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;
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[] = {
438 { trace_clock_local, "local" },
439 { trace_clock_global, "global" },
445 * trace_parser_get_init - gets the buffer for trace parser
447 int trace_parser_get_init(struct trace_parser *parser, int size)
449 memset(parser, 0, sizeof(*parser));
451 parser->buffer = kmalloc(size, GFP_KERNEL);
460 * trace_parser_put - frees the buffer for trace parser
462 void trace_parser_put(struct trace_parser *parser)
464 kfree(parser->buffer);
468 * trace_get_user - reads the user input string separated by space
469 * (matched by isspace(ch))
471 * For each string found the 'struct trace_parser' is updated,
472 * and the function returns.
474 * Returns number of bytes read.
476 * See kernel/trace/trace.h for 'struct trace_parser' details.
478 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
479 size_t cnt, loff_t *ppos)
486 trace_parser_clear(parser);
488 ret = get_user(ch, ubuf++);
496 * The parser is not finished with the last write,
497 * continue reading the user input without skipping spaces.
500 /* skip white space */
501 while (cnt && isspace(ch)) {
502 ret = get_user(ch, ubuf++);
509 /* only spaces were written */
519 /* read the non-space input */
520 while (cnt && !isspace(ch)) {
521 if (parser->idx < parser->size - 1)
522 parser->buffer[parser->idx++] = ch;
527 ret = get_user(ch, ubuf++);
534 /* We either got finished input or we have to wait for another call. */
536 parser->buffer[parser->idx] = 0;
537 parser->cont = false;
540 parser->buffer[parser->idx++] = ch;
550 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
558 if (s->len <= s->readpos)
561 len = s->len - s->readpos;
564 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
574 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
579 if (s->len <= s->readpos)
582 len = s->len - s->readpos;
585 ret = memcpy(buf, s->buffer + s->readpos, cnt);
594 * ftrace_max_lock is used to protect the swapping of buffers
595 * when taking a max snapshot. The buffers themselves are
596 * protected by per_cpu spinlocks. But the action of the swap
597 * needs its own lock.
599 * This is defined as a arch_spinlock_t in order to help
600 * with performance when lockdep debugging is enabled.
602 * It is also used in other places outside the update_max_tr
603 * so it needs to be defined outside of the
604 * CONFIG_TRACER_MAX_TRACE.
606 static arch_spinlock_t ftrace_max_lock =
607 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
609 unsigned long __read_mostly tracing_thresh;
611 #ifdef CONFIG_TRACER_MAX_TRACE
612 unsigned long __read_mostly tracing_max_latency;
615 * Copy the new maximum trace into the separate maximum-trace
616 * structure. (this way the maximum trace is permanently saved,
617 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
620 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
622 struct trace_array_cpu *data = tr->data[cpu];
623 struct trace_array_cpu *max_data;
626 max_tr.time_start = data->preempt_timestamp;
628 max_data = max_tr.data[cpu];
629 max_data->saved_latency = tracing_max_latency;
630 max_data->critical_start = data->critical_start;
631 max_data->critical_end = data->critical_end;
633 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
634 max_data->pid = tsk->pid;
635 max_data->uid = task_uid(tsk);
636 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
637 max_data->policy = tsk->policy;
638 max_data->rt_priority = tsk->rt_priority;
640 /* record this tasks comm */
641 tracing_record_cmdline(tsk);
645 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
647 * @tsk: the task with the latency
648 * @cpu: The cpu that initiated the trace.
650 * Flip the buffers between the @tr and the max_tr and record information
651 * about which task was the cause of this latency.
654 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
656 struct ring_buffer *buf = tr->buffer;
658 if (trace_stop_count)
661 WARN_ON_ONCE(!irqs_disabled());
662 arch_spin_lock(&ftrace_max_lock);
664 tr->buffer = max_tr.buffer;
667 __update_max_tr(tr, tsk, cpu);
668 arch_spin_unlock(&ftrace_max_lock);
672 * update_max_tr_single - only copy one trace over, and reset the rest
674 * @tsk - task with the latency
675 * @cpu - the cpu of the buffer to copy.
677 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
680 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
684 if (trace_stop_count)
687 WARN_ON_ONCE(!irqs_disabled());
688 arch_spin_lock(&ftrace_max_lock);
690 ftrace_disable_cpu();
692 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
696 * We failed to swap the buffer due to a commit taking
697 * place on this CPU. We fail to record, but we reset
698 * the max trace buffer (no one writes directly to it)
699 * and flag that it failed.
701 trace_array_printk(&max_tr, _THIS_IP_,
702 "Failed to swap buffers due to commit in progress\n");
707 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
709 __update_max_tr(tr, tsk, cpu);
710 arch_spin_unlock(&ftrace_max_lock);
712 #endif /* CONFIG_TRACER_MAX_TRACE */
715 * register_tracer - register a tracer with the ftrace system.
716 * @type - the plugin for the tracer
718 * Register a new plugin tracer.
720 int register_tracer(struct tracer *type)
721 __releases(kernel_lock)
722 __acquires(kernel_lock)
728 pr_info("Tracer must have a name\n");
732 if (strlen(type->name) > MAX_TRACER_SIZE) {
733 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
738 * When this gets called we hold the BKL which means that
739 * preemption is disabled. Various trace selftests however
740 * need to disable and enable preemption for successful tests.
741 * So we drop the BKL here and grab it after the tests again.
744 mutex_lock(&trace_types_lock);
746 tracing_selftest_running = true;
748 for (t = trace_types; t; t = t->next) {
749 if (strcmp(type->name, t->name) == 0) {
751 pr_info("Tracer %s already registered\n",
759 type->set_flag = &dummy_set_flag;
761 type->flags = &dummy_tracer_flags;
763 if (!type->flags->opts)
764 type->flags->opts = dummy_tracer_opt;
765 if (!type->wait_pipe)
766 type->wait_pipe = default_wait_pipe;
769 #ifdef CONFIG_FTRACE_STARTUP_TEST
770 if (type->selftest && !tracing_selftest_disabled) {
771 struct tracer *saved_tracer = current_trace;
772 struct trace_array *tr = &global_trace;
775 * Run a selftest on this tracer.
776 * Here we reset the trace buffer, and set the current
777 * tracer to be this tracer. The tracer can then run some
778 * internal tracing to verify that everything is in order.
779 * If we fail, we do not register this tracer.
781 tracing_reset_online_cpus(tr);
783 current_trace = type;
784 /* the test is responsible for initializing and enabling */
785 pr_info("Testing tracer %s: ", type->name);
786 ret = type->selftest(type, tr);
787 /* the test is responsible for resetting too */
788 current_trace = saved_tracer;
790 printk(KERN_CONT "FAILED!\n");
793 /* Only reset on passing, to avoid touching corrupted buffers */
794 tracing_reset_online_cpus(tr);
796 printk(KERN_CONT "PASSED\n");
800 type->next = trace_types;
804 tracing_selftest_running = false;
805 mutex_unlock(&trace_types_lock);
807 if (ret || !default_bootup_tracer)
810 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
813 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
814 /* Do we want this tracer to start on bootup? */
815 tracing_set_tracer(type->name);
816 default_bootup_tracer = NULL;
817 /* disable other selftests, since this will break it. */
818 tracing_selftest_disabled = 1;
819 #ifdef CONFIG_FTRACE_STARTUP_TEST
820 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
829 void unregister_tracer(struct tracer *type)
833 mutex_lock(&trace_types_lock);
834 for (t = &trace_types; *t; t = &(*t)->next) {
838 pr_info("Tracer %s not registered\n", type->name);
844 if (type == current_trace && tracer_enabled) {
847 if (current_trace->stop)
848 current_trace->stop(&global_trace);
849 current_trace = &nop_trace;
852 mutex_unlock(&trace_types_lock);
855 static void __tracing_reset(struct ring_buffer *buffer, int cpu)
857 ftrace_disable_cpu();
858 ring_buffer_reset_cpu(buffer, cpu);
862 void tracing_reset(struct trace_array *tr, int cpu)
864 struct ring_buffer *buffer = tr->buffer;
866 ring_buffer_record_disable(buffer);
868 /* Make sure all commits have finished */
870 __tracing_reset(buffer, cpu);
872 ring_buffer_record_enable(buffer);
875 void tracing_reset_online_cpus(struct trace_array *tr)
877 struct ring_buffer *buffer = tr->buffer;
880 ring_buffer_record_disable(buffer);
882 /* Make sure all commits have finished */
885 tr->time_start = ftrace_now(tr->cpu);
887 for_each_online_cpu(cpu)
888 __tracing_reset(buffer, cpu);
890 ring_buffer_record_enable(buffer);
893 void tracing_reset_current(int cpu)
895 tracing_reset(&global_trace, cpu);
898 void tracing_reset_current_online_cpus(void)
900 tracing_reset_online_cpus(&global_trace);
903 #define SAVED_CMDLINES 128
904 #define NO_CMDLINE_MAP UINT_MAX
905 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
906 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
907 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
908 static int cmdline_idx;
909 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
911 /* temporary disable recording */
912 static atomic_t trace_record_cmdline_disabled __read_mostly;
914 static void trace_init_cmdlines(void)
916 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
917 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
921 int is_tracing_stopped(void)
923 return trace_stop_count;
927 * ftrace_off_permanent - disable all ftrace code permanently
929 * This should only be called when a serious anomally has
930 * been detected. This will turn off the function tracing,
931 * ring buffers, and other tracing utilites. It takes no
932 * locks and can be called from any context.
934 void ftrace_off_permanent(void)
936 tracing_disabled = 1;
938 tracing_off_permanent();
942 * tracing_start - quick start of the tracer
944 * If tracing is enabled but was stopped by tracing_stop,
945 * this will start the tracer back up.
