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;
1408 int cpu, len = 0, size, pc;
1410 if (unlikely(tracing_selftest_running || tracing_disabled))
1413 /* Don't pollute graph traces with trace_vprintk internals */
1414 pause_graph_tracing();
1416 pc = preempt_count();
1417 resched = ftrace_preempt_disable();
1418 cpu = raw_smp_processor_id();
1419 data = tr->data[cpu];
1421 disable = atomic_inc_return(&data->disabled);
1422 if (unlikely(disable != 1))
1425 /* Lockdep uses trace_printk for lock tracing */
1426 local_irq_save(flags);
1427 arch_spin_lock(&trace_buf_lock);
1428 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1430 if (len > TRACE_BUF_SIZE || len < 0)
1433 size = sizeof(*entry) + sizeof(u32) * len;
1434 buffer = tr->buffer;
1435 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1439 entry = ring_buffer_event_data(event);
1443 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1444 if (!filter_check_discard(call, entry, buffer, event)) {
1445 ring_buffer_unlock_commit(buffer, event);
1446 ftrace_trace_stack(buffer, flags, 6, pc);
1450 arch_spin_unlock(&trace_buf_lock);
1451 local_irq_restore(flags);
1454 atomic_dec_return(&data->disabled);
1455 ftrace_preempt_enable(resched);
1456 unpause_graph_tracing();
1460 EXPORT_SYMBOL_GPL(trace_vbprintk);
1462 int trace_array_printk(struct trace_array *tr,
1463 unsigned long ip, const char *fmt, ...)
1468 if (!(trace_flags & TRACE_ITER_PRINTK))
1472 ret = trace_array_vprintk(tr, ip, fmt, ap);
1477 int trace_array_vprintk(struct trace_array *tr,
1478 unsigned long ip, const char *fmt, va_list args)
1480 static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1481 static char trace_buf[TRACE_BUF_SIZE];
1483 struct ftrace_event_call *call = &event_print;
1484 struct ring_buffer_event *event;
1485 struct ring_buffer *buffer;
1486 struct trace_array_cpu *data;
1487 int cpu, len = 0, size, pc;
1488 struct print_entry *entry;
1489 unsigned long irq_flags;
1492 if (tracing_disabled || tracing_selftest_running)
1495 pc = preempt_count();
1496 preempt_disable_notrace();
1497 cpu = raw_smp_processor_id();
1498 data = tr->data[cpu];
1500 disable = atomic_inc_return(&data->disabled);
1501 if (unlikely(disable != 1))
1504 pause_graph_tracing();
1505 raw_local_irq_save(irq_flags);
1506 arch_spin_lock(&trace_buf_lock);
1507 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1509 size = sizeof(*entry) + len + 1;
1510 buffer = tr->buffer;
1511 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1515 entry = ring_buffer_event_data(event);
1518 memcpy(&entry->buf, trace_buf, len);
1519 entry->buf[len] = '\0';
1520 if (!filter_check_discard(call, entry, buffer, event)) {
1521 ring_buffer_unlock_commit(buffer, event);
1522 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1526 arch_spin_unlock(&trace_buf_lock);
1527 raw_local_irq_restore(irq_flags);
1528 unpause_graph_tracing();
1530 atomic_dec_return(&data->disabled);
1531 preempt_enable_notrace();
1536 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1538 return trace_array_vprintk(&global_trace, ip, fmt, args);
1540 EXPORT_SYMBOL_GPL(trace_vprintk);
1542 enum trace_file_type {
1543 TRACE_FILE_LAT_FMT = 1,
1544 TRACE_FILE_ANNOTATE = 2,
1547 static void trace_iterator_increment(struct trace_iterator *iter)
1549 /* Don't allow ftrace to trace into the ring buffers */
1550 ftrace_disable_cpu();
1553 if (iter->buffer_iter[iter->cpu])
1554 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1556 ftrace_enable_cpu();
1559 static struct trace_entry *
1560 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1561 unsigned long *lost_events)
1563 struct ring_buffer_event *event;
1564 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1566 /* Don't allow ftrace to trace into the ring buffers */
1567 ftrace_disable_cpu();
1570 event = ring_buffer_iter_peek(buf_iter, ts);
1572 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1575 ftrace_enable_cpu();
1577 return event ? ring_buffer_event_data(event) : NULL;
1580 static struct trace_entry *
1581 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1582 unsigned long *missing_events, u64 *ent_ts)
1584 struct ring_buffer *buffer = iter->tr->buffer;
1585 struct trace_entry *ent, *next = NULL;
1586 unsigned long lost_events = 0, next_lost = 0;
1587 int cpu_file = iter->cpu_file;
1588 u64 next_ts = 0, ts;
1593 * If we are in a per_cpu trace file, don't bother by iterating over
1594 * all cpu and peek directly.
1596 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1597 if (ring_buffer_empty_cpu(buffer, cpu_file))
1599 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1601 *ent_cpu = cpu_file;
1606 for_each_tracing_cpu(cpu) {
1608 if (ring_buffer_empty_cpu(buffer, cpu))
1611 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1614 * Pick the entry with the smallest timestamp:
1616 if (ent && (!next || ts < next_ts)) {
1620 next_lost = lost_events;
1625 *ent_cpu = next_cpu;
1631 *missing_events = next_lost;
1636 /* Find the next real entry, without updating the iterator itself */
1637 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1638 int *ent_cpu, u64 *ent_ts)
1640 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1643 /* Find the next real entry, and increment the iterator to the next entry */
1644 static void *find_next_entry_inc(struct trace_iterator *iter)
1646 iter->ent = __find_next_entry(iter, &iter->cpu,
1647 &iter->lost_events, &iter->ts);
1650 trace_iterator_increment(iter);
1652 return iter->ent ? iter : NULL;
1655 static void trace_consume(struct trace_iterator *iter)
1657 /* Don't allow ftrace to trace into the ring buffers */
1658 ftrace_disable_cpu();
1659 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1660 &iter->lost_events);
1661 ftrace_enable_cpu();
1664 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1666 struct trace_iterator *iter = m->private;
1670 WARN_ON_ONCE(iter->leftover);
1674 /* can't go backwards */
1679 ent = find_next_entry_inc(iter);
1683 while (ent && iter->idx < i)
1684 ent = find_next_entry_inc(iter);
1691 static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1693 struct trace_array *tr = iter->tr;
1694 struct ring_buffer_event *event;
1695 struct ring_buffer_iter *buf_iter;
1696 unsigned long entries = 0;
1699 tr->data[cpu]->skipped_entries = 0;
1701 if (!iter->buffer_iter[cpu])
1704 buf_iter = iter->buffer_iter[cpu];
1705 ring_buffer_iter_reset(buf_iter);
1708 * We could have the case with the max latency tracers
1709 * that a reset never took place on a cpu. This is evident
1710 * by the timestamp being before the start of the buffer.
