2 * ring buffer based function tracer
4 * Copyright (C) 2007-2012 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 Nadia Yvette Chambers
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/notifier.h>
21 #include <linux/irqflags.h>
22 #include <linux/debugfs.h>
23 #include <linux/pagemap.h>
24 #include <linux/hardirq.h>
25 #include <linux/linkage.h>
26 #include <linux/uaccess.h>
27 #include <linux/kprobes.h>
28 #include <linux/ftrace.h>
29 #include <linux/module.h>
30 #include <linux/percpu.h>
31 #include <linux/splice.h>
32 #include <linux/kdebug.h>
33 #include <linux/string.h>
34 #include <linux/rwsem.h>
35 #include <linux/slab.h>
36 #include <linux/ctype.h>
37 #include <linux/init.h>
38 #include <linux/poll.h>
39 #include <linux/nmi.h>
41 #include <linux/sched/rt.h>
44 #include "trace_output.h"
47 * On boot up, the ring buffer is set to the minimum size, so that
48 * we do not waste memory on systems that are not using tracing.
50 bool ring_buffer_expanded;
53 * We need to change this state when a selftest is running.
54 * A selftest will lurk into the ring-buffer to count the
55 * entries inserted during the selftest although some concurrent
56 * insertions into the ring-buffer such as trace_printk could occurred
57 * at the same time, giving false positive or negative results.
59 static bool __read_mostly tracing_selftest_running;
62 * If a tracer is running, we do not want to run SELFTEST.
64 bool __read_mostly tracing_selftest_disabled;
66 /* For tracers that don't implement custom flags */
67 static struct tracer_opt dummy_tracer_opt[] = {
71 static struct tracer_flags dummy_tracer_flags = {
73 .opts = dummy_tracer_opt
76 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
82 * To prevent the comm cache from being overwritten when no
83 * tracing is active, only save the comm when a trace event
86 static DEFINE_PER_CPU(bool, trace_cmdline_save);
89 * Kill all tracing for good (never come back).
90 * It is initialized to 1 but will turn to zero if the initialization
91 * of the tracer is successful. But that is the only place that sets
94 static int tracing_disabled = 1;
96 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
98 cpumask_var_t __read_mostly tracing_buffer_mask;
101 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
103 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
104 * is set, then ftrace_dump is called. This will output the contents
105 * of the ftrace buffers to the console. This is very useful for
106 * capturing traces that lead to crashes and outputing it to a
109 * It is default off, but you can enable it with either specifying
110 * "ftrace_dump_on_oops" in the kernel command line, or setting
111 * /proc/sys/kernel/ftrace_dump_on_oops
112 * Set 1 if you want to dump buffers of all CPUs
113 * Set 2 if you want to dump the buffer of the CPU that triggered oops
116 enum ftrace_dump_mode ftrace_dump_on_oops;
118 /* When set, tracing will stop when a WARN*() is hit */
119 int __disable_trace_on_warning;
121 static int tracing_set_tracer(const char *buf);
123 #define MAX_TRACER_SIZE 100
124 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
125 static char *default_bootup_tracer;
127 static bool allocate_snapshot;
129 static int __init set_cmdline_ftrace(char *str)
131 strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
132 default_bootup_tracer = bootup_tracer_buf;
133 /* We are using ftrace early, expand it */
134 ring_buffer_expanded = true;
137 __setup("ftrace=", set_cmdline_ftrace);
139 static int __init set_ftrace_dump_on_oops(char *str)
141 if (*str++ != '=' || !*str) {
142 ftrace_dump_on_oops = DUMP_ALL;
146 if (!strcmp("orig_cpu", str)) {
147 ftrace_dump_on_oops = DUMP_ORIG;
153 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
155 static int __init stop_trace_on_warning(char *str)
157 __disable_trace_on_warning = 1;
160 __setup("traceoff_on_warning=", stop_trace_on_warning);
162 static int __init boot_alloc_snapshot(char *str)
164 allocate_snapshot = true;
165 /* We also need the main ring buffer expanded */
166 ring_buffer_expanded = true;
169 __setup("alloc_snapshot", boot_alloc_snapshot);
172 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
173 static char *trace_boot_options __initdata;
175 static int __init set_trace_boot_options(char *str)
177 strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
178 trace_boot_options = trace_boot_options_buf;
181 __setup("trace_options=", set_trace_boot_options);
184 unsigned long long ns2usecs(cycle_t nsec)
192 * The global_trace is the descriptor that holds the tracing
193 * buffers for the live tracing. For each CPU, it contains
194 * a link list of pages that will store trace entries. The
195 * page descriptor of the pages in the memory is used to hold
196 * the link list by linking the lru item in the page descriptor
197 * to each of the pages in the buffer per CPU.
199 * For each active CPU there is a data field that holds the
200 * pages for the buffer for that CPU. Each CPU has the same number
201 * of pages allocated for its buffer.
203 static struct trace_array global_trace;
205 LIST_HEAD(ftrace_trace_arrays);
207 int trace_array_get(struct trace_array *this_tr)
209 struct trace_array *tr;
212 mutex_lock(&trace_types_lock);
213 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
220 mutex_unlock(&trace_types_lock);
225 static void __trace_array_put(struct trace_array *this_tr)
227 WARN_ON(!this_tr->ref);
231 void trace_array_put(struct trace_array *this_tr)
233 mutex_lock(&trace_types_lock);
234 __trace_array_put(this_tr);
235 mutex_unlock(&trace_types_lock);
238 int filter_check_discard(struct ftrace_event_file *file, void *rec,
239 struct ring_buffer *buffer,
240 struct ring_buffer_event *event)
242 if (unlikely(file->flags & FTRACE_EVENT_FL_FILTERED) &&
243 !filter_match_preds(file->filter, rec)) {
244 ring_buffer_discard_commit(buffer, event);
250 EXPORT_SYMBOL_GPL(filter_check_discard);
252 int call_filter_check_discard(struct ftrace_event_call *call, void *rec,
253 struct ring_buffer *buffer,
254 struct ring_buffer_event *event)
256 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
257 !filter_match_preds(call->filter, rec)) {
258 ring_buffer_discard_commit(buffer, event);
264 EXPORT_SYMBOL_GPL(call_filter_check_discard);
266 cycle_t buffer_ftrace_now(struct trace_buffer *buf, int cpu)
270 /* Early boot up does not have a buffer yet */
272 return trace_clock_local();
274 ts = ring_buffer_time_stamp(buf->buffer, cpu);
275 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
280 cycle_t ftrace_now(int cpu)
282 return buffer_ftrace_now(&global_trace.trace_buffer, cpu);
286 * tracing_is_enabled - Show if global_trace has been disabled
288 * Shows if the global trace has been enabled or not. It uses the
289 * mirror flag "buffer_disabled" to be used in fast paths such as for
290 * the irqsoff tracer. But it may be inaccurate due to races. If you
291 * need to know the accurate state, use tracing_is_on() which is a little
292 * slower, but accurate.
294 int tracing_is_enabled(void)
297 * For quick access (irqsoff uses this in fast path), just
298 * return the mirror variable of the state of the ring buffer.
299 * It's a little racy, but we don't really care.
302 return !global_trace.buffer_disabled;
306 * trace_buf_size is the size in bytes that is allocated
307 * for a buffer. Note, the number of bytes is always rounded
310 * This number is purposely set to a low number of 16384.
311 * If the dump on oops happens, it will be much appreciated
312 * to not have to wait for all that output. Anyway this can be
313 * boot time and run time configurable.
315 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
317 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
319 /* trace_types holds a link list of available tracers. */
320 static struct tracer *trace_types __read_mostly;
323 * trace_types_lock is used to protect the trace_types list.
325 DEFINE_MUTEX(trace_types_lock);
328 * serialize the access of the ring buffer
330 * ring buffer serializes readers, but it is low level protection.
331 * The validity of the events (which returns by ring_buffer_peek() ..etc)
332 * are not protected by ring buffer.
334 * The content of events may become garbage if we allow other process consumes
335 * these events concurrently:
336 * A) the page of the consumed events may become a normal page
337 * (not reader page) in ring buffer, and this page will be rewrited
338 * by events producer.
339 * B) The page of the consumed events may become a page for splice_read,
340 * and this page will be returned to system.
342 * These primitives allow multi process access to different cpu ring buffer
345 * These primitives don't distinguish read-only and read-consume access.
346 * Multi read-only access are also serialized.
350 static DECLARE_RWSEM(all_cpu_access_lock);
351 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
353 static inline void trace_access_lock(int cpu)
355 if (cpu == RING_BUFFER_ALL_CPUS) {
356 /* gain it for accessing the whole ring buffer. */
357 down_write(&all_cpu_access_lock);
359 /* gain it for accessing a cpu ring buffer. */
361 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
362 down_read(&all_cpu_access_lock);
364 /* Secondly block other access to this @cpu ring buffer. */
365 mutex_lock(&per_cpu(cpu_access_lock, cpu));
369 static inline void trace_access_unlock(int cpu)
371 if (cpu == RING_BUFFER_ALL_CPUS) {
372 up_write(&all_cpu_access_lock);
374 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
375 up_read(&all_cpu_access_lock);
379 static inline void trace_access_lock_init(void)
383 for_each_possible_cpu(cpu)
384 mutex_init(&per_cpu(cpu_access_lock, cpu));
389 static DEFINE_MUTEX(access_lock);
391 static inline void trace_access_lock(int cpu)
394 mutex_lock(&access_lock);
397 static inline void trace_access_unlock(int cpu)
400 mutex_unlock(&access_lock);
403 static inline void trace_access_lock_init(void)
409 /* trace_flags holds trace_options default values */
410 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
411 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
412 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE |
413 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | TRACE_ITER_FUNCTION;
415 static void tracer_tracing_on(struct trace_array *tr)
417 if (tr->trace_buffer.buffer)
418 ring_buffer_record_on(tr->trace_buffer.buffer);
420 * This flag is looked at when buffers haven't been allocated
421 * yet, or by some tracers (like irqsoff), that just want to
422 * know if the ring buffer has been disabled, but it can handle
423 * races of where it gets disabled but we still do a record.
424 * As the check is in the fast path of the tracers, it is more
425 * important to be fast than accurate.
427 tr->buffer_disabled = 0;
428 /* Make the flag seen by readers */
433 * tracing_on - enable tracing buffers
435 * This function enables tracing buffers that may have been
436 * disabled with tracing_off.
438 void tracing_on(void)
440 tracer_tracing_on(&global_trace);
442 EXPORT_SYMBOL_GPL(tracing_on);
445 * __trace_puts - write a constant string into the trace buffer.
446 * @ip: The address of the caller
447 * @str: The constant string to write
448 * @size: The size of the string.
450 int __trace_puts(unsigned long ip, const char *str, int size)
452 struct ring_buffer_event *event;
453 struct ring_buffer *buffer;
454 struct print_entry *entry;
455 unsigned long irq_flags;
458 alloc = sizeof(*entry) + size + 2; /* possible \n added */
460 local_save_flags(irq_flags);
461 buffer = global_trace.trace_buffer.buffer;
462 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
463 irq_flags, preempt_count());
467 entry = ring_buffer_event_data(event);
470 memcpy(&entry->buf, str, size);
472 /* Add a newline if necessary */
473 if (entry->buf[size - 1] != '\n') {
474 entry->buf[size] = '\n';
475 entry->buf[size + 1] = '\0';
477 entry->buf[size] = '\0';
479 __buffer_unlock_commit(buffer, event);
483 EXPORT_SYMBOL_GPL(__trace_puts);
486 * __trace_bputs - write the pointer to a constant string into trace buffer
487 * @ip: The address of the caller
488 * @str: The constant string to write to the buffer to
490 int __trace_bputs(unsigned long ip, const char *str)
492 struct ring_buffer_event *event;
493 struct ring_buffer *buffer;
494 struct bputs_entry *entry;
495 unsigned long irq_flags;
496 int size = sizeof(struct bputs_entry);
498 local_save_flags(irq_flags);
499 buffer = global_trace.trace_buffer.buffer;
500 event = trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
501 irq_flags, preempt_count());
505 entry = ring_buffer_event_data(event);
509 __buffer_unlock_commit(buffer, event);
513 EXPORT_SYMBOL_GPL(__trace_bputs);
515 #ifdef CONFIG_TRACER_SNAPSHOT
517 * trace_snapshot - take a snapshot of the current buffer.
