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/smp_lock.h>
21 #include <linux/notifier.h>
22 #include <linux/irqflags.h>
23 #include <linux/debugfs.h>
24 #include <linux/pagemap.h>
25 #include <linux/hardirq.h>
26 #include <linux/linkage.h>
27 #include <linux/uaccess.h>
28 #include <linux/kprobes.h>
29 #include <linux/ftrace.h>
30 #include <linux/module.h>
31 #include <linux/percpu.h>
32 #include <linux/splice.h>
33 #include <linux/kdebug.h>
34 #include <linux/string.h>
35 #include <linux/rwsem.h>
36 #include <linux/slab.h>
37 #include <linux/ctype.h>
38 #include <linux/init.h>
39 #include <linux/poll.h>
43 #include "trace_output.h"
45 #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
48 * On boot up, the ring buffer is set to the minimum size, so that
49 * we do not waste memory on systems that are not using tracing.
51 int ring_buffer_expanded;
54 * We need to change this state when a selftest is running.
55 * A selftest will lurk into the ring-buffer to count the
56 * entries inserted during the selftest although some concurrent
57 * insertions into the ring-buffer such as trace_printk could occurred
58 * at the same time, giving false positive or negative results.
60 static bool __read_mostly tracing_selftest_running;
63 * If a tracer is running, we do not want to run SELFTEST.
65 bool __read_mostly tracing_selftest_disabled;
67 /* For tracers that don't implement custom flags */
68 static struct tracer_opt dummy_tracer_opt[] = {
72 static struct tracer_flags dummy_tracer_flags = {
74 .opts = dummy_tracer_opt
77 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
83 * Kill all tracing for good (never come back).
84 * It is initialized to 1 but will turn to zero if the initialization
85 * of the tracer is successful. But that is the only place that sets
88 static int tracing_disabled = 1;
90 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
92 static inline void ftrace_disable_cpu(void)
95 __this_cpu_inc(ftrace_cpu_disabled);
98 static inline void ftrace_enable_cpu(void)
100 __this_cpu_dec(ftrace_cpu_disabled);
104 cpumask_var_t __read_mostly tracing_buffer_mask;
107 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
109 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
110 * is set, then ftrace_dump is called. This will output the contents
111 * of the ftrace buffers to the console. This is very useful for
112 * capturing traces that lead to crashes and outputing it to a
115 * It is default off, but you can enable it with either specifying
116 * "ftrace_dump_on_oops" in the kernel command line, or setting
117 * /proc/sys/kernel/ftrace_dump_on_oops
118 * Set 1 if you want to dump buffers of all CPUs
119 * Set 2 if you want to dump the buffer of the CPU that triggered oops
122 enum ftrace_dump_mode ftrace_dump_on_oops;
124 static int tracing_set_tracer(const char *buf);
126 #define MAX_TRACER_SIZE 100
127 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
128 static char *default_bootup_tracer;
130 static int __init set_cmdline_ftrace(char *str)
132 strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
133 default_bootup_tracer = bootup_tracer_buf;
134 /* We are using ftrace early, expand it */
135 ring_buffer_expanded = 1;
138 __setup("ftrace=", set_cmdline_ftrace);
140 static int __init set_ftrace_dump_on_oops(char *str)
142 if (*str++ != '=' || !*str) {
143 ftrace_dump_on_oops = DUMP_ALL;
147 if (!strcmp("orig_cpu", str)) {
148 ftrace_dump_on_oops = DUMP_ORIG;
154 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
156 unsigned long long ns2usecs(cycle_t nsec)
164 * The global_trace is the descriptor that holds the tracing
165 * buffers for the live tracing. For each CPU, it contains
166 * a link list of pages that will store trace entries. The
167 * page descriptor of the pages in the memory is used to hold
168 * the link list by linking the lru item in the page descriptor
169 * to each of the pages in the buffer per CPU.
171 * For each active CPU there is a data field that holds the
172 * pages for the buffer for that CPU. Each CPU has the same number
173 * of pages allocated for its buffer.
175 static struct trace_array global_trace;
177 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
179 int filter_current_check_discard(struct ring_buffer *buffer,
180 struct ftrace_event_call *call, void *rec,
181 struct ring_buffer_event *event)
183 return filter_check_discard(call, rec, buffer, event);
185 EXPORT_SYMBOL_GPL(filter_current_check_discard);
187 cycle_t ftrace_now(int cpu)
191 /* Early boot up does not have a buffer yet */
192 if (!global_trace.buffer)
193 return trace_clock_local();
195 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
196 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
202 * The max_tr is used to snapshot the global_trace when a maximum
203 * latency is reached. Some tracers will use this to store a maximum
204 * trace while it continues examining live traces.
206 * The buffers for the max_tr are set up the same as the global_trace.
207 * When a snapshot is taken, the link list of the max_tr is swapped
208 * with the link list of the global_trace and the buffers are reset for
209 * the global_trace so the tracing can continue.
211 static struct trace_array max_tr;
213 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
215 /* tracer_enabled is used to toggle activation of a tracer */
216 static int tracer_enabled = 1;
219 * tracing_is_enabled - return tracer_enabled status
221 * This function is used by other tracers to know the status
222 * of the tracer_enabled flag. Tracers may use this function
223 * to know if it should enable their features when starting
224 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
226 int tracing_is_enabled(void)
228 return tracer_enabled;
232 * trace_buf_size is the size in bytes that is allocated
233 * for a buffer. Note, the number of bytes is always rounded
236 * This number is purposely set to a low number of 16384.
237 * If the dump on oops happens, it will be much appreciated
238 * to not have to wait for all that output. Anyway this can be
239 * boot time and run time configurable.
241 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
243 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
245 /* trace_types holds a link list of available tracers. */
246 static struct tracer *trace_types __read_mostly;
248 /* current_trace points to the tracer that is currently active */
249 static struct tracer *current_trace __read_mostly;
252 * trace_types_lock is used to protect the trace_types list.
254 static DEFINE_MUTEX(trace_types_lock);
257 * serialize the access of the ring buffer
259 * ring buffer serializes readers, but it is low level protection.
260 * The validity of the events (which returns by ring_buffer_peek() ..etc)
261 * are not protected by ring buffer.
263 * The content of events may become garbage if we allow other process consumes
264 * these events concurrently:
265 * A) the page of the consumed events may become a normal page
266 * (not reader page) in ring buffer, and this page will be rewrited
267 * by events producer.
268 * B) The page of the consumed events may become a page for splice_read,
269 * and this page will be returned to system.
271 * These primitives allow multi process access to different cpu ring buffer
274 * These primitives don't distinguish read-only and read-consume access.
275 * Multi read-only access are also serialized.
279 static DECLARE_RWSEM(all_cpu_access_lock);
280 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
282 static inline void trace_access_lock(int cpu)
284 if (cpu == TRACE_PIPE_ALL_CPU) {
285 /* gain it for accessing the whole ring buffer. */
286 down_write(&all_cpu_access_lock);
288 /* gain it for accessing a cpu ring buffer. */
290 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
291 down_read(&all_cpu_access_lock);
293 /* Secondly block other access to this @cpu ring buffer. */
294 mutex_lock(&per_cpu(cpu_access_lock, cpu));
298 static inline void trace_access_unlock(int cpu)
300 if (cpu == TRACE_PIPE_ALL_CPU) {
301 up_write(&all_cpu_access_lock);
303 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
304 up_read(&all_cpu_access_lock);
308 static inline void trace_access_lock_init(void)
312 for_each_possible_cpu(cpu)
313 mutex_init(&per_cpu(cpu_access_lock, cpu));
318 static DEFINE_MUTEX(access_lock);
320 static inline void trace_access_lock(int cpu)
323 mutex_lock(&access_lock);
326 static inline void trace_access_unlock(int cpu)
329 mutex_unlock(&access_lock);
332 static inline void trace_access_lock_init(void)
338 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
339 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
341 /* trace_flags holds trace_options default values */
342 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
343 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
344 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD;
346 static int trace_stop_count;
347 static DEFINE_SPINLOCK(tracing_start_lock);
350 * trace_wake_up - wake up tasks waiting for trace input
352 * Simply wakes up any task that is blocked on the trace_wait
353 * queue. These is used with trace_poll for tasks polling the trace.
355 void trace_wake_up(void)
359 if (trace_flags & TRACE_ITER_BLOCK)
362 * The runqueue_is_locked() can fail, but this is the best we
366 if (!runqueue_is_locked(cpu))
367 wake_up(&trace_wait);
371 static int __init set_buf_size(char *str)
373 unsigned long buf_size;
377 buf_size = memparse(str, &str);
378 /* nr_entries can not be zero */
381 trace_buf_size = buf_size;
384 __setup("trace_buf_size=", set_buf_size);
386 static int __init set_tracing_thresh(char *str)
388 unsigned long threshhold;
393 ret = strict_strtoul(str, 0, &threshhold);
396 tracing_thresh = threshhold * 1000;
399 __setup("tracing_thresh=", set_tracing_thresh);
401 unsigned long nsecs_to_usecs(unsigned long nsecs)
406 /* These must match the bit postions in trace_iterator_flags */
407 static const char *trace_options[] = {
436 { trace_clock_local, "local" },
437 { trace_clock_global, "global" },
443 * trace_parser_get_init - gets the buffer for trace parser
445 int trace_parser_get_init(struct trace_parser *parser, int size)
447 memset(parser, 0, sizeof(*parser));
449 parser->buffer = kmalloc(size, GFP_KERNEL);
458 * trace_parser_put - frees the buffer for trace parser
460 void trace_parser_put(struct trace_parser *parser)
462 kfree(parser->buffer);
466 * trace_get_user - reads the user input string separated by space
467 * (matched by isspace(ch))
469 * For each string found the 'struct trace_parser' is updated,
470 * and the function returns.
472 * Returns number of bytes read.
474 * See kernel/trace/trace.h for 'struct trace_parser' details.
476 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
477 size_t cnt, loff_t *ppos)
484 trace_parser_clear(parser);
486 ret = get_user(ch, ubuf++);
494 * The parser is not finished with the last write,
495 * continue reading the user input without skipping spaces.
498 /* skip white space */
499 while (cnt && isspace(ch)) {
500 ret = get_user(ch, ubuf++);
507 /* only spaces were written */
517 /* read the non-space input */
518 while (cnt && !isspace(ch)) {
519 if (parser->idx < parser->size - 1)
520 parser->buffer[parser->idx++] = ch;
525 ret = get_user(ch, ubuf++);
532 /* We either got finished input or we have to wait for another call. */
534 parser->buffer[parser->idx] = 0;
535 parser->cont = false;
538 parser->buffer[parser->idx++] = ch;
548 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
556 if (s->len <= s->readpos)
559 len = s->len - s->readpos;
562 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
572 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
577 if (s->len <= s->readpos)
580 len = s->len - s->readpos;
583 ret = memcpy(buf, s->buffer + s->readpos, cnt);
592 * ftrace_max_lock is used to protect the swapping of buffers
593 * when taking a max snapshot. The buffers themselves are
594 * protected by per_cpu spinlocks. But the action of the swap
595 * needs its own lock.
597 * This is defined as a arch_spinlock_t in order to help
598 * with performance when lockdep debugging is enabled.
600 * It is also used in other places outside the update_max_tr
601 * so it needs to be defined outside of the
602 * CONFIG_TRACER_MAX_TRACE.