1712 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1713 if (ts >= iter->tr->time_start)
1716 ring_buffer_read(buf_iter, NULL);
1719 tr->data[cpu]->skipped_entries = entries;
1723 * The current tracer is copied to avoid a global locking
1726 static void *s_start(struct seq_file *m, loff_t *pos)
1728 struct trace_iterator *iter = m->private;
1729 static struct tracer *old_tracer;
1730 int cpu_file = iter->cpu_file;
1735 /* copy the tracer to avoid using a global lock all around */
1736 mutex_lock(&trace_types_lock);
1737 if (unlikely(old_tracer != current_trace && current_trace)) {
1738 old_tracer = current_trace;
1739 *iter->trace = *current_trace;
1741 mutex_unlock(&trace_types_lock);
1743 atomic_inc(&trace_record_cmdline_disabled);
1745 if (*pos != iter->pos) {
1750 ftrace_disable_cpu();
1752 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1753 for_each_tracing_cpu(cpu)
1754 tracing_iter_reset(iter, cpu);
1756 tracing_iter_reset(iter, cpu_file);
1758 ftrace_enable_cpu();
1761 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1766 * If we overflowed the seq_file before, then we want
1767 * to just reuse the trace_seq buffer again.
1773 p = s_next(m, p, &l);
1777 trace_event_read_lock();
1778 trace_access_lock(cpu_file);
1782 static void s_stop(struct seq_file *m, void *p)
1784 struct trace_iterator *iter = m->private;
1786 atomic_dec(&trace_record_cmdline_disabled);
1787 trace_access_unlock(iter->cpu_file);
1788 trace_event_read_unlock();
1791 static void print_lat_help_header(struct seq_file *m)
1793 seq_puts(m, "# _------=> CPU# \n");
1794 seq_puts(m, "# / _-----=> irqs-off \n");
1795 seq_puts(m, "# | / _----=> need-resched \n");
1796 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1797 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1798 seq_puts(m, "# |||| /_--=> lock-depth \n");
1799 seq_puts(m, "# |||||/ delay \n");
1800 seq_puts(m, "# cmd pid |||||| time | caller \n");
1801 seq_puts(m, "# \\ / |||||| \\ | / \n");
1804 static void print_func_help_header(struct seq_file *m)
1806 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1807 seq_puts(m, "# | | | | |\n");
1812 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1814 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1815 struct trace_array *tr = iter->tr;
1816 struct trace_array_cpu *data = tr->data[tr->cpu];
1817 struct tracer *type = current_trace;
1818 unsigned long entries = 0;
1819 unsigned long total = 0;
1820 unsigned long count;
1821 const char *name = "preemption";
1828 for_each_tracing_cpu(cpu) {
1829 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1831 * If this buffer has skipped entries, then we hold all
1832 * entries for the trace and we need to ignore the
1833 * ones before the time stamp.
1835 if (tr->data[cpu]->skipped_entries) {
1836 count -= tr->data[cpu]->skipped_entries;
1837 /* total is the same as the entries */
1841 ring_buffer_overrun_cpu(tr->buffer, cpu);
1845 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1847 seq_puts(m, "# -----------------------------------"
1848 "---------------------------------\n");
1849 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1850 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1851 nsecs_to_usecs(data->saved_latency),
1855 #if defined(CONFIG_PREEMPT_NONE)
1857 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1859 #elif defined(CONFIG_PREEMPT)
1864 /* These are reserved for later use */
1867 seq_printf(m, " #P:%d)\n", num_online_cpus());
1871 seq_puts(m, "# -----------------\n");
1872 seq_printf(m, "# | task: %.16s-%d "
1873 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1874 data->comm, data->pid, data->uid, data->nice,
1875 data->policy, data->rt_priority);
1876 seq_puts(m, "# -----------------\n");
1878 if (data->critical_start) {
1879 seq_puts(m, "# => started at: ");
1880 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1881 trace_print_seq(m, &iter->seq);
1882 seq_puts(m, "\n# => ended at: ");
1883 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1884 trace_print_seq(m, &iter->seq);
1885 seq_puts(m, "\n#\n");
1891 static void test_cpu_buff_start(struct trace_iterator *iter)
1893 struct trace_seq *s = &iter->seq;
1895 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1898 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1901 if (cpumask_test_cpu(iter->cpu, iter->started))
1904 if (iter->tr->data[iter->cpu]->skipped_entries)
1907 cpumask_set_cpu(iter->cpu, iter->started);
1909 /* Don't print started cpu buffer for the first entry of the trace */
1911 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1915 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1917 struct trace_seq *s = &iter->seq;
1918 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1919 struct trace_entry *entry;
1920 struct trace_event *event;
1924 test_cpu_buff_start(iter);
1926 event = ftrace_find_event(entry->type);
1928 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1929 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1930 if (!trace_print_lat_context(iter))
1933 if (!trace_print_context(iter))
1939 return event->trace(iter, sym_flags);
1941 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1944 return TRACE_TYPE_HANDLED;
1946 return TRACE_TYPE_PARTIAL_LINE;
1949 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1951 struct trace_seq *s = &iter->seq;
1952 struct trace_entry *entry;
1953 struct trace_event *event;
1957 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1958 if (!trace_seq_printf(s, "%d %d %llu ",
1959 entry->pid, iter->cpu, iter->ts))
1963 event = ftrace_find_event(entry->type);
1965 return event->raw(iter, 0);
1967 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1970 return TRACE_TYPE_HANDLED;
1972 return TRACE_TYPE_PARTIAL_LINE;
1975 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1977 struct trace_seq *s = &iter->seq;
1978 unsigned char newline = '\n';
1979 struct trace_entry *entry;
1980 struct trace_event *event;
1984 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1985 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1986 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1987 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1990 event = ftrace_find_event(entry->type);
1992 enum print_line_t ret = event->hex(iter, 0);
1993 if (ret != TRACE_TYPE_HANDLED)
1997 SEQ_PUT_FIELD_RET(s, newline);
1999 return TRACE_TYPE_HANDLED;
2002 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2004 struct trace_seq *s = &iter->seq;
2005 struct trace_entry *entry;
2006 struct trace_event *event;
2010 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2011 SEQ_PUT_FIELD_RET(s, entry->pid);
2012 SEQ_PUT_FIELD_RET(s, iter->cpu);
2013 SEQ_PUT_FIELD_RET(s, iter->ts);
2016 event = ftrace_find_event(entry->type);
2017 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
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) {
2205 iter->buffer_iter[cpu] =
2206 ring_buffer_read_start(iter->tr->buffer, cpu);
2207 tracing_iter_reset(iter, cpu);
2210 cpu = iter->cpu_file;
2211 iter->buffer_iter[cpu] =
2212 ring_buffer_read_start(iter->tr->buffer, cpu);
2213 tracing_iter_reset(iter, cpu);
2216 ret = seq_open(file, &tracer_seq_ops);
2218 fail_ret = ERR_PTR(ret);
2222 m = file->private_data;
2225 mutex_unlock(&trace_types_lock);
2230 for_each_tracing_cpu(cpu) {
2231 if (iter->buffer_iter[cpu])
2232 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2234 free_cpumask_var(iter->started);
2237 mutex_unlock(&trace_types_lock);
2244 int tracing_open_generic(struct inode *inode, struct file *filp)
2246 if (tracing_disabled)
2249 filp->private_data = inode->i_private;
2253 static int tracing_release(struct inode *inode, struct file *file)
2255 struct seq_file *m = (struct seq_file *)file->private_data;
2256 struct trace_iterator *iter;
2259 if (!(file->f_mode & FMODE_READ))
2264 mutex_lock(&trace_types_lock);
2265 for_each_tracing_cpu(cpu) {
2266 if (iter->buffer_iter[cpu])
2267 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2270 if (iter->trace && iter->trace->close)
2271 iter->trace->close(iter);
2273 /* reenable tracing if it was previously enabled */
2275 mutex_unlock(&trace_types_lock);
2277 seq_release(inode, file);
2278 mutex_destroy(&iter->mutex);
2279 free_cpumask_var(iter->started);
2285 static int tracing_open(struct inode *inode, struct file *file)
2287 struct trace_iterator *iter;
2290 /* If this file was open for write, then erase contents */
2291 if ((file->f_mode & FMODE_WRITE) &&
2292 (file->f_flags & O_TRUNC)) {
2293 long cpu = (long) inode->i_private;
2295 if (cpu == TRACE_PIPE_ALL_CPU)
2296 tracing_reset_online_cpus(&global_trace);
2298 tracing_reset(&global_trace, cpu);
2301 if (file->f_mode & FMODE_READ) {
