr);
3175 if (current_trace && current_trace->use_max_tr) {
3177 * We don't free the ring buffer. instead, resize it because
3178 * The max_tr ring buffer has some state (e.g. ring->clock) and
3179 * we want preserve it.
3181 ring_buffer_resize(max_tr.buffer, 1, RING_BUFFER_ALL_CPUS);
3182 set_buffer_entries(&max_tr, 1);
3184 destroy_trace_option_files(topts);
3188 topts = create_trace_option_files(current_trace);
3189 if (current_trace->use_max_tr) {
3191 /* we need to make per cpu buffer sizes equivalent */
3192 for_each_tracing_cpu(cpu) {
3193 ret = ring_buffer_resize(max_tr.buffer,
3194 global_trace.data[cpu]->entries,
3198 max_tr.data[cpu]->entries =
3199 global_trace.data[cpu]->entries;
3204 ret = tracer_init(t, tr);
3209 trace_branch_enable(tr);
3211 mutex_unlock(&trace_types_lock);
3217 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3218 size_t cnt, loff_t *ppos)
3220 char buf[MAX_TRACER_SIZE+1];
3227 if (cnt > MAX_TRACER_SIZE)
3228 cnt = MAX_TRACER_SIZE;
3230 if (copy_from_user(&buf, ubuf, cnt))
3235 /* strip ending whitespace. */
3236 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3239 err = tracing_set_tracer(buf);
3249 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3250 size_t cnt, loff_t *ppos)
3252 unsigned long *ptr = filp->private_data;
3256 r = snprintf(buf, sizeof(buf), "%ld\n",
3257 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3258 if (r > sizeof(buf))
3260 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3264 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3265 size_t cnt, loff_t *ppos)
3267 unsigned long *ptr = filp->private_data;
3271 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3280 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3282 long cpu_file = (long) inode->i_private;
3283 struct trace_iterator *iter;
3286 if (tracing_disabled)
3289 mutex_lock(&trace_types_lock);
3291 /* create a buffer to store the information to pass to userspace */
3292 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3299 * We make a copy of the current tracer to avoid concurrent
3300 * changes on it while we are reading.
3302 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3308 *iter->trace = *current_trace;
3310 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3315 /* trace pipe does not show start of buffer */
3316 cpumask_setall(iter->started);
3318 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3319 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3321 iter->cpu_file = cpu_file;
3322 iter->tr = &global_trace;
3323 mutex_init(&iter->mutex);
3324 filp->private_data = iter;
3326 if (iter->trace->pipe_open)
3327 iter->trace->pipe_open(iter);
3329 nonseekable_open(inode, filp);
3331 mutex_unlock(&trace_types_lock);
3337 mutex_unlock(&trace_types_lock);
3341 static int tracing_release_pipe(struct inode *inode, struct file *file)
3343 struct trace_iterator *iter = file->private_data;
3345 mutex_lock(&trace_types_lock);
3347 if (iter->trace->pipe_close)
3348 iter->trace->pipe_close(iter);
3350 mutex_unlock(&trace_types_lock);
3352 free_cpumask_var(iter->started);
3353 mutex_destroy(&iter->mutex);
3361 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3363 struct trace_iterator *iter = filp->private_data;
3365 if (trace_flags & TRACE_ITER_BLOCK) {
3367 * Always select as readable when in blocking mode
3369 return POLLIN | POLLRDNORM;
3371 if (!trace_empty(iter))
3372 return POLLIN | POLLRDNORM;
3373 poll_wait(filp, &trace_wait, poll_table);
3374 if (!trace_empty(iter))
3375 return POLLIN | POLLRDNORM;
3382 void default_wait_pipe(struct trace_iterator *iter)
3386 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3388 if (trace_empty(iter))
3391 finish_wait(&trace_wait, &wait);
3395 * This is a make-shift waitqueue.
3396 * A tracer might use this callback on some rare cases:
3398 * 1) the current tracer might hold the runqueue lock when it wakes up
3399 * a reader, hence a deadlock (sched, function, and function graph tracers)
3400 * 2) the function tracers, trace all functions, we don't want
3401 * the overhead of calling wake_up and friends
3402 * (and tracing them too)
3404 * Anyway, this is really very primitive wakeup.
3406 void poll_wait_pipe(struct trace_iterator *iter)
3408 set_current_state(TASK_INTERRUPTIBLE);
3409 /* sleep for 100 msecs, and try again. */
3410 schedule_timeout(HZ / 10);
3413 /* Must be called with trace_types_lock mutex held. */
3414 static int tracing_wait_pipe(struct file *filp)
3416 struct trace_iterator *iter = filp->private_data;
3418 while (trace_empty(iter)) {
3420 if ((filp->f_flags & O_NONBLOCK)) {
3424 mutex_unlock(&iter->mutex);
3426 iter->trace->wait_pipe(iter);
3428 mutex_lock(&iter->mutex);
3430 if (signal_pending(current))
3434 * We block until we read something and tracing is disabled.
3435 * We still block if tracing is disabled, but we have never
3436 * read anything. This allows a user to cat this file, and
3437 * then enable tracing. But after we have read something,
3438 * we give an EOF when tracing is again disabled.
3440 * iter->pos will be 0 if we haven't read anything.