604 static arch_spinlock_t ftrace_max_lock =
605 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
607 unsigned long __read_mostly tracing_thresh;
609 #ifdef CONFIG_TRACER_MAX_TRACE
610 unsigned long __read_mostly tracing_max_latency;
613 * Copy the new maximum trace into the separate maximum-trace
614 * structure. (this way the maximum trace is permanently saved,
615 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
618 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
620 struct trace_array_cpu *data = tr->data[cpu];
621 struct trace_array_cpu *max_data;
624 max_tr.time_start = data->preempt_timestamp;
626 max_data = max_tr.data[cpu];
627 max_data->saved_latency = tracing_max_latency;
628 max_data->critical_start = data->critical_start;
629 max_data->critical_end = data->critical_end;
631 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
632 max_data->pid = tsk->pid;
633 max_data->uid = task_uid(tsk);
634 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
635 max_data->policy = tsk->policy;
636 max_data->rt_priority = tsk->rt_priority;
638 /* record this tasks comm */
639 tracing_record_cmdline(tsk);
643 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
645 * @tsk: the task with the latency
646 * @cpu: The cpu that initiated the trace.
648 * Flip the buffers between the @tr and the max_tr and record information
649 * about which task was the cause of this latency.
652 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
654 struct ring_buffer *buf = tr->buffer;
656 if (trace_stop_count)
659 WARN_ON_ONCE(!irqs_disabled());
660 if (!current_trace->use_max_tr) {
664 arch_spin_lock(&ftrace_max_lock);
666 tr->buffer = max_tr.buffer;
669 __update_max_tr(tr, tsk, cpu);
670 arch_spin_unlock(&ftrace_max_lock);
674 * update_max_tr_single - only copy one trace over, and reset the rest
676 * @tsk - task with the latency
677 * @cpu - the cpu of the buffer to copy.
679 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
682 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
686 if (trace_stop_count)
689 WARN_ON_ONCE(!irqs_disabled());
690 if (!current_trace->use_max_tr) {
695 arch_spin_lock(&ftrace_max_lock);
697 ftrace_disable_cpu();
699 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
703 * We failed to swap the buffer due to a commit taking
704 * place on this CPU. We fail to record, but we reset
705 * the max trace buffer (no one writes directly to it)
706 * and flag that it failed.
708 trace_array_printk(&max_tr, _THIS_IP_,
709 "Failed to swap buffers due to commit in progress\n");
714 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
716 __update_max_tr(tr, tsk, cpu);
717 arch_spin_unlock(&ftrace_max_lock);
719 #endif /* CONFIG_TRACER_MAX_TRACE */
722 * register_tracer - register a tracer with the ftrace system.
723 * @type - the plugin for the tracer
725 * Register a new plugin tracer.
727 int register_tracer(struct tracer *type)
728 __releases(kernel_lock)
729 __acquires(kernel_lock)
735 pr_info("Tracer must have a name\n");
739 if (strlen(type->name) >= MAX_TRACER_SIZE) {
740 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
745 * When this gets called we hold the BKL which means that
746 * preemption is disabled. Various trace selftests however
747 * need to disable and enable preemption for successful tests.
748 * So we drop the BKL here and grab it after the tests again.
751 mutex_lock(&trace_types_lock);
753 tracing_selftest_running = true;
755 for (t = trace_types; t; t = t->next) {
756 if (strcmp(type->name, t->name) == 0) {
758 pr_info("Tracer %s already registered\n",
766 type->set_flag = &dummy_set_flag;
768 type->flags = &dummy_tracer_flags;
770 if (!type->flags->opts)
771 type->flags->opts = dummy_tracer_opt;
772 if (!type->wait_pipe)
773 type->wait_pipe = default_wait_pipe;
776 #ifdef CONFIG_FTRACE_STARTUP_TEST
777 if (type->selftest && !tracing_selftest_disabled) {
778 struct tracer *saved_tracer = current_trace;
779 struct trace_array *tr = &global_trace;
782 * Run a selftest on this tracer.
783 * Here we reset the trace buffer, and set the current
784 * tracer to be this tracer. The tracer can then run some
785 * internal tracing to verify that everything is in order.
786 * If we fail, we do not register this tracer.
788 tracing_reset_online_cpus(tr);
790 current_trace = type;
791 /* the test is responsible for initializing and enabling */
792 pr_info("Testing tracer %s: ", type->name);
793 ret = type->selftest(type, tr);
794 /* the test is responsible for resetting too */
795 current_trace = saved_tracer;
797 printk(KERN_CONT "FAILED!\n");
800 /* Only reset on passing, to avoid touching corrupted buffers */
801 tracing_reset_online_cpus(tr);
803 printk(KERN_CONT "PASSED\n");
807 type->next = trace_types;
811 tracing_selftest_running = false;
812 mutex_unlock(&trace_types_lock);
814 if (ret || !default_bootup_tracer)
817 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
820 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
821 /* Do we want this tracer to start on bootup? */
822 tracing_set_tracer(type->name);
823 default_bootup_tracer = NULL;
824 /* disable other selftests, since this will break it. */
825 tracing_selftest_disabled = 1;
826 #ifdef CONFIG_FTRACE_STARTUP_TEST
827 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
836 void unregister_tracer(struct tracer *type)
840 mutex_lock(&trace_types_lock);
841 for (t = &trace_types; *t; t = &(*t)->next) {
845 pr_info("Tracer %s not registered\n", type->name);
851 if (type == current_trace && tracer_enabled) {
854 if (current_trace->stop)
855 current_trace->stop(&global_trace);
856 current_trace = &nop_trace;
859 mutex_unlock(&trace_types_lock);
862 static void __tracing_reset(struct ring_buffer *buffer, int cpu)
864 ftrace_disable_cpu();
865 ring_buffer_reset_cpu(buffer, cpu);
869 void tracing_reset(struct trace_array *tr, int cpu)
871 struct ring_buffer *buffer = tr->buffer;
873 ring_buffer_record_disable(buffer);
875 /* Make sure all commits have finished */
877 __tracing_reset(buffer, cpu);
879 ring_buffer_record_enable(buffer);
882 void tracing_reset_online_cpus(struct trace_array *tr)
884 struct ring_buffer *buffer = tr->buffer;
887 ring_buffer_record_disable(buffer);
889 /* Make sure all commits have finished */
892 tr->time_start = ftrace_now(tr->cpu);
894 for_each_online_cpu(cpu)
895 __tracing_reset(buffer, cpu);
897 ring_buffer_record_enable(buffer);
900 void tracing_reset_current(int cpu)
902 tracing_reset(&global_trace, cpu);
905 void tracing_reset_current_online_cpus(void)
907 tracing_reset_online_cpus(&global_trace);
910 #define SAVED_CMDLINES 128
911 #define NO_CMDLINE_MAP UINT_MAX
912 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
913 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
914 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
915 static int cmdline_idx;
916 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
918 /* temporary disable recording */
919 static atomic_t trace_record_cmdline_disabled __read_mostly;
921 static void trace_init_cmdlines(void)
923 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
924 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
928 int is_tracing_stopped(void)
930 return trace_stop_count;
934 * ftrace_off_permanent - disable all ftrace code permanently
936 * This should only be called when a serious anomally has
937 * been detected. This will turn off the function tracing,
938 * ring buffers, and other tracing utilites. It takes no
939 * locks and can be called from any context.
941 void ftrace_off_permanent(void)
943 tracing_disabled = 1;
945 tracing_off_permanent();
949 * tracing_start - quick start of the tracer
951 * If tracing is enabled but was stopped by tracing_stop,
952 * this will start the tracer back up.
954 void tracing_start(void)
956 struct ring_buffer *buffer;
959 if (tracing_disabled)
962 spin_lock_irqsave(&tracing_start_lock, flags);
963 if (--trace_stop_count) {
964 if (trace_stop_count < 0) {
965 /* Someone screwed up their debugging */
967 trace_stop_count = 0;
972 /* Prevent the buffers from switching */
973 arch_spin_lock(&ftrace_max_lock);
975 buffer = global_trace.buffer;
977 ring_buffer_record_enable(buffer);
979 buffer = max_tr.buffer;
981 ring_buffer_record_enable(buffer);
983 arch_spin_unlock(&ftrace_max_lock);
987 spin_unlock_irqrestore(&tracing_start_lock, flags);
991 * tracing_stop - quick stop of the tracer
993 * Light weight way to stop tracing. Use in conjunction with
996 void tracing_stop(void)
998 struct ring_buffer *buffer;
1002 spin_lock_irqsave(&tracing_start_lock, flags);
1003 if (trace_stop_count++)
1006 /* Prevent the buffers from switching */
1007 arch_spin_lock(&ftrace_max_lock);
1009 buffer = global_trace.buffer;
1011 ring_buffer_record_disable(buffer);
1013 buffer = max_tr.buffer;
1015 ring_buffer_record_disable(buffer);
1017 arch_spin_unlock(&ftrace_max_lock);
1020 spin_unlock_irqrestore(&tracing_start_lock, flags);
1023 void trace_stop_cmdline_recording(void);
1025 static void trace_save_cmdline(struct task_struct *tsk)
1029 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1033 * It's not the end of the world if we don't get
1034 * the lock, but we also don't want to spin
1035 * nor do we want to disable interrupts,
1036 * so if we miss here, then better luck next time.
1038 if (!arch_spin_trylock(&trace_cmdline_lock))
1041 idx = map_pid_to_cmdline[tsk->pid];
1042 if (idx == NO_CMDLINE_MAP) {
1043 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1046 * Check whether the cmdline buffer at idx has a pid
1047 * mapped. We are going to overwrite that entry so we
1048 * need to clear the map_pid_to_cmdline. Otherwise we
1049 * would read the new comm for the old pid.