2302 iter = __tracing_open(inode, file);
2304 ret = PTR_ERR(iter);
2305 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2306 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2312 t_next(struct seq_file *m, void *v, loff_t *pos)
2314 struct tracer *t = v;
2324 static void *t_start(struct seq_file *m, loff_t *pos)
2329 mutex_lock(&trace_types_lock);
2330 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2336 static void t_stop(struct seq_file *m, void *p)
2338 mutex_unlock(&trace_types_lock);
2341 static int t_show(struct seq_file *m, void *v)
2343 struct tracer *t = v;
2348 seq_printf(m, "%s", t->name);
2357 static const struct seq_operations show_traces_seq_ops = {
2364 static int show_traces_open(struct inode *inode, struct file *file)
2366 if (tracing_disabled)
2369 return seq_open(file, &show_traces_seq_ops);
2373 tracing_write_stub(struct file *filp, const char __user *ubuf,
2374 size_t count, loff_t *ppos)
2379 static const struct file_operations tracing_fops = {
2380 .open = tracing_open,
2382 .write = tracing_write_stub,
2383 .llseek = seq_lseek,
2384 .release = tracing_release,
2387 static const struct file_operations show_traces_fops = {
2388 .open = show_traces_open,
2390 .release = seq_release,
2394 * Only trace on a CPU if the bitmask is set:
2396 static cpumask_var_t tracing_cpumask;
2399 * The tracer itself will not take this lock, but still we want
2400 * to provide a consistent cpumask to user-space:
2402 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2405 * Temporary storage for the character representation of the
2406 * CPU bitmask (and one more byte for the newline):
2408 static char mask_str[NR_CPUS + 1];
2411 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2412 size_t count, loff_t *ppos)
2416 mutex_lock(&tracing_cpumask_update_lock);
2418 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2419 if (count - len < 2) {
2423 len += sprintf(mask_str + len, "\n");
2424 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2427 mutex_unlock(&tracing_cpumask_update_lock);
2433 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2434 size_t count, loff_t *ppos)
2437 cpumask_var_t tracing_cpumask_new;
2439 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2442 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2446 mutex_lock(&tracing_cpumask_update_lock);
2448 local_irq_disable();
2449 arch_spin_lock(&ftrace_max_lock);
2450 for_each_tracing_cpu(cpu) {
2452 * Increase/decrease the disabled counter if we are
2453 * about to flip a bit in the cpumask:
2455 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2456 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2457 atomic_inc(&global_trace.data[cpu]->disabled);
2459 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2460 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2461 atomic_dec(&global_trace.data[cpu]->disabled);
2464 arch_spin_unlock(&ftrace_max_lock);
2467 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2469 mutex_unlock(&tracing_cpumask_update_lock);
2470 free_cpumask_var(tracing_cpumask_new);
2475 free_cpumask_var(tracing_cpumask_new);
2480 static const struct file_operations tracing_cpumask_fops = {
2481 .open = tracing_open_generic,
2482 .read = tracing_cpumask_read,
2483 .write = tracing_cpumask_write,
2486 static int tracing_trace_options_show(struct seq_file *m, void *v)
2488 struct tracer_opt *trace_opts;
2492 mutex_lock(&trace_types_lock);
2493 tracer_flags = current_trace->flags->val;
2494 trace_opts = current_trace->flags->opts;
2496 for (i = 0; trace_options[i]; i++) {
2497 if (trace_flags & (1 << i))
2498 seq_printf(m, "%s\n", trace_options[i]);
2500 seq_printf(m, "no%s\n", trace_options[i]);
2503 for (i = 0; trace_opts[i].name; i++) {
2504 if (tracer_flags & trace_opts[i].bit)
2505 seq_printf(m, "%s\n", trace_opts[i].name);
2507 seq_printf(m, "no%s\n", trace_opts[i].name);
2509 mutex_unlock(&trace_types_lock);
2514 static int __set_tracer_option(struct tracer *trace,
2515 struct tracer_flags *tracer_flags,
2516 struct tracer_opt *opts, int neg)
2520 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2525 tracer_flags->val &= ~opts->bit;
2527 tracer_flags->val |= opts->bit;
2531 /* Try to assign a tracer specific option */
2532 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2534 struct tracer_flags *tracer_flags = trace->flags;
2535 struct tracer_opt *opts = NULL;
2538 for (i = 0; tracer_flags->opts[i].name; i++) {
2539 opts = &tracer_flags->opts[i];
2541 if (strcmp(cmp, opts->name) == 0)
2542 return __set_tracer_option(trace, trace->flags,
2549 static void set_tracer_flags(unsigned int mask, int enabled)
2551 /* do nothing if flag is already set */
2552 if (!!(trace_flags & mask) == !!enabled)
2556 trace_flags |= mask;
2558 trace_flags &= ~mask;
2562 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2563 size_t cnt, loff_t *ppos)
2571 if (cnt >= sizeof(buf))
2574 if (copy_from_user(&buf, ubuf, cnt))
2578 cmp = strstrip(buf);
2580 if (strncmp(cmp, "no", 2) == 0) {
2585 for (i = 0; trace_options[i]; i++) {
2586 if (strcmp(cmp, trace_options[i]) == 0) {
2587 set_tracer_flags(1 << i, !neg);
2592 /* If no option could be set, test the specific tracer options */
2593 if (!trace_options[i]) {
2594 mutex_lock(&trace_types_lock);
2595 ret = set_tracer_option(current_trace, cmp, neg);
2596 mutex_unlock(&trace_types_lock);
2606 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2608 if (tracing_disabled)
2610 return single_open(file, tracing_trace_options_show, NULL);
2613 static const struct file_operations tracing_iter_fops = {
2614 .open = tracing_trace_options_open,
2616 .llseek = seq_lseek,
2617 .release = single_release,
2618 .write = tracing_trace_options_write,
2621 static const char readme_msg[] =
2622 "tracing mini-HOWTO:\n\n"
2623 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2624 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2625 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2626 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2628 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2629 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2631 "# cat /sys/kernel/debug/tracing/trace_options\n"
2632 "noprint-parent nosym-offset nosym-addr noverbose\n"
2633 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2634 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2635 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2636 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2640 tracing_readme_read(struct file *filp, char __user *ubuf,
2641 size_t cnt, loff_t *ppos)
2643 return simple_read_from_buffer(ubuf, cnt, ppos,
2644 readme_msg, strlen(readme_msg));
2647 static const struct file_operations tracing_readme_fops = {
2648 .open = tracing_open_generic,
2649 .read = tracing_readme_read,
2653 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2654 size_t cnt, loff_t *ppos)
2663 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2667 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2675 for (i = 0; i < SAVED_CMDLINES; i++) {
2678 pid = map_cmdline_to_pid[i];
2679 if (pid == -1 || pid == NO_CMDLINE_MAP)
2682 trace_find_cmdline(pid, buf_comm);
2683 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2688 len = simple_read_from_buffer(ubuf, cnt, ppos,
2697 static const struct file_operations tracing_saved_cmdlines_fops = {
2698 .open = tracing_open_generic,
2699 .read = tracing_saved_cmdlines_read,
2703 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2704 size_t cnt, loff_t *ppos)
2709 r = sprintf(buf, "%u\n", tracer_enabled);
2710 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2714 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2715 size_t cnt, loff_t *ppos)
2717 struct trace_array *tr = filp->private_data;
2722 if (cnt >= sizeof(buf))
2725 if (copy_from_user(&buf, ubuf, cnt))
2730 ret = strict_strtoul(buf, 10, &val);
2736 mutex_lock(&trace_types_lock);
2737 if (tracer_enabled ^ val) {
2740 if (current_trace->start)
2741 current_trace->start(tr);
2746 if (current_trace->stop)
2747 current_trace->stop(tr);
2750 mutex_unlock(&trace_types_lock);
2758 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2759 size_t cnt, loff_t *ppos)
2761 char buf[MAX_TRACER_SIZE+2];
2764 mutex_lock(&trace_types_lock);
2766 r = sprintf(buf, "%s\n", current_trace->name);
2768 r = sprintf(buf, "\n");
2769 mutex_unlock(&trace_types_lock);
2771 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2774 int tracer_init(struct tracer *t, struct trace_array *tr)
2776 tracing_reset_online_cpus(tr);
2780 static int tracing_resize_ring_buffer(unsigned long size)
2785 * If kernel or user changes the size of the ring buffer
2786 * we use the size that was given, and we can forget about
2787 * expanding it later.
