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 static cpumask_var_t __read_mostly tracing_buffer_mask;
106 #define for_each_tracing_cpu(cpu) \
107 for_each_cpu(cpu, tracing_buffer_mask)
110 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
112 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
113 * is set, then ftrace_dump is called. This will output the contents
114 * of the ftrace buffers to the console. This is very useful for
115 * capturing traces that lead to crashes and outputing it to a
118 * It is default off, but you can enable it with either specifying
119 * "ftrace_dump_on_oops" in the kernel command line, or setting
120 * /proc/sys/kernel/ftrace_dump_on_oops
121 * Set 1 if you want to dump buffers of all CPUs
122 * Set 2 if you want to dump the buffer of the CPU that triggered oops
125 enum ftrace_dump_mode ftrace_dump_on_oops;
127 static int tracing_set_tracer(const char *buf);
129 #define MAX_TRACER_SIZE 100
130 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
131 static char *default_bootup_tracer;
133 static int __init set_cmdline_ftrace(char *str)
135 strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
136 default_bootup_tracer = bootup_tracer_buf;
137 /* We are using ftrace early, expand it */
138 ring_buffer_expanded = 1;
141 __setup("ftrace=", set_cmdline_ftrace);
143 static int __init set_ftrace_dump_on_oops(char *str)
145 if (*str++ != '=' || !*str) {
146 ftrace_dump_on_oops = DUMP_ALL;
150 if (!strcmp("orig_cpu", str)) {
151 ftrace_dump_on_oops = DUMP_ORIG;
157 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
159 unsigned long long ns2usecs(cycle_t nsec)
167 * The global_trace is the descriptor that holds the tracing
168 * buffers for the live tracing. For each CPU, it contains
169 * a link list of pages that will store trace entries. The
170 * page descriptor of the pages in the memory is used to hold
171 * the link list by linking the lru item in the page descriptor
172 * to each of the pages in the buffer per CPU.
174 * For each active CPU there is a data field that holds the
175 * pages for the buffer for that CPU. Each CPU has the same number
176 * of pages allocated for its buffer.
178 static struct trace_array global_trace;
180 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
182 int filter_current_check_discard(struct ring_buffer *buffer,
183 struct ftrace_event_call *call, void *rec,
184 struct ring_buffer_event *event)
186 return filter_check_discard(call, rec, buffer, event);
188 EXPORT_SYMBOL_GPL(filter_current_check_discard);
190 cycle_t ftrace_now(int cpu)
194 /* Early boot up does not have a buffer yet */
195 if (!global_trace.buffer)
196 return trace_clock_local();
198 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
199 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
205 * The max_tr is used to snapshot the global_trace when a maximum
206 * latency is reached. Some tracers will use this to store a maximum
207 * trace while it continues examining live traces.
209 * The buffers for the max_tr are set up the same as the global_trace.
210 * When a snapshot is taken, the link list of the max_tr is swapped
211 * with the link list of the global_trace and the buffers are reset for
212 * the global_trace so the tracing can continue.
214 static struct trace_array max_tr;
216 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
218 /* tracer_enabled is used to toggle activation of a tracer */
219 static int tracer_enabled = 1;
222 * tracing_is_enabled - return tracer_enabled status
224 * This function is used by other tracers to know the status
225 * of the tracer_enabled flag. Tracers may use this function
226 * to know if it should enable their features when starting
227 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
229 int tracing_is_enabled(void)
231 return tracer_enabled;
235 * trace_buf_size is the size in bytes that is allocated
236 * for a buffer. Note, the number of bytes is always rounded
239 * This number is purposely set to a low number of 16384.
240 * If the dump on oops happens, it will be much appreciated
241 * to not have to wait for all that output. Anyway this can be
242 * boot time and run time configurable.
244 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
246 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
248 /* trace_types holds a link list of available tracers. */
249 static struct tracer *trace_types __read_mostly;
251 /* current_trace points to the tracer that is currently active */
252 static struct tracer *current_trace __read_mostly;
255 * trace_types_lock is used to protect the trace_types list.
257 static DEFINE_MUTEX(trace_types_lock);
260 * serialize the access of the ring buffer
262 * ring buffer serializes readers, but it is low level protection.
263 * The validity of the events (which returns by ring_buffer_peek() ..etc)
264 * are not protected by ring buffer.
266 * The content of events may become garbage if we allow other process consumes
267 * these events concurrently:
268 * A) the page of the consumed events may become a normal page
269 * (not reader page) in ring buffer, and this page will be rewrited
270 * by events producer.
271 * B) The page of the consumed events may become a page for splice_read,
272 * and this page will be returned to system.
274 * These primitives allow multi process access to different cpu ring buffer
277 * These primitives don't distinguish read-only and read-consume access.
278 * Multi read-only access are also serialized.
282 static DECLARE_RWSEM(all_cpu_access_lock);
283 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
285 static inline void trace_access_lock(int cpu)
287 if (cpu == TRACE_PIPE_ALL_CPU) {
288 /* gain it for accessing the whole ring buffer. */
289 down_write(&all_cpu_access_lock);
291 /* gain it for accessing a cpu ring buffer. */
293 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
294 down_read(&all_cpu_access_lock);
296 /* Secondly block other access to this @cpu ring buffer. */
297 mutex_lock(&per_cpu(cpu_access_lock, cpu));
301 static inline void trace_access_unlock(int cpu)
303 if (cpu == TRACE_PIPE_ALL_CPU) {
304 up_write(&all_cpu_access_lock);
306 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
307 up_read(&all_cpu_access_lock);
311 static inline void trace_access_lock_init(void)
315 for_each_possible_cpu(cpu)
316 mutex_init(&per_cpu(cpu_access_lock, cpu));
321 static DEFINE_MUTEX(access_lock);
323 static inline void trace_access_lock(int cpu)
326 mutex_lock(&access_lock);
329 static inline void trace_access_unlock(int cpu)
332 mutex_unlock(&access_lock);
335 static inline void trace_access_lock_init(void)
341 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
342 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
344 /* trace_flags holds trace_options default values */
345 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
346 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
347 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD;
349 static int trace_stop_count;
350 static DEFINE_SPINLOCK(tracing_start_lock);
353 * trace_wake_up - wake up tasks waiting for trace input
355 * Simply wakes up any task that is blocked on the trace_wait
356 * queue. These is used with trace_poll for tasks polling the trace.
358 void trace_wake_up(void)
362 if (trace_flags & TRACE_ITER_BLOCK)
365 * The runqueue_is_locked() can fail, but this is the best we
369 if (!runqueue_is_locked(cpu))
370 wake_up(&trace_wait);
374 static int __init set_buf_size(char *str)
376 unsigned long buf_size;
380 buf_size = memparse(str, &str);
381 /* nr_entries can not be zero */
384 trace_buf_size = buf_size;
387 __setup("trace_buf_size=", set_buf_size);
389 static int __init set_tracing_thresh(char *str)
391 unsigned long threshhold;
396 ret = strict_strtoul(str, 0, &threshhold);
399 tracing_thresh = threshhold * 1000;
402 __setup("tracing_thresh=", set_tracing_thresh);
404 unsigned long nsecs_to_usecs(unsigned long nsecs)
409 /* These must match the bit postions in trace_iterator_flags */
410 static const char *trace_options[] = {
439 { trace_clock_local, "local" },
440 { trace_clock_global, "global" },
446 * trace_parser_get_init - gets the buffer for trace parser
448 int trace_parser_get_init(struct trace_parser *parser, int size)
450 memset(parser, 0, sizeof(*parser));
452 parser->buffer = kmalloc(size, GFP_KERNEL);
461 * trace_parser_put - frees the buffer for trace parser
463 void trace_parser_put(struct trace_parser *parser)
465 kfree(parser->buffer);
469 * trace_get_user - reads the user input string separated by space
470 * (matched by isspace(ch))
472 * For each string found the 'struct trace_parser' is updated,
473 * and the function returns.
475 * Returns number of bytes read.
477 * See kernel/trace/trace.h for 'struct trace_parser' details.
479 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
480 size_t cnt, loff_t *ppos)
487 trace_parser_clear(parser);
489 ret = get_user(ch, ubuf++);
497 * The parser is not finished with the last write,
498 * continue reading the user input without skipping spaces.
501 /* skip white space */
502 while (cnt && isspace(ch)) {
503 ret = get_user(ch, ubuf++);
510 /* only spaces were written */
520 /* read the non-space input */
521 while (cnt && !isspace(ch)) {
522 if (parser->idx < parser->size - 1)
523 parser->buffer[parser->idx++] = ch;
528 ret = get_user(ch, ubuf++);
535 /* We either got finished input or we have to wait for another call. */
537 parser->buffer[parser->idx] = 0;
538 parser->cont = false;
541 parser->buffer[parser->idx++] = ch;
551 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
559 if (s->len <= s->readpos)
562 len = s->len - s->readpos;
565 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
575 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
580 if (s->len <= s->readpos)
583 len = s->len - s->readpos;
586 ret = memcpy(buf, s->buffer + s->readpos, cnt);
595 * ftrace_max_lock is used to protect the swapping of buffers
596 * when taking a max snapshot. The buffers themselves are
597 * protected by per_cpu spinlocks. But the action of the swap
598 * needs its own lock.
600 * This is defined as a arch_spinlock_t in order to help
601 * with performance when lockdep debugging is enabled.
603 * It is also used in other places outside the update_max_tr
604 * so it needs to be defined outside of the
605 * CONFIG_TRACER_MAX_TRACE.
607 static arch_spinlock_t ftrace_max_lock =
608 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
610 unsigned long __read_mostly tracing_thresh;
612 #ifdef CONFIG_TRACER_MAX_TRACE
613 unsigned long __read_mostly tracing_max_latency;
616 * Copy the new maximum trace into the separate maximum-trace
617 * structure. (this way the maximum trace is permanently saved,
618 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
621 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
623 struct trace_array_cpu *data = tr->data[cpu];
624 struct trace_array_cpu *max_data;
627 max_tr.time_start = data->preempt_timestamp;
629 max_data = max_tr.data[cpu];
630 max_data->saved_latency = tracing_max_latency;
631 max_data->critical_start = data->critical_start;
632 max_data->critical_end = data->critical_end;
634 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
635 max_data->pid = tsk->pid;
636 max_data->uid = task_uid(tsk);
637 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
638 max_data->policy = tsk->policy;
639 max_data->rt_priority = tsk->rt_priority;
641 /* record this tasks comm */
642 tracing_record_cmdline(tsk);
646 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
648 * @tsk: the task with the latency
649 * @cpu: The cpu that initiated the trace.
651 * Flip the buffers between the @tr and the max_tr and record information
652 * about which task was the cause of this latency.
655 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
657 struct ring_buffer *buf = tr->buffer;
659 if (trace_stop_count)
662 WARN_ON_ONCE(!irqs_disabled());
663 if (!current_trace->use_max_tr) {
667 arch_spin_lock(&ftrace_max_lock);
669 tr->buffer = max_tr.buffer;
672 __update_max_tr(tr, tsk, cpu);
673 arch_spin_unlock(&ftrace_max_lock);
677 * update_max_tr_single - only copy one trace over, and reset the rest
679 * @tsk - task with the latency
680 * @cpu - the cpu of the buffer to copy.