947 void tracing_start(void)
949 struct ring_buffer *buffer;
952 if (tracing_disabled)
955 spin_lock_irqsave(&tracing_start_lock, flags);
956 if (--trace_stop_count) {
957 if (trace_stop_count < 0) {
958 /* Someone screwed up their debugging */
960 trace_stop_count = 0;
965 /* Prevent the buffers from switching */
966 arch_spin_lock(&ftrace_max_lock);
968 buffer = global_trace.buffer;
970 ring_buffer_record_enable(buffer);
972 buffer = max_tr.buffer;
974 ring_buffer_record_enable(buffer);
976 arch_spin_unlock(&ftrace_max_lock);
980 spin_unlock_irqrestore(&tracing_start_lock, flags);
984 * tracing_stop - quick stop of the tracer
986 * Light weight way to stop tracing. Use in conjunction with
989 void tracing_stop(void)
991 struct ring_buffer *buffer;
995 spin_lock_irqsave(&tracing_start_lock, flags);
996 if (trace_stop_count++)
999 /* Prevent the buffers from switching */
1000 arch_spin_lock(&ftrace_max_lock);
1002 buffer = global_trace.buffer;
1004 ring_buffer_record_disable(buffer);
1006 buffer = max_tr.buffer;
1008 ring_buffer_record_disable(buffer);
1010 arch_spin_unlock(&ftrace_max_lock);
1013 spin_unlock_irqrestore(&tracing_start_lock, flags);
1016 void trace_stop_cmdline_recording(void);
1018 static void trace_save_cmdline(struct task_struct *tsk)
1022 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1026 * It's not the end of the world if we don't get
1027 * the lock, but we also don't want to spin
1028 * nor do we want to disable interrupts,
1029 * so if we miss here, then better luck next time.
1031 if (!arch_spin_trylock(&trace_cmdline_lock))
1034 idx = map_pid_to_cmdline[tsk->pid];
1035 if (idx == NO_CMDLINE_MAP) {
1036 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1039 * Check whether the cmdline buffer at idx has a pid
1040 * mapped. We are going to overwrite that entry so we
1041 * need to clear the map_pid_to_cmdline. Otherwise we
1042 * would read the new comm for the old pid.
1044 pid = map_cmdline_to_pid[idx];
1045 if (pid != NO_CMDLINE_MAP)
1046 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1048 map_cmdline_to_pid[idx] = tsk->pid;
1049 map_pid_to_cmdline[tsk->pid] = idx;
1054 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1056 arch_spin_unlock(&trace_cmdline_lock);
1059 void trace_find_cmdline(int pid, char comm[])
1064 strcpy(comm, "<idle>");
1068 if (WARN_ON_ONCE(pid < 0)) {
1069 strcpy(comm, "<XXX>");
1073 if (pid > PID_MAX_DEFAULT) {
1074 strcpy(comm, "<...>");
1079 arch_spin_lock(&trace_cmdline_lock);
1080 map = map_pid_to_cmdline[pid];
1081 if (map != NO_CMDLINE_MAP)
1082 strcpy(comm, saved_cmdlines[map]);
1084 strcpy(comm, "<...>");
1086 arch_spin_unlock(&trace_cmdline_lock);
1090 void tracing_record_cmdline(struct task_struct *tsk)
1092 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1096 trace_save_cmdline(tsk);
1100 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1103 struct task_struct *tsk = current;
1105 entry->preempt_count = pc & 0xff;
1106 entry->pid = (tsk) ? tsk->pid : 0;
1107 entry->lock_depth = (tsk) ? tsk->lock_depth : 0;
1109 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1110 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1112 TRACE_FLAG_IRQS_NOSUPPORT |
1114 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1115 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1116 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1118 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1120 struct ring_buffer_event *
1121 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1124 unsigned long flags, int pc)
1126 struct ring_buffer_event *event;
1128 event = ring_buffer_lock_reserve(buffer, len);
1129 if (event != NULL) {
1130 struct trace_entry *ent = ring_buffer_event_data(event);
1132 tracing_generic_entry_update(ent, flags, pc);
1140 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1141 struct ring_buffer_event *event,
1142 unsigned long flags, int pc,
1145 ring_buffer_unlock_commit(buffer, event);
1147 ftrace_trace_stack(buffer, flags, 6, pc);
1148 ftrace_trace_userstack(buffer, flags, pc);
1154 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1155 struct ring_buffer_event *event,
1156 unsigned long flags, int pc)
1158 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1161 struct ring_buffer_event *
1162 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1163 int type, unsigned long len,
1164 unsigned long flags, int pc)
1166 *current_rb = global_trace.buffer;
1167 return trace_buffer_lock_reserve(*current_rb,
1168 type, len, flags, pc);
1170 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1172 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1173 struct ring_buffer_event *event,
1174 unsigned long flags, int pc)
1176 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1178 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1180 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1181 struct ring_buffer_event *event,
1182 unsigned long flags, int pc)
1184 __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1186 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1188 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1189 struct ring_buffer_event *event)
1191 ring_buffer_discard_commit(buffer, event);
1193 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1196 trace_function(struct trace_array *tr,
1197 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1200 struct ftrace_event_call *call = &event_function;
1201 struct ring_buffer *buffer = tr->buffer;
1202 struct ring_buffer_event *event;
1203 struct ftrace_entry *entry;
1205 /* If we are reading the ring buffer, don't trace */
1206 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1209 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1213 entry = ring_buffer_event_data(event);
1215 entry->parent_ip = parent_ip;
1217 if (!filter_check_discard(call, entry, buffer, event))
1218 ring_buffer_unlock_commit(buffer, event);
1222 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1223 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1226 if (likely(!atomic_read(&data->disabled)))
1227 trace_function(tr, ip, parent_ip, flags, pc);
1230 #ifdef CONFIG_STACKTRACE
1231 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1232 unsigned long flags,
1235 struct ftrace_event_call *call = &event_kernel_stack;
1236 struct ring_buffer_event *event;
1237 struct stack_entry *entry;
1238 struct stack_trace trace;
1240 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1241 sizeof(*entry), flags, pc);
1244 entry = ring_buffer_event_data(event);
1245 memset(&entry->caller, 0, sizeof(entry->caller));
1247 trace.nr_entries = 0;
1248 trace.max_entries = FTRACE_STACK_ENTRIES;
1250 trace.entries = entry->caller;
1252 save_stack_trace(&trace);
1253 if (!filter_check_discard(call, entry, buffer, event))
1254 ring_buffer_unlock_commit(buffer, event);
1257 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1260 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1263 __ftrace_trace_stack(buffer, flags, skip, pc);
1266 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1269 __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1273 * trace_dump_stack - record a stack back trace in the trace buffer
1275 void trace_dump_stack(void)
1277 unsigned long flags;
1279 if (tracing_disabled || tracing_selftest_running)
1282 local_save_flags(flags);
1284 /* skipping 3 traces, seems to get us at the caller of this function */
1285 __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count());
1289 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1291 struct ftrace_event_call *call = &event_user_stack;
1292 struct ring_buffer_event *event;
1293 struct userstack_entry *entry;
1294 struct stack_trace trace;
1296 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1300 * NMIs can not handle page faults, even with fix ups.
1301 * The save user stack can (and often does) fault.
1303 if (unlikely(in_nmi()))
1306 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1307 sizeof(*entry), flags, pc);
1310 entry = ring_buffer_event_data(event);
1312 entry->tgid = current->tgid;
1313 memset(&entry->caller, 0, sizeof(entry->caller));
1315 trace.nr_entries = 0;
1316 trace.max_entries = FTRACE_STACK_ENTRIES;
1318 trace.entries = entry->caller;
1320 save_stack_trace_user(&trace);
1321 if (!filter_check_discard(call, entry, buffer, event))
1322 ring_buffer_unlock_commit(buffer, event);
1326 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1328 ftrace_trace_userstack(tr, flags, preempt_count());
1332 #endif /* CONFIG_STACKTRACE */
1335 ftrace_trace_special(void *__tr,
1336 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1339 struct ftrace_event_call *call = &event_special;
1340 struct ring_buffer_event *event;
1341 struct trace_array *tr = __tr;
1342 struct ring_buffer *buffer = tr->buffer;
1343 struct special_entry *entry;
1345 event = trace_buffer_lock_reserve(buffer, TRACE_SPECIAL,
1346 sizeof(*entry), 0, pc);
1349 entry = ring_buffer_event_data(event);
1354 if (!filter_check_discard(call, entry, buffer, event))
1355 trace_buffer_unlock_commit(buffer, event, 0, pc);
1359 __trace_special(void *__tr, void *__data,
1360 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1362 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1366 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1368 struct trace_array *tr = &global_trace;
1369 struct trace_array_cpu *data;
1370 unsigned long flags;
1374 if (tracing_disabled)
1377 pc = preempt_count();
1378 local_irq_save(flags);
1379 cpu = raw_smp_processor_id();
1380 data = tr->data[cpu];
1382 if (likely(atomic_inc_return(&data->disabled) == 1))
1383 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1385 atomic_dec(&data->disabled);
1386 local_irq_restore(flags);
1390 * trace_vbprintk - write binary msg to tracing buffer
1393 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1395 static arch_spinlock_t trace_buf_lock =
1396 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1397 static u32 trace_buf[TRACE_BUF_SIZE];
1399 struct ftrace_event_call *call = &event_bprint;
1400 struct ring_buffer_event *event;
1401 struct ring_buffer *buffer;
1402 struct trace_array *tr = &global_trace;
1403 struct trace_array_cpu *data;
1404 struct bprint_entry *entry;
1405 unsigned long flags;
1407 int cpu, len = 0, size, pc;
1409 if (unlikely(tracing_selftest_running || tracing_disabled))
1412 /* Don't pollute graph traces with trace_vprintk internals */
1413 pause_graph_tracing();
1415 pc = preempt_count();
1416 preempt_disable_notrace();
1417 cpu = raw_smp_processor_id();
1418 data = tr->data[cpu];
1420 disable = atomic_inc_return(&data->disabled);
1421 if (unlikely(disable != 1))
1424 /* Lockdep uses trace_printk for lock tracing */
1425 local_irq_save(flags);
1426 arch_spin_lock(&trace_buf_lock);
1427 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1429 if (len > TRACE_BUF_SIZE || len < 0)
1432 size = sizeof(*entry) + sizeof(u32) * len;
1433 buffer = tr->buffer;
1434 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1438 entry = ring_buffer_event_data(event);
1442 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1443 if (!filter_check_discard(call, entry, buffer, event)) {
1444 ring_buffer_unlock_commit(buffer, event);
1445 ftrace_trace_stack(buffer, flags, 6, pc);
1449 arch_spin_unlock(&trace_buf_lock);
1450 local_irq_restore(flags);
1453 atomic_dec_return(&data->disabled);
1454 preempt_enable_notrace();
1455 unpause_graph_tracing();
1459 EXPORT_SYMBOL_GPL(trace_vbprintk);
1461 int trace_array_printk(struct trace_array *tr,
1462 unsigned long ip, const char *fmt, ...)