1051 pid = map_cmdline_to_pid[idx];
1052 if (pid != NO_CMDLINE_MAP)
1053 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1055 map_cmdline_to_pid[idx] = tsk->pid;
1056 map_pid_to_cmdline[tsk->pid] = idx;
1061 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1063 arch_spin_unlock(&trace_cmdline_lock);
1066 void trace_find_cmdline(int pid, char comm[])
1071 strcpy(comm, "<idle>");
1075 if (WARN_ON_ONCE(pid < 0)) {
1076 strcpy(comm, "<XXX>");
1080 if (pid > PID_MAX_DEFAULT) {
1081 strcpy(comm, "<...>");
1086 arch_spin_lock(&trace_cmdline_lock);
1087 map = map_pid_to_cmdline[pid];
1088 if (map != NO_CMDLINE_MAP)
1089 strcpy(comm, saved_cmdlines[map]);
1091 strcpy(comm, "<...>");
1093 arch_spin_unlock(&trace_cmdline_lock);
1097 void tracing_record_cmdline(struct task_struct *tsk)
1099 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1103 trace_save_cmdline(tsk);
1107 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1110 struct task_struct *tsk = current;
1112 entry->preempt_count = pc & 0xff;
1113 entry->pid = (tsk) ? tsk->pid : 0;
1114 entry->lock_depth = (tsk) ? tsk->lock_depth : 0;
1116 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1117 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1119 TRACE_FLAG_IRQS_NOSUPPORT |
1121 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1122 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1123 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1125 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1127 struct ring_buffer_event *
1128 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1131 unsigned long flags, int pc)
1133 struct ring_buffer_event *event;
1135 event = ring_buffer_lock_reserve(buffer, len);
1136 if (event != NULL) {
1137 struct trace_entry *ent = ring_buffer_event_data(event);
1139 tracing_generic_entry_update(ent, flags, pc);
1147 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1148 struct ring_buffer_event *event,
1149 unsigned long flags, int pc,
1152 ring_buffer_unlock_commit(buffer, event);
1154 ftrace_trace_stack(buffer, flags, 6, pc);
1155 ftrace_trace_userstack(buffer, flags, pc);
1161 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1162 struct ring_buffer_event *event,
1163 unsigned long flags, int pc)
1165 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1168 struct ring_buffer_event *
1169 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1170 int type, unsigned long len,
1171 unsigned long flags, int pc)
1173 *current_rb = global_trace.buffer;
1174 return trace_buffer_lock_reserve(*current_rb,
1175 type, len, flags, pc);
1177 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1179 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1180 struct ring_buffer_event *event,
1181 unsigned long flags, int pc)
1183 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1185 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1187 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1188 struct ring_buffer_event *event,
1189 unsigned long flags, int pc)
1191 __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1193 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1195 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1196 struct ring_buffer_event *event)
1198 ring_buffer_discard_commit(buffer, event);
1200 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1203 trace_function(struct trace_array *tr,
1204 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1207 struct ftrace_event_call *call = &event_function;
1208 struct ring_buffer *buffer = tr->buffer;
1209 struct ring_buffer_event *event;
1210 struct ftrace_entry *entry;
1212 /* If we are reading the ring buffer, don't trace */
1213 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1216 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1220 entry = ring_buffer_event_data(event);
1222 entry->parent_ip = parent_ip;
1224 if (!filter_check_discard(call, entry, buffer, event))
1225 ring_buffer_unlock_commit(buffer, event);
1229 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1230 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1233 if (likely(!atomic_read(&data->disabled)))
1234 trace_function(tr, ip, parent_ip, flags, pc);
1237 #ifdef CONFIG_STACKTRACE
1238 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1239 unsigned long flags,
1242 struct ftrace_event_call *call = &event_kernel_stack;
1243 struct ring_buffer_event *event;
1244 struct stack_entry *entry;
1245 struct stack_trace trace;
1247 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1248 sizeof(*entry), flags, pc);
1251 entry = ring_buffer_event_data(event);
1252 memset(&entry->caller, 0, sizeof(entry->caller));
1254 trace.nr_entries = 0;
1255 trace.max_entries = FTRACE_STACK_ENTRIES;
1257 trace.entries = entry->caller;
1259 save_stack_trace(&trace);
1260 if (!filter_check_discard(call, entry, buffer, event))
1261 ring_buffer_unlock_commit(buffer, event);
1264 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1267 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1270 __ftrace_trace_stack(buffer, flags, skip, pc);
1273 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1276 __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1280 * trace_dump_stack - record a stack back trace in the trace buffer
1282 void trace_dump_stack(void)
1284 unsigned long flags;
1286 if (tracing_disabled || tracing_selftest_running)
1289 local_save_flags(flags);
1291 /* skipping 3 traces, seems to get us at the caller of this function */
1292 __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count());
1296 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1298 struct ftrace_event_call *call = &event_user_stack;
1299 struct ring_buffer_event *event;
1300 struct userstack_entry *entry;
1301 struct stack_trace trace;
1303 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1307 * NMIs can not handle page faults, even with fix ups.
1308 * The save user stack can (and often does) fault.
1310 if (unlikely(in_nmi()))
1313 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1314 sizeof(*entry), flags, pc);
1317 entry = ring_buffer_event_data(event);
1319 entry->tgid = current->tgid;
1320 memset(&entry->caller, 0, sizeof(entry->caller));
1322 trace.nr_entries = 0;
1323 trace.max_entries = FTRACE_STACK_ENTRIES;
1325 trace.entries = entry->caller;
1327 save_stack_trace_user(&trace);
1328 if (!filter_check_discard(call, entry, buffer, event))
1329 ring_buffer_unlock_commit(buffer, event);
1333 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1335 ftrace_trace_userstack(tr, flags, preempt_count());
1339 #endif /* CONFIG_STACKTRACE */
1342 * trace_vbprintk - write binary msg to tracing buffer
1345 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1347 static arch_spinlock_t trace_buf_lock =
1348 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1349 static u32 trace_buf[TRACE_BUF_SIZE];
1351 struct ftrace_event_call *call = &event_bprint;
1352 struct ring_buffer_event *event;
1353 struct ring_buffer *buffer;
1354 struct trace_array *tr = &global_trace;
1355 struct trace_array_cpu *data;
1356 struct bprint_entry *entry;
1357 unsigned long flags;
1359 int cpu, len = 0, size, pc;
1361 if (unlikely(tracing_selftest_running || tracing_disabled))
1364 /* Don't pollute graph traces with trace_vprintk internals */
1365 pause_graph_tracing();
1367 pc = preempt_count();
1368 preempt_disable_notrace();
1369 cpu = raw_smp_processor_id();
1370 data = tr->data[cpu];
1372 disable = atomic_inc_return(&data->disabled);
1373 if (unlikely(disable != 1))
1376 /* Lockdep uses trace_printk for lock tracing */
1377 local_irq_save(flags);
1378 arch_spin_lock(&trace_buf_lock);
1379 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1381 if (len > TRACE_BUF_SIZE || len < 0)
1384 size = sizeof(*entry) + sizeof(u32) * len;
1385 buffer = tr->buffer;
1386 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1390 entry = ring_buffer_event_data(event);
1394 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1395 if (!filter_check_discard(call, entry, buffer, event)) {
1396 ring_buffer_unlock_commit(buffer, event);
1397 ftrace_trace_stack(buffer, flags, 6, pc);
1401 arch_spin_unlock(&trace_buf_lock);
1402 local_irq_restore(flags);
1405 atomic_dec_return(&data->disabled);
1406 preempt_enable_notrace();
1407 unpause_graph_tracing();
1411 EXPORT_SYMBOL_GPL(trace_vbprintk);
1413 int trace_array_printk(struct trace_array *tr,
1414 unsigned long ip, const char *fmt, ...)
1419 if (!(trace_flags & TRACE_ITER_PRINTK))
1423 ret = trace_array_vprintk(tr, ip, fmt, ap);
1428 int trace_array_vprintk(struct trace_array *tr,
1429 unsigned long ip, const char *fmt, va_list args)
1431 static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1432 static char trace_buf[TRACE_BUF_SIZE];
1434 struct ftrace_event_call *call = &event_print;
1435 struct ring_buffer_event *event;
1436 struct ring_buffer *buffer;
1437 struct trace_array_cpu *data;
1438 int cpu, len = 0, size, pc;
1439 struct print_entry *entry;
1440 unsigned long irq_flags;
1443 if (tracing_disabled || tracing_selftest_running)
1446 pc = preempt_count();
1447 preempt_disable_notrace();
1448 cpu = raw_smp_processor_id();
1449 data = tr->data[cpu];
1451 disable = atomic_inc_return(&data->disabled);
1452 if (unlikely(disable != 1))
1455 pause_graph_tracing();
1456 raw_local_irq_save(irq_flags);
1457 arch_spin_lock(&trace_buf_lock);
1458 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1460 size = sizeof(*entry) + len + 1;
1461 buffer = tr->buffer;
1462 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1466 entry = ring_buffer_event_data(event);
1469 memcpy(&entry->buf, trace_buf, len);
1470 entry->buf[len] = '\0';
1471 if (!filter_check_discard(call, entry, buffer, event)) {
1472 ring_buffer_unlock_commit(buffer, event);
1473 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1477 arch_spin_unlock(&trace_buf_lock);
1478 raw_local_irq_restore(irq_flags);
1479 unpause_graph_tracing();
1481 atomic_dec_return(&data->disabled);
1482 preempt_enable_notrace();
1487 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1489 return trace_array_vprintk(&global_trace, ip, fmt, args);
1491 EXPORT_SYMBOL_GPL(trace_vprintk);
1493 static void trace_iterator_increment(struct trace_iterator *iter)
1495 /* Don't allow ftrace to trace into the ring buffers */
1496 ftrace_disable_cpu();
1499 if (iter->buffer_iter[iter->cpu])
1500 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1502 ftrace_enable_cpu();
1505 static struct trace_entry *
1506 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1507 unsigned long *lost_events)
1509 struct ring_buffer_event *event;
1510 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1512 /* Don't allow ftrace to trace into the ring buffers */
1513 ftrace_disable_cpu();
1516 event = ring_buffer_iter_peek(buf_iter, ts);
1518 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1521 ftrace_enable_cpu();
1523 return event ? ring_buffer_event_data(event) : NULL;
1526 static struct trace_entry *
1527 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1528 unsigned long *missing_events, u64 *ent_ts)
1530 struct ring_buffer *buffer = iter->tr->buffer;
1531 struct trace_entry *ent, *next = NULL;
1532 unsigned long lost_events = 0, next_lost = 0;
1533 int cpu_file = iter->cpu_file;
1534 u64 next_ts = 0, ts;
1539 * If we are in a per_cpu trace file, don't bother by iterating over
1540 * all cpu and peek directly.
1542 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1543 if (ring_buffer_empty_cpu(buffer, cpu_file))
1545 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1547 *ent_cpu = cpu_file;
1552 for_each_tracing_cpu(cpu) {
1554 if (ring_buffer_empty_cpu(buffer, cpu))
1557 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1560 * Pick the entry with the smallest timestamp:
1562 if (ent && (!next || ts < next_ts)) {
1566 next_lost = lost_events;
1571 *ent_cpu = next_cpu;
1577 *missing_events = next_lost;
1582 /* Find the next real entry, without updating the iterator itself */
1583 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1584 int *ent_cpu, u64 *ent_ts)
1586 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1589 /* Find the next real entry, and increment the iterator to the next entry */
1590 void *trace_find_next_entry_inc(struct trace_iterator *iter)
1592 iter->ent = __find_next_entry(iter, &iter->cpu,
1593 &iter->lost_events, &iter->ts);
1596 trace_iterator_increment(iter);
1598 return iter->ent ? iter : NULL;
1601 static void trace_consume(struct trace_iterator *iter)
1603 /* Don't allow ftrace to trace into the ring buffers */
1604 ftrace_disable_cpu();
1605 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1606 &iter->lost_events);
1607 ftrace_enable_cpu();
1610 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1612 struct trace_iterator *iter = m->private;
1616 WARN_ON_ONCE(iter->leftover);
1620 /* can't go backwards */
1625 ent = trace_find_next_entry_inc(iter);
1629 while (ent && iter->idx < i)
1630 ent = trace_find_next_entry_inc(iter);
1637 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1639 struct trace_array *tr = iter->tr;
1640 struct ring_buffer_event *event;
1641 struct ring_buffer_iter *buf_iter;
1642 unsigned long entries = 0;
1645 tr->data[cpu]->skipped_entries = 0;
1647 if (!iter->buffer_iter[cpu])
1650 buf_iter = iter->buffer_iter[cpu];
1651 ring_buffer_iter_reset(buf_iter);
1654 * We could have the case with the max latency tracers
1655 * that a reset never took place on a cpu. This is evident
1656 * by the timestamp being before the start of the buffer.
