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 |
331 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS;
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 threshold;
434 ret = kstrtoul(str, 0, &threshold);
437 tracing_thresh = threshold * 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[] = {
481 { trace_clock_local, "local" },
482 { trace_clock_global, "global" },
483 { trace_clock_counter, "counter" },
489 * trace_parser_get_init - gets the buffer for trace parser
491 int trace_parser_get_init(struct trace_parser *parser, int size)
493 memset(parser, 0, sizeof(*parser));
495 parser->buffer = kmalloc(size, GFP_KERNEL);
504 * trace_parser_put - frees the buffer for trace parser
506 void trace_parser_put(struct trace_parser *parser)
508 kfree(parser->buffer);
512 * trace_get_user - reads the user input string separated by space
513 * (matched by isspace(ch))
515 * For each string found the 'struct trace_parser' is updated,
516 * and the function returns.
518 * Returns number of bytes read.
520 * See kernel/trace/trace.h for 'struct trace_parser' details.
522 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
523 size_t cnt, loff_t *ppos)
530 trace_parser_clear(parser);
532 ret = get_user(ch, ubuf++);
540 * The parser is not finished with the last write,
541 * continue reading the user input without skipping spaces.
544 /* skip white space */
545 while (cnt && isspace(ch)) {
546 ret = get_user(ch, ubuf++);
553 /* only spaces were written */
563 /* read the non-space input */
564 while (cnt && !isspace(ch)) {
565 if (parser->idx < parser->size - 1)
566 parser->buffer[parser->idx++] = ch;
571 ret = get_user(ch, ubuf++);
578 /* We either got finished input or we have to wait for another call. */
580 parser->buffer[parser->idx] = 0;
581 parser->cont = false;
584 parser->buffer[parser->idx++] = ch;
594 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
602 if (s->len <= s->readpos)
605 len = s->len - s->readpos;
608 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
618 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
622 if (s->len <= s->readpos)
625 len = s->len - s->readpos;
628 memcpy(buf, s->buffer + s->readpos, cnt);
635 * ftrace_max_lock is used to protect the swapping of buffers
636 * when taking a max snapshot. The buffers themselves are
637 * protected by per_cpu spinlocks. But the action of the swap
638 * needs its own lock.
640 * This is defined as a arch_spinlock_t in order to help
641 * with performance when lockdep debugging is enabled.
643 * It is also used in other places outside the update_max_tr
644 * so it needs to be defined outside of the
645 * CONFIG_TRACER_MAX_TRACE.
647 static arch_spinlock_t ftrace_max_lock =
648 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
650 unsigned long __read_mostly tracing_thresh;
652 #ifdef CONFIG_TRACER_MAX_TRACE
653 unsigned long __read_mostly tracing_max_latency;
656 * Copy the new maximum trace into the separate maximum-trace
657 * structure. (this way the maximum trace is permanently saved,
658 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
661 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
663 struct trace_array_cpu *data = tr->data[cpu];
664 struct trace_array_cpu *max_data;
667 max_tr.time_start = data->preempt_timestamp;
669 max_data = max_tr.data[cpu];
670 max_data->saved_latency = tracing_max_latency;
671 max_data->critical_start = data->critical_start;
672 max_data->critical_end = data->critical_end;
674 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
675 max_data->pid = tsk->pid;
676 max_data->uid = task_uid(tsk);
677 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
678 max_data->policy = tsk->policy;
679 max_data->rt_priority = tsk->rt_priority;
681 /* record this tasks comm */
682 tracing_record_cmdline(tsk);
686 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
688 * @tsk: the task with the latency
689 * @cpu: The cpu that initiated the trace.
691 * Flip the buffers between the @tr and the max_tr and record information
692 * about which task was the cause of this latency.
695 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
697 struct ring_buffer *buf = tr->buffer;
699 if (trace_stop_count)
702 WARN_ON_ONCE(!irqs_disabled());
703 if (!current_trace->use_max_tr) {
707 arch_spin_lock(&ftrace_max_lock);
709 tr->buffer = max_tr.buffer;
712 __update_max_tr(tr, tsk, cpu);
713 arch_spin_unlock(&ftrace_max_lock);
717 * update_max_tr_single - only copy one trace over, and reset the rest
719 * @tsk - task with the latency
720 * @cpu - the cpu of the buffer to copy.
722 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
725 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
729 if (trace_stop_count)
732 WARN_ON_ONCE(!irqs_disabled());
733 if (!current_trace->use_max_tr) {
738 arch_spin_lock(&ftrace_max_lock);
740 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
744 * We failed to swap the buffer due to a commit taking
745 * place on this CPU. We fail to record, but we reset
746 * the max trace buffer (no one writes directly to it)
747 * and flag that it failed.
749 trace_array_printk(&max_tr, _THIS_IP_,
750 "Failed to swap buffers due to commit in progress\n");
753 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
755 __update_max_tr(tr, tsk, cpu);
756 arch_spin_unlock(&ftrace_max_lock);
758 #endif /* CONFIG_TRACER_MAX_TRACE */
761 * register_tracer - register a tracer with the ftrace system.
762 * @type - the plugin for the tracer
764 * Register a new plugin tracer.
766 int register_tracer(struct tracer *type)
772 pr_info("Tracer must have a name\n");
776 if (strlen(type->name) >= MAX_TRACER_SIZE) {
777 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
781 mutex_lock(&trace_types_lock);
783 tracing_selftest_running = true;
785 for (t = trace_types; t; t = t->next) {
786 if (strcmp(type->name, t->name) == 0) {
788 pr_info("Tracer %s already registered\n",
796 type->set_flag = &dummy_set_flag;
798 type->flags = &dummy_tracer_flags;
800 if (!type->flags->opts)
801 type->flags->opts = dummy_tracer_opt;
802 if (!type->wait_pipe)
803 type->wait_pipe = default_wait_pipe;
806 #ifdef CONFIG_FTRACE_STARTUP_TEST
807 if (type->selftest && !tracing_selftest_disabled) {
808 struct tracer *saved_tracer = current_trace;
809 struct trace_array *tr = &global_trace;
812 * Run a selftest on this tracer.
813 * Here we reset the trace buffer, and set the current
814 * tracer to be this tracer. The tracer can then run some
815 * internal tracing to verify that everything is in order.
816 * If we fail, we do not register this tracer.
818 tracing_reset_online_cpus(tr);
820 current_trace = type;
822 /* If we expanded the buffers, make sure the max is expanded too */
823 if (ring_buffer_expanded && type->use_max_tr)
824 ring_buffer_resize(max_tr.buffer, trace_buf_size,
825 RING_BUFFER_ALL_CPUS);
827 /* the test is responsible for initializing and enabling */
828 pr_info("Testing tracer %s: ", type->name);
829 ret = type->selftest(type, tr);
830 /* the test is responsible for resetting too */
831 current_trace = saved_tracer;
833 printk(KERN_CONT "FAILED!\n");
834 /* Add the warning after printing 'FAILED' */
838 /* Only reset on passing, to avoid touching corrupted buffers */
839 tracing_reset_online_cpus(tr);
841 /* Shrink the max buffer again */
842 if (ring_buffer_expanded && type->use_max_tr)
843 ring_buffer_resize(max_tr.buffer, 1,
844 RING_BUFFER_ALL_CPUS);
846 printk(KERN_CONT "PASSED\n");
850 type->next = trace_types;
854 tracing_selftest_running = false;
855 mutex_unlock(&trace_types_lock);
857 if (ret || !default_bootup_tracer)
860 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
863 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
864 /* Do we want this tracer to start on bootup? */
865 tracing_set_tracer(type->name);
866 default_bootup_tracer = NULL;
867 /* disable other selftests, since this will break it. */
868 tracing_selftest_disabled = 1;
869 #ifdef CONFIG_FTRACE_STARTUP_TEST
870 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
878 void unregister_tracer(struct tracer *type)
882 mutex_lock(&trace_types_lock);
883 for (t = &trace_types; *t; t = &(*t)->next) {
887 pr_info("Tracer %s not registered\n", type->name);
893 if (type == current_trace && tracer_enabled) {
896 if (current_trace->stop)
897 current_trace->stop(&global_trace);
898 current_trace = &nop_trace;
901 mutex_unlock(&trace_types_lock);
904 void tracing_reset(struct trace_array *tr, int cpu)
906 struct ring_buffer *buffer = tr->buffer;
908 ring_buffer_record_disable(buffer);
910 /* Make sure all commits have finished */
912 ring_buffer_reset_cpu(buffer, cpu);
914 ring_buffer_record_enable(buffer);
917 void tracing_reset_online_cpus(struct trace_array *tr)
919 struct ring_buffer *buffer = tr->buffer;
922 ring_buffer_record_disable(buffer);
924 /* Make sure all commits have finished */
927 tr->time_start = ftrace_now(tr->cpu);
929 for_each_online_cpu(cpu)
930 ring_buffer_reset_cpu(buffer, cpu);
932 ring_buffer_record_enable(buffer);
935 void tracing_reset_current(int cpu)
937 tracing_reset(&global_trace, cpu);
940 void tracing_reset_current_online_cpus(void)
942 tracing_reset_online_cpus(&global_trace);
945 #define SAVED_CMDLINES 128
946 #define NO_CMDLINE_MAP UINT_MAX
947 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
948 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
949 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
950 static int cmdline_idx;
951 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
953 /* temporary disable recording */
954 static atomic_t trace_record_cmdline_disabled __read_mostly;
956 static void trace_init_cmdlines(void)
958 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
959 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
963 int is_tracing_stopped(void)
965 return trace_stop_count;
969 * ftrace_off_permanent - disable all ftrace code permanently
971 * This should only be called when a serious anomally has
972 * been detected. This will turn off the function tracing,
973 * ring buffers, and other tracing utilites. It takes no
974 * locks and can be called from any context.
976 void ftrace_off_permanent(void)
978 tracing_disabled = 1;
980 tracing_off_permanent();
984 * tracing_start - quick start of the tracer
986 * If tracing is enabled but was stopped by tracing_stop,
987 * this will start the tracer back up.
989 void tracing_start(void)
991 struct ring_buffer *buffer;
994 if (tracing_disabled)
997 raw_spin_lock_irqsave(&tracing_start_lock, flags);
998 if (--trace_stop_count) {
999 if (trace_stop_count < 0) {
1000 /* Someone screwed up their debugging */
1002 trace_stop_count = 0;
1007 /* Prevent the buffers from switching */
1008 arch_spin_lock(&ftrace_max_lock);
1010 buffer = global_trace.buffer;
1012 ring_buffer_record_enable(buffer);
1014 buffer = max_tr.buffer;
1016 ring_buffer_record_enable(buffer);
1018 arch_spin_unlock(&ftrace_max_lock);
1022 raw_spin_unlock_irqrestore(&tracing_start_lock, flags);
1026 * tracing_stop - quick stop of the tracer
1028 * Light weight way to stop tracing. Use in conjunction with
1031 void tracing_stop(void)
1033 struct ring_buffer *buffer;
1034 unsigned long flags;
1037 raw_spin_lock_irqsave(&tracing_start_lock, flags);
1038 if (trace_stop_count++)
1041 /* Prevent the buffers from switching */
1042 arch_spin_lock(&ftrace_max_lock);
1044 buffer = global_trace.buffer;
1046 ring_buffer_record_disable(buffer);
1048 buffer = max_tr.buffer;
1050 ring_buffer_record_disable(buffer);
1052 arch_spin_unlock(&ftrace_max_lock);
1055 raw_spin_unlock_irqrestore(&tracing_start_lock, flags);
1058 void trace_stop_cmdline_recording(void);
1060 static void trace_save_cmdline(struct task_struct *tsk)
1064 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1068 * It's not the end of the world if we don't get
1069 * the lock, but we also don't want to spin
1070 * nor do we want to disable interrupts,
1071 * so if we miss here, then better luck next time.
1073 if (!arch_spin_trylock(&trace_cmdline_lock))
1076 idx = map_pid_to_cmdline[tsk->pid];
1077 if (idx == NO_CMDLINE_MAP) {
1078 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1081 * Check whether the cmdline buffer at idx has a pid
1082 * mapped. We are going to overwrite that entry so we
1083 * need to clear the map_pid_to_cmdline. Otherwise we
1084 * would read the new comm for the old pid.