1467 if (!(trace_flags & TRACE_ITER_PRINTK))
1471 ret = trace_array_vprintk(tr, ip, fmt, ap);
1476 int trace_array_vprintk(struct trace_array *tr,
1477 unsigned long ip, const char *fmt, va_list args)
1479 static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1480 static char trace_buf[TRACE_BUF_SIZE];
1482 struct ftrace_event_call *call = &event_print;
1483 struct ring_buffer_event *event;
1484 struct ring_buffer *buffer;
1485 struct trace_array_cpu *data;
1486 int cpu, len = 0, size, pc;
1487 struct print_entry *entry;
1488 unsigned long irq_flags;
1491 if (tracing_disabled || tracing_selftest_running)
1494 pc = preempt_count();
1495 preempt_disable_notrace();
1496 cpu = raw_smp_processor_id();
1497 data = tr->data[cpu];
1499 disable = atomic_inc_return(&data->disabled);
1500 if (unlikely(disable != 1))
1503 pause_graph_tracing();
1504 raw_local_irq_save(irq_flags);
1505 arch_spin_lock(&trace_buf_lock);
1506 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1508 size = sizeof(*entry) + len + 1;
1509 buffer = tr->buffer;
1510 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1514 entry = ring_buffer_event_data(event);
1517 memcpy(&entry->buf, trace_buf, len);
1518 entry->buf[len] = '\0';
1519 if (!filter_check_discard(call, entry, buffer, event)) {
1520 ring_buffer_unlock_commit(buffer, event);
1521 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1525 arch_spin_unlock(&trace_buf_lock);
1526 raw_local_irq_restore(irq_flags);
1527 unpause_graph_tracing();
1529 atomic_dec_return(&data->disabled);
1530 preempt_enable_notrace();
1535 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1537 return trace_array_vprintk(&global_trace, ip, fmt, args);
1539 EXPORT_SYMBOL_GPL(trace_vprintk);
1541 enum trace_file_type {
1542 TRACE_FILE_LAT_FMT = 1,
1543 TRACE_FILE_ANNOTATE = 2,
1546 static void trace_iterator_increment(struct trace_iterator *iter)
1548 /* Don't allow ftrace to trace into the ring buffers */
1549 ftrace_disable_cpu();
1552 if (iter->buffer_iter[iter->cpu])
1553 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1555 ftrace_enable_cpu();
1558 static struct trace_entry *
1559 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1560 unsigned long *lost_events)
1562 struct ring_buffer_event *event;
1563 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1565 /* Don't allow ftrace to trace into the ring buffers */
1566 ftrace_disable_cpu();
1569 event = ring_buffer_iter_peek(buf_iter, ts);
1571 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1574 ftrace_enable_cpu();
1576 return event ? ring_buffer_event_data(event) : NULL;
1579 static struct trace_entry *
1580 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1581 unsigned long *missing_events, u64 *ent_ts)
1583 struct ring_buffer *buffer = iter->tr->buffer;
1584 struct trace_entry *ent, *next = NULL;
1585 unsigned long lost_events = 0, next_lost = 0;
1586 int cpu_file = iter->cpu_file;
1587 u64 next_ts = 0, ts;
1592 * If we are in a per_cpu trace file, don't bother by iterating over
1593 * all cpu and peek directly.
1595 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1596 if (ring_buffer_empty_cpu(buffer, cpu_file))
1598 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1600 *ent_cpu = cpu_file;
1605 for_each_tracing_cpu(cpu) {
1607 if (ring_buffer_empty_cpu(buffer, cpu))
1610 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1613 * Pick the entry with the smallest timestamp:
1615 if (ent && (!next || ts < next_ts)) {
1619 next_lost = lost_events;
1624 *ent_cpu = next_cpu;
1630 *missing_events = next_lost;
1635 /* Find the next real entry, without updating the iterator itself */
1636 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1637 int *ent_cpu, u64 *ent_ts)
1639 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1642 /* Find the next real entry, and increment the iterator to the next entry */
1643 static void *find_next_entry_inc(struct trace_iterator *iter)
1645 iter->ent = __find_next_entry(iter, &iter->cpu,
1646 &iter->lost_events, &iter->ts);
1649 trace_iterator_increment(iter);
1651 return iter->ent ? iter : NULL;
1654 static void trace_consume(struct trace_iterator *iter)
1656 /* Don't allow ftrace to trace into the ring buffers */
1657 ftrace_disable_cpu();
1658 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1659 &iter->lost_events);
1660 ftrace_enable_cpu();
1663 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1665 struct trace_iterator *iter = m->private;
1669 WARN_ON_ONCE(iter->leftover);
1673 /* can't go backwards */
1678 ent = find_next_entry_inc(iter);
1682 while (ent && iter->idx < i)
1683 ent = find_next_entry_inc(iter);
1690 static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1692 struct trace_array *tr = iter->tr;
1693 struct ring_buffer_event *event;
1694 struct ring_buffer_iter *buf_iter;
1695 unsigned long entries = 0;
1698 tr->data[cpu]->skipped_entries = 0;
1700 if (!iter->buffer_iter[cpu])
1703 buf_iter = iter->buffer_iter[cpu];
1704 ring_buffer_iter_reset(buf_iter);
1707 * We could have the case with the max latency tracers
1708 * that a reset never took place on a cpu. This is evident
1709 * by the timestamp being before the start of the buffer.
1711 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1712 if (ts >= iter->tr->time_start)
1715 ring_buffer_read(buf_iter, NULL);
1718 tr->data[cpu]->skipped_entries = entries;
1722 * The current tracer is copied to avoid a global locking
1725 static void *s_start(struct seq_file *m, loff_t *pos)
1727 struct trace_iterator *iter = m->private;
1728 static struct tracer *old_tracer;
1729 int cpu_file = iter->cpu_file;
1734 /* copy the tracer to avoid using a global lock all around */
1735 mutex_lock(&trace_types_lock);
1736 if (unlikely(old_tracer != current_trace && current_trace)) {
1737 old_tracer = current_trace;
1738 *iter->trace = *current_trace;
1740 mutex_unlock(&trace_types_lock);
1742 atomic_inc(&trace_record_cmdline_disabled);
1744 if (*pos != iter->pos) {
1749 ftrace_disable_cpu();
1751 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1752 for_each_tracing_cpu(cpu)
1753 tracing_iter_reset(iter, cpu);
1755 tracing_iter_reset(iter, cpu_file);
1757 ftrace_enable_cpu();
1760 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1765 * If we overflowed the seq_file before, then we want
1766 * to just reuse the trace_seq buffer again.
1772 p = s_next(m, p, &l);
1776 trace_event_read_lock();
1777 trace_access_lock(cpu_file);
1781 static void s_stop(struct seq_file *m, void *p)
1783 struct trace_iterator *iter = m->private;
1785 atomic_dec(&trace_record_cmdline_disabled);
1786 trace_access_unlock(iter->cpu_file);
1787 trace_event_read_unlock();
1790 static void print_lat_help_header(struct seq_file *m)
1792 seq_puts(m, "# _------=> CPU# \n");
1793 seq_puts(m, "# / _-----=> irqs-off \n");
1794 seq_puts(m, "# | / _----=> need-resched \n");
1795 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1796 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1797 seq_puts(m, "# |||| /_--=> lock-depth \n");
1798 seq_puts(m, "# |||||/ delay \n");
1799 seq_puts(m, "# cmd pid |||||| time | caller \n");
1800 seq_puts(m, "# \\ / |||||| \\ | / \n");
1803 static void print_func_help_header(struct seq_file *m)
1805 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1806 seq_puts(m, "# | | | | |\n");
1811 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1813 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1814 struct trace_array *tr = iter->tr;
1815 struct trace_array_cpu *data = tr->data[tr->cpu];
1816 struct tracer *type = current_trace;
1817 unsigned long entries = 0;
1818 unsigned long total = 0;
1819 unsigned long count;
1820 const char *name = "preemption";
1827 for_each_tracing_cpu(cpu) {
1828 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1830 * If this buffer has skipped entries, then we hold all
1831 * entries for the trace and we need to ignore the
1832 * ones before the time stamp.