682 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
685 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
689 if (trace_stop_count)
692 WARN_ON_ONCE(!irqs_disabled());
693 if (!current_trace->use_max_tr) {
698 arch_spin_lock(&ftrace_max_lock);
700 ftrace_disable_cpu();
702 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
706 * We failed to swap the buffer due to a commit taking
707 * place on this CPU. We fail to record, but we reset
708 * the max trace buffer (no one writes directly to it)
709 * and flag that it failed.
711 trace_array_printk(&max_tr, _THIS_IP_,
712 "Failed to swap buffers due to commit in progress\n");
717 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
719 __update_max_tr(tr, tsk, cpu);
720 arch_spin_unlock(&ftrace_max_lock);
722 #endif /* CONFIG_TRACER_MAX_TRACE */
725 * register_tracer - register a tracer with the ftrace system.
726 * @type - the plugin for the tracer
728 * Register a new plugin tracer.
730 int register_tracer(struct tracer *type)
731 __releases(kernel_lock)
732 __acquires(kernel_lock)
738 pr_info("Tracer must have a name\n");
742 if (strlen(type->name) >= MAX_TRACER_SIZE) {
743 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
748 * When this gets called we hold the BKL which means that
749 * preemption is disabled. Various trace selftests however
750 * need to disable and enable preemption for successful tests.
751 * So we drop the BKL here and grab it after the tests again.
754 mutex_lock(&trace_types_lock);
756 tracing_selftest_running = true;
758 for (t = trace_types; t; t = t->next) {
759 if (strcmp(type->name, t->name) == 0) {
761 pr_info("Tracer %s already registered\n",
769 type->set_flag = &dummy_set_flag;
771 type->flags = &dummy_tracer_flags;
773 if (!type->flags->opts)
774 type->flags->opts = dummy_tracer_opt;
775 if (!type->wait_pipe)
776 type->wait_pipe = default_wait_pipe;
779 #ifdef CONFIG_FTRACE_STARTUP_TEST
780 if (type->selftest && !tracing_selftest_disabled) {
781 struct tracer *saved_tracer = current_trace;
782 struct trace_array *tr = &global_trace;
785 * Run a selftest on this tracer.
786 * Here we reset the trace buffer, and set the current
787 * tracer to be this tracer. The tracer can then run some
788 * internal tracing to verify that everything is in order.
789 * If we fail, we do not register this tracer.
791 tracing_reset_online_cpus(tr);
793 current_trace = type;
794 /* the test is responsible for initializing and enabling */
795 pr_info("Testing tracer %s: ", type->name);
796 ret = type->selftest(type, tr);
797 /* the test is responsible for resetting too */
798 current_trace = saved_tracer;
800 printk(KERN_CONT "FAILED!\n");
803 /* Only reset on passing, to avoid touching corrupted buffers */
804 tracing_reset_online_cpus(tr);
806 printk(KERN_CONT "PASSED\n");
810 type->next = trace_types;
814 tracing_selftest_running = false;
815 mutex_unlock(&trace_types_lock);
817 if (ret || !default_bootup_tracer)
820 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
823 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
824 /* Do we want this tracer to start on bootup? */
825 tracing_set_tracer(type->name);
826 default_bootup_tracer = NULL;
827 /* disable other selftests, since this will break it. */
828 tracing_selftest_disabled = 1;
829 #ifdef CONFIG_FTRACE_STARTUP_TEST
830 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
839 void unregister_tracer(struct tracer *type)
843 mutex_lock(&trace_types_lock);
844 for (t = &trace_types; *t; t = &(*t)->next) {
848 pr_info("Tracer %s not registered\n", type->name);
854 if (type == current_trace && tracer_enabled) {
857 if (current_trace->stop)
858 current_trace->stop(&global_trace);
859 current_trace = &nop_trace;
862 mutex_unlock(&trace_types_lock);
865 static void __tracing_reset(struct ring_buffer *buffer, int cpu)
867 ftrace_disable_cpu();
868 ring_buffer_reset_cpu(buffer, cpu);
872 void tracing_reset(struct trace_array *tr, int cpu)
874 struct ring_buffer *buffer = tr->buffer;
876 ring_buffer_record_disable(buffer);
878 /* Make sure all commits have finished */
880 __tracing_reset(buffer, cpu);
882 ring_buffer_record_enable(buffer);
885 void tracing_reset_online_cpus(struct trace_array *tr)
887 struct ring_buffer *buffer = tr->buffer;
890 ring_buffer_record_disable(buffer);
892 /* Make sure all commits have finished */
895 tr->time_start = ftrace_now(tr->cpu);
897 for_each_online_cpu(cpu)
898 __tracing_reset(buffer, cpu);
900 ring_buffer_record_enable(buffer);
903 void tracing_reset_current(int cpu)
905 tracing_reset(&global_trace, cpu);
908 void tracing_reset_current_online_cpus(void)
910 tracing_reset_online_cpus(&global_trace);
913 #define SAVED_CMDLINES 128
914 #define NO_CMDLINE_MAP UINT_MAX
915 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
916 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
917 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
918 static int cmdline_idx;
919 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
921 /* temporary disable recording */
922 static atomic_t trace_record_cmdline_disabled __read_mostly;
924 static void trace_init_cmdlines(void)
926 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
927 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
931 int is_tracing_stopped(void)
933 return trace_stop_count;
937 * ftrace_off_permanent - disable all ftrace code permanently
939 * This should only be called when a serious anomally has
940 * been detected. This will turn off the function tracing,
941 * ring buffers, and other tracing utilites. It takes no
942 * locks and can be called from any context.
944 void ftrace_off_permanent(void)
946 tracing_disabled = 1;
948 tracing_off_permanent();
952 * tracing_start - quick start of the tracer
954 * If tracing is enabled but was stopped by tracing_stop,
955 * this will start the tracer back up.
957 void tracing_start(void)
959 struct ring_buffer *buffer;
962 if (tracing_disabled)
965 spin_lock_irqsave(&tracing_start_lock, flags);
966 if (--trace_stop_count) {
967 if (trace_stop_count < 0) {
968 /* Someone screwed up their debugging */
970 trace_stop_count = 0;
975 /* Prevent the buffers from switching */
976 arch_spin_lock(&ftrace_max_lock);
978 buffer = global_trace.buffer;
980 ring_buffer_record_enable(buffer);
982 buffer = max_tr.buffer;
984 ring_buffer_record_enable(buffer);
986 arch_spin_unlock(&ftrace_max_lock);
990 spin_unlock_irqrestore(&tracing_start_lock, flags);
994 * tracing_stop - quick stop of the tracer
996 * Light weight way to stop tracing. Use in conjunction with
999 void tracing_stop(void)
1001 struct ring_buffer *buffer;
1002 unsigned long flags;
1005 spin_lock_irqsave(&tracing_start_lock, flags);
1006 if (trace_stop_count++)
1009 /* Prevent the buffers from switching */
1010 arch_spin_lock(&ftrace_max_lock);
1012 buffer = global_trace.buffer;
1014 ring_buffer_record_disable(buffer);
1016 buffer = max_tr.buffer;
1018 ring_buffer_record_disable(buffer);
1020 arch_spin_unlock(&ftrace_max_lock);
1023 spin_unlock_irqrestore(&tracing_start_lock, flags);
1026 void trace_stop_cmdline_recording(void);
1028 static void trace_save_cmdline(struct task_struct *tsk)
1032 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1036 * It's not the end of the world if we don't get
1037 * the lock, but we also don't want to spin
1038 * nor do we want to disable interrupts,
1039 * so if we miss here, then better luck next time.
1041 if (!arch_spin_trylock(&trace_cmdline_lock))
1044 idx = map_pid_to_cmdline[tsk->pid];
1045 if (idx == NO_CMDLINE_MAP) {
1046 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1049 * Check whether the cmdline buffer at idx has a pid
1050 * mapped. We are going to overwrite that entry so we
1051 * need to clear the map_pid_to_cmdline. Otherwise we
1052 * would read the new comm for the old pid.
1054 pid = map_cmdline_to_pid[idx];
1055 if (pid != NO_CMDLINE_MAP)
1056 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1058 map_cmdline_to_pid[idx] = tsk->pid;
1059 map_pid_to_cmdline[tsk->pid] = idx;
1064 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1066 arch_spin_unlock(&trace_cmdline_lock);
1069 void trace_find_cmdline(int pid, char comm[])
1074 strcpy(comm, "<idle>");
1078 if (WARN_ON_ONCE(pid < 0)) {
1079 strcpy(comm, "<XXX>");
1083 if (pid > PID_MAX_DEFAULT) {
1084 strcpy(comm, "<...>");
1089 arch_spin_lock(&trace_cmdline_lock);
1090 map = map_pid_to_cmdline[pid];
1091 if (map != NO_CMDLINE_MAP)
1092 strcpy(comm, saved_cmdlines[map]);
1094 strcpy(comm, "<...>");
1096 arch_spin_unlock(&trace_cmdline_lock);
1100 void tracing_record_cmdline(struct task_struct *tsk)
1102 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1106 trace_save_cmdline(tsk);
1110 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1113 struct task_struct *tsk = current;
1115 entry->preempt_count = pc & 0xff;
1116 entry->pid = (tsk) ? tsk->pid : 0;
1117 entry->lock_depth = (tsk) ? tsk->lock_depth : 0;
1119 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1120 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1122 TRACE_FLAG_IRQS_NOSUPPORT |
1124 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1125 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1126 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1128 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1130 struct ring_buffer_event *
1131 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1134 unsigned long flags, int pc)
1136 struct ring_buffer_event *event;
1138 event = ring_buffer_lock_reserve(buffer, len);
1139 if (event != NULL) {
1140 struct trace_entry *ent = ring_buffer_event_data(event);
1142 tracing_generic_entry_update(ent, flags, pc);
1150 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1151 struct ring_buffer_event *event,
1152 unsigned long flags, int pc,
1155 ring_buffer_unlock_commit(buffer, event);
1157 ftrace_trace_stack(buffer, flags, 6, pc);
1158 ftrace_trace_userstack(buffer, flags, pc);
1164 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1165 struct ring_buffer_event *event,
1166 unsigned long flags, int pc)
1168 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1171 struct ring_buffer_event *
1172 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1173 int type, unsigned long len,
1174 unsigned long flags, int pc)
1176 *current_rb = global_trace.buffer;
1177 return trace_buffer_lock_reserve(*current_rb,
1178 type, len, flags, pc);
1180 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1182 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1183 struct ring_buffer_event *event,
1184 unsigned long flags, int pc)
1186 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1188 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1190 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1191 struct ring_buffer_event *event,
1192 unsigned long flags, int pc)
1194 __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1196 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1198 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1199 struct ring_buffer_event *event)
1201 ring_buffer_discard_commit(buffer, event);
1203 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1206 trace_function(struct trace_array *tr,
1207 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1210 struct ftrace_event_call *call = &event_function;
1211 struct ring_buffer *buffer = tr->buffer;
1212 struct ring_buffer_event *event;
1213 struct ftrace_entry *entry;
1215 /* If we are reading the ring buffer, don't trace */
1216 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1219 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1223 entry = ring_buffer_event_data(event);
1225 entry->parent_ip = parent_ip;
1227 if (!filter_check_discard(call, entry, buffer, event))
1228 ring_buffer_unlock_commit(buffer, event);
1232 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1233 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1236 if (likely(!atomic_read(&data->disabled)))
1237 trace_function(tr, ip, parent_ip, flags, pc);
1240 #ifdef CONFIG_STACKTRACE
1241 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1242 unsigned long flags,
1245 struct ftrace_event_call *call = &event_kernel_stack;
1246 struct ring_buffer_event *event;
1247 struct stack_entry *entry;
1248 struct stack_trace trace;
1250 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1251 sizeof(*entry), flags, pc);
1254 entry = ring_buffer_event_data(event);
1255 memset(&entry->caller, 0, sizeof(entry->caller));
1257 trace.nr_entries = 0;
1258 trace.max_entries = FTRACE_STACK_ENTRIES;
1260 trace.entries = entry->caller;
1262 save_stack_trace(&trace);
1263 if (!filter_check_discard(call, entry, buffer, event))
1264 ring_buffer_unlock_commit(buffer, event);
1267 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1270 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1273 __ftrace_trace_stack(buffer, flags, skip, pc);
1276 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1279 __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1283 * trace_dump_stack - record a stack back trace in the trace buffer
1285 void trace_dump_stack(void)
1287 unsigned long flags;
1289 if (tracing_disabled || tracing_selftest_running)
1292 local_save_flags(flags);
1294 /* skipping 3 traces, seems to get us at the caller of this function */
1295 __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count());
1299 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1301 struct ftrace_event_call *call = &event_user_stack;
1302 struct ring_buffer_event *event;
1303 struct userstack_entry *entry;
1304 struct stack_trace trace;
1306 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1310 * NMIs can not handle page faults, even with fix ups.
