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/ctype.h>
37 #include <linux/init.h>
38 #include <linux/poll.h>
39 #include <linux/gfp.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(per_cpu_var(ftrace_cpu_disabled));
98 static inline void ftrace_enable_cpu(void)
100 __this_cpu_dec(per_cpu_var(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 to true.
122 int ftrace_dump_on_oops;
124 static int tracing_set_tracer(const char *buf);
126 #define MAX_TRACER_SIZE 100
127 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
128 static char *default_bootup_tracer;
130 static int __init set_cmdline_ftrace(char *str)
132 strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
133 default_bootup_tracer = bootup_tracer_buf;
134 /* We are using ftrace early, expand it */
135 ring_buffer_expanded = 1;
138 __setup("ftrace=", set_cmdline_ftrace);
140 static int __init set_ftrace_dump_on_oops(char *str)
142 ftrace_dump_on_oops = 1;
145 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
147 unsigned long long ns2usecs(cycle_t nsec)
155 * The global_trace is the descriptor that holds the tracing
156 * buffers for the live tracing. For each CPU, it contains
157 * a link list of pages that will store trace entries. The
158 * page descriptor of the pages in the memory is used to hold
159 * the link list by linking the lru item in the page descriptor
160 * to each of the pages in the buffer per CPU.
162 * For each active CPU there is a data field that holds the
163 * pages for the buffer for that CPU. Each CPU has the same number
164 * of pages allocated for its buffer.
166 static struct trace_array global_trace;
168 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
170 int filter_current_check_discard(struct ring_buffer *buffer,
171 struct ftrace_event_call *call, void *rec,
172 struct ring_buffer_event *event)
174 return filter_check_discard(call, rec, buffer, event);
176 EXPORT_SYMBOL_GPL(filter_current_check_discard);
178 cycle_t ftrace_now(int cpu)
182 /* Early boot up does not have a buffer yet */
183 if (!global_trace.buffer)
184 return trace_clock_local();
186 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
187 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
193 * The max_tr is used to snapshot the global_trace when a maximum
194 * latency is reached. Some tracers will use this to store a maximum
195 * trace while it continues examining live traces.
197 * The buffers for the max_tr are set up the same as the global_trace.
198 * When a snapshot is taken, the link list of the max_tr is swapped
199 * with the link list of the global_trace and the buffers are reset for
200 * the global_trace so the tracing can continue.
202 static struct trace_array max_tr;
204 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
206 /* tracer_enabled is used to toggle activation of a tracer */
207 static int tracer_enabled = 1;
210 * tracing_is_enabled - return tracer_enabled status
212 * This function is used by other tracers to know the status
213 * of the tracer_enabled flag. Tracers may use this function
214 * to know if it should enable their features when starting
215 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
217 int tracing_is_enabled(void)
219 return tracer_enabled;
223 * trace_buf_size is the size in bytes that is allocated
224 * for a buffer. Note, the number of bytes is always rounded
227 * This number is purposely set to a low number of 16384.
228 * If the dump on oops happens, it will be much appreciated
229 * to not have to wait for all that output. Anyway this can be
230 * boot time and run time configurable.
232 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
234 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
236 /* trace_types holds a link list of available tracers. */
237 static struct tracer *trace_types __read_mostly;
239 /* current_trace points to the tracer that is currently active */
240 static struct tracer *current_trace __read_mostly;
243 * trace_types_lock is used to protect the trace_types list.
245 static DEFINE_MUTEX(trace_types_lock);
248 * serialize the access of the ring buffer
250 * ring buffer serializes readers, but it is low level protection.
251 * The validity of the events (which returns by ring_buffer_peek() ..etc)
252 * are not protected by ring buffer.
254 * The content of events may become garbage if we allow other process consumes
255 * these events concurrently:
256 * A) the page of the consumed events may become a normal page
257 * (not reader page) in ring buffer, and this page will be rewrited
258 * by events producer.
259 * B) The page of the consumed events may become a page for splice_read,
260 * and this page will be returned to system.
262 * These primitives allow multi process access to different cpu ring buffer
265 * These primitives don't distinguish read-only and read-consume access.
266 * Multi read-only access are also serialized.
270 static DECLARE_RWSEM(all_cpu_access_lock);
271 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
273 static inline void trace_access_lock(int cpu)
275 if (cpu == TRACE_PIPE_ALL_CPU) {
276 /* gain it for accessing the whole ring buffer. */
277 down_write(&all_cpu_access_lock);
279 /* gain it for accessing a cpu ring buffer. */
281 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
282 down_read(&all_cpu_access_lock);
284 /* Secondly block other access to this @cpu ring buffer. */
285 mutex_lock(&per_cpu(cpu_access_lock, cpu));
289 static inline void trace_access_unlock(int cpu)
291 if (cpu == TRACE_PIPE_ALL_CPU) {
292 up_write(&all_cpu_access_lock);
294 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
295 up_read(&all_cpu_access_lock);
299 static inline void trace_access_lock_init(void)
303 for_each_possible_cpu(cpu)
304 mutex_init(&per_cpu(cpu_access_lock, cpu));
309 static DEFINE_MUTEX(access_lock);
311 static inline void trace_access_lock(int cpu)
314 mutex_lock(&access_lock);
317 static inline void trace_access_unlock(int cpu)
320 mutex_unlock(&access_lock);
323 static inline void trace_access_lock_init(void)
329 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
330 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
332 /* trace_flags holds trace_options default values */
333 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
334 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
335 TRACE_ITER_GRAPH_TIME;
337 static int trace_stop_count;
338 static DEFINE_SPINLOCK(tracing_start_lock);
341 * trace_wake_up - wake up tasks waiting for trace input
343 * Simply wakes up any task that is blocked on the trace_wait
344 * queue. These is used with trace_poll for tasks polling the trace.
346 void trace_wake_up(void)
350 if (trace_flags & TRACE_ITER_BLOCK)
353 * The runqueue_is_locked() can fail, but this is the best we
357 if (!runqueue_is_locked(cpu))
358 wake_up(&trace_wait);
362 static int __init set_buf_size(char *str)
364 unsigned long buf_size;
368 buf_size = memparse(str, &str);
369 /* nr_entries can not be zero */
372 trace_buf_size = buf_size;
375 __setup("trace_buf_size=", set_buf_size);
377 static int __init set_tracing_thresh(char *str)
379 unsigned long threshhold;
384 ret = strict_strtoul(str, 0, &threshhold);
387 tracing_thresh = threshhold * 1000;
390 __setup("tracing_thresh=", set_tracing_thresh);
392 unsigned long nsecs_to_usecs(unsigned long nsecs)
397 /* These must match the bit postions in trace_iterator_flags */
398 static const char *trace_options[] = {
426 { trace_clock_local, "local" },
427 { trace_clock_global, "global" },
433 * trace_parser_get_init - gets the buffer for trace parser
435 int trace_parser_get_init(struct trace_parser *parser, int size)
437 memset(parser, 0, sizeof(*parser));
439 parser->buffer = kmalloc(size, GFP_KERNEL);
448 * trace_parser_put - frees the buffer for trace parser
450 void trace_parser_put(struct trace_parser *parser)
452 kfree(parser->buffer);
456 * trace_get_user - reads the user input string separated by space
457 * (matched by isspace(ch))
459 * For each string found the 'struct trace_parser' is updated,
460 * and the function returns.
462 * Returns number of bytes read.
464 * See kernel/trace/trace.h for 'struct trace_parser' details.
466 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
467 size_t cnt, loff_t *ppos)
474 trace_parser_clear(parser);
476 ret = get_user(ch, ubuf++);
484 * The parser is not finished with the last write,
485 * continue reading the user input without skipping spaces.
488 /* skip white space */
489 while (cnt && isspace(ch)) {
490 ret = get_user(ch, ubuf++);
497 /* only spaces were written */
507 /* read the non-space input */
508 while (cnt && !isspace(ch)) {
509 if (parser->idx < parser->size - 1)
510 parser->buffer[parser->idx++] = ch;
515 ret = get_user(ch, ubuf++);
522 /* We either got finished input or we have to wait for another call. */
524 parser->buffer[parser->idx] = 0;
525 parser->cont = false;
528 parser->buffer[parser->idx++] = ch;
538 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
546 if (s->len <= s->readpos)
549 len = s->len - s->readpos;
552 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
562 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
567 if (s->len <= s->readpos)
570 len = s->len - s->readpos;
573 ret = memcpy(buf, s->buffer + s->readpos, cnt);
582 * ftrace_max_lock is used to protect the swapping of buffers
583 * when taking a max snapshot. The buffers themselves are
584 * protected by per_cpu spinlocks. But the action of the swap
585 * needs its own lock.
587 * This is defined as a arch_spinlock_t in order to help
588 * with performance when lockdep debugging is enabled.
590 * It is also used in other places outside the update_max_tr
591 * so it needs to be defined outside of the
592 * CONFIG_TRACER_MAX_TRACE.
594 static arch_spinlock_t ftrace_max_lock =
595 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
597 unsigned long __read_mostly tracing_thresh;
599 #ifdef CONFIG_TRACER_MAX_TRACE
600 unsigned long __read_mostly tracing_max_latency;
603 * Copy the new maximum trace into the separate maximum-trace
604 * structure. (this way the maximum trace is permanently saved,
605 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
608 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
610 struct trace_array_cpu *data = tr->data[cpu];
611 struct trace_array_cpu *max_data;
614 max_tr.time_start = data->preempt_timestamp;
616 max_data = max_tr.data[cpu];
617 max_data->saved_latency = tracing_max_latency;
618 max_data->critical_start = data->critical_start;
619 max_data->critical_end = data->critical_end;
621 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
622 max_data->pid = tsk->pid;
623 max_data->uid = task_uid(tsk);
624 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
625 max_data->policy = tsk->policy;
626 max_data->rt_priority = tsk->rt_priority;
628 /* record this tasks comm */
629 tracing_record_cmdline(tsk);
633 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
635 * @tsk: the task with the latency
636 * @cpu: The cpu that initiated the trace.
638 * Flip the buffers between the @tr and the max_tr and record information
639 * about which task was the cause of this latency.
642 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
644 struct ring_buffer *buf = tr->buffer;
646 if (trace_stop_count)
649 WARN_ON_ONCE(!irqs_disabled());
650 arch_spin_lock(&ftrace_max_lock);
652 tr->buffer = max_tr.buffer;
655 __update_max_tr(tr, tsk, cpu);
656 arch_spin_unlock(&ftrace_max_lock);
660 * update_max_tr_single - only copy one trace over, and reset the rest
662 * @tsk - task with the latency
663 * @cpu - the cpu of the buffer to copy.
665 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
668 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
672 if (trace_stop_count)
675 WARN_ON_ONCE(!irqs_disabled());
676 arch_spin_lock(&ftrace_max_lock);
678 ftrace_disable_cpu();
680 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
684 * We failed to swap the buffer due to a commit taking
685 * place on this CPU. We fail to record, but we reset
686 * the max trace buffer (no one writes directly to it)
687 * and flag that it failed.