1834 if (tr->data[cpu]->skipped_entries) {
1835 count -= tr->data[cpu]->skipped_entries;
1836 /* total is the same as the entries */
1840 ring_buffer_overrun_cpu(tr->buffer, cpu);
1844 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1846 seq_puts(m, "# -----------------------------------"
1847 "---------------------------------\n");
1848 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1849 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1850 nsecs_to_usecs(data->saved_latency),
1854 #if defined(CONFIG_PREEMPT_NONE)
1856 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1858 #elif defined(CONFIG_PREEMPT)
1863 /* These are reserved for later use */
1866 seq_printf(m, " #P:%d)\n", num_online_cpus());
1870 seq_puts(m, "# -----------------\n");
1871 seq_printf(m, "# | task: %.16s-%d "
1872 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1873 data->comm, data->pid, data->uid, data->nice,
1874 data->policy, data->rt_priority);
1875 seq_puts(m, "# -----------------\n");
1877 if (data->critical_start) {
1878 seq_puts(m, "# => started at: ");
1879 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1880 trace_print_seq(m, &iter->seq);
1881 seq_puts(m, "\n# => ended at: ");
1882 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1883 trace_print_seq(m, &iter->seq);
1884 seq_puts(m, "\n#\n");
1890 static void test_cpu_buff_start(struct trace_iterator *iter)
1892 struct trace_seq *s = &iter->seq;
1894 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1897 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1900 if (cpumask_test_cpu(iter->cpu, iter->started))
1903 if (iter->tr->data[iter->cpu]->skipped_entries)
1906 cpumask_set_cpu(iter->cpu, iter->started);
1908 /* Don't print started cpu buffer for the first entry of the trace */
1910 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1914 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1916 struct trace_seq *s = &iter->seq;
1917 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1918 struct trace_entry *entry;
1919 struct trace_event *event;
1923 test_cpu_buff_start(iter);
1925 event = ftrace_find_event(entry->type);
1927 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1928 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1929 if (!trace_print_lat_context(iter))
1932 if (!trace_print_context(iter))
1938 return event->funcs->trace(iter, sym_flags, event);
1940 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1943 return TRACE_TYPE_HANDLED;
1945 return TRACE_TYPE_PARTIAL_LINE;
1948 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1950 struct trace_seq *s = &iter->seq;
1951 struct trace_entry *entry;
1952 struct trace_event *event;
1956 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1957 if (!trace_seq_printf(s, "%d %d %llu ",
1958 entry->pid, iter->cpu, iter->ts))
1962 event = ftrace_find_event(entry->type);
1964 return event->funcs->raw(iter, 0, event);
1966 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1969 return TRACE_TYPE_HANDLED;
1971 return TRACE_TYPE_PARTIAL_LINE;
1974 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1976 struct trace_seq *s = &iter->seq;
1977 unsigned char newline = '\n';
1978 struct trace_entry *entry;
1979 struct trace_event *event;
1983 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1984 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1985 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1986 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1989 event = ftrace_find_event(entry->type);
1991 enum print_line_t ret = event->funcs->hex(iter, 0, event);
1992 if (ret != TRACE_TYPE_HANDLED)
1996 SEQ_PUT_FIELD_RET(s, newline);
1998 return TRACE_TYPE_HANDLED;
2001 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2003 struct trace_seq *s = &iter->seq;
2004 struct trace_entry *entry;
2005 struct trace_event *event;
2009 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2010 SEQ_PUT_FIELD_RET(s, entry->pid);
2011 SEQ_PUT_FIELD_RET(s, iter->cpu);
2012 SEQ_PUT_FIELD_RET(s, iter->ts);
2015 event = ftrace_find_event(entry->type);
2016 return event ? event->funcs->binary(iter, 0, event) :
2020 int trace_empty(struct trace_iterator *iter)
2024 /* If we are looking at one CPU buffer, only check that one */
2025 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
2026 cpu = iter->cpu_file;
2027 if (iter->buffer_iter[cpu]) {
2028 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2031 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2037 for_each_tracing_cpu(cpu) {
2038 if (iter->buffer_iter[cpu]) {
2039 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2042 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2050 /* Called with trace_event_read_lock() held. */
2051 static enum print_line_t print_trace_line(struct trace_iterator *iter)
2053 enum print_line_t ret;
2055 if (iter->lost_events)
2056 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2057 iter->cpu, iter->lost_events);
2059 if (iter->trace && iter->trace->print_line) {
2060 ret = iter->trace->print_line(iter);
2061 if (ret != TRACE_TYPE_UNHANDLED)
2065 if (iter->ent->type == TRACE_BPRINT &&
2066 trace_flags & TRACE_ITER_PRINTK &&
2067 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2068 return trace_print_bprintk_msg_only(iter);
2070 if (iter->ent->type == TRACE_PRINT &&
2071 trace_flags & TRACE_ITER_PRINTK &&
2072 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2073 return trace_print_printk_msg_only(iter);
2075 if (trace_flags & TRACE_ITER_BIN)
2076 return print_bin_fmt(iter);
2078 if (trace_flags & TRACE_ITER_HEX)
2079 return print_hex_fmt(iter);
2081 if (trace_flags & TRACE_ITER_RAW)
2082 return print_raw_fmt(iter);
2084 return print_trace_fmt(iter);
2087 void trace_default_header(struct seq_file *m)
2089 struct trace_iterator *iter = m->private;
2091 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2092 /* print nothing if the buffers are empty */
2093 if (trace_empty(iter))
2095 print_trace_header(m, iter);
2096 if (!(trace_flags & TRACE_ITER_VERBOSE))
2097 print_lat_help_header(m);
2099 if (!(trace_flags & TRACE_ITER_VERBOSE))
2100 print_func_help_header(m);
2104 static int s_show(struct seq_file *m, void *v)
2106 struct trace_iterator *iter = v;
2109 if (iter->ent == NULL) {
2111 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2114 if (iter->trace && iter->trace->print_header)
2115 iter->trace->print_header(m);
2117 trace_default_header(m);
2119 } else if (iter->leftover) {
2121 * If we filled the seq_file buffer earlier, we
2122 * want to just show it now.
2124 ret = trace_print_seq(m, &iter->seq);
2126 /* ret should this time be zero, but you never know */
2127 iter->leftover = ret;
2130 print_trace_line(iter);
2131 ret = trace_print_seq(m, &iter->seq);
2133 * If we overflow the seq_file buffer, then it will
2134 * ask us for this data again at start up.
2136 * ret is 0 if seq_file write succeeded.
2139 iter->leftover = ret;
2145 static const struct seq_operations tracer_seq_ops = {
2152 static struct trace_iterator *
2153 __tracing_open(struct inode *inode, struct file *file)
2155 long cpu_file = (long) inode->i_private;
2156 void *fail_ret = ERR_PTR(-ENOMEM);
2157 struct trace_iterator *iter;
2161 if (tracing_disabled)
2162 return ERR_PTR(-ENODEV);
2164 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2166 return ERR_PTR(-ENOMEM);
2169 * We make a copy of the current tracer to avoid concurrent
2170 * changes on it while we are reading.
2172 mutex_lock(&trace_types_lock);
2173 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2178 *iter->trace = *current_trace;
2180 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2183 if (current_trace && current_trace->print_max)
2186 iter->tr = &global_trace;
2188 mutex_init(&iter->mutex);
2189 iter->cpu_file = cpu_file;
2191 /* Notify the tracer early; before we stop tracing. */
2192 if (iter->trace && iter->trace->open)
2193 iter->trace->open(iter);
2195 /* Annotate start of buffers if we had overruns */
2196 if (ring_buffer_overruns(iter->tr->buffer))
2197 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2199 /* stop the trace while dumping */
2202 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2203 for_each_tracing_cpu(cpu) {
2204 iter->buffer_iter[cpu] =
2205 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2207 ring_buffer_read_prepare_sync();
2208 for_each_tracing_cpu(cpu) {
2209 ring_buffer_read_start(iter->buffer_iter[cpu]);
2210 tracing_iter_reset(iter, cpu);
2213 cpu = iter->cpu_file;
2214 iter->buffer_iter[cpu] =
2215 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2216 ring_buffer_read_prepare_sync();
2217 ring_buffer_read_start(iter->buffer_iter[cpu]);
2218 tracing_iter_reset(iter, cpu);
2221 ret = seq_open(file, &tracer_seq_ops);
2223 fail_ret = ERR_PTR(ret);
2227 m = file->private_data;
2230 mutex_unlock(&trace_types_lock);
2235 for_each_tracing_cpu(cpu) {
2236 if (iter->buffer_iter[cpu])
2237 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2239 free_cpumask_var(iter->started);
2242 mutex_unlock(&trace_types_lock);
2249 int tracing_open_generic(struct inode *inode, struct file *filp)
2251 if (tracing_disabled)
2254 filp->private_data = inode->i_private;
2258 static int tracing_release(struct inode *inode, struct file *file)
2260 struct seq_file *m = (struct seq_file *)file->private_data;
2261 struct trace_iterator *iter;
2264 if (!(file->f_mode & FMODE_READ))
2269 mutex_lock(&trace_types_lock);
2270 for_each_tracing_cpu(cpu) {
2271 if (iter->buffer_iter[cpu])
2272 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2275 if (iter->trace && iter->trace->close)
2276 iter->trace->close(iter);
2278 /* reenable tracing if it was previously enabled */
2280 mutex_unlock(&trace_types_lock);
2282 seq_release(inode, file);
2283 mutex_destroy(&iter->mutex);
2284 free_cpumask_var(iter->started);
2290 static int tracing_open(struct inode *inode, struct file *file)
2292 struct trace_iterator *iter;
2295 /* If this file was open for write, then erase contents */
2296 if ((file->f_mode & FMODE_WRITE) &&
2297 (file->f_flags & O_TRUNC)) {
2298 long cpu = (long) inode->i_private;
2300 if (cpu == TRACE_PIPE_ALL_CPU)
2301 tracing_reset_online_cpus(&global_trace);
2303 tracing_reset(&global_trace, cpu);
2306 if (file->f_mode & FMODE_READ) {
2307 iter = __tracing_open(inode, file);
2309 ret = PTR_ERR(iter);
2310 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2311 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2317 t_next(struct seq_file *m, void *v, loff_t *pos)
2319 struct tracer *t = v;
2329 static void *t_start(struct seq_file *m, loff_t *pos)
2334 mutex_lock(&trace_types_lock);
2335 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2341 static void t_stop(struct seq_file *m, void *p)
2343 mutex_unlock(&trace_types_lock);
2346 static int t_show(struct seq_file *m, void *v)
2348 struct tracer *t = v;
2353 seq_printf(m, "%s", t->name);
2362 static const struct seq_operations show_traces_seq_ops = {
2369 static int show_traces_open(struct inode *inode, struct file *file)
2371 if (tracing_disabled)
2374 return seq_open(file, &show_traces_seq_ops);
2378 tracing_write_stub(struct file *filp, const char __user *ubuf,
2379 size_t count, loff_t *ppos)
2384 static const struct file_operations tracing_fops = {
2385 .open = tracing_open,
2387 .write = tracing_write_stub,
2388 .llseek = seq_lseek,