1658 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1659 if (ts >= iter->tr->time_start)
1662 ring_buffer_read(buf_iter, NULL);
1665 tr->data[cpu]->skipped_entries = entries;
1669 * The current tracer is copied to avoid a global locking
1672 static void *s_start(struct seq_file *m, loff_t *pos)
1674 struct trace_iterator *iter = m->private;
1675 static struct tracer *old_tracer;
1676 int cpu_file = iter->cpu_file;
1681 /* copy the tracer to avoid using a global lock all around */
1682 mutex_lock(&trace_types_lock);
1683 if (unlikely(old_tracer != current_trace && current_trace)) {
1684 old_tracer = current_trace;
1685 *iter->trace = *current_trace;
1687 mutex_unlock(&trace_types_lock);
1689 atomic_inc(&trace_record_cmdline_disabled);
1691 if (*pos != iter->pos) {
1696 ftrace_disable_cpu();
1698 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1699 for_each_tracing_cpu(cpu)
1700 tracing_iter_reset(iter, cpu);
1702 tracing_iter_reset(iter, cpu_file);
1704 ftrace_enable_cpu();
1707 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1712 * If we overflowed the seq_file before, then we want
1713 * to just reuse the trace_seq buffer again.
1719 p = s_next(m, p, &l);
1723 trace_event_read_lock();
1724 trace_access_lock(cpu_file);
1728 static void s_stop(struct seq_file *m, void *p)
1730 struct trace_iterator *iter = m->private;
1732 atomic_dec(&trace_record_cmdline_disabled);
1733 trace_access_unlock(iter->cpu_file);
1734 trace_event_read_unlock();
1737 static void print_lat_help_header(struct seq_file *m)
1739 seq_puts(m, "# _------=> CPU# \n");
1740 seq_puts(m, "# / _-----=> irqs-off \n");
1741 seq_puts(m, "# | / _----=> need-resched \n");
1742 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1743 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1744 seq_puts(m, "# |||| /_--=> lock-depth \n");
1745 seq_puts(m, "# |||||/ delay \n");
1746 seq_puts(m, "# cmd pid |||||| time | caller \n");
1747 seq_puts(m, "# \\ / |||||| \\ | / \n");
1750 static void print_func_help_header(struct seq_file *m)
1752 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1753 seq_puts(m, "# | | | | |\n");
1758 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1760 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1761 struct trace_array *tr = iter->tr;
1762 struct trace_array_cpu *data = tr->data[tr->cpu];
1763 struct tracer *type = current_trace;
1764 unsigned long entries = 0;
1765 unsigned long total = 0;
1766 unsigned long count;
1767 const char *name = "preemption";
1774 for_each_tracing_cpu(cpu) {
1775 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1777 * If this buffer has skipped entries, then we hold all
1778 * entries for the trace and we need to ignore the
1779 * ones before the time stamp.
1781 if (tr->data[cpu]->skipped_entries) {
1782 count -= tr->data[cpu]->skipped_entries;
1783 /* total is the same as the entries */
1787 ring_buffer_overrun_cpu(tr->buffer, cpu);
1791 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1793 seq_puts(m, "# -----------------------------------"
1794 "---------------------------------\n");
1795 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1796 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1797 nsecs_to_usecs(data->saved_latency),
1801 #if defined(CONFIG_PREEMPT_NONE)
1803 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1805 #elif defined(CONFIG_PREEMPT)
1810 /* These are reserved for later use */
1813 seq_printf(m, " #P:%d)\n", num_online_cpus());
1817 seq_puts(m, "# -----------------\n");
1818 seq_printf(m, "# | task: %.16s-%d "
1819 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1820 data->comm, data->pid, data->uid, data->nice,
1821 data->policy, data->rt_priority);
1822 seq_puts(m, "# -----------------\n");
1824 if (data->critical_start) {
1825 seq_puts(m, "# => started at: ");
1826 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1827 trace_print_seq(m, &iter->seq);
1828 seq_puts(m, "\n# => ended at: ");
1829 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1830 trace_print_seq(m, &iter->seq);
1831 seq_puts(m, "\n#\n");
1837 static void test_cpu_buff_start(struct trace_iterator *iter)
1839 struct trace_seq *s = &iter->seq;
1841 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1844 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1847 if (cpumask_test_cpu(iter->cpu, iter->started))
1850 if (iter->tr->data[iter->cpu]->skipped_entries)
1853 cpumask_set_cpu(iter->cpu, iter->started);
1855 /* Don't print started cpu buffer for the first entry of the trace */
1857 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1861 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1863 struct trace_seq *s = &iter->seq;
1864 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1865 struct trace_entry *entry;
1866 struct trace_event *event;
1870 test_cpu_buff_start(iter);
1872 event = ftrace_find_event(entry->type);
1874 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1875 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1876 if (!trace_print_lat_context(iter))
1879 if (!trace_print_context(iter))
1885 return event->funcs->trace(iter, sym_flags, event);
1887 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1890 return TRACE_TYPE_HANDLED;
1892 return TRACE_TYPE_PARTIAL_LINE;
1895 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1897 struct trace_seq *s = &iter->seq;
1898 struct trace_entry *entry;
1899 struct trace_event *event;
1903 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1904 if (!trace_seq_printf(s, "%d %d %llu ",
1905 entry->pid, iter->cpu, iter->ts))
1909 event = ftrace_find_event(entry->type);
1911 return event->funcs->raw(iter, 0, event);
1913 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1916 return TRACE_TYPE_HANDLED;
1918 return TRACE_TYPE_PARTIAL_LINE;
1921 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1923 struct trace_seq *s = &iter->seq;
1924 unsigned char newline = '\n';
1925 struct trace_entry *entry;
1926 struct trace_event *event;
1930 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1931 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1932 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1933 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1936 event = ftrace_find_event(entry->type);
1938 enum print_line_t ret = event->funcs->hex(iter, 0, event);
1939 if (ret != TRACE_TYPE_HANDLED)
1943 SEQ_PUT_FIELD_RET(s, newline);
1945 return TRACE_TYPE_HANDLED;
1948 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1950 struct trace_seq *s = &iter->seq;
1951 struct trace_entry *entry;
1952 struct trace_event *event;
1956 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1957 SEQ_PUT_FIELD_RET(s, entry->pid);
1958 SEQ_PUT_FIELD_RET(s, iter->cpu);
1959 SEQ_PUT_FIELD_RET(s, iter->ts);
1962 event = ftrace_find_event(entry->type);
1963 return event ? event->funcs->binary(iter, 0, event) :
1967 int trace_empty(struct trace_iterator *iter)
1971 /* If we are looking at one CPU buffer, only check that one */
1972 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1973 cpu = iter->cpu_file;
1974 if (iter->buffer_iter[cpu]) {
1975 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1978 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1984 for_each_tracing_cpu(cpu) {
1985 if (iter->buffer_iter[cpu]) {
1986 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1989 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1997 /* Called with trace_event_read_lock() held. */
1998 enum print_line_t print_trace_line(struct trace_iterator *iter)
2000 enum print_line_t ret;
2002 if (iter->lost_events)
2003 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2004 iter->cpu, iter->lost_events);
2006 if (iter->trace && iter->trace->print_line) {
2007 ret = iter->trace->print_line(iter);
2008 if (ret != TRACE_TYPE_UNHANDLED)
2012 if (iter->ent->type == TRACE_BPRINT &&
2013 trace_flags & TRACE_ITER_PRINTK &&
2014 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2015 return trace_print_bprintk_msg_only(iter);
2017 if (iter->ent->type == TRACE_PRINT &&
2018 trace_flags & TRACE_ITER_PRINTK &&
2019 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2020 return trace_print_printk_msg_only(iter);
2022 if (trace_flags & TRACE_ITER_BIN)
2023 return print_bin_fmt(iter);
2025 if (trace_flags & TRACE_ITER_HEX)
2026 return print_hex_fmt(iter);
2028 if (trace_flags & TRACE_ITER_RAW)
2029 return print_raw_fmt(iter);
2031 return print_trace_fmt(iter);
2034 void trace_default_header(struct seq_file *m)
2036 struct trace_iterator *iter = m->private;
2038 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2039 /* print nothing if the buffers are empty */
2040 if (trace_empty(iter))
2042 print_trace_header(m, iter);
2043 if (!(trace_flags & TRACE_ITER_VERBOSE))
2044 print_lat_help_header(m);
2046 if (!(trace_flags & TRACE_ITER_VERBOSE))
2047 print_func_help_header(m);
2051 static int s_show(struct seq_file *m, void *v)
2053 struct trace_iterator *iter = v;
2056 if (iter->ent == NULL) {
2058 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2061 if (iter->trace && iter->trace->print_header)
2062 iter->trace->print_header(m);
2064 trace_default_header(m);
2066 } else if (iter->leftover) {
2068 * If we filled the seq_file buffer earlier, we
2069 * want to just show it now.
2071 ret = trace_print_seq(m, &iter->seq);
2073 /* ret should this time be zero, but you never know */
2074 iter->leftover = ret;
2077 print_trace_line(iter);
2078 ret = trace_print_seq(m, &iter->seq);
2080 * If we overflow the seq_file buffer, then it will
2081 * ask us for this data again at start up.
2083 * ret is 0 if seq_file write succeeded.
2086 iter->leftover = ret;
2092 static const struct seq_operations tracer_seq_ops = {
2099 static struct trace_iterator *
2100 __tracing_open(struct inode *inode, struct file *file)
2102 long cpu_file = (long) inode->i_private;
2103 void *fail_ret = ERR_PTR(-ENOMEM);
2104 struct trace_iterator *iter;
2108 if (tracing_disabled)
2109 return ERR_PTR(-ENODEV);
2111 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2113 return ERR_PTR(-ENOMEM);
2116 * We make a copy of the current tracer to avoid concurrent
2117 * changes on it while we are reading.