689 trace_array_printk(&max_tr, _THIS_IP_,
690 "Failed to swap buffers due to commit in progress\n");
695 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
697 __update_max_tr(tr, tsk, cpu);
698 arch_spin_unlock(&ftrace_max_lock);
700 #endif /* CONFIG_TRACER_MAX_TRACE */
703 * register_tracer - register a tracer with the ftrace system.
704 * @type - the plugin for the tracer
706 * Register a new plugin tracer.
708 int register_tracer(struct tracer *type)
709 __releases(kernel_lock)
710 __acquires(kernel_lock)
716 pr_info("Tracer must have a name\n");
720 if (strlen(type->name) > MAX_TRACER_SIZE) {
721 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
726 * When this gets called we hold the BKL which means that
727 * preemption is disabled. Various trace selftests however
728 * need to disable and enable preemption for successful tests.
729 * So we drop the BKL here and grab it after the tests again.
732 mutex_lock(&trace_types_lock);
734 tracing_selftest_running = true;
736 for (t = trace_types; t; t = t->next) {
737 if (strcmp(type->name, t->name) == 0) {
739 pr_info("Tracer %s already registered\n",
747 type->set_flag = &dummy_set_flag;
749 type->flags = &dummy_tracer_flags;
751 if (!type->flags->opts)
752 type->flags->opts = dummy_tracer_opt;
753 if (!type->wait_pipe)
754 type->wait_pipe = default_wait_pipe;
757 #ifdef CONFIG_FTRACE_STARTUP_TEST
758 if (type->selftest && !tracing_selftest_disabled) {
759 struct tracer *saved_tracer = current_trace;
760 struct trace_array *tr = &global_trace;
763 * Run a selftest on this tracer.
764 * Here we reset the trace buffer, and set the current
765 * tracer to be this tracer. The tracer can then run some
766 * internal tracing to verify that everything is in order.
767 * If we fail, we do not register this tracer.
769 tracing_reset_online_cpus(tr);
771 current_trace = type;
772 /* the test is responsible for initializing and enabling */
773 pr_info("Testing tracer %s: ", type->name);
774 ret = type->selftest(type, tr);
775 /* the test is responsible for resetting too */
776 current_trace = saved_tracer;
778 printk(KERN_CONT "FAILED!\n");
781 /* Only reset on passing, to avoid touching corrupted buffers */
782 tracing_reset_online_cpus(tr);
784 printk(KERN_CONT "PASSED\n");
788 type->next = trace_types;
792 tracing_selftest_running = false;
793 mutex_unlock(&trace_types_lock);
795 if (ret || !default_bootup_tracer)
798 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
801 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
802 /* Do we want this tracer to start on bootup? */
803 tracing_set_tracer(type->name);
804 default_bootup_tracer = NULL;
805 /* disable other selftests, since this will break it. */
806 tracing_selftest_disabled = 1;
807 #ifdef CONFIG_FTRACE_STARTUP_TEST
808 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
817 void unregister_tracer(struct tracer *type)
821 mutex_lock(&trace_types_lock);
822 for (t = &trace_types; *t; t = &(*t)->next) {
826 pr_info("Tracer %s not registered\n", type->name);
832 if (type == current_trace && tracer_enabled) {
835 if (current_trace->stop)
836 current_trace->stop(&global_trace);
837 current_trace = &nop_trace;
840 mutex_unlock(&trace_types_lock);
843 static void __tracing_reset(struct trace_array *tr, int cpu)
845 ftrace_disable_cpu();
846 ring_buffer_reset_cpu(tr->buffer, cpu);
850 void tracing_reset(struct trace_array *tr, int cpu)
852 struct ring_buffer *buffer = tr->buffer;
854 ring_buffer_record_disable(buffer);
856 /* Make sure all commits have finished */
858 __tracing_reset(tr, cpu);
860 ring_buffer_record_enable(buffer);
863 void tracing_reset_online_cpus(struct trace_array *tr)
865 struct ring_buffer *buffer = tr->buffer;
868 ring_buffer_record_disable(buffer);
870 /* Make sure all commits have finished */
873 tr->time_start = ftrace_now(tr->cpu);
875 for_each_online_cpu(cpu)
876 __tracing_reset(tr, cpu);
878 ring_buffer_record_enable(buffer);
881 void tracing_reset_current(int cpu)
883 tracing_reset(&global_trace, cpu);
886 void tracing_reset_current_online_cpus(void)
888 tracing_reset_online_cpus(&global_trace);
891 #define SAVED_CMDLINES 128
892 #define NO_CMDLINE_MAP UINT_MAX
893 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
894 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
895 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
896 static int cmdline_idx;
897 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
899 /* temporary disable recording */
900 static atomic_t trace_record_cmdline_disabled __read_mostly;
902 static void trace_init_cmdlines(void)
904 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
905 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
909 int is_tracing_stopped(void)
911 return trace_stop_count;
915 * ftrace_off_permanent - disable all ftrace code permanently
917 * This should only be called when a serious anomally has
918 * been detected. This will turn off the function tracing,
919 * ring buffers, and other tracing utilites. It takes no
920 * locks and can be called from any context.
922 void ftrace_off_permanent(void)
924 tracing_disabled = 1;
926 tracing_off_permanent();
930 * tracing_start - quick start of the tracer
932 * If tracing is enabled but was stopped by tracing_stop,
933 * this will start the tracer back up.
935 void tracing_start(void)
937 struct ring_buffer *buffer;
940 if (tracing_disabled)
943 spin_lock_irqsave(&tracing_start_lock, flags);
944 if (--trace_stop_count) {
945 if (trace_stop_count < 0) {
946 /* Someone screwed up their debugging */
948 trace_stop_count = 0;
954 buffer = global_trace.buffer;
956 ring_buffer_record_enable(buffer);
958 buffer = max_tr.buffer;
960 ring_buffer_record_enable(buffer);
964 spin_unlock_irqrestore(&tracing_start_lock, flags);
968 * tracing_stop - quick stop of the tracer
970 * Light weight way to stop tracing. Use in conjunction with
973 void tracing_stop(void)
975 struct ring_buffer *buffer;
979 spin_lock_irqsave(&tracing_start_lock, flags);
980 if (trace_stop_count++)
983 buffer = global_trace.buffer;
985 ring_buffer_record_disable(buffer);
987 buffer = max_tr.buffer;
989 ring_buffer_record_disable(buffer);
992 spin_unlock_irqrestore(&tracing_start_lock, flags);
995 void trace_stop_cmdline_recording(void);
997 static void trace_save_cmdline(struct task_struct *tsk)
1001 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1005 * It's not the end of the world if we don't get
1006 * the lock, but we also don't want to spin
1007 * nor do we want to disable interrupts,
1008 * so if we miss here, then better luck next time.
1010 if (!arch_spin_trylock(&trace_cmdline_lock))
1013 idx = map_pid_to_cmdline[tsk->pid];
1014 if (idx == NO_CMDLINE_MAP) {
1015 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1018 * Check whether the cmdline buffer at idx has a pid
1019 * mapped. We are going to overwrite that entry so we
1020 * need to clear the map_pid_to_cmdline. Otherwise we
1021 * would read the new comm for the old pid.
1023 pid = map_cmdline_to_pid[idx];
1024 if (pid != NO_CMDLINE_MAP)
1025 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1027 map_cmdline_to_pid[idx] = tsk->pid;
1028 map_pid_to_cmdline[tsk->pid] = idx;
1033 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1035 arch_spin_unlock(&trace_cmdline_lock);
1038 void trace_find_cmdline(int pid, char comm[])
1043 strcpy(comm, "<idle>");
1047 if (WARN_ON_ONCE(pid < 0)) {
1048 strcpy(comm, "<XXX>");
1052 if (pid > PID_MAX_DEFAULT) {
1053 strcpy(comm, "<...>");
1058 arch_spin_lock(&trace_cmdline_lock);
1059 map = map_pid_to_cmdline[pid];
1060 if (map != NO_CMDLINE_MAP)
1061 strcpy(comm, saved_cmdlines[map]);
1063 strcpy(comm, "<...>");
1065 arch_spin_unlock(&trace_cmdline_lock);
1069 void tracing_record_cmdline(struct task_struct *tsk)
1071 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1075 trace_save_cmdline(tsk);
1079 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1082 struct task_struct *tsk = current;
1084 entry->preempt_count = pc & 0xff;
1085 entry->pid = (tsk) ? tsk->pid : 0;
1086 entry->lock_depth = (tsk) ? tsk->lock_depth : 0;
1088 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1089 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1091 TRACE_FLAG_IRQS_NOSUPPORT |
1093 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1094 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1095 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1097 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1099 struct ring_buffer_event *
1100 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1103 unsigned long flags, int pc)
1105 struct ring_buffer_event *event;
1107 event = ring_buffer_lock_reserve(buffer, len);
1108 if (event != NULL) {
1109 struct trace_entry *ent = ring_buffer_event_data(event);
1111 tracing_generic_entry_update(ent, flags, pc);
1119 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1120 struct ring_buffer_event *event,
1121 unsigned long flags, int pc,
1124 ring_buffer_unlock_commit(buffer, event);
1126 ftrace_trace_stack(buffer, flags, 6, pc);
1127 ftrace_trace_userstack(buffer, flags, pc);
1133 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1134 struct ring_buffer_event *event,
1135 unsigned long flags, int pc)
1137 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1140 struct ring_buffer_event *
1141 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1142 int type, unsigned long len,
1143 unsigned long flags, int pc)
1145 *current_rb = global_trace.buffer;
1146 return trace_buffer_lock_reserve(*current_rb,
1147 type, len, flags, pc);
1149 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1151 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1152 struct ring_buffer_event *event,
1153 unsigned long flags, int pc)
1155 __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1157 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1159 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1160 struct ring_buffer_event *event,
1161 unsigned long flags, int pc)
1163 __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1165 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1167 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1168 struct ring_buffer_event *event)
1170 ring_buffer_discard_commit(buffer, event);
1172 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1175 trace_function(struct trace_array *tr,
1176 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1179 struct ftrace_event_call *call = &event_function;
1180 struct ring_buffer *buffer = tr->buffer;
1181 struct ring_buffer_event *event;
1182 struct ftrace_entry *entry;
1184 /* If we are reading the ring buffer, don't trace */
1185 if (unlikely(__this_cpu_read(per_cpu_var(ftrace_cpu_disabled))))
1188 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1192 entry = ring_buffer_event_data(event);
1194 entry->parent_ip = parent_ip;
1196 if (!filter_check_discard(call, entry, buffer, event))
1197 ring_buffer_unlock_commit(buffer, event);
1201 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1202 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1205 if (likely(!atomic_read(&data->disabled)))
1206 trace_function(tr, ip, parent_ip, flags, pc);
1209 #ifdef CONFIG_STACKTRACE
1210 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1211 unsigned long flags,