2789 ring_buffer_expanded = 1;
2791 ret = ring_buffer_resize(global_trace.buffer, size);
2795 ret = ring_buffer_resize(max_tr.buffer, size);
2799 r = ring_buffer_resize(global_trace.buffer,
2800 global_trace.entries);
2803 * AARGH! We are left with different
2804 * size max buffer!!!!
2805 * The max buffer is our "snapshot" buffer.
2806 * When a tracer needs a snapshot (one of the
2807 * latency tracers), it swaps the max buffer
2808 * with the saved snap shot. We succeeded to
2809 * update the size of the main buffer, but failed to
2810 * update the size of the max buffer. But when we tried
2811 * to reset the main buffer to the original size, we
2812 * failed there too. This is very unlikely to
2813 * happen, but if it does, warn and kill all
2817 tracing_disabled = 1;
2822 global_trace.entries = size;
2828 * tracing_update_buffers - used by tracing facility to expand ring buffers
2830 * To save on memory when the tracing is never used on a system with it
2831 * configured in. The ring buffers are set to a minimum size. But once
2832 * a user starts to use the tracing facility, then they need to grow
2833 * to their default size.
2835 * This function is to be called when a tracer is about to be used.
2837 int tracing_update_buffers(void)
2841 mutex_lock(&trace_types_lock);
2842 if (!ring_buffer_expanded)
2843 ret = tracing_resize_ring_buffer(trace_buf_size);
2844 mutex_unlock(&trace_types_lock);
2849 struct trace_option_dentry;
2851 static struct trace_option_dentry *
2852 create_trace_option_files(struct tracer *tracer);
2855 destroy_trace_option_files(struct trace_option_dentry *topts);
2857 static int tracing_set_tracer(const char *buf)
2859 static struct trace_option_dentry *topts;
2860 struct trace_array *tr = &global_trace;
2864 mutex_lock(&trace_types_lock);
2866 if (!ring_buffer_expanded) {
2867 ret = tracing_resize_ring_buffer(trace_buf_size);
2873 for (t = trace_types; t; t = t->next) {
2874 if (strcmp(t->name, buf) == 0)
2881 if (t == current_trace)
2884 trace_branch_disable();
2885 if (current_trace && current_trace->reset)
2886 current_trace->reset(tr);
2888 destroy_trace_option_files(topts);
2892 topts = create_trace_option_files(current_trace);
2895 ret = tracer_init(t, tr);
2900 trace_branch_enable(tr);
2902 mutex_unlock(&trace_types_lock);
2908 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2909 size_t cnt, loff_t *ppos)
2911 char buf[MAX_TRACER_SIZE+1];
2918 if (cnt > MAX_TRACER_SIZE)
2919 cnt = MAX_TRACER_SIZE;
2921 if (copy_from_user(&buf, ubuf, cnt))
2926 /* strip ending whitespace. */
2927 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2930 err = tracing_set_tracer(buf);
2940 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2941 size_t cnt, loff_t *ppos)
2943 unsigned long *ptr = filp->private_data;
2947 r = snprintf(buf, sizeof(buf), "%ld\n",
2948 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2949 if (r > sizeof(buf))
2951 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2955 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2956 size_t cnt, loff_t *ppos)
2958 unsigned long *ptr = filp->private_data;
2963 if (cnt >= sizeof(buf))
2966 if (copy_from_user(&buf, ubuf, cnt))
2971 ret = strict_strtoul(buf, 10, &val);
2980 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2982 long cpu_file = (long) inode->i_private;
2983 struct trace_iterator *iter;
2986 if (tracing_disabled)
2989 mutex_lock(&trace_types_lock);
2991 /* create a buffer to store the information to pass to userspace */
2992 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2999 * We make a copy of the current tracer to avoid concurrent
3000 * changes on it while we are reading.
3002 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3008 *iter->trace = *current_trace;
3010 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3015 /* trace pipe does not show start of buffer */
3016 cpumask_setall(iter->started);
3018 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3019 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3021 iter->cpu_file = cpu_file;
3022 iter->tr = &global_trace;
3023 mutex_init(&iter->mutex);
3024 filp->private_data = iter;
3026 if (iter->trace->pipe_open)
3027 iter->trace->pipe_open(iter);
3030 mutex_unlock(&trace_types_lock);
3036 mutex_unlock(&trace_types_lock);
3040 static int tracing_release_pipe(struct inode *inode, struct file *file)
3042 struct trace_iterator *iter = file->private_data;
3044 mutex_lock(&trace_types_lock);
3046 if (iter->trace->pipe_close)
3047 iter->trace->pipe_close(iter);
3049 mutex_unlock(&trace_types_lock);
3051 free_cpumask_var(iter->started);
3052 mutex_destroy(&iter->mutex);
3060 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3062 struct trace_iterator *iter = filp->private_data;
3064 if (trace_flags & TRACE_ITER_BLOCK) {
3066 * Always select as readable when in blocking mode
3068 return POLLIN | POLLRDNORM;
3070 if (!trace_empty(iter))
3071 return POLLIN | POLLRDNORM;
3072 poll_wait(filp, &trace_wait, poll_table);
3073 if (!trace_empty(iter))
3074 return POLLIN | POLLRDNORM;
3081 void default_wait_pipe(struct trace_iterator *iter)
3085 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3087 if (trace_empty(iter))
3090 finish_wait(&trace_wait, &wait);
3094 * This is a make-shift waitqueue.
3095 * A tracer might use this callback on some rare cases:
3097 * 1) the current tracer might hold the runqueue lock when it wakes up
3098 * a reader, hence a deadlock (sched, function, and function graph tracers)
3099 * 2) the function tracers, trace all functions, we don't want
3100 * the overhead of calling wake_up and friends
3101 * (and tracing them too)
3103 * Anyway, this is really very primitive wakeup.