3442 if (!tracer_enabled && iter->pos)
3453 tracing_read_pipe(struct file *filp, char __user *ubuf,
3454 size_t cnt, loff_t *ppos)
3456 struct trace_iterator *iter = filp->private_data;
3457 static struct tracer *old_tracer;
3460 /* return any leftover data */
3461 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3465 trace_seq_init(&iter->seq);
3467 /* copy the tracer to avoid using a global lock all around */
3468 mutex_lock(&trace_types_lock);
3469 if (unlikely(old_tracer != current_trace && current_trace)) {
3470 old_tracer = current_trace;
3471 *iter->trace = *current_trace;
3473 mutex_unlock(&trace_types_lock);
3476 * Avoid more than one consumer on a single file descriptor
3477 * This is just a matter of traces coherency, the ring buffer itself
3480 mutex_lock(&iter->mutex);
3481 if (iter->trace->read) {
3482 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3488 sret = tracing_wait_pipe(filp);
3492 /* stop when tracing is finished */
3493 if (trace_empty(iter)) {
3498 if (cnt >= PAGE_SIZE)
3499 cnt = PAGE_SIZE - 1;
3501 /* reset all but tr, trace, and overruns */
3502 memset(&iter->seq, 0,
3503 sizeof(struct trace_iterator) -
3504 offsetof(struct trace_iterator, seq));
3507 trace_event_read_lock();
3508 trace_access_lock(iter->cpu_file);
3509 while (trace_find_next_entry_inc(iter) != NULL) {
3510 enum print_line_t ret;
3511 int len = iter->seq.len;
3513 ret = print_trace_line(iter);
3514 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3515 /* don't print partial lines */
3516 iter->seq.len = len;
3519 if (ret != TRACE_TYPE_NO_CONSUME)
3520 trace_consume(iter);
3522 if (iter->seq.len >= cnt)
3526 * Setting the full flag means we reached the trace_seq buffer
3527 * size and we should leave by partial output condition above.
3528 * One of the trace_seq_* functions is not used properly.
3530 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
3533 trace_access_unlock(iter->cpu_file);
3534 trace_event_read_unlock();
3536 /* Now copy what we have to the user */
3537 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3538 if (iter->seq.readpos >= iter->seq.len)
3539 trace_seq_init(&iter->seq);
3542 * If there was nothing to send to user, in spite of consuming trace
3543 * entries, go back to wait for more entries.
3549 mutex_unlock(&iter->mutex);
3554 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3555 struct pipe_buffer *buf)
3557 __free_page(buf->page);
3560 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3563 __free_page(spd->pages[idx]);
3566 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3568 .map = generic_pipe_buf_map,
3569 .unmap = generic_pipe_buf_unmap,
3570 .confirm = generic_pipe_buf_confirm,
3571 .release = tracing_pipe_buf_release,
3572 .steal = generic_pipe_buf_steal,
3573 .get = generic_pipe_buf_get,
3577 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3582 /* Seq buffer is page-sized, exactly what we need. */
3584 count = iter->seq.len;
3585 ret = print_trace_line(iter);
3586 count = iter->seq.len - count;
3589 iter->seq.len -= count;
3592 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3593 iter->seq.len -= count;
3597 if (ret != TRACE_TYPE_NO_CONSUME)
3598 trace_consume(iter);
3600 if (!trace_find_next_entry_inc(iter)) {
3610 static ssize_t tracing_splice_read_pipe(struct file *filp,
3612 struct pipe_inode_info *pipe,
3616 struct page *pages_def[PIPE_DEF_BUFFERS];
3617 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3618 struct trace_iterator *iter = filp->private_data;
3619 struct splice_pipe_desc spd = {
3621 .partial = partial_def,
3622 .nr_pages = 0, /* This gets updated below. */
3623 .nr_pages_max = PIPE_DEF_BUFFERS,
3625 .ops = &tracing_pipe_buf_ops,
3626 .spd_release = tracing_spd_release_pipe,
3628 static struct tracer *old_tracer;
3633 if (splice_grow_spd(pipe, &spd))
3636 /* copy the tracer to avoid using a global lock all around */
3637 mutex_lock(&trace_types_lock);
3638 if (unlikely(old_tracer != current_trace && current_trace)) {
3639 old_tracer = current_trace;
3640 *iter->trace = *current_trace;
3642 mutex_unlock(&trace_types_lock);
3644 mutex_lock(&iter->mutex);
3646 if (iter->trace->splice_read) {
3647 ret = iter->trace->splice_read(iter, filp,
3648 ppos, pipe, len, flags);
3653 ret = tracing_wait_pipe(filp);
3657 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
3662 trace_event_read_lock();