1214 struct ftrace_event_call *call = &event_kernel_stack;
1215 struct ring_buffer_event *event;
1216 struct stack_entry *entry;
1217 struct stack_trace trace;
1219 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1220 sizeof(*entry), flags, pc);
1223 entry = ring_buffer_event_data(event);
1224 memset(&entry->caller, 0, sizeof(entry->caller));
1226 trace.nr_entries = 0;
1227 trace.max_entries = FTRACE_STACK_ENTRIES;
1229 trace.entries = entry->caller;
1231 save_stack_trace(&trace);
1232 if (!filter_check_discard(call, entry, buffer, event))
1233 ring_buffer_unlock_commit(buffer, event);
1236 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1239 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1242 __ftrace_trace_stack(buffer, flags, skip, pc);
1245 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1248 __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1252 * trace_dump_stack - record a stack back trace in the trace buffer
1254 void trace_dump_stack(void)
1256 unsigned long flags;
1258 if (tracing_disabled || tracing_selftest_running)
1261 local_save_flags(flags);
1263 /* skipping 3 traces, seems to get us at the caller of this function */
1264 __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count());
1268 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1270 struct ftrace_event_call *call = &event_user_stack;
1271 struct ring_buffer_event *event;
1272 struct userstack_entry *entry;
1273 struct stack_trace trace;
1275 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1278 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1279 sizeof(*entry), flags, pc);
1282 entry = ring_buffer_event_data(event);
1284 entry->tgid = current->tgid;
1285 memset(&entry->caller, 0, sizeof(entry->caller));
1287 trace.nr_entries = 0;
1288 trace.max_entries = FTRACE_STACK_ENTRIES;
1290 trace.entries = entry->caller;
1292 save_stack_trace_user(&trace);
1293 if (!filter_check_discard(call, entry, buffer, event))
1294 ring_buffer_unlock_commit(buffer, event);
1298 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1300 ftrace_trace_userstack(tr, flags, preempt_count());
1304 #endif /* CONFIG_STACKTRACE */
1307 ftrace_trace_special(void *__tr,
1308 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1311 struct ftrace_event_call *call = &event_special;
1312 struct ring_buffer_event *event;
1313 struct trace_array *tr = __tr;
1314 struct ring_buffer *buffer = tr->buffer;
1315 struct special_entry *entry;
1317 event = trace_buffer_lock_reserve(buffer, TRACE_SPECIAL,
1318 sizeof(*entry), 0, pc);
1321 entry = ring_buffer_event_data(event);
1326 if (!filter_check_discard(call, entry, buffer, event))
1327 trace_buffer_unlock_commit(buffer, event, 0, pc);
1331 __trace_special(void *__tr, void *__data,
1332 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1334 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1338 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1340 struct trace_array *tr = &global_trace;
1341 struct trace_array_cpu *data;
1342 unsigned long flags;
1346 if (tracing_disabled)
1349 pc = preempt_count();
1350 local_irq_save(flags);
1351 cpu = raw_smp_processor_id();
1352 data = tr->data[cpu];
1354 if (likely(atomic_inc_return(&data->disabled) == 1))
1355 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1357 atomic_dec(&data->disabled);
1358 local_irq_restore(flags);
1362 * trace_vbprintk - write binary msg to tracing buffer
1365 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1367 static arch_spinlock_t trace_buf_lock =
1368 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1369 static u32 trace_buf[TRACE_BUF_SIZE];
1371 struct ftrace_event_call *call = &event_bprint;
1372 struct ring_buffer_event *event;
1373 struct ring_buffer *buffer;
1374 struct trace_array *tr = &global_trace;
1375 struct trace_array_cpu *data;
1376 struct bprint_entry *entry;
1377 unsigned long flags;
1380 int cpu, len = 0, size, pc;
1382 if (unlikely(tracing_selftest_running || tracing_disabled))
1385 /* Don't pollute graph traces with trace_vprintk internals */
1386 pause_graph_tracing();
1388 pc = preempt_count();
1389 resched = ftrace_preempt_disable();
1390 cpu = raw_smp_processor_id();
1391 data = tr->data[cpu];
1393 disable = atomic_inc_return(&data->disabled);
1394 if (unlikely(disable != 1))
1397 /* Lockdep uses trace_printk for lock tracing */
1398 local_irq_save(flags);
1399 arch_spin_lock(&trace_buf_lock);
1400 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1402 if (len > TRACE_BUF_SIZE || len < 0)
1405 size = sizeof(*entry) + sizeof(u32) * len;
1406 buffer = tr->buffer;
1407 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1411 entry = ring_buffer_event_data(event);
1415 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1416 if (!filter_check_discard(call, entry, buffer, event)) {
1417 ring_buffer_unlock_commit(buffer, event);
1418 ftrace_trace_stack(buffer, flags, 6, pc);
1422 arch_spin_unlock(&trace_buf_lock);
1423 local_irq_restore(flags);
1426 atomic_dec_return(&data->disabled);
1427 ftrace_preempt_enable(resched);
1428 unpause_graph_tracing();
1432 EXPORT_SYMBOL_GPL(trace_vbprintk);
1434 int trace_array_printk(struct trace_array *tr,
1435 unsigned long ip, const char *fmt, ...)
1440 if (!(trace_flags & TRACE_ITER_PRINTK))
1444 ret = trace_array_vprintk(tr, ip, fmt, ap);
1449 int trace_array_vprintk(struct trace_array *tr,
1450 unsigned long ip, const char *fmt, va_list args)
1452 static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1453 static char trace_buf[TRACE_BUF_SIZE];
1455 struct ftrace_event_call *call = &event_print;
1456 struct ring_buffer_event *event;
1457 struct ring_buffer *buffer;
1458 struct trace_array_cpu *data;
1459 int cpu, len = 0, size, pc;
1460 struct print_entry *entry;
1461 unsigned long irq_flags;
1464 if (tracing_disabled || tracing_selftest_running)
1467 pc = preempt_count();
1468 preempt_disable_notrace();
1469 cpu = raw_smp_processor_id();
1470 data = tr->data[cpu];
1472 disable = atomic_inc_return(&data->disabled);
1473 if (unlikely(disable != 1))
1476 pause_graph_tracing();
1477 raw_local_irq_save(irq_flags);
1478 arch_spin_lock(&trace_buf_lock);
1479 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1481 size = sizeof(*entry) + len + 1;
1482 buffer = tr->buffer;
1483 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1487 entry = ring_buffer_event_data(event);
1490 memcpy(&entry->buf, trace_buf, len);
1491 entry->buf[len] = '\0';
1492 if (!filter_check_discard(call, entry, buffer, event)) {
1493 ring_buffer_unlock_commit(buffer, event);
1494 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1498 arch_spin_unlock(&trace_buf_lock);
1499 raw_local_irq_restore(irq_flags);
1500 unpause_graph_tracing();
1502 atomic_dec_return(&data->disabled);
1503 preempt_enable_notrace();
1508 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1510 return trace_array_vprintk(&global_trace, ip, fmt, args);
1512 EXPORT_SYMBOL_GPL(trace_vprintk);
1514 enum trace_file_type {
1515 TRACE_FILE_LAT_FMT = 1,
1516 TRACE_FILE_ANNOTATE = 2,
1519 static void trace_iterator_increment(struct trace_iterator *iter)
1521 /* Don't allow ftrace to trace into the ring buffers */
1522 ftrace_disable_cpu();
1525 if (iter->buffer_iter[iter->cpu])
1526 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1528 ftrace_enable_cpu();
1531 static struct trace_entry *
1532 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1534 struct ring_buffer_event *event;
1535 struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1537 /* Don't allow ftrace to trace into the ring buffers */
1538 ftrace_disable_cpu();
1541 event = ring_buffer_iter_peek(buf_iter, ts);
1543 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1545 ftrace_enable_cpu();
1547 return event ? ring_buffer_event_data(event) : NULL;
1550 static struct trace_entry *
1551 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1553 struct ring_buffer *buffer = iter->tr->buffer;
1554 struct trace_entry *ent, *next = NULL;
1555 int cpu_file = iter->cpu_file;
1556 u64 next_ts = 0, ts;
1561 * If we are in a per_cpu trace file, don't bother by iterating over
1562 * all cpu and peek directly.
1564 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1565 if (ring_buffer_empty_cpu(buffer, cpu_file))
1567 ent = peek_next_entry(iter, cpu_file, ent_ts);
1569 *ent_cpu = cpu_file;
1574 for_each_tracing_cpu(cpu) {
1576 if (ring_buffer_empty_cpu(buffer, cpu))
1579 ent = peek_next_entry(iter, cpu, &ts);
1582 * Pick the entry with the smallest timestamp:
1584 if (ent && (!next || ts < next_ts)) {
1592 *ent_cpu = next_cpu;
1600 /* Find the next real entry, without updating the iterator itself */
1601 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1602 int *ent_cpu, u64 *ent_ts)
1604 return __find_next_entry(iter, ent_cpu, ent_ts);
1607 /* Find the next real entry, and increment the iterator to the next entry */
1608 static void *find_next_entry_inc(struct trace_iterator *iter)
1610 iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1613 trace_iterator_increment(iter);
1615 return iter->ent ? iter : NULL;
1618 static void trace_consume(struct trace_iterator *iter)
1620 /* Don't allow ftrace to trace into the ring buffers */
1621 ftrace_disable_cpu();
1622 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1623 ftrace_enable_cpu();
1626 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1628 struct trace_iterator *iter = m->private;
1632 WARN_ON_ONCE(iter->leftover);
1636 /* can't go backwards */
1641 ent = find_next_entry_inc(iter);
1645 while (ent && iter->idx < i)
1646 ent = find_next_entry_inc(iter);
1653 static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1655 struct trace_array *tr = iter->tr;
1656 struct ring_buffer_event *event;
1657 struct ring_buffer_iter *buf_iter;
1658 unsigned long entries = 0;
1661 tr->data[cpu]->skipped_entries = 0;
1663 if (!iter->buffer_iter[cpu])
1666 buf_iter = iter->buffer_iter[cpu];
1667 ring_buffer_iter_reset(buf_iter);
1670 * We could have the case with the max latency tracers
1671 * that a reset never took place on a cpu. This is evident
1672 * by the timestamp being before the start of the buffer.
1674 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1675 if (ts >= iter->tr->time_start)
1678 ring_buffer_read(buf_iter, NULL);
1681 tr->data[cpu]->skipped_entries = entries;
1685 * The current tracer is copied to avoid a global locking
1688 static void *s_start(struct seq_file *m, loff_t *pos)
1690 struct trace_iterator *iter = m->private;
1691 static struct tracer *old_tracer;
1692 int cpu_file = iter->cpu_file;
1697 /* copy the tracer to avoid using a global lock all around */
1698 mutex_lock(&trace_types_lock);
1699 if (unlikely(old_tracer != current_trace && current_trace)) {
1700 old_tracer = current_trace;
1701 *iter->trace = *current_trace;
1703 mutex_unlock(&trace_types_lock);
1705 atomic_inc(&trace_record_cmdline_disabled);
1707 if (*pos != iter->pos) {
1712 ftrace_disable_cpu();
1714 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1715 for_each_tracing_cpu(cpu)
1716 tracing_iter_reset(iter, cpu);
1718 tracing_iter_reset(iter, cpu_file);
1720 ftrace_enable_cpu();
1722 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1727 * If we overflowed the seq_file before, then we want
1728 * to just reuse the trace_seq buffer again.