3105 void poll_wait_pipe(struct trace_iterator *iter)
3107 set_current_state(TASK_INTERRUPTIBLE);
3108 /* sleep for 100 msecs, and try again. */
3109 schedule_timeout(HZ / 10);
3112 /* Must be called with trace_types_lock mutex held. */
3113 static int tracing_wait_pipe(struct file *filp)
3115 struct trace_iterator *iter = filp->private_data;
3117 while (trace_empty(iter)) {
3119 if ((filp->f_flags & O_NONBLOCK)) {
3123 mutex_unlock(&iter->mutex);
3125 iter->trace->wait_pipe(iter);
3127 mutex_lock(&iter->mutex);
3129 if (signal_pending(current))
3133 * We block until we read something and tracing is disabled.
3134 * We still block if tracing is disabled, but we have never
3135 * read anything. This allows a user to cat this file, and
3136 * then enable tracing. But after we have read something,
3137 * we give an EOF when tracing is again disabled.
3139 * iter->pos will be 0 if we haven't read anything.
3141 if (!tracer_enabled && iter->pos)
3152 tracing_read_pipe(struct file *filp, char __user *ubuf,
3153 size_t cnt, loff_t *ppos)
3155 struct trace_iterator *iter = filp->private_data;
3156 static struct tracer *old_tracer;
3159 /* return any leftover data */
3160 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3164 trace_seq_init(&iter->seq);
3166 /* copy the tracer to avoid using a global lock all around */
3167 mutex_lock(&trace_types_lock);
3168 if (unlikely(old_tracer != current_trace && current_trace)) {
3169 old_tracer = current_trace;
3170 *iter->trace = *current_trace;
3172 mutex_unlock(&trace_types_lock);
3175 * Avoid more than one consumer on a single file descriptor
3176 * This is just a matter of traces coherency, the ring buffer itself
3179 mutex_lock(&iter->mutex);
3180 if (iter->trace->read) {
3181 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3187 sret = tracing_wait_pipe(filp);
3191 /* stop when tracing is finished */
3192 if (trace_empty(iter)) {
3197 if (cnt >= PAGE_SIZE)
3198 cnt = PAGE_SIZE - 1;
3200 /* reset all but tr, trace, and overruns */
3201 memset(&iter->seq, 0,
3202 sizeof(struct trace_iterator) -
3203 offsetof(struct trace_iterator, seq));
3206 trace_event_read_lock();
3207 trace_access_lock(iter->cpu_file);
3208 while (find_next_entry_inc(iter) != NULL) {
3209 enum print_line_t ret;
3210 int len = iter->seq.len;
3212 ret = print_trace_line(iter);
3213 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3214 /* don't print partial lines */
3215 iter->seq.len = len;
3218 if (ret != TRACE_TYPE_NO_CONSUME)
3219 trace_consume(iter);
3221 if (iter->seq.len >= cnt)
3224 trace_access_unlock(iter->cpu_file);
3225 trace_event_read_unlock();
3227 /* Now copy what we have to the user */
3228 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3229 if (iter->seq.readpos >= iter->seq.len)
3230 trace_seq_init(&iter->seq);
3233 * If there was nothing to send to user, inspite of consuming trace
3234 * entries, go back to wait for more entries.
3240 mutex_unlock(&iter->mutex);
3245 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3246 struct pipe_buffer *buf)
3248 __free_page(buf->page);
3251 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3254 __free_page(spd->pages[idx]);
3257 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3259 .map = generic_pipe_buf_map,
3260 .unmap = generic_pipe_buf_unmap,
3261 .confirm = generic_pipe_buf_confirm,
3262 .release = tracing_pipe_buf_release,
3263 .steal = generic_pipe_buf_steal,
3264 .get = generic_pipe_buf_get,
3268 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3273 /* Seq buffer is page-sized, exactly what we need. */
3275 count = iter->seq.len;
3276 ret = print_trace_line(iter);
3277 count = iter->seq.len - count;
3280 iter->seq.len -= count;
3283 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3284 iter->seq.len -= count;
3288 if (ret != TRACE_TYPE_NO_CONSUME)
3289 trace_consume(iter);
3291 if (!find_next_entry_inc(iter)) {
3301 static ssize_t tracing_splice_read_pipe(struct file *filp,
3303 struct pipe_inode_info *pipe,
3307 struct page *pages[PIPE_BUFFERS];
3308 struct partial_page partial[PIPE_BUFFERS];
3309 struct trace_iterator *iter = filp->private_data;
3310 struct splice_pipe_desc spd = {
3313 .nr_pages = 0, /* This gets updated below. */
3315 .ops = &tracing_pipe_buf_ops,
3316 .spd_release = tracing_spd_release_pipe,
3318 static struct tracer *old_tracer;
3323 /* copy the tracer to avoid using a global lock all around */
3324 mutex_lock(&trace_types_lock);
3325 if (unlikely(old_tracer != current_trace && current_trace)) {
3326 old_tracer = current_trace;
3327 *iter->trace = *current_trace;
3329 mutex_unlock(&trace_types_lock);
3331 mutex_lock(&iter->mutex);
3333 if (iter->trace->splice_read) {
3334 ret = iter->trace->splice_read(iter, filp,
3335 ppos, pipe, len, flags);
3340 ret = tracing_wait_pipe(filp);
3344 if (!iter->ent && !find_next_entry_inc(iter)) {
3349 trace_event_read_lock();
3350 trace_access_lock(iter->cpu_file);
3352 /* Fill as many pages as possible. */
3353 for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3354 pages[i] = alloc_page(GFP_KERNEL);
3358 rem = tracing_fill_pipe_page(rem, iter);
3360 /* Copy the data into the page, so we can start over. */
3361 ret = trace_seq_to_buffer(&iter->seq,
3362 page_address(pages[i]),
3365 __free_page(pages[i]);
3368 partial[i].offset = 0;
3369 partial[i].len = iter->seq.len;
3371 trace_seq_init(&iter->seq);
3374 trace_access_unlock(iter->cpu_file);
3375 trace_event_read_unlock();
3376 mutex_unlock(&iter->mutex);
3380 return splice_to_pipe(pipe, &spd);
3383 mutex_unlock(&iter->mutex);
3389 tracing_entries_read(struct file *filp, char __user *ubuf,
3390 size_t cnt, loff_t *ppos)
3392 struct trace_array *tr = filp->private_data;
3396 mutex_lock(&trace_types_lock);
3397 if (!ring_buffer_expanded)
3398 r = sprintf(buf, "%lu (expanded: %lu)\n",
3400 trace_buf_size >> 10);
3402 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3403 mutex_unlock(&trace_types_lock);
3405 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3409 tracing_entries_write(struct file *filp, const char __user *ubuf,
3410 size_t cnt, loff_t *ppos)
3416 if (cnt >= sizeof(buf))
3419 if (copy_from_user(&buf, ubuf, cnt))
3424 ret = strict_strtoul(buf, 10, &val);
3428 /* must have at least 1 entry */
3432 mutex_lock(&trace_types_lock);
3436 /* disable all cpu buffers */
3437 for_each_tracing_cpu(cpu) {
3438 if (global_trace.data[cpu])
3439 atomic_inc(&global_trace.data[cpu]->disabled);
3440 if (max_tr.data[cpu])
3441 atomic_inc(&max_tr.data[cpu]->disabled);
3444 /* value is in KB */
3447 if (val != global_trace.entries) {
3448 ret = tracing_resize_ring_buffer(val);
3457 /* If check pages failed, return ENOMEM */
3458 if (tracing_disabled)
3461 for_each_tracing_cpu(cpu) {
3462 if (global_trace.data[cpu])
3463 atomic_dec(&global_trace.data[cpu]->disabled);
3464 if (max_tr.data[cpu])
3465 atomic_dec(&max_tr.data[cpu]->disabled);
3469 max_tr.entries = global_trace.entries;
3470 mutex_unlock(&trace_types_lock);
3475 static int mark_printk(const char *fmt, ...)