1311 * The save user stack can (and often does) fault.
1313 if (unlikely(in_nmi()))
1316 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1317 sizeof(*entry), flags, pc);
1320 entry = ring_buffer_event_data(event);
1322 entry->tgid = current->tgid;
1323 memset(&entry->caller, 0, sizeof(entry->caller));
1325 trace.nr_entries = 0;
1326 trace.max_entries = FTRACE_STACK_ENTRIES;
1328 trace.entries = entry->caller;
1330 save_stack_trace_user(&trace);
1331 if (!filter_check_discard(call, entry, buffer, event))
1332 ring_buffer_unlock_commit(buffer, event);
1336 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1338 ftrace_trace_userstack(tr, flags, preempt_count());
1342 #endif /* CONFIG_STACKTRACE */
1345 * trace_vbprintk - write binary msg to tracing buffer
1348 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1350 static arch_spinlock_t trace_buf_lock =
1351 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1352 static u32 trace_buf[TRACE_BUF_SIZE];
1354 struct ftrace_event_call *call = &event_bprint;
1355 struct ring_buffer_event *event;
1356 struct ring_buffer *buffer;
1357 struct trace_array *tr = &global_trace;
1358 struct trace_array_cpu *data;
1359 struct bprint_entry *entry;
1360 unsigned long flags;
1362 int cpu, len = 0, size, pc;
1364 if (unlikely(tracing_selftest_running || tracing_disabled))
1367 /* Don't pollute graph traces with trace_vprintk internals */
1368 pause_graph_tracing();
1370 pc = preempt_count();
1371 preempt_disable_notrace();
1372 cpu = raw_smp_processor_id();
1373 data = tr->data[cpu];
1375 disable = atomic_inc_return(&data->disabled);
1376 if (unlikely(disable != 1))
1379 /* Lockdep uses trace_printk for lock tracing */
1380 local_irq_save(flags);
1381 arch_spin_lock(&trace_buf_lock);
1382 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1384 if (len > TRACE_BUF_SIZE || len < 0)
1387 size = sizeof(*entry) + sizeof(u32) * len;
1388 buffer = tr->buffer;
1389 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1393 entry = ring_buffer_event_data(event);
1397 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1398 if (!filter_check_discard(call, entry, buffer, event)) {
1399 ring_buffer_unlock_commit(buffer, event);
1400 ftrace_trace_stack(buffer, flags, 6, pc);
1404 arch_spin_unlock(&trace_buf_lock);
1405 local_irq_restore(flags);
1408 atomic_dec_return(&data->disabled);
1409 preempt_enable_notrace();
1410 unpause_graph_tracing();
1414 EXPORT_SYMBOL_GPL(trace_vbprintk);
1416 int trace_array_printk(struct trace_array *tr,
1417 unsigned long ip, const char *fmt, ...)
1422 if (!(trace_flags & TRACE_ITER_PRINTK))
1426 ret = trace_array_vprintk(tr, ip, fmt, ap);
1431 int trace_array_vprintk(struct trace_array *tr,
1432 unsigned long ip, const char *fmt, va_list args)
1434 static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1435 static char trace_buf[TRACE_BUF_SIZE];
1437 struct ftrace_event_call *call = &event_print;
1438 struct ring_buffer_event *event;
1439 struct ring_buffer *buffer;
1440 struct trace_array_cpu *data;
1441 int cpu, len = 0, size, pc;
1442 struct print_entry *entry;
1443 unsigned long irq_flags;
1446 if (tracing_disabled || tracing_selftest_running)
1449 pc = preempt_count();
1450 preempt_disable_notrace();
1451 cpu = raw_smp_processor_id();
1452 data = tr->data[cpu];
1454 disable = atomic_inc_return(&data->disabled);
1455 if (unlikely(disable != 1))
1458 pause_graph_tracing();
1459 raw_local_irq_save(irq_flags);
1460 arch_spin_lock(&trace_buf_lock);
1461 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1463 size = sizeof(*entry) + len + 1;
1464 buffer = tr->buffer;
1465 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1469 entry = ring_buffer_event_data(event);
1472 memcpy(&entry->buf, trace_buf, len);
1473 entry->buf[len] = '\0';
1474 if (!filter_check_discard(call, entry, buffer, event)) {
1475 ring_buffer_unlock_commit(buffer, event);
1476 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1480 arch_spin_unlock(&trace_buf_lock);
1481 raw_local_irq_restore(irq_flags);
1482 unpause_graph_tracing();
1484 atomic_dec_return(&data->disabled);
1485 preempt_enable_notrace();
1490 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1492 return trace_array_vprintk(&global_trace, ip, fmt, args);
1494 EXPORT_SYMBOL_GPL(trace_vprintk);
1496 enum trace_file_type {
1497 TRACE_FILE_LAT_FMT = 1,
1498 TRACE_FILE_ANNOTATE = 2,
1501 static void trace_iterator_increment(struct trace_iterator *iter)
1503 /* Don't allow ftrace to trace into the ring buffers */
1504 ftrace_disable_cpu();
1507 if (iter->buffer_iter[iter->cpu])
1508 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1510 ftrace_enable_cpu();
1513 static struct trace_entry *
1514 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1515 unsigned long *lost_events)
1517 struct ring_buffer_event *event;
1518 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1520 /* Don't allow ftrace to trace into the ring buffers */
1521 ftrace_disable_cpu();
1524 event = ring_buffer_iter_peek(buf_iter, ts);
1526 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1529 ftrace_enable_cpu();
1531 return event ? ring_buffer_event_data(event) : NULL;
1534 static struct trace_entry *
1535 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1536 unsigned long *missing_events, u64 *ent_ts)
1538 struct ring_buffer *buffer = iter->tr->buffer;
1539 struct trace_entry *ent, *next = NULL;
1540 unsigned long lost_events = 0, next_lost = 0;
1541 int cpu_file = iter->cpu_file;
1542 u64 next_ts = 0, ts;
1547 * If we are in a per_cpu trace file, don't bother by iterating over
1548 * all cpu and peek directly.
1550 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1551 if (ring_buffer_empty_cpu(buffer, cpu_file))
1553 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1555 *ent_cpu = cpu_file;
1560 for_each_tracing_cpu(cpu) {
1562 if (ring_buffer_empty_cpu(buffer, cpu))
1565 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1568 * Pick the entry with the smallest timestamp:
1570 if (ent && (!next || ts < next_ts)) {
1574 next_lost = lost_events;
1579 *ent_cpu = next_cpu;
1585 *missing_events = next_lost;
1590 /* Find the next real entry, without updating the iterator itself */
1591 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1592 int *ent_cpu, u64 *ent_ts)
1594 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1597 /* Find the next real entry, and increment the iterator to the next entry */
1598 static void *find_next_entry_inc(struct trace_iterator *iter)
1600 iter->ent = __find_next_entry(iter, &iter->cpu,
1601 &iter->lost_events, &iter->ts);
1604 trace_iterator_increment(iter);
1606 return iter->ent ? iter : NULL;
1609 static void trace_consume(struct trace_iterator *iter)
1611 /* Don't allow ftrace to trace into the ring buffers */
1612 ftrace_disable_cpu();
1613 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1614 &iter->lost_events);
1615 ftrace_enable_cpu();
1618 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1620 struct trace_iterator *iter = m->private;
1624 WARN_ON_ONCE(iter->leftover);
1628 /* can't go backwards */
1633 ent = find_next_entry_inc(iter);
1637 while (ent && iter->idx < i)
1638 ent = find_next_entry_inc(iter);
1645 static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1647 struct trace_array *tr = iter->tr;
1648 struct ring_buffer_event *event;
1649 struct ring_buffer_iter *buf_iter;
1650 unsigned long entries = 0;
1653 tr->data[cpu]->skipped_entries = 0;
1655 if (!iter->buffer_iter[cpu])
1658 buf_iter = iter->buffer_iter[cpu];
1659 ring_buffer_iter_reset(buf_iter);
1662 * We could have the case with the max latency tracers
1663 * that a reset never took place on a cpu. This is evident
1664 * by the timestamp being before the start of the buffer.
1666 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1667 if (ts >= iter->tr->time_start)
1670 ring_buffer_read(buf_iter, NULL);
1673 tr->data[cpu]->skipped_entries = entries;
1677 * The current tracer is copied to avoid a global locking
1680 static void *s_start(struct seq_file *m, loff_t *pos)
1682 struct trace_iterator *iter = m->private;
1683 static struct tracer *old_tracer;
1684 int cpu_file = iter->cpu_file;
1689 /* copy the tracer to avoid using a global lock all around */
1690 mutex_lock(&trace_types_lock);
1691 if (unlikely(old_tracer != current_trace && current_trace)) {
1692 old_tracer = current_trace;
1693 *iter->trace = *current_trace;
1695 mutex_unlock(&trace_types_lock);
1697 atomic_inc(&trace_record_cmdline_disabled);
1699 if (*pos != iter->pos) {
1704 ftrace_disable_cpu();
1706 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1707 for_each_tracing_cpu(cpu)
1708 tracing_iter_reset(iter, cpu);
1710 tracing_iter_reset(iter, cpu_file);
1712 ftrace_enable_cpu();
1715 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1720 * If we overflowed the seq_file before, then we want
1721 * to just reuse the trace_seq buffer again.