3663 trace_access_lock(iter->cpu_file);
3665 /* Fill as many pages as possible. */
3666 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3667 spd.pages[i] = alloc_page(GFP_KERNEL);
3671 rem = tracing_fill_pipe_page(rem, iter);
3673 /* Copy the data into the page, so we can start over. */
3674 ret = trace_seq_to_buffer(&iter->seq,
3675 page_address(spd.pages[i]),
3678 __free_page(spd.pages[i]);
3681 spd.partial[i].offset = 0;
3682 spd.partial[i].len = iter->seq.len;
3684 trace_seq_init(&iter->seq);
3687 trace_access_unlock(iter->cpu_file);
3688 trace_event_read_unlock();
3689 mutex_unlock(&iter->mutex);
3693 ret = splice_to_pipe(pipe, &spd);
3695 splice_shrink_spd(&spd);
3699 mutex_unlock(&iter->mutex);
3703 struct ftrace_entries_info {
3704 struct trace_array *tr;
3708 static int tracing_entries_open(struct inode *inode, struct file *filp)
3710 struct ftrace_entries_info *info;
3712 if (tracing_disabled)
3715 info = kzalloc(sizeof(*info), GFP_KERNEL);
3719 info->tr = &global_trace;
3720 info->cpu = (unsigned long)inode->i_private;
3722 filp->private_data = info;
3728 tracing_entries_read(struct file *filp, char __user *ubuf,
3729 size_t cnt, loff_t *ppos)
3731 struct ftrace_entries_info *info = filp->private_data;
3732 struct trace_array *tr = info->tr;
3737 mutex_lock(&trace_types_lock);
3739 if (info->cpu == RING_BUFFER_ALL_CPUS) {
3740 int cpu, buf_size_same;
3745 /* check if all cpu sizes are same */
3746 for_each_tracing_cpu(cpu) {
3747 /* fill in the size from first enabled cpu */
3749 size = tr->data[cpu]->entries;
3750 if (size != tr->data[cpu]->entries) {
3756 if (buf_size_same) {
3757 if (!ring_buffer_expanded)
3758 r = sprintf(buf, "%lu (expanded: %lu)\n",
3760 trace_buf_size >> 10);
3762 r = sprintf(buf, "%lu\n", size >> 10);
3764 r = sprintf(buf, "X\n");
3766 r = sprintf(buf, "%lu\n", tr->data[info->cpu]->entries >> 10);
3768 mutex_unlock(&trace_types_lock);
3770 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3775 tracing_entries_write(struct file *filp, const char __user *ubuf,
3776 size_t cnt, loff_t *ppos)
3778 struct ftrace_entries_info *info = filp->private_data;
3782 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3786 /* must have at least 1 entry */
3790 /* value is in KB */
3793 ret = tracing_resize_ring_buffer(val, info->cpu);
3803 tracing_entries_release(struct inode *inode, struct file *filp)
3805 struct ftrace_entries_info *info = filp->private_data;
3813 tracing_total_entries_read(struct file *filp, char __user *ubuf,
3814 size_t cnt, loff_t *ppos)
3816 struct trace_array *tr = filp->private_data;
3819 unsigned long size = 0, expanded_size = 0;
3821 mutex_lock(&trace_types_lock);
3822 for_each_tracing_cpu(cpu) {
3823 size += tr->data[cpu]->entries >> 10;
3824 if (!ring_buffer_expanded)
3825 expanded_size += trace_buf_size >> 10;
3827 if (ring_buffer_expanded)
3828 r = sprintf(buf, "%lu\n", size);
3830 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
3831 mutex_unlock(&trace_types_lock);
3833 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3837 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
3838 size_t cnt, loff_t *ppos)
3841 * There is no need to read what the user has written, this function
3842 * is just to make sure that there is no error when "echo" is used
3851 tracing_free_buffer_release(struct inode *inode, struct file *filp)
3853 /* disable tracing ? */
3854 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
3856 /* resize the ring buffer to 0 */
3857 tracing_resize_ring_buffer(0, RING_BUFFER_ALL_CPUS);
3863 tracing_mark_write(struct file *filp, const char __user *ubuf,
3864 size_t cnt, loff_t *fpos)
3866 unsigned long addr = (unsigned long)ubuf;
3867 struct ring_buffer_event *event;
3868 struct ring_buffer *buffer;
3869 struct print_entry *entry;
3870 unsigned long irq_flags;
3871 struct page *pages[2];
3881 if (tracing_disabled)
3884 if (cnt > TRACE_BUF_SIZE)
3885 cnt = TRACE_BUF_SIZE;
3888 * Userspace is injecting traces into the kernel trace buffer.
3889 * We want to be as non intrusive as possible.
3890 * To do so, we do not want to allocate any special buffers
3891 * or take any locks, but instead write the userspace data
3892 * straight into the ring buffer.
3894 * First we need to pin the userspace buffer into memory,
3895 * which, most likely it is, because it just referenced it.