1086 pid = map_cmdline_to_pid[idx];
1087 if (pid != NO_CMDLINE_MAP)
1088 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1090 map_cmdline_to_pid[idx] = tsk->pid;
1091 map_pid_to_cmdline[tsk->pid] = idx;
1096 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1098 arch_spin_unlock(&trace_cmdline_lock);
1101 void trace_find_cmdline(int pid, char comm[])
1106 strcpy(comm, "<idle>");
1110 if (WARN_ON_ONCE(pid < 0)) {
1111 strcpy(comm, "<XXX>");
1115 if (pid > PID_MAX_DEFAULT) {
1116 strcpy(comm, "<...>");
1121 arch_spin_lock(&trace_cmdline_lock);
1122 map = map_pid_to_cmdline[pid];
1123 if (map != NO_CMDLINE_MAP)
1124 strcpy(comm, saved_cmdlines[map]);
1126 strcpy(comm, "<...>");
1128 arch_spin_unlock(&trace_cmdline_lock);
1132 void tracing_record_cmdline(struct task_struct *tsk)
1134 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1138 trace_save_cmdline(tsk);
1142 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1145 struct task_struct *tsk = current;
1147 entry->preempt_count = pc & 0xff;
1148 entry->pid = (tsk) ? tsk->pid : 0;
1151 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1152 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1154 TRACE_FLAG_IRQS_NOSUPPORT |
1156 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1157 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1158 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1160 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1162 struct ring_buffer_event *
1163 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1166 unsigned long flags, int pc)
1168 struct ring_buffer_event *event;
1170 event = ring_buffer_lock_reserve(buffer, len);
1171 if (event != NULL) {
1172 struct trace_entry *ent = ring_buffer_event_data(event);
1174 tracing_generic_entry_update(ent, flags, pc);
1182 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1183 struct ring_buffer_event *event,
1184 unsigned long flags, int pc,
1187 ring_buffer_unlock_commit(buffer, event);
1189 ftrace_trace_stack(buffer, flags, 6, pc);
1190 ftrace_trace_userstack(buffer, flags, pc);
1196 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1197 struct ring_buffer_event *event,
1198 unsigned long flags, int pc)
1200 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1203 struct ring_buffer_event *
1204 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1205 int type, unsigned long len,
1206 unsigned long flags, int pc)
1208 *current_rb = global_trace.buffer;
1209 return trace_buffer_lock_reserve(*current_rb,
1210 type, len, flags, pc);
1212 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1214 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1215 struct ring_buffer_event *event,
1216 unsigned long flags, int pc)
1218 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1220 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1222 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1223 struct ring_buffer_event *event,
1224 unsigned long flags, int pc)
1226 __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1228 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1230 void trace_nowake_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1231 struct ring_buffer_event *event,
1232 unsigned long flags, int pc,
1233 struct pt_regs *regs)
1235 ring_buffer_unlock_commit(buffer, event);
1237 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1238 ftrace_trace_userstack(buffer, flags, pc);
1240 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit_regs);
1242 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1243 struct ring_buffer_event *event)
1245 ring_buffer_discard_commit(buffer, event);
1247 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1250 trace_function(struct trace_array *tr,
1251 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1254 struct ftrace_event_call *call = &event_function;
1255 struct ring_buffer *buffer = tr->buffer;
1256 struct ring_buffer_event *event;
1257 struct ftrace_entry *entry;
1259 /* If we are reading the ring buffer, don't trace */
1260 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1263 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1267 entry = ring_buffer_event_data(event);
1269 entry->parent_ip = parent_ip;
1271 if (!filter_check_discard(call, entry, buffer, event))
1272 ring_buffer_unlock_commit(buffer, event);
1276 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1277 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1280 if (likely(!atomic_read(&data->disabled)))
1281 trace_function(tr, ip, parent_ip, flags, pc);
1284 #ifdef CONFIG_STACKTRACE
1286 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1287 struct ftrace_stack {
1288 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1291 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1292 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1294 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1295 unsigned long flags,
1296 int skip, int pc, struct pt_regs *regs)
1298 struct ftrace_event_call *call = &event_kernel_stack;
1299 struct ring_buffer_event *event;
1300 struct stack_entry *entry;
1301 struct stack_trace trace;
1303 int size = FTRACE_STACK_ENTRIES;
1305 trace.nr_entries = 0;
1309 * Since events can happen in NMIs there's no safe way to
1310 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1311 * or NMI comes in, it will just have to use the default
1312 * FTRACE_STACK_SIZE.
1314 preempt_disable_notrace();
1316 use_stack = ++__get_cpu_var(ftrace_stack_reserve);
1318 * We don't need any atomic variables, just a barrier.
1319 * If an interrupt comes in, we don't care, because it would
1320 * have exited and put the counter back to what we want.
1321 * We just need a barrier to keep gcc from moving things
1325 if (use_stack == 1) {
1326 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1327 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1330 save_stack_trace_regs(regs, &trace);
1332 save_stack_trace(&trace);
1334 if (trace.nr_entries > size)
1335 size = trace.nr_entries;
1337 /* From now on, use_stack is a boolean */
1340 size *= sizeof(unsigned long);
1342 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1343 sizeof(*entry) + size, flags, pc);
1346 entry = ring_buffer_event_data(event);
1348 memset(&entry->caller, 0, size);
1351 memcpy(&entry->caller, trace.entries,
1352 trace.nr_entries * sizeof(unsigned long));
1354 trace.max_entries = FTRACE_STACK_ENTRIES;
1355 trace.entries = entry->caller;
1357 save_stack_trace_regs(regs, &trace);
1359 save_stack_trace(&trace);
1362 entry->size = trace.nr_entries;
1364 if (!filter_check_discard(call, entry, buffer, event))
1365 ring_buffer_unlock_commit(buffer, event);
1368 /* Again, don't let gcc optimize things here */
1370 __get_cpu_var(ftrace_stack_reserve)--;
1371 preempt_enable_notrace();
1375 void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1376 int skip, int pc, struct pt_regs *regs)
1378 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1381 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1384 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1387 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1390 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1393 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1396 __ftrace_trace_stack(tr->buffer, flags, skip, pc, NULL);
1400 * trace_dump_stack - record a stack back trace in the trace buffer
1402 void trace_dump_stack(void)
1404 unsigned long flags;
1406 if (tracing_disabled || tracing_selftest_running)
1409 local_save_flags(flags);
1411 /* skipping 3 traces, seems to get us at the caller of this function */
1412 __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count(), NULL);
1415 static DEFINE_PER_CPU(int, user_stack_count);
1418 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1420 struct ftrace_event_call *call = &event_user_stack;
1421 struct ring_buffer_event *event;
1422 struct userstack_entry *entry;
1423 struct stack_trace trace;
1425 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1429 * NMIs can not handle page faults, even with fix ups.
1430 * The save user stack can (and often does) fault.
1432 if (unlikely(in_nmi()))
1436 * prevent recursion, since the user stack tracing may
1437 * trigger other kernel events.
1440 if (__this_cpu_read(user_stack_count))
1443 __this_cpu_inc(user_stack_count);
1445 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1446 sizeof(*entry), flags, pc);
1448 goto out_drop_count;
1449 entry = ring_buffer_event_data(event);
1451 entry->tgid = current->tgid;
1452 memset(&entry->caller, 0, sizeof(entry->caller));
1454 trace.nr_entries = 0;
1455 trace.max_entries = FTRACE_STACK_ENTRIES;
1457 trace.entries = entry->caller;
1459 save_stack_trace_user(&trace);
1460 if (!filter_check_discard(call, entry, buffer, event))
1461 ring_buffer_unlock_commit(buffer, event);
1464 __this_cpu_dec(user_stack_count);
1470 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1472 ftrace_trace_userstack(tr, flags, preempt_count());
1476 #endif /* CONFIG_STACKTRACE */
1478 /* created for use with alloc_percpu */
1479 struct trace_buffer_struct {
1480 char buffer[TRACE_BUF_SIZE];
1483 static struct trace_buffer_struct *trace_percpu_buffer;
1484 static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1485 static struct trace_buffer_struct *trace_percpu_irq_buffer;
1486 static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1489 * The buffer used is dependent on the context. There is a per cpu
1490 * buffer for normal context, softirq contex, hard irq context and
1491 * for NMI context. Thise allows for lockless recording.
1493 * Note, if the buffers failed to be allocated, then this returns NULL
1495 static char *get_trace_buf(void)
1497 struct trace_buffer_struct *percpu_buffer;
1498 struct trace_buffer_struct *buffer;
1501 * If we have allocated per cpu buffers, then we do not
1502 * need to do any locking.
1505 percpu_buffer = trace_percpu_nmi_buffer;
1507 percpu_buffer = trace_percpu_irq_buffer;
1508 else if (in_softirq())
1509 percpu_buffer = trace_percpu_sirq_buffer;
1511 percpu_buffer = trace_percpu_buffer;
1516 buffer = per_cpu_ptr(percpu_buffer, smp_processor_id());
1518 return buffer->buffer;
1521 static int alloc_percpu_trace_buffer(void)
1523 struct trace_buffer_struct *buffers;
1524 struct trace_buffer_struct *sirq_buffers;
1525 struct trace_buffer_struct *irq_buffers;
1526 struct trace_buffer_struct *nmi_buffers;
1528 buffers = alloc_percpu(struct trace_buffer_struct);
1532 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
1536 irq_buffers = alloc_percpu(struct trace_buffer_struct);
1540 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
1544 trace_percpu_buffer = buffers;
1545 trace_percpu_sirq_buffer = sirq_buffers;
1546 trace_percpu_irq_buffer = irq_buffers;
1547 trace_percpu_nmi_buffer = nmi_buffers;
1552 free_percpu(irq_buffers);
1554 free_percpu(sirq_buffers);
1556 free_percpu(buffers);
1558 WARN(1, "Could not allocate percpu trace_printk buffer");
1562 void trace_printk_init_buffers(void)
1564 static int buffers_allocated;
1566 if (buffers_allocated)
1569 if (alloc_percpu_trace_buffer())
1572 pr_info("ftrace: Allocated trace_printk buffers\n");
1574 /* Expand the buffers to set size */
1575 tracing_update_buffers();
1577 buffers_allocated = 1;
1581 * trace_vbprintk - write binary msg to tracing buffer
1584 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1586 struct ftrace_event_call *call = &event_bprint;
1587 struct ring_buffer_event *event;
1588 struct ring_buffer *buffer;
1589 struct trace_array *tr = &global_trace;
1590 struct bprint_entry *entry;
1591 unsigned long flags;
1593 int len = 0, size, pc;
1595 if (unlikely(tracing_selftest_running || tracing_disabled))
1598 /* Don't pollute graph traces with trace_vprintk internals */
1599 pause_graph_tracing();
1601 pc = preempt_count();
1602 preempt_disable_notrace();
1604 tbuffer = get_trace_buf();
1610 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
1612 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
1615 local_save_flags(flags);
1616 size = sizeof(*entry) + sizeof(u32) * len;
1617 buffer = tr->buffer;
1618 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1622 entry = ring_buffer_event_data(event);
1626 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
1627 if (!filter_check_discard(call, entry, buffer, event)) {
1628 ring_buffer_unlock_commit(buffer, event);
1629 ftrace_trace_stack(buffer, flags, 6, pc);
1633 preempt_enable_notrace();
1634 unpause_graph_tracing();
1638 EXPORT_SYMBOL_GPL(trace_vbprintk);
1640 int trace_array_printk(struct trace_array *tr,
1641 unsigned long ip, const char *fmt, ...)