519 * This causes a swap between the snapshot buffer and the current live
520 * tracing buffer. You can use this to take snapshots of the live
521 * trace when some condition is triggered, but continue to trace.
523 * Note, make sure to allocate the snapshot with either
524 * a tracing_snapshot_alloc(), or by doing it manually
525 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
527 * If the snapshot buffer is not allocated, it will stop tracing.
528 * Basically making a permanent snapshot.
530 void tracing_snapshot(void)
532 struct trace_array *tr = &global_trace;
533 struct tracer *tracer = tr->current_trace;
537 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
538 internal_trace_puts("*** snapshot is being ignored ***\n");
542 if (!tr->allocated_snapshot) {
543 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
544 internal_trace_puts("*** stopping trace here! ***\n");
549 /* Note, snapshot can not be used when the tracer uses it */
550 if (tracer->use_max_tr) {
551 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
552 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
556 local_irq_save(flags);
557 update_max_tr(tr, current, smp_processor_id());
558 local_irq_restore(flags);
560 EXPORT_SYMBOL_GPL(tracing_snapshot);
562 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
563 struct trace_buffer *size_buf, int cpu_id);
564 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val);
566 static int alloc_snapshot(struct trace_array *tr)
570 if (!tr->allocated_snapshot) {
572 /* allocate spare buffer */
573 ret = resize_buffer_duplicate_size(&tr->max_buffer,
574 &tr->trace_buffer, RING_BUFFER_ALL_CPUS);
578 tr->allocated_snapshot = true;
584 void free_snapshot(struct trace_array *tr)
587 * We don't free the ring buffer. instead, resize it because
588 * The max_tr ring buffer has some state (e.g. ring->clock) and
589 * we want preserve it.
591 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
592 set_buffer_entries(&tr->max_buffer, 1);
593 tracing_reset_online_cpus(&tr->max_buffer);
594 tr->allocated_snapshot = false;
598 * trace_snapshot_alloc - allocate and take a snapshot of the current buffer.
600 * This is similar to trace_snapshot(), but it will allocate the
601 * snapshot buffer if it isn't already allocated. Use this only
602 * where it is safe to sleep, as the allocation may sleep.
604 * This causes a swap between the snapshot buffer and the current live
605 * tracing buffer. You can use this to take snapshots of the live
606 * trace when some condition is triggered, but continue to trace.
608 void tracing_snapshot_alloc(void)
610 struct trace_array *tr = &global_trace;
613 ret = alloc_snapshot(tr);
614 if (WARN_ON(ret < 0))
619 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
621 void tracing_snapshot(void)
623 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
625 EXPORT_SYMBOL_GPL(tracing_snapshot);
626 void tracing_snapshot_alloc(void)
631 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
632 #endif /* CONFIG_TRACER_SNAPSHOT */
634 static void tracer_tracing_off(struct trace_array *tr)
636 if (tr->trace_buffer.buffer)
637 ring_buffer_record_off(tr->trace_buffer.buffer);
639 * This flag is looked at when buffers haven't been allocated
640 * yet, or by some tracers (like irqsoff), that just want to
641 * know if the ring buffer has been disabled, but it can handle
642 * races of where it gets disabled but we still do a record.
643 * As the check is in the fast path of the tracers, it is more
644 * important to be fast than accurate.
646 tr->buffer_disabled = 1;
647 /* Make the flag seen by readers */
652 * tracing_off - turn off tracing buffers
654 * This function stops the tracing buffers from recording data.
655 * It does not disable any overhead the tracers themselves may
656 * be causing. This function simply causes all recording to
657 * the ring buffers to fail.
659 void tracing_off(void)
661 tracer_tracing_off(&global_trace);
663 EXPORT_SYMBOL_GPL(tracing_off);
665 void disable_trace_on_warning(void)
667 if (__disable_trace_on_warning)
672 * tracer_tracing_is_on - show real state of ring buffer enabled
673 * @tr : the trace array to know if ring buffer is enabled
675 * Shows real state of the ring buffer if it is enabled or not.
677 static int tracer_tracing_is_on(struct trace_array *tr)
679 if (tr->trace_buffer.buffer)
680 return ring_buffer_record_is_on(tr->trace_buffer.buffer);
681 return !tr->buffer_disabled;
685 * tracing_is_on - show state of ring buffers enabled
687 int tracing_is_on(void)
689 return tracer_tracing_is_on(&global_trace);
691 EXPORT_SYMBOL_GPL(tracing_is_on);
693 static int __init set_buf_size(char *str)
695 unsigned long buf_size;
699 buf_size = memparse(str, &str);
700 /* nr_entries can not be zero */
703 trace_buf_size = buf_size;
706 __setup("trace_buf_size=", set_buf_size);
708 static int __init set_tracing_thresh(char *str)
710 unsigned long threshold;
715 ret = kstrtoul(str, 0, &threshold);
718 tracing_thresh = threshold * 1000;
721 __setup("tracing_thresh=", set_tracing_thresh);
723 unsigned long nsecs_to_usecs(unsigned long nsecs)
728 /* These must match the bit postions in trace_iterator_flags */
729 static const char *trace_options[] = {
762 int in_ns; /* is this clock in nanoseconds? */
764 { trace_clock_local, "local", 1 },
765 { trace_clock_global, "global", 1 },
766 { trace_clock_counter, "counter", 0 },
767 { trace_clock_jiffies, "uptime", 1 },
768 { trace_clock, "perf", 1 },
773 * trace_parser_get_init - gets the buffer for trace parser
775 int trace_parser_get_init(struct trace_parser *parser, int size)
777 memset(parser, 0, sizeof(*parser));
779 parser->buffer = kmalloc(size, GFP_KERNEL);
788 * trace_parser_put - frees the buffer for trace parser
790 void trace_parser_put(struct trace_parser *parser)
792 kfree(parser->buffer);
796 * trace_get_user - reads the user input string separated by space
797 * (matched by isspace(ch))
799 * For each string found the 'struct trace_parser' is updated,
800 * and the function returns.
802 * Returns number of bytes read.
804 * See kernel/trace/trace.h for 'struct trace_parser' details.
806 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
807 size_t cnt, loff_t *ppos)
814 trace_parser_clear(parser);
816 ret = get_user(ch, ubuf++);
824 * The parser is not finished with the last write,
825 * continue reading the user input without skipping spaces.
828 /* skip white space */
829 while (cnt && isspace(ch)) {
830 ret = get_user(ch, ubuf++);
837 /* only spaces were written */
847 /* read the non-space input */
848 while (cnt && !isspace(ch)) {
849 if (parser->idx < parser->size - 1)
850 parser->buffer[parser->idx++] = ch;
855 ret = get_user(ch, ubuf++);
862 /* We either got finished input or we have to wait for another call. */
864 parser->buffer[parser->idx] = 0;
865 parser->cont = false;
866 } else if (parser->idx < parser->size - 1) {
868 parser->buffer[parser->idx++] = ch;
881 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
889 if (s->len <= s->readpos)
892 len = s->len - s->readpos;
895 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
905 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
909 if (s->len <= s->readpos)
912 len = s->len - s->readpos;
915 memcpy(buf, s->buffer + s->readpos, cnt);
922 * ftrace_max_lock is used to protect the swapping of buffers
923 * when taking a max snapshot. The buffers themselves are
924 * protected by per_cpu spinlocks. But the action of the swap
925 * needs its own lock.
927 * This is defined as a arch_spinlock_t in order to help
928 * with performance when lockdep debugging is enabled.
930 * It is also used in other places outside the update_max_tr
931 * so it needs to be defined outside of the
932 * CONFIG_TRACER_MAX_TRACE.
934 static arch_spinlock_t ftrace_max_lock =
935 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
937 unsigned long __read_mostly tracing_thresh;
939 #ifdef CONFIG_TRACER_MAX_TRACE
940 unsigned long __read_mostly tracing_max_latency;
943 * Copy the new maximum trace into the separate maximum-trace
944 * structure. (this way the maximum trace is permanently saved,
945 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
948 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
950 struct trace_buffer *trace_buf = &tr->trace_buffer;
951 struct trace_buffer *max_buf = &tr->max_buffer;
952 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
953 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
956 max_buf->time_start = data->preempt_timestamp;
958 max_data->saved_latency = tracing_max_latency;
959 max_data->critical_start = data->critical_start;
960 max_data->critical_end = data->critical_end;
962 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
963 max_data->pid = tsk->pid;
965 * If tsk == current, then use current_uid(), as that does not use
966 * RCU. The irq tracer can be called out of RCU scope.
969 max_data->uid = current_uid();
971 max_data->uid = task_uid(tsk);
973 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
974 max_data->policy = tsk->policy;
975 max_data->rt_priority = tsk->rt_priority;
977 /* record this tasks comm */
978 tracing_record_cmdline(tsk);
982 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
984 * @tsk: the task with the latency
985 * @cpu: The cpu that initiated the trace.
987 * Flip the buffers between the @tr and the max_tr and record information
988 * about which task was the cause of this latency.
991 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
993 struct ring_buffer *buf;
998 WARN_ON_ONCE(!irqs_disabled());
1000 if (!tr->allocated_snapshot) {
1001 /* Only the nop tracer should hit this when disabling */
1002 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1006 arch_spin_lock(&ftrace_max_lock);
1008 buf = tr->trace_buffer.buffer;
1009 tr->trace_buffer.buffer = tr->max_buffer.buffer;
1010 tr->max_buffer.buffer = buf;
1012 __update_max_tr(tr, tsk, cpu);
1013 arch_spin_unlock(&ftrace_max_lock);
1017 * update_max_tr_single - only copy one trace over, and reset the rest
1019 * @tsk - task with the latency
1020 * @cpu - the cpu of the buffer to copy.
1022 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1025 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1032 WARN_ON_ONCE(!irqs_disabled());
1033 if (!tr->allocated_snapshot) {
1034 /* Only the nop tracer should hit this when disabling */
1035 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1039 arch_spin_lock(&ftrace_max_lock);
1041 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->trace_buffer.buffer, cpu);
1043 if (ret == -EBUSY) {
1045 * We failed to swap the buffer due to a commit taking
1046 * place on this CPU. We fail to record, but we reset
1047 * the max trace buffer (no one writes directly to it)
1048 * and flag that it failed.
1050 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1051 "Failed to swap buffers due to commit in progress\n");
1054 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1056 __update_max_tr(tr, tsk, cpu);
1057 arch_spin_unlock(&ftrace_max_lock);
1059 #endif /* CONFIG_TRACER_MAX_TRACE */
1061 static void default_wait_pipe(struct trace_iterator *iter)
1063 /* Iterators are static, they should be filled or empty */
1064 if (trace_buffer_iter(iter, iter->cpu_file))
1067 ring_buffer_wait(iter->trace_buffer->buffer, iter->cpu_file);
1070 #ifdef CONFIG_FTRACE_STARTUP_TEST
1071 static int run_tracer_selftest(struct tracer *type)
1073 struct trace_array *tr = &global_trace;
1074 struct tracer *saved_tracer = tr->current_trace;
1077 if (!type->selftest || tracing_selftest_disabled)
1081 * Run a selftest on this tracer.
1082 * Here we reset the trace buffer, and set the current
1083 * tracer to be this tracer. The tracer can then run some
1084 * internal tracing to verify that everything is in order.
1085 * If we fail, we do not register this tracer.