1734 p = s_next(m, p, &l);
1738 trace_event_read_lock();
1739 trace_access_lock(cpu_file);
1743 static void s_stop(struct seq_file *m, void *p)
1745 struct trace_iterator *iter = m->private;
1747 atomic_dec(&trace_record_cmdline_disabled);
1748 trace_access_unlock(iter->cpu_file);
1749 trace_event_read_unlock();
1752 static void print_lat_help_header(struct seq_file *m)
1754 seq_puts(m, "# _------=> CPU# \n");
1755 seq_puts(m, "# / _-----=> irqs-off \n");
1756 seq_puts(m, "# | / _----=> need-resched \n");
1757 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1758 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1759 seq_puts(m, "# |||| /_--=> lock-depth \n");
1760 seq_puts(m, "# |||||/ delay \n");
1761 seq_puts(m, "# cmd pid |||||| time | caller \n");
1762 seq_puts(m, "# \\ / |||||| \\ | / \n");
1765 static void print_func_help_header(struct seq_file *m)
1767 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1768 seq_puts(m, "# | | | | |\n");
1773 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1775 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1776 struct trace_array *tr = iter->tr;
1777 struct trace_array_cpu *data = tr->data[tr->cpu];
1778 struct tracer *type = current_trace;
1779 unsigned long entries = 0;
1780 unsigned long total = 0;
1781 unsigned long count;
1782 const char *name = "preemption";
1789 for_each_tracing_cpu(cpu) {
1790 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1792 * If this buffer has skipped entries, then we hold all
1793 * entries for the trace and we need to ignore the
1794 * ones before the time stamp.
1796 if (tr->data[cpu]->skipped_entries) {
1797 count -= tr->data[cpu]->skipped_entries;
1798 /* total is the same as the entries */
1802 ring_buffer_overrun_cpu(tr->buffer, cpu);
1806 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1808 seq_puts(m, "# -----------------------------------"
1809 "---------------------------------\n");
1810 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1811 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1812 nsecs_to_usecs(data->saved_latency),
1816 #if defined(CONFIG_PREEMPT_NONE)
1818 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1820 #elif defined(CONFIG_PREEMPT)
1825 /* These are reserved for later use */
1828 seq_printf(m, " #P:%d)\n", num_online_cpus());
1832 seq_puts(m, "# -----------------\n");
1833 seq_printf(m, "# | task: %.16s-%d "
1834 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1835 data->comm, data->pid, data->uid, data->nice,
1836 data->policy, data->rt_priority);
1837 seq_puts(m, "# -----------------\n");
1839 if (data->critical_start) {
1840 seq_puts(m, "# => started at: ");
1841 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1842 trace_print_seq(m, &iter->seq);
1843 seq_puts(m, "\n# => ended at: ");
1844 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1845 trace_print_seq(m, &iter->seq);
1846 seq_puts(m, "\n#\n");
1852 static void test_cpu_buff_start(struct trace_iterator *iter)
1854 struct trace_seq *s = &iter->seq;
1856 if (!(trace_flags & TRACE_ITER_ANNOTATE))
1859 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1862 if (cpumask_test_cpu(iter->cpu, iter->started))
1865 if (iter->tr->data[iter->cpu]->skipped_entries)
1868 cpumask_set_cpu(iter->cpu, iter->started);
1870 /* Don't print started cpu buffer for the first entry of the trace */
1872 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1876 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1878 struct trace_seq *s = &iter->seq;
1879 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1880 struct trace_entry *entry;
1881 struct trace_event *event;
1885 test_cpu_buff_start(iter);
1887 event = ftrace_find_event(entry->type);
1889 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1890 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1891 if (!trace_print_lat_context(iter))
1894 if (!trace_print_context(iter))
1900 return event->trace(iter, sym_flags);
1902 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1905 return TRACE_TYPE_HANDLED;
1907 return TRACE_TYPE_PARTIAL_LINE;
1910 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1912 struct trace_seq *s = &iter->seq;
1913 struct trace_entry *entry;
1914 struct trace_event *event;
1918 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1919 if (!trace_seq_printf(s, "%d %d %llu ",
1920 entry->pid, iter->cpu, iter->ts))
1924 event = ftrace_find_event(entry->type);
1926 return event->raw(iter, 0);
1928 if (!trace_seq_printf(s, "%d ?\n", entry->type))
1931 return TRACE_TYPE_HANDLED;
1933 return TRACE_TYPE_PARTIAL_LINE;
1936 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1938 struct trace_seq *s = &iter->seq;
1939 unsigned char newline = '\n';
1940 struct trace_entry *entry;
1941 struct trace_event *event;
1945 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1946 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1947 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1948 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1951 event = ftrace_find_event(entry->type);
1953 enum print_line_t ret = event->hex(iter, 0);
1954 if (ret != TRACE_TYPE_HANDLED)
1958 SEQ_PUT_FIELD_RET(s, newline);
1960 return TRACE_TYPE_HANDLED;
1963 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1965 struct trace_seq *s = &iter->seq;
1966 struct trace_entry *entry;
1967 struct trace_event *event;
1971 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1972 SEQ_PUT_FIELD_RET(s, entry->pid);
1973 SEQ_PUT_FIELD_RET(s, iter->cpu);
1974 SEQ_PUT_FIELD_RET(s, iter->ts);
1977 event = ftrace_find_event(entry->type);
1978 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1981 static int trace_empty(struct trace_iterator *iter)
1985 /* If we are looking at one CPU buffer, only check that one */
1986 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1987 cpu = iter->cpu_file;
1988 if (iter->buffer_iter[cpu]) {
1989 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1992 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1998 for_each_tracing_cpu(cpu) {
1999 if (iter->buffer_iter[cpu]) {
2000 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2003 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2011 /* Called with trace_event_read_lock() held. */
2012 static enum print_line_t print_trace_line(struct trace_iterator *iter)
2014 enum print_line_t ret;
2016 if (iter->trace && iter->trace->print_line) {
2017 ret = iter->trace->print_line(iter);
2018 if (ret != TRACE_TYPE_UNHANDLED)
2022 if (iter->ent->type == TRACE_BPRINT &&
2023 trace_flags & TRACE_ITER_PRINTK &&
2024 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2025 return trace_print_bprintk_msg_only(iter);
2027 if (iter->ent->type == TRACE_PRINT &&
2028 trace_flags & TRACE_ITER_PRINTK &&
2029 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2030 return trace_print_printk_msg_only(iter);
2032 if (trace_flags & TRACE_ITER_BIN)
2033 return print_bin_fmt(iter);
2035 if (trace_flags & TRACE_ITER_HEX)
2036 return print_hex_fmt(iter);
2038 if (trace_flags & TRACE_ITER_RAW)
2039 return print_raw_fmt(iter);
2041 return print_trace_fmt(iter);
2044 static int s_show(struct seq_file *m, void *v)
2046 struct trace_iterator *iter = v;
2049 if (iter->ent == NULL) {
2051 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2054 if (iter->trace && iter->trace->print_header)
2055 iter->trace->print_header(m);
2056 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2057 /* print nothing if the buffers are empty */
2058 if (trace_empty(iter))
2060 print_trace_header(m, iter);
2061 if (!(trace_flags & TRACE_ITER_VERBOSE))
2062 print_lat_help_header(m);
2064 if (!(trace_flags & TRACE_ITER_VERBOSE))
2065 print_func_help_header(m);
2067 } else if (iter->leftover) {
2069 * If we filled the seq_file buffer earlier, we
2070 * want to just show it now.
2072 ret = trace_print_seq(m, &iter->seq);
2074 /* ret should this time be zero, but you never know */
2075 iter->leftover = ret;
2078 print_trace_line(iter);
2079 ret = trace_print_seq(m, &iter->seq);
2081 * If we overflow the seq_file buffer, then it will
2082 * ask us for this data again at start up.
2084 * ret is 0 if seq_file write succeeded.
2087 iter->leftover = ret;
2093 static const struct seq_operations tracer_seq_ops = {
2100 static struct trace_iterator *
2101 __tracing_open(struct inode *inode, struct file *file)
2103 long cpu_file = (long) inode->i_private;
2104 void *fail_ret = ERR_PTR(-ENOMEM);
2105 struct trace_iterator *iter;
2109 if (tracing_disabled)
2110 return ERR_PTR(-ENODEV);
2112 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2114 return ERR_PTR(-ENOMEM);
2117 * We make a copy of the current tracer to avoid concurrent
2118 * changes on it while we are reading.