3896 * But there's no guarantee that it is. By using get_user_pages_fast()
3897 * and kmap_atomic/kunmap_atomic() we can get access to the
3898 * pages directly. We then write the data directly into the
3901 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
3903 /* check if we cross pages */
3904 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
3907 offset = addr & (PAGE_SIZE - 1);
3910 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
3911 if (ret < nr_pages) {
3913 put_page(pages[ret]);
3918 for (i = 0; i < nr_pages; i++)
3919 map_page[i] = kmap_atomic(pages[i]);
3921 local_save_flags(irq_flags);
3922 size = sizeof(*entry) + cnt + 2; /* possible \n added */
3923 buffer = global_trace.buffer;
3924 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3925 irq_flags, preempt_count());
3927 /* Ring buffer disabled, return as if not open for write */
3932 entry = ring_buffer_event_data(event);
3933 entry->ip = _THIS_IP_;
3935 if (nr_pages == 2) {
3936 len = PAGE_SIZE - offset;
3937 memcpy(&entry->buf, map_page[0] + offset, len);
3938 memcpy(&entry->buf[len], map_page[1], cnt - len);
3940 memcpy(&entry->buf, map_page[0] + offset, cnt);
3942 if (entry->buf[cnt - 1] != '\n') {
3943 entry->buf[cnt] = '\n';
3944 entry->buf[cnt + 1] = '\0';
3946 entry->buf[cnt] = '\0';
3948 ring_buffer_unlock_commit(buffer, event);
3955 for (i = 0; i < nr_pages; i++){
3956 kunmap_atomic(map_page[i]);
3963 static int tracing_clock_show(struct seq_file *m, void *v)
3967 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3969 "%s%s%s%s", i ? " " : "",
3970 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3971 i == trace_clock_id ? "]" : "");
3977 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3978 size_t cnt, loff_t *fpos)
3981 const char *clockstr;
3984 if (cnt >= sizeof(buf))
3987 if (copy_from_user(&buf, ubuf, cnt))
3992 clockstr = strstrip(buf);
3994 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3995 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3998 if (i == ARRAY_SIZE(trace_clocks))
4003 mutex_lock(&trace_types_lock);
4005 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
4007 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
4009 mutex_unlock(&trace_types_lock);
4016 static int tracing_clock_open(struct inode *inode, struct file *file)
4018 if (tracing_disabled)
4020 return single_open(file, tracing_clock_show, NULL);
4023 static const struct file_operations tracing_max_lat_fops = {
4024 .open = tracing_open_generic,
4025 .read = tracing_max_lat_read,
4026 .write = tracing_max_lat_write,
4027 .llseek = generic_file_llseek,
4030 static const struct file_operations tracing_ctrl_fops = {
4031 .open = tracing_open_generic,
4032 .read = tracing_ctrl_read,
4033 .write = tracing_ctrl_write,
4034 .llseek = generic_file_llseek,
4037 static const struct file_operations set_tracer_fops = {
4038 .open = tracing_open_generic,
4039 .read = tracing_set_trace_read,
4040 .write = tracing_set_trace_write,
4041 .llseek = generic_file_llseek,
4044 static const struct file_operations tracing_pipe_fops = {
4045 .open = tracing_open_pipe,
4046 .poll = tracing_poll_pipe,
4047 .read = tracing_read_pipe,
4048 .splice_read = tracing_splice_read_pipe,
4049 .release = tracing_release_pipe,
4050 .llseek = no_llseek,
4053 static const struct file_operations tracing_entries_fops = {
4054 .open = tracing_entries_open,
4055 .read = tracing_entries_read,
4056 .write = tracing_entries_write,
4057 .release = tracing_entries_release,
4058 .llseek = generic_file_llseek,
4061 static const struct file_operations tracing_total_entries_fops = {
4062 .open = tracing_open_generic,
4063 .read = tracing_total_entries_read,
4064 .llseek = generic_file_llseek,
4067 static const struct file_operations tracing_free_buffer_fops = {
4068 .write = tracing_free_buffer_write,
4069 .release = tracing_free_buffer_release,
4072 static const struct file_operations tracing_mark_fops = {
4073 .open = tracing_open_generic,
4074 .write = tracing_mark_write,
4075 .llseek = generic_file_llseek,
4078 static const struct file_operations trace_clock_fops = {
4079 .open = tracing_clock_open,
4081 .llseek = seq_lseek,
4082 .release = single_release,
4083 .write = tracing_clock_write,
4086 struct ftrace_buffer_info {
4087 struct trace_array *tr;
4093 static int tracing_buffers_open(struct inode *inode, struct file *filp)
4095 int cpu = (int)(long)inode->i_private;
4096 struct ftrace_buffer_info *info;
4098 if (tracing_disabled)
4101 info = kzalloc(sizeof(*info), GFP_KERNEL);
4105 info->tr = &global_trace;
4108 /* Force reading ring buffer for first read */
4109 info->read = (unsigned int)-1;
4111 filp->private_data = info;
4113 return nonseekable_open(inode, filp);
4117 tracing_buffers_read(struct file *filp, char __user *ubuf,
4118 size_t count, loff_t *ppos)
4120 struct ftrace_buffer_info *info = filp->private_data;
4128 info->spare = ring_buffer_alloc_read_page(info->tr->buffer, info->cpu);
4132 /* Do we have previous read data to read? */
4133 if (info->read < PAGE_SIZE)
4136 trace_access_lock(info->cpu);
4137 ret = ring_buffer_read_page(info->tr->buffer,
4141 trace_access_unlock(info->cpu);
4148 size = PAGE_SIZE - info->read;
4152 ret = copy_to_user(ubuf, info->spare + info->read, size);
4163 static int tracing_buffers_release(struct inode *inode, struct file *file)
4165 struct ftrace_buffer_info *info = file->private_data;
4168 ring_buffer_free_read_page(info->tr->buffer, info->spare);
4175 struct ring_buffer *buffer;
4180 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
4181 struct pipe_buffer *buf)
4183 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4188 ring_buffer_free_read_page(ref->buffer, ref->page);
4193 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
4194 struct pipe_buffer *buf)
4199 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
4200 struct pipe_buffer *buf)
4202 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4207 /* Pipe buffer operations for a buffer. */
4208 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
4210 .map = generic_pipe_buf_map,
4211 .unmap = generic_pipe_buf_unmap,
4212 .confirm = generic_pipe_buf_confirm,
4213 .release = buffer_pipe_buf_release,
4214 .steal = buffer_pipe_buf_steal,
4215 .get = buffer_pipe_buf_get,
4219 * Callback from splice_to_pipe(), if we need to release some pages
4220 * at the end of the spd in case we error'ed out in filling the pipe.