1087 tracing_reset_online_cpus(&tr->trace_buffer);
1089 tr->current_trace = type;
1091 #ifdef CONFIG_TRACER_MAX_TRACE
1092 if (type->use_max_tr) {
1093 /* If we expanded the buffers, make sure the max is expanded too */
1094 if (ring_buffer_expanded)
1095 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1096 RING_BUFFER_ALL_CPUS);
1097 tr->allocated_snapshot = true;
1101 /* the test is responsible for initializing and enabling */
1102 pr_info("Testing tracer %s: ", type->name);
1103 ret = type->selftest(type, tr);
1104 /* the test is responsible for resetting too */
1105 tr->current_trace = saved_tracer;
1107 printk(KERN_CONT "FAILED!\n");
1108 /* Add the warning after printing 'FAILED' */
1112 /* Only reset on passing, to avoid touching corrupted buffers */
1113 tracing_reset_online_cpus(&tr->trace_buffer);
1115 #ifdef CONFIG_TRACER_MAX_TRACE
1116 if (type->use_max_tr) {
1117 tr->allocated_snapshot = false;
1119 /* Shrink the max buffer again */
1120 if (ring_buffer_expanded)
1121 ring_buffer_resize(tr->max_buffer.buffer, 1,
1122 RING_BUFFER_ALL_CPUS);
1126 printk(KERN_CONT "PASSED\n");
1130 static inline int run_tracer_selftest(struct tracer *type)
1134 #endif /* CONFIG_FTRACE_STARTUP_TEST */
1137 * register_tracer - register a tracer with the ftrace system.
1138 * @type - the plugin for the tracer
1140 * Register a new plugin tracer.
1142 int register_tracer(struct tracer *type)
1148 pr_info("Tracer must have a name\n");
1152 if (strlen(type->name) >= MAX_TRACER_SIZE) {
1153 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
1157 mutex_lock(&trace_types_lock);
1159 tracing_selftest_running = true;
1161 for (t = trace_types; t; t = t->next) {
1162 if (strcmp(type->name, t->name) == 0) {
1164 pr_info("Tracer %s already registered\n",
1171 if (!type->set_flag)
1172 type->set_flag = &dummy_set_flag;
1174 type->flags = &dummy_tracer_flags;
1176 if (!type->flags->opts)
1177 type->flags->opts = dummy_tracer_opt;
1178 if (!type->wait_pipe)
1179 type->wait_pipe = default_wait_pipe;
1181 ret = run_tracer_selftest(type);
1185 type->next = trace_types;
1189 tracing_selftest_running = false;
1190 mutex_unlock(&trace_types_lock);
1192 if (ret || !default_bootup_tracer)
1195 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
1198 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
1199 /* Do we want this tracer to start on bootup? */
1200 tracing_set_tracer(type->name);
1201 default_bootup_tracer = NULL;
1202 /* disable other selftests, since this will break it. */
1203 tracing_selftest_disabled = true;
1204 #ifdef CONFIG_FTRACE_STARTUP_TEST
1205 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
1213 void tracing_reset(struct trace_buffer *buf, int cpu)
1215 struct ring_buffer *buffer = buf->buffer;
1220 ring_buffer_record_disable(buffer);
1222 /* Make sure all commits have finished */
1223 synchronize_sched();
1224 ring_buffer_reset_cpu(buffer, cpu);
1226 ring_buffer_record_enable(buffer);
1229 void tracing_reset_online_cpus(struct trace_buffer *buf)
1231 struct ring_buffer *buffer = buf->buffer;
1237 ring_buffer_record_disable(buffer);
1239 /* Make sure all commits have finished */
1240 synchronize_sched();
1242 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
1244 for_each_online_cpu(cpu)
1245 ring_buffer_reset_cpu(buffer, cpu);
1247 ring_buffer_record_enable(buffer);
1250 /* Must have trace_types_lock held */
1251 void tracing_reset_all_online_cpus(void)
1253 struct trace_array *tr;
1255 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1256 tracing_reset_online_cpus(&tr->trace_buffer);
1257 #ifdef CONFIG_TRACER_MAX_TRACE
1258 tracing_reset_online_cpus(&tr->max_buffer);
1263 #define SAVED_CMDLINES 128
1264 #define NO_CMDLINE_MAP UINT_MAX
1265 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
1266 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
1267 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
1268 static int cmdline_idx;
1269 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1271 /* temporary disable recording */
1272 static atomic_t trace_record_cmdline_disabled __read_mostly;
1274 static void trace_init_cmdlines(void)
1276 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
1277 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
1281 int is_tracing_stopped(void)
1283 return global_trace.stop_count;
1287 * ftrace_off_permanent - disable all ftrace code permanently
1289 * This should only be called when a serious anomally has
1290 * been detected. This will turn off the function tracing,
1291 * ring buffers, and other tracing utilites. It takes no
1292 * locks and can be called from any context.
1294 void ftrace_off_permanent(void)
1296 tracing_disabled = 1;
1298 tracing_off_permanent();
1302 * tracing_start - quick start of the tracer
1304 * If tracing is enabled but was stopped by tracing_stop,
1305 * this will start the tracer back up.
1307 void tracing_start(void)
1309 struct ring_buffer *buffer;
1310 unsigned long flags;
1312 if (tracing_disabled)
1315 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1316 if (--global_trace.stop_count) {
1317 if (global_trace.stop_count < 0) {
1318 /* Someone screwed up their debugging */
1320 global_trace.stop_count = 0;
1325 /* Prevent the buffers from switching */
1326 arch_spin_lock(&ftrace_max_lock);
1328 buffer = global_trace.trace_buffer.buffer;
1330 ring_buffer_record_enable(buffer);
1332 #ifdef CONFIG_TRACER_MAX_TRACE
1333 buffer = global_trace.max_buffer.buffer;
1335 ring_buffer_record_enable(buffer);
1338 arch_spin_unlock(&ftrace_max_lock);
1342 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1345 static void tracing_start_tr(struct trace_array *tr)
1347 struct ring_buffer *buffer;
1348 unsigned long flags;
1350 if (tracing_disabled)
1353 /* If global, we need to also start the max tracer */
1354 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1355 return tracing_start();
1357 raw_spin_lock_irqsave(&tr->start_lock, flags);
1359 if (--tr->stop_count) {
1360 if (tr->stop_count < 0) {
1361 /* Someone screwed up their debugging */
1368 buffer = tr->trace_buffer.buffer;
1370 ring_buffer_record_enable(buffer);
1373 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1377 * tracing_stop - quick stop of the tracer
1379 * Light weight way to stop tracing. Use in conjunction with
1382 void tracing_stop(void)
1384 struct ring_buffer *buffer;
1385 unsigned long flags;
1388 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1389 if (global_trace.stop_count++)
1392 /* Prevent the buffers from switching */
1393 arch_spin_lock(&ftrace_max_lock);
1395 buffer = global_trace.trace_buffer.buffer;
1397 ring_buffer_record_disable(buffer);
1399 #ifdef CONFIG_TRACER_MAX_TRACE
1400 buffer = global_trace.max_buffer.buffer;
1402 ring_buffer_record_disable(buffer);
1405 arch_spin_unlock(&ftrace_max_lock);
1408 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1411 static void tracing_stop_tr(struct trace_array *tr)
1413 struct ring_buffer *buffer;
1414 unsigned long flags;
1416 /* If global, we need to also stop the max tracer */
1417 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1418 return tracing_stop();
1420 raw_spin_lock_irqsave(&tr->start_lock, flags);
1421 if (tr->stop_count++)
1424 buffer = tr->trace_buffer.buffer;
1426 ring_buffer_record_disable(buffer);
1429 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1432 void trace_stop_cmdline_recording(void);
1434 static void trace_save_cmdline(struct task_struct *tsk)
1438 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1442 * It's not the end of the world if we don't get
1443 * the lock, but we also don't want to spin
1444 * nor do we want to disable interrupts,
1445 * so if we miss here, then better luck next time.
1447 if (!arch_spin_trylock(&trace_cmdline_lock))
1450 idx = map_pid_to_cmdline[tsk->pid];
1451 if (idx == NO_CMDLINE_MAP) {
1452 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1455 * Check whether the cmdline buffer at idx has a pid
1456 * mapped. We are going to overwrite that entry so we
1457 * need to clear the map_pid_to_cmdline. Otherwise we
1458 * would read the new comm for the old pid.
1460 pid = map_cmdline_to_pid[idx];
1461 if (pid != NO_CMDLINE_MAP)
1462 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1464 map_cmdline_to_pid[idx] = tsk->pid;
1465 map_pid_to_cmdline[tsk->pid] = idx;
1470 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1472 arch_spin_unlock(&trace_cmdline_lock);
1475 void trace_find_cmdline(int pid, char comm[])
1480 strcpy(comm, "<idle>");
1484 if (WARN_ON_ONCE(pid < 0)) {
1485 strcpy(comm, "<XXX>");
1489 if (pid > PID_MAX_DEFAULT) {
1490 strcpy(comm, "<...>");
1495 arch_spin_lock(&trace_cmdline_lock);
1496 map = map_pid_to_cmdline[pid];
1497 if (map != NO_CMDLINE_MAP)
1498 strcpy(comm, saved_cmdlines[map]);
1500 strcpy(comm, "<...>");
1502 arch_spin_unlock(&trace_cmdline_lock);
1506 void tracing_record_cmdline(struct task_struct *tsk)
1508 if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on())
1511 if (!__this_cpu_read(trace_cmdline_save))
1514 __this_cpu_write(trace_cmdline_save, false);
1516 trace_save_cmdline(tsk);
1520 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1523 struct task_struct *tsk = current;
1525 entry->preempt_count = pc & 0xff;
1526 entry->pid = (tsk) ? tsk->pid : 0;
1528 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1529 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1531 TRACE_FLAG_IRQS_NOSUPPORT |
1533 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1534 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1535 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1537 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1539 struct ring_buffer_event *
1540 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1543 unsigned long flags, int pc)
1545 struct ring_buffer_event *event;
1547 event = ring_buffer_lock_reserve(buffer, len);
1548 if (event != NULL) {
1549 struct trace_entry *ent = ring_buffer_event_data(event);
1551 tracing_generic_entry_update(ent, flags, pc);
1559 __buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event)
1561 __this_cpu_write(trace_cmdline_save, true);
1562 ring_buffer_unlock_commit(buffer, event);
1566 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1567 struct ring_buffer_event *event,
1568 unsigned long flags, int pc)
1570 __buffer_unlock_commit(buffer, event);
1572 ftrace_trace_stack(buffer, flags, 6, pc);
1573 ftrace_trace_userstack(buffer, flags, pc);
1576 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1577 struct ring_buffer_event *event,
1578 unsigned long flags, int pc)
1580 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1582 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit);
1584 struct ring_buffer_event *
1585 trace_event_buffer_lock_reserve(struct ring_buffer **current_rb,
1586 struct ftrace_event_file *ftrace_file,
1587 int type, unsigned long len,
1588 unsigned long flags, int pc)
1590 *current_rb = ftrace_file->tr->trace_buffer.buffer;
1591 return trace_buffer_lock_reserve(*current_rb,
1592 type, len, flags, pc);
1594 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
1596 struct ring_buffer_event *
1597 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1598 int type, unsigned long len,
1599 unsigned long flags, int pc)
1601 *current_rb = global_trace.trace_buffer.buffer;
1602 return trace_buffer_lock_reserve(*current_rb,
1603 type, len, flags, pc);
1605 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1607 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1608 struct ring_buffer_event *event,
1609 unsigned long flags, int pc)
1611 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1613 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1615 void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1616 struct ring_buffer_event *event,
1617 unsigned long flags, int pc,
1618 struct pt_regs *regs)
1620 __buffer_unlock_commit(buffer, event);
1622 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1623 ftrace_trace_userstack(buffer, flags, pc);
1625 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs);
1627 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1628 struct ring_buffer_event *event)
1630 ring_buffer_discard_commit(buffer, event);
1632 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1635 trace_function(struct trace_array *tr,
1636 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1639 struct ftrace_event_call *call = &event_function;
1640 struct ring_buffer *buffer = tr->trace_buffer.buffer;
1641 struct ring_buffer_event *event;
1642 struct ftrace_entry *entry;
1644 /* If we are reading the ring buffer, don't trace */
1645 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1648 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1652 entry = ring_buffer_event_data(event);
1654 entry->parent_ip = parent_ip;
1656 if (!call_filter_check_discard(call, entry, buffer, event))
1657 __buffer_unlock_commit(buffer, event);
1660 #ifdef CONFIG_STACKTRACE
1662 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1663 struct ftrace_stack {
1664 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1667 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1668 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1670 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1671 unsigned long flags,
1672 int skip, int pc, struct pt_regs *regs)
1674 struct ftrace_event_call *call = &event_kernel_stack;
1675 struct ring_buffer_event *event;
1676 struct stack_entry *entry;
1677 struct stack_trace trace;
1679 int size = FTRACE_STACK_ENTRIES;
1681 trace.nr_entries = 0;
1685 * Since events can happen in NMIs there's no safe way to
1686 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1687 * or NMI comes in, it will just have to use the default
1688 * FTRACE_STACK_SIZE.