2389 .release = tracing_release,
2392 static const struct file_operations show_traces_fops = {
2393 .open = show_traces_open,
2395 .release = seq_release,
2399 * Only trace on a CPU if the bitmask is set:
2401 static cpumask_var_t tracing_cpumask;
2404 * The tracer itself will not take this lock, but still we want
2405 * to provide a consistent cpumask to user-space:
2407 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2410 * Temporary storage for the character representation of the
2411 * CPU bitmask (and one more byte for the newline):
2413 static char mask_str[NR_CPUS + 1];
2416 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2417 size_t count, loff_t *ppos)
2421 mutex_lock(&tracing_cpumask_update_lock);
2423 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2424 if (count - len < 2) {
2428 len += sprintf(mask_str + len, "\n");
2429 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2432 mutex_unlock(&tracing_cpumask_update_lock);
2438 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2439 size_t count, loff_t *ppos)
2442 cpumask_var_t tracing_cpumask_new;
2444 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2447 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2451 mutex_lock(&tracing_cpumask_update_lock);
2453 local_irq_disable();
2454 arch_spin_lock(&ftrace_max_lock);
2455 for_each_tracing_cpu(cpu) {
2457 * Increase/decrease the disabled counter if we are
2458 * about to flip a bit in the cpumask:
2460 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2461 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2462 atomic_inc(&global_trace.data[cpu]->disabled);
2464 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2465 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2466 atomic_dec(&global_trace.data[cpu]->disabled);
2469 arch_spin_unlock(&ftrace_max_lock);
2472 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2474 mutex_unlock(&tracing_cpumask_update_lock);
2475 free_cpumask_var(tracing_cpumask_new);
2480 free_cpumask_var(tracing_cpumask_new);
2485 static const struct file_operations tracing_cpumask_fops = {
2486 .open = tracing_open_generic,
2487 .read = tracing_cpumask_read,
2488 .write = tracing_cpumask_write,
2491 static int tracing_trace_options_show(struct seq_file *m, void *v)
2493 struct tracer_opt *trace_opts;
2497 mutex_lock(&trace_types_lock);
2498 tracer_flags = current_trace->flags->val;
2499 trace_opts = current_trace->flags->opts;
2501 for (i = 0; trace_options[i]; i++) {
2502 if (trace_flags & (1 << i))
2503 seq_printf(m, "%s\n", trace_options[i]);
2505 seq_printf(m, "no%s\n", trace_options[i]);
2508 for (i = 0; trace_opts[i].name; i++) {
2509 if (tracer_flags & trace_opts[i].bit)
2510 seq_printf(m, "%s\n", trace_opts[i].name);
2512 seq_printf(m, "no%s\n", trace_opts[i].name);
2514 mutex_unlock(&trace_types_lock);
2519 static int __set_tracer_option(struct tracer *trace,
2520 struct tracer_flags *tracer_flags,
2521 struct tracer_opt *opts, int neg)
2525 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2530 tracer_flags->val &= ~opts->bit;
2532 tracer_flags->val |= opts->bit;
2536 /* Try to assign a tracer specific option */
2537 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2539 struct tracer_flags *tracer_flags = trace->flags;
2540 struct tracer_opt *opts = NULL;
2543 for (i = 0; tracer_flags->opts[i].name; i++) {
2544 opts = &tracer_flags->opts[i];
2546 if (strcmp(cmp, opts->name) == 0)
2547 return __set_tracer_option(trace, trace->flags,
2554 static void set_tracer_flags(unsigned int mask, int enabled)
2556 /* do nothing if flag is already set */
2557 if (!!(trace_flags & mask) == !!enabled)
2561 trace_flags |= mask;
2563 trace_flags &= ~mask;
2567 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2568 size_t cnt, loff_t *ppos)
2576 if (cnt >= sizeof(buf))
2579 if (copy_from_user(&buf, ubuf, cnt))
2583 cmp = strstrip(buf);
2585 if (strncmp(cmp, "no", 2) == 0) {
2590 for (i = 0; trace_options[i]; i++) {
2591 if (strcmp(cmp, trace_options[i]) == 0) {
2592 set_tracer_flags(1 << i, !neg);
2597 /* If no option could be set, test the specific tracer options */
2598 if (!trace_options[i]) {
2599 mutex_lock(&trace_types_lock);
2600 ret = set_tracer_option(current_trace, cmp, neg);
2601 mutex_unlock(&trace_types_lock);
2611 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2613 if (tracing_disabled)
2615 return single_open(file, tracing_trace_options_show, NULL);
2618 static const struct file_operations tracing_iter_fops = {
2619 .open = tracing_trace_options_open,
2621 .llseek = seq_lseek,
2622 .release = single_release,
2623 .write = tracing_trace_options_write,
2626 static const char readme_msg[] =
2627 "tracing mini-HOWTO:\n\n"
2628 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2629 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2630 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2631 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2633 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2634 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2636 "# cat /sys/kernel/debug/tracing/trace_options\n"
2637 "noprint-parent nosym-offset nosym-addr noverbose\n"
2638 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2639 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2640 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2641 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2645 tracing_readme_read(struct file *filp, char __user *ubuf,
2646 size_t cnt, loff_t *ppos)
2648 return simple_read_from_buffer(ubuf, cnt, ppos,
2649 readme_msg, strlen(readme_msg));
2652 static const struct file_operations tracing_readme_fops = {
2653 .open = tracing_open_generic,
2654 .read = tracing_readme_read,
2658 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2659 size_t cnt, loff_t *ppos)
2668 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2672 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2680 for (i = 0; i < SAVED_CMDLINES; i++) {
2683 pid = map_cmdline_to_pid[i];
2684 if (pid == -1 || pid == NO_CMDLINE_MAP)
2687 trace_find_cmdline(pid, buf_comm);
2688 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2693 len = simple_read_from_buffer(ubuf, cnt, ppos,
2702 static const struct file_operations tracing_saved_cmdlines_fops = {
2703 .open = tracing_open_generic,
2704 .read = tracing_saved_cmdlines_read,
2708 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2709 size_t cnt, loff_t *ppos)
2714 r = sprintf(buf, "%u\n", tracer_enabled);
2715 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2719 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2720 size_t cnt, loff_t *ppos)
2722 struct trace_array *tr = filp->private_data;
2727 if (cnt >= sizeof(buf))
2730 if (copy_from_user(&buf, ubuf, cnt))
2735 ret = strict_strtoul(buf, 10, &val);
2741 mutex_lock(&trace_types_lock);
2742 if (tracer_enabled ^ val) {
2745 if (current_trace->start)
2746 current_trace->start(tr);
2751 if (current_trace->stop)
2752 current_trace->stop(tr);
2755 mutex_unlock(&trace_types_lock);
2763 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2764 size_t cnt, loff_t *ppos)
2766 char buf[MAX_TRACER_SIZE+2];
2769 mutex_lock(&trace_types_lock);
2771 r = sprintf(buf, "%s\n", current_trace->name);
2773 r = sprintf(buf, "\n");
2774 mutex_unlock(&trace_types_lock);
2776 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2779 int tracer_init(struct tracer *t, struct trace_array *tr)
2781 tracing_reset_online_cpus(tr);
2785 static int tracing_resize_ring_buffer(unsigned long size)
2790 * If kernel or user changes the size of the ring buffer
2791 * we use the size that was given, and we can forget about
2792 * expanding it later.
2794 ring_buffer_expanded = 1;
2796 ret = ring_buffer_resize(global_trace.buffer, size);
2800 ret = ring_buffer_resize(max_tr.buffer, size);
2804 r = ring_buffer_resize(global_trace.buffer,
2805 global_trace.entries);
2808 * AARGH! We are left with different
2809 * size max buffer!!!!
2810 * The max buffer is our "snapshot" buffer.
2811 * When a tracer needs a snapshot (one of the
2812 * latency tracers), it swaps the max buffer
2813 * with the saved snap shot. We succeeded to
2814 * update the size of the main buffer, but failed to
2815 * update the size of the max buffer. But when we tried
2816 * to reset the main buffer to the original size, we
2817 * failed there too. This is very unlikely to
2818 * happen, but if it does, warn and kill all
2822 tracing_disabled = 1;
2827 global_trace.entries = size;
2833 * tracing_update_buffers - used by tracing facility to expand ring buffers
2835 * To save on memory when the tracing is never used on a system with it
2836 * configured in. The ring buffers are set to a minimum size. But once
2837 * a user starts to use the tracing facility, then they need to grow
2838 * to their default size.
2840 * This function is to be called when a tracer is about to be used.
2842 int tracing_update_buffers(void)
2846 mutex_lock(&trace_types_lock);
2847 if (!ring_buffer_expanded)
2848 ret = tracing_resize_ring_buffer(trace_buf_size);
2849 mutex_unlock(&trace_types_lock);
2854 struct trace_option_dentry;
2856 static struct trace_option_dentry *
2857 create_trace_option_files(struct tracer *tracer);
2860 destroy_trace_option_files(struct trace_option_dentry *topts);
2862 static int tracing_set_tracer(const char *buf)
2864 static struct trace_option_dentry *topts;
2865 struct trace_array *tr = &global_trace;
2869 mutex_lock(&trace_types_lock);
2871 if (!ring_buffer_expanded) {
2872 ret = tracing_resize_ring_buffer(trace_buf_size);
2878 for (t = trace_types; t; t = t->next) {
2879 if (strcmp(t->name, buf) == 0)
2886 if (t == current_trace)
2889 trace_branch_disable();
2890 if (current_trace && current_trace->reset)
2891 current_trace->reset(tr);
2893 destroy_trace_option_files(topts);
2897 topts = create_trace_option_files(current_trace);
2900 ret = tracer_init(t, tr);
2905 trace_branch_enable(tr);
2907 mutex_unlock(&trace_types_lock);
2913 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2914 size_t cnt, loff_t *ppos)
2916 char buf[MAX_TRACER_SIZE+1];
2923 if (cnt > MAX_TRACER_SIZE)
2924 cnt = MAX_TRACER_SIZE;
2926 if (copy_from_user(&buf, ubuf, cnt))
2931 /* strip ending whitespace. */
2932 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2935 err = tracing_set_tracer(buf);
2945 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2946 size_t cnt, loff_t *ppos)
2948 unsigned long *ptr = filp->private_data;
2952 r = snprintf(buf, sizeof(buf), "%ld\n",
2953 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2954 if (r > sizeof(buf))
2956 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2960 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2961 size_t cnt, loff_t *ppos)
2963 unsigned long *ptr = filp->private_data;
2968 if (cnt >= sizeof(buf))
2971 if (copy_from_user(&buf, ubuf, cnt))
2976 ret = strict_strtoul(buf, 10, &val);
2985 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2987 long cpu_file = (long) inode->i_private;
2988 struct trace_iterator *iter;
2991 if (tracing_disabled)
2994 mutex_lock(&trace_types_lock);
2996 /* create a buffer to store the information to pass to userspace */
2997 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3004 * We make a copy of the current tracer to avoid concurrent
3005 * changes on it while we are reading.