4222 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
4224 struct buffer_ref *ref =
4225 (struct buffer_ref *)spd->partial[i].private;
4230 ring_buffer_free_read_page(ref->buffer, ref->page);
4232 spd->partial[i].private = 0;
4236 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4237 struct pipe_inode_info *pipe, size_t len,
4240 struct ftrace_buffer_info *info = file->private_data;
4241 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4242 struct page *pages_def[PIPE_DEF_BUFFERS];
4243 struct splice_pipe_desc spd = {
4245 .partial = partial_def,
4246 .nr_pages_max = PIPE_DEF_BUFFERS,
4248 .ops = &buffer_pipe_buf_ops,
4249 .spd_release = buffer_spd_release,
4251 struct buffer_ref *ref;
4252 int entries, size, i;
4255 if (splice_grow_spd(pipe, &spd))
4258 if (*ppos & (PAGE_SIZE - 1)) {
4259 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
4264 if (len & (PAGE_SIZE - 1)) {
4265 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
4266 if (len < PAGE_SIZE) {
4273 trace_access_lock(info->cpu);
4274 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
4276 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
4280 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
4285 ref->buffer = info->tr->buffer;
4286 ref->page = ring_buffer_alloc_read_page(ref->buffer, info->cpu);
4292 r = ring_buffer_read_page(ref->buffer, &ref->page,
4295 ring_buffer_free_read_page(ref->buffer, ref->page);
4301 * zero out any left over data, this is going to
4304 size = ring_buffer_page_len(ref->page);
4305 if (size < PAGE_SIZE)
4306 memset(ref->page + size, 0, PAGE_SIZE - size);
4308 page = virt_to_page(ref->page);
4310 spd.pages[i] = page;
4311 spd.partial[i].len = PAGE_SIZE;
4312 spd.partial[i].offset = 0;
4313 spd.partial[i].private = (unsigned long)ref;
4317 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
4320 trace_access_unlock(info->cpu);
4323 /* did we read anything? */
4324 if (!spd.nr_pages) {
4325 if (flags & SPLICE_F_NONBLOCK)
4333 ret = splice_to_pipe(pipe, &spd);
4334 splice_shrink_spd(&spd);
4339 static const struct file_operations tracing_buffers_fops = {
4340 .open = tracing_buffers_open,
4341 .read = tracing_buffers_read,
4342 .release = tracing_buffers_release,
4343 .splice_read = tracing_buffers_splice_read,
4344 .llseek = no_llseek,
4348 tracing_stats_read(struct file *filp, char __user *ubuf,
4349 size_t count, loff_t *ppos)
4351 unsigned long cpu = (unsigned long)filp->private_data;
4352 struct trace_array *tr = &global_trace;
4353 struct trace_seq *s;
4355 unsigned long long t;
4356 unsigned long usec_rem;
4358 s = kmalloc(sizeof(*s), GFP_KERNEL);
4364 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
4365 trace_seq_printf(s, "entries: %ld\n", cnt);
4367 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
4368 trace_seq_printf(s, "overrun: %ld\n", cnt);
4370 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
4371 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
4373 cnt = ring_buffer_bytes_cpu(tr->buffer, cpu);
4374 trace_seq_printf(s, "bytes: %ld\n", cnt);
4376 t = ns2usecs(ring_buffer_oldest_event_ts(tr->buffer, cpu));
4377 usec_rem = do_div(t, USEC_PER_SEC);
4378 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n", t, usec_rem);
4380 t = ns2usecs(ring_buffer_time_stamp(tr->buffer, cpu));
4381 usec_rem = do_div(t, USEC_PER_SEC);
4382 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
4384 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
4391 static const struct file_operations tracing_stats_fops = {
4392 .open = tracing_open_generic,
4393 .read = tracing_stats_read,
4394 .llseek = generic_file_llseek,
4397 #ifdef CONFIG_DYNAMIC_FTRACE
4399 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
4405 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
4406 size_t cnt, loff_t *ppos)
4408 static char ftrace_dyn_info_buffer[1024];
4409 static DEFINE_MUTEX(dyn_info_mutex);
4410 unsigned long *p = filp->private_data;
4411 char *buf = ftrace_dyn_info_buffer;
4412 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
4415 mutex_lock(&dyn_info_mutex);
4416 r = sprintf(buf, "%ld ", *p);
4418 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
4421 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4423 mutex_unlock(&dyn_info_mutex);
4428 static const struct file_operations tracing_dyn_info_fops = {
4429 .open = tracing_open_generic,
4430 .read = tracing_read_dyn_info,