1690 preempt_disable_notrace();
1692 use_stack = __this_cpu_inc_return(ftrace_stack_reserve);
1694 * We don't need any atomic variables, just a barrier.
1695 * If an interrupt comes in, we don't care, because it would
1696 * have exited and put the counter back to what we want.
1697 * We just need a barrier to keep gcc from moving things
1701 if (use_stack == 1) {
1702 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1703 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1706 save_stack_trace_regs(regs, &trace);
1708 save_stack_trace(&trace);
1710 if (trace.nr_entries > size)
1711 size = trace.nr_entries;
1713 /* From now on, use_stack is a boolean */
1716 size *= sizeof(unsigned long);
1718 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1719 sizeof(*entry) + size, flags, pc);
1722 entry = ring_buffer_event_data(event);
1724 memset(&entry->caller, 0, size);
1727 memcpy(&entry->caller, trace.entries,
1728 trace.nr_entries * sizeof(unsigned long));
1730 trace.max_entries = FTRACE_STACK_ENTRIES;
1731 trace.entries = entry->caller;
1733 save_stack_trace_regs(regs, &trace);
1735 save_stack_trace(&trace);
1738 entry->size = trace.nr_entries;
1740 if (!call_filter_check_discard(call, entry, buffer, event))
1741 __buffer_unlock_commit(buffer, event);
1744 /* Again, don't let gcc optimize things here */
1746 __this_cpu_dec(ftrace_stack_reserve);
1747 preempt_enable_notrace();
1751 void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1752 int skip, int pc, struct pt_regs *regs)
1754 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1757 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1760 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1763 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1766 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1769 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1772 __ftrace_trace_stack(tr->trace_buffer.buffer, flags, skip, pc, NULL);
1776 * trace_dump_stack - record a stack back trace in the trace buffer
1777 * @skip: Number of functions to skip (helper handlers)
1779 void trace_dump_stack(int skip)
1781 unsigned long flags;
1783 if (tracing_disabled || tracing_selftest_running)
1786 local_save_flags(flags);
1789 * Skip 3 more, seems to get us at the caller of
1793 __ftrace_trace_stack(global_trace.trace_buffer.buffer,
1794 flags, skip, preempt_count(), NULL);
1797 static DEFINE_PER_CPU(int, user_stack_count);
1800 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1802 struct ftrace_event_call *call = &event_user_stack;
1803 struct ring_buffer_event *event;
1804 struct userstack_entry *entry;
1805 struct stack_trace trace;
1807 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1811 * NMIs can not handle page faults, even with fix ups.
1812 * The save user stack can (and often does) fault.
1814 if (unlikely(in_nmi()))
1818 * prevent recursion, since the user stack tracing may
1819 * trigger other kernel events.
1822 if (__this_cpu_read(user_stack_count))
1825 __this_cpu_inc(user_stack_count);
1827 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1828 sizeof(*entry), flags, pc);
1830 goto out_drop_count;
1831 entry = ring_buffer_event_data(event);
1833 entry->tgid = current->tgid;
1834 memset(&entry->caller, 0, sizeof(entry->caller));
1836 trace.nr_entries = 0;
1837 trace.max_entries = FTRACE_STACK_ENTRIES;
1839 trace.entries = entry->caller;
1841 save_stack_trace_user(&trace);
1842 if (!call_filter_check_discard(call, entry, buffer, event))
1843 __buffer_unlock_commit(buffer, event);
1846 __this_cpu_dec(user_stack_count);
1852 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1854 ftrace_trace_userstack(tr, flags, preempt_count());
1858 #endif /* CONFIG_STACKTRACE */
1860 /* created for use with alloc_percpu */
1861 struct trace_buffer_struct {
1862 char buffer[TRACE_BUF_SIZE];
1865 static struct trace_buffer_struct *trace_percpu_buffer;
1866 static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1867 static struct trace_buffer_struct *trace_percpu_irq_buffer;
1868 static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1871 * The buffer used is dependent on the context. There is a per cpu
1872 * buffer for normal context, softirq contex, hard irq context and
1873 * for NMI context. Thise allows for lockless recording.
1875 * Note, if the buffers failed to be allocated, then this returns NULL
1877 static char *get_trace_buf(void)
1879 struct trace_buffer_struct *percpu_buffer;
1882 * If we have allocated per cpu buffers, then we do not
1883 * need to do any locking.
1886 percpu_buffer = trace_percpu_nmi_buffer;
1888 percpu_buffer = trace_percpu_irq_buffer;
1889 else if (in_softirq())
1890 percpu_buffer = trace_percpu_sirq_buffer;
1892 percpu_buffer = trace_percpu_buffer;
1897 return this_cpu_ptr(&percpu_buffer->buffer[0]);
1900 static int alloc_percpu_trace_buffer(void)
1902 struct trace_buffer_struct *buffers;
1903 struct trace_buffer_struct *sirq_buffers;
1904 struct trace_buffer_struct *irq_buffers;
1905 struct trace_buffer_struct *nmi_buffers;
1907 buffers = alloc_percpu(struct trace_buffer_struct);
1911 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
1915 irq_buffers = alloc_percpu(struct trace_buffer_struct);
1919 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
1923 trace_percpu_buffer = buffers;
1924 trace_percpu_sirq_buffer = sirq_buffers;
1925 trace_percpu_irq_buffer = irq_buffers;
1926 trace_percpu_nmi_buffer = nmi_buffers;
1931 free_percpu(irq_buffers);
1933 free_percpu(sirq_buffers);
1935 free_percpu(buffers);
1937 WARN(1, "Could not allocate percpu trace_printk buffer");
1941 static int buffers_allocated;
1943 void trace_printk_init_buffers(void)
1945 if (buffers_allocated)
1948 if (alloc_percpu_trace_buffer())
1951 pr_info("ftrace: Allocated trace_printk buffers\n");
1953 /* Expand the buffers to set size */
1954 tracing_update_buffers();
1956 buffers_allocated = 1;
1959 * trace_printk_init_buffers() can be called by modules.
1960 * If that happens, then we need to start cmdline recording
1961 * directly here. If the global_trace.buffer is already
1962 * allocated here, then this was called by module code.
1964 if (global_trace.trace_buffer.buffer)
1965 tracing_start_cmdline_record();
1968 void trace_printk_start_comm(void)
1970 /* Start tracing comms if trace printk is set */
1971 if (!buffers_allocated)
1973 tracing_start_cmdline_record();
1976 static void trace_printk_start_stop_comm(int enabled)
1978 if (!buffers_allocated)
1982 tracing_start_cmdline_record();
1984 tracing_stop_cmdline_record();
1988 * trace_vbprintk - write binary msg to tracing buffer
1991 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1993 struct ftrace_event_call *call = &event_bprint;
1994 struct ring_buffer_event *event;
1995 struct ring_buffer *buffer;
1996 struct trace_array *tr = &global_trace;
1997 struct bprint_entry *entry;
1998 unsigned long flags;
2000 int len = 0, size, pc;
2002 if (unlikely(tracing_selftest_running || tracing_disabled))
2005 /* Don't pollute graph traces with trace_vprintk internals */
2006 pause_graph_tracing();
2008 pc = preempt_count();
2009 preempt_disable_notrace();
2011 tbuffer = get_trace_buf();
2017 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
2019 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
2022 local_save_flags(flags);
2023 size = sizeof(*entry) + sizeof(u32) * len;
2024 buffer = tr->trace_buffer.buffer;
2025 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
2029 entry = ring_buffer_event_data(event);
2033 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
2034 if (!call_filter_check_discard(call, entry, buffer, event)) {
2035 __buffer_unlock_commit(buffer, event);
2036 ftrace_trace_stack(buffer, flags, 6, pc);
2040 preempt_enable_notrace();
2041 unpause_graph_tracing();
2045 EXPORT_SYMBOL_GPL(trace_vbprintk);
2048 __trace_array_vprintk(struct ring_buffer *buffer,
2049 unsigned long ip, const char *fmt, va_list args)
2051 struct ftrace_event_call *call = &event_print;
2052 struct ring_buffer_event *event;
2053 int len = 0, size, pc;
2054 struct print_entry *entry;
2055 unsigned long flags;
2058 if (tracing_disabled || tracing_selftest_running)
2061 /* Don't pollute graph traces with trace_vprintk internals */
2062 pause_graph_tracing();
2064 pc = preempt_count();
2065 preempt_disable_notrace();
2068 tbuffer = get_trace_buf();
2074 len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
2075 if (len > TRACE_BUF_SIZE)
2078 local_save_flags(flags);
2079 size = sizeof(*entry) + len + 1;
2080 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
2084 entry = ring_buffer_event_data(event);
2087 memcpy(&entry->buf, tbuffer, len);
2088 entry->buf[len] = '\0';
2089 if (!call_filter_check_discard(call, entry, buffer, event)) {
2090 __buffer_unlock_commit(buffer, event);
2091 ftrace_trace_stack(buffer, flags, 6, pc);
2094 preempt_enable_notrace();
2095 unpause_graph_tracing();
2100 int trace_array_vprintk(struct trace_array *tr,
2101 unsigned long ip, const char *fmt, va_list args)
2103 return __trace_array_vprintk(tr->trace_buffer.buffer, ip, fmt, args);
2106 int trace_array_printk(struct trace_array *tr,
2107 unsigned long ip, const char *fmt, ...)
2112 if (!(trace_flags & TRACE_ITER_PRINTK))
2116 ret = trace_array_vprintk(tr, ip, fmt, ap);
2121 int trace_array_printk_buf(struct ring_buffer *buffer,
2122 unsigned long ip, const char *fmt, ...)
2127 if (!(trace_flags & TRACE_ITER_PRINTK))
2131 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
2136 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
2138 return trace_array_vprintk(&global_trace, ip, fmt, args);
2140 EXPORT_SYMBOL_GPL(trace_vprintk);
2142 static void trace_iterator_increment(struct trace_iterator *iter)
2144 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
2148 ring_buffer_read(buf_iter, NULL);
2151 static struct trace_entry *
2152 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
2153 unsigned long *lost_events)
2155 struct ring_buffer_event *event;
2156 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
2159 event = ring_buffer_iter_peek(buf_iter, ts);
2161 event = ring_buffer_peek(iter->trace_buffer->buffer, cpu, ts,
2165 iter->ent_size = ring_buffer_event_length(event);
2166 return ring_buffer_event_data(event);
2172 static struct trace_entry *
2173 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
2174 unsigned long *missing_events, u64 *ent_ts)
2176 struct ring_buffer *buffer = iter->trace_buffer->buffer;
2177 struct trace_entry *ent, *next = NULL;
2178 unsigned long lost_events = 0, next_lost = 0;
2179 int cpu_file = iter->cpu_file;
2180 u64 next_ts = 0, ts;
2186 * If we are in a per_cpu trace file, don't bother by iterating over
2187 * all cpu and peek directly.