2119 mutex_lock(&trace_types_lock);
2120 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2125 *iter->trace = *current_trace;
2127 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2130 if (current_trace && current_trace->print_max)
2133 iter->tr = &global_trace;
2135 mutex_init(&iter->mutex);
2136 iter->cpu_file = cpu_file;
2138 /* Notify the tracer early; before we stop tracing. */
2139 if (iter->trace && iter->trace->open)
2140 iter->trace->open(iter);
2142 /* Annotate start of buffers if we had overruns */
2143 if (ring_buffer_overruns(iter->tr->buffer))
2144 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2146 /* stop the trace while dumping */
2149 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2150 for_each_tracing_cpu(cpu) {
2151 iter->buffer_iter[cpu] =
2152 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2154 ring_buffer_read_prepare_sync();
2155 for_each_tracing_cpu(cpu) {
2156 ring_buffer_read_start(iter->buffer_iter[cpu]);
2157 tracing_iter_reset(iter, cpu);
2160 cpu = iter->cpu_file;
2161 iter->buffer_iter[cpu] =
2162 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2163 ring_buffer_read_prepare_sync();
2164 ring_buffer_read_start(iter->buffer_iter[cpu]);
2165 tracing_iter_reset(iter, cpu);
2168 ret = seq_open(file, &tracer_seq_ops);
2170 fail_ret = ERR_PTR(ret);
2174 m = file->private_data;
2177 mutex_unlock(&trace_types_lock);
2182 for_each_tracing_cpu(cpu) {
2183 if (iter->buffer_iter[cpu])
2184 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2186 free_cpumask_var(iter->started);
2189 mutex_unlock(&trace_types_lock);
2196 int tracing_open_generic(struct inode *inode, struct file *filp)
2198 if (tracing_disabled)
2201 filp->private_data = inode->i_private;
2205 static int tracing_release(struct inode *inode, struct file *file)
2207 struct seq_file *m = (struct seq_file *)file->private_data;
2208 struct trace_iterator *iter;
2211 if (!(file->f_mode & FMODE_READ))
2216 mutex_lock(&trace_types_lock);
2217 for_each_tracing_cpu(cpu) {
2218 if (iter->buffer_iter[cpu])
2219 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2222 if (iter->trace && iter->trace->close)
2223 iter->trace->close(iter);
2225 /* reenable tracing if it was previously enabled */
2227 mutex_unlock(&trace_types_lock);
2229 seq_release(inode, file);
2230 mutex_destroy(&iter->mutex);
2231 free_cpumask_var(iter->started);
2237 static int tracing_open(struct inode *inode, struct file *file)
2239 struct trace_iterator *iter;
2242 /* If this file was open for write, then erase contents */
2243 if ((file->f_mode & FMODE_WRITE) &&
2244 (file->f_flags & O_TRUNC)) {
2245 long cpu = (long) inode->i_private;
2247 if (cpu == TRACE_PIPE_ALL_CPU)
2248 tracing_reset_online_cpus(&global_trace);
2250 tracing_reset(&global_trace, cpu);
2253 if (file->f_mode & FMODE_READ) {
2254 iter = __tracing_open(inode, file);
2256 ret = PTR_ERR(iter);
2257 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2258 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2264 t_next(struct seq_file *m, void *v, loff_t *pos)
2266 struct tracer *t = v;
2276 static void *t_start(struct seq_file *m, loff_t *pos)
2281 mutex_lock(&trace_types_lock);
2282 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2288 static void t_stop(struct seq_file *m, void *p)
2290 mutex_unlock(&trace_types_lock);
2293 static int t_show(struct seq_file *m, void *v)
2295 struct tracer *t = v;
2300 seq_printf(m, "%s", t->name);
2309 static const struct seq_operations show_traces_seq_ops = {
2316 static int show_traces_open(struct inode *inode, struct file *file)
2318 if (tracing_disabled)
2321 return seq_open(file, &show_traces_seq_ops);
2325 tracing_write_stub(struct file *filp, const char __user *ubuf,
2326 size_t count, loff_t *ppos)
2331 static const struct file_operations tracing_fops = {
2332 .open = tracing_open,
2334 .write = tracing_write_stub,
2335 .llseek = seq_lseek,
2336 .release = tracing_release,
2339 static const struct file_operations show_traces_fops = {
2340 .open = show_traces_open,
2342 .release = seq_release,
2343 .llseek = seq_lseek,
2347 * Only trace on a CPU if the bitmask is set:
2349 static cpumask_var_t tracing_cpumask;
2352 * The tracer itself will not take this lock, but still we want
2353 * to provide a consistent cpumask to user-space:
2355 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2358 * Temporary storage for the character representation of the
2359 * CPU bitmask (and one more byte for the newline):
2361 static char mask_str[NR_CPUS + 1];
2364 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2365 size_t count, loff_t *ppos)
2369 mutex_lock(&tracing_cpumask_update_lock);
2371 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2372 if (count - len < 2) {
2376 len += sprintf(mask_str + len, "\n");
2377 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2380 mutex_unlock(&tracing_cpumask_update_lock);
2386 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2387 size_t count, loff_t *ppos)
2390 cpumask_var_t tracing_cpumask_new;
2392 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2395 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2399 mutex_lock(&tracing_cpumask_update_lock);
2401 local_irq_disable();
2402 arch_spin_lock(&ftrace_max_lock);
2403 for_each_tracing_cpu(cpu) {
2405 * Increase/decrease the disabled counter if we are
2406 * about to flip a bit in the cpumask:
2408 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2409 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2410 atomic_inc(&global_trace.data[cpu]->disabled);
2412 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2413 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2414 atomic_dec(&global_trace.data[cpu]->disabled);
2417 arch_spin_unlock(&ftrace_max_lock);
2420 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2422 mutex_unlock(&tracing_cpumask_update_lock);
2423 free_cpumask_var(tracing_cpumask_new);
2428 free_cpumask_var(tracing_cpumask_new);
2433 static const struct file_operations tracing_cpumask_fops = {
2434 .open = tracing_open_generic,
2435 .read = tracing_cpumask_read,
2436 .write = tracing_cpumask_write,
2437 .llseek = generic_file_llseek,
2440 static int tracing_trace_options_show(struct seq_file *m, void *v)
2442 struct tracer_opt *trace_opts;
2446 mutex_lock(&trace_types_lock);
2447 tracer_flags = current_trace->flags->val;
2448 trace_opts = current_trace->flags->opts;
2450 for (i = 0; trace_options[i]; i++) {
2451 if (trace_flags & (1 << i))
2452 seq_printf(m, "%s\n", trace_options[i]);
2454 seq_printf(m, "no%s\n", trace_options[i]);
2457 for (i = 0; trace_opts[i].name; i++) {
2458 if (tracer_flags & trace_opts[i].bit)
2459 seq_printf(m, "%s\n", trace_opts[i].name);
2461 seq_printf(m, "no%s\n", trace_opts[i].name);
2463 mutex_unlock(&trace_types_lock);
2468 static int __set_tracer_option(struct tracer *trace,
2469 struct tracer_flags *tracer_flags,
2470 struct tracer_opt *opts, int neg)
2474 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2479 tracer_flags->val &= ~opts->bit;
2481 tracer_flags->val |= opts->bit;
2485 /* Try to assign a tracer specific option */
2486 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2488 struct tracer_flags *tracer_flags = trace->flags;
2489 struct tracer_opt *opts = NULL;
2492 for (i = 0; tracer_flags->opts[i].name; i++) {
2493 opts = &tracer_flags->opts[i];
2495 if (strcmp(cmp, opts->name) == 0)
2496 return __set_tracer_option(trace, trace->flags,
2503 static void set_tracer_flags(unsigned int mask, int enabled)
2505 /* do nothing if flag is already set */
2506 if (!!(trace_flags & mask) == !!enabled)
2510 trace_flags |= mask;
2512 trace_flags &= ~mask;
2514 if (mask == TRACE_ITER_RECORD_CMD)
2515 trace_event_enable_cmd_record(enabled);
2519 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2520 size_t cnt, loff_t *ppos)
2528 if (cnt >= sizeof(buf))
2531 if (copy_from_user(&buf, ubuf, cnt))
2535 cmp = strstrip(buf);
2537 if (strncmp(cmp, "no", 2) == 0) {
2542 for (i = 0; trace_options[i]; i++) {
2543 if (strcmp(cmp, trace_options[i]) == 0) {
2544 set_tracer_flags(1 << i, !neg);
2549 /* If no option could be set, test the specific tracer options */
2550 if (!trace_options[i]) {
2551 mutex_lock(&trace_types_lock);
2552 ret = set_tracer_option(current_trace, cmp, neg);
2553 mutex_unlock(&trace_types_lock);
2563 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2565 if (tracing_disabled)
2567 return single_open(file, tracing_trace_options_show, NULL);
2570 static const struct file_operations tracing_iter_fops = {
2571 .open = tracing_trace_options_open,
2573 .llseek = seq_lseek,
2574 .release = single_release,
2575 .write = tracing_trace_options_write,
2578 static const char readme_msg[] =
2579 "tracing mini-HOWTO:\n\n"
2580 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2581 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2582 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2583 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2585 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2586 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2588 "# cat /sys/kernel/debug/tracing/trace_options\n"
2589 "noprint-parent nosym-offset nosym-addr noverbose\n"
2590 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2591 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2592 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2593 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2597 tracing_readme_read(struct file *filp, char __user *ubuf,
2598 size_t cnt, loff_t *ppos)
2600 return simple_read_from_buffer(ubuf, cnt, ppos,
2601 readme_msg, strlen(readme_msg));
2604 static const struct file_operations tracing_readme_fops = {
2605 .open = tracing_open_generic,
2606 .read = tracing_readme_read,
2607 .llseek = generic_file_llseek,
2611 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2612 size_t cnt, loff_t *ppos)
2621 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2625 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2633 for (i = 0; i < SAVED_CMDLINES; i++) {
2636 pid = map_cmdline_to_pid[i];
2637 if (pid == -1 || pid == NO_CMDLINE_MAP)
2640 trace_find_cmdline(pid, buf_comm);
2641 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2646 len = simple_read_from_buffer(ubuf, cnt, ppos,
2655 static const struct file_operations tracing_saved_cmdlines_fops = {
2656 .open = tracing_open_generic,
2657 .read = tracing_saved_cmdlines_read,
2658 .llseek = generic_file_llseek,
2662 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2663 size_t cnt, loff_t *ppos)
2668 r = sprintf(buf, "%u\n", tracer_enabled);
2669 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2673 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2674 size_t cnt, loff_t *ppos)
2676 struct trace_array *tr = filp->private_data;
2681 if (cnt >= sizeof(buf))
2684 if (copy_from_user(&buf, ubuf, cnt))
2689 ret = strict_strtoul(buf, 10, &val);
2695 mutex_lock(&trace_types_lock);
2696 if (tracer_enabled ^ val) {
2699 if (current_trace->start)
2700 current_trace->start(tr);
2705 if (current_trace->stop)
2706 current_trace->stop(tr);
2709 mutex_unlock(&trace_types_lock);
2717 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2718 size_t cnt, loff_t *ppos)
2720 char buf[MAX_TRACER_SIZE+2];
2723 mutex_lock(&trace_types_lock);
2725 r = sprintf(buf, "%s\n", current_trace->name);
2727 r = sprintf(buf, "\n");
2728 mutex_unlock(&trace_types_lock);
2730 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2733 int tracer_init(struct tracer *t, struct trace_array *tr)
2735 tracing_reset_online_cpus(tr);
2739 static int tracing_resize_ring_buffer(unsigned long size)
2744 * If kernel or user changes the size of the ring buffer
2745 * we use the size that was given, and we can forget about
2746 * expanding it later.
2748 ring_buffer_expanded = 1;
2750 ret = ring_buffer_resize(global_trace.buffer, size);
2754 if (!current_trace->use_max_tr)
2757 ret = ring_buffer_resize(max_tr.buffer, size);
2761 r = ring_buffer_resize(global_trace.buffer,
2762 global_trace.entries);
2765 * AARGH! We are left with different
2766 * size max buffer!!!!
2767 * The max buffer is our "snapshot" buffer.
2768 * When a tracer needs a snapshot (one of the
2769 * latency tracers), it swaps the max buffer
2770 * with the saved snap shot. We succeeded to
2771 * update the size of the main buffer, but failed to
2772 * update the size of the max buffer. But when we tried
2773 * to reset the main buffer to the original size, we
2774 * failed there too. This is very unlikely to
2775 * happen, but if it does, warn and kill all
2779 tracing_disabled = 1;
2784 max_tr.entries = size;
2786 global_trace.entries = size;
2793 * tracing_update_buffers - used by tracing facility to expand ring buffers
2795 * To save on memory when the tracing is never used on a system with it
2796 * configured in. The ring buffers are set to a minimum size. But once
2797 * a user starts to use the tracing facility, then they need to grow
2798 * to their default size.
2800 * This function is to be called when a tracer is about to be used.
2802 int tracing_update_buffers(void)
2806 mutex_lock(&trace_types_lock);
2807 if (!ring_buffer_expanded)
2808 ret = tracing_resize_ring_buffer(trace_buf_size);
2809 mutex_unlock(&trace_types_lock);
2814 struct trace_option_dentry;
2816 static struct trace_option_dentry *
2817 create_trace_option_files(struct tracer *tracer);
2820 destroy_trace_option_files(struct trace_option_dentry *topts);
2822 static int tracing_set_tracer(const char *buf)
2824 static struct trace_option_dentry *topts;
2825 struct trace_array *tr = &global_trace;
2829 mutex_lock(&trace_types_lock);
2831 if (!ring_buffer_expanded) {
2832 ret = tracing_resize_ring_buffer(trace_buf_size);
2838 for (t = trace_types; t; t = t->next) {
2839 if (strcmp(t->name, buf) == 0)
2846 if (t == current_trace)
2849 trace_branch_disable();
2850 if (current_trace && current_trace->reset)
2851 current_trace->reset(tr);
2852 if (current_trace && current_trace->use_max_tr) {
2854 * We don't free the ring buffer. instead, resize it because
2855 * The max_tr ring buffer has some state (e.g. ring->clock) and
2856 * we want preserve it.