1646 if (!(trace_flags & TRACE_ITER_PRINTK))
1650 ret = trace_array_vprintk(tr, ip, fmt, ap);
1655 int trace_array_vprintk(struct trace_array *tr,
1656 unsigned long ip, const char *fmt, va_list args)
1658 struct ftrace_event_call *call = &event_print;
1659 struct ring_buffer_event *event;
1660 struct ring_buffer *buffer;
1661 int len = 0, size, pc;
1662 struct print_entry *entry;
1663 unsigned long flags;
1666 if (tracing_disabled || tracing_selftest_running)
1669 /* Don't pollute graph traces with trace_vprintk internals */
1670 pause_graph_tracing();
1672 pc = preempt_count();
1673 preempt_disable_notrace();
1676 tbuffer = get_trace_buf();
1682 len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
1683 if (len > TRACE_BUF_SIZE)
1686 local_save_flags(flags);
1687 size = sizeof(*entry) + len + 1;
1688 buffer = tr->buffer;
1689 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1693 entry = ring_buffer_event_data(event);
1696 memcpy(&entry->buf, tbuffer, len);
1697 entry->buf[len] = '\0';
1698 if (!filter_check_discard(call, entry, buffer, event)) {
1699 ring_buffer_unlock_commit(buffer, event);
1700 ftrace_trace_stack(buffer, flags, 6, pc);
1703 preempt_enable_notrace();
1704 unpause_graph_tracing();
1709 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1711 return trace_array_vprintk(&global_trace, ip, fmt, args);
1713 EXPORT_SYMBOL_GPL(trace_vprintk);
1715 static void trace_iterator_increment(struct trace_iterator *iter)
1717 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
1721 ring_buffer_read(buf_iter, NULL);
1724 static struct trace_entry *
1725 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1726 unsigned long *lost_events)
1728 struct ring_buffer_event *event;
1729 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
1732 event = ring_buffer_iter_peek(buf_iter, ts);
1734 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1738 iter->ent_size = ring_buffer_event_length(event);
1739 return ring_buffer_event_data(event);
1745 static struct trace_entry *
1746 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1747 unsigned long *missing_events, u64 *ent_ts)
1749 struct ring_buffer *buffer = iter->tr->buffer;
1750 struct trace_entry *ent, *next = NULL;
1751 unsigned long lost_events = 0, next_lost = 0;
1752 int cpu_file = iter->cpu_file;
1753 u64 next_ts = 0, ts;
1759 * If we are in a per_cpu trace file, don't bother by iterating over
1760 * all cpu and peek directly.
1762 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1763 if (ring_buffer_empty_cpu(buffer, cpu_file))
1765 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1767 *ent_cpu = cpu_file;
1772 for_each_tracing_cpu(cpu) {
1774 if (ring_buffer_empty_cpu(buffer, cpu))
1777 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1780 * Pick the entry with the smallest timestamp:
1782 if (ent && (!next || ts < next_ts)) {
1786 next_lost = lost_events;
1787 next_size = iter->ent_size;
1791 iter->ent_size = next_size;
1794 *ent_cpu = next_cpu;
1800 *missing_events = next_lost;
1805 /* Find the next real entry, without updating the iterator itself */
1806 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1807 int *ent_cpu, u64 *ent_ts)
1809 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1812 /* Find the next real entry, and increment the iterator to the next entry */
1813 void *trace_find_next_entry_inc(struct trace_iterator *iter)
1815 iter->ent = __find_next_entry(iter, &iter->cpu,
1816 &iter->lost_events, &iter->ts);
1819 trace_iterator_increment(iter);
1821 return iter->ent ? iter : NULL;
1824 static void trace_consume(struct trace_iterator *iter)
1826 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1827 &iter->lost_events);
1830 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1832 struct trace_iterator *iter = m->private;
1836 WARN_ON_ONCE(iter->leftover);
1840 /* can't go backwards */
1845 ent = trace_find_next_entry_inc(iter);
1849 while (ent && iter->idx < i)
1850 ent = trace_find_next_entry_inc(iter);
1857 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1859 struct trace_array *tr = iter->tr;
1860 struct ring_buffer_event *event;
1861 struct ring_buffer_iter *buf_iter;
1862 unsigned long entries = 0;
1865 tr->data[cpu]->skipped_entries = 0;
1867 buf_iter = trace_buffer_iter(iter, cpu);
1871 ring_buffer_iter_reset(buf_iter);
1874 * We could have the case with the max latency tracers
1875 * that a reset never took place on a cpu. This is evident
1876 * by the timestamp being before the start of the buffer.
1878 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1879 if (ts >= iter->tr->time_start)
1882 ring_buffer_read(buf_iter, NULL);
1885 tr->data[cpu]->skipped_entries = entries;
1889 * The current tracer is copied to avoid a global locking
1892 static void *s_start(struct seq_file *m, loff_t *pos)
1894 struct trace_iterator *iter = m->private;
1895 static struct tracer *old_tracer;
1896 int cpu_file = iter->cpu_file;
1901 /* copy the tracer to avoid using a global lock all around */
1902 mutex_lock(&trace_types_lock);
1903 if (unlikely(old_tracer != current_trace && current_trace)) {
1904 old_tracer = current_trace;
1905 *iter->trace = *current_trace;
1907 mutex_unlock(&trace_types_lock);
1909 atomic_inc(&trace_record_cmdline_disabled);
1911 if (*pos != iter->pos) {
1916 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1917 for_each_tracing_cpu(cpu)
1918 tracing_iter_reset(iter, cpu);
1920 tracing_iter_reset(iter, cpu_file);
1923 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1928 * If we overflowed the seq_file before, then we want
1929 * to just reuse the trace_seq buffer again.
1935 p = s_next(m, p, &l);
1939 trace_event_read_lock();
1940 trace_access_lock(cpu_file);
1944 static void s_stop(struct seq_file *m, void *p)
1946 struct trace_iterator *iter = m->private;
1948 atomic_dec(&trace_record_cmdline_disabled);
1949 trace_access_unlock(iter->cpu_file);
1950 trace_event_read_unlock();
1954 get_total_entries(struct trace_array *tr, unsigned long *total, unsigned long *entries)
1956 unsigned long count;
1962 for_each_tracing_cpu(cpu) {
1963 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1965 * If this buffer has skipped entries, then we hold all
1966 * entries for the trace and we need to ignore the
1967 * ones before the time stamp.
1969 if (tr->data[cpu]->skipped_entries) {
1970 count -= tr->data[cpu]->skipped_entries;
1971 /* total is the same as the entries */
1975 ring_buffer_overrun_cpu(tr->buffer, cpu);
1980 static void print_lat_help_header(struct seq_file *m)
1982 seq_puts(m, "# _------=> CPU# \n");
1983 seq_puts(m, "# / _-----=> irqs-off \n");
1984 seq_puts(m, "# | / _----=> need-resched \n");
1985 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1986 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1987 seq_puts(m, "# |||| / delay \n");
1988 seq_puts(m, "# cmd pid ||||| time | caller \n");
1989 seq_puts(m, "# \\ / ||||| \\ | / \n");
1992 static void print_event_info(struct trace_array *tr, struct seq_file *m)
1994 unsigned long total;
1995 unsigned long entries;
1997 get_total_entries(tr, &total, &entries);
1998 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
1999 entries, total, num_online_cpus());
2003 static void print_func_help_header(struct trace_array *tr, struct seq_file *m)
2005 print_event_info(tr, m);
2006 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2007 seq_puts(m, "# | | | | |\n");
2010 static void print_func_help_header_irq(struct trace_array *tr, struct seq_file *m)
2012 print_event_info(tr, m);
2013 seq_puts(m, "# _-----=> irqs-off\n");
2014 seq_puts(m, "# / _----=> need-resched\n");
2015 seq_puts(m, "# | / _---=> hardirq/softirq\n");
2016 seq_puts(m, "# || / _--=> preempt-depth\n");
2017 seq_puts(m, "# ||| / delay\n");
2018 seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2019 seq_puts(m, "# | | | |||| | |\n");
2023 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2025 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2026 struct trace_array *tr = iter->tr;
2027 struct trace_array_cpu *data = tr->data[tr->cpu];
2028 struct tracer *type = current_trace;
2029 unsigned long entries;
2030 unsigned long total;
2031 const char *name = "preemption";
2036 get_total_entries(tr, &total, &entries);
2038 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2040 seq_puts(m, "# -----------------------------------"
2041 "---------------------------------\n");
2042 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2043 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2044 nsecs_to_usecs(data->saved_latency),
2048 #if defined(CONFIG_PREEMPT_NONE)
2050 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2052 #elif defined(CONFIG_PREEMPT)
2057 /* These are reserved for later use */
2060 seq_printf(m, " #P:%d)\n", num_online_cpus());
2064 seq_puts(m, "# -----------------\n");
2065 seq_printf(m, "# | task: %.16s-%d "
2066 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2067 data->comm, data->pid,
2068 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
2069 data->policy, data->rt_priority);
2070 seq_puts(m, "# -----------------\n");
2072 if (data->critical_start) {
2073 seq_puts(m, "# => started at: ");
2074 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2075 trace_print_seq(m, &iter->seq);
2076 seq_puts(m, "\n# => ended at: ");
2077 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2078 trace_print_seq(m, &iter->seq);
2079 seq_puts(m, "\n#\n");
2085 static void test_cpu_buff_start(struct trace_iterator *iter)
2087 struct trace_seq *s = &iter->seq;
2089 if (!(trace_flags & TRACE_ITER_ANNOTATE))
2092 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2095 if (cpumask_test_cpu(iter->cpu, iter->started))
2098 if (iter->tr->data[iter->cpu]->skipped_entries)
2101 cpumask_set_cpu(iter->cpu, iter->started);
2103 /* Don't print started cpu buffer for the first entry of the trace */
2105 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2109 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2111 struct trace_seq *s = &iter->seq;
2112 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2113 struct trace_entry *entry;
2114 struct trace_event *event;
2118 test_cpu_buff_start(iter);
2120 event = ftrace_find_event(entry->type);
2122 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2123 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2124 if (!trace_print_lat_context(iter))
2127 if (!trace_print_context(iter))
2133 return event->funcs->trace(iter, sym_flags, event);
2135 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2138 return TRACE_TYPE_HANDLED;
2140 return TRACE_TYPE_PARTIAL_LINE;
2143 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2145 struct trace_seq *s = &iter->seq;
2146 struct trace_entry *entry;
2147 struct trace_event *event;
2151 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2152 if (!trace_seq_printf(s, "%d %d %llu ",
2153 entry->pid, iter->cpu, iter->ts))
2157 event = ftrace_find_event(entry->type);
2159 return event->funcs->raw(iter, 0, event);
2161 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2164 return TRACE_TYPE_HANDLED;
2166 return TRACE_TYPE_PARTIAL_LINE;
2169 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2171 struct trace_seq *s = &iter->seq;
2172 unsigned char newline = '\n';
2173 struct trace_entry *entry;
2174 struct trace_event *event;
2178 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2179 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2180 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2181 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2184 event = ftrace_find_event(entry->type);
2186 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2187 if (ret != TRACE_TYPE_HANDLED)
2191 SEQ_PUT_FIELD_RET(s, newline);
2193 return TRACE_TYPE_HANDLED;
2196 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2198 struct trace_seq *s = &iter->seq;
2199 struct trace_entry *entry;
2200 struct trace_event *event;
2204 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2205 SEQ_PUT_FIELD_RET(s, entry->pid);
2206 SEQ_PUT_FIELD_RET(s, iter->cpu);
2207 SEQ_PUT_FIELD_RET(s, iter->ts);
2210 event = ftrace_find_event(entry->type);
2211 return event ? event->funcs->binary(iter, 0, event) :
2215 int trace_empty(struct trace_iterator *iter)
2217 struct ring_buffer_iter *buf_iter;
2220 /* If we are looking at one CPU buffer, only check that one */
2221 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
2222 cpu = iter->cpu_file;
2223 buf_iter = trace_buffer_iter(iter, cpu);
2225 if (!ring_buffer_iter_empty(buf_iter))
2228 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2234 for_each_tracing_cpu(cpu) {
2235 buf_iter = trace_buffer_iter(iter, cpu);
2237 if (!ring_buffer_iter_empty(buf_iter))
2240 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2248 /* Called with trace_event_read_lock() held. */
2249 enum print_line_t print_trace_line(struct trace_iterator *iter)
2251 enum print_line_t ret;
2253 if (iter->lost_events &&
2254 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2255 iter->cpu, iter->lost_events))
2256 return TRACE_TYPE_PARTIAL_LINE;
2258 if (iter->trace && iter->trace->print_line) {
2259 ret = iter->trace->print_line(iter);
2260 if (ret != TRACE_TYPE_UNHANDLED)
2264 if (iter->ent->type == TRACE_BPRINT &&
2265 trace_flags & TRACE_ITER_PRINTK &&
2266 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2267 return trace_print_bprintk_msg_only(iter);
2269 if (iter->ent->type == TRACE_PRINT &&
2270 trace_flags & TRACE_ITER_PRINTK &&
2271 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2272 return trace_print_printk_msg_only(iter);
2274 if (trace_flags & TRACE_ITER_BIN)
2275 return print_bin_fmt(iter);
2277 if (trace_flags & TRACE_ITER_HEX)
2278 return print_hex_fmt(iter);
2280 if (trace_flags & TRACE_ITER_RAW)
2281 return print_raw_fmt(iter);
2283 return print_trace_fmt(iter);
2286 void trace_latency_header(struct seq_file *m)
2288 struct trace_iterator *iter = m->private;
2290 /* print nothing if the buffers are empty */
2291 if (trace_empty(iter))
2294 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2295 print_trace_header(m, iter);
2297 if (!(trace_flags & TRACE_ITER_VERBOSE))
2298 print_lat_help_header(m);
2301 void trace_default_header(struct seq_file *m)
2303 struct trace_iterator *iter = m->private;
2305 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2308 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2309 /* print nothing if the buffers are empty */
2310 if (trace_empty(iter))
2312 print_trace_header(m, iter);
2313 if (!(trace_flags & TRACE_ITER_VERBOSE))
2314 print_lat_help_header(m);
2316 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2317 if (trace_flags & TRACE_ITER_IRQ_INFO)
2318 print_func_help_header_irq(iter->tr, m);
2320 print_func_help_header(iter->tr, m);
2325 static void test_ftrace_alive(struct seq_file *m)
2327 if (!ftrace_is_dead())
2329 seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2330 seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2333 static int s_show(struct seq_file *m, void *v)
2335 struct trace_iterator *iter = v;
2338 if (iter->ent == NULL) {
2340 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2342 test_ftrace_alive(m);
2344 if (iter->trace && iter->trace->print_header)
2345 iter->trace->print_header(m);
2347 trace_default_header(m);
2349 } else if (iter->leftover) {
2351 * If we filled the seq_file buffer earlier, we
2352 * want to just show it now.