2189 if (cpu_file > RING_BUFFER_ALL_CPUS) {
2190 if (ring_buffer_empty_cpu(buffer, cpu_file))
2192 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
2194 *ent_cpu = cpu_file;
2199 for_each_tracing_cpu(cpu) {
2201 if (ring_buffer_empty_cpu(buffer, cpu))
2204 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
2207 * Pick the entry with the smallest timestamp:
2209 if (ent && (!next || ts < next_ts)) {
2213 next_lost = lost_events;
2214 next_size = iter->ent_size;
2218 iter->ent_size = next_size;
2221 *ent_cpu = next_cpu;
2227 *missing_events = next_lost;
2232 /* Find the next real entry, without updating the iterator itself */
2233 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
2234 int *ent_cpu, u64 *ent_ts)
2236 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
2239 /* Find the next real entry, and increment the iterator to the next entry */
2240 void *trace_find_next_entry_inc(struct trace_iterator *iter)
2242 iter->ent = __find_next_entry(iter, &iter->cpu,
2243 &iter->lost_events, &iter->ts);
2246 trace_iterator_increment(iter);
2248 return iter->ent ? iter : NULL;
2251 static void trace_consume(struct trace_iterator *iter)
2253 ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu, &iter->ts,
2254 &iter->lost_events);
2257 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
2259 struct trace_iterator *iter = m->private;
2263 WARN_ON_ONCE(iter->leftover);
2267 /* can't go backwards */
2272 ent = trace_find_next_entry_inc(iter);
2276 while (ent && iter->idx < i)
2277 ent = trace_find_next_entry_inc(iter);
2284 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
2286 struct ring_buffer_event *event;
2287 struct ring_buffer_iter *buf_iter;
2288 unsigned long entries = 0;
2291 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = 0;
2293 buf_iter = trace_buffer_iter(iter, cpu);
2297 ring_buffer_iter_reset(buf_iter);
2300 * We could have the case with the max latency tracers
2301 * that a reset never took place on a cpu. This is evident
2302 * by the timestamp being before the start of the buffer.
2304 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
2305 if (ts >= iter->trace_buffer->time_start)
2308 ring_buffer_read(buf_iter, NULL);
2311 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = entries;
2315 * The current tracer is copied to avoid a global locking
2318 static void *s_start(struct seq_file *m, loff_t *pos)
2320 struct trace_iterator *iter = m->private;
2321 struct trace_array *tr = iter->tr;
2322 int cpu_file = iter->cpu_file;
2328 * copy the tracer to avoid using a global lock all around.
2329 * iter->trace is a copy of current_trace, the pointer to the
2330 * name may be used instead of a strcmp(), as iter->trace->name
2331 * will point to the same string as current_trace->name.
2333 mutex_lock(&trace_types_lock);
2334 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
2335 *iter->trace = *tr->current_trace;
2336 mutex_unlock(&trace_types_lock);
2338 #ifdef CONFIG_TRACER_MAX_TRACE
2339 if (iter->snapshot && iter->trace->use_max_tr)
2340 return ERR_PTR(-EBUSY);
2343 if (!iter->snapshot)
2344 atomic_inc(&trace_record_cmdline_disabled);
2346 if (*pos != iter->pos) {
2351 if (cpu_file == RING_BUFFER_ALL_CPUS) {
2352 for_each_tracing_cpu(cpu)
2353 tracing_iter_reset(iter, cpu);
2355 tracing_iter_reset(iter, cpu_file);
2358 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
2363 * If we overflowed the seq_file before, then we want
2364 * to just reuse the trace_seq buffer again.
2370 p = s_next(m, p, &l);
2374 trace_event_read_lock();
2375 trace_access_lock(cpu_file);
2379 static void s_stop(struct seq_file *m, void *p)
2381 struct trace_iterator *iter = m->private;
2383 #ifdef CONFIG_TRACER_MAX_TRACE
2384 if (iter->snapshot && iter->trace->use_max_tr)
2388 if (!iter->snapshot)
2389 atomic_dec(&trace_record_cmdline_disabled);
2391 trace_access_unlock(iter->cpu_file);
2392 trace_event_read_unlock();
2396 get_total_entries(struct trace_buffer *buf,
2397 unsigned long *total, unsigned long *entries)
2399 unsigned long count;
2405 for_each_tracing_cpu(cpu) {
2406 count = ring_buffer_entries_cpu(buf->buffer, cpu);
2408 * If this buffer has skipped entries, then we hold all
2409 * entries for the trace and we need to ignore the
2410 * ones before the time stamp.
2412 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
2413 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
2414 /* total is the same as the entries */
2418 ring_buffer_overrun_cpu(buf->buffer, cpu);
2423 static void print_lat_help_header(struct seq_file *m)
2425 seq_puts(m, "# _------=> CPU# \n");
2426 seq_puts(m, "# / _-----=> irqs-off \n");
2427 seq_puts(m, "# | / _----=> need-resched \n");
2428 seq_puts(m, "# || / _---=> hardirq/softirq \n");
2429 seq_puts(m, "# ||| / _--=> preempt-depth \n");
2430 seq_puts(m, "# |||| / delay \n");
2431 seq_puts(m, "# cmd pid ||||| time | caller \n");
2432 seq_puts(m, "# \\ / ||||| \\ | / \n");
2435 static void print_event_info(struct trace_buffer *buf, struct seq_file *m)
2437 unsigned long total;
2438 unsigned long entries;
2440 get_total_entries(buf, &total, &entries);
2441 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2442 entries, total, num_online_cpus());
2446 static void print_func_help_header(struct trace_buffer *buf, struct seq_file *m)
2448 print_event_info(buf, m);
2449 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2450 seq_puts(m, "# | | | | |\n");
2453 static void print_func_help_header_irq(struct trace_buffer *buf, struct seq_file *m)
2455 print_event_info(buf, m);
2456 seq_puts(m, "# _-----=> irqs-off\n");
2457 seq_puts(m, "# / _----=> need-resched\n");
2458 seq_puts(m, "# | / _---=> hardirq/softirq\n");
2459 seq_puts(m, "# || / _--=> preempt-depth\n");
2460 seq_puts(m, "# ||| / delay\n");
2461 seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2462 seq_puts(m, "# | | | |||| | |\n");
2466 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2468 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2469 struct trace_buffer *buf = iter->trace_buffer;
2470 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
2471 struct tracer *type = iter->trace;
2472 unsigned long entries;
2473 unsigned long total;
2474 const char *name = "preemption";
2478 get_total_entries(buf, &total, &entries);
2480 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2482 seq_puts(m, "# -----------------------------------"
2483 "---------------------------------\n");
2484 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2485 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2486 nsecs_to_usecs(data->saved_latency),
2490 #if defined(CONFIG_PREEMPT_NONE)
2492 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2494 #elif defined(CONFIG_PREEMPT)
2499 /* These are reserved for later use */
2502 seq_printf(m, " #P:%d)\n", num_online_cpus());
2506 seq_puts(m, "# -----------------\n");
2507 seq_printf(m, "# | task: %.16s-%d "
2508 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2509 data->comm, data->pid,
2510 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
2511 data->policy, data->rt_priority);
2512 seq_puts(m, "# -----------------\n");
2514 if (data->critical_start) {
2515 seq_puts(m, "# => started at: ");
2516 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2517 trace_print_seq(m, &iter->seq);
2518 seq_puts(m, "\n# => ended at: ");
2519 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2520 trace_print_seq(m, &iter->seq);
2521 seq_puts(m, "\n#\n");
2527 static void test_cpu_buff_start(struct trace_iterator *iter)
2529 struct trace_seq *s = &iter->seq;
2531 if (!(trace_flags & TRACE_ITER_ANNOTATE))
2534 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2537 if (cpumask_test_cpu(iter->cpu, iter->started))
2540 if (per_cpu_ptr(iter->trace_buffer->data, iter->cpu)->skipped_entries)
2543 cpumask_set_cpu(iter->cpu, iter->started);
2545 /* Don't print started cpu buffer for the first entry of the trace */
2547 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2551 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2553 struct trace_seq *s = &iter->seq;
2554 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2555 struct trace_entry *entry;
2556 struct trace_event *event;
2560 test_cpu_buff_start(iter);
2562 event = ftrace_find_event(entry->type);
2564 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2565 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2566 if (!trace_print_lat_context(iter))
2569 if (!trace_print_context(iter))
2575 return event->funcs->trace(iter, sym_flags, event);
2577 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2580 return TRACE_TYPE_HANDLED;
2582 return TRACE_TYPE_PARTIAL_LINE;
2585 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2587 struct trace_seq *s = &iter->seq;
2588 struct trace_entry *entry;
2589 struct trace_event *event;
2593 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2594 if (!trace_seq_printf(s, "%d %d %llu ",
2595 entry->pid, iter->cpu, iter->ts))
2599 event = ftrace_find_event(entry->type);
2601 return event->funcs->raw(iter, 0, event);
2603 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2606 return TRACE_TYPE_HANDLED;
2608 return TRACE_TYPE_PARTIAL_LINE;
2611 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2613 struct trace_seq *s = &iter->seq;
2614 unsigned char newline = '\n';
2615 struct trace_entry *entry;
2616 struct trace_event *event;
2620 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2621 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2622 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2623 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2626 event = ftrace_find_event(entry->type);
2628 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2629 if (ret != TRACE_TYPE_HANDLED)
2633 SEQ_PUT_FIELD_RET(s, newline);
2635 return TRACE_TYPE_HANDLED;
2638 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2640 struct trace_seq *s = &iter->seq;
2641 struct trace_entry *entry;
2642 struct trace_event *event;
2646 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2647 SEQ_PUT_FIELD_RET(s, entry->pid);
2648 SEQ_PUT_FIELD_RET(s, iter->cpu);
2649 SEQ_PUT_FIELD_RET(s, iter->ts);
2652 event = ftrace_find_event(entry->type);
2653 return event ? event->funcs->binary(iter, 0, event) :
2657 int trace_empty(struct trace_iterator *iter)
2659 struct ring_buffer_iter *buf_iter;
2662 /* If we are looking at one CPU buffer, only check that one */
2663 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
2664 cpu = iter->cpu_file;
2665 buf_iter = trace_buffer_iter(iter, cpu);
2667 if (!ring_buffer_iter_empty(buf_iter))
2670 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2676 for_each_tracing_cpu(cpu) {
2677 buf_iter = trace_buffer_iter(iter, cpu);
2679 if (!ring_buffer_iter_empty(buf_iter))
2682 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2690 /* Called with trace_event_read_lock() held. */
2691 enum print_line_t print_trace_line(struct trace_iterator *iter)
2693 enum print_line_t ret;
2695 if (iter->lost_events &&
2696 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2697 iter->cpu, iter->lost_events))
2698 return TRACE_TYPE_PARTIAL_LINE;
2700 if (iter->trace && iter->trace->print_line) {
2701 ret = iter->trace->print_line(iter);
2702 if (ret != TRACE_TYPE_UNHANDLED)
2706 if (iter->ent->type == TRACE_BPUTS &&
2707 trace_flags & TRACE_ITER_PRINTK &&
2708 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2709 return trace_print_bputs_msg_only(iter);
2711 if (iter->ent->type == TRACE_BPRINT &&
2712 trace_flags & TRACE_ITER_PRINTK &&
2713 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2714 return trace_print_bprintk_msg_only(iter);
2716 if (iter->ent->type == TRACE_PRINT &&
2717 trace_flags & TRACE_ITER_PRINTK &&
2718 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2719 return trace_print_printk_msg_only(iter);
2721 if (trace_flags & TRACE_ITER_BIN)
2722 return print_bin_fmt(iter);
2724 if (trace_flags & TRACE_ITER_HEX)
2725 return print_hex_fmt(iter);
2727 if (trace_flags & TRACE_ITER_RAW)
2728 return print_raw_fmt(iter);
2730 return print_trace_fmt(iter);
2733 void trace_latency_header(struct seq_file *m)
2735 struct trace_iterator *iter = m->private;
2737 /* print nothing if the buffers are empty */
2738 if (trace_empty(iter))
2741 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2742 print_trace_header(m, iter);
2744 if (!(trace_flags & TRACE_ITER_VERBOSE))
2745 print_lat_help_header(m);
2748 void trace_default_header(struct seq_file *m)
2750 struct trace_iterator *iter = m->private;
2752 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2755 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2756 /* print nothing if the buffers are empty */
2757 if (trace_empty(iter))
2759 print_trace_header(m, iter);
2760 if (!(trace_flags & TRACE_ITER_VERBOSE))
2761 print_lat_help_header(m);
2763 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2764 if (trace_flags & TRACE_ITER_IRQ_INFO)
2765 print_func_help_header_irq(iter->trace_buffer, m);
2767 print_func_help_header(iter->trace_buffer, m);
2772 static void test_ftrace_alive(struct seq_file *m)
2774 if (!ftrace_is_dead())
2776 seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2777 seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2780 #ifdef CONFIG_TRACER_MAX_TRACE
2781 static void show_snapshot_main_help(struct seq_file *m)
2783 seq_printf(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n");
2784 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2785 seq_printf(m, "# Takes a snapshot of the main buffer.\n");
2786 seq_printf(m, "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n");
2787 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2788 seq_printf(m, "# is not a '0' or '1')\n");
2791 static void show_snapshot_percpu_help(struct seq_file *m)
2793 seq_printf(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
2794 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
2795 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2796 seq_printf(m, "# Takes a snapshot of the main buffer for this cpu.\n");
2798 seq_printf(m, "# echo 1 > snapshot : Not supported with this kernel.\n");
2799 seq_printf(m, "# Must use main snapshot file to allocate.\n");
2801 seq_printf(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n");
2802 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2803 seq_printf(m, "# is not a '0' or '1')\n");
2806 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
2808 if (iter->tr->allocated_snapshot)
2809 seq_printf(m, "#\n# * Snapshot is allocated *\n#\n");
2811 seq_printf(m, "#\n# * Snapshot is freed *\n#\n");
2813 seq_printf(m, "# Snapshot commands:\n");
2814 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
2815 show_snapshot_main_help(m);
2817 show_snapshot_percpu_help(m);
2820 /* Should never be called */
2821 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
2824 static int s_show(struct seq_file *m, void *v)
2826 struct trace_iterator *iter = v;
2829 if (iter->ent == NULL) {
2831 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2833 test_ftrace_alive(m);
2835 if (iter->snapshot && trace_empty(iter))
2836 print_snapshot_help(m, iter);
2837 else if (iter->trace && iter->trace->print_header)
2838 iter->trace->print_header(m);
2840 trace_default_header(m);
2842 } else if (iter->leftover) {
2844 * If we filled the seq_file buffer earlier, we
2845 * want to just show it now.