2858 ring_buffer_resize(max_tr.buffer, 1);
2861 destroy_trace_option_files(topts);
2865 topts = create_trace_option_files(current_trace);
2866 if (current_trace->use_max_tr) {
2867 ret = ring_buffer_resize(max_tr.buffer, global_trace.entries);
2870 max_tr.entries = global_trace.entries;
2874 ret = tracer_init(t, tr);
2879 trace_branch_enable(tr);
2881 mutex_unlock(&trace_types_lock);
2887 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2888 size_t cnt, loff_t *ppos)
2890 char buf[MAX_TRACER_SIZE+1];
2897 if (cnt > MAX_TRACER_SIZE)
2898 cnt = MAX_TRACER_SIZE;
2900 if (copy_from_user(&buf, ubuf, cnt))
2905 /* strip ending whitespace. */
2906 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2909 err = tracing_set_tracer(buf);
2919 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2920 size_t cnt, loff_t *ppos)
2922 unsigned long *ptr = filp->private_data;
2926 r = snprintf(buf, sizeof(buf), "%ld\n",
2927 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2928 if (r > sizeof(buf))
2930 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2934 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2935 size_t cnt, loff_t *ppos)
2937 unsigned long *ptr = filp->private_data;
2942 if (cnt >= sizeof(buf))
2945 if (copy_from_user(&buf, ubuf, cnt))
2950 ret = strict_strtoul(buf, 10, &val);
2959 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2961 long cpu_file = (long) inode->i_private;
2962 struct trace_iterator *iter;
2965 if (tracing_disabled)
2968 mutex_lock(&trace_types_lock);
2970 /* create a buffer to store the information to pass to userspace */
2971 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2978 * We make a copy of the current tracer to avoid concurrent
2979 * changes on it while we are reading.
2981 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2987 *iter->trace = *current_trace;
2989 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2994 /* trace pipe does not show start of buffer */
2995 cpumask_setall(iter->started);
2997 if (trace_flags & TRACE_ITER_LATENCY_FMT)
2998 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3000 iter->cpu_file = cpu_file;
3001 iter->tr = &global_trace;
3002 mutex_init(&iter->mutex);
3003 filp->private_data = iter;
3005 if (iter->trace->pipe_open)
3006 iter->trace->pipe_open(iter);
3008 nonseekable_open(inode, filp);
3010 mutex_unlock(&trace_types_lock);
3016 mutex_unlock(&trace_types_lock);
3020 static int tracing_release_pipe(struct inode *inode, struct file *file)
3022 struct trace_iterator *iter = file->private_data;
3024 mutex_lock(&trace_types_lock);
3026 if (iter->trace->pipe_close)
3027 iter->trace->pipe_close(iter);
3029 mutex_unlock(&trace_types_lock);
3031 free_cpumask_var(iter->started);
3032 mutex_destroy(&iter->mutex);
3040 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3042 struct trace_iterator *iter = filp->private_data;
3044 if (trace_flags & TRACE_ITER_BLOCK) {
3046 * Always select as readable when in blocking mode
3048 return POLLIN | POLLRDNORM;
3050 if (!trace_empty(iter))
3051 return POLLIN | POLLRDNORM;
3052 poll_wait(filp, &trace_wait, poll_table);
3053 if (!trace_empty(iter))
3054 return POLLIN | POLLRDNORM;
3061 void default_wait_pipe(struct trace_iterator *iter)
3065 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3067 if (trace_empty(iter))
3070 finish_wait(&trace_wait, &wait);
3074 * This is a make-shift waitqueue.
3075 * A tracer might use this callback on some rare cases:
3077 * 1) the current tracer might hold the runqueue lock when it wakes up
3078 * a reader, hence a deadlock (sched, function, and function graph tracers)
3079 * 2) the function tracers, trace all functions, we don't want
3080 * the overhead of calling wake_up and friends
3081 * (and tracing them too)
3083 * Anyway, this is really very primitive wakeup.
3085 void poll_wait_pipe(struct trace_iterator *iter)
3087 set_current_state(TASK_INTERRUPTIBLE);
3088 /* sleep for 100 msecs, and try again. */
3089 schedule_timeout(HZ / 10);
3092 /* Must be called with trace_types_lock mutex held. */
3093 static int tracing_wait_pipe(struct file *filp)
3095 struct trace_iterator *iter = filp->private_data;
3097 while (trace_empty(iter)) {
3099 if ((filp->f_flags & O_NONBLOCK)) {
3103 mutex_unlock(&iter->mutex);
3105 iter->trace->wait_pipe(iter);
3107 mutex_lock(&iter->mutex);
3109 if (signal_pending(current))
3113 * We block until we read something and tracing is disabled.
3114 * We still block if tracing is disabled, but we have never
3115 * read anything. This allows a user to cat this file, and
3116 * then enable tracing. But after we have read something,
3117 * we give an EOF when tracing is again disabled.
3119 * iter->pos will be 0 if we haven't read anything.
3121 if (!tracer_enabled && iter->pos)
3132 tracing_read_pipe(struct file *filp, char __user *ubuf,
3133 size_t cnt, loff_t *ppos)
3135 struct trace_iterator *iter = filp->private_data;
3136 static struct tracer *old_tracer;
3139 /* return any leftover data */
3140 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3144 trace_seq_init(&iter->seq);
3146 /* copy the tracer to avoid using a global lock all around */
3147 mutex_lock(&trace_types_lock);
3148 if (unlikely(old_tracer != current_trace && current_trace)) {
3149 old_tracer = current_trace;
3150 *iter->trace = *current_trace;
3152 mutex_unlock(&trace_types_lock);
3155 * Avoid more than one consumer on a single file descriptor
3156 * This is just a matter of traces coherency, the ring buffer itself
3159 mutex_lock(&iter->mutex);
3160 if (iter->trace->read) {
3161 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3167 sret = tracing_wait_pipe(filp);
3171 /* stop when tracing is finished */
3172 if (trace_empty(iter)) {
3177 if (cnt >= PAGE_SIZE)
3178 cnt = PAGE_SIZE - 1;
3180 /* reset all but tr, trace, and overruns */
3181 memset(&iter->seq, 0,
3182 sizeof(struct trace_iterator) -
3183 offsetof(struct trace_iterator, seq));
3186 trace_event_read_lock();
3187 trace_access_lock(iter->cpu_file);
3188 while (trace_find_next_entry_inc(iter) != NULL) {
3189 enum print_line_t ret;
3190 int len = iter->seq.len;
3192 ret = print_trace_line(iter);
3193 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3194 /* don't print partial lines */
3195 iter->seq.len = len;
3198 if (ret != TRACE_TYPE_NO_CONSUME)
3199 trace_consume(iter);
3201 if (iter->seq.len >= cnt)
3204 trace_access_unlock(iter->cpu_file);
3205 trace_event_read_unlock();
3207 /* Now copy what we have to the user */
3208 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3209 if (iter->seq.readpos >= iter->seq.len)
3210 trace_seq_init(&iter->seq);
3213 * If there was nothing to send to user, inspite of consuming trace
3214 * entries, go back to wait for more entries.
3220 mutex_unlock(&iter->mutex);
3225 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3226 struct pipe_buffer *buf)
3228 __free_page(buf->page);
3231 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3234 __free_page(spd->pages[idx]);
3237 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3239 .map = generic_pipe_buf_map,
3240 .unmap = generic_pipe_buf_unmap,
3241 .confirm = generic_pipe_buf_confirm,
3242 .release = tracing_pipe_buf_release,
3243 .steal = generic_pipe_buf_steal,
3244 .get = generic_pipe_buf_get,
3248 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3253 /* Seq buffer is page-sized, exactly what we need. */
3255 count = iter->seq.len;
3256 ret = print_trace_line(iter);
3257 count = iter->seq.len - count;
3260 iter->seq.len -= count;
3263 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3264 iter->seq.len -= count;
3268 if (ret != TRACE_TYPE_NO_CONSUME)
3269 trace_consume(iter);
3271 if (!trace_find_next_entry_inc(iter)) {
3281 static ssize_t tracing_splice_read_pipe(struct file *filp,
3283 struct pipe_inode_info *pipe,
3287 struct page *pages_def[PIPE_DEF_BUFFERS];
3288 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3289 struct trace_iterator *iter = filp->private_data;
3290 struct splice_pipe_desc spd = {
3292 .partial = partial_def,
3293 .nr_pages = 0, /* This gets updated below. */
3295 .ops = &tracing_pipe_buf_ops,
3296 .spd_release = tracing_spd_release_pipe,
3298 static struct tracer *old_tracer;
3303 if (splice_grow_spd(pipe, &spd))
3306 /* copy the tracer to avoid using a global lock all around */
3307 mutex_lock(&trace_types_lock);
3308 if (unlikely(old_tracer != current_trace && current_trace)) {
3309 old_tracer = current_trace;
3310 *iter->trace = *current_trace;
3312 mutex_unlock(&trace_types_lock);
3314 mutex_lock(&iter->mutex);
3316 if (iter->trace->splice_read) {
3317 ret = iter->trace->splice_read(iter, filp,
3318 ppos, pipe, len, flags);
3323 ret = tracing_wait_pipe(filp);
3327 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
3332 trace_event_read_lock();
3333 trace_access_lock(iter->cpu_file);
3335 /* Fill as many pages as possible. */
3336 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3337 spd.pages[i] = alloc_page(GFP_KERNEL);
3341 rem = tracing_fill_pipe_page(rem, iter);
3343 /* Copy the data into the page, so we can start over. */
3344 ret = trace_seq_to_buffer(&iter->seq,
3345 page_address(spd.pages[i]),
3348 __free_page(spd.pages[i]);
3351 spd.partial[i].offset = 0;
3352 spd.partial[i].len = iter->seq.len;
3354 trace_seq_init(&iter->seq);
3357 trace_access_unlock(iter->cpu_file);
3358 trace_event_read_unlock();
3359 mutex_unlock(&iter->mutex);
3363 ret = splice_to_pipe(pipe, &spd);
3365 splice_shrink_spd(pipe, &spd);
3369 mutex_unlock(&iter->mutex);
3374 tracing_entries_read(struct file *filp, char __user *ubuf,
3375 size_t cnt, loff_t *ppos)
3377 struct trace_array *tr = filp->private_data;
3381 mutex_lock(&trace_types_lock);
3382 if (!ring_buffer_expanded)
3383 r = sprintf(buf, "%lu (expanded: %lu)\n",
3385 trace_buf_size >> 10);
3387 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3388 mutex_unlock(&trace_types_lock);
3390 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3394 tracing_entries_write(struct file *filp, const char __user *ubuf,
3395 size_t cnt, loff_t *ppos)
3401 if (cnt >= sizeof(buf))
3404 if (copy_from_user(&buf, ubuf, cnt))
3409 ret = strict_strtoul(buf, 10, &val);
3413 /* must have at least 1 entry */
3417 mutex_lock(&trace_types_lock);
3421 /* disable all cpu buffers */
3422 for_each_tracing_cpu(cpu) {
3423 if (global_trace.data[cpu])
3424 atomic_inc(&global_trace.data[cpu]->disabled);
3425 if (max_tr.data[cpu])
3426 atomic_inc(&max_tr.data[cpu]->disabled);
3429 /* value is in KB */
3432 if (val != global_trace.entries) {
3433 ret = tracing_resize_ring_buffer(val);
3442 /* If check pages failed, return ENOMEM */
3443 if (tracing_disabled)
3446 for_each_tracing_cpu(cpu) {
3447 if (global_trace.data[cpu])
3448 atomic_dec(&global_trace.data[cpu]->disabled);
3449 if (max_tr.data[cpu])
3450 atomic_dec(&max_tr.data[cpu]->disabled);
3454 mutex_unlock(&trace_types_lock);
3459 static int mark_printk(const char *fmt, ...)