2354 ret = trace_print_seq(m, &iter->seq);
2356 /* ret should this time be zero, but you never know */
2357 iter->leftover = ret;
2360 print_trace_line(iter);
2361 ret = trace_print_seq(m, &iter->seq);
2363 * If we overflow the seq_file buffer, then it will
2364 * ask us for this data again at start up.
2366 * ret is 0 if seq_file write succeeded.
2369 iter->leftover = ret;
2375 static const struct seq_operations tracer_seq_ops = {
2382 static struct trace_iterator *
2383 __tracing_open(struct inode *inode, struct file *file)
2385 long cpu_file = (long) inode->i_private;
2386 struct trace_iterator *iter;
2389 if (tracing_disabled)
2390 return ERR_PTR(-ENODEV);
2392 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
2394 return ERR_PTR(-ENOMEM);
2396 iter->buffer_iter = kzalloc(sizeof(*iter->buffer_iter) * num_possible_cpus(),
2398 if (!iter->buffer_iter)
2402 * We make a copy of the current tracer to avoid concurrent
2403 * changes on it while we are reading.
2405 mutex_lock(&trace_types_lock);
2406 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2411 *iter->trace = *current_trace;
2413 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2416 if (current_trace && current_trace->print_max)
2419 iter->tr = &global_trace;
2421 mutex_init(&iter->mutex);
2422 iter->cpu_file = cpu_file;
2424 /* Notify the tracer early; before we stop tracing. */
2425 if (iter->trace && iter->trace->open)
2426 iter->trace->open(iter);
2428 /* Annotate start of buffers if we had overruns */
2429 if (ring_buffer_overruns(iter->tr->buffer))
2430 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2432 /* stop the trace while dumping */
2435 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2436 for_each_tracing_cpu(cpu) {
2437 iter->buffer_iter[cpu] =
2438 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2440 ring_buffer_read_prepare_sync();
2441 for_each_tracing_cpu(cpu) {
2442 ring_buffer_read_start(iter->buffer_iter[cpu]);
2443 tracing_iter_reset(iter, cpu);
2446 cpu = iter->cpu_file;
2447 iter->buffer_iter[cpu] =
2448 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2449 ring_buffer_read_prepare_sync();
2450 ring_buffer_read_start(iter->buffer_iter[cpu]);
2451 tracing_iter_reset(iter, cpu);
2454 mutex_unlock(&trace_types_lock);
2459 mutex_unlock(&trace_types_lock);
2461 kfree(iter->buffer_iter);
2463 seq_release_private(inode, file);
2464 return ERR_PTR(-ENOMEM);
2467 int tracing_open_generic(struct inode *inode, struct file *filp)
2469 if (tracing_disabled)
2472 filp->private_data = inode->i_private;
2476 static int tracing_release(struct inode *inode, struct file *file)
2478 struct seq_file *m = file->private_data;
2479 struct trace_iterator *iter;
2482 if (!(file->f_mode & FMODE_READ))
2487 mutex_lock(&trace_types_lock);
2488 for_each_tracing_cpu(cpu) {
2489 if (iter->buffer_iter[cpu])
2490 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2493 if (iter->trace && iter->trace->close)
2494 iter->trace->close(iter);
2496 /* reenable tracing if it was previously enabled */
2498 mutex_unlock(&trace_types_lock);
2500 mutex_destroy(&iter->mutex);
2501 free_cpumask_var(iter->started);
2503 kfree(iter->buffer_iter);
2504 seq_release_private(inode, file);
2508 static int tracing_open(struct inode *inode, struct file *file)
2510 struct trace_iterator *iter;
2513 /* If this file was open for write, then erase contents */
2514 if ((file->f_mode & FMODE_WRITE) &&
2515 (file->f_flags & O_TRUNC)) {
2516 long cpu = (long) inode->i_private;
2518 if (cpu == TRACE_PIPE_ALL_CPU)
2519 tracing_reset_online_cpus(&global_trace);
2521 tracing_reset(&global_trace, cpu);
2524 if (file->f_mode & FMODE_READ) {
2525 iter = __tracing_open(inode, file);
2527 ret = PTR_ERR(iter);
2528 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2529 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2535 t_next(struct seq_file *m, void *v, loff_t *pos)
2537 struct tracer *t = v;
2547 static void *t_start(struct seq_file *m, loff_t *pos)
2552 mutex_lock(&trace_types_lock);
2553 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2559 static void t_stop(struct seq_file *m, void *p)
2561 mutex_unlock(&trace_types_lock);
2564 static int t_show(struct seq_file *m, void *v)
2566 struct tracer *t = v;
2571 seq_printf(m, "%s", t->name);
2580 static const struct seq_operations show_traces_seq_ops = {
2587 static int show_traces_open(struct inode *inode, struct file *file)
2589 if (tracing_disabled)
2592 return seq_open(file, &show_traces_seq_ops);
2596 tracing_write_stub(struct file *filp, const char __user *ubuf,
2597 size_t count, loff_t *ppos)
2602 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
2604 if (file->f_mode & FMODE_READ)
2605 return seq_lseek(file, offset, origin);
2610 static const struct file_operations tracing_fops = {
2611 .open = tracing_open,
2613 .write = tracing_write_stub,
2614 .llseek = tracing_seek,
2615 .release = tracing_release,
2618 static const struct file_operations show_traces_fops = {
2619 .open = show_traces_open,
2621 .release = seq_release,
2622 .llseek = seq_lseek,
2626 * Only trace on a CPU if the bitmask is set:
2628 static cpumask_var_t tracing_cpumask;
2631 * The tracer itself will not take this lock, but still we want
2632 * to provide a consistent cpumask to user-space:
2634 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2637 * Temporary storage for the character representation of the
2638 * CPU bitmask (and one more byte for the newline):
2640 static char mask_str[NR_CPUS + 1];
2643 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2644 size_t count, loff_t *ppos)
2648 mutex_lock(&tracing_cpumask_update_lock);
2650 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2651 if (count - len < 2) {
2655 len += sprintf(mask_str + len, "\n");
2656 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2659 mutex_unlock(&tracing_cpumask_update_lock);
2665 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2666 size_t count, loff_t *ppos)
2669 cpumask_var_t tracing_cpumask_new;
2671 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2674 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2678 mutex_lock(&tracing_cpumask_update_lock);
2680 local_irq_disable();
2681 arch_spin_lock(&ftrace_max_lock);
2682 for_each_tracing_cpu(cpu) {
2684 * Increase/decrease the disabled counter if we are
2685 * about to flip a bit in the cpumask:
2687 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2688 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2689 atomic_inc(&global_trace.data[cpu]->disabled);
2690 ring_buffer_record_disable_cpu(global_trace.buffer, cpu);
2692 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2693 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2694 atomic_dec(&global_trace.data[cpu]->disabled);
2695 ring_buffer_record_enable_cpu(global_trace.buffer, cpu);
2698 arch_spin_unlock(&ftrace_max_lock);
2701 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2703 mutex_unlock(&tracing_cpumask_update_lock);
2704 free_cpumask_var(tracing_cpumask_new);
2709 free_cpumask_var(tracing_cpumask_new);
2714 static const struct file_operations tracing_cpumask_fops = {
2715 .open = tracing_open_generic,
2716 .read = tracing_cpumask_read,
2717 .write = tracing_cpumask_write,
2718 .llseek = generic_file_llseek,
2721 static int tracing_trace_options_show(struct seq_file *m, void *v)
2723 struct tracer_opt *trace_opts;
2727 mutex_lock(&trace_types_lock);
2728 tracer_flags = current_trace->flags->val;
2729 trace_opts = current_trace->flags->opts;
2731 for (i = 0; trace_options[i]; i++) {
2732 if (trace_flags & (1 << i))
2733 seq_printf(m, "%s\n", trace_options[i]);
2735 seq_printf(m, "no%s\n", trace_options[i]);
2738 for (i = 0; trace_opts[i].name; i++) {
2739 if (tracer_flags & trace_opts[i].bit)
2740 seq_printf(m, "%s\n", trace_opts[i].name);
2742 seq_printf(m, "no%s\n", trace_opts[i].name);
2744 mutex_unlock(&trace_types_lock);
2749 static int __set_tracer_option(struct tracer *trace,
2750 struct tracer_flags *tracer_flags,
2751 struct tracer_opt *opts, int neg)
2755 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2760 tracer_flags->val &= ~opts->bit;
2762 tracer_flags->val |= opts->bit;
2766 /* Try to assign a tracer specific option */
2767 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2769 struct tracer_flags *tracer_flags = trace->flags;
2770 struct tracer_opt *opts = NULL;
2773 for (i = 0; tracer_flags->opts[i].name; i++) {
2774 opts = &tracer_flags->opts[i];
2776 if (strcmp(cmp, opts->name) == 0)
2777 return __set_tracer_option(trace, trace->flags,
2784 static void set_tracer_flags(unsigned int mask, int enabled)
2786 /* do nothing if flag is already set */
2787 if (!!(trace_flags & mask) == !!enabled)
2791 trace_flags |= mask;
2793 trace_flags &= ~mask;
2795 if (mask == TRACE_ITER_RECORD_CMD)
2796 trace_event_enable_cmd_record(enabled);
2798 if (mask == TRACE_ITER_OVERWRITE)
2799 ring_buffer_change_overwrite(global_trace.buffer, enabled);
2803 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2804 size_t cnt, loff_t *ppos)
2812 if (cnt >= sizeof(buf))
2815 if (copy_from_user(&buf, ubuf, cnt))
2819 cmp = strstrip(buf);
2821 if (strncmp(cmp, "no", 2) == 0) {
2826 for (i = 0; trace_options[i]; i++) {
2827 if (strcmp(cmp, trace_options[i]) == 0) {
2828 set_tracer_flags(1 << i, !neg);
2833 /* If no option could be set, test the specific tracer options */
2834 if (!trace_options[i]) {
2835 mutex_lock(&trace_types_lock);
2836 ret = set_tracer_option(current_trace, cmp, neg);
2837 mutex_unlock(&trace_types_lock);
2847 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2849 if (tracing_disabled)
2851 return single_open(file, tracing_trace_options_show, NULL);
2854 static const struct file_operations tracing_iter_fops = {
2855 .open = tracing_trace_options_open,
2857 .llseek = seq_lseek,
2858 .release = single_release,
2859 .write = tracing_trace_options_write,
2862 static const char readme_msg[] =
2863 "tracing mini-HOWTO:\n\n"
2864 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2865 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2866 "wakeup wakeup_rt preemptirqsoff preemptoff irqsoff function nop\n\n"
2867 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2869 "# echo wakeup > /sys/kernel/debug/tracing/current_tracer\n"
2870 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2872 "# cat /sys/kernel/debug/tracing/trace_options\n"
2873 "noprint-parent nosym-offset nosym-addr noverbose\n"
2874 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2875 "# echo 1 > /sys/kernel/debug/tracing/tracing_on\n"