4431 .llseek = generic_file_llseek,
4435 static struct dentry *d_tracer;
4437 struct dentry *tracing_init_dentry(void)
4444 if (!debugfs_initialized())
4447 d_tracer = debugfs_create_dir("tracing", NULL);
4449 if (!d_tracer && !once) {
4451 pr_warning("Could not create debugfs directory 'tracing'\n");
4458 static struct dentry *d_percpu;
4460 struct dentry *tracing_dentry_percpu(void)
4463 struct dentry *d_tracer;
4468 d_tracer = tracing_init_dentry();
4473 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
4475 if (!d_percpu && !once) {
4477 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4484 static void tracing_init_debugfs_percpu(long cpu)
4486 struct dentry *d_percpu = tracing_dentry_percpu();
4487 struct dentry *d_cpu;
4488 char cpu_dir[30]; /* 30 characters should be more than enough */
4493 snprintf(cpu_dir, 30, "cpu%ld", cpu);
4494 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4496 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4500 /* per cpu trace_pipe */
4501 trace_create_file("trace_pipe", 0444, d_cpu,
4502 (void *) cpu, &tracing_pipe_fops);
4505 trace_create_file("trace", 0644, d_cpu,
4506 (void *) cpu, &tracing_fops);
4508 trace_create_file("trace_pipe_raw", 0444, d_cpu,
4509 (void *) cpu, &tracing_buffers_fops);
4511 trace_create_file("stats", 0444, d_cpu,
4512 (void *) cpu, &tracing_stats_fops);
4514 trace_create_file("buffer_size_kb", 0444, d_cpu,
4515 (void *) cpu, &tracing_entries_fops);
4518 #ifdef CONFIG_FTRACE_SELFTEST
4519 /* Let selftest have access to static functions in this file */
4520 #include "trace_selftest.c"
4523 struct trace_option_dentry {
4524 struct tracer_opt *opt;
4525 struct tracer_flags *flags;
4526 struct dentry *entry;
4530 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4533 struct trace_option_dentry *topt = filp->private_data;
4536 if (topt->flags->val & topt->opt->bit)
4541 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4545 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4548 struct trace_option_dentry *topt = filp->private_data;
4552 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4556 if (val != 0 && val != 1)
4559 if (!!(topt->flags->val & topt->opt->bit) != val) {
4560 mutex_lock(&trace_types_lock);
4561 ret = __set_tracer_option(current_trace, topt->flags,
4563 mutex_unlock(&trace_types_lock);
4574 static const struct file_operations trace_options_fops = {
4575 .open = tracing_open_generic,
4576 .read = trace_options_read,
4577 .write = trace_options_write,
4578 .llseek = generic_file_llseek,
4582 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4585 long index = (long)filp->private_data;
4588 if (trace_flags & (1 << index))
4593 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4597 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4600 long index = (long)filp->private_data;
4604 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4608 if (val != 0 && val != 1)
4610 set_tracer_flags(1 << index, val);
4617 static const struct file_operations trace_options_core_fops = {
4618 .open = tracing_open_generic,
4619 .read = trace_options_core_read,
4620 .write = trace_options_core_write,
4621 .llseek = generic_file_llseek,
4624 struct dentry *trace_create_file(const char *name,
4626 struct dentry *parent,
4628 const struct file_operations *fops)
4632 ret = debugfs_create_file(name, mode, parent, data, fops);
4634 pr_warning("Could not create debugfs '%s' entry\n", name);
4640 static struct dentry *trace_options_init_dentry(void)
4642 struct dentry *d_tracer;
4643 static struct dentry *t_options;
4648 d_tracer = tracing_init_dentry();
4652 t_options = debugfs_create_dir("options", d_tracer);
4654 pr_warning("Could not create debugfs directory 'options'\n");
4662 create_trace_option_file(struct trace_option_dentry *topt,
4663 struct tracer_flags *flags,
4664 struct tracer_opt *opt)
4666 struct dentry *t_options;
4668 t_options = trace_options_init_dentry();
4672 topt->flags = flags;
4675 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4676 &trace_options_fops);
4680 static struct trace_option_dentry *
4681 create_trace_option_files(struct tracer *tracer)
4683 struct trace_option_dentry *topts;
4684 struct tracer_flags *flags;
4685 struct tracer_opt *opts;
4691 flags = tracer->flags;
4693 if (!flags || !flags->opts)
4698 for (cnt = 0; opts[cnt].name; cnt++)
4701 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4705 for (cnt = 0; opts[cnt].name; cnt++)