3463 va_start(args, fmt);
3464 ret = trace_vprintk(0, fmt, args);
3470 tracing_mark_write(struct file *filp, const char __user *ubuf,
3471 size_t cnt, loff_t *fpos)
3475 if (tracing_disabled)
3478 if (cnt > TRACE_BUF_SIZE)
3479 cnt = TRACE_BUF_SIZE;
3481 buf = kmalloc(cnt + 2, GFP_KERNEL);
3485 if (copy_from_user(buf, ubuf, cnt)) {
3489 if (buf[cnt-1] != '\n') {
3495 cnt = mark_printk("%s", buf);
3502 static int tracing_clock_show(struct seq_file *m, void *v)
3506 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3508 "%s%s%s%s", i ? " " : "",
3509 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3510 i == trace_clock_id ? "]" : "");
3516 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3517 size_t cnt, loff_t *fpos)
3520 const char *clockstr;
3523 if (cnt >= sizeof(buf))
3526 if (copy_from_user(&buf, ubuf, cnt))
3531 clockstr = strstrip(buf);
3533 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3534 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3537 if (i == ARRAY_SIZE(trace_clocks))
3542 mutex_lock(&trace_types_lock);
3544 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3546 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3548 mutex_unlock(&trace_types_lock);
3555 static int tracing_clock_open(struct inode *inode, struct file *file)
3557 if (tracing_disabled)
3559 return single_open(file, tracing_clock_show, NULL);
3562 static const struct file_operations tracing_max_lat_fops = {
3563 .open = tracing_open_generic,
3564 .read = tracing_max_lat_read,
3565 .write = tracing_max_lat_write,
3566 .llseek = generic_file_llseek,
3569 static const struct file_operations tracing_ctrl_fops = {
3570 .open = tracing_open_generic,
3571 .read = tracing_ctrl_read,
3572 .write = tracing_ctrl_write,
3573 .llseek = generic_file_llseek,
3576 static const struct file_operations set_tracer_fops = {
3577 .open = tracing_open_generic,
3578 .read = tracing_set_trace_read,
3579 .write = tracing_set_trace_write,
3580 .llseek = generic_file_llseek,
3583 static const struct file_operations tracing_pipe_fops = {
3584 .open = tracing_open_pipe,
3585 .poll = tracing_poll_pipe,
3586 .read = tracing_read_pipe,
3587 .splice_read = tracing_splice_read_pipe,
3588 .release = tracing_release_pipe,
3589 .llseek = no_llseek,
3592 static const struct file_operations tracing_entries_fops = {
3593 .open = tracing_open_generic,
3594 .read = tracing_entries_read,
3595 .write = tracing_entries_write,
3596 .llseek = generic_file_llseek,
3599 static const struct file_operations tracing_mark_fops = {
3600 .open = tracing_open_generic,
3601 .write = tracing_mark_write,
3602 .llseek = generic_file_llseek,
3605 static const struct file_operations trace_clock_fops = {
3606 .open = tracing_clock_open,
3608 .llseek = seq_lseek,
3609 .release = single_release,
3610 .write = tracing_clock_write,
3613 struct ftrace_buffer_info {
3614 struct trace_array *tr;
3620 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3622 int cpu = (int)(long)inode->i_private;
3623 struct ftrace_buffer_info *info;
3625 if (tracing_disabled)
3628 info = kzalloc(sizeof(*info), GFP_KERNEL);
3632 info->tr = &global_trace;
3635 /* Force reading ring buffer for first read */
3636 info->read = (unsigned int)-1;
3638 filp->private_data = info;
3640 return nonseekable_open(inode, filp);
3644 tracing_buffers_read(struct file *filp, char __user *ubuf,
3645 size_t count, loff_t *ppos)
3647 struct ftrace_buffer_info *info = filp->private_data;
3655 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3659 /* Do we have previous read data to read? */
3660 if (info->read < PAGE_SIZE)
3665 trace_access_lock(info->cpu);
3666 ret = ring_buffer_read_page(info->tr->buffer,
3670 trace_access_unlock(info->cpu);
3675 size = PAGE_SIZE - info->read;
3679 ret = copy_to_user(ubuf, info->spare + info->read, size);
3690 static int tracing_buffers_release(struct inode *inode, struct file *file)
3692 struct ftrace_buffer_info *info = file->private_data;
3695 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3702 struct ring_buffer *buffer;
3707 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3708 struct pipe_buffer *buf)
3710 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3715 ring_buffer_free_read_page(ref->buffer, ref->page);
3720 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3721 struct pipe_buffer *buf)
3726 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3727 struct pipe_buffer *buf)
3729 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3734 /* Pipe buffer operations for a buffer. */
3735 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
3737 .map = generic_pipe_buf_map,
3738 .unmap = generic_pipe_buf_unmap,
3739 .confirm = generic_pipe_buf_confirm,
3740 .release = buffer_pipe_buf_release,
3741 .steal = buffer_pipe_buf_steal,
3742 .get = buffer_pipe_buf_get,
3746 * Callback from splice_to_pipe(), if we need to release some pages
3747 * at the end of the spd in case we error'ed out in filling the pipe.
3749 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3751 struct buffer_ref *ref =
3752 (struct buffer_ref *)spd->partial[i].private;
3757 ring_buffer_free_read_page(ref->buffer, ref->page);
3759 spd->partial[i].private = 0;
3763 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3764 struct pipe_inode_info *pipe, size_t len,
3767 struct ftrace_buffer_info *info = file->private_data;
3768 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3769 struct page *pages_def[PIPE_DEF_BUFFERS];
3770 struct splice_pipe_desc spd = {
3772 .partial = partial_def,
3774 .ops = &buffer_pipe_buf_ops,
3775 .spd_release = buffer_spd_release,
3777 struct buffer_ref *ref;
3778 int entries, size, i;
3781 if (splice_grow_spd(pipe, &spd))
3784 if (*ppos & (PAGE_SIZE - 1)) {
3785 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3790 if (len & (PAGE_SIZE - 1)) {
3791 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3792 if (len < PAGE_SIZE) {
3799 trace_access_lock(info->cpu);
3800 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3802 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
3806 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3811 ref->buffer = info->tr->buffer;
3812 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3818 r = ring_buffer_read_page(ref->buffer, &ref->page,
3821 ring_buffer_free_read_page(ref->buffer,
3828 * zero out any left over data, this is going to
3831 size = ring_buffer_page_len(ref->page);
3832 if (size < PAGE_SIZE)
3833 memset(ref->page + size, 0, PAGE_SIZE - size);
3835 page = virt_to_page(ref->page);
3837 spd.pages[i] = page;
3838 spd.partial[i].len = PAGE_SIZE;
3839 spd.partial[i].offset = 0;
3840 spd.partial[i].private = (unsigned long)ref;
3844 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3847 trace_access_unlock(info->cpu);
3850 /* did we read anything? */
3851 if (!spd.nr_pages) {
3852 if (flags & SPLICE_F_NONBLOCK)
3860 ret = splice_to_pipe(pipe, &spd);
3861 splice_shrink_spd(pipe, &spd);
3866 static const struct file_operations tracing_buffers_fops = {
3867 .open = tracing_buffers_open,
3868 .read = tracing_buffers_read,
3869 .release = tracing_buffers_release,
3870 .splice_read = tracing_buffers_splice_read,
3871 .llseek = no_llseek,
3875 tracing_stats_read(struct file *filp, char __user *ubuf,
3876 size_t count, loff_t *ppos)
3878 unsigned long cpu = (unsigned long)filp->private_data;
3879 struct trace_array *tr = &global_trace;
3880 struct trace_seq *s;
3883 s = kmalloc(sizeof(*s), GFP_KERNEL);
3889 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3890 trace_seq_printf(s, "entries: %ld\n", cnt);
3892 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3893 trace_seq_printf(s, "overrun: %ld\n", cnt);
3895 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3896 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3898 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3905 static const struct file_operations tracing_stats_fops = {
3906 .open = tracing_open_generic,
3907 .read = tracing_stats_read,
3908 .llseek = generic_file_llseek,
3911 #ifdef CONFIG_DYNAMIC_FTRACE
3913 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3919 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3920 size_t cnt, loff_t *ppos)
3922 static char ftrace_dyn_info_buffer[1024];
3923 static DEFINE_MUTEX(dyn_info_mutex);
3924 unsigned long *p = filp->private_data;
3925 char *buf = ftrace_dyn_info_buffer;
3926 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3929 mutex_lock(&dyn_info_mutex);
3930 r = sprintf(buf, "%ld ", *p);
3932 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3935 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3937 mutex_unlock(&dyn_info_mutex);
3942 static const struct file_operations tracing_dyn_info_fops = {
3943 .open = tracing_open_generic,
3944 .read = tracing_read_dyn_info,
3945 .llseek = generic_file_llseek,
3949 static struct dentry *d_tracer;
3951 struct dentry *tracing_init_dentry(void)
3958 if (!debugfs_initialized())
3961 d_tracer = debugfs_create_dir("tracing", NULL);
3963 if (!d_tracer && !once) {
3965 pr_warning("Could not create debugfs directory 'tracing'\n");
3972 static struct dentry *d_percpu;
3974 struct dentry *tracing_dentry_percpu(void)
3977 struct dentry *d_tracer;
3982 d_tracer = tracing_init_dentry();
3987 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3989 if (!d_percpu && !once) {
3991 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3998 static void tracing_init_debugfs_percpu(long cpu)
4000 struct dentry *d_percpu = tracing_dentry_percpu();
4001 struct dentry *d_cpu;
4002 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
4005 if (cpu > 999 || cpu < 0)
4008 sprintf(cpu_dir, "cpu%ld", cpu);
4009 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4011 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4015 /* per cpu trace_pipe */
4016 trace_create_file("trace_pipe", 0444, d_cpu,
4017 (void *) cpu, &tracing_pipe_fops);
4020 trace_create_file("trace", 0644, d_cpu,
4021 (void *) cpu, &tracing_fops);
4023 trace_create_file("trace_pipe_raw", 0444, d_cpu,
4024 (void *) cpu, &tracing_buffers_fops);
4026 trace_create_file("stats", 0444, d_cpu,
4027 (void *) cpu, &tracing_stats_fops);
4030 #ifdef CONFIG_FTRACE_SELFTEST
4031 /* Let selftest have access to static functions in this file */
4032 #include "trace_selftest.c"
4035 struct trace_option_dentry {
4036 struct tracer_opt *opt;
4037 struct tracer_flags *flags;
4038 struct dentry *entry;
4042 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4045 struct trace_option_dentry *topt = filp->private_data;
4048 if (topt->flags->val & topt->opt->bit)
4053 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4057 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4060 struct trace_option_dentry *topt = filp->private_data;
4065 if (cnt >= sizeof(buf))
4068 if (copy_from_user(&buf, ubuf, cnt))