1727 p = s_next(m, p, &l);
1731 trace_event_read_lock();
1732 trace_access_lock(cpu_file);
1736 static void s_stop(struct seq_file *m, void *p)
1738 struct trace_iterator *iter = m->private;
1740 atomic_dec(&trace_record_cmdline_disabled);
1741 trace_access_unlock(iter->cpu_file);
1742 trace_event_read_unlock();
1745 static void print_lat_help_header(struct seq_file *m)
1747 seq_puts(m, "# _------=> CPU# \n");
1748 seq_puts(m, "# / _-----=> irqs-off \n");
1749 seq_puts(m, "# | / _----=> need-resched \n");
1750 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1751 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1752 seq_puts(m, "# |||| /_--=> lock-depth \n");
1753 seq_puts(m, "# |||||/ delay \n");
1754 seq_puts(m, "# cmd pid |||||| time | caller \n");
1755 seq_puts(m, "# \\ / |||||| \\ | / \n");
1758 static void print_func_help_header(struct seq_file *m)
1760 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1761 seq_puts(m, "# | | | | |\n");
1766 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1768 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1769 struct trace_array *tr = iter->tr;
1770 struct trace_array_cpu *data = tr->data[tr->cpu];
1771 struct tracer *type = current_trace;
1772 unsigned long entries = 0;
1773 unsigned long total = 0;
1774 unsigned long count;
1775 const char *name = "preemption";
1782 for_each_tracing_cpu(cpu) {
1783 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1785 * If this buffer has skipped entries, then we hold all
1786 * entries for the trace and we need to ignore the
1787 * ones before the time stamp.
1789 if (tr->data[cpu]->skipped_entries) {
1790 count -= tr->data[cpu]->skipped_entries;
1791 /* total is the same as the entries */
1795 ring_buffer_overrun_cpu(tr->buffer, cpu);
1799 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1801 seq_puts(m, "# -----------------------------------"
1802 "---------------------------------\n");
1803 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1804 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1805 nsecs_to_usecs(data->saved_latency),
1809 #if defined(CONFIG_PREEMPT_NONE)
1811 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1813 #elif defined(CONFIG_PREEMPT)
1818 /* These are reserved for later use */
1821 seq_printf(m, " #P:%d)\n", num_online_cpus());
1825 seq_puts(m, "# -----------------\n");
1826 seq_printf(m, "# | task: %.16s-%d "
1827 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1828 data->comm, data->pid, data->uid, data->nice,
1829 data->policy, data->rt_priority);
1830 seq_puts(m, "# -----------------\n");
1832 if (data->critical_start) {
1833 seq_puts(m, "# => started at: ");
1834 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1835 trace_print_seq(m, &iter->seq);
1836 seq_puts(m, "\n# => ended at: ");
1837 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1838 trace_print_seq(m, &iter->seq);
1839 seq_puts(m, "\n#\n");
1845 static void test_cpu_buff_start(struct trace_iterator *iter)
1847 struct trace_seq *s = &iter->seq;
1849 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1852 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1855 if (cpumask_test_cpu(iter->cpu, iter->started))
1858 if (iter->tr->data[iter->cpu]->skipped_entries)
1861 cpumask_set_cpu(iter->cpu, iter->started);
1863 /* Don't print started cpu buffer for the first entry of the trace */
1865 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1869 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1871 struct trace_seq *s = &iter->seq;
1872 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1873 struct trace_entry *entry;
1874 struct trace_event *event;
1878 test_cpu_buff_start(iter);
1880 event = ftrace_find_event(entry->type);
1882 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1883 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1884 if (!trace_print_lat_context(iter))
1887 if (!trace_print_context(iter))
1893 return event->funcs->trace(iter, sym_flags, event);
1895 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1898 return TRACE_TYPE_HANDLED;
1900 return TRACE_TYPE_PARTIAL_LINE;
1903 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1905 struct trace_seq *s = &iter->seq;
1906 struct trace_entry *entry;
1907 struct trace_event *event;
1911 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1912 if (!trace_seq_printf(s, "%d %d %llu ",
1913 entry->pid, iter->cpu, iter->ts))
1917 event = ftrace_find_event(entry->type);
1919 return event->funcs->raw(iter, 0, event);
1921 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1924 return TRACE_TYPE_HANDLED;
1926 return TRACE_TYPE_PARTIAL_LINE;
1929 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1931 struct trace_seq *s = &iter->seq;
1932 unsigned char newline = '\n';
1933 struct trace_entry *entry;
1934 struct trace_event *event;
1938 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1939 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1940 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1941 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1944 event = ftrace_find_event(entry->type);
1946 enum print_line_t ret = event->funcs->hex(iter, 0, event);
1947 if (ret != TRACE_TYPE_HANDLED)
1951 SEQ_PUT_FIELD_RET(s, newline);
1953 return TRACE_TYPE_HANDLED;
1956 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1958 struct trace_seq *s = &iter->seq;
1959 struct trace_entry *entry;
1960 struct trace_event *event;
1964 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1965 SEQ_PUT_FIELD_RET(s, entry->pid);
1966 SEQ_PUT_FIELD_RET(s, iter->cpu);
1967 SEQ_PUT_FIELD_RET(s, iter->ts);
1970 event = ftrace_find_event(entry->type);
1971 return event ? event->funcs->binary(iter, 0, event) :
1975 int trace_empty(struct trace_iterator *iter)
1979 /* If we are looking at one CPU buffer, only check that one */
1980 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1981 cpu = iter->cpu_file;
1982 if (iter->buffer_iter[cpu]) {
1983 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1986 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1992 for_each_tracing_cpu(cpu) {
1993 if (iter->buffer_iter[cpu]) {
1994 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1997 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2005 /* Called with trace_event_read_lock() held. */
2006 static enum print_line_t print_trace_line(struct trace_iterator *iter)
2008 enum print_line_t ret;
2010 if (iter->lost_events)
2011 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2012 iter->cpu, iter->lost_events);
2014 if (iter->trace && iter->trace->print_line) {
2015 ret = iter->trace->print_line(iter);
2016 if (ret != TRACE_TYPE_UNHANDLED)
2020 if (iter->ent->type == TRACE_BPRINT &&
2021 trace_flags & TRACE_ITER_PRINTK &&
2022 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2023 return trace_print_bprintk_msg_only(iter);
2025 if (iter->ent->type == TRACE_PRINT &&
2026 trace_flags & TRACE_ITER_PRINTK &&
2027 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2028 return trace_print_printk_msg_only(iter);
2030 if (trace_flags & TRACE_ITER_BIN)
2031 return print_bin_fmt(iter);
2033 if (trace_flags & TRACE_ITER_HEX)
2034 return print_hex_fmt(iter);
2036 if (trace_flags & TRACE_ITER_RAW)
2037 return print_raw_fmt(iter);
2039 return print_trace_fmt(iter);
2042 void trace_default_header(struct seq_file *m)
2044 struct trace_iterator *iter = m->private;
2046 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2047 /* print nothing if the buffers are empty */
2048 if (trace_empty(iter))
2050 print_trace_header(m, iter);
2051 if (!(trace_flags & TRACE_ITER_VERBOSE))
2052 print_lat_help_header(m);
2054 if (!(trace_flags & TRACE_ITER_VERBOSE))
2055 print_func_help_header(m);
2059 static int s_show(struct seq_file *m, void *v)
2061 struct trace_iterator *iter = v;
2064 if (iter->ent == NULL) {
2066 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2069 if (iter->trace && iter->trace->print_header)
2070 iter->trace->print_header(m);
2072 trace_default_header(m);
2074 } else if (iter->leftover) {
2076 * If we filled the seq_file buffer earlier, we
2077 * want to just show it now.
2079 ret = trace_print_seq(m, &iter->seq);
2081 /* ret should this time be zero, but you never know */
2082 iter->leftover = ret;
2085 print_trace_line(iter);
2086 ret = trace_print_seq(m, &iter->seq);
2088 * If we overflow the seq_file buffer, then it will
2089 * ask us for this data again at start up.
2091 * ret is 0 if seq_file write succeeded.
2094 iter->leftover = ret;
2100 static const struct seq_operations tracer_seq_ops = {
2107 static struct trace_iterator *
2108 __tracing_open(struct inode *inode, struct file *file)
2110 long cpu_file = (long) inode->i_private;
2111 void *fail_ret = ERR_PTR(-ENOMEM);
2112 struct trace_iterator *iter;
2116 if (tracing_disabled)
2117 return ERR_PTR(-ENODEV);
2119 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2121 return ERR_PTR(-ENOMEM);
2124 * We make a copy of the current tracer to avoid concurrent
2125 * changes on it while we are reading.