2876 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2877 "# echo 0 > /sys/kernel/debug/tracing/tracing_on\n"
2881 tracing_readme_read(struct file *filp, char __user *ubuf,
2882 size_t cnt, loff_t *ppos)
2884 return simple_read_from_buffer(ubuf, cnt, ppos,
2885 readme_msg, strlen(readme_msg));
2888 static const struct file_operations tracing_readme_fops = {
2889 .open = tracing_open_generic,
2890 .read = tracing_readme_read,
2891 .llseek = generic_file_llseek,
2895 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2896 size_t cnt, loff_t *ppos)
2905 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2909 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2917 for (i = 0; i < SAVED_CMDLINES; i++) {
2920 pid = map_cmdline_to_pid[i];
2921 if (pid == -1 || pid == NO_CMDLINE_MAP)
2924 trace_find_cmdline(pid, buf_comm);
2925 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2930 len = simple_read_from_buffer(ubuf, cnt, ppos,
2939 static const struct file_operations tracing_saved_cmdlines_fops = {
2940 .open = tracing_open_generic,
2941 .read = tracing_saved_cmdlines_read,
2942 .llseek = generic_file_llseek,
2946 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2947 size_t cnt, loff_t *ppos)
2952 r = sprintf(buf, "%u\n", tracer_enabled);
2953 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2957 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2958 size_t cnt, loff_t *ppos)
2960 struct trace_array *tr = filp->private_data;
2964 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
2970 mutex_lock(&trace_types_lock);
2971 if (tracer_enabled ^ val) {
2973 /* Only need to warn if this is used to change the state */
2974 WARN_ONCE(1, "tracing_enabled is deprecated. Use tracing_on");
2978 if (current_trace->start)
2979 current_trace->start(tr);
2984 if (current_trace->stop)
2985 current_trace->stop(tr);
2988 mutex_unlock(&trace_types_lock);
2996 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2997 size_t cnt, loff_t *ppos)
2999 char buf[MAX_TRACER_SIZE+2];
3002 mutex_lock(&trace_types_lock);
3004 r = sprintf(buf, "%s\n", current_trace->name);
3006 r = sprintf(buf, "\n");
3007 mutex_unlock(&trace_types_lock);
3009 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3012 int tracer_init(struct tracer *t, struct trace_array *tr)
3014 tracing_reset_online_cpus(tr);
3018 static void set_buffer_entries(struct trace_array *tr, unsigned long val)
3021 for_each_tracing_cpu(cpu)
3022 tr->data[cpu]->entries = val;
3025 static int __tracing_resize_ring_buffer(unsigned long size, int cpu)
3030 * If kernel or user changes the size of the ring buffer
3031 * we use the size that was given, and we can forget about
3032 * expanding it later.
3034 ring_buffer_expanded = 1;
3036 /* May be called before buffers are initialized */
3037 if (!global_trace.buffer)
3040 ret = ring_buffer_resize(global_trace.buffer, size, cpu);
3044 if (!current_trace->use_max_tr)
3047 ret = ring_buffer_resize(max_tr.buffer, size, cpu);
3051 if (cpu == RING_BUFFER_ALL_CPUS) {
3053 for_each_tracing_cpu(i) {
3054 r = ring_buffer_resize(global_trace.buffer,
3055 global_trace.data[i]->entries,
3061 r = ring_buffer_resize(global_trace.buffer,
3062 global_trace.data[cpu]->entries,
3068 * AARGH! We are left with different
3069 * size max buffer!!!!
3070 * The max buffer is our "snapshot" buffer.
3071 * When a tracer needs a snapshot (one of the
3072 * latency tracers), it swaps the max buffer
3073 * with the saved snap shot. We succeeded to
3074 * update the size of the main buffer, but failed to
3075 * update the size of the max buffer. But when we tried
3076 * to reset the main buffer to the original size, we
3077 * failed there too. This is very unlikely to
3078 * happen, but if it does, warn and kill all
3082 tracing_disabled = 1;
3087 if (cpu == RING_BUFFER_ALL_CPUS)
3088 set_buffer_entries(&max_tr, size);
3090 max_tr.data[cpu]->entries = size;
3093 if (cpu == RING_BUFFER_ALL_CPUS)
3094 set_buffer_entries(&global_trace, size);
3096 global_trace.data[cpu]->entries = size;
3101 static ssize_t tracing_resize_ring_buffer(unsigned long size, int cpu_id)
3105 mutex_lock(&trace_types_lock);
3107 if (cpu_id != RING_BUFFER_ALL_CPUS) {
3108 /* make sure, this cpu is enabled in the mask */
3109 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
3115 ret = __tracing_resize_ring_buffer(size, cpu_id);
3120 mutex_unlock(&trace_types_lock);
3127 * tracing_update_buffers - used by tracing facility to expand ring buffers
3129 * To save on memory when the tracing is never used on a system with it
3130 * configured in. The ring buffers are set to a minimum size. But once
3131 * a user starts to use the tracing facility, then they need to grow
3132 * to their default size.
3134 * This function is to be called when a tracer is about to be used.
3136 int tracing_update_buffers(void)
3140 mutex_lock(&trace_types_lock);
3141 if (!ring_buffer_expanded)
3142 ret = __tracing_resize_ring_buffer(trace_buf_size,
3143 RING_BUFFER_ALL_CPUS);
3144 mutex_unlock(&trace_types_lock);
3149 struct trace_option_dentry;
3151 static struct trace_option_dentry *
3152 create_trace_option_files(struct tracer *tracer);
3155 destroy_trace_option_files(struct trace_option_dentry *topts);
3157 static int tracing_set_tracer(const char *buf)
3159 static struct trace_option_dentry *topts;
3160 struct trace_array *tr = &global_trace;
3164 mutex_lock(&trace_types_lock);
3166 if (!ring_buffer_expanded) {
3167 ret = __tracing_resize_ring_buffer(trace_buf_size,
3168 RING_BUFFER_ALL_CPUS);
3174 for (t = trace_types; t; t = t->next) {
3175 if (strcmp(t->name, buf) == 0)
3182 if (t == current_trace)
3185 trace_branch_disable();
3186 if (current_trace && current_trace->reset)
3187 current_trace->reset(tr);
3188 if (current_trace && current_trace->use_max_tr) {
3190 * We don't free the ring buffer. instead, resize it because
3191 * The max_tr ring buffer has some state (e.g. ring->clock) and
3192 * we want preserve it.
3194 ring_buffer_resize(max_tr.buffer, 1, RING_BUFFER_ALL_CPUS);
3195 set_buffer_entries(&max_tr, 1);
3197 destroy_trace_option_files(topts);
3199 current_trace = &nop_trace;
3201 topts = create_trace_option_files(t);
3202 if (t->use_max_tr) {
3204 /* we need to make per cpu buffer sizes equivalent */
3205 for_each_tracing_cpu(cpu) {
3206 ret = ring_buffer_resize(max_tr.buffer,
3207 global_trace.data[cpu]->entries,
3211 max_tr.data[cpu]->entries =
3212 global_trace.data[cpu]->entries;
3217 ret = tracer_init(t, tr);
3223 trace_branch_enable(tr);
3225 mutex_unlock(&trace_types_lock);
3231 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3232 size_t cnt, loff_t *ppos)
3234 char buf[MAX_TRACER_SIZE+1];
3241 if (cnt > MAX_TRACER_SIZE)
3242 cnt = MAX_TRACER_SIZE;
3244 if (copy_from_user(&buf, ubuf, cnt))
3249 /* strip ending whitespace. */
3250 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3253 err = tracing_set_tracer(buf);
3263 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3264 size_t cnt, loff_t *ppos)
3266 unsigned long *ptr = filp->private_data;
3270 r = snprintf(buf, sizeof(buf), "%ld\n",
3271 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3272 if (r > sizeof(buf))
3274 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3278 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3279 size_t cnt, loff_t *ppos)
3281 unsigned long *ptr = filp->private_data;
3285 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3294 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3296 long cpu_file = (long) inode->i_private;
3297 struct trace_iterator *iter;
3300 if (tracing_disabled)
3303 mutex_lock(&trace_types_lock);
3305 /* create a buffer to store the information to pass to userspace */
3306 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3313 * We make a copy of the current tracer to avoid concurrent
3314 * changes on it while we are reading.
3316 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3322 *iter->trace = *current_trace;
3324 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3329 /* trace pipe does not show start of buffer */
3330 cpumask_setall(iter->started);
3332 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3333 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3335 iter->cpu_file = cpu_file;
3336 iter->tr = &global_trace;
3337 mutex_init(&iter->mutex);
3338 filp->private_data = iter;
3340 if (iter->trace->pipe_open)
3341 iter->trace->pipe_open(iter);
3343 nonseekable_open(inode, filp);
3345 mutex_unlock(&trace_types_lock);
3351 mutex_unlock(&trace_types_lock);
3355 static int tracing_release_pipe(struct inode *inode, struct file *file)
3357 struct trace_iterator *iter = file->private_data;
3359 mutex_lock(&trace_types_lock);
3361 if (iter->trace->pipe_close)
3362 iter->trace->pipe_close(iter);
3364 mutex_unlock(&trace_types_lock);
3366 free_cpumask_var(iter->started);
3367 mutex_destroy(&iter->mutex);
3375 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3377 struct trace_iterator *iter = filp->private_data;
3379 if (trace_flags & TRACE_ITER_BLOCK) {
3381 * Always select as readable when in blocking mode
3383 return POLLIN | POLLRDNORM;
3385 if (!trace_empty(iter))
3386 return POLLIN | POLLRDNORM;
3387 poll_wait(filp, &trace_wait, poll_table);
3388 if (!trace_empty(iter))
3389 return POLLIN | POLLRDNORM;
3396 void default_wait_pipe(struct trace_iterator *iter)
3400 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3402 if (trace_empty(iter))
3405 finish_wait(&trace_wait, &wait);
3409 * This is a make-shift waitqueue.
3410 * A tracer might use this callback on some rare cases:
3412 * 1) the current tracer might hold the runqueue lock when it wakes up
3413 * a reader, hence a deadlock (sched, function, and function graph tracers)
3414 * 2) the function tracers, trace all functions, we don't want
3415 * the overhead of calling wake_up and friends
3416 * (and tracing them too)
3418 * Anyway, this is really very primitive wakeup.
3420 void poll_wait_pipe(struct trace_iterator *iter)
3422 set_current_state(TASK_INTERRUPTIBLE);
3423 /* sleep for 100 msecs, and try again. */
3424 schedule_timeout(HZ / 10);
3427 /* Must be called with trace_types_lock mutex held. */
3428 static int tracing_wait_pipe(struct file *filp)
3430 struct trace_iterator *iter = filp->private_data;
3432 while (trace_empty(iter)) {
3434 if ((filp->f_flags & O_NONBLOCK)) {
3438 mutex_unlock(&iter->mutex);
3440 iter->trace->wait_pipe(iter);
3442 mutex_lock(&iter->mutex);
3444 if (signal_pending(current))
3448 * We block until we read something and tracing is disabled.