3007 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3013 *iter->trace = *current_trace;
3015 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3020 /* trace pipe does not show start of buffer */
3021 cpumask_setall(iter->started);
3023 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3024 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3026 iter->cpu_file = cpu_file;
3027 iter->tr = &global_trace;
3028 mutex_init(&iter->mutex);
3029 filp->private_data = iter;
3031 if (iter->trace->pipe_open)
3032 iter->trace->pipe_open(iter);
3035 mutex_unlock(&trace_types_lock);
3041 mutex_unlock(&trace_types_lock);
3045 static int tracing_release_pipe(struct inode *inode, struct file *file)
3047 struct trace_iterator *iter = file->private_data;
3049 mutex_lock(&trace_types_lock);
3051 if (iter->trace->pipe_close)
3052 iter->trace->pipe_close(iter);
3054 mutex_unlock(&trace_types_lock);
3056 free_cpumask_var(iter->started);
3057 mutex_destroy(&iter->mutex);
3065 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3067 struct trace_iterator *iter = filp->private_data;
3069 if (trace_flags & TRACE_ITER_BLOCK) {
3071 * Always select as readable when in blocking mode
3073 return POLLIN | POLLRDNORM;
3075 if (!trace_empty(iter))
3076 return POLLIN | POLLRDNORM;
3077 poll_wait(filp, &trace_wait, poll_table);
3078 if (!trace_empty(iter))
3079 return POLLIN | POLLRDNORM;
3086 void default_wait_pipe(struct trace_iterator *iter)
3090 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3092 if (trace_empty(iter))
3095 finish_wait(&trace_wait, &wait);
3099 * This is a make-shift waitqueue.
3100 * A tracer might use this callback on some rare cases:
3102 * 1) the current tracer might hold the runqueue lock when it wakes up
3103 * a reader, hence a deadlock (sched, function, and function graph tracers)
3104 * 2) the function tracers, trace all functions, we don't want
3105 * the overhead of calling wake_up and friends
3106 * (and tracing them too)
3108 * Anyway, this is really very primitive wakeup.
3110 void poll_wait_pipe(struct trace_iterator *iter)
3112 set_current_state(TASK_INTERRUPTIBLE);
3113 /* sleep for 100 msecs, and try again. */
3114 schedule_timeout(HZ / 10);
3117 /* Must be called with trace_types_lock mutex held. */
3118 static int tracing_wait_pipe(struct file *filp)
3120 struct trace_iterator *iter = filp->private_data;
3122 while (trace_empty(iter)) {
3124 if ((filp->f_flags & O_NONBLOCK)) {
3128 mutex_unlock(&iter->mutex);
3130 iter->trace->wait_pipe(iter);
3132 mutex_lock(&iter->mutex);
3134 if (signal_pending(current))
3138 * We block until we read something and tracing is disabled.
3139 * We still block if tracing is disabled, but we have never
3140 * read anything. This allows a user to cat this file, and
3141 * then enable tracing. But after we have read something,
3142 * we give an EOF when tracing is again disabled.
3144 * iter->pos will be 0 if we haven't read anything.
3146 if (!tracer_enabled && iter->pos)
3157 tracing_read_pipe(struct file *filp, char __user *ubuf,
3158 size_t cnt, loff_t *ppos)
3160 struct trace_iterator *iter = filp->private_data;
3161 static struct tracer *old_tracer;
3164 /* return any leftover data */
3165 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3169 trace_seq_init(&iter->seq);
3171 /* copy the tracer to avoid using a global lock all around */
3172 mutex_lock(&trace_types_lock);
3173 if (unlikely(old_tracer != current_trace && current_trace)) {
3174 old_tracer = current_trace;
3175 *iter->trace = *current_trace;
3177 mutex_unlock(&trace_types_lock);
3180 * Avoid more than one consumer on a single file descriptor
3181 * This is just a matter of traces coherency, the ring buffer itself
3184 mutex_lock(&iter->mutex);
3185 if (iter->trace->read) {
3186 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3192 sret = tracing_wait_pipe(filp);
3196 /* stop when tracing is finished */
3197 if (trace_empty(iter)) {
3202 if (cnt >= PAGE_SIZE)
3203 cnt = PAGE_SIZE - 1;
3205 /* reset all but tr, trace, and overruns */
3206 memset(&iter->seq, 0,
3207 sizeof(struct trace_iterator) -
3208 offsetof(struct trace_iterator, seq));
3211 trace_event_read_lock();
3212 trace_access_lock(iter->cpu_file);
3213 while (find_next_entry_inc(iter) != NULL) {
3214 enum print_line_t ret;
3215 int len = iter->seq.len;
3217 ret = print_trace_line(iter);
3218 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3219 /* don't print partial lines */
3220 iter->seq.len = len;
3223 if (ret != TRACE_TYPE_NO_CONSUME)
3224 trace_consume(iter);
3226 if (iter->seq.len >= cnt)
3229 trace_access_unlock(iter->cpu_file);
3230 trace_event_read_unlock();
3232 /* Now copy what we have to the user */
3233 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3234 if (iter->seq.readpos >= iter->seq.len)
3235 trace_seq_init(&iter->seq);
3238 * If there was nothing to send to user, inspite of consuming trace
3239 * entries, go back to wait for more entries.
3245 mutex_unlock(&iter->mutex);
3250 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3251 struct pipe_buffer *buf)
3253 __free_page(buf->page);
3256 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3259 __free_page(spd->pages[idx]);
3262 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3264 .map = generic_pipe_buf_map,
3265 .unmap = generic_pipe_buf_unmap,
3266 .confirm = generic_pipe_buf_confirm,
3267 .release = tracing_pipe_buf_release,
3268 .steal = generic_pipe_buf_steal,
3269 .get = generic_pipe_buf_get,
3273 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3278 /* Seq buffer is page-sized, exactly what we need. */
3280 count = iter->seq.len;
3281 ret = print_trace_line(iter);
3282 count = iter->seq.len - count;
3285 iter->seq.len -= count;
3288 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3289 iter->seq.len -= count;
3293 if (ret != TRACE_TYPE_NO_CONSUME)
3294 trace_consume(iter);
3296 if (!find_next_entry_inc(iter)) {
3306 static ssize_t tracing_splice_read_pipe(struct file *filp,
3308 struct pipe_inode_info *pipe,
3312 struct page *pages_def[PIPE_DEF_BUFFERS];
3313 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3314 struct trace_iterator *iter = filp->private_data;
3315 struct splice_pipe_desc spd = {
3317 .partial = partial_def,
3318 .nr_pages = 0, /* This gets updated below. */
3320 .ops = &tracing_pipe_buf_ops,
3321 .spd_release = tracing_spd_release_pipe,
3323 static struct tracer *old_tracer;
3328 if (splice_grow_spd(pipe, &spd))
3331 /* copy the tracer to avoid using a global lock all around */
3332 mutex_lock(&trace_types_lock);
3333 if (unlikely(old_tracer != current_trace && current_trace)) {
3334 old_tracer = current_trace;
3335 *iter->trace = *current_trace;
3337 mutex_unlock(&trace_types_lock);
3339 mutex_lock(&iter->mutex);
3341 if (iter->trace->splice_read) {
3342 ret = iter->trace->splice_read(iter, filp,
3343 ppos, pipe, len, flags);
3348 ret = tracing_wait_pipe(filp);
3352 if (!iter->ent && !find_next_entry_inc(iter)) {
3357 trace_event_read_lock();
3358 trace_access_lock(iter->cpu_file);
3360 /* Fill as many pages as possible. */
3361 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3362 spd.pages[i] = alloc_page(GFP_KERNEL);
3366 rem = tracing_fill_pipe_page(rem, iter);
3368 /* Copy the data into the page, so we can start over. */
3369 ret = trace_seq_to_buffer(&iter->seq,
3370 page_address(spd.pages[i]),
3373 __free_page(spd.pages[i]);
3376 spd.partial[i].offset = 0;
3377 spd.partial[i].len = iter->seq.len;
3379 trace_seq_init(&iter->seq);
3382 trace_access_unlock(iter->cpu_file);
3383 trace_event_read_unlock();
3384 mutex_unlock(&iter->mutex);
3388 ret = splice_to_pipe(pipe, &spd);
3390 splice_shrink_spd(pipe, &spd);
3394 mutex_unlock(&iter->mutex);
3399 tracing_entries_read(struct file *filp, char __user *ubuf,
3400 size_t cnt, loff_t *ppos)
3402 struct trace_array *tr = filp->private_data;
3406 mutex_lock(&trace_types_lock);
3407 if (!ring_buffer_expanded)
3408 r = sprintf(buf, "%lu (expanded: %lu)\n",
3410 trace_buf_size >> 10);
3412 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3413 mutex_unlock(&trace_types_lock);
3415 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3419 tracing_entries_write(struct file *filp, const char __user *ubuf,
3420 size_t cnt, loff_t *ppos)
3426 if (cnt >= sizeof(buf))
3429 if (copy_from_user(&buf, ubuf, cnt))
3434 ret = strict_strtoul(buf, 10, &val);
3438 /* must have at least 1 entry */
3442 mutex_lock(&trace_types_lock);
3446 /* disable all cpu buffers */
3447 for_each_tracing_cpu(cpu) {
3448 if (global_trace.data[cpu])
3449 atomic_inc(&global_trace.data[cpu]->disabled);
3450 if (max_tr.data[cpu])
3451 atomic_inc(&max_tr.data[cpu]->disabled);
3454 /* value is in KB */
3457 if (val != global_trace.entries) {
3458 ret = tracing_resize_ring_buffer(val);
3467 /* If check pages failed, return ENOMEM */
3468 if (tracing_disabled)
3471 for_each_tracing_cpu(cpu) {
3472 if (global_trace.data[cpu])
3473 atomic_dec(&global_trace.data[cpu]->disabled);
3474 if (max_tr.data[cpu])
3475 atomic_dec(&max_tr.data[cpu]->disabled);
3479 max_tr.entries = global_trace.entries;
3480 mutex_unlock(&trace_types_lock);
3485 static int mark_printk(const char *fmt, ...)