2120 mutex_lock(&trace_types_lock);
2121 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2126 *iter->trace = *current_trace;
2128 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2131 if (current_trace && current_trace->print_max)
2134 iter->tr = &global_trace;
2136 mutex_init(&iter->mutex);
2137 iter->cpu_file = cpu_file;
2139 /* Notify the tracer early; before we stop tracing. */
2140 if (iter->trace && iter->trace->open)
2141 iter->trace->open(iter);
2143 /* Annotate start of buffers if we had overruns */
2144 if (ring_buffer_overruns(iter->tr->buffer))
2145 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2147 /* stop the trace while dumping */
2150 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2151 for_each_tracing_cpu(cpu) {
2153 iter->buffer_iter[cpu] =
2154 ring_buffer_read_start(iter->tr->buffer, cpu);
2155 tracing_iter_reset(iter, cpu);
2158 cpu = iter->cpu_file;
2159 iter->buffer_iter[cpu] =
2160 ring_buffer_read_start(iter->tr->buffer, cpu);
2161 tracing_iter_reset(iter, cpu);
2164 ret = seq_open(file, &tracer_seq_ops);
2166 fail_ret = ERR_PTR(ret);
2170 m = file->private_data;
2173 mutex_unlock(&trace_types_lock);
2178 for_each_tracing_cpu(cpu) {
2179 if (iter->buffer_iter[cpu])
2180 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2182 free_cpumask_var(iter->started);
2185 mutex_unlock(&trace_types_lock);
2192 int tracing_open_generic(struct inode *inode, struct file *filp)
2194 if (tracing_disabled)
2197 filp->private_data = inode->i_private;
2201 static int tracing_release(struct inode *inode, struct file *file)
2203 struct seq_file *m = (struct seq_file *)file->private_data;
2204 struct trace_iterator *iter;
2207 if (!(file->f_mode & FMODE_READ))
2212 mutex_lock(&trace_types_lock);
2213 for_each_tracing_cpu(cpu) {
2214 if (iter->buffer_iter[cpu])
2215 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2218 if (iter->trace && iter->trace->close)
2219 iter->trace->close(iter);
2221 /* reenable tracing if it was previously enabled */
2223 mutex_unlock(&trace_types_lock);
2225 seq_release(inode, file);
2226 mutex_destroy(&iter->mutex);
2227 free_cpumask_var(iter->started);
2233 static int tracing_open(struct inode *inode, struct file *file)
2235 struct trace_iterator *iter;
2238 /* If this file was open for write, then erase contents */
2239 if ((file->f_mode & FMODE_WRITE) &&
2240 (file->f_flags & O_TRUNC)) {
2241 long cpu = (long) inode->i_private;
2243 if (cpu == TRACE_PIPE_ALL_CPU)
2244 tracing_reset_online_cpus(&global_trace);
2246 tracing_reset(&global_trace, cpu);
2249 if (file->f_mode & FMODE_READ) {
2250 iter = __tracing_open(inode, file);
2252 ret = PTR_ERR(iter);
2253 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2254 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2260 t_next(struct seq_file *m, void *v, loff_t *pos)
2262 struct tracer *t = v;
2272 static void *t_start(struct seq_file *m, loff_t *pos)
2277 mutex_lock(&trace_types_lock);
2278 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2284 static void t_stop(struct seq_file *m, void *p)
2286 mutex_unlock(&trace_types_lock);
2289 static int t_show(struct seq_file *m, void *v)
2291 struct tracer *t = v;
2296 seq_printf(m, "%s", t->name);
2305 static const struct seq_operations show_traces_seq_ops = {
2312 static int show_traces_open(struct inode *inode, struct file *file)
2314 if (tracing_disabled)
2317 return seq_open(file, &show_traces_seq_ops);
2321 tracing_write_stub(struct file *filp, const char __user *ubuf,
2322 size_t count, loff_t *ppos)
2327 static const struct file_operations tracing_fops = {
2328 .open = tracing_open,
2330 .write = tracing_write_stub,
2331 .llseek = seq_lseek,
2332 .release = tracing_release,
2335 static const struct file_operations show_traces_fops = {
2336 .open = show_traces_open,
2338 .release = seq_release,
2342 * Only trace on a CPU if the bitmask is set:
2344 static cpumask_var_t tracing_cpumask;
2347 * The tracer itself will not take this lock, but still we want
2348 * to provide a consistent cpumask to user-space:
2350 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2353 * Temporary storage for the character representation of the
2354 * CPU bitmask (and one more byte for the newline):
2356 static char mask_str[NR_CPUS + 1];
2359 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2360 size_t count, loff_t *ppos)
2364 mutex_lock(&tracing_cpumask_update_lock);
2366 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2367 if (count - len < 2) {
2371 len += sprintf(mask_str + len, "\n");
2372 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2375 mutex_unlock(&tracing_cpumask_update_lock);
2381 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2382 size_t count, loff_t *ppos)
2385 cpumask_var_t tracing_cpumask_new;
2387 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2390 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2394 mutex_lock(&tracing_cpumask_update_lock);
2396 local_irq_disable();
2397 arch_spin_lock(&ftrace_max_lock);
2398 for_each_tracing_cpu(cpu) {
2400 * Increase/decrease the disabled counter if we are
2401 * about to flip a bit in the cpumask:
2403 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2404 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2405 atomic_inc(&global_trace.data[cpu]->disabled);
2407 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2408 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2409 atomic_dec(&global_trace.data[cpu]->disabled);
2412 arch_spin_unlock(&ftrace_max_lock);
2415 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2417 mutex_unlock(&tracing_cpumask_update_lock);
2418 free_cpumask_var(tracing_cpumask_new);
2423 free_cpumask_var(tracing_cpumask_new);
2428 static const struct file_operations tracing_cpumask_fops = {
2429 .open = tracing_open_generic,
2430 .read = tracing_cpumask_read,
2431 .write = tracing_cpumask_write,
2434 static int tracing_trace_options_show(struct seq_file *m, void *v)
2436 struct tracer_opt *trace_opts;
2440 mutex_lock(&trace_types_lock);
2441 tracer_flags = current_trace->flags->val;
2442 trace_opts = current_trace->flags->opts;
2444 for (i = 0; trace_options[i]; i++) {
2445 if (trace_flags & (1 << i))
2446 seq_printf(m, "%s\n", trace_options[i]);
2448 seq_printf(m, "no%s\n", trace_options[i]);
2451 for (i = 0; trace_opts[i].name; i++) {
2452 if (tracer_flags & trace_opts[i].bit)
2453 seq_printf(m, "%s\n", trace_opts[i].name);
2455 seq_printf(m, "no%s\n", trace_opts[i].name);
2457 mutex_unlock(&trace_types_lock);
2462 static int __set_tracer_option(struct tracer *trace,
2463 struct tracer_flags *tracer_flags,
2464 struct tracer_opt *opts, int neg)
2468 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2473 tracer_flags->val &= ~opts->bit;
2475 tracer_flags->val |= opts->bit;
2479 /* Try to assign a tracer specific option */
2480 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2482 struct tracer_flags *tracer_flags = trace->flags;
2483 struct tracer_opt *opts = NULL;
2486 for (i = 0; tracer_flags->opts[i].name; i++) {
2487 opts = &tracer_flags->opts[i];
2489 if (strcmp(cmp, opts->name) == 0)
2490 return __set_tracer_option(trace, trace->flags,
2497 static void set_tracer_flags(unsigned int mask, int enabled)
2499 /* do nothing if flag is already set */
2500 if (!!(trace_flags & mask) == !!enabled)
2504 trace_flags |= mask;
2506 trace_flags &= ~mask;
2510 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2511 size_t cnt, loff_t *ppos)
2519 if (cnt >= sizeof(buf))
2522 if (copy_from_user(&buf, ubuf, cnt))
2526 cmp = strstrip(buf);
2528 if (strncmp(cmp, "no", 2) == 0) {
2533 for (i = 0; trace_options[i]; i++) {
2534 if (strcmp(cmp, trace_options[i]) == 0) {
2535 set_tracer_flags(1 << i, !neg);
2540 /* If no option could be set, test the specific tracer options */
2541 if (!trace_options[i]) {
2542 mutex_lock(&trace_types_lock);
2543 ret = set_tracer_option(current_trace, cmp, neg);
2544 mutex_unlock(&trace_types_lock);
2554 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2556 if (tracing_disabled)
2558 return single_open(file, tracing_trace_options_show, NULL);
2561 static const struct file_operations tracing_iter_fops = {
2562 .open = tracing_trace_options_open,
2564 .llseek = seq_lseek,
2565 .release = single_release,
2566 .write = tracing_trace_options_write,
2569 static const char readme_msg[] =
2570 "tracing mini-HOWTO:\n\n"
2571 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2572 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2573 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2574 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2576 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2577 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2579 "# cat /sys/kernel/debug/tracing/trace_options\n"
2580 "noprint-parent nosym-offset nosym-addr noverbose\n"
2581 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2582 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2583 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2584 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2588 tracing_readme_read(struct file *filp, char __user *ubuf,
2589 size_t cnt, loff_t *ppos)
2591 return simple_read_from_buffer(ubuf, cnt, ppos,
2592 readme_msg, strlen(readme_msg));
2595 static const struct file_operations tracing_readme_fops = {
2596 .open = tracing_open_generic,
2597 .read = tracing_readme_read,
2601 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2602 size_t cnt, loff_t *ppos)
2611 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2615 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2623 for (i = 0; i < SAVED_CMDLINES; i++) {
2626 pid = map_cmdline_to_pid[i];
2627 if (pid == -1 || pid == NO_CMDLINE_MAP)
2630 trace_find_cmdline(pid, buf_comm);
2631 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2636 len = simple_read_from_buffer(ubuf, cnt, ppos,
2645 static const struct file_operations tracing_saved_cmdlines_fops = {
2646 .open = tracing_open_generic,
2647 .read = tracing_saved_cmdlines_read,
2651 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2652 size_t cnt, loff_t *ppos)
2657 r = sprintf(buf, "%u\n", tracer_enabled);
2658 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2662 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2663 size_t cnt, loff_t *ppos)
2665 struct trace_array *tr = filp->private_data;
2670 if (cnt >= sizeof(buf))
2673 if (copy_from_user(&buf, ubuf, cnt))
2678 ret = strict_strtoul(buf, 10, &val);
2684 mutex_lock(&trace_types_lock);
2685 if (tracer_enabled ^ val) {
2688 if (current_trace->start)
2689 current_trace->start(tr);
2694 if (current_trace->stop)
2695 current_trace->stop(tr);
2698 mutex_unlock(&trace_types_lock);
2706 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2707 size_t cnt, loff_t *ppos)
2709 char buf[MAX_TRACER_SIZE+2];
2712 mutex_lock(&trace_types_lock);
2714 r = sprintf(buf, "%s\n", current_trace->name);
2716 r = sprintf(buf, "\n");
2717 mutex_unlock(&trace_types_lock);
2719 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2722 int tracer_init(struct tracer *t, struct trace_array *tr)
2724 tracing_reset_online_cpus(tr);
2728 static int tracing_resize_ring_buffer(unsigned long size)
2733 * If kernel or user changes the size of the ring buffer
2734 * we use the size that was given, and we can forget about
2735 * expanding it later.
2737 ring_buffer_expanded = 1;
2739 ret = ring_buffer_resize(global_trace.buffer, size);
2743 ret = ring_buffer_resize(max_tr.buffer, size);
2747 r = ring_buffer_resize(global_trace.buffer,
2748 global_trace.entries);
2751 * AARGH! We are left with different
2752 * size max buffer!!!!
2753 * The max buffer is our "snapshot" buffer.
2754 * When a tracer needs a snapshot (one of the
2755 * latency tracers), it swaps the max buffer
2756 * with the saved snap shot. We succeeded to
2757 * update the size of the main buffer, but failed to
2758 * update the size of the max buffer. But when we tried
2759 * to reset the main buffer to the original size, we
2760 * failed there too. This is very unlikely to
2761 * happen, but if it does, warn and kill all
2765 tracing_disabled = 1;
2770 global_trace.entries = size;
2776 * tracing_update_buffers - used by tracing facility to expand ring buffers
2778 * To save on memory when the tracing is never used on a system with it
2779 * configured in. The ring buffers are set to a minimum size. But once
2780 * a user starts to use the tracing facility, then they need to grow
2781 * to their default size.
2783 * This function is to be called when a tracer is about to be used.