3449 * We still block if tracing is disabled, but we have never
3450 * read anything. This allows a user to cat this file, and
3451 * then enable tracing. But after we have read something,
3452 * we give an EOF when tracing is again disabled.
3454 * iter->pos will be 0 if we haven't read anything.
3456 if (!tracer_enabled && iter->pos)
3467 tracing_read_pipe(struct file *filp, char __user *ubuf,
3468 size_t cnt, loff_t *ppos)
3470 struct trace_iterator *iter = filp->private_data;
3471 static struct tracer *old_tracer;
3474 /* return any leftover data */
3475 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3479 trace_seq_init(&iter->seq);
3481 /* copy the tracer to avoid using a global lock all around */
3482 mutex_lock(&trace_types_lock);
3483 if (unlikely(old_tracer != current_trace && current_trace)) {
3484 old_tracer = current_trace;
3485 *iter->trace = *current_trace;
3487 mutex_unlock(&trace_types_lock);
3490 * Avoid more than one consumer on a single file descriptor
3491 * This is just a matter of traces coherency, the ring buffer itself
3494 mutex_lock(&iter->mutex);
3495 if (iter->trace->read) {
3496 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3502 sret = tracing_wait_pipe(filp);
3506 /* stop when tracing is finished */
3507 if (trace_empty(iter)) {
3512 if (cnt >= PAGE_SIZE)
3513 cnt = PAGE_SIZE - 1;
3515 /* reset all but tr, trace, and overruns */
3516 memset(&iter->seq, 0,
3517 sizeof(struct trace_iterator) -
3518 offsetof(struct trace_iterator, seq));
3521 trace_event_read_lock();
3522 trace_access_lock(iter->cpu_file);
3523 while (trace_find_next_entry_inc(iter) != NULL) {
3524 enum print_line_t ret;
3525 int len = iter->seq.len;
3527 ret = print_trace_line(iter);
3528 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3529 /* don't print partial lines */
3530 iter->seq.len = len;
3533 if (ret != TRACE_TYPE_NO_CONSUME)
3534 trace_consume(iter);
3536 if (iter->seq.len >= cnt)
3540 * Setting the full flag means we reached the trace_seq buffer
3541 * size and we should leave by partial output condition above.
3542 * One of the trace_seq_* functions is not used properly.
3544 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
3547 trace_access_unlock(iter->cpu_file);
3548 trace_event_read_unlock();
3550 /* Now copy what we have to the user */
3551 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3552 if (iter->seq.readpos >= iter->seq.len)
3553 trace_seq_init(&iter->seq);
3556 * If there was nothing to send to user, in spite of consuming trace
3557 * entries, go back to wait for more entries.
3563 mutex_unlock(&iter->mutex);
3568 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3569 struct pipe_buffer *buf)
3571 __free_page(buf->page);
3574 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3577 __free_page(spd->pages[idx]);
3580 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3582 .map = generic_pipe_buf_map,
3583 .unmap = generic_pipe_buf_unmap,
3584 .confirm = generic_pipe_buf_confirm,
3585 .release = tracing_pipe_buf_release,
3586 .steal = generic_pipe_buf_steal,
3587 .get = generic_pipe_buf_get,
3591 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3596 /* Seq buffer is page-sized, exactly what we need. */
3598 count = iter->seq.len;
3599 ret = print_trace_line(iter);
3600 count = iter->seq.len - count;
3603 iter->seq.len -= count;
3606 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3607 iter->seq.len -= count;
3611 if (ret != TRACE_TYPE_NO_CONSUME)
3612 trace_consume(iter);
3614 if (!trace_find_next_entry_inc(iter)) {
3624 static ssize_t tracing_splice_read_pipe(struct file *filp,
3626 struct pipe_inode_info *pipe,
3630 struct page *pages_def[PIPE_DEF_BUFFERS];
3631 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3632 struct trace_iterator *iter = filp->private_data;
3633 struct splice_pipe_desc spd = {
3635 .partial = partial_def,
3636 .nr_pages = 0, /* This gets updated below. */
3637 .nr_pages_max = PIPE_DEF_BUFFERS,
3639 .ops = &tracing_pipe_buf_ops,
3640 .spd_release = tracing_spd_release_pipe,
3642 static struct tracer *old_tracer;
3647 if (splice_grow_spd(pipe, &spd))
3650 /* copy the tracer to avoid using a global lock all around */
3651 mutex_lock(&trace_types_lock);
3652 if (unlikely(old_tracer != current_trace && current_trace)) {
3653 old_tracer = current_trace;
3654 *iter->trace = *current_trace;
3656 mutex_unlock(&trace_types_lock);
3658 mutex_lock(&iter->mutex);
3660 if (iter->trace->splice_read) {
3661 ret = iter->trace->splice_read(iter, filp,
3662 ppos, pipe, len, flags);
3667 ret = tracing_wait_pipe(filp);
3671 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
3676 trace_event_read_lock();
3677 trace_access_lock(iter->cpu_file);
3679 /* Fill as many pages as possible. */
3680 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3681 spd.pages[i] = alloc_page(GFP_KERNEL);
3685 rem = tracing_fill_pipe_page(rem, iter);
3687 /* Copy the data into the page, so we can start over. */
3688 ret = trace_seq_to_buffer(&iter->seq,
3689 page_address(spd.pages[i]),
3692 __free_page(spd.pages[i]);
3695 spd.partial[i].offset = 0;
3696 spd.partial[i].len = iter->seq.len;
3698 trace_seq_init(&iter->seq);
3701 trace_access_unlock(iter->cpu_file);
3702 trace_event_read_unlock();
3703 mutex_unlock(&iter->mutex);
3707 ret = splice_to_pipe(pipe, &spd);
3709 splice_shrink_spd(&spd);
3713 mutex_unlock(&iter->mutex);
3717 struct ftrace_entries_info {
3718 struct trace_array *tr;
3722 static int tracing_entries_open(struct inode *inode, struct file *filp)
3724 struct ftrace_entries_info *info;
3726 if (tracing_disabled)
3729 info = kzalloc(sizeof(*info), GFP_KERNEL);
3733 info->tr = &global_trace;
3734 info->cpu = (unsigned long)inode->i_private;
3736 filp->private_data = info;
3742 tracing_entries_read(struct file *filp, char __user *ubuf,
3743 size_t cnt, loff_t *ppos)
3745 struct ftrace_entries_info *info = filp->private_data;
3746 struct trace_array *tr = info->tr;
3751 mutex_lock(&trace_types_lock);
3753 if (info->cpu == RING_BUFFER_ALL_CPUS) {
3754 int cpu, buf_size_same;
3759 /* check if all cpu sizes are same */
3760 for_each_tracing_cpu(cpu) {
3761 /* fill in the size from first enabled cpu */
3763 size = tr->data[cpu]->entries;
3764 if (size != tr->data[cpu]->entries) {
3770 if (buf_size_same) {
3771 if (!ring_buffer_expanded)
3772 r = sprintf(buf, "%lu (expanded: %lu)\n",
3774 trace_buf_size >> 10);
3776 r = sprintf(buf, "%lu\n", size >> 10);
3778 r = sprintf(buf, "X\n");
3780 r = sprintf(buf, "%lu\n", tr->data[info->cpu]->entries >> 10);
3782 mutex_unlock(&trace_types_lock);
3784 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3789 tracing_entries_write(struct file *filp, const char __user *ubuf,
3790 size_t cnt, loff_t *ppos)
3792 struct ftrace_entries_info *info = filp->private_data;
3796 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3800 /* must have at least 1 entry */
3804 /* value is in KB */
3807 ret = tracing_resize_ring_buffer(val, info->cpu);
3817 tracing_entries_release(struct inode *inode, struct file *filp)
3819 struct ftrace_entries_info *info = filp->private_data;
3827 tracing_total_entries_read(struct file *filp, char __user *ubuf,
3828 size_t cnt, loff_t *ppos)
3830 struct trace_array *tr = filp->private_data;
3833 unsigned long size = 0, expanded_size = 0;
3835 mutex_lock(&trace_types_lock);
3836 for_each_tracing_cpu(cpu) {
3837 size += tr->data[cpu]->entries >> 10;
3838 if (!ring_buffer_expanded)
3839 expanded_size += trace_buf_size >> 10;
3841 if (ring_buffer_expanded)
3842 r = sprintf(buf, "%lu\n", size);
3844 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
3845 mutex_unlock(&trace_types_lock);
3847 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3851 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
3852 size_t cnt, loff_t *ppos)
3855 * There is no need to read what the user has written, this function
3856 * is just to make sure that there is no error when "echo" is used
3865 tracing_free_buffer_release(struct inode *inode, struct file *filp)
3867 /* disable tracing ? */
3868 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
3870 /* resize the ring buffer to 0 */
3871 tracing_resize_ring_buffer(0, RING_BUFFER_ALL_CPUS);
3877 tracing_mark_write(struct file *filp, const char __user *ubuf,
3878 size_t cnt, loff_t *fpos)
3880 unsigned long addr = (unsigned long)ubuf;
3881 struct ring_buffer_event *event;
3882 struct ring_buffer *buffer;
3883 struct print_entry *entry;
3884 unsigned long irq_flags;
3885 struct page *pages[2];
3895 if (tracing_disabled)
3898 if (!(trace_flags & TRACE_ITER_MARKERS))
3901 if (cnt > TRACE_BUF_SIZE)
3902 cnt = TRACE_BUF_SIZE;
3905 * Userspace is injecting traces into the kernel trace buffer.
3906 * We want to be as non intrusive as possible.
3907 * To do so, we do not want to allocate any special buffers
3908 * or take any locks, but instead write the userspace data
3909 * straight into the ring buffer.
3911 * First we need to pin the userspace buffer into memory,
3912 * which, most likely it is, because it just referenced it.