3479 va_start(args, fmt);
3480 ret = trace_vprintk(0, fmt, args);
3486 tracing_mark_write(struct file *filp, const char __user *ubuf,
3487 size_t cnt, loff_t *fpos)
3491 if (tracing_disabled)
3494 if (cnt > TRACE_BUF_SIZE)
3495 cnt = TRACE_BUF_SIZE;
3497 buf = kmalloc(cnt + 2, GFP_KERNEL);
3501 if (copy_from_user(buf, ubuf, cnt)) {
3505 if (buf[cnt-1] != '\n') {
3511 cnt = mark_printk("%s", buf);
3518 static int tracing_clock_show(struct seq_file *m, void *v)
3522 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3524 "%s%s%s%s", i ? " " : "",
3525 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3526 i == trace_clock_id ? "]" : "");
3532 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3533 size_t cnt, loff_t *fpos)
3536 const char *clockstr;
3539 if (cnt >= sizeof(buf))
3542 if (copy_from_user(&buf, ubuf, cnt))
3547 clockstr = strstrip(buf);
3549 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3550 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3553 if (i == ARRAY_SIZE(trace_clocks))
3558 mutex_lock(&trace_types_lock);
3560 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3562 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3564 mutex_unlock(&trace_types_lock);
3571 static int tracing_clock_open(struct inode *inode, struct file *file)
3573 if (tracing_disabled)
3575 return single_open(file, tracing_clock_show, NULL);
3578 static const struct file_operations tracing_max_lat_fops = {
3579 .open = tracing_open_generic,
3580 .read = tracing_max_lat_read,
3581 .write = tracing_max_lat_write,
3584 static const struct file_operations tracing_ctrl_fops = {
3585 .open = tracing_open_generic,
3586 .read = tracing_ctrl_read,
3587 .write = tracing_ctrl_write,
3590 static const struct file_operations set_tracer_fops = {
3591 .open = tracing_open_generic,
3592 .read = tracing_set_trace_read,
3593 .write = tracing_set_trace_write,
3596 static const struct file_operations tracing_pipe_fops = {
3597 .open = tracing_open_pipe,
3598 .poll = tracing_poll_pipe,
3599 .read = tracing_read_pipe,
3600 .splice_read = tracing_splice_read_pipe,
3601 .release = tracing_release_pipe,
3604 static const struct file_operations tracing_entries_fops = {
3605 .open = tracing_open_generic,
3606 .read = tracing_entries_read,
3607 .write = tracing_entries_write,
3610 static const struct file_operations tracing_mark_fops = {
3611 .open = tracing_open_generic,
3612 .write = tracing_mark_write,
3615 static const struct file_operations trace_clock_fops = {
3616 .open = tracing_clock_open,
3618 .llseek = seq_lseek,
3619 .release = single_release,
3620 .write = tracing_clock_write,
3623 struct ftrace_buffer_info {
3624 struct trace_array *tr;
3630 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3632 int cpu = (int)(long)inode->i_private;
3633 struct ftrace_buffer_info *info;
3635 if (tracing_disabled)
3638 info = kzalloc(sizeof(*info), GFP_KERNEL);
3642 info->tr = &global_trace;
3645 /* Force reading ring buffer for first read */
3646 info->read = (unsigned int)-1;
3648 filp->private_data = info;
3650 return nonseekable_open(inode, filp);
3654 tracing_buffers_read(struct file *filp, char __user *ubuf,
3655 size_t count, loff_t *ppos)
3657 struct ftrace_buffer_info *info = filp->private_data;
3666 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3670 /* Do we have previous read data to read? */
3671 if (info->read < PAGE_SIZE)
3676 trace_access_lock(info->cpu);
3677 ret = ring_buffer_read_page(info->tr->buffer,
3681 trace_access_unlock(info->cpu);
3685 pos = ring_buffer_page_len(info->spare);
3687 if (pos < PAGE_SIZE)
3688 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3691 size = PAGE_SIZE - info->read;
3695 ret = copy_to_user(ubuf, info->spare + info->read, size);
3706 static int tracing_buffers_release(struct inode *inode, struct file *file)
3708 struct ftrace_buffer_info *info = file->private_data;
3711 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3718 struct ring_buffer *buffer;
3723 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3724 struct pipe_buffer *buf)
3726 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3731 ring_buffer_free_read_page(ref->buffer, ref->page);
3736 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3737 struct pipe_buffer *buf)
3742 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3743 struct pipe_buffer *buf)
3745 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3750 /* Pipe buffer operations for a buffer. */
3751 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
3753 .map = generic_pipe_buf_map,
3754 .unmap = generic_pipe_buf_unmap,
3755 .confirm = generic_pipe_buf_confirm,
3756 .release = buffer_pipe_buf_release,
3757 .steal = buffer_pipe_buf_steal,
3758 .get = buffer_pipe_buf_get,
3762 * Callback from splice_to_pipe(), if we need to release some pages
3763 * at the end of the spd in case we error'ed out in filling the pipe.