2785 int tracing_update_buffers(void)
2789 mutex_lock(&trace_types_lock);
2790 if (!ring_buffer_expanded)
2791 ret = tracing_resize_ring_buffer(trace_buf_size);
2792 mutex_unlock(&trace_types_lock);
2797 struct trace_option_dentry;
2799 static struct trace_option_dentry *
2800 create_trace_option_files(struct tracer *tracer);
2803 destroy_trace_option_files(struct trace_option_dentry *topts);
2805 static int tracing_set_tracer(const char *buf)
2807 static struct trace_option_dentry *topts;
2808 struct trace_array *tr = &global_trace;
2812 mutex_lock(&trace_types_lock);
2814 if (!ring_buffer_expanded) {
2815 ret = tracing_resize_ring_buffer(trace_buf_size);
2821 for (t = trace_types; t; t = t->next) {
2822 if (strcmp(t->name, buf) == 0)
2829 if (t == current_trace)
2832 trace_branch_disable();
2833 if (current_trace && current_trace->reset)
2834 current_trace->reset(tr);
2836 destroy_trace_option_files(topts);
2840 topts = create_trace_option_files(current_trace);
2843 ret = tracer_init(t, tr);
2848 trace_branch_enable(tr);
2850 mutex_unlock(&trace_types_lock);
2856 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2857 size_t cnt, loff_t *ppos)
2859 char buf[MAX_TRACER_SIZE+1];
2866 if (cnt > MAX_TRACER_SIZE)
2867 cnt = MAX_TRACER_SIZE;
2869 if (copy_from_user(&buf, ubuf, cnt))
2874 /* strip ending whitespace. */
2875 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2878 err = tracing_set_tracer(buf);
2888 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2889 size_t cnt, loff_t *ppos)
2891 unsigned long *ptr = filp->private_data;
2895 r = snprintf(buf, sizeof(buf), "%ld\n",
2896 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2897 if (r > sizeof(buf))
2899 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2903 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2904 size_t cnt, loff_t *ppos)
2906 unsigned long *ptr = filp->private_data;
2911 if (cnt >= sizeof(buf))
2914 if (copy_from_user(&buf, ubuf, cnt))
2919 ret = strict_strtoul(buf, 10, &val);
2928 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2930 long cpu_file = (long) inode->i_private;
2931 struct trace_iterator *iter;
2934 if (tracing_disabled)
2937 mutex_lock(&trace_types_lock);
2939 /* create a buffer to store the information to pass to userspace */
2940 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2947 * We make a copy of the current tracer to avoid concurrent
2948 * changes on it while we are reading.
2950 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2956 *iter->trace = *current_trace;
2958 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2963 /* trace pipe does not show start of buffer */
2964 cpumask_setall(iter->started);
2966 if (trace_flags & TRACE_ITER_LATENCY_FMT)
2967 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2969 iter->cpu_file = cpu_file;
2970 iter->tr = &global_trace;
2971 mutex_init(&iter->mutex);
2972 filp->private_data = iter;
2974 if (iter->trace->pipe_open)
2975 iter->trace->pipe_open(iter);
2978 mutex_unlock(&trace_types_lock);
2984 mutex_unlock(&trace_types_lock);
2988 static int tracing_release_pipe(struct inode *inode, struct file *file)
2990 struct trace_iterator *iter = file->private_data;
2992 mutex_lock(&trace_types_lock);
2994 if (iter->trace->pipe_close)
2995 iter->trace->pipe_close(iter);
2997 mutex_unlock(&trace_types_lock);
2999 free_cpumask_var(iter->started);
3000 mutex_destroy(&iter->mutex);
3008 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3010 struct trace_iterator *iter = filp->private_data;
3012 if (trace_flags & TRACE_ITER_BLOCK) {
3014 * Always select as readable when in blocking mode
3016 return POLLIN | POLLRDNORM;
3018 if (!trace_empty(iter))
3019 return POLLIN | POLLRDNORM;
3020 poll_wait(filp, &trace_wait, poll_table);
3021 if (!trace_empty(iter))
3022 return POLLIN | POLLRDNORM;
3029 void default_wait_pipe(struct trace_iterator *iter)
3033 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3035 if (trace_empty(iter))
3038 finish_wait(&trace_wait, &wait);
3042 * This is a make-shift waitqueue.
3043 * A tracer might use this callback on some rare cases:
3045 * 1) the current tracer might hold the runqueue lock when it wakes up
3046 * a reader, hence a deadlock (sched, function, and function graph tracers)
3047 * 2) the function tracers, trace all functions, we don't want
3048 * the overhead of calling wake_up and friends
3049 * (and tracing them too)
3051 * Anyway, this is really very primitive wakeup.
3053 void poll_wait_pipe(struct trace_iterator *iter)
3055 set_current_state(TASK_INTERRUPTIBLE);
3056 /* sleep for 100 msecs, and try again. */
3057 schedule_timeout(HZ / 10);
3060 /* Must be called with trace_types_lock mutex held. */
3061 static int tracing_wait_pipe(struct file *filp)
3063 struct trace_iterator *iter = filp->private_data;
3065 while (trace_empty(iter)) {
3067 if ((filp->f_flags & O_NONBLOCK)) {
3071 mutex_unlock(&iter->mutex);
3073 iter->trace->wait_pipe(iter);
3075 mutex_lock(&iter->mutex);
3077 if (signal_pending(current))
3081 * We block until we read something and tracing is disabled.
3082 * We still block if tracing is disabled, but we have never
3083 * read anything. This allows a user to cat this file, and
3084 * then enable tracing. But after we have read something,
3085 * we give an EOF when tracing is again disabled.
3087 * iter->pos will be 0 if we haven't read anything.
3089 if (!tracer_enabled && iter->pos)
3100 tracing_read_pipe(struct file *filp, char __user *ubuf,
3101 size_t cnt, loff_t *ppos)
3103 struct trace_iterator *iter = filp->private_data;
3104 static struct tracer *old_tracer;
3107 /* return any leftover data */
3108 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3112 trace_seq_init(&iter->seq);
3114 /* copy the tracer to avoid using a global lock all around */
3115 mutex_lock(&trace_types_lock);
3116 if (unlikely(old_tracer != current_trace && current_trace)) {
3117 old_tracer = current_trace;
3118 *iter->trace = *current_trace;
3120 mutex_unlock(&trace_types_lock);
3123 * Avoid more than one consumer on a single file descriptor
3124 * This is just a matter of traces coherency, the ring buffer itself
3127 mutex_lock(&iter->mutex);
3128 if (iter->trace->read) {
3129 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3135 sret = tracing_wait_pipe(filp);
3139 /* stop when tracing is finished */
3140 if (trace_empty(iter)) {
3145 if (cnt >= PAGE_SIZE)
3146 cnt = PAGE_SIZE - 1;
3148 /* reset all but tr, trace, and overruns */
3149 memset(&iter->seq, 0,
3150 sizeof(struct trace_iterator) -
3151 offsetof(struct trace_iterator, seq));
3154 trace_event_read_lock();
3155 trace_access_lock(iter->cpu_file);
3156 while (find_next_entry_inc(iter) != NULL) {
3157 enum print_line_t ret;
3158 int len = iter->seq.len;
3160 ret = print_trace_line(iter);
3161 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3162 /* don't print partial lines */
3163 iter->seq.len = len;
3166 if (ret != TRACE_TYPE_NO_CONSUME)
3167 trace_consume(iter);
3169 if (iter->seq.len >= cnt)
3172 trace_access_unlock(iter->cpu_file);
3173 trace_event_read_unlock();
3175 /* Now copy what we have to the user */
3176 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3177 if (iter->seq.readpos >= iter->seq.len)
3178 trace_seq_init(&iter->seq);
3181 * If there was nothing to send to user, inspite of consuming trace
3182 * entries, go back to wait for more entries.
3188 mutex_unlock(&iter->mutex);
3193 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3194 struct pipe_buffer *buf)
3196 __free_page(buf->page);
3199 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3202 __free_page(spd->pages[idx]);
3205 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3207 .map = generic_pipe_buf_map,
3208 .unmap = generic_pipe_buf_unmap,
3209 .confirm = generic_pipe_buf_confirm,
3210 .release = tracing_pipe_buf_release,
3211 .steal = generic_pipe_buf_steal,
3212 .get = generic_pipe_buf_get,
3216 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3221 /* Seq buffer is page-sized, exactly what we need. */
3223 count = iter->seq.len;
3224 ret = print_trace_line(iter);
3225 count = iter->seq.len - count;
3228 iter->seq.len -= count;
3231 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3232 iter->seq.len -= count;
3236 if (ret != TRACE_TYPE_NO_CONSUME)
3237 trace_consume(iter);
3239 if (!find_next_entry_inc(iter)) {
3249 static ssize_t tracing_splice_read_pipe(struct file *filp,
3251 struct pipe_inode_info *pipe,
3255 struct page *pages[PIPE_BUFFERS];
3256 struct partial_page partial[PIPE_BUFFERS];
3257 struct trace_iterator *iter = filp->private_data;
3258 struct splice_pipe_desc spd = {
3261 .nr_pages = 0, /* This gets updated below. */
3263 .ops = &tracing_pipe_buf_ops,
3264 .spd_release = tracing_spd_release_pipe,
3266 static struct tracer *old_tracer;
3271 /* copy the tracer to avoid using a global lock all around */
3272 mutex_lock(&trace_types_lock);
3273 if (unlikely(old_tracer != current_trace && current_trace)) {
3274 old_tracer = current_trace;
3275 *iter->trace = *current_trace;
3277 mutex_unlock(&trace_types_lock);
3279 mutex_lock(&iter->mutex);
3281 if (iter->trace->splice_read) {
3282 ret = iter->trace->splice_read(iter, filp,
3283 ppos, pipe, len, flags);
3288 ret = tracing_wait_pipe(filp);
3292 if (!iter->ent && !find_next_entry_inc(iter)) {
3297 trace_event_read_lock();
3298 trace_access_lock(iter->cpu_file);
3300 /* Fill as many pages as possible. */
3301 for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3302 pages[i] = alloc_page(GFP_KERNEL);
3306 rem = tracing_fill_pipe_page(rem, iter);
3308 /* Copy the data into the page, so we can start over. */
3309 ret = trace_seq_to_buffer(&iter->seq,
3310 page_address(pages[i]),
3313 __free_page(pages[i]);
3316 partial[i].offset = 0;
3317 partial[i].len = iter->seq.len;
3319 trace_seq_init(&iter->seq);
3322 trace_access_unlock(iter->cpu_file);
3323 trace_event_read_unlock();
3324 mutex_unlock(&iter->mutex);
3328 return splice_to_pipe(pipe, &spd);
3331 mutex_unlock(&iter->mutex);
3337 tracing_entries_read(struct file *filp, char __user *ubuf,
3338 size_t cnt, loff_t *ppos)
3340 struct trace_array *tr = filp->private_data;
3344 mutex_lock(&trace_types_lock);
3345 if (!ring_buffer_expanded)
3346 r = sprintf(buf, "%lu (expanded: %lu)\n",
3348 trace_buf_size >> 10);
3350 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3351 mutex_unlock(&trace_types_lock);
3353 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3357 tracing_entries_write(struct file *filp, const char __user *ubuf,
3358 size_t cnt, loff_t *ppos)
3364 if (cnt >= sizeof(buf))
3367 if (copy_from_user(&buf, ubuf, cnt))
3372 ret = strict_strtoul(buf, 10, &val);
3376 /* must have at least 1 entry */
3380 mutex_lock(&trace_types_lock);
3384 /* disable all cpu buffers */
3385 for_each_tracing_cpu(cpu) {
3386 if (global_trace.data[cpu])
3387 atomic_inc(&global_trace.data[cpu]->disabled);
3388 if (max_tr.data[cpu])
3389 atomic_inc(&max_tr.data[cpu]->disabled);
3392 /* value is in KB */
3395 if (val != global_trace.entries) {
3396 ret = tracing_resize_ring_buffer(val);
3405 /* If check pages failed, return ENOMEM */
3406 if (tracing_disabled)
3409 for_each_tracing_cpu(cpu) {
3410 if (global_trace.data[cpu])
3411 atomic_dec(&global_trace.data[cpu]->disabled);
3412 if (max_tr.data[cpu])
3413 atomic_dec(&max_tr.data[cpu]->disabled);
3417 max_tr.entries = global_trace.entries;
3418 mutex_unlock(&trace_types_lock);
3423 static int mark_printk(const char *fmt, ...)