2127 mutex_lock(&trace_types_lock);
2128 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2133 *iter->trace = *current_trace;
2135 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2138 if (current_trace && current_trace->print_max)
2141 iter->tr = &global_trace;
2143 mutex_init(&iter->mutex);
2144 iter->cpu_file = cpu_file;
2146 /* Notify the tracer early; before we stop tracing. */
2147 if (iter->trace && iter->trace->open)
2148 iter->trace->open(iter);
2150 /* Annotate start of buffers if we had overruns */
2151 if (ring_buffer_overruns(iter->tr->buffer))
2152 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2154 /* stop the trace while dumping */
2157 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2158 for_each_tracing_cpu(cpu) {
2159 iter->buffer_iter[cpu] =
2160 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2162 ring_buffer_read_prepare_sync();
2163 for_each_tracing_cpu(cpu) {
2164 ring_buffer_read_start(iter->buffer_iter[cpu]);
2165 tracing_iter_reset(iter, cpu);
2168 cpu = iter->cpu_file;
2169 iter->buffer_iter[cpu] =
2170 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2171 ring_buffer_read_prepare_sync();
2172 ring_buffer_read_start(iter->buffer_iter[cpu]);
2173 tracing_iter_reset(iter, cpu);
2176 ret = seq_open(file, &tracer_seq_ops);
2178 fail_ret = ERR_PTR(ret);
2182 m = file->private_data;
2185 mutex_unlock(&trace_types_lock);
2190 for_each_tracing_cpu(cpu) {
2191 if (iter->buffer_iter[cpu])
2192 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2194 free_cpumask_var(iter->started);
2197 mutex_unlock(&trace_types_lock);
2204 int tracing_open_generic(struct inode *inode, struct file *filp)
2206 if (tracing_disabled)
2209 filp->private_data = inode->i_private;
2213 static int tracing_release(struct inode *inode, struct file *file)
2215 struct seq_file *m = (struct seq_file *)file->private_data;
2216 struct trace_iterator *iter;
2219 if (!(file->f_mode & FMODE_READ))
2224 mutex_lock(&trace_types_lock);
2225 for_each_tracing_cpu(cpu) {
2226 if (iter->buffer_iter[cpu])
2227 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2230 if (iter->trace && iter->trace->close)
2231 iter->trace->close(iter);
2233 /* reenable tracing if it was previously enabled */
2235 mutex_unlock(&trace_types_lock);
2237 seq_release(inode, file);
2238 mutex_destroy(&iter->mutex);
2239 free_cpumask_var(iter->started);
2245 static int tracing_open(struct inode *inode, struct file *file)
2247 struct trace_iterator *iter;
2250 /* If this file was open for write, then erase contents */
2251 if ((file->f_mode & FMODE_WRITE) &&
2252 (file->f_flags & O_TRUNC)) {
2253 long cpu = (long) inode->i_private;
2255 if (cpu == TRACE_PIPE_ALL_CPU)
2256 tracing_reset_online_cpus(&global_trace);
2258 tracing_reset(&global_trace, cpu);
2261 if (file->f_mode & FMODE_READ) {
2262 iter = __tracing_open(inode, file);
2264 ret = PTR_ERR(iter);
2265 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2266 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2272 t_next(struct seq_file *m, void *v, loff_t *pos)
2274 struct tracer *t = v;
2284 static void *t_start(struct seq_file *m, loff_t *pos)
2289 mutex_lock(&trace_types_lock);
2290 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2296 static void t_stop(struct seq_file *m, void *p)
2298 mutex_unlock(&trace_types_lock);
2301 static int t_show(struct seq_file *m, void *v)
2303 struct tracer *t = v;
2308 seq_printf(m, "%s", t->name);
2317 static const struct seq_operations show_traces_seq_ops = {
2324 static int show_traces_open(struct inode *inode, struct file *file)
2326 if (tracing_disabled)
2329 return seq_open(file, &show_traces_seq_ops);
2333 tracing_write_stub(struct file *filp, const char __user *ubuf,
2334 size_t count, loff_t *ppos)
2339 static const struct file_operations tracing_fops = {
2340 .open = tracing_open,
2342 .write = tracing_write_stub,
2343 .llseek = seq_lseek,
2344 .release = tracing_release,
2347 static const struct file_operations show_traces_fops = {
2348 .open = show_traces_open,
2350 .release = seq_release,
2351 .llseek = seq_lseek,
2355 * Only trace on a CPU if the bitmask is set:
2357 static cpumask_var_t tracing_cpumask;
2360 * The tracer itself will not take this lock, but still we want
2361 * to provide a consistent cpumask to user-space:
2363 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2366 * Temporary storage for the character representation of the
2367 * CPU bitmask (and one more byte for the newline):
2369 static char mask_str[NR_CPUS + 1];
2372 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2373 size_t count, loff_t *ppos)
2377 mutex_lock(&tracing_cpumask_update_lock);
2379 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2380 if (count - len < 2) {
2384 len += sprintf(mask_str + len, "\n");
2385 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2388 mutex_unlock(&tracing_cpumask_update_lock);
2394 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2395 size_t count, loff_t *ppos)
2398 cpumask_var_t tracing_cpumask_new;
2400 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2403 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2407 mutex_lock(&tracing_cpumask_update_lock);
2409 local_irq_disable();
2410 arch_spin_lock(&ftrace_max_lock);
2411 for_each_tracing_cpu(cpu) {
2413 * Increase/decrease the disabled counter if we are
2414 * about to flip a bit in the cpumask:
2416 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2417 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2418 atomic_inc(&global_trace.data[cpu]->disabled);
2420 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2421 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2422 atomic_dec(&global_trace.data[cpu]->disabled);
2425 arch_spin_unlock(&ftrace_max_lock);
2428 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2430 mutex_unlock(&tracing_cpumask_update_lock);
2431 free_cpumask_var(tracing_cpumask_new);
2436 free_cpumask_var(tracing_cpumask_new);
2441 static const struct file_operations tracing_cpumask_fops = {
2442 .open = tracing_open_generic,
2443 .read = tracing_cpumask_read,
2444 .write = tracing_cpumask_write,
2445 .llseek = generic_file_llseek,
2448 static int tracing_trace_options_show(struct seq_file *m, void *v)
2450 struct tracer_opt *trace_opts;
2454 mutex_lock(&trace_types_lock);
2455 tracer_flags = current_trace->flags->val;
2456 trace_opts = current_trace->flags->opts;
2458 for (i = 0; trace_options[i]; i++) {
2459 if (trace_flags & (1 << i))
2460 seq_printf(m, "%s\n", trace_options[i]);
2462 seq_printf(m, "no%s\n", trace_options[i]);
2465 for (i = 0; trace_opts[i].name; i++) {
2466 if (tracer_flags & trace_opts[i].bit)
2467 seq_printf(m, "%s\n", trace_opts[i].name);
2469 seq_printf(m, "no%s\n", trace_opts[i].name);
2471 mutex_unlock(&trace_types_lock);
2476 static int __set_tracer_option(struct tracer *trace,
2477 struct tracer_flags *tracer_flags,
2478 struct tracer_opt *opts, int neg)
2482 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2487 tracer_flags->val &= ~opts->bit;
2489 tracer_flags->val |= opts->bit;
2493 /* Try to assign a tracer specific option */
2494 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2496 struct tracer_flags *tracer_flags = trace->flags;
2497 struct tracer_opt *opts = NULL;
2500 for (i = 0; tracer_flags->opts[i].name; i++) {
2501 opts = &tracer_flags->opts[i];
2503 if (strcmp(cmp, opts->name) == 0)
2504 return __set_tracer_option(trace, trace->flags,
2511 static void set_tracer_flags(unsigned int mask, int enabled)
2513 /* do nothing if flag is already set */
2514 if (!!(trace_flags & mask) == !!enabled)
2518 trace_flags |= mask;
2520 trace_flags &= ~mask;
2522 if (mask == TRACE_ITER_RECORD_CMD)
2523 trace_event_enable_cmd_record(enabled);
2527 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2528 size_t cnt, loff_t *ppos)
2536 if (cnt >= sizeof(buf))
2539 if (copy_from_user(&buf, ubuf, cnt))
2543 cmp = strstrip(buf);
2545 if (strncmp(cmp, "no", 2) == 0) {
2550 for (i = 0; trace_options[i]; i++) {
2551 if (strcmp(cmp, trace_options[i]) == 0) {
2552 set_tracer_flags(1 << i, !neg);
2557 /* If no option could be set, test the specific tracer options */
2558 if (!trace_options[i]) {
2559 mutex_lock(&trace_types_lock);
2560 ret = set_tracer_option(current_trace, cmp, neg);
2561 mutex_unlock(&trace_types_lock);
2571 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2573 if (tracing_disabled)
2575 return single_open(file, tracing_trace_options_show, NULL);
2578 static const struct file_operations tracing_iter_fops = {
2579 .open = tracing_trace_options_open,
2581 .llseek = seq_lseek,
2582 .release = single_release,
2583 .write = tracing_trace_options_write,
2586 static const char readme_msg[] =
2587 "tracing mini-HOWTO:\n\n"
2588 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2589 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2590 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2591 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2593 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2594 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2596 "# cat /sys/kernel/debug/tracing/trace_options\n"
2597 "noprint-parent nosym-offset nosym-addr noverbose\n"
2598 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2599 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2600 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2601 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2605 tracing_readme_read(struct file *filp, char __user *ubuf,
2606 size_t cnt, loff_t *ppos)
2608 return simple_read_from_buffer(ubuf, cnt, ppos,
2609 readme_msg, strlen(readme_msg));
2612 static const struct file_operations tracing_readme_fops = {
2613 .open = tracing_open_generic,
2614 .read = tracing_readme_read,
2615 .llseek = generic_file_llseek,
2619 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2620 size_t cnt, loff_t *ppos)
2629 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2633 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2641 for (i = 0; i < SAVED_CMDLINES; i++) {
2644 pid = map_cmdline_to_pid[i];
2645 if (pid == -1 || pid == NO_CMDLINE_MAP)
2648 trace_find_cmdline(pid, buf_comm);
2649 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2654 len = simple_read_from_buffer(ubuf, cnt, ppos,
2663 static const struct file_operations tracing_saved_cmdlines_fops = {
2664 .open = tracing_open_generic,
2665 .read = tracing_saved_cmdlines_read,
2666 .llseek = generic_file_llseek,
2670 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2671 size_t cnt, loff_t *ppos)
2676 r = sprintf(buf, "%u\n", tracer_enabled);
2677 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2681 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2682 size_t cnt, loff_t *ppos)
2684 struct trace_array *tr = filp->private_data;
2689 if (cnt >= sizeof(buf))
2692 if (copy_from_user(&buf, ubuf, cnt))
2697 ret = strict_strtoul(buf, 10, &val);
2703 mutex_lock(&trace_types_lock);
2704 if (tracer_enabled ^ val) {
2707 if (current_trace->start)
2708 current_trace->start(tr);
2713 if (current_trace->stop)
2714 current_trace->stop(tr);
2717 mutex_unlock(&trace_types_lock);
2725 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2726 size_t cnt, loff_t *ppos)
2728 char buf[MAX_TRACER_SIZE+2];
2731 mutex_lock(&trace_types_lock);
2733 r = sprintf(buf, "%s\n", current_trace->name);
2735 r = sprintf(buf, "\n");
2736 mutex_unlock(&trace_types_lock);
2738 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2741 int tracer_init(struct tracer *t, struct trace_array *tr)
2743 tracing_reset_online_cpus(tr);
2747 static int tracing_resize_ring_buffer(unsigned long size)
2752 * If kernel or user changes the size of the ring buffer
2753 * we use the size that was given, and we can forget about
2754 * expanding it later.
2756 ring_buffer_expanded = 1;
2758 ret = ring_buffer_resize(global_trace.buffer, size);
2762 if (!current_trace->use_max_tr)
2765 ret = ring_buffer_resize(max_tr.buffer, size);
2769 r = ring_buffer_resize(global_trace.buffer,
2770 global_trace.entries);
2773 * AARGH! We are left with different
2774 * size max buffer!!!!
2775 * The max buffer is our "snapshot" buffer.
2776 * When a tracer needs a snapshot (one of the
2777 * latency tracers), it swaps the max buffer
2778 * with the saved snap shot. We succeeded to
2779 * update the size of the main buffer, but failed to
2780 * update the size of the max buffer. But when we tried
2781 * to reset the main buffer to the original size, we
2782 * failed there too. This is very unlikely to
2783 * happen, but if it does, warn and kill all
2787 tracing_disabled = 1;
2792 max_tr.entries = size;
2794 global_trace.entries = size;
2801 * tracing_update_buffers - used by tracing facility to expand ring buffers
2803 * To save on memory when the tracing is never used on a system with it
2804 * configured in. The ring buffers are set to a minimum size. But once
2805 * a user starts to use the tracing facility, then they need to grow
2806 * to their default size.
2808 * This function is to be called when a tracer is about to be used.
2810 int tracing_update_buffers(void)
2814 mutex_lock(&trace_types_lock);
2815 if (!ring_buffer_expanded)
2816 ret = tracing_resize_ring_buffer(trace_buf_size);
2817 mutex_unlock(&trace_types_lock);
2822 struct trace_option_dentry;
2824 static struct trace_option_dentry *
2825 create_trace_option_files(struct tracer *tracer);
2828 destroy_trace_option_files(struct trace_option_dentry *topts);
2830 static int tracing_set_tracer(const char *buf)
2832 static struct trace_option_dentry *topts;
2833 struct trace_array *tr = &global_trace;
2837 mutex_lock(&trace_types_lock);
2839 if (!ring_buffer_expanded) {
2840 ret = tracing_resize_ring_buffer(trace_buf_size);
2846 for (t = trace_types; t; t = t->next) {
2847 if (strcmp(t->name, buf) == 0)
2854 if (t == current_trace)
2857 trace_branch_disable();
2858 if (current_trace && current_trace->reset)
2859 current_trace->reset(tr);
2860 if (current_trace && current_trace->use_max_tr) {
2862 * We don't free the ring buffer. instead, resize it because
2863 * The max_tr ring buffer has some state (e.g. ring->clock) and
2864 * we want preserve it.