2847 ret = trace_print_seq(m, &iter->seq);
2849 /* ret should this time be zero, but you never know */
2850 iter->leftover = ret;
2853 print_trace_line(iter);
2854 ret = trace_print_seq(m, &iter->seq);
2856 * If we overflow the seq_file buffer, then it will
2857 * ask us for this data again at start up.
2859 * ret is 0 if seq_file write succeeded.
2862 iter->leftover = ret;
2869 * Should be used after trace_array_get(), trace_types_lock
2870 * ensures that i_cdev was already initialized.
2872 static inline int tracing_get_cpu(struct inode *inode)
2874 if (inode->i_cdev) /* See trace_create_cpu_file() */
2875 return (long)inode->i_cdev - 1;
2876 return RING_BUFFER_ALL_CPUS;
2879 static const struct seq_operations tracer_seq_ops = {
2886 static struct trace_iterator *
2887 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
2889 struct trace_array *tr = inode->i_private;
2890 struct trace_iterator *iter;
2893 if (tracing_disabled)
2894 return ERR_PTR(-ENODEV);
2896 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
2898 return ERR_PTR(-ENOMEM);
2900 iter->buffer_iter = kzalloc(sizeof(*iter->buffer_iter) * num_possible_cpus(),
2902 if (!iter->buffer_iter)
2906 * We make a copy of the current tracer to avoid concurrent
2907 * changes on it while we are reading.
2909 mutex_lock(&trace_types_lock);
2910 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2914 *iter->trace = *tr->current_trace;
2916 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2921 #ifdef CONFIG_TRACER_MAX_TRACE
2922 /* Currently only the top directory has a snapshot */
2923 if (tr->current_trace->print_max || snapshot)
2924 iter->trace_buffer = &tr->max_buffer;
2927 iter->trace_buffer = &tr->trace_buffer;
2928 iter->snapshot = snapshot;
2930 iter->cpu_file = tracing_get_cpu(inode);
2931 mutex_init(&iter->mutex);
2933 /* Notify the tracer early; before we stop tracing. */
2934 if (iter->trace && iter->trace->open)
2935 iter->trace->open(iter);
2937 /* Annotate start of buffers if we had overruns */
2938 if (ring_buffer_overruns(iter->trace_buffer->buffer))
2939 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2941 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
2942 if (trace_clocks[tr->clock_id].in_ns)
2943 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
2945 /* stop the trace while dumping if we are not opening "snapshot" */
2946 if (!iter->snapshot)
2947 tracing_stop_tr(tr);
2949 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
2950 for_each_tracing_cpu(cpu) {
2951 iter->buffer_iter[cpu] =
2952 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
2954 ring_buffer_read_prepare_sync();
2955 for_each_tracing_cpu(cpu) {
2956 ring_buffer_read_start(iter->buffer_iter[cpu]);
2957 tracing_iter_reset(iter, cpu);
2960 cpu = iter->cpu_file;
2961 iter->buffer_iter[cpu] =
2962 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
2963 ring_buffer_read_prepare_sync();
2964 ring_buffer_read_start(iter->buffer_iter[cpu]);
2965 tracing_iter_reset(iter, cpu);
2968 mutex_unlock(&trace_types_lock);
2973 mutex_unlock(&trace_types_lock);
2975 kfree(iter->buffer_iter);
2977 seq_release_private(inode, file);
2978 return ERR_PTR(-ENOMEM);
2981 int tracing_open_generic(struct inode *inode, struct file *filp)
2983 if (tracing_disabled)
2986 filp->private_data = inode->i_private;
2991 * Open and update trace_array ref count.
2992 * Must have the current trace_array passed to it.
2994 static int tracing_open_generic_tr(struct inode *inode, struct file *filp)
2996 struct trace_array *tr = inode->i_private;
2998 if (tracing_disabled)
3001 if (trace_array_get(tr) < 0)
3004 filp->private_data = inode->i_private;
3009 static int tracing_release(struct inode *inode, struct file *file)
3011 struct trace_array *tr = inode->i_private;
3012 struct seq_file *m = file->private_data;
3013 struct trace_iterator *iter;
3016 if (!(file->f_mode & FMODE_READ)) {
3017 trace_array_put(tr);
3021 /* Writes do not use seq_file */
3023 mutex_lock(&trace_types_lock);
3025 for_each_tracing_cpu(cpu) {
3026 if (iter->buffer_iter[cpu])
3027 ring_buffer_read_finish(iter->buffer_iter[cpu]);
3030 if (iter->trace && iter->trace->close)
3031 iter->trace->close(iter);
3033 if (!iter->snapshot)
3034 /* reenable tracing if it was previously enabled */
3035 tracing_start_tr(tr);
3037 __trace_array_put(tr);
3039 mutex_unlock(&trace_types_lock);
3041 mutex_destroy(&iter->mutex);
3042 free_cpumask_var(iter->started);
3044 kfree(iter->buffer_iter);
3045 seq_release_private(inode, file);
3050 static int tracing_release_generic_tr(struct inode *inode, struct file *file)
3052 struct trace_array *tr = inode->i_private;
3054 trace_array_put(tr);
3058 static int tracing_single_release_tr(struct inode *inode, struct file *file)
3060 struct trace_array *tr = inode->i_private;
3062 trace_array_put(tr);
3064 return single_release(inode, file);
3067 static int tracing_open(struct inode *inode, struct file *file)
3069 struct trace_array *tr = inode->i_private;
3070 struct trace_iterator *iter;
3073 if (trace_array_get(tr) < 0)
3076 /* If this file was open for write, then erase contents */
3077 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
3078 int cpu = tracing_get_cpu(inode);
3080 if (cpu == RING_BUFFER_ALL_CPUS)
3081 tracing_reset_online_cpus(&tr->trace_buffer);
3083 tracing_reset(&tr->trace_buffer, cpu);
3086 if (file->f_mode & FMODE_READ) {
3087 iter = __tracing_open(inode, file, false);
3089 ret = PTR_ERR(iter);
3090 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
3091 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3095 trace_array_put(tr);
3101 t_next(struct seq_file *m, void *v, loff_t *pos)
3103 struct tracer *t = v;
3113 static void *t_start(struct seq_file *m, loff_t *pos)
3118 mutex_lock(&trace_types_lock);
3119 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
3125 static void t_stop(struct seq_file *m, void *p)
3127 mutex_unlock(&trace_types_lock);
3130 static int t_show(struct seq_file *m, void *v)
3132 struct tracer *t = v;
3137 seq_printf(m, "%s", t->name);
3146 static const struct seq_operations show_traces_seq_ops = {
3153 static int show_traces_open(struct inode *inode, struct file *file)
3155 if (tracing_disabled)
3158 return seq_open(file, &show_traces_seq_ops);
3162 tracing_write_stub(struct file *filp, const char __user *ubuf,
3163 size_t count, loff_t *ppos)
3168 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
3170 if (file->f_mode & FMODE_READ)
3171 return seq_lseek(file, offset, origin);
3176 static const struct file_operations tracing_fops = {
3177 .open = tracing_open,
3179 .write = tracing_write_stub,
3180 .llseek = tracing_seek,
3181 .release = tracing_release,
3184 static const struct file_operations show_traces_fops = {
3185 .open = show_traces_open,
3187 .release = seq_release,
3188 .llseek = seq_lseek,
3192 * The tracer itself will not take this lock, but still we want
3193 * to provide a consistent cpumask to user-space:
3195 static DEFINE_MUTEX(tracing_cpumask_update_lock);
3198 * Temporary storage for the character representation of the
3199 * CPU bitmask (and one more byte for the newline):
3201 static char mask_str[NR_CPUS + 1];
3204 tracing_cpumask_read(struct file *filp, char __user *ubuf,
3205 size_t count, loff_t *ppos)
3207 struct trace_array *tr = file_inode(filp)->i_private;
3210 mutex_lock(&tracing_cpumask_update_lock);
3212 len = cpumask_scnprintf(mask_str, count, tr->tracing_cpumask);
3213 if (count - len < 2) {
3217 len += sprintf(mask_str + len, "\n");
3218 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
3221 mutex_unlock(&tracing_cpumask_update_lock);
3227 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
3228 size_t count, loff_t *ppos)
3230 struct trace_array *tr = file_inode(filp)->i_private;
3231 cpumask_var_t tracing_cpumask_new;
3234 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
3237 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
3241 mutex_lock(&tracing_cpumask_update_lock);
3243 local_irq_disable();
3244 arch_spin_lock(&ftrace_max_lock);
3245 for_each_tracing_cpu(cpu) {
3247 * Increase/decrease the disabled counter if we are
3248 * about to flip a bit in the cpumask:
3250 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
3251 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3252 atomic_inc(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3253 ring_buffer_record_disable_cpu(tr->trace_buffer.buffer, cpu);
3255 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
3256 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3257 atomic_dec(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3258 ring_buffer_record_enable_cpu(tr->trace_buffer.buffer, cpu);
3261 arch_spin_unlock(&ftrace_max_lock);
3264 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
3266 mutex_unlock(&tracing_cpumask_update_lock);
3267 free_cpumask_var(tracing_cpumask_new);
3272 free_cpumask_var(tracing_cpumask_new);
3277 static const struct file_operations tracing_cpumask_fops = {
3278 .open = tracing_open_generic_tr,
3279 .read = tracing_cpumask_read,
3280 .write = tracing_cpumask_write,
3281 .release = tracing_release_generic_tr,
3282 .llseek = generic_file_llseek,
3285 static int tracing_trace_options_show(struct seq_file *m, void *v)
3287 struct tracer_opt *trace_opts;
3288 struct trace_array *tr = m->private;
3292 mutex_lock(&trace_types_lock);
3293 tracer_flags = tr->current_trace->flags->val;
3294 trace_opts = tr->current_trace->flags->opts;
3296 for (i = 0; trace_options[i]; i++) {
3297 if (trace_flags & (1 << i))
3298 seq_printf(m, "%s\n", trace_options[i]);
3300 seq_printf(m, "no%s\n", trace_options[i]);
3303 for (i = 0; trace_opts[i].name; i++) {
3304 if (tracer_flags & trace_opts[i].bit)
3305 seq_printf(m, "%s\n", trace_opts[i].name);
3307 seq_printf(m, "no%s\n", trace_opts[i].name);
3309 mutex_unlock(&trace_types_lock);
3314 static int __set_tracer_option(struct tracer *trace,
3315 struct tracer_flags *tracer_flags,
3316 struct tracer_opt *opts, int neg)
3320 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
3325 tracer_flags->val &= ~opts->bit;
3327 tracer_flags->val |= opts->bit;
3331 /* Try to assign a tracer specific option */
3332 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
3334 struct tracer_flags *tracer_flags = trace->flags;
3335 struct tracer_opt *opts = NULL;
3338 for (i = 0; tracer_flags->opts[i].name; i++) {
3339 opts = &tracer_flags->opts[i];
3341 if (strcmp(cmp, opts->name) == 0)
3342 return __set_tracer_option(trace, trace->flags,
3349 /* Some tracers require overwrite to stay enabled */
3350 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
3352 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
3358 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