3765 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3767 struct buffer_ref *ref =
3768 (struct buffer_ref *)spd->partial[i].private;
3773 ring_buffer_free_read_page(ref->buffer, ref->page);
3775 spd->partial[i].private = 0;
3779 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3780 struct pipe_inode_info *pipe, size_t len,
3783 struct ftrace_buffer_info *info = file->private_data;
3784 struct partial_page partial[PIPE_BUFFERS];
3785 struct page *pages[PIPE_BUFFERS];
3786 struct splice_pipe_desc spd = {
3790 .ops = &buffer_pipe_buf_ops,
3791 .spd_release = buffer_spd_release,
3793 struct buffer_ref *ref;
3794 int entries, size, i;
3797 if (*ppos & (PAGE_SIZE - 1)) {
3798 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3802 if (len & (PAGE_SIZE - 1)) {
3803 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3804 if (len < PAGE_SIZE)
3809 trace_access_lock(info->cpu);
3810 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3812 for (i = 0; i < PIPE_BUFFERS && len && entries; i++, len -= PAGE_SIZE) {
3816 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3821 ref->buffer = info->tr->buffer;
3822 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3828 r = ring_buffer_read_page(ref->buffer, &ref->page,
3831 ring_buffer_free_read_page(ref->buffer,
3838 * zero out any left over data, this is going to
3841 size = ring_buffer_page_len(ref->page);
3842 if (size < PAGE_SIZE)
3843 memset(ref->page + size, 0, PAGE_SIZE - size);
3845 page = virt_to_page(ref->page);
3847 spd.pages[i] = page;
3848 spd.partial[i].len = PAGE_SIZE;
3849 spd.partial[i].offset = 0;
3850 spd.partial[i].private = (unsigned long)ref;
3854 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3857 trace_access_unlock(info->cpu);
3860 /* did we read anything? */
3861 if (!spd.nr_pages) {
3862 if (flags & SPLICE_F_NONBLOCK)
3870 ret = splice_to_pipe(pipe, &spd);
3875 static const struct file_operations tracing_buffers_fops = {
3876 .open = tracing_buffers_open,
3877 .read = tracing_buffers_read,
3878 .release = tracing_buffers_release,
3879 .splice_read = tracing_buffers_splice_read,
3880 .llseek = no_llseek,
3884 tracing_stats_read(struct file *filp, char __user *ubuf,
3885 size_t count, loff_t *ppos)
3887 unsigned long cpu = (unsigned long)filp->private_data;
3888 struct trace_array *tr = &global_trace;
3889 struct trace_seq *s;
3892 s = kmalloc(sizeof(*s), GFP_KERNEL);
3898 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3899 trace_seq_printf(s, "entries: %ld\n", cnt);
3901 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3902 trace_seq_printf(s, "overrun: %ld\n", cnt);
3904 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3905 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3907 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3914 static const struct file_operations tracing_stats_fops = {
3915 .open = tracing_open_generic,
3916 .read = tracing_stats_read,
3919 #ifdef CONFIG_DYNAMIC_FTRACE
3921 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3927 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3928 size_t cnt, loff_t *ppos)
3930 static char ftrace_dyn_info_buffer[1024];
3931 static DEFINE_MUTEX(dyn_info_mutex);
3932 unsigned long *p = filp->private_data;
3933 char *buf = ftrace_dyn_info_buffer;
3934 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3937 mutex_lock(&dyn_info_mutex);
3938 r = sprintf(buf, "%ld ", *p);
3940 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3943 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3945 mutex_unlock(&dyn_info_mutex);
3950 static const struct file_operations tracing_dyn_info_fops = {
3951 .open = tracing_open_generic,
3952 .read = tracing_read_dyn_info,
3956 static struct dentry *d_tracer;
3958 struct dentry *tracing_init_dentry(void)
3965 if (!debugfs_initialized())
3968 d_tracer = debugfs_create_dir("tracing", NULL);
3970 if (!d_tracer && !once) {
3972 pr_warning("Could not create debugfs directory 'tracing'\n");
3979 static struct dentry *d_percpu;
3981 struct dentry *tracing_dentry_percpu(void)
3984 struct dentry *d_tracer;
3989 d_tracer = tracing_init_dentry();
3994 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3996 if (!d_percpu && !once) {
3998 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4005 static void tracing_init_debugfs_percpu(long cpu)
4007 struct dentry *d_percpu = tracing_dentry_percpu();
4008 struct dentry *d_cpu;
4009 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
4012 if (cpu > 999 || cpu < 0)
4015 sprintf(cpu_dir, "cpu%ld", cpu);
4016 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4018 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4022 /* per cpu trace_pipe */
4023 trace_create_file("trace_pipe", 0444, d_cpu,
4024 (void *) cpu, &tracing_pipe_fops);
4027 trace_create_file("trace", 0644, d_cpu,
4028 (void *) cpu, &tracing_fops);
4030 trace_create_file("trace_pipe_raw", 0444, d_cpu,
4031 (void *) cpu, &tracing_buffers_fops);
4033 trace_create_file("stats", 0444, d_cpu,
4034 (void *) cpu, &tracing_stats_fops);
4037 #ifdef CONFIG_FTRACE_SELFTEST
4038 /* Let selftest have access to static functions in this file */
4039 #include "trace_selftest.c"
4042 struct trace_option_dentry {
4043 struct tracer_opt *opt;
4044 struct tracer_flags *flags;
4045 struct dentry *entry;
4049 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4052 struct trace_option_dentry *topt = filp->private_data;
4055 if (topt->flags->val & topt->opt->bit)
4060 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4064 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4067 struct trace_option_dentry *topt = filp->private_data;
4072 if (cnt >= sizeof(buf))
4075 if (copy_from_user(&buf, ubuf, cnt))
4080 ret = strict_strtoul(buf, 10, &val);
4084 if (val != 0 && val != 1)
4087 if (!!(topt->flags->val & topt->opt->bit) != val) {
4088 mutex_lock(&trace_types_lock);
4089 ret = __set_tracer_option(current_trace, topt->flags,
4091 mutex_unlock(&trace_types_lock);
4102 static const struct file_operations trace_options_fops = {
4103 .open = tracing_open_generic,
4104 .read = trace_options_read,
4105 .write = trace_options_write,
4109 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4112 long index = (long)filp->private_data;
4115 if (trace_flags & (1 << index))
4120 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4124 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4127 long index = (long)filp->private_data;
4132 if (cnt >= sizeof(buf))
4135 if (copy_from_user(&buf, ubuf, cnt))
4140 ret = strict_strtoul(buf, 10, &val);
4144 if (val != 0 && val != 1)
4146 set_tracer_flags(1 << index, val);
4153 static const struct file_operations trace_options_core_fops = {
4154 .open = tracing_open_generic,
4155 .read = trace_options_core_read,
4156 .write = trace_options_core_write,
4159 struct dentry *trace_create_file(const char *name,
4161 struct dentry *parent,
4163 const struct file_operations *fops)
4167 ret = debugfs_create_file(name, mode, parent, data, fops);
4169 pr_warning("Could not create debugfs '%s' entry\n", name);
4175 static struct dentry *trace_options_init_dentry(void)
4177 struct dentry *d_tracer;
4178 static struct dentry *t_options;
4183 d_tracer = tracing_init_dentry();
4187 t_options = debugfs_create_dir("options", d_tracer);
4189 pr_warning("Could not create debugfs directory 'options'\n");
4197 create_trace_option_file(struct trace_option_dentry *topt,
4198 struct tracer_flags *flags,
4199 struct tracer_opt *opt)
4201 struct dentry *t_options;
4203 t_options = trace_options_init_dentry();
4207 topt->flags = flags;
4210 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4211 &trace_options_fops);
4215 static struct trace_option_dentry *
4216 create_trace_option_files(struct tracer *tracer)
4218 struct trace_option_dentry *topts;
4219 struct tracer_flags *flags;
4220 struct tracer_opt *opts;
4226 flags = tracer->flags;
4228 if (!flags || !flags->opts)
4233 for (cnt = 0; opts[cnt].name; cnt++)
4236 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4240 for (cnt = 0; opts[cnt].name; cnt++)
4241 create_trace_option_file(&topts[cnt], flags,
4248 destroy_trace_option_files(struct trace_option_dentry *topts)
4255 for (cnt = 0; topts[cnt].opt; cnt++) {
4256 if (topts[cnt].entry)