3913 * But there's no guarantee that it is. By using get_user_pages_fast()
3914 * and kmap_atomic/kunmap_atomic() we can get access to the
3915 * pages directly. We then write the data directly into the
3918 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
3920 /* check if we cross pages */
3921 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
3924 offset = addr & (PAGE_SIZE - 1);
3927 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
3928 if (ret < nr_pages) {
3930 put_page(pages[ret]);
3935 for (i = 0; i < nr_pages; i++)
3936 map_page[i] = kmap_atomic(pages[i]);
3938 local_save_flags(irq_flags);
3939 size = sizeof(*entry) + cnt + 2; /* possible \n added */
3940 buffer = global_trace.buffer;
3941 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3942 irq_flags, preempt_count());
3944 /* Ring buffer disabled, return as if not open for write */
3949 entry = ring_buffer_event_data(event);
3950 entry->ip = _THIS_IP_;
3952 if (nr_pages == 2) {
3953 len = PAGE_SIZE - offset;
3954 memcpy(&entry->buf, map_page[0] + offset, len);
3955 memcpy(&entry->buf[len], map_page[1], cnt - len);
3957 memcpy(&entry->buf, map_page[0] + offset, cnt);
3959 if (entry->buf[cnt - 1] != '\n') {
3960 entry->buf[cnt] = '\n';
3961 entry->buf[cnt + 1] = '\0';
3963 entry->buf[cnt] = '\0';
3965 ring_buffer_unlock_commit(buffer, event);
3972 for (i = 0; i < nr_pages; i++){
3973 kunmap_atomic(map_page[i]);
3980 static int tracing_clock_show(struct seq_file *m, void *v)
3984 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3986 "%s%s%s%s", i ? " " : "",
3987 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3988 i == trace_clock_id ? "]" : "");
3994 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3995 size_t cnt, loff_t *fpos)
3998 const char *clockstr;
4001 if (cnt >= sizeof(buf))
4004 if (copy_from_user(&buf, ubuf, cnt))
4009 clockstr = strstrip(buf);
4011 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
4012 if (strcmp(trace_clocks[i].name, clockstr) == 0)
4015 if (i == ARRAY_SIZE(trace_clocks))
4020 mutex_lock(&trace_types_lock);
4022 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
4024 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
4026 mutex_unlock(&trace_types_lock);
4033 static int tracing_clock_open(struct inode *inode, struct file *file)
4035 if (tracing_disabled)
4037 return single_open(file, tracing_clock_show, NULL);
4040 static const struct file_operations tracing_max_lat_fops = {
4041 .open = tracing_open_generic,
4042 .read = tracing_max_lat_read,
4043 .write = tracing_max_lat_write,
4044 .llseek = generic_file_llseek,
4047 static const struct file_operations tracing_ctrl_fops = {
4048 .open = tracing_open_generic,
4049 .read = tracing_ctrl_read,
4050 .write = tracing_ctrl_write,
4051 .llseek = generic_file_llseek,
4054 static const struct file_operations set_tracer_fops = {
4055 .open = tracing_open_generic,
4056 .read = tracing_set_trace_read,
4057 .write = tracing_set_trace_write,
4058 .llseek = generic_file_llseek,
4061 static const struct file_operations tracing_pipe_fops = {
4062 .open = tracing_open_pipe,
4063 .poll = tracing_poll_pipe,
4064 .read = tracing_read_pipe,
4065 .splice_read = tracing_splice_read_pipe,
4066 .release = tracing_release_pipe,
4067 .llseek = no_llseek,
4070 static const struct file_operations tracing_entries_fops = {
4071 .open = tracing_entries_open,
4072 .read = tracing_entries_read,
4073 .write = tracing_entries_write,
4074 .release = tracing_entries_release,
4075 .llseek = generic_file_llseek,
4078 static const struct file_operations tracing_total_entries_fops = {
4079 .open = tracing_open_generic,
4080 .read = tracing_total_entries_read,
4081 .llseek = generic_file_llseek,
4084 static const struct file_operations tracing_free_buffer_fops = {
4085 .write = tracing_free_buffer_write,
4086 .release = tracing_free_buffer_release,
4089 static const struct file_operations tracing_mark_fops = {
4090 .open = tracing_open_generic,
4091 .write = tracing_mark_write,
4092 .llseek = generic_file_llseek,
4095 static const struct file_operations trace_clock_fops = {
4096 .open = tracing_clock_open,
4098 .llseek = seq_lseek,
4099 .release = single_release,
4100 .write = tracing_clock_write,
4103 struct ftrace_buffer_info {
4104 struct trace_array *tr;
4110 static int tracing_buffers_open(struct inode *inode, struct file *filp)
4112 int cpu = (int)(long)inode->i_private;
4113 struct ftrace_buffer_info *info;
4115 if (tracing_disabled)
4118 info = kzalloc(sizeof(*info), GFP_KERNEL);
4122 info->tr = &global_trace;
4125 /* Force reading ring buffer for first read */
4126 info->read = (unsigned int)-1;
4128 filp->private_data = info;
4130 return nonseekable_open(inode, filp);
4134 tracing_buffers_read(struct file *filp, char __user *ubuf,
4135 size_t count, loff_t *ppos)
4137 struct ftrace_buffer_info *info = filp->private_data;
4145 info->spare = ring_buffer_alloc_read_page(info->tr->buffer, info->cpu);
4149 /* Do we have previous read data to read? */
4150 if (info->read < PAGE_SIZE)
4153 trace_access_lock(info->cpu);
4154 ret = ring_buffer_read_page(info->tr->buffer,
4158 trace_access_unlock(info->cpu);
4165 size = PAGE_SIZE - info->read;
4169 ret = copy_to_user(ubuf, info->spare + info->read, size);
4180 static int tracing_buffers_release(struct inode *inode, struct file *file)
4182 struct ftrace_buffer_info *info = file->private_data;
4185 ring_buffer_free_read_page(info->tr->buffer, info->spare);
4192 struct ring_buffer *buffer;
4197 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
4198 struct pipe_buffer *buf)
4200 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4205 ring_buffer_free_read_page(ref->buffer, ref->page);
4210 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
4211 struct pipe_buffer *buf)
4213 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4218 /* Pipe buffer operations for a buffer. */
4219 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
4221 .map = generic_pipe_buf_map,
4222 .unmap = generic_pipe_buf_unmap,
4223 .confirm = generic_pipe_buf_confirm,
4224 .release = buffer_pipe_buf_release,
4225 .steal = generic_pipe_buf_steal,
4226 .get = buffer_pipe_buf_get,
4230 * Callback from splice_to_pipe(), if we need to release some pages
4231 * at the end of the spd in case we error'ed out in filling the pipe.
4233 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
4235 struct buffer_ref *ref =
4236 (struct buffer_ref *)spd->partial[i].private;
4241 ring_buffer_free_read_page(ref->buffer, ref->page);
4243 spd->partial[i].private = 0;
4247 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4248 struct pipe_inode_info *pipe, size_t len,
4251 struct ftrace_buffer_info *info = file->private_data;
4252 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4253 struct page *pages_def[PIPE_DEF_BUFFERS];
4254 struct splice_pipe_desc spd = {
4256 .partial = partial_def,
4257 .nr_pages_max = PIPE_DEF_BUFFERS,
4259 .ops = &buffer_pipe_buf_ops,
4260 .spd_release = buffer_spd_release,
4262 struct buffer_ref *ref;
4263 int entries, size, i;
4266 if (splice_grow_spd(pipe, &spd))
4269 if (*ppos & (PAGE_SIZE - 1)) {
4270 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
4275 if (len & (PAGE_SIZE - 1)) {
4276 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
4277 if (len < PAGE_SIZE) {
4284 trace_access_lock(info->cpu);
4285 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
4287 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
4291 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
4296 ref->buffer = info->tr->buffer;
4297 ref->page = ring_buffer_alloc_read_page(ref->buffer, info->cpu);
4303 r = ring_buffer_read_page(ref->buffer, &ref->page,
4306 ring_buffer_free_read_page(ref->buffer, ref->page);
4312 * zero out any left over data, this is going to
4315 size = ring_buffer_page_len(ref->page);
4316 if (size < PAGE_SIZE)
4317 memset(ref->page + size, 0, PAGE_SIZE - size);
4319 page = virt_to_page(ref->page);
4321 spd.pages[i] = page;
4322 spd.partial[i].len = PAGE_SIZE;
4323 spd.partial[i].offset = 0;
4324 spd.partial[i].private = (unsigned long)ref;
4328 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
4331 trace_access_unlock(info->cpu);
4334 /* did we read anything? */
4335 if (!spd.nr_pages) {
4336 if (flags & SPLICE_F_NONBLOCK)
4344 ret = splice_to_pipe(pipe, &spd);
4345 splice_shrink_spd(&spd);
4350 static const struct file_operations tracing_buffers_fops = {
4351 .open = tracing_buffers_open,
4352 .read = tracing_buffers_read,
4353 .release = tracing_buffers_release,
4354 .splice_read = tracing_buffers_splice_read,
4355 .llseek = no_llseek,
4359 tracing_stats_read(struct file *filp, char __user *ubuf,
4360 size_t count, loff_t *ppos)
4362 unsigned long cpu = (unsigned long)filp->private_data;
4363 struct trace_array *tr = &global_trace;
4364 struct trace_seq *s;
4366 unsigned long long t;
4367 unsigned long usec_rem;
4369 s = kmalloc(sizeof(*s), GFP_KERNEL);
4375 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
4376 trace_seq_printf(s, "entries: %ld\n", cnt);
4378 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
4379 trace_seq_printf(s, "overrun: %ld\n", cnt);
4381 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
4382 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
4384 cnt = ring_buffer_bytes_cpu(tr->buffer, cpu);
4385 trace_seq_printf(s, "bytes: %ld\n", cnt);
4387 t = ns2usecs(ring_buffer_oldest_event_ts(tr->buffer, cpu));
4388 usec_rem = do_div(t, USEC_PER_SEC);
4389 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n", t, usec_rem);
4391 t = ns2usecs(ring_buffer_time_stamp(tr->buffer, cpu));
4392 usec_rem = do_div(t, USEC_PER_SEC);
4393 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
4395 cnt = ring_buffer_dropped_events_cpu(tr->buffer, cpu);
4396 trace_seq_printf(s, "dropped events: %ld\n", cnt);
4398 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
4405 static const struct file_operations tracing_stats_fops = {
4406 .open = tracing_open_generic,
4407 .read = tracing_stats_read,
4408 .llseek = generic_file_llseek,
4411 #ifdef CONFIG_DYNAMIC_FTRACE
4413 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
4419 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
4420 size_t cnt, loff_t *ppos)
4422 static char ftrace_dyn_info_buffer[1024];
4423 static DEFINE_MUTEX(dyn_info_mutex);
4424 unsigned long *p = filp->private_data;
4425 char *buf = ftrace_dyn_info_buffer;
4426 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
4429 mutex_lock(&dyn_info_mutex);
4430 r = sprintf(buf, "%ld ", *p);
4432 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
4435 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4437 mutex_unlock(&dyn_info_mutex);
4442 static const struct file_operations tracing_dyn_info_fops = {
4443 .open = tracing_open_generic,
4444 .read = tracing_read_dyn_info,
4445 .llseek = generic_file_llseek,
4449 static struct dentry *d_tracer;
4451 struct dentry *tracing_init_dentry(void)
4458 if (!debugfs_initialized())
4461 d_tracer = debugfs_create_dir("tracing", NULL);
4463 if (!d_tracer && !once) {
4465 pr_warning("Could not create debugfs directory 'tracing'\n");
4472 static struct dentry *d_percpu;
4474 struct dentry *tracing_dentry_percpu(void)
4477 struct dentry *d_tracer;
4482 d_tracer = tracing_init_dentry();
4487 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
4489 if (!d_percpu && !once) {
4491 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4498 static void tracing_init_debugfs_percpu(long cpu)
4500 struct dentry *d_percpu = tracing_dentry_percpu();
4501 struct dentry *d_cpu;
4502 char cpu_dir[30]; /* 30 characters should be more than enough */
4507 snprintf(cpu_dir, 30, "cpu%ld", cpu);
4508 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4510 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4514 /* per cpu trace_pipe */
4515 trace_create_file("trace_pipe", 0444, d_cpu,
4516 (void *) cpu, &tracing_pipe_fops);
4519 trace_create_file("trace", 0644, d_cpu,
4520 (void *) cpu, &tracing_fops);
4522 trace_create_file("trace_pipe_raw", 0444, d_cpu,
4523 (void *) cpu, &tracing_buffers_fops);
4525 trace_create_file("stats", 0444, d_cpu,
4526 (void *) cpu, &tracing_stats_fops);
4528 trace_create_file("buffer_size_kb", 0444, d_cpu,
4529 (void *) cpu, &tracing_entries_fops);
4532 #ifdef CONFIG_FTRACE_SELFTEST
4533 /* Let selftest have access to static functions in this file */
4534 #include "trace_selftest.c"
4537 struct trace_option_dentry {
4538 struct tracer_opt *opt;
4539 struct tracer_flags *flags;
4540 struct dentry *entry;
4544 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4547 struct trace_option_dentry *topt = filp->private_data;
4550 if (topt->flags->val & topt->opt->bit)