2866 ring_buffer_resize(max_tr.buffer, 1);
2869 destroy_trace_option_files(topts);
2873 topts = create_trace_option_files(current_trace);
2874 if (current_trace->use_max_tr) {
2875 ret = ring_buffer_resize(max_tr.buffer, global_trace.entries);
2878 max_tr.entries = global_trace.entries;
2882 ret = tracer_init(t, tr);
2887 trace_branch_enable(tr);
2889 mutex_unlock(&trace_types_lock);
2895 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2896 size_t cnt, loff_t *ppos)
2898 char buf[MAX_TRACER_SIZE+1];
2905 if (cnt > MAX_TRACER_SIZE)
2906 cnt = MAX_TRACER_SIZE;
2908 if (copy_from_user(&buf, ubuf, cnt))
2913 /* strip ending whitespace. */
2914 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2917 err = tracing_set_tracer(buf);
2927 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2928 size_t cnt, loff_t *ppos)
2930 unsigned long *ptr = filp->private_data;
2934 r = snprintf(buf, sizeof(buf), "%ld\n",
2935 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2936 if (r > sizeof(buf))
2938 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2942 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2943 size_t cnt, loff_t *ppos)
2945 unsigned long *ptr = filp->private_data;
2950 if (cnt >= sizeof(buf))
2953 if (copy_from_user(&buf, ubuf, cnt))
2958 ret = strict_strtoul(buf, 10, &val);
2967 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2969 long cpu_file = (long) inode->i_private;
2970 struct trace_iterator *iter;
2973 if (tracing_disabled)
2976 mutex_lock(&trace_types_lock);
2978 /* create a buffer to store the information to pass to userspace */
2979 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2986 * We make a copy of the current tracer to avoid concurrent
2987 * changes on it while we are reading.
2989 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2995 *iter->trace = *current_trace;
2997 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3002 /* trace pipe does not show start of buffer */
3003 cpumask_setall(iter->started);
3005 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3006 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3008 iter->cpu_file = cpu_file;
3009 iter->tr = &global_trace;
3010 mutex_init(&iter->mutex);
3011 filp->private_data = iter;
3013 if (iter->trace->pipe_open)
3014 iter->trace->pipe_open(iter);
3016 nonseekable_open(inode, filp);
3018 mutex_unlock(&trace_types_lock);
3024 mutex_unlock(&trace_types_lock);
3028 static int tracing_release_pipe(struct inode *inode, struct file *file)
3030 struct trace_iterator *iter = file->private_data;
3032 mutex_lock(&trace_types_lock);
3034 if (iter->trace->pipe_close)
3035 iter->trace->pipe_close(iter);
3037 mutex_unlock(&trace_types_lock);
3039 free_cpumask_var(iter->started);
3040 mutex_destroy(&iter->mutex);
3048 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3050 struct trace_iterator *iter = filp->private_data;
3052 if (trace_flags & TRACE_ITER_BLOCK) {
3054 * Always select as readable when in blocking mode
3056 return POLLIN | POLLRDNORM;
3058 if (!trace_empty(iter))
3059 return POLLIN | POLLRDNORM;
3060 poll_wait(filp, &trace_wait, poll_table);
3061 if (!trace_empty(iter))
3062 return POLLIN | POLLRDNORM;
3069 void default_wait_pipe(struct trace_iterator *iter)
3073 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3075 if (trace_empty(iter))
3078 finish_wait(&trace_wait, &wait);
3082 * This is a make-shift waitqueue.
3083 * A tracer might use this callback on some rare cases:
3085 * 1) the current tracer might hold the runqueue lock when it wakes up
3086 * a reader, hence a deadlock (sched, function, and function graph tracers)
3087 * 2) the function tracers, trace all functions, we don't want
3088 * the overhead of calling wake_up and friends
3089 * (and tracing them too)
3091 * Anyway, this is really very primitive wakeup.
3093 void poll_wait_pipe(struct trace_iterator *iter)
3095 set_current_state(TASK_INTERRUPTIBLE);
3096 /* sleep for 100 msecs, and try again. */
3097 schedule_timeout(HZ / 10);
3100 /* Must be called with trace_types_lock mutex held. */
3101 static int tracing_wait_pipe(struct file *filp)
3103 struct trace_iterator *iter = filp->private_data;
3105 while (trace_empty(iter)) {
3107 if ((filp->f_flags & O_NONBLOCK)) {
3111 mutex_unlock(&iter->mutex);
3113 iter->trace->wait_pipe(iter);
3115 mutex_lock(&iter->mutex);
3117 if (signal_pending(current))
3121 * We block until we read something and tracing is disabled.
3122 * We still block if tracing is disabled, but we have never
3123 * read anything. This allows a user to cat this file, and
3124 * then enable tracing. But after we have read something,
3125 * we give an EOF when tracing is again disabled.
3127 * iter->pos will be 0 if we haven't read anything.
3129 if (!tracer_enabled && iter->pos)
3140 tracing_read_pipe(struct file *filp, char __user *ubuf,
3141 size_t cnt, loff_t *ppos)
3143 struct trace_iterator *iter = filp->private_data;
3144 static struct tracer *old_tracer;
3147 /* return any leftover data */
3148 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3152 trace_seq_init(&iter->seq);
3154 /* copy the tracer to avoid using a global lock all around */
3155 mutex_lock(&trace_types_lock);
3156 if (unlikely(old_tracer != current_trace && current_trace)) {
3157 old_tracer = current_trace;
3158 *iter->trace = *current_trace;
3160 mutex_unlock(&trace_types_lock);
3163 * Avoid more than one consumer on a single file descriptor
3164 * This is just a matter of traces coherency, the ring buffer itself
3167 mutex_lock(&iter->mutex);
3168 if (iter->trace->read) {
3169 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3175 sret = tracing_wait_pipe(filp);
3179 /* stop when tracing is finished */
3180 if (trace_empty(iter)) {
3185 if (cnt >= PAGE_SIZE)
3186 cnt = PAGE_SIZE - 1;
3188 /* reset all but tr, trace, and overruns */
3189 memset(&iter->seq, 0,
3190 sizeof(struct trace_iterator) -
3191 offsetof(struct trace_iterator, seq));
3194 trace_event_read_lock();
3195 trace_access_lock(iter->cpu_file);
3196 while (find_next_entry_inc(iter) != NULL) {
3197 enum print_line_t ret;
3198 int len = iter->seq.len;
3200 ret = print_trace_line(iter);
3201 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3202 /* don't print partial lines */
3203 iter->seq.len = len;
3206 if (ret != TRACE_TYPE_NO_CONSUME)
3207 trace_consume(iter);
3209 if (iter->seq.len >= cnt)
3212 trace_access_unlock(iter->cpu_file);
3213 trace_event_read_unlock();
3215 /* Now copy what we have to the user */
3216 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3217 if (iter->seq.readpos >= iter->seq.len)
3218 trace_seq_init(&iter->seq);
3221 * If there was nothing to send to user, inspite of consuming trace
3222 * entries, go back to wait for more entries.
3228 mutex_unlock(&iter->mutex);
3233 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3234 struct pipe_buffer *buf)
3236 __free_page(buf->page);
3239 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3242 __free_page(spd->pages[idx]);
3245 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3247 .map = generic_pipe_buf_map,
3248 .unmap = generic_pipe_buf_unmap,
3249 .confirm = generic_pipe_buf_confirm,
3250 .release = tracing_pipe_buf_release,
3251 .steal = generic_pipe_buf_steal,
3252 .get = generic_pipe_buf_get,
3256 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3261 /* Seq buffer is page-sized, exactly what we need. */
3263 count = iter->seq.len;
3264 ret = print_trace_line(iter);
3265 count = iter->seq.len - count;
3268 iter->seq.len -= count;
3271 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3272 iter->seq.len -= count;
3276 if (ret != TRACE_TYPE_NO_CONSUME)
3277 trace_consume(iter);
3279 if (!find_next_entry_inc(iter)) {
3289 static ssize_t tracing_splice_read_pipe(struct file *filp,
3291 struct pipe_inode_info *pipe,
3295 struct page *pages_def[PIPE_DEF_BUFFERS];
3296 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3297 struct trace_iterator *iter = filp->private_data;
3298 struct splice_pipe_desc spd = {
3300 .partial = partial_def,
3301 .nr_pages = 0, /* This gets updated below. */
3303 .ops = &tracing_pipe_buf_ops,
3304 .spd_release = tracing_spd_release_pipe,
3306 static struct tracer *old_tracer;
3311 if (splice_grow_spd(pipe, &spd))
3314 /* copy the tracer to avoid using a global lock all around */
3315 mutex_lock(&trace_types_lock);
3316 if (unlikely(old_tracer != current_trace && current_trace)) {
3317 old_tracer = current_trace;
3318 *iter->trace = *current_trace;
3320 mutex_unlock(&trace_types_lock);
3322 mutex_lock(&iter->mutex);
3324 if (iter->trace->splice_read) {
3325 ret = iter->trace->splice_read(iter, filp,
3326 ppos, pipe, len, flags);
3331 ret = tracing_wait_pipe(filp);
3335 if (!iter->ent && !find_next_entry_inc(iter)) {
3340 trace_event_read_lock();
3341 trace_access_lock(iter->cpu_file);
3343 /* Fill as many pages as possible. */
3344 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3345 spd.pages[i] = alloc_page(GFP_KERNEL);
3349 rem = tracing_fill_pipe_page(rem, iter);
3351 /* Copy the data into the page, so we can start over. */
3352 ret = trace_seq_to_buffer(&iter->seq,
3353 page_address(spd.pages[i]),
3356 __free_page(spd.pages[i]);
3359 spd.partial[i].offset = 0;
3360 spd.partial[i].len = iter->seq.len;
3362 trace_seq_init(&iter->seq);
3365 trace_access_unlock(iter->cpu_file);
3366 trace_event_read_unlock();
3367 mutex_unlock(&iter->mutex);
3371 ret = splice_to_pipe(pipe, &spd);
3373 splice_shrink_spd(pipe, &spd);
3377 mutex_unlock(&iter->mutex);
3382 tracing_entries_read(struct file *filp, char __user *ubuf,
3383 size_t cnt, loff_t *ppos)
3385 struct trace_array *tr = filp->private_data;
3389 mutex_lock(&trace_types_lock);
3390 if (!ring_buffer_expanded)
3391 r = sprintf(buf, "%lu (expanded: %lu)\n",
3393 trace_buf_size >> 10);
3395 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3396 mutex_unlock(&trace_types_lock);
3398 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3402 tracing_entries_write(struct file *filp, const char __user *ubuf,
3403 size_t cnt, loff_t *ppos)
3409 if (cnt >= sizeof(buf))
3412 if (copy_from_user(&buf, ubuf, cnt))
3417 ret = strict_strtoul(buf, 10, &val);
3421 /* must have at least 1 entry */
3425 mutex_lock(&trace_types_lock);
3429 /* disable all cpu buffers */
3430 for_each_tracing_cpu(cpu) {
3431 if (global_trace.data[cpu])
3432 atomic_inc(&global_trace.data[cpu]->disabled);
3433 if (max_tr.data[cpu])
3434 atomic_inc(&max_tr.data[cpu]->disabled);
3437 /* value is in KB */
3440 if (val != global_trace.entries) {
3441 ret = tracing_resize_ring_buffer(val);
3450 /* If check pages failed, return ENOMEM */
3451 if (tracing_disabled)
3454 for_each_tracing_cpu(cpu) {
3455 if (global_trace.data[cpu])
3456 atomic_dec(&global_trace.data[cpu]->disabled);
3457 if (max_tr.data[cpu])
3458 atomic_dec(&max_tr.data[cpu]->disabled);
3462 mutex_unlock(&trace_types_lock);
3467 static int mark_printk(const char *fmt, ...)