3360 /* do nothing if flag is already set */
3361 if (!!(trace_flags & mask) == !!enabled)
3364 /* Give the tracer a chance to approve the change */
3365 if (tr->current_trace->flag_changed)
3366 if (tr->current_trace->flag_changed(tr->current_trace, mask, !!enabled))
3370 trace_flags |= mask;
3372 trace_flags &= ~mask;
3374 if (mask == TRACE_ITER_RECORD_CMD)
3375 trace_event_enable_cmd_record(enabled);
3377 if (mask == TRACE_ITER_OVERWRITE) {
3378 ring_buffer_change_overwrite(tr->trace_buffer.buffer, enabled);
3379 #ifdef CONFIG_TRACER_MAX_TRACE
3380 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
3384 if (mask == TRACE_ITER_PRINTK)
3385 trace_printk_start_stop_comm(enabled);
3390 static int trace_set_options(struct trace_array *tr, char *option)
3397 cmp = strstrip(option);
3399 if (strncmp(cmp, "no", 2) == 0) {
3404 mutex_lock(&trace_types_lock);
3406 for (i = 0; trace_options[i]; i++) {
3407 if (strcmp(cmp, trace_options[i]) == 0) {
3408 ret = set_tracer_flag(tr, 1 << i, !neg);
3413 /* If no option could be set, test the specific tracer options */
3414 if (!trace_options[i])
3415 ret = set_tracer_option(tr->current_trace, cmp, neg);
3417 mutex_unlock(&trace_types_lock);
3423 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
3424 size_t cnt, loff_t *ppos)
3426 struct seq_file *m = filp->private_data;
3427 struct trace_array *tr = m->private;
3431 if (cnt >= sizeof(buf))
3434 if (copy_from_user(&buf, ubuf, cnt))
3439 ret = trace_set_options(tr, buf);
3448 static int tracing_trace_options_open(struct inode *inode, struct file *file)
3450 struct trace_array *tr = inode->i_private;
3453 if (tracing_disabled)
3456 if (trace_array_get(tr) < 0)
3459 ret = single_open(file, tracing_trace_options_show, inode->i_private);
3461 trace_array_put(tr);
3466 static const struct file_operations tracing_iter_fops = {
3467 .open = tracing_trace_options_open,
3469 .llseek = seq_lseek,
3470 .release = tracing_single_release_tr,
3471 .write = tracing_trace_options_write,
3474 static const char readme_msg[] =
3475 "tracing mini-HOWTO:\n\n"
3476 "# echo 0 > tracing_on : quick way to disable tracing\n"
3477 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
3478 " Important files:\n"
3479 " trace\t\t\t- The static contents of the buffer\n"
3480 "\t\t\t To clear the buffer write into this file: echo > trace\n"
3481 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
3482 " current_tracer\t- function and latency tracers\n"
3483 " available_tracers\t- list of configured tracers for current_tracer\n"
3484 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
3485 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
3486 " trace_clock\t\t-change the clock used to order events\n"
3487 " local: Per cpu clock but may not be synced across CPUs\n"
3488 " global: Synced across CPUs but slows tracing down.\n"
3489 " counter: Not a clock, but just an increment\n"
3490 " uptime: Jiffy counter from time of boot\n"
3491 " perf: Same clock that perf events use\n"
3492 #ifdef CONFIG_X86_64
3493 " x86-tsc: TSC cycle counter\n"
3495 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
3496 " tracing_cpumask\t- Limit which CPUs to trace\n"
3497 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
3498 "\t\t\t Remove sub-buffer with rmdir\n"
3499 " trace_options\t\t- Set format or modify how tracing happens\n"
3500 "\t\t\t Disable an option by adding a suffix 'no' to the option name\n"
3501 #ifdef CONFIG_DYNAMIC_FTRACE
3502 "\n available_filter_functions - list of functions that can be filtered on\n"
3503 " set_ftrace_filter\t- echo function name in here to only trace these functions\n"
3504 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3505 " modules: Can select a group via module\n"
3506 " Format: :mod:<module-name>\n"
3507 " example: echo :mod:ext3 > set_ftrace_filter\n"
3508 " triggers: a command to perform when function is hit\n"
3509 " Format: <function>:<trigger>[:count]\n"
3510 " trigger: traceon, traceoff\n"
3511 " enable_event:<system>:<event>\n"
3512 " disable_event:<system>:<event>\n"
3513 #ifdef CONFIG_STACKTRACE
3516 #ifdef CONFIG_TRACER_SNAPSHOT
3519 " example: echo do_fault:traceoff > set_ftrace_filter\n"
3520 " echo do_trap:traceoff:3 > set_ftrace_filter\n"
3521 " The first one will disable tracing every time do_fault is hit\n"
3522 " The second will disable tracing at most 3 times when do_trap is hit\n"
3523 " The first time do trap is hit and it disables tracing, the counter\n"
3524 " will decrement to 2. If tracing is already disabled, the counter\n"
3525 " will not decrement. It only decrements when the trigger did work\n"
3526 " To remove trigger without count:\n"
3527 " echo '!<function>:<trigger> > set_ftrace_filter\n"
3528 " To remove trigger with a count:\n"
3529 " echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
3530 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
3531 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3532 " modules: Can select a group via module command :mod:\n"
3533 " Does not accept triggers\n"
3534 #endif /* CONFIG_DYNAMIC_FTRACE */
3535 #ifdef CONFIG_FUNCTION_TRACER
3536 " set_ftrace_pid\t- Write pid(s) to only function trace those pids (function)\n"
3538 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3539 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
3540 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
3542 #ifdef CONFIG_TRACER_SNAPSHOT
3543 "\n snapshot\t\t- Like 'trace' but shows the content of the static snapshot buffer\n"
3544 "\t\t\t Read the contents for more information\n"
3546 #ifdef CONFIG_STACK_TRACER
3547 " stack_trace\t\t- Shows the max stack trace when active\n"
3548 " stack_max_size\t- Shows current max stack size that was traced\n"
3549 "\t\t\t Write into this file to reset the max size (trigger a new trace)\n"
3550 #ifdef CONFIG_DYNAMIC_FTRACE
3551 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace traces\n"
3553 #endif /* CONFIG_STACK_TRACER */
3557 tracing_readme_read(struct file *filp, char __user *ubuf,
3558 size_t cnt, loff_t *ppos)
3560 return simple_read_from_buffer(ubuf, cnt, ppos,
3561 readme_msg, strlen(readme_msg));
3564 static const struct file_operations tracing_readme_fops = {
3565 .open = tracing_open_generic,
3566 .read = tracing_readme_read,
3567 .llseek = generic_file_llseek,
3571 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
3572 size_t cnt, loff_t *ppos)
3581 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
3585 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
3593 for (i = 0; i < SAVED_CMDLINES; i++) {
3596 pid = map_cmdline_to_pid[i];
3597 if (pid == -1 || pid == NO_CMDLINE_MAP)
3600 trace_find_cmdline(pid, buf_comm);
3601 r = sprintf(buf, "%d %s\n", pid, buf_comm);
3606 len = simple_read_from_buffer(ubuf, cnt, ppos,
3615 static const struct file_operations tracing_saved_cmdlines_fops = {
3616 .open = tracing_open_generic,
3617 .read = tracing_saved_cmdlines_read,
3618 .llseek = generic_file_llseek,
3622 tracing_set_trace_read(struct file *filp, char __user *ubuf,
3623 size_t cnt, loff_t *ppos)
3625 struct trace_array *tr = filp->private_data;
3626 char buf[MAX_TRACER_SIZE+2];
3629 mutex_lock(&trace_types_lock);
3630 r = sprintf(buf, "%s\n", tr->current_trace->name);
3631 mutex_unlock(&trace_types_lock);
3633 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3636 int tracer_init(struct tracer *t, struct trace_array *tr)
3638 tracing_reset_online_cpus(&tr->trace_buffer);
3642 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val)
3646 for_each_tracing_cpu(cpu)
3647 per_cpu_ptr(buf->data, cpu)->entries = val;
3650 #ifdef CONFIG_TRACER_MAX_TRACE
3651 /* resize @tr's buffer to the size of @size_tr's entries */
3652 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
3653 struct trace_buffer *size_buf, int cpu_id)
3657 if (cpu_id == RING_BUFFER_ALL_CPUS) {
3658 for_each_tracing_cpu(cpu) {
3659 ret = ring_buffer_resize(trace_buf->buffer,
3660 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
3663 per_cpu_ptr(trace_buf->data, cpu)->entries =
3664 per_cpu_ptr(size_buf->data, cpu)->entries;
3667 ret = ring_buffer_resize(trace_buf->buffer,
3668 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
3670 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
3671 per_cpu_ptr(size_buf->data, cpu_id)->entries;
3676 #endif /* CONFIG_TRACER_MAX_TRACE */
3678 static int __tracing_resize_ring_buffer(struct trace_array *tr,
3679 unsigned long size, int cpu)
3684 * If kernel or user changes the size of the ring buffer
3685 * we use the size that was given, and we can forget about
3686 * expanding it later.
3688 ring_buffer_expanded = true;
3690 /* May be called before buffers are initialized */
3691 if (!tr->trace_buffer.buffer)
3694 ret = ring_buffer_resize(tr->trace_buffer.buffer, size, cpu);
3698 #ifdef CONFIG_TRACER_MAX_TRACE
3699 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
3700 !tr->current_trace->use_max_tr)
3703 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
3705 int r = resize_buffer_duplicate_size(&tr->trace_buffer,
3706 &tr->trace_buffer, cpu);
3709 * AARGH! We are left with different
3710 * size max buffer!!!!
3711 * The max buffer is our "snapshot" buffer.
3712 * When a tracer needs a snapshot (one of the
3713 * latency tracers), it swaps the max buffer
3714 * with the saved snap shot. We succeeded to
3715 * update the size of the main buffer, but failed to
3716 * update the size of the max buffer. But when we tried
3717 * to reset the main buffer to the original size, we
3718 * failed there too. This is very unlikely to
3719 * happen, but if it does, warn and kill all
3723 tracing_disabled = 1;
3728 if (cpu == RING_BUFFER_ALL_CPUS)
3729 set_buffer_entries(&tr->max_buffer, size);
3731 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
3734 #endif /* CONFIG_TRACER_MAX_TRACE */
3736 if (cpu == RING_BUFFER_ALL_CPUS)
3737 set_buffer_entries(&tr->trace_buffer, size);
3739 per_cpu_ptr(tr->trace_buffer.data, cpu)->entries = size;
3744 static ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
3745 unsigned long size, int cpu_id)
3749 mutex_lock(&trace_types_lock);
3751 if (cpu_id != RING_BUFFER_ALL_CPUS) {
3752 /* make sure, this cpu is enabled in the mask */
3753 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
3759 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
3764 mutex_unlock(&trace_types_lock);
3771 * tracing_update_buffers - used by tracing facility to expand ring buffers
3773 * To save on memory when the tracing is never used on a system with it
3774 * configured in. The ring buffers are set to a minimum size. But once
3775 * a user starts to use the tracing facility, then they need to grow
3776 * to their default size.