4073 ret = strict_strtoul(buf, 10, &val);
4077 if (val != 0 && val != 1)
4080 if (!!(topt->flags->val & topt->opt->bit) != val) {
4081 mutex_lock(&trace_types_lock);
4082 ret = __set_tracer_option(current_trace, topt->flags,
4084 mutex_unlock(&trace_types_lock);
4095 static const struct file_operations trace_options_fops = {
4096 .open = tracing_open_generic,
4097 .read = trace_options_read,
4098 .write = trace_options_write,
4099 .llseek = generic_file_llseek,
4103 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4106 long index = (long)filp->private_data;
4109 if (trace_flags & (1 << index))
4114 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4118 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4121 long index = (long)filp->private_data;
4126 if (cnt >= sizeof(buf))
4129 if (copy_from_user(&buf, ubuf, cnt))
4134 ret = strict_strtoul(buf, 10, &val);
4138 if (val != 0 && val != 1)
4140 set_tracer_flags(1 << index, val);
4147 static const struct file_operations trace_options_core_fops = {
4148 .open = tracing_open_generic,
4149 .read = trace_options_core_read,
4150 .write = trace_options_core_write,
4151 .llseek = generic_file_llseek,
4154 struct dentry *trace_create_file(const char *name,
4156 struct dentry *parent,
4158 const struct file_operations *fops)
4162 ret = debugfs_create_file(name, mode, parent, data, fops);
4164 pr_warning("Could not create debugfs '%s' entry\n", name);
4170 static struct dentry *trace_options_init_dentry(void)
4172 struct dentry *d_tracer;
4173 static struct dentry *t_options;
4178 d_tracer = tracing_init_dentry();
4182 t_options = debugfs_create_dir("options", d_tracer);
4184 pr_warning("Could not create debugfs directory 'options'\n");
4192 create_trace_option_file(struct trace_option_dentry *topt,
4193 struct tracer_flags *flags,
4194 struct tracer_opt *opt)
4196 struct dentry *t_options;
4198 t_options = trace_options_init_dentry();
4202 topt->flags = flags;
4205 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4206 &trace_options_fops);
4210 static struct trace_option_dentry *
4211 create_trace_option_files(struct tracer *tracer)
4213 struct trace_option_dentry *topts;
4214 struct tracer_flags *flags;
4215 struct tracer_opt *opts;
4221 flags = tracer->flags;
4223 if (!flags || !flags->opts)
4228 for (cnt = 0; opts[cnt].name; cnt++)
4231 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4235 for (cnt = 0; opts[cnt].name; cnt++)
4236 create_trace_option_file(&topts[cnt], flags,
4243 destroy_trace_option_files(struct trace_option_dentry *topts)
4250 for (cnt = 0; topts[cnt].opt; cnt++) {
4251 if (topts[cnt].entry)
4252 debugfs_remove(topts[cnt].entry);
4258 static struct dentry *
4259 create_trace_option_core_file(const char *option, long index)
4261 struct dentry *t_options;
4263 t_options = trace_options_init_dentry();
4267 return trace_create_file(option, 0644, t_options, (void *)index,
4268 &trace_options_core_fops);
4271 static __init void create_trace_options_dir(void)
4273 struct dentry *t_options;
4276 t_options = trace_options_init_dentry();
4280 for (i = 0; trace_options[i]; i++)
4281 create_trace_option_core_file(trace_options[i], i);
4284 static __init int tracer_init_debugfs(void)
4286 struct dentry *d_tracer;
4289 trace_access_lock_init();
4291 d_tracer = tracing_init_dentry();
4293 trace_create_file("tracing_enabled", 0644, d_tracer,
4294 &global_trace, &tracing_ctrl_fops);
4296 trace_create_file("trace_options", 0644, d_tracer,
4297 NULL, &tracing_iter_fops);
4299 trace_create_file("tracing_cpumask", 0644, d_tracer,
4300 NULL, &tracing_cpumask_fops);
4302 trace_create_file("trace", 0644, d_tracer,
4303 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4305 trace_create_file("available_tracers", 0444, d_tracer,
4306 &global_trace, &show_traces_fops);
4308 trace_create_file("current_tracer", 0644, d_tracer,
4309 &global_trace, &set_tracer_fops);
4311 #ifdef CONFIG_TRACER_MAX_TRACE
4312 trace_create_file("tracing_max_latency", 0644, d_tracer,
4313 &tracing_max_latency, &tracing_max_lat_fops);
4316 trace_create_file("tracing_thresh", 0644, d_tracer,
4317 &tracing_thresh, &tracing_max_lat_fops);
4319 trace_create_file("README", 0444, d_tracer,
4320 NULL, &tracing_readme_fops);
4322 trace_create_file("trace_pipe", 0444, d_tracer,
4323 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4325 trace_create_file("buffer_size_kb", 0644, d_tracer,
4326 &global_trace, &tracing_entries_fops);
4328 trace_create_file("trace_marker", 0220, d_tracer,
4329 NULL, &tracing_mark_fops);
4331 trace_create_file("saved_cmdlines", 0444, d_tracer,
4332 NULL, &tracing_saved_cmdlines_fops);
4334 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4337 #ifdef CONFIG_DYNAMIC_FTRACE
4338 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4339 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4342 create_trace_options_dir();
4344 for_each_tracing_cpu(cpu)
4345 tracing_init_debugfs_percpu(cpu);
4350 static int trace_panic_handler(struct notifier_block *this,
4351 unsigned long event, void *unused)
4353 if (ftrace_dump_on_oops)
4354 ftrace_dump(ftrace_dump_on_oops);
4358 static struct notifier_block trace_panic_notifier = {
4359 .notifier_call = trace_panic_handler,
4361 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4364 static int trace_die_handler(struct notifier_block *self,
4370 if (ftrace_dump_on_oops)
4371 ftrace_dump(ftrace_dump_on_oops);
4379 static struct notifier_block trace_die_notifier = {
4380 .notifier_call = trace_die_handler,
4385 * printk is set to max of 1024, we really don't need it that big.
4386 * Nothing should be printing 1000 characters anyway.
4388 #define TRACE_MAX_PRINT 1000
4391 * Define here KERN_TRACE so that we have one place to modify
4392 * it if we decide to change what log level the ftrace dump
4395 #define KERN_TRACE KERN_EMERG
4398 trace_printk_seq(struct trace_seq *s)
4400 /* Probably should print a warning here. */
4404 /* should be zero ended, but we are paranoid. */
4405 s->buffer[s->len] = 0;
4407 printk(KERN_TRACE "%s", s->buffer);
4412 void trace_init_global_iter(struct trace_iterator *iter)
4414 iter->tr = &global_trace;
4415 iter->trace = current_trace;
4416 iter->cpu_file = TRACE_PIPE_ALL_CPU;
4420 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4422 static arch_spinlock_t ftrace_dump_lock =
4423 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4424 /* use static because iter can be a bit big for the stack */
4425 static struct trace_iterator iter;
4426 unsigned int old_userobj;
4427 static int dump_ran;
4428 unsigned long flags;
4432 local_irq_save(flags);
4433 arch_spin_lock(&ftrace_dump_lock);
4441 if (disable_tracing)
4444 trace_init_global_iter(&iter);
4446 for_each_tracing_cpu(cpu) {
4447 atomic_inc(&iter.tr->data[cpu]->disabled);
4450 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4452 /* don't look at user memory in panic mode */
4453 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4455 /* Simulate the iterator */
4456 iter.tr = &global_trace;
4457 iter.trace = current_trace;
4459 switch (oops_dump_mode) {
4461 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4464 iter.cpu_file = raw_smp_processor_id();
4469 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
4470 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4473 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4476 * We need to stop all tracing on all CPUS to read the
4477 * the next buffer. This is a bit expensive, but is
4478 * not done often. We fill all what we can read,
4479 * and then release the locks again.
4482 while (!trace_empty(&iter)) {
4485 printk(KERN_TRACE "---------------------------------\n");
4489 /* reset all but tr, trace, and overruns */
4490 memset(&iter.seq, 0,
4491 sizeof(struct trace_iterator) -
4492 offsetof(struct trace_iterator, seq));
4493 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4496 if (trace_find_next_entry_inc(&iter) != NULL) {
4499 ret = print_trace_line(&iter);
4500 if (ret != TRACE_TYPE_NO_CONSUME)
4501 trace_consume(&iter);
4504 trace_printk_seq(&iter.seq);
4508 printk(KERN_TRACE " (ftrace buffer empty)\n");
4510 printk(KERN_TRACE "---------------------------------\n");
4513 /* Re-enable tracing if requested */
4514 if (!disable_tracing) {
4515 trace_flags |= old_userobj;
4517 for_each_tracing_cpu(cpu) {
4518 atomic_dec(&iter.tr->data[cpu]->disabled);
4524 arch_spin_unlock(&ftrace_dump_lock);
4525 local_irq_restore(flags);
4528 /* By default: disable tracing after the dump */
4529 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
4531 __ftrace_dump(true, oops_dump_mode);
4534 __init static int tracer_alloc_buffers(void)
4540 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4543 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4544 goto out_free_buffer_mask;
4546 /* To save memory, keep the ring buffer size to its minimum */
4547 if (ring_buffer_expanded)
4548 ring_buf_size = trace_buf_size;
4552 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4553 cpumask_copy(tracing_cpumask, cpu_all_mask);
4555 /* TODO: make the number of buffers hot pluggable with CPUS */
4556 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4557 TRACE_BUFFER_FLAGS);
4558 if (!global_trace.buffer) {
4559 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4561 goto out_free_cpumask;
4563 global_trace.entries = ring_buffer_size(global_trace.buffer);
4566 #ifdef CONFIG_TRACER_MAX_TRACE
4567 max_tr.buffer = ring_buffer_alloc(1, TRACE_BUFFER_FLAGS);
4568 if (!max_tr.buffer) {
4569 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4571 ring_buffer_free(global_trace.buffer);
4572 goto out_free_cpumask;
4577 /* Allocate the first page for all buffers */
4578 for_each_tracing_cpu(i) {
4579 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4580 max_tr.data[i] = &per_cpu(max_tr_data, i);
4583 trace_init_cmdlines();
4585 register_tracer(&nop_trace);
4586 current_trace = &nop_trace;
4587 /* All seems OK, enable tracing */
4588 tracing_disabled = 0;
4590 atomic_notifier_chain_register(&panic_notifier_list,
4591 &trace_panic_notifier);
4593 register_die_notifier(&trace_die_notifier);
4598 free_cpumask_var(tracing_cpumask);
4599 out_free_buffer_mask:
4600 free_cpumask_var(tracing_buffer_mask);
4605 __init static int clear_boot_tracer(void)
4608 * The default tracer at boot buffer is an init section.
4609 * This function is called in lateinit. If we did not
4610 * find the boot tracer, then clear it out, to prevent
4611 * later registration from accessing the buffer that is
4612 * about to be freed.
4614 if (!default_bootup_tracer)
4617 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4618 default_bootup_tracer);
4619 default_bootup_tracer = NULL;
4624 early_initcall(tracer_alloc_buffers);
4625 fs_initcall(tracer_init_debugfs);
4626 late_initcall(clear_boot_tracer);