3427 va_start(args, fmt);
3428 ret = trace_vprintk(0, fmt, args);
3434 tracing_mark_write(struct file *filp, const char __user *ubuf,
3435 size_t cnt, loff_t *fpos)
3439 if (tracing_disabled)
3442 if (cnt > TRACE_BUF_SIZE)
3443 cnt = TRACE_BUF_SIZE;
3445 buf = kmalloc(cnt + 2, GFP_KERNEL);
3449 if (copy_from_user(buf, ubuf, cnt)) {
3453 if (buf[cnt-1] != '\n') {
3459 cnt = mark_printk("%s", buf);
3466 static int tracing_clock_show(struct seq_file *m, void *v)
3470 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3472 "%s%s%s%s", i ? " " : "",
3473 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3474 i == trace_clock_id ? "]" : "");
3480 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3481 size_t cnt, loff_t *fpos)
3484 const char *clockstr;
3487 if (cnt >= sizeof(buf))
3490 if (copy_from_user(&buf, ubuf, cnt))
3495 clockstr = strstrip(buf);
3497 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3498 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3501 if (i == ARRAY_SIZE(trace_clocks))
3506 mutex_lock(&trace_types_lock);
3508 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3510 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3512 mutex_unlock(&trace_types_lock);
3519 static int tracing_clock_open(struct inode *inode, struct file *file)
3521 if (tracing_disabled)
3523 return single_open(file, tracing_clock_show, NULL);
3526 static const struct file_operations tracing_max_lat_fops = {
3527 .open = tracing_open_generic,
3528 .read = tracing_max_lat_read,
3529 .write = tracing_max_lat_write,
3532 static const struct file_operations tracing_ctrl_fops = {
3533 .open = tracing_open_generic,
3534 .read = tracing_ctrl_read,
3535 .write = tracing_ctrl_write,
3538 static const struct file_operations set_tracer_fops = {
3539 .open = tracing_open_generic,
3540 .read = tracing_set_trace_read,
3541 .write = tracing_set_trace_write,
3544 static const struct file_operations tracing_pipe_fops = {
3545 .open = tracing_open_pipe,
3546 .poll = tracing_poll_pipe,
3547 .read = tracing_read_pipe,
3548 .splice_read = tracing_splice_read_pipe,
3549 .release = tracing_release_pipe,
3552 static const struct file_operations tracing_entries_fops = {
3553 .open = tracing_open_generic,
3554 .read = tracing_entries_read,
3555 .write = tracing_entries_write,
3558 static const struct file_operations tracing_mark_fops = {
3559 .open = tracing_open_generic,
3560 .write = tracing_mark_write,
3563 static const struct file_operations trace_clock_fops = {
3564 .open = tracing_clock_open,
3566 .llseek = seq_lseek,
3567 .release = single_release,
3568 .write = tracing_clock_write,
3571 struct ftrace_buffer_info {
3572 struct trace_array *tr;
3578 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3580 int cpu = (int)(long)inode->i_private;
3581 struct ftrace_buffer_info *info;
3583 if (tracing_disabled)
3586 info = kzalloc(sizeof(*info), GFP_KERNEL);
3590 info->tr = &global_trace;
3593 /* Force reading ring buffer for first read */
3594 info->read = (unsigned int)-1;
3596 filp->private_data = info;
3598 return nonseekable_open(inode, filp);
3602 tracing_buffers_read(struct file *filp, char __user *ubuf,
3603 size_t count, loff_t *ppos)
3605 struct ftrace_buffer_info *info = filp->private_data;
3614 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3618 /* Do we have previous read data to read? */
3619 if (info->read < PAGE_SIZE)
3624 trace_access_lock(info->cpu);
3625 ret = ring_buffer_read_page(info->tr->buffer,
3629 trace_access_unlock(info->cpu);
3633 pos = ring_buffer_page_len(info->spare);
3635 if (pos < PAGE_SIZE)
3636 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3639 size = PAGE_SIZE - info->read;
3643 ret = copy_to_user(ubuf, info->spare + info->read, size);
3654 static int tracing_buffers_release(struct inode *inode, struct file *file)
3656 struct ftrace_buffer_info *info = file->private_data;
3659 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3666 struct ring_buffer *buffer;
3671 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3672 struct pipe_buffer *buf)
3674 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3679 ring_buffer_free_read_page(ref->buffer, ref->page);
3684 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3685 struct pipe_buffer *buf)
3690 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3691 struct pipe_buffer *buf)
3693 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3698 /* Pipe buffer operations for a buffer. */
3699 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
3701 .map = generic_pipe_buf_map,
3702 .unmap = generic_pipe_buf_unmap,
3703 .confirm = generic_pipe_buf_confirm,
3704 .release = buffer_pipe_buf_release,
3705 .steal = buffer_pipe_buf_steal,
3706 .get = buffer_pipe_buf_get,
3710 * Callback from splice_to_pipe(), if we need to release some pages
3711 * at the end of the spd in case we error'ed out in filling the pipe.
3713 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3715 struct buffer_ref *ref =
3716 (struct buffer_ref *)spd->partial[i].private;
3721 ring_buffer_free_read_page(ref->buffer, ref->page);
3723 spd->partial[i].private = 0;
3727 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3728 struct pipe_inode_info *pipe, size_t len,
3731 struct ftrace_buffer_info *info = file->private_data;
3732 struct partial_page partial[PIPE_BUFFERS];
3733 struct page *pages[PIPE_BUFFERS];
3734 struct splice_pipe_desc spd = {
3738 .ops = &buffer_pipe_buf_ops,
3739 .spd_release = buffer_spd_release,
3741 struct buffer_ref *ref;
3742 int entries, size, i;
3745 if (*ppos & (PAGE_SIZE - 1)) {
3746 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3750 if (len & (PAGE_SIZE - 1)) {
3751 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3752 if (len < PAGE_SIZE)
3757 trace_access_lock(info->cpu);
3758 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3760 for (i = 0; i < PIPE_BUFFERS && len && entries; i++, len -= PAGE_SIZE) {
3764 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3769 ref->buffer = info->tr->buffer;
3770 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3776 r = ring_buffer_read_page(ref->buffer, &ref->page,
3779 ring_buffer_free_read_page(ref->buffer,
3786 * zero out any left over data, this is going to
3789 size = ring_buffer_page_len(ref->page);
3790 if (size < PAGE_SIZE)
3791 memset(ref->page + size, 0, PAGE_SIZE - size);
3793 page = virt_to_page(ref->page);
3795 spd.pages[i] = page;
3796 spd.partial[i].len = PAGE_SIZE;
3797 spd.partial[i].offset = 0;
3798 spd.partial[i].private = (unsigned long)ref;
3802 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3805 trace_access_unlock(info->cpu);
3808 /* did we read anything? */
3809 if (!spd.nr_pages) {
3810 if (flags & SPLICE_F_NONBLOCK)
3818 ret = splice_to_pipe(pipe, &spd);
3823 static const struct file_operations tracing_buffers_fops = {
3824 .open = tracing_buffers_open,
3825 .read = tracing_buffers_read,
3826 .release = tracing_buffers_release,
3827 .splice_read = tracing_buffers_splice_read,
3828 .llseek = no_llseek,
3832 tracing_stats_read(struct file *filp, char __user *ubuf,
3833 size_t count, loff_t *ppos)
3835 unsigned long cpu = (unsigned long)filp->private_data;
3836 struct trace_array *tr = &global_trace;
3837 struct trace_seq *s;
3840 s = kmalloc(sizeof(*s), GFP_KERNEL);
3846 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3847 trace_seq_printf(s, "entries: %ld\n", cnt);
3849 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3850 trace_seq_printf(s, "overrun: %ld\n", cnt);
3852 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3853 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3855 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3862 static const struct file_operations tracing_stats_fops = {
3863 .open = tracing_open_generic,
3864 .read = tracing_stats_read,
3867 #ifdef CONFIG_DYNAMIC_FTRACE
3869 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3875 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3876 size_t cnt, loff_t *ppos)
3878 static char ftrace_dyn_info_buffer[1024];
3879 static DEFINE_MUTEX(dyn_info_mutex);
3880 unsigned long *p = filp->private_data;
3881 char *buf = ftrace_dyn_info_buffer;
3882 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3885 mutex_lock(&dyn_info_mutex);
3886 r = sprintf(buf, "%ld ", *p);
3888 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3891 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3893 mutex_unlock(&dyn_info_mutex);
3898 static const struct file_operations tracing_dyn_info_fops = {
3899 .open = tracing_open_generic,
3900 .read = tracing_read_dyn_info,
3904 static struct dentry *d_tracer;
3906 struct dentry *tracing_init_dentry(void)
3913 if (!debugfs_initialized())
3916 d_tracer = debugfs_create_dir("tracing", NULL);
3918 if (!d_tracer && !once) {
3920 pr_warning("Could not create debugfs directory 'tracing'\n");
3927 static struct dentry *d_percpu;
3929 struct dentry *tracing_dentry_percpu(void)
3932 struct dentry *d_tracer;
3937 d_tracer = tracing_init_dentry();
3942 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3944 if (!d_percpu && !once) {
3946 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3953 static void tracing_init_debugfs_percpu(long cpu)
3955 struct dentry *d_percpu = tracing_dentry_percpu();
3956 struct dentry *d_cpu;
3957 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3960 if (cpu > 999 || cpu < 0)
3963 sprintf(cpu_dir, "cpu%ld", cpu);
3964 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3966 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3970 /* per cpu trace_pipe */
3971 trace_create_file("trace_pipe", 0444, d_cpu,
3972 (void *) cpu, &tracing_pipe_fops);
3975 trace_create_file("trace", 0644, d_cpu,
3976 (void *) cpu, &tracing_fops);
3978 trace_create_file("trace_pipe_raw", 0444, d_cpu,
3979 (void *) cpu, &tracing_buffers_fops);
3981 trace_create_file("stats", 0444, d_cpu,
3982 (void *) cpu, &tracing_stats_fops);
3985 #ifdef CONFIG_FTRACE_SELFTEST
3986 /* Let selftest have access to static functions in this file */
3987 #include "trace_selftest.c"
3990 struct trace_option_dentry {
3991 struct tracer_opt *opt;
3992 struct tracer_flags *flags;
3993 struct dentry *entry;
3997 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4000 struct trace_option_dentry *topt = filp->private_data;
4003 if (topt->flags->val & topt->opt->bit)
4008 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4012 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4015 struct trace_option_dentry *topt = filp->private_data;
4020 if (cnt >= sizeof(buf))