4706 create_trace_option_file(&topts[cnt], flags,
4713 destroy_trace_option_files(struct trace_option_dentry *topts)
4720 for (cnt = 0; topts[cnt].opt; cnt++) {
4721 if (topts[cnt].entry)
4722 debugfs_remove(topts[cnt].entry);
4728 static struct dentry *
4729 create_trace_option_core_file(const char *option, long index)
4731 struct dentry *t_options;
4733 t_options = trace_options_init_dentry();
4737 return trace_create_file(option, 0644, t_options, (void *)index,
4738 &trace_options_core_fops);
4741 static __init void create_trace_options_dir(void)
4743 struct dentry *t_options;
4746 t_options = trace_options_init_dentry();
4750 for (i = 0; trace_options[i]; i++)
4751 create_trace_option_core_file(trace_options[i], i);
4755 rb_simple_read(struct file *filp, char __user *ubuf,
4756 size_t cnt, loff_t *ppos)
4758 struct trace_array *tr = filp->private_data;
4759 struct ring_buffer *buffer = tr->buffer;
4764 r = ring_buffer_record_is_on(buffer);
4768 r = sprintf(buf, "%d\n", r);
4770 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4774 rb_simple_write(struct file *filp, const char __user *ubuf,
4775 size_t cnt, loff_t *ppos)
4777 struct trace_array *tr = filp->private_data;
4778 struct ring_buffer *buffer = tr->buffer;
4782 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4788 ring_buffer_record_on(buffer);
4790 ring_buffer_record_off(buffer);
4798 static const struct file_operations rb_simple_fops = {
4799 .open = tracing_open_generic,
4800 .read = rb_simple_read,
4801 .write = rb_simple_write,
4802 .llseek = default_llseek,
4805 static __init int tracer_init_debugfs(void)
4807 struct dentry *d_tracer;
4810 trace_access_lock_init();
4812 d_tracer = tracing_init_dentry();
4814 trace_create_file("tracing_enabled", 0644, d_tracer,
4815 &global_trace, &tracing_ctrl_fops);
4817 trace_create_file("trace_options", 0644, d_tracer,
4818 NULL, &tracing_iter_fops);
4820 trace_create_file("tracing_cpumask", 0644, d_tracer,
4821 NULL, &tracing_cpumask_fops);
4823 trace_create_file("trace", 0644, d_tracer,
4824 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4826 trace_create_file("available_tracers", 0444, d_tracer,
4827 &global_trace, &show_traces_fops);
4829 trace_create_file("current_tracer", 0644, d_tracer,
4830 &global_trace, &set_tracer_fops);
4832 #ifdef CONFIG_TRACER_MAX_TRACE
4833 trace_create_file("tracing_max_latency", 0644, d_tracer,
4834 &tracing_max_latency, &tracing_max_lat_fops);
4837 trace_create_file("tracing_thresh", 0644, d_tracer,
4838 &tracing_thresh, &tracing_max_lat_fops);
4840 trace_create_file("README", 0444, d_tracer,
4841 NULL, &tracing_readme_fops);
4843 trace_create_file("trace_pipe", 0444, d_tracer,
4844 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4846 trace_create_file("buffer_size_kb", 0644, d_tracer,
4847 (void *) RING_BUFFER_ALL_CPUS, &tracing_entries_fops);
4849 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
4850 &global_trace, &tracing_total_entries_fops);
4852 trace_create_file("free_buffer", 0644, d_tracer,
4853 &global_trace, &tracing_free_buffer_fops);
4855 trace_create_file("trace_marker", 0220, d_tracer,
4856 NULL, &tracing_mark_fops);
4858 trace_create_file("saved_cmdlines", 0444, d_tracer,
4859 NULL, &tracing_saved_cmdlines_fops);
4861 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4864 trace_create_file("tracing_on", 0644, d_tracer,
4865 &global_trace, &rb_simple_fops);
4867 #ifdef CONFIG_DYNAMIC_FTRACE
4868 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4869 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4872 create_trace_options_dir();
4874 for_each_tracing_cpu(cpu)
4875 tracing_init_debugfs_percpu(cpu);
4880 static int trace_panic_handler(struct notifier_block *this,
4881 unsigned long event, void *unused)
4883 if (ftrace_dump_on_oops)
4884 ftrace_dump(ftrace_dump_on_oops);
4888 static struct notifier_block trace_panic_notifier = {
4889 .notifier_call = trace_panic_handler,
4891 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4894 static int trace_die_handler(struct notifier_block *self,
4900 if (ftrace_dump_on_oops)
4901 ftrace_dump(ftrace_dump_on_oops);
4909 static struct notifier_block trace_die_notifier = {
4910 .notifier_call = trace_die_handler,
4915 * printk is set to max of 1024, we really don't need it that big.
4916 * Nothing should be printing 1000 characters anyway.