3471 va_start(args, fmt);
3472 ret = trace_vprintk(0, fmt, args);
3478 tracing_mark_write(struct file *filp, const char __user *ubuf,
3479 size_t cnt, loff_t *fpos)
3483 if (tracing_disabled)
3486 if (cnt > TRACE_BUF_SIZE)
3487 cnt = TRACE_BUF_SIZE;
3489 buf = kmalloc(cnt + 2, GFP_KERNEL);
3493 if (copy_from_user(buf, ubuf, cnt)) {
3497 if (buf[cnt-1] != '\n') {
3503 cnt = mark_printk("%s", buf);
3510 static int tracing_clock_show(struct seq_file *m, void *v)
3514 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3516 "%s%s%s%s", i ? " " : "",
3517 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3518 i == trace_clock_id ? "]" : "");
3524 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3525 size_t cnt, loff_t *fpos)
3528 const char *clockstr;
3531 if (cnt >= sizeof(buf))
3534 if (copy_from_user(&buf, ubuf, cnt))
3539 clockstr = strstrip(buf);
3541 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3542 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3545 if (i == ARRAY_SIZE(trace_clocks))
3550 mutex_lock(&trace_types_lock);
3552 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3554 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3556 mutex_unlock(&trace_types_lock);
3563 static int tracing_clock_open(struct inode *inode, struct file *file)
3565 if (tracing_disabled)
3567 return single_open(file, tracing_clock_show, NULL);
3570 static const struct file_operations tracing_max_lat_fops = {
3571 .open = tracing_open_generic,
3572 .read = tracing_max_lat_read,
3573 .write = tracing_max_lat_write,
3574 .llseek = generic_file_llseek,
3577 static const struct file_operations tracing_ctrl_fops = {
3578 .open = tracing_open_generic,
3579 .read = tracing_ctrl_read,
3580 .write = tracing_ctrl_write,
3581 .llseek = generic_file_llseek,
3584 static const struct file_operations set_tracer_fops = {
3585 .open = tracing_open_generic,
3586 .read = tracing_set_trace_read,
3587 .write = tracing_set_trace_write,
3588 .llseek = generic_file_llseek,
3591 static const struct file_operations tracing_pipe_fops = {
3592 .open = tracing_open_pipe,
3593 .poll = tracing_poll_pipe,
3594 .read = tracing_read_pipe,
3595 .splice_read = tracing_splice_read_pipe,
3596 .release = tracing_release_pipe,
3597 .llseek = no_llseek,
3600 static const struct file_operations tracing_entries_fops = {
3601 .open = tracing_open_generic,
3602 .read = tracing_entries_read,
3603 .write = tracing_entries_write,
3604 .llseek = generic_file_llseek,
3607 static const struct file_operations tracing_mark_fops = {
3608 .open = tracing_open_generic,
3609 .write = tracing_mark_write,
3610 .llseek = generic_file_llseek,
3613 static const struct file_operations trace_clock_fops = {
3614 .open = tracing_clock_open,
3616 .llseek = seq_lseek,
3617 .release = single_release,
3618 .write = tracing_clock_write,
3621 struct ftrace_buffer_info {
3622 struct trace_array *tr;
3628 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3630 int cpu = (int)(long)inode->i_private;
3631 struct ftrace_buffer_info *info;
3633 if (tracing_disabled)
3636 info = kzalloc(sizeof(*info), GFP_KERNEL);
3640 info->tr = &global_trace;
3643 /* Force reading ring buffer for first read */
3644 info->read = (unsigned int)-1;
3646 filp->private_data = info;
3648 return nonseekable_open(inode, filp);
3652 tracing_buffers_read(struct file *filp, char __user *ubuf,
3653 size_t count, loff_t *ppos)
3655 struct ftrace_buffer_info *info = filp->private_data;
3663 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3667 /* Do we have previous read data to read? */
3668 if (info->read < PAGE_SIZE)
3673 trace_access_lock(info->cpu);
3674 ret = ring_buffer_read_page(info->tr->buffer,
3678 trace_access_unlock(info->cpu);
3683 size = PAGE_SIZE - info->read;
3687 ret = copy_to_user(ubuf, info->spare + info->read, size);
3698 static int tracing_buffers_release(struct inode *inode, struct file *file)
3700 struct ftrace_buffer_info *info = file->private_data;
3703 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3710 struct ring_buffer *buffer;
3715 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3716 struct pipe_buffer *buf)
3718 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3723 ring_buffer_free_read_page(ref->buffer, ref->page);
3728 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3729 struct pipe_buffer *buf)
3734 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3735 struct pipe_buffer *buf)
3737 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3742 /* Pipe buffer operations for a buffer. */
3743 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
3745 .map = generic_pipe_buf_map,
3746 .unmap = generic_pipe_buf_unmap,
3747 .confirm = generic_pipe_buf_confirm,
3748 .release = buffer_pipe_buf_release,
3749 .steal = buffer_pipe_buf_steal,
3750 .get = buffer_pipe_buf_get,
3754 * Callback from splice_to_pipe(), if we need to release some pages
3755 * at the end of the spd in case we error'ed out in filling the pipe.
3757 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3759 struct buffer_ref *ref =
3760 (struct buffer_ref *)spd->partial[i].private;
3765 ring_buffer_free_read_page(ref->buffer, ref->page);
3767 spd->partial[i].private = 0;
3771 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3772 struct pipe_inode_info *pipe, size_t len,
3775 struct ftrace_buffer_info *info = file->private_data;
3776 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3777 struct page *pages_def[PIPE_DEF_BUFFERS];
3778 struct splice_pipe_desc spd = {
3780 .partial = partial_def,
3782 .ops = &buffer_pipe_buf_ops,
3783 .spd_release = buffer_spd_release,
3785 struct buffer_ref *ref;
3786 int entries, size, i;
3789 if (splice_grow_spd(pipe, &spd))
3792 if (*ppos & (PAGE_SIZE - 1)) {
3793 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3798 if (len & (PAGE_SIZE - 1)) {
3799 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3800 if (len < PAGE_SIZE) {
3807 trace_access_lock(info->cpu);
3808 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3810 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
3814 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3819 ref->buffer = info->tr->buffer;
3820 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3826 r = ring_buffer_read_page(ref->buffer, &ref->page,
3829 ring_buffer_free_read_page(ref->buffer,
3836 * zero out any left over data, this is going to
3839 size = ring_buffer_page_len(ref->page);
3840 if (size < PAGE_SIZE)
3841 memset(ref->page + size, 0, PAGE_SIZE - size);
3843 page = virt_to_page(ref->page);
3845 spd.pages[i] = page;
3846 spd.partial[i].len = PAGE_SIZE;
3847 spd.partial[i].offset = 0;
3848 spd.partial[i].private = (unsigned long)ref;
3852 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3855 trace_access_unlock(info->cpu);
3858 /* did we read anything? */
3859 if (!spd.nr_pages) {
3860 if (flags & SPLICE_F_NONBLOCK)
3868 ret = splice_to_pipe(pipe, &spd);
3869 splice_shrink_spd(pipe, &spd);
3874 static const struct file_operations tracing_buffers_fops = {
3875 .open = tracing_buffers_open,
3876 .read = tracing_buffers_read,
3877 .release = tracing_buffers_release,
3878 .splice_read = tracing_buffers_splice_read,
3879 .llseek = no_llseek,
3883 tracing_stats_read(struct file *filp, char __user *ubuf,
3884 size_t count, loff_t *ppos)
3886 unsigned long cpu = (unsigned long)filp->private_data;
3887 struct trace_array *tr = &global_trace;
3888 struct trace_seq *s;
3891 s = kmalloc(sizeof(*s), GFP_KERNEL);
3897 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3898 trace_seq_printf(s, "entries: %ld\n", cnt);
3900 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3901 trace_seq_printf(s, "overrun: %ld\n", cnt);
3903 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3904 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3906 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3913 static const struct file_operations tracing_stats_fops = {
3914 .open = tracing_open_generic,
3915 .read = tracing_stats_read,
3916 .llseek = generic_file_llseek,
3919 #ifdef CONFIG_DYNAMIC_FTRACE
3921 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3927 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3928 size_t cnt, loff_t *ppos)
3930 static char ftrace_dyn_info_buffer[1024];
3931 static DEFINE_MUTEX(dyn_info_mutex);
3932 unsigned long *p = filp->private_data;
3933 char *buf = ftrace_dyn_info_buffer;
3934 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3937 mutex_lock(&dyn_info_mutex);
3938 r = sprintf(buf, "%ld ", *p);
3940 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3943 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3945 mutex_unlock(&dyn_info_mutex);
3950 static const struct file_operations tracing_dyn_info_fops = {
3951 .open = tracing_open_generic,
3952 .read = tracing_read_dyn_info,
3953 .llseek = generic_file_llseek,
3957 static struct dentry *d_tracer;
3959 struct dentry *tracing_init_dentry(void)
3966 if (!debugfs_initialized())
3969 d_tracer = debugfs_create_dir("tracing", NULL);
3971 if (!d_tracer && !once) {
3973 pr_warning("Could not create debugfs directory 'tracing'\n");
3980 static struct dentry *d_percpu;
3982 struct dentry *tracing_dentry_percpu(void)
3985 struct dentry *d_tracer;
3990 d_tracer = tracing_init_dentry();
3995 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3997 if (!d_percpu && !once) {
3999 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4006 static void tracing_init_debugfs_percpu(long cpu)
4008 struct dentry *d_percpu = tracing_dentry_percpu();
4009 struct dentry *d_cpu;
4010 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
4013 if (cpu > 999 || cpu < 0)
4016 sprintf(cpu_dir, "cpu%ld", cpu);
4017 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4019 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4023 /* per cpu trace_pipe */
4024 trace_create_file("trace_pipe", 0444, d_cpu,
4025 (void *) cpu, &tracing_pipe_fops);
4028 trace_create_file("trace", 0644, d_cpu,
4029 (void *) cpu, &tracing_fops);
4031 trace_create_file("trace_pipe_raw", 0444, d_cpu,
4032 (void *) cpu, &tracing_buffers_fops);
4034 trace_create_file("stats", 0444, d_cpu,
4035 (void *) cpu, &tracing_stats_fops);
4038 #ifdef CONFIG_FTRACE_SELFTEST
4039 /* Let selftest have access to static functions in this file */
4040 #include "trace_selftest.c"
4043 struct trace_option_dentry {
4044 struct tracer_opt *opt;
4045 struct tracer_flags *flags;
4046 struct dentry *entry;
4050 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4053 struct trace_option_dentry *topt = filp->private_data;
4056 if (topt->flags->val & topt->opt->bit)
4061 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4065 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4068 struct trace_option_dentry *topt = filp->private_data;