4555 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4559 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4562 struct trace_option_dentry *topt = filp->private_data;
4566 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4570 if (val != 0 && val != 1)
4573 if (!!(topt->flags->val & topt->opt->bit) != val) {
4574 mutex_lock(&trace_types_lock);
4575 ret = __set_tracer_option(current_trace, topt->flags,
4577 mutex_unlock(&trace_types_lock);
4588 static const struct file_operations trace_options_fops = {
4589 .open = tracing_open_generic,
4590 .read = trace_options_read,
4591 .write = trace_options_write,
4592 .llseek = generic_file_llseek,
4596 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4599 long index = (long)filp->private_data;
4602 if (trace_flags & (1 << index))
4607 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4611 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4614 long index = (long)filp->private_data;
4618 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4622 if (val != 0 && val != 1)
4624 set_tracer_flags(1 << index, val);
4631 static const struct file_operations trace_options_core_fops = {
4632 .open = tracing_open_generic,
4633 .read = trace_options_core_read,
4634 .write = trace_options_core_write,
4635 .llseek = generic_file_llseek,
4638 struct dentry *trace_create_file(const char *name,
4640 struct dentry *parent,
4642 const struct file_operations *fops)
4646 ret = debugfs_create_file(name, mode, parent, data, fops);
4648 pr_warning("Could not create debugfs '%s' entry\n", name);
4654 static struct dentry *trace_options_init_dentry(void)
4656 struct dentry *d_tracer;
4657 static struct dentry *t_options;
4662 d_tracer = tracing_init_dentry();
4666 t_options = debugfs_create_dir("options", d_tracer);
4668 pr_warning("Could not create debugfs directory 'options'\n");
4676 create_trace_option_file(struct trace_option_dentry *topt,
4677 struct tracer_flags *flags,
4678 struct tracer_opt *opt)
4680 struct dentry *t_options;
4682 t_options = trace_options_init_dentry();
4686 topt->flags = flags;
4689 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4690 &trace_options_fops);
4694 static struct trace_option_dentry *
4695 create_trace_option_files(struct tracer *tracer)
4697 struct trace_option_dentry *topts;
4698 struct tracer_flags *flags;
4699 struct tracer_opt *opts;
4705 flags = tracer->flags;
4707 if (!flags || !flags->opts)
4712 for (cnt = 0; opts[cnt].name; cnt++)
4715 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4719 for (cnt = 0; opts[cnt].name; cnt++)
4720 create_trace_option_file(&topts[cnt], flags,
4727 destroy_trace_option_files(struct trace_option_dentry *topts)
4734 for (cnt = 0; topts[cnt].opt; cnt++) {
4735 if (topts[cnt].entry)
4736 debugfs_remove(topts[cnt].entry);
4742 static struct dentry *
4743 create_trace_option_core_file(const char *option, long index)
4745 struct dentry *t_options;
4747 t_options = trace_options_init_dentry();
4751 return trace_create_file(option, 0644, t_options, (void *)index,
4752 &trace_options_core_fops);
4755 static __init void create_trace_options_dir(void)
4757 struct dentry *t_options;
4760 t_options = trace_options_init_dentry();
4764 for (i = 0; trace_options[i]; i++)
4765 create_trace_option_core_file(trace_options[i], i);
4769 rb_simple_read(struct file *filp, char __user *ubuf,
4770 size_t cnt, loff_t *ppos)
4772 struct trace_array *tr = filp->private_data;
4773 struct ring_buffer *buffer = tr->buffer;
4778 r = ring_buffer_record_is_on(buffer);
4782 r = sprintf(buf, "%d\n", r);
4784 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4788 rb_simple_write(struct file *filp, const char __user *ubuf,
4789 size_t cnt, loff_t *ppos)
4791 struct trace_array *tr = filp->private_data;
4792 struct ring_buffer *buffer = tr->buffer;
4796 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4802 ring_buffer_record_on(buffer);
4804 ring_buffer_record_off(buffer);
4812 static const struct file_operations rb_simple_fops = {
4813 .open = tracing_open_generic,
4814 .read = rb_simple_read,
4815 .write = rb_simple_write,
4816 .llseek = default_llseek,
4819 static __init int tracer_init_debugfs(void)
4821 struct dentry *d_tracer;
4824 trace_access_lock_init();
4826 d_tracer = tracing_init_dentry();
4828 trace_create_file("tracing_enabled", 0644, d_tracer,
4829 &global_trace, &tracing_ctrl_fops);
4831 trace_create_file("trace_options", 0644, d_tracer,
4832 NULL, &tracing_iter_fops);
4834 trace_create_file("tracing_cpumask", 0644, d_tracer,
4835 NULL, &tracing_cpumask_fops);
4837 trace_create_file("trace", 0644, d_tracer,
4838 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4840 trace_create_file("available_tracers", 0444, d_tracer,
4841 &global_trace, &show_traces_fops);
4843 trace_create_file("current_tracer", 0644, d_tracer,
4844 &global_trace, &set_tracer_fops);
4846 #ifdef CONFIG_TRACER_MAX_TRACE
4847 trace_create_file("tracing_max_latency", 0644, d_tracer,
4848 &tracing_max_latency, &tracing_max_lat_fops);
4851 trace_create_file("tracing_thresh", 0644, d_tracer,
4852 &tracing_thresh, &tracing_max_lat_fops);
4854 trace_create_file("README", 0444, d_tracer,
4855 NULL, &tracing_readme_fops);
4857 trace_create_file("trace_pipe", 0444, d_tracer,
4858 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4860 trace_create_file("buffer_size_kb", 0644, d_tracer,
4861 (void *) RING_BUFFER_ALL_CPUS, &tracing_entries_fops);
4863 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
4864 &global_trace, &tracing_total_entries_fops);
4866 trace_create_file("free_buffer", 0644, d_tracer,
4867 &global_trace, &tracing_free_buffer_fops);
4869 trace_create_file("trace_marker", 0220, d_tracer,
4870 NULL, &tracing_mark_fops);
4872 trace_create_file("saved_cmdlines", 0444, d_tracer,
4873 NULL, &tracing_saved_cmdlines_fops);
4875 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4878 trace_create_file("tracing_on", 0644, d_tracer,
4879 &global_trace, &rb_simple_fops);
4881 #ifdef CONFIG_DYNAMIC_FTRACE
4882 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4883 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4886 create_trace_options_dir();
4888 for_each_tracing_cpu(cpu)
4889 tracing_init_debugfs_percpu(cpu);
4894 static int trace_panic_handler(struct notifier_block *this,
4895 unsigned long event, void *unused)
4897 if (ftrace_dump_on_oops)
4898 ftrace_dump(ftrace_dump_on_oops);
4902 static struct notifier_block trace_panic_notifier = {
4903 .notifier_call = trace_panic_handler,
4905 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4908 static int trace_die_handler(struct notifier_block *self,
4914 if (ftrace_dump_on_oops)
4915 ftrace_dump(ftrace_dump_on_oops);
4923 static struct notifier_block trace_die_notifier = {
4924 .notifier_call = trace_die_handler,
4929 * printk is set to max of 1024, we really don't need it that big.
4930 * Nothing should be printing 1000 characters anyway.
4932 #define TRACE_MAX_PRINT 1000
4935 * Define here KERN_TRACE so that we have one place to modify
4936 * it if we decide to change what log level the ftrace dump
4939 #define KERN_TRACE KERN_EMERG
4942 trace_printk_seq(struct trace_seq *s)
4944 /* Probably should print a warning here. */
4948 /* should be zero ended, but we are paranoid. */
4949 s->buffer[s->len] = 0;
4951 printk(KERN_TRACE "%s", s->buffer);
4956 void trace_init_global_iter(struct trace_iterator *iter)
4958 iter->tr = &global_trace;
4959 iter->trace = current_trace;
4960 iter->cpu_file = TRACE_PIPE_ALL_CPU;
4964 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4966 static arch_spinlock_t ftrace_dump_lock =
4967 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4968 /* use static because iter can be a bit big for the stack */
4969 static struct trace_iterator iter;
4970 unsigned int old_userobj;
4971 static int dump_ran;
4972 unsigned long flags;
4976 local_irq_save(flags);
4977 arch_spin_lock(&ftrace_dump_lock);
4985 /* Did function tracer already get disabled? */
4986 if (ftrace_is_dead()) {
4987 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
4988 printk("# MAY BE MISSING FUNCTION EVENTS\n");
4991 if (disable_tracing)
4994 trace_init_global_iter(&iter);
4996 for_each_tracing_cpu(cpu) {
4997 atomic_inc(&iter.tr->data[cpu]->disabled);
5000 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
5002 /* don't look at user memory in panic mode */
5003 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
5005 /* Simulate the iterator */
5006 iter.tr = &global_trace;
5007 iter.trace = current_trace;
5009 switch (oops_dump_mode) {
5011 iter.cpu_file = TRACE_PIPE_ALL_CPU;
5014 iter.cpu_file = raw_smp_processor_id();
5019 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
5020 iter.cpu_file = TRACE_PIPE_ALL_CPU;
5023 printk(KERN_TRACE "Dumping ftrace buffer:\n");
5026 * We need to stop all tracing on all CPUS to read the
5027 * the next buffer. This is a bit expensive, but is
5028 * not done often. We fill all what we can read,
5029 * and then release the locks again.
5032 while (!trace_empty(&iter)) {
5035 printk(KERN_TRACE "---------------------------------\n");
5039 /* reset all but tr, trace, and overruns */
5040 memset(&iter.seq, 0,
5041 sizeof(struct trace_iterator) -
5042 offsetof(struct trace_iterator, seq));
5043 iter.iter_flags |= TRACE_FILE_LAT_FMT;
5046 if (trace_find_next_entry_inc(&iter) != NULL) {
5049 ret = print_trace_line(&iter);
5050 if (ret != TRACE_TYPE_NO_CONSUME)
5051 trace_consume(&iter);
5053 touch_nmi_watchdog();
5055 trace_printk_seq(&iter.seq);
5059 printk(KERN_TRACE " (ftrace buffer empty)\n");
5061 printk(KERN_TRACE "---------------------------------\n");
5064 /* Re-enable tracing if requested */
5065 if (!disable_tracing) {
5066 trace_flags |= old_userobj;
5068 for_each_tracing_cpu(cpu) {
5069 atomic_dec(&iter.tr->data[cpu]->disabled);
5075 arch_spin_unlock(&ftrace_dump_lock);
5076 local_irq_restore(flags);
5079 /* By default: disable tracing after the dump */
5080 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
5082 __ftrace_dump(true, oops_dump_mode);
5084 EXPORT_SYMBOL_GPL(ftrace_dump);
5086 __init static int tracer_alloc_buffers(void)
5089 enum ring_buffer_flags rb_flags;
5094 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
5097 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
5098 goto out_free_buffer_mask;
5100 /* Only allocate trace_printk buffers if a trace_printk exists */
5101 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
5102 trace_printk_init_buffers();
5104 /* To save memory, keep the ring buffer size to its minimum */
5105 if (ring_buffer_expanded)
5106 ring_buf_size = trace_buf_size;
5110 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
5112 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
5113 cpumask_copy(tracing_cpumask, cpu_all_mask);
5115 /* TODO: make the number of buffers hot pluggable with CPUS */
5116 global_trace.buffer = ring_buffer_alloc(ring_buf_size, rb_flags);
5117 if (!global_trace.buffer) {
5118 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
5120 goto out_free_cpumask;
5122 if (global_trace.buffer_disabled)
5126 #ifdef CONFIG_TRACER_MAX_TRACE
5127 max_tr.buffer = ring_buffer_alloc(1, rb_flags);
5128 if (!max_tr.buffer) {
5129 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
5131 ring_buffer_free(global_trace.buffer);
5132 goto out_free_cpumask;
5136 /* Allocate the first page for all buffers */
5137 for_each_tracing_cpu(i) {
5138 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
5139 max_tr.data[i] = &per_cpu(max_tr_data, i);
5142 set_buffer_entries(&global_trace,
5143 ring_buffer_size(global_trace.buffer, 0));
5144 #ifdef CONFIG_TRACER_MAX_TRACE
5145 set_buffer_entries(&max_tr, 1);
5148 trace_init_cmdlines();
5150 register_tracer(&nop_trace);
5151 current_trace = &nop_trace;
5152 /* All seems OK, enable tracing */
5153 tracing_disabled = 0;
5155 atomic_notifier_chain_register(&panic_notifier_list,
5156 &trace_panic_notifier);
5158 register_die_notifier(&trace_die_notifier);
5163 free_cpumask_var(tracing_cpumask);
5164 out_free_buffer_mask:
5165 free_cpumask_var(tracing_buffer_mask);
5170 __init static int clear_boot_tracer(void)
5173 * The default tracer at boot buffer is an init section.
5174 * This function is called in lateinit. If we did not
5175 * find the boot tracer, then clear it out, to prevent
5176 * later registration from accessing the buffer that is
5177 * about to be freed.
5179 if (!default_bootup_tracer)
5182 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
5183 default_bootup_tracer);
5184 default_bootup_tracer = NULL;
5189 early_initcall(tracer_alloc_buffers);
5190 fs_initcall(tracer_init_debugfs);
5191 late_initcall(clear_boot_tracer);