4257 debugfs_remove(topts[cnt].entry);
4263 static struct dentry *
4264 create_trace_option_core_file(const char *option, long index)
4266 struct dentry *t_options;
4268 t_options = trace_options_init_dentry();
4272 return trace_create_file(option, 0644, t_options, (void *)index,
4273 &trace_options_core_fops);
4276 static __init void create_trace_options_dir(void)
4278 struct dentry *t_options;
4281 t_options = trace_options_init_dentry();
4285 for (i = 0; trace_options[i]; i++)
4286 create_trace_option_core_file(trace_options[i], i);
4289 static __init int tracer_init_debugfs(void)
4291 struct dentry *d_tracer;
4294 trace_access_lock_init();
4296 d_tracer = tracing_init_dentry();
4298 trace_create_file("tracing_enabled", 0644, d_tracer,
4299 &global_trace, &tracing_ctrl_fops);
4301 trace_create_file("trace_options", 0644, d_tracer,
4302 NULL, &tracing_iter_fops);
4304 trace_create_file("tracing_cpumask", 0644, d_tracer,
4305 NULL, &tracing_cpumask_fops);
4307 trace_create_file("trace", 0644, d_tracer,
4308 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4310 trace_create_file("available_tracers", 0444, d_tracer,
4311 &global_trace, &show_traces_fops);
4313 trace_create_file("current_tracer", 0644, d_tracer,
4314 &global_trace, &set_tracer_fops);
4316 #ifdef CONFIG_TRACER_MAX_TRACE
4317 trace_create_file("tracing_max_latency", 0644, d_tracer,
4318 &tracing_max_latency, &tracing_max_lat_fops);
4321 trace_create_file("tracing_thresh", 0644, d_tracer,
4322 &tracing_thresh, &tracing_max_lat_fops);
4324 trace_create_file("README", 0444, d_tracer,
4325 NULL, &tracing_readme_fops);
4327 trace_create_file("trace_pipe", 0444, d_tracer,
4328 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4330 trace_create_file("buffer_size_kb", 0644, d_tracer,
4331 &global_trace, &tracing_entries_fops);
4333 trace_create_file("trace_marker", 0220, d_tracer,
4334 NULL, &tracing_mark_fops);
4336 trace_create_file("saved_cmdlines", 0444, d_tracer,
4337 NULL, &tracing_saved_cmdlines_fops);
4339 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4342 #ifdef CONFIG_DYNAMIC_FTRACE
4343 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4344 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4346 #ifdef CONFIG_SYSPROF_TRACER
4347 init_tracer_sysprof_debugfs(d_tracer);
4350 create_trace_options_dir();
4352 for_each_tracing_cpu(cpu)
4353 tracing_init_debugfs_percpu(cpu);
4358 static int trace_panic_handler(struct notifier_block *this,
4359 unsigned long event, void *unused)
4361 if (ftrace_dump_on_oops)
4362 ftrace_dump(ftrace_dump_on_oops);
4366 static struct notifier_block trace_panic_notifier = {
4367 .notifier_call = trace_panic_handler,
4369 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4372 static int trace_die_handler(struct notifier_block *self,
4378 if (ftrace_dump_on_oops)
4379 ftrace_dump(ftrace_dump_on_oops);
4387 static struct notifier_block trace_die_notifier = {
4388 .notifier_call = trace_die_handler,
4393 * printk is set to max of 1024, we really don't need it that big.
4394 * Nothing should be printing 1000 characters anyway.
4396 #define TRACE_MAX_PRINT 1000
4399 * Define here KERN_TRACE so that we have one place to modify
4400 * it if we decide to change what log level the ftrace dump
4403 #define KERN_TRACE KERN_EMERG
4406 trace_printk_seq(struct trace_seq *s)
4408 /* Probably should print a warning here. */
4412 /* should be zero ended, but we are paranoid. */
4413 s->buffer[s->len] = 0;
4415 printk(KERN_TRACE "%s", s->buffer);
4421 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4423 static arch_spinlock_t ftrace_dump_lock =
4424 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4425 /* use static because iter can be a bit big for the stack */
4426 static struct trace_iterator iter;
4427 unsigned int old_userobj;
4428 static int dump_ran;
4429 unsigned long flags;
4433 local_irq_save(flags);
4434 arch_spin_lock(&ftrace_dump_lock);
4442 if (disable_tracing)
4445 for_each_tracing_cpu(cpu) {
4446 atomic_inc(&global_trace.data[cpu]->disabled);
4449 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4451 /* don't look at user memory in panic mode */
4452 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4454 /* Simulate the iterator */
4455 iter.tr = &global_trace;
4456 iter.trace = current_trace;
4458 switch (oops_dump_mode) {
4460 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4463 iter.cpu_file = raw_smp_processor_id();
4468 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
4469 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4472 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4475 * We need to stop all tracing on all CPUS to read the
4476 * the next buffer. This is a bit expensive, but is
4477 * not done often. We fill all what we can read,
4478 * and then release the locks again.
4481 while (!trace_empty(&iter)) {
4484 printk(KERN_TRACE "---------------------------------\n");
4488 /* reset all but tr, trace, and overruns */
4489 memset(&iter.seq, 0,
4490 sizeof(struct trace_iterator) -
4491 offsetof(struct trace_iterator, seq));
4492 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4495 if (find_next_entry_inc(&iter) != NULL) {
4498 ret = print_trace_line(&iter);
4499 if (ret != TRACE_TYPE_NO_CONSUME)
4500 trace_consume(&iter);
4503 trace_printk_seq(&iter.seq);
4507 printk(KERN_TRACE " (ftrace buffer empty)\n");
4509 printk(KERN_TRACE "---------------------------------\n");
4512 /* Re-enable tracing if requested */
4513 if (!disable_tracing) {
4514 trace_flags |= old_userobj;
4516 for_each_tracing_cpu(cpu) {
4517 atomic_dec(&global_trace.data[cpu]->disabled);
4523 arch_spin_unlock(&ftrace_dump_lock);
4524 local_irq_restore(flags);
4527 /* By default: disable tracing after the dump */
4528 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
4530 __ftrace_dump(true, oops_dump_mode);
4533 __init static int tracer_alloc_buffers(void)
4539 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4542 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4543 goto out_free_buffer_mask;
4545 /* To save memory, keep the ring buffer size to its minimum */
4546 if (ring_buffer_expanded)
4547 ring_buf_size = trace_buf_size;
4551 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4552 cpumask_copy(tracing_cpumask, cpu_all_mask);
4554 /* TODO: make the number of buffers hot pluggable with CPUS */
4555 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4556 TRACE_BUFFER_FLAGS);
4557 if (!global_trace.buffer) {
4558 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4560 goto out_free_cpumask;
4562 global_trace.entries = ring_buffer_size(global_trace.buffer);
4565 #ifdef CONFIG_TRACER_MAX_TRACE
4566 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4567 TRACE_BUFFER_FLAGS);
4568 if (!max_tr.buffer) {
4569 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4571 ring_buffer_free(global_trace.buffer);
4572 goto out_free_cpumask;
4574 max_tr.entries = ring_buffer_size(max_tr.buffer);
4575 WARN_ON(max_tr.entries != global_trace.entries);
4578 /* Allocate the first page for all buffers */
4579 for_each_tracing_cpu(i) {
4580 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4581 max_tr.data[i] = &per_cpu(max_tr_data, i);
4584 trace_init_cmdlines();
4586 register_tracer(&nop_trace);
4587 current_trace = &nop_trace;
4588 #ifdef CONFIG_BOOT_TRACER
4589 register_tracer(&boot_tracer);
4591 /* All seems OK, enable tracing */
4592 tracing_disabled = 0;
4594 atomic_notifier_chain_register(&panic_notifier_list,
4595 &trace_panic_notifier);
4597 register_die_notifier(&trace_die_notifier);
4602 free_cpumask_var(tracing_cpumask);
4603 out_free_buffer_mask:
4604 free_cpumask_var(tracing_buffer_mask);
4609 __init static int clear_boot_tracer(void)
4612 * The default tracer at boot buffer is an init section.
4613 * This function is called in lateinit. If we did not
4614 * find the boot tracer, then clear it out, to prevent
4615 * later registration from accessing the buffer that is
4616 * about to be freed.
4618 if (!default_bootup_tracer)
4621 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4622 default_bootup_tracer);
4623 default_bootup_tracer = NULL;
4628 early_initcall(tracer_alloc_buffers);
4629 fs_initcall(tracer_init_debugfs);
4630 late_initcall(clear_boot_tracer);