4023 if (copy_from_user(&buf, ubuf, cnt))
4028 ret = strict_strtoul(buf, 10, &val);
4032 if (val != 0 && val != 1)
4035 if (!!(topt->flags->val & topt->opt->bit) != val) {
4036 mutex_lock(&trace_types_lock);
4037 ret = __set_tracer_option(current_trace, topt->flags,
4039 mutex_unlock(&trace_types_lock);
4050 static const struct file_operations trace_options_fops = {
4051 .open = tracing_open_generic,
4052 .read = trace_options_read,
4053 .write = trace_options_write,
4057 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4060 long index = (long)filp->private_data;
4063 if (trace_flags & (1 << index))
4068 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4072 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4075 long index = (long)filp->private_data;
4080 if (cnt >= sizeof(buf))
4083 if (copy_from_user(&buf, ubuf, cnt))
4088 ret = strict_strtoul(buf, 10, &val);
4092 if (val != 0 && val != 1)
4094 set_tracer_flags(1 << index, val);
4101 static const struct file_operations trace_options_core_fops = {
4102 .open = tracing_open_generic,
4103 .read = trace_options_core_read,
4104 .write = trace_options_core_write,
4107 struct dentry *trace_create_file(const char *name,
4109 struct dentry *parent,
4111 const struct file_operations *fops)
4115 ret = debugfs_create_file(name, mode, parent, data, fops);
4117 pr_warning("Could not create debugfs '%s' entry\n", name);
4123 static struct dentry *trace_options_init_dentry(void)
4125 struct dentry *d_tracer;
4126 static struct dentry *t_options;
4131 d_tracer = tracing_init_dentry();
4135 t_options = debugfs_create_dir("options", d_tracer);
4137 pr_warning("Could not create debugfs directory 'options'\n");
4145 create_trace_option_file(struct trace_option_dentry *topt,
4146 struct tracer_flags *flags,
4147 struct tracer_opt *opt)
4149 struct dentry *t_options;
4151 t_options = trace_options_init_dentry();
4155 topt->flags = flags;
4158 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4159 &trace_options_fops);
4163 static struct trace_option_dentry *
4164 create_trace_option_files(struct tracer *tracer)
4166 struct trace_option_dentry *topts;
4167 struct tracer_flags *flags;
4168 struct tracer_opt *opts;
4174 flags = tracer->flags;
4176 if (!flags || !flags->opts)
4181 for (cnt = 0; opts[cnt].name; cnt++)
4184 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4188 for (cnt = 0; opts[cnt].name; cnt++)
4189 create_trace_option_file(&topts[cnt], flags,
4196 destroy_trace_option_files(struct trace_option_dentry *topts)
4203 for (cnt = 0; topts[cnt].opt; cnt++) {
4204 if (topts[cnt].entry)
4205 debugfs_remove(topts[cnt].entry);
4211 static struct dentry *
4212 create_trace_option_core_file(const char *option, long index)
4214 struct dentry *t_options;
4216 t_options = trace_options_init_dentry();
4220 return trace_create_file(option, 0644, t_options, (void *)index,
4221 &trace_options_core_fops);
4224 static __init void create_trace_options_dir(void)
4226 struct dentry *t_options;
4229 t_options = trace_options_init_dentry();
4233 for (i = 0; trace_options[i]; i++)
4234 create_trace_option_core_file(trace_options[i], i);
4237 static __init int tracer_init_debugfs(void)
4239 struct dentry *d_tracer;
4242 trace_access_lock_init();
4244 d_tracer = tracing_init_dentry();
4246 trace_create_file("tracing_enabled", 0644, d_tracer,
4247 &global_trace, &tracing_ctrl_fops);
4249 trace_create_file("trace_options", 0644, d_tracer,
4250 NULL, &tracing_iter_fops);
4252 trace_create_file("tracing_cpumask", 0644, d_tracer,
4253 NULL, &tracing_cpumask_fops);
4255 trace_create_file("trace", 0644, d_tracer,
4256 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4258 trace_create_file("available_tracers", 0444, d_tracer,
4259 &global_trace, &show_traces_fops);
4261 trace_create_file("current_tracer", 0644, d_tracer,
4262 &global_trace, &set_tracer_fops);
4264 #ifdef CONFIG_TRACER_MAX_TRACE
4265 trace_create_file("tracing_max_latency", 0644, d_tracer,
4266 &tracing_max_latency, &tracing_max_lat_fops);
4269 trace_create_file("tracing_thresh", 0644, d_tracer,
4270 &tracing_thresh, &tracing_max_lat_fops);
4272 trace_create_file("README", 0444, d_tracer,
4273 NULL, &tracing_readme_fops);
4275 trace_create_file("trace_pipe", 0444, d_tracer,
4276 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4278 trace_create_file("buffer_size_kb", 0644, d_tracer,
4279 &global_trace, &tracing_entries_fops);
4281 trace_create_file("trace_marker", 0220, d_tracer,
4282 NULL, &tracing_mark_fops);
4284 trace_create_file("saved_cmdlines", 0444, d_tracer,
4285 NULL, &tracing_saved_cmdlines_fops);
4287 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4290 #ifdef CONFIG_DYNAMIC_FTRACE
4291 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4292 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4294 #ifdef CONFIG_SYSPROF_TRACER
4295 init_tracer_sysprof_debugfs(d_tracer);
4298 create_trace_options_dir();
4300 for_each_tracing_cpu(cpu)
4301 tracing_init_debugfs_percpu(cpu);
4306 static int trace_panic_handler(struct notifier_block *this,
4307 unsigned long event, void *unused)
4309 if (ftrace_dump_on_oops)
4314 static struct notifier_block trace_panic_notifier = {
4315 .notifier_call = trace_panic_handler,
4317 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4320 static int trace_die_handler(struct notifier_block *self,
4326 if (ftrace_dump_on_oops)
4335 static struct notifier_block trace_die_notifier = {
4336 .notifier_call = trace_die_handler,
4341 * printk is set to max of 1024, we really don't need it that big.
4342 * Nothing should be printing 1000 characters anyway.
4344 #define TRACE_MAX_PRINT 1000
4347 * Define here KERN_TRACE so that we have one place to modify
4348 * it if we decide to change what log level the ftrace dump
4351 #define KERN_TRACE KERN_EMERG
4354 trace_printk_seq(struct trace_seq *s)
4356 /* Probably should print a warning here. */
4360 /* should be zero ended, but we are paranoid. */
4361 s->buffer[s->len] = 0;
4363 printk(KERN_TRACE "%s", s->buffer);
4368 static void __ftrace_dump(bool disable_tracing)
4370 static arch_spinlock_t ftrace_dump_lock =
4371 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4372 /* use static because iter can be a bit big for the stack */
4373 static struct trace_iterator iter;
4374 unsigned int old_userobj;
4375 static int dump_ran;
4376 unsigned long flags;
4380 local_irq_save(flags);
4381 arch_spin_lock(&ftrace_dump_lock);
4389 if (disable_tracing)
4392 for_each_tracing_cpu(cpu) {
4393 atomic_inc(&global_trace.data[cpu]->disabled);
4396 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4398 /* don't look at user memory in panic mode */
4399 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4401 printk(KERN_TRACE "Dumping ftrace buffer:\n");
4403 /* Simulate the iterator */
4404 iter.tr = &global_trace;
4405 iter.trace = current_trace;
4406 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4409 * We need to stop all tracing on all CPUS to read the
4410 * the next buffer. This is a bit expensive, but is
4411 * not done often. We fill all what we can read,
4412 * and then release the locks again.
4415 while (!trace_empty(&iter)) {
4418 printk(KERN_TRACE "---------------------------------\n");
4422 /* reset all but tr, trace, and overruns */
4423 memset(&iter.seq, 0,
4424 sizeof(struct trace_iterator) -
4425 offsetof(struct trace_iterator, seq));
4426 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4429 if (find_next_entry_inc(&iter) != NULL) {
4432 ret = print_trace_line(&iter);
4433 if (ret != TRACE_TYPE_NO_CONSUME)
4434 trace_consume(&iter);
4437 trace_printk_seq(&iter.seq);
4441 printk(KERN_TRACE " (ftrace buffer empty)\n");
4443 printk(KERN_TRACE "---------------------------------\n");
4445 /* Re-enable tracing if requested */
4446 if (!disable_tracing) {
4447 trace_flags |= old_userobj;
4449 for_each_tracing_cpu(cpu) {
4450 atomic_dec(&global_trace.data[cpu]->disabled);
4456 arch_spin_unlock(&ftrace_dump_lock);
4457 local_irq_restore(flags);
4460 /* By default: disable tracing after the dump */
4461 void ftrace_dump(void)
4463 __ftrace_dump(true);
4466 __init static int tracer_alloc_buffers(void)
4472 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4475 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4476 goto out_free_buffer_mask;
4478 /* To save memory, keep the ring buffer size to its minimum */
4479 if (ring_buffer_expanded)
4480 ring_buf_size = trace_buf_size;
4484 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4485 cpumask_copy(tracing_cpumask, cpu_all_mask);
4487 /* TODO: make the number of buffers hot pluggable with CPUS */
4488 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4489 TRACE_BUFFER_FLAGS);
4490 if (!global_trace.buffer) {
4491 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4493 goto out_free_cpumask;
4495 global_trace.entries = ring_buffer_size(global_trace.buffer);
4498 #ifdef CONFIG_TRACER_MAX_TRACE
4499 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4500 TRACE_BUFFER_FLAGS);
4501 if (!max_tr.buffer) {
4502 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4504 ring_buffer_free(global_trace.buffer);
4505 goto out_free_cpumask;
4507 max_tr.entries = ring_buffer_size(max_tr.buffer);
4508 WARN_ON(max_tr.entries != global_trace.entries);
4511 /* Allocate the first page for all buffers */
4512 for_each_tracing_cpu(i) {
4513 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4514 max_tr.data[i] = &per_cpu(max_tr_data, i);
4517 trace_init_cmdlines();
4519 register_tracer(&nop_trace);
4520 current_trace = &nop_trace;
4521 #ifdef CONFIG_BOOT_TRACER
4522 register_tracer(&boot_tracer);
4524 /* All seems OK, enable tracing */
4525 tracing_disabled = 0;
4527 atomic_notifier_chain_register(&panic_notifier_list,
4528 &trace_panic_notifier);
4530 register_die_notifier(&trace_die_notifier);
4535 free_cpumask_var(tracing_cpumask);
4536 out_free_buffer_mask:
4537 free_cpumask_var(tracing_buffer_mask);
4542 __init static int clear_boot_tracer(void)
4545 * The default tracer at boot buffer is an init section.
4546 * This function is called in lateinit. If we did not
4547 * find the boot tracer, then clear it out, to prevent
4548 * later registration from accessing the buffer that is
4549 * about to be freed.
4551 if (!default_bootup_tracer)
4554 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4555 default_bootup_tracer);
4556 default_bootup_tracer = NULL;
4561 early_initcall(tracer_alloc_buffers);
4562 fs_initcall(tracer_init_debugfs);
4563 late_initcall(clear_boot_tracer);