3489 va_start(args, fmt);
3490 ret = trace_vprintk(0, fmt, args);
3496 tracing_mark_write(struct file *filp, const char __user *ubuf,
3497 size_t cnt, loff_t *fpos)
3501 if (tracing_disabled)
3504 if (cnt > TRACE_BUF_SIZE)
3505 cnt = TRACE_BUF_SIZE;
3507 buf = kmalloc(cnt + 2, GFP_KERNEL);
3511 if (copy_from_user(buf, ubuf, cnt)) {
3515 if (buf[cnt-1] != '\n') {
3521 cnt = mark_printk("%s", buf);
3528 static int tracing_clock_show(struct seq_file *m, void *v)
3532 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3534 "%s%s%s%s", i ? " " : "",
3535 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3536 i == trace_clock_id ? "]" : "");
3542 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3543 size_t cnt, loff_t *fpos)
3546 const char *clockstr;
3549 if (cnt >= sizeof(buf))
3552 if (copy_from_user(&buf, ubuf, cnt))
3557 clockstr = strstrip(buf);
3559 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3560 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3563 if (i == ARRAY_SIZE(trace_clocks))
3568 mutex_lock(&trace_types_lock);
3570 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3572 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3574 mutex_unlock(&trace_types_lock);
3581 static int tracing_clock_open(struct inode *inode, struct file *file)
3583 if (tracing_disabled)
3585 return single_open(file, tracing_clock_show, NULL);
3588 static const struct file_operations tracing_max_lat_fops = {
3589 .open = tracing_open_generic,
3590 .read = tracing_max_lat_read,
3591 .write = tracing_max_lat_write,
3594 static const struct file_operations tracing_ctrl_fops = {
3595 .open = tracing_open_generic,
3596 .read = tracing_ctrl_read,
3597 .write = tracing_ctrl_write,
3600 static const struct file_operations set_tracer_fops = {
3601 .open = tracing_open_generic,
3602 .read = tracing_set_trace_read,
3603 .write = tracing_set_trace_write,
3606 static const struct file_operations tracing_pipe_fops = {
3607 .open = tracing_open_pipe,
3608 .poll = tracing_poll_pipe,
3609 .read = tracing_read_pipe,
3610 .splice_read = tracing_splice_read_pipe,
3611 .release = tracing_release_pipe,
3614 static const struct file_operations tracing_entries_fops = {
3615 .open = tracing_open_generic,
3616 .read = tracing_entries_read,
3617 .write = tracing_entries_write,
3620 static const struct file_operations tracing_mark_fops = {
3621 .open = tracing_open_generic,
3622 .write = tracing_mark_write,
3625 static const struct file_operations trace_clock_fops = {
3626 .open = tracing_clock_open,
3628 .llseek = seq_lseek,
3629 .release = single_release,
3630 .write = tracing_clock_write,
3633 struct ftrace_buffer_info {
3634 struct trace_array *tr;
3640 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3642 int cpu = (int)(long)inode->i_private;
3643 struct ftrace_buffer_info *info;
3645 if (tracing_disabled)
3648 info = kzalloc(sizeof(*info), GFP_KERNEL);
3652 info->tr = &global_trace;
3655 /* Force reading ring buffer for first read */
3656 info->read = (unsigned int)-1;
3658 filp->private_data = info;
3660 return nonseekable_open(inode, filp);
3664 tracing_buffers_read(struct file *filp, char __user *ubuf,
3665 size_t count, loff_t *ppos)
3667 struct ftrace_buffer_info *info = filp->private_data;
3675 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3679 /* Do we have previous read data to read? */
3680 if (info->read < PAGE_SIZE)
3685 trace_access_lock(info->cpu);
3686 ret = ring_buffer_read_page(info->tr->buffer,
3690 trace_access_unlock(info->cpu);
3695 size = PAGE_SIZE - info->read;
3699 ret = copy_to_user(ubuf, info->spare + info->read, size);
3710 static int tracing_buffers_release(struct inode *inode, struct file *file)
3712 struct ftrace_buffer_info *info = file->private_data;
3715 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3722 struct ring_buffer *buffer;
3727 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3728 struct pipe_buffer *buf)
3730 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3735 ring_buffer_free_read_page(ref->buffer, ref->page);
3740 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3741 struct pipe_buffer *buf)
3746 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3747 struct pipe_buffer *buf)
3749 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3754 /* Pipe buffer operations for a buffer. */
3755 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
3757 .map = generic_pipe_buf_map,
3758 .unmap = generic_pipe_buf_unmap,
3759 .confirm = generic_pipe_buf_confirm,
3760 .release = buffer_pipe_buf_release,
3761 .steal = buffer_pipe_buf_steal,
3762 .get = buffer_pipe_buf_get,
3766 * Callback from splice_to_pipe(), if we need to release some pages
3767 * at the end of the spd in case we error'ed out in filling the pipe.
3769 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3771 struct buffer_ref *ref =
3772 (struct buffer_ref *)spd->partial[i].private;
3777 ring_buffer_free_read_page(ref->buffer, ref->page);
3779 spd->partial[i].private = 0;
3783 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3784 struct pipe_inode_info *pipe, size_t len,
3787 struct ftrace_buffer_info *info = file->private_data;
3788 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3789 struct page *pages_def[PIPE_DEF_BUFFERS];
3790 struct splice_pipe_desc spd = {
3792 .partial = partial_def,
3794 .ops = &buffer_pipe_buf_ops,
3795 .spd_release = buffer_spd_release,
3797 struct buffer_ref *ref;
3798 int entries, size, i;
3801 if (splice_grow_spd(pipe, &spd))
3804 if (*ppos & (PAGE_SIZE - 1)) {
3805 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3810 if (len & (PAGE_SIZE - 1)) {
3811 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3812 if (len < PAGE_SIZE) {
3819 trace_access_lock(info->cpu);
3820 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3822 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
3826 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3831 ref->buffer = info->tr->buffer;
3832 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3838 r = ring_buffer_read_page(ref->buffer, &ref->page,
3841 ring_buffer_free_read_page(ref->buffer,
3848 * zero out any left over data, this is going to
3851 size = ring_buffer_page_len(ref->page);
3852 if (size < PAGE_SIZE)
3853 memset(ref->page + size, 0, PAGE_SIZE - size);
3855 page = virt_to_page(ref->page);
3857 spd.pages[i] = page;
3858 spd.partial[i].len = PAGE_SIZE;
3859 spd.partial[i].offset = 0;
3860 spd.partial[i].private = (unsigned long)ref;
3864 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3867 trace_access_unlock(info->cpu);
3870 /* did we read anything? */
3871 if (!spd.nr_pages) {
3872 if (flags & SPLICE_F_NONBLOCK)
3880 ret = splice_to_pipe(pipe, &spd);
3881 splice_shrink_spd(pipe, &spd);
3886 static const struct file_operations tracing_buffers_fops = {
3887 .open = tracing_buffers_open,
3888 .read = tracing_buffers_read,
3889 .release = tracing_buffers_release,
3890 .splice_read = tracing_buffers_splice_read,
3891 .llseek = no_llseek,
3895 tracing_stats_read(struct file *filp, char __user *ubuf,
3896 size_t count, loff_t *ppos)
3898 unsigned long cpu = (unsigned long)filp->private_data;
3899 struct trace_array *tr = &global_trace;
3900 struct trace_seq *s;
3903 s = kmalloc(sizeof(*s), GFP_KERNEL);
3909 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3910 trace_seq_printf(s, "entries: %ld\n", cnt);
3912 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3913 trace_seq_printf(s, "overrun: %ld\n", cnt);
3915 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3916 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3918 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3925 static const struct file_operations tracing_stats_fops = {
3926 .open = tracing_open_generic,
3927 .read = tracing_stats_read,
3930 #ifdef CONFIG_DYNAMIC_FTRACE
3932 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3938 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3939 size_t cnt, loff_t *ppos)
3941 static char ftrace_dyn_info_buffer[1024];
3942 static DEFINE_MUTEX(dyn_info_mutex);
3943 unsigned long *p = filp->private_data;
3944 char *buf = ftrace_dyn_info_buffer;
3945 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3948 mutex_lock(&dyn_info_mutex);
3949 r = sprintf(buf, "%ld ", *p);
3951 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3954 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3956 mutex_unlock(&dyn_info_mutex);
3961 static const struct file_operations tracing_dyn_info_fops = {
3962 .open = tracing_open_generic,
3963 .read = tracing_read_dyn_info,
3967 static struct dentry *d_tracer;
3969 struct dentry *tracing_init_dentry(void)
3976 if (!debugfs_initialized())
3979 d_tracer = debugfs_create_dir("tracing", NULL);
3981 if (!d_tracer && !once) {
3983 pr_warning("Could not create debugfs directory 'tracing'\n");
3990 static struct dentry *d_percpu;
3992 struct dentry *tracing_dentry_percpu(void)
3995 struct dentry *d_tracer;
4000 d_tracer = tracing_init_dentry();
4005 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
4007 if (!d_percpu && !once) {
4009 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4016 static void tracing_init_debugfs_percpu(long cpu)
4018 struct dentry *d_percpu = tracing_dentry_percpu();
4019 struct dentry *d_cpu;
4020 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
4023 if (cpu > 999 || cpu < 0)
4026 sprintf(cpu_dir, "cpu%ld", cpu);
4027 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4029 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4033 /* per cpu trace_pipe */
4034 trace_create_file("trace_pipe", 0444, d_cpu,
4035 (void *) cpu, &tracing_pipe_fops);
4038 trace_create_file("trace", 0644, d_cpu,
4039 (void *) cpu, &tracing_fops);
4041 trace_create_file("trace_pipe_raw", 0444, d_cpu,
4042 (void *) cpu, &tracing_buffers_fops);
4044 trace_create_file("stats", 0444, d_cpu,
4045 (void *) cpu, &tracing_stats_fops);
4048 #ifdef CONFIG_FTRACE_SELFTEST
4049 /* Let selftest have access to static functions in this file */
4050 #include "trace_selftest.c"
4053 struct trace_option_dentry {
4054 struct tracer_opt *opt;
4055 struct tracer_flags *flags;
4056 struct dentry *entry;
4060 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4063 struct trace_option_dentry *topt = filp->private_data;
4066 if (topt->flags->val & topt->opt->bit)
4071 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4075 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4078 struct trace_option_dentry *topt = filp->private_data;
4083 if (cnt >= sizeof(buf))
4086 if (copy_from_user(&buf, ubuf, cnt))
4091 ret = strict_strtoul(buf, 10, &val);
4095 if (val != 0 && val != 1)