4918 #define TRACE_MAX_PRINT 1000
4921 * Define here KERN_TRACE so that we have one place to modify
4922 * it if we decide to change what log level the ftrace dump
4925 #define KERN_TRACE KERN_EMERG
4928 trace_printk_seq(struct trace_seq *s)
4930 /* Probably should print a warning here. */
4934 /* should be zero ended, but we are paranoid. */
4935 s->buffer[s->len] = 0;
4937 printk(KERN_TRACE "%s", s->buffer);
4942 void trace_init_global_iter(struct trace_iterator *iter)
4944 iter->tr = &global_trace;
4945 iter->trace = current_trace;
4946 iter->cpu_file = TRACE_PIPE_ALL_CPU;
4950 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4952 static arch_spinlock_t ftrace_dump_lock =
4953 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4954 /* use static because iter can be a bit big for the stack */
4955 static struct trace_iterator iter;
4956 unsigned int old_userobj;
4957 static int dump_ran;
4958 unsigned long flags;
4962 local_irq_save(flags);
4963 arch_spin_lock(&ftrace_dump_lock);
4971 /* Did function tracer already get disabled? */
4972 if (ftrace_is_dead()) {
4973 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
4974 printk("# MAY BE MISSING FUNCTION EVENTS\n");
4977 if (disable_tracing)
4980 trace_init_global_iter(&iter);
4982 for_each_tracing_cpu(cpu) {
4983 atomic_inc(&iter.tr->data[cpu]->disabled);
4986 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4988 /* don't look at user memory in panic mode */
4989 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4991 /* Simulate the iterator */
4992 iter.tr = &global_trace;
4993 iter.trace = current_trace;
4995 switch (oops_dump_mode) {
4997 iter.cpu_file = TRACE_PIPE_ALL_CPU;
5000 iter.cpu_file = raw_smp_processor_id();
5005 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
5006 iter.cpu_file = TRACE_PIPE_ALL_CPU;
5009 printk(KERN_TRACE "Dumping ftrace buffer:\n");
5012 * We need to stop all tracing on all CPUS to read the
5013 * the next buffer. This is a bit expensive, but is
5014 * not done often. We fill all what we can read,
5015 * and then release the locks again.
5018 while (!trace_empty(&iter)) {
5021 printk(KERN_TRACE "---------------------------------\n");
5025 /* reset all but tr, trace, and overruns */
5026 memset(&iter.seq, 0,
5027 sizeof(struct trace_iterator) -
5028 offsetof(struct trace_iterator, seq));
5029 iter.iter_flags |= TRACE_FILE_LAT_FMT;
5032 if (trace_find_next_entry_inc(&iter) != NULL) {
5035 ret = print_trace_line(&iter);
5036 if (ret != TRACE_TYPE_NO_CONSUME)
5037 trace_consume(&iter);
5039 touch_nmi_watchdog();
5041 trace_printk_seq(&iter.seq);
5045 printk(KERN_TRACE " (ftrace buffer empty)\n");
5047 printk(KERN_TRACE "---------------------------------\n");
5050 /* Re-enable tracing if requested */
5051 if (!disable_tracing) {
5052 trace_flags |= old_userobj;
5054 for_each_tracing_cpu(cpu) {
5055 atomic_dec(&iter.tr->data[cpu]->disabled);
5061 arch_spin_unlock(&ftrace_dump_lock);
5062 local_irq_restore(flags);
5065 /* By default: disable tracing after the dump */
5066 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
5068 __ftrace_dump(true, oops_dump_mode);
5070 EXPORT_SYMBOL_GPL(ftrace_dump);
5072 __init static int tracer_alloc_buffers(void)
5075 enum ring_buffer_flags rb_flags;
5080 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
5083 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
5084 goto out_free_buffer_mask;
5086 /* Only allocate trace_printk buffers if a trace_printk exists */
5087 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
5088 trace_printk_init_buffers();
5090 /* To save memory, keep the ring buffer size to its minimum */
5091 if (ring_buffer_expanded)
5092 ring_buf_size = trace_buf_size;
5096 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
5098 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
5099 cpumask_copy(tracing_cpumask, cpu_all_mask);
5101 /* TODO: make the number of buffers hot pluggable with CPUS */
5102 global_trace.buffer = ring_buffer_alloc(ring_buf_size, rb_flags);
5103 if (!global_trace.buffer) {
5104 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
5106 goto out_free_cpumask;
5108 if (global_trace.buffer_disabled)
5112 #ifdef CONFIG_TRACER_MAX_TRACE
5113 max_tr.buffer = ring_buffer_alloc(1, rb_flags);
5114 if (!max_tr.buffer) {
5115 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
5117 ring_buffer_free(global_trace.buffer);
5118 goto out_free_cpumask;
5122 /* Allocate the first page for all buffers */
5123 for_each_tracing_cpu(i) {
5124 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
5125 max_tr.data[i] = &per_cpu(max_tr_data, i);
5128 set_buffer_entries(&global_trace,
5129 ring_buffer_size(global_trace.buffer, 0));
5130 #ifdef CONFIG_TRACER_MAX_TRACE
5131 set_buffer_entries(&max_tr, 1);
5134 trace_init_cmdlines();
5136 register_tracer(&nop_trace);
5137 current_trace = &nop_trace;
5138 /* All seems OK, enable tracing */
5139 tracing_disabled = 0;
5141 atomic_notifier_chain_register(&panic_notifier_list,
5142 &trace_panic_notifier);
5144 register_die_notifier(&trace_die_notifier);
5149 free_cpumask_var(tracing_cpumask);
5150 out_free_buffer_mask:
5151 free_cpumask_var(tracing_buffer_mask);
5156 __init static int clear_boot_tracer(void)
5159 * The default tracer at boot buffer is an init section.
5160 * This function is called in lateinit. If we did not
5161 * find the boot tracer, then clear it out, to prevent
5162 * later registration from accessing the buffer that is
5163 * about to be freed.
5165 if (!default_bootup_tracer)
5168 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
5169 default_bootup_tracer);
5170 default_bootup_tracer = NULL;
5175 early_initcall(tracer_alloc_buffers);
5176 fs_initcall(tracer_init_debugfs);
5177 late_initcall(clear_boot_tracer);