4073 if (cnt >= sizeof(buf))
4076 if (copy_from_user(&buf, ubuf, cnt))
4081 ret = strict_strtoul(buf, 10, &val);
4085 if (val != 0 && val != 1)
4088 if (!!(topt->flags->val & topt->opt->bit) != val) {
4089 mutex_lock(&trace_types_lock);
4090 ret = __set_tracer_option(current_trace, topt->flags,
4092 mutex_unlock(&trace_types_lock);
4103 static const struct file_operations trace_options_fops = {
4104 .open = tracing_open_generic,
4105 .read = trace_options_read,
4106 .write = trace_options_write,
4107 .llseek = generic_file_llseek,
4111 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4114 long index = (long)filp->private_data;
4117 if (trace_flags & (1 << index))
4122 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4126 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4129 long index = (long)filp->private_data;
4134 if (cnt >= sizeof(buf))
4137 if (copy_from_user(&buf, ubuf, cnt))
4142 ret = strict_strtoul(buf, 10, &val);
4146 if (val != 0 && val != 1)
4148 set_tracer_flags(1 << index, val);
4155 static const struct file_operations trace_options_core_fops = {
4156 .open = tracing_open_generic,
4157 .read = trace_options_core_read,
4158 .write = trace_options_core_write,
4159 .llseek = generic_file_llseek,
4162 struct dentry *trace_create_file(const char *name,
4164 struct dentry *parent,
4166 const struct file_operations *fops)
4170 ret = debugfs_create_file(name, mode, parent, data, fops);
4172 pr_warning("Could not create debugfs '%s' entry\n", name);
4178 static struct dentry *trace_options_init_dentry(void)
4180 struct dentry *d_tracer;
4181 static struct dentry *t_options;
4186 d_tracer = tracing_init_dentry();
4190 t_options = debugfs_create_dir("options", d_tracer);
4192 pr_warning("Could not create debugfs directory 'options'\n");
4200 create_trace_option_file(struct trace_option_dentry *topt,
4201 struct tracer_flags *flags,
4202 struct tracer_opt *opt)
4204 struct dentry *t_options;
4206 t_options = trace_options_init_dentry();
4210 topt->flags = flags;
4213 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4214 &trace_options_fops);
4218 static struct trace_option_dentry *
4219 create_trace_option_files(struct tracer *tracer)
4221 struct trace_option_dentry *topts;
4222 struct tracer_flags *flags;
4223 struct tracer_opt *opts;
4229 flags = tracer->flags;
4231 if (!flags || !flags->opts)
4236 for (cnt = 0; opts[cnt].name; cnt++)
4239 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4243 for (cnt = 0; opts[cnt].name; cnt++)
4244 create_trace_option_file(&topts[cnt], flags,
4251 destroy_trace_option_files(struct trace_option_dentry *topts)
4258 for (cnt = 0; topts[cnt].opt; cnt++) {
4259 if (topts[cnt].entry)
4260 debugfs_remove(topts[cnt].entry);
4266 static struct dentry *
4267 create_trace_option_core_file(const char *option, long index)
4269 struct dentry *t_options;
4271 t_options = trace_options_init_dentry();
4275 return trace_create_file(option, 0644, t_options, (void *)index,
4276 &trace_options_core_fops);
4279 static __init void create_trace_options_dir(void)
4281 struct dentry *t_options;
4284 t_options = trace_options_init_dentry();
4288 for (i = 0; trace_options[i]; i++)
4289 create_trace_option_core_file(trace_options[i], i);
4292 static __init int tracer_init_debugfs(void)
4294 struct dentry *d_tracer;
4297 trace_access_lock_init();
4299 d_tracer = tracing_init_dentry();
4301 trace_create_file("tracing_enabled", 0644, d_tracer,
4302 &global_trace, &tracing_ctrl_fops);
4304 trace_create_file("trace_options", 0644, d_tracer,
4305 NULL, &tracing_iter_fops);
4307 trace_create_file("tracing_cpumask", 0644, d_tracer,
4308 NULL, &tracing_cpumask_fops);
4310 trace_create_file("trace", 0644, d_tracer,
4311 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4313 trace_create_file("available_tracers", 0444, d_tracer,
4314 &global_trace, &show_traces_fops);
4316 trace_create_file("current_tracer", 0644, d_tracer,
4317 &global_trace, &set_tracer_fops);
4319 #ifdef CONFIG_TRACER_MAX_TRACE
4320 trace_create_file("tracing_max_latency", 0644, d_tracer,
4321 &tracing_max_latency, &tracing_max_lat_fops);
4324 trace_create_file("tracing_thresh", 0644, d_tracer,
4325 &tracing_thresh, &tracing_max_lat_fops);
4327 trace_create_file("README", 0444, d_tracer,
4328 NULL, &tracing_readme_fops);
4330 trace_create_file("trace_pipe", 0444, d_tracer,
4331 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4333 trace_create_file("buffer_size_kb", 0644, d_tracer,
4334 &global_trace, &tracing_entries_fops);
4336 trace_create_file("trace_marker", 0220, d_tracer,
4337 NULL, &tracing_mark_fops);
4339 trace_create_file("saved_cmdlines", 0444, d_tracer,
4340 NULL, &tracing_saved_cmdlines_fops);
4342 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4345 #ifdef CONFIG_DYNAMIC_FTRACE
4346 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4347 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4350 create_trace_options_dir();
4352 for_each_tracing_cpu(cpu)
4353 tracing_init_debugfs_percpu(cpu);
4358 static int trace_panic_handler(struct notifier_block *this,
4359 unsigned long event, void *unused)
4361 if (ftrace_dump_on_oops)
4362 ftrace_dump(ftrace_dump_on_oops);
4366 static struct notifier_block trace_panic_notifier = {
4367 .notifier_call = trace_panic_handler,
4369 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4372 static int trace_die_handler(struct notifier_block *self,
4378 if (ftrace_dump_on_oops)
4379 ftrace_dump(ftrace_dump_on_oops);
4387 static struct notifier_block trace_die_notifier = {
4388 .notifier_call = trace_die_handler,
4393 * printk is set to max of 1024, we really don't need it that big.
4394 * Nothing should be printing 1000 characters anyway.
4396 #define TRACE_MAX_PRINT 1000
4399 * Define here KERN_TRACE so that we have one place to modify
4400 * it if we decide to change what log level the ftrace dump
4403 #define KERN_TRACE KERN_EMERG
4406 trace_printk_seq(struct trace_seq *s)
4408 /* Probably should print a warning here. */
4412 /* should be zero ended, but we are paranoid. */
4413 s->buffer[s->len] = 0;
4415 printk(KERN_TRACE "%s", s->buffer);
4421 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4423 static arch_spinlock_t ftrace_dump_lock =
4424 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4425 /* use static because iter can be a bit big for the stack */
4426 static struct trace_iterator iter;
4427 unsigned int old_userobj;
4428 static int dump_ran;
4429 unsigned long flags;
4433 local_irq_save(flags);
4434 arch_spin_lock(&ftrace_dump_lock);
4442 if (disable_tracing)
4445 for_each_tracing_cpu(cpu) {
4446 atomic_inc(&global_trace.data[cpu]->disabled);
4449 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4451 /* don't look at user memory in panic mode */
4452 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4454 /* Simulate the iterator */
4455 iter.tr = &global_trace;
4456 iter.trace = current_trace;
4458 switch (oops_dump_mode) {
4460 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4463 iter.cpu_file = raw_smp_processor_id();
4468 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
4469 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4472 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4475 * We need to stop all tracing on all CPUS to read the
4476 * the next buffer. This is a bit expensive, but is
4477 * not done often. We fill all what we can read,
4478 * and then release the locks again.
4481 while (!trace_empty(&iter)) {
4484 printk(KERN_TRACE "---------------------------------\n");
4488 /* reset all but tr, trace, and overruns */
4489 memset(&iter.seq, 0,
4490 sizeof(struct trace_iterator) -
4491 offsetof(struct trace_iterator, seq));
4492 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4495 if (find_next_entry_inc(&iter) != NULL) {
4498 ret = print_trace_line(&iter);
4499 if (ret != TRACE_TYPE_NO_CONSUME)
4500 trace_consume(&iter);
4503 trace_printk_seq(&iter.seq);
4507 printk(KERN_TRACE " (ftrace buffer empty)\n");
4509 printk(KERN_TRACE "---------------------------------\n");
4512 /* Re-enable tracing if requested */
4513 if (!disable_tracing) {
4514 trace_flags |= old_userobj;
4516 for_each_tracing_cpu(cpu) {
4517 atomic_dec(&global_trace.data[cpu]->disabled);
4523 arch_spin_unlock(&ftrace_dump_lock);
4524 local_irq_restore(flags);
4527 /* By default: disable tracing after the dump */
4528 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
4530 __ftrace_dump(true, oops_dump_mode);
4533 __init static int tracer_alloc_buffers(void)
4539 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4542 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4543 goto out_free_buffer_mask;
4545 /* To save memory, keep the ring buffer size to its minimum */
4546 if (ring_buffer_expanded)
4547 ring_buf_size = trace_buf_size;
4551 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4552 cpumask_copy(tracing_cpumask, cpu_all_mask);
4554 /* TODO: make the number of buffers hot pluggable with CPUS */
4555 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4556 TRACE_BUFFER_FLAGS);
4557 if (!global_trace.buffer) {
4558 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4560 goto out_free_cpumask;
4562 global_trace.entries = ring_buffer_size(global_trace.buffer);
4565 #ifdef CONFIG_TRACER_MAX_TRACE
4566 max_tr.buffer = ring_buffer_alloc(1, TRACE_BUFFER_FLAGS);
4567 if (!max_tr.buffer) {
4568 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4570 ring_buffer_free(global_trace.buffer);
4571 goto out_free_cpumask;
4576 /* Allocate the first page for all buffers */
4577 for_each_tracing_cpu(i) {
4578 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4579 max_tr.data[i] = &per_cpu(max_tr_data, i);
4582 trace_init_cmdlines();
4584 register_tracer(&nop_trace);
4585 current_trace = &nop_trace;
4586 /* All seems OK, enable tracing */
4587 tracing_disabled = 0;
4589 atomic_notifier_chain_register(&panic_notifier_list,
4590 &trace_panic_notifier);
4592 register_die_notifier(&trace_die_notifier);
4597 free_cpumask_var(tracing_cpumask);
4598 out_free_buffer_mask:
4599 free_cpumask_var(tracing_buffer_mask);
4604 __init static int clear_boot_tracer(void)
4607 * The default tracer at boot buffer is an init section.
4608 * This function is called in lateinit. If we did not
4609 * find the boot tracer, then clear it out, to prevent
4610 * later registration from accessing the buffer that is
4611 * about to be freed.
4613 if (!default_bootup_tracer)
4616 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4617 default_bootup_tracer);
4618 default_bootup_tracer = NULL;
4623 early_initcall(tracer_alloc_buffers);
4624 fs_initcall(tracer_init_debugfs);
4625 late_initcall(clear_boot_tracer);