4098 if (!!(topt->flags->val & topt->opt->bit) != val) {
4099 mutex_lock(&trace_types_lock);
4100 ret = __set_tracer_option(current_trace, topt->flags,
4102 mutex_unlock(&trace_types_lock);
4113 static const struct file_operations trace_options_fops = {
4114 .open = tracing_open_generic,
4115 .read = trace_options_read,
4116 .write = trace_options_write,
4120 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4123 long index = (long)filp->private_data;
4126 if (trace_flags & (1 << index))
4131 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4135 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4138 long index = (long)filp->private_data;
4143 if (cnt >= sizeof(buf))
4146 if (copy_from_user(&buf, ubuf, cnt))
4151 ret = strict_strtoul(buf, 10, &val);
4155 if (val != 0 && val != 1)
4157 set_tracer_flags(1 << index, val);
4164 static const struct file_operations trace_options_core_fops = {
4165 .open = tracing_open_generic,
4166 .read = trace_options_core_read,
4167 .write = trace_options_core_write,
4170 struct dentry *trace_create_file(const char *name,
4172 struct dentry *parent,
4174 const struct file_operations *fops)
4178 ret = debugfs_create_file(name, mode, parent, data, fops);
4180 pr_warning("Could not create debugfs '%s' entry\n", name);
4186 static struct dentry *trace_options_init_dentry(void)
4188 struct dentry *d_tracer;
4189 static struct dentry *t_options;
4194 d_tracer = tracing_init_dentry();
4198 t_options = debugfs_create_dir("options", d_tracer);
4200 pr_warning("Could not create debugfs directory 'options'\n");
4208 create_trace_option_file(struct trace_option_dentry *topt,
4209 struct tracer_flags *flags,
4210 struct tracer_opt *opt)
4212 struct dentry *t_options;
4214 t_options = trace_options_init_dentry();
4218 topt->flags = flags;
4221 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4222 &trace_options_fops);
4226 static struct trace_option_dentry *
4227 create_trace_option_files(struct tracer *tracer)
4229 struct trace_option_dentry *topts;
4230 struct tracer_flags *flags;
4231 struct tracer_opt *opts;
4237 flags = tracer->flags;
4239 if (!flags || !flags->opts)
4244 for (cnt = 0; opts[cnt].name; cnt++)
4247 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4251 for (cnt = 0; opts[cnt].name; cnt++)
4252 create_trace_option_file(&topts[cnt], flags,
4259 destroy_trace_option_files(struct trace_option_dentry *topts)
4266 for (cnt = 0; topts[cnt].opt; cnt++) {
4267 if (topts[cnt].entry)
4268 debugfs_remove(topts[cnt].entry);
4274 static struct dentry *
4275 create_trace_option_core_file(const char *option, long index)
4277 struct dentry *t_options;
4279 t_options = trace_options_init_dentry();
4283 return trace_create_file(option, 0644, t_options, (void *)index,
4284 &trace_options_core_fops);
4287 static __init void create_trace_options_dir(void)
4289 struct dentry *t_options;
4292 t_options = trace_options_init_dentry();
4296 for (i = 0; trace_options[i]; i++)
4297 create_trace_option_core_file(trace_options[i], i);
4300 static __init int tracer_init_debugfs(void)
4302 struct dentry *d_tracer;
4305 trace_access_lock_init();
4307 d_tracer = tracing_init_dentry();
4309 trace_create_file("tracing_enabled", 0644, d_tracer,
4310 &global_trace, &tracing_ctrl_fops);
4312 trace_create_file("trace_options", 0644, d_tracer,
4313 NULL, &tracing_iter_fops);
4315 trace_create_file("tracing_cpumask", 0644, d_tracer,
4316 NULL, &tracing_cpumask_fops);
4318 trace_create_file("trace", 0644, d_tracer,
4319 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4321 trace_create_file("available_tracers", 0444, d_tracer,
4322 &global_trace, &show_traces_fops);
4324 trace_create_file("current_tracer", 0644, d_tracer,
4325 &global_trace, &set_tracer_fops);
4327 #ifdef CONFIG_TRACER_MAX_TRACE
4328 trace_create_file("tracing_max_latency", 0644, d_tracer,
4329 &tracing_max_latency, &tracing_max_lat_fops);
4332 trace_create_file("tracing_thresh", 0644, d_tracer,
4333 &tracing_thresh, &tracing_max_lat_fops);
4335 trace_create_file("README", 0444, d_tracer,
4336 NULL, &tracing_readme_fops);
4338 trace_create_file("trace_pipe", 0444, d_tracer,
4339 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4341 trace_create_file("buffer_size_kb", 0644, d_tracer,
4342 &global_trace, &tracing_entries_fops);
4344 trace_create_file("trace_marker", 0220, d_tracer,
4345 NULL, &tracing_mark_fops);
4347 trace_create_file("saved_cmdlines", 0444, d_tracer,
4348 NULL, &tracing_saved_cmdlines_fops);
4350 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4353 #ifdef CONFIG_DYNAMIC_FTRACE
4354 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4355 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4357 #ifdef CONFIG_SYSPROF_TRACER
4358 init_tracer_sysprof_debugfs(d_tracer);
4361 create_trace_options_dir();
4363 for_each_tracing_cpu(cpu)
4364 tracing_init_debugfs_percpu(cpu);
4369 static int trace_panic_handler(struct notifier_block *this,
4370 unsigned long event, void *unused)
4372 if (ftrace_dump_on_oops)
4373 ftrace_dump(ftrace_dump_on_oops);
4377 static struct notifier_block trace_panic_notifier = {
4378 .notifier_call = trace_panic_handler,
4380 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4383 static int trace_die_handler(struct notifier_block *self,
4389 if (ftrace_dump_on_oops)
4390 ftrace_dump(ftrace_dump_on_oops);
4398 static struct notifier_block trace_die_notifier = {
4399 .notifier_call = trace_die_handler,
4404 * printk is set to max of 1024, we really don't need it that big.
4405 * Nothing should be printing 1000 characters anyway.
4407 #define TRACE_MAX_PRINT 1000
4410 * Define here KERN_TRACE so that we have one place to modify
4411 * it if we decide to change what log level the ftrace dump
4414 #define KERN_TRACE KERN_EMERG
4417 trace_printk_seq(struct trace_seq *s)
4419 /* Probably should print a warning here. */
4423 /* should be zero ended, but we are paranoid. */
4424 s->buffer[s->len] = 0;
4426 printk(KERN_TRACE "%s", s->buffer);
4432 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4434 static arch_spinlock_t ftrace_dump_lock =
4435 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4436 /* use static because iter can be a bit big for the stack */
4437 static struct trace_iterator iter;
4438 unsigned int old_userobj;
4439 static int dump_ran;
4440 unsigned long flags;
4444 local_irq_save(flags);
4445 arch_spin_lock(&ftrace_dump_lock);
4453 if (disable_tracing)
4456 for_each_tracing_cpu(cpu) {
4457 atomic_inc(&global_trace.data[cpu]->disabled);
4460 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4462 /* don't look at user memory in panic mode */
4463 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4465 /* Simulate the iterator */
4466 iter.tr = &global_trace;
4467 iter.trace = current_trace;
4469 switch (oops_dump_mode) {
4471 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4474 iter.cpu_file = raw_smp_processor_id();
4479 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
4480 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4483 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4486 * We need to stop all tracing on all CPUS to read the
4487 * the next buffer. This is a bit expensive, but is
4488 * not done often. We fill all what we can read,
4489 * and then release the locks again.
4492 while (!trace_empty(&iter)) {
4495 printk(KERN_TRACE "---------------------------------\n");
4499 /* reset all but tr, trace, and overruns */
4500 memset(&iter.seq, 0,
4501 sizeof(struct trace_iterator) -
4502 offsetof(struct trace_iterator, seq));
4503 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4506 if (find_next_entry_inc(&iter) != NULL) {
4509 ret = print_trace_line(&iter);
4510 if (ret != TRACE_TYPE_NO_CONSUME)
4511 trace_consume(&iter);
4514 trace_printk_seq(&iter.seq);
4518 printk(KERN_TRACE " (ftrace buffer empty)\n");
4520 printk(KERN_TRACE "---------------------------------\n");
4523 /* Re-enable tracing if requested */
4524 if (!disable_tracing) {
4525 trace_flags |= old_userobj;
4527 for_each_tracing_cpu(cpu) {
4528 atomic_dec(&global_trace.data[cpu]->disabled);
4534 arch_spin_unlock(&ftrace_dump_lock);
4535 local_irq_restore(flags);
4538 /* By default: disable tracing after the dump */
4539 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
4541 __ftrace_dump(true, oops_dump_mode);
4544 __init static int tracer_alloc_buffers(void)
4550 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4553 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4554 goto out_free_buffer_mask;
4556 /* To save memory, keep the ring buffer size to its minimum */
4557 if (ring_buffer_expanded)
4558 ring_buf_size = trace_buf_size;
4562 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4563 cpumask_copy(tracing_cpumask, cpu_all_mask);
4565 /* TODO: make the number of buffers hot pluggable with CPUS */
4566 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4567 TRACE_BUFFER_FLAGS);
4568 if (!global_trace.buffer) {
4569 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4571 goto out_free_cpumask;
4573 global_trace.entries = ring_buffer_size(global_trace.buffer);
4576 #ifdef CONFIG_TRACER_MAX_TRACE
4577 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4578 TRACE_BUFFER_FLAGS);
4579 if (!max_tr.buffer) {
4580 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4582 ring_buffer_free(global_trace.buffer);
4583 goto out_free_cpumask;
4585 max_tr.entries = ring_buffer_size(max_tr.buffer);
4586 WARN_ON(max_tr.entries != global_trace.entries);
4589 /* Allocate the first page for all buffers */
4590 for_each_tracing_cpu(i) {
4591 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4592 max_tr.data[i] = &per_cpu(max_tr_data, i);
4595 trace_init_cmdlines();
4597 register_tracer(&nop_trace);
4598 current_trace = &nop_trace;
4599 /* All seems OK, enable tracing */
4600 tracing_disabled = 0;
4602 atomic_notifier_chain_register(&panic_notifier_list,
4603 &trace_panic_notifier);
4605 register_die_notifier(&trace_die_notifier);
4610 free_cpumask_var(tracing_cpumask);
4611 out_free_buffer_mask:
4612 free_cpumask_var(tracing_buffer_mask);
4617 __init static int clear_boot_tracer(void)
4620 * The default tracer at boot buffer is an init section.
4621 * This function is called in lateinit. If we did not
4622 * find the boot tracer, then clear it out, to prevent
4623 * later registration from accessing the buffer that is
4624 * about to be freed.
4626 if (!default_bootup_tracer)
4629 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4630 default_bootup_tracer);
4631 default_bootup_tracer = NULL;
4636 early_initcall(tracer_alloc_buffers);
4637 fs_initcall(tracer_init_debugfs);
4638 late_initcall(clear_boot_tracer);