3778 * This function is to be called when a tracer is about to be used.
3780 int tracing_update_buffers(void)
3784 mutex_lock(&trace_types_lock);
3785 if (!ring_buffer_expanded)
3786 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
3787 RING_BUFFER_ALL_CPUS);
3788 mutex_unlock(&trace_types_lock);
3793 struct trace_option_dentry;
3795 static struct trace_option_dentry *
3796 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
3799 destroy_trace_option_files(struct trace_option_dentry *topts);
3801 static int tracing_set_tracer(const char *buf)
3803 static struct trace_option_dentry *topts;
3804 struct trace_array *tr = &global_trace;
3806 #ifdef CONFIG_TRACER_MAX_TRACE
3811 mutex_lock(&trace_types_lock);
3813 if (!ring_buffer_expanded) {
3814 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
3815 RING_BUFFER_ALL_CPUS);
3821 for (t = trace_types; t; t = t->next) {
3822 if (strcmp(t->name, buf) == 0)
3829 if (t == tr->current_trace)
3832 trace_branch_disable();
3834 tr->current_trace->enabled = false;
3836 if (tr->current_trace->reset)
3837 tr->current_trace->reset(tr);
3839 /* Current trace needs to be nop_trace before synchronize_sched */
3840 tr->current_trace = &nop_trace;
3842 #ifdef CONFIG_TRACER_MAX_TRACE
3843 had_max_tr = tr->allocated_snapshot;
3845 if (had_max_tr && !t->use_max_tr) {
3847 * We need to make sure that the update_max_tr sees that
3848 * current_trace changed to nop_trace to keep it from
3849 * swapping the buffers after we resize it.
3850 * The update_max_tr is called from interrupts disabled
3851 * so a synchronized_sched() is sufficient.
3853 synchronize_sched();
3857 destroy_trace_option_files(topts);
3859 topts = create_trace_option_files(tr, t);
3861 #ifdef CONFIG_TRACER_MAX_TRACE
3862 if (t->use_max_tr && !had_max_tr) {
3863 ret = alloc_snapshot(tr);
3870 ret = tracer_init(t, tr);
3875 tr->current_trace = t;
3876 tr->current_trace->enabled = true;
3877 trace_branch_enable(tr);
3879 mutex_unlock(&trace_types_lock);
3885 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3886 size_t cnt, loff_t *ppos)
3888 char buf[MAX_TRACER_SIZE+1];
3895 if (cnt > MAX_TRACER_SIZE)
3896 cnt = MAX_TRACER_SIZE;
3898 if (copy_from_user(&buf, ubuf, cnt))
3903 /* strip ending whitespace. */
3904 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3907 err = tracing_set_tracer(buf);
3917 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3918 size_t cnt, loff_t *ppos)
3920 unsigned long *ptr = filp->private_data;
3924 r = snprintf(buf, sizeof(buf), "%ld\n",
3925 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3926 if (r > sizeof(buf))
3928 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3932 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3933 size_t cnt, loff_t *ppos)
3935 unsigned long *ptr = filp->private_data;
3939 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3948 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3950 struct trace_array *tr = inode->i_private;
3951 struct trace_iterator *iter;
3954 if (tracing_disabled)
3957 if (trace_array_get(tr) < 0)
3960 mutex_lock(&trace_types_lock);
3962 /* create a buffer to store the information to pass to userspace */
3963 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3966 __trace_array_put(tr);
3971 * We make a copy of the current tracer to avoid concurrent
3972 * changes on it while we are reading.
3974 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3979 *iter->trace = *tr->current_trace;
3981 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3986 /* trace pipe does not show start of buffer */
3987 cpumask_setall(iter->started);
3989 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3990 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3992 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3993 if (trace_clocks[tr->clock_id].in_ns)
3994 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
3997 iter->trace_buffer = &tr->trace_buffer;
3998 iter->cpu_file = tracing_get_cpu(inode);
3999 mutex_init(&iter->mutex);
4000 filp->private_data = iter;
4002 if (iter->trace->pipe_open)
4003 iter->trace->pipe_open(iter);
4005 nonseekable_open(inode, filp);
4007 mutex_unlock(&trace_types_lock);
4013 __trace_array_put(tr);
4014 mutex_unlock(&trace_types_lock);
4018 static int tracing_release_pipe(struct inode *inode, struct file *file)
4020 struct trace_iterator *iter = file->private_data;
4021 struct trace_array *tr = inode->i_private;
4023 mutex_lock(&trace_types_lock);
4025 if (iter->trace->pipe_close)
4026 iter->trace->pipe_close(iter);
4028 mutex_unlock(&trace_types_lock);
4030 free_cpumask_var(iter->started);
4031 mutex_destroy(&iter->mutex);
4035 trace_array_put(tr);
4041 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
4043 /* Iterators are static, they should be filled or empty */
4044 if (trace_buffer_iter(iter, iter->cpu_file))
4045 return POLLIN | POLLRDNORM;
4047 if (trace_flags & TRACE_ITER_BLOCK)
4049 * Always select as readable when in blocking mode
4051 return POLLIN | POLLRDNORM;
4053 return ring_buffer_poll_wait(iter->trace_buffer->buffer, iter->cpu_file,
4058 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
4060 struct trace_iterator *iter = filp->private_data;
4062 return trace_poll(iter, filp, poll_table);
4066 * This is a make-shift waitqueue.
4067 * A tracer might use this callback on some rare cases:
4069 * 1) the current tracer might hold the runqueue lock when it wakes up
4070 * a reader, hence a deadlock (sched, function, and function graph tracers)
4071 * 2) the function tracers, trace all functions, we don't want
4072 * the overhead of calling wake_up and friends
4073 * (and tracing them too)
4075 * Anyway, this is really very primitive wakeup.
4077 void poll_wait_pipe(struct trace_iterator *iter)
4079 set_current_state(TASK_INTERRUPTIBLE);
4080 /* sleep for 100 msecs, and try again. */
4081 schedule_timeout(HZ / 10);
4084 /* Must be called with trace_types_lock mutex held. */
4085 static int tracing_wait_pipe(struct file *filp)
4087 struct trace_iterator *iter = filp->private_data;
4089 while (trace_empty(iter)) {
4091 if ((filp->f_flags & O_NONBLOCK)) {
4095 mutex_unlock(&iter->mutex);
4097 iter->trace->wait_pipe(iter);
4099 mutex_lock(&iter->mutex);
4101 if (signal_pending(current))
4105 * We block until we read something and tracing is disabled.
4106 * We still block if tracing is disabled, but we have never
4107 * read anything. This allows a user to cat this file, and
4108 * then enable tracing. But after we have read something,
4109 * we give an EOF when tracing is again disabled.
4111 * iter->pos will be 0 if we haven't read anything.
4113 if (!tracing_is_on() && iter->pos)
4124 tracing_read_pipe(struct file *filp, char __user *ubuf,
4125 size_t cnt, loff_t *ppos)
4127 struct trace_iterator *iter = filp->private_data;
4128 struct trace_array *tr = iter->tr;
4131 /* return any leftover data */
4132 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4136 trace_seq_init(&iter->seq);
4138 /* copy the tracer to avoid using a global lock all around */
4139 mutex_lock(&trace_types_lock);
4140 if (unlikely(iter->trace->name != tr->current_trace->name))
4141 *iter->trace = *tr->current_trace;
4142 mutex_unlock(&trace_types_lock);
4145 * Avoid more than one consumer on a single file descriptor
4146 * This is just a matter of traces coherency, the ring buffer itself
4149 mutex_lock(&iter->mutex);
4150 if (iter->trace->read) {
4151 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
4157 sret = tracing_wait_pipe(filp);
4161 /* stop when tracing is finished */
4162 if (trace_empty(iter)) {
4167 if (cnt >= PAGE_SIZE)
4168 cnt = PAGE_SIZE - 1;
4170 /* reset all but tr, trace, and overruns */
4171 memset(&iter->seq, 0,
4172 sizeof(struct trace_iterator) -
4173 offsetof(struct trace_iterator, seq));
4174 cpumask_clear(iter->started);
4177 trace_event_read_lock();
4178 trace_access_lock(iter->cpu_file);
4179 while (trace_find_next_entry_inc(iter) != NULL) {
4180 enum print_line_t ret;
4181 int len = iter->seq.len;
4183 ret = print_trace_line(iter);
4184 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4185 /* don't print partial lines */
4186 iter->seq.len = len;
4189 if (ret != TRACE_TYPE_NO_CONSUME)
4190 trace_consume(iter);
4192 if (iter->seq.len >= cnt)
4196 * Setting the full flag means we reached the trace_seq buffer
4197 * size and we should leave by partial output condition above.
4198 * One of the trace_seq_* functions is not used properly.
4200 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
4203 trace_access_unlock(iter->cpu_file);
4204 trace_event_read_unlock();
4206 /* Now copy what we have to the user */
4207 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4208 if (iter->seq.readpos >= iter->seq.len)
4209 trace_seq_init(&iter->seq);
4212 * If there was nothing to send to user, in spite of consuming trace
4213 * entries, go back to wait for more entries.
4219 mutex_unlock(&iter->mutex);
4224 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
4225 struct pipe_buffer *buf)
4227 __free_page(buf->page);
4230 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
4233 __free_page(spd->pages[idx]);
4236 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
4238 .map = generic_pipe_buf_map,
4239 .unmap = generic_pipe_buf_unmap,
4240 .confirm = generic_pipe_buf_confirm,
4241 .release = tracing_pipe_buf_release,
4242 .steal = generic_pipe_buf_steal,
4243 .get = generic_pipe_buf_get,
4247 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
4252 /* Seq buffer is page-sized, exactly what we need. */
4254 count = iter->seq.len;
4255 ret = print_trace_line(iter);
4256 count = iter->seq.len - count;
4259 iter->seq.len -= count;
4262 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4263 iter->seq.len -= count;
4267 if (ret != TRACE_TYPE_NO_CONSUME)
4268 trace_consume(iter);
4270 if (!trace_find_next_entry_inc(iter)) {
4280 static ssize_t tracing_splice_read_pipe(struct file *filp,
4282 struct pipe_inode_info *pipe,
4286 struct page *pages_def[PIPE_DEF_BUFFERS];
4287 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4288 struct trace_iterator *iter = filp->private_data;
4289 struct splice_pipe_desc spd = {
4291 .partial = partial_def,
4292 .nr_pages = 0, /* This gets updated below. */
4293 .nr_pages_max = PIPE_DEF_BUFFERS,
4295 .ops = &tracing_pipe_buf_ops,
4296 .spd_release = tracing_spd_release_pipe,
4298 struct trace_array *tr = iter->tr;
4303 if (splice_grow_spd(pipe, &spd))
4306 /* copy the tracer to avoid using a global lock all around */
4307 mutex_lock(&trace_types_lock);
4308 if (unlikely(iter->trace->name != tr->current_trace->name))
4309 *iter->trace = *tr->current_trace;
4310 mutex_unlock(&trace_types_lock);
4312 mutex_lock(&iter->mutex);
4314 if (iter->trace->splice_read) {
4315 ret = iter->trace->splice_read(iter, filp,
4316 ppos, pipe, len, flags);
4321 ret = tracing_wait_pipe(filp);
4325 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
4330 trace_event_read_lock();
4331 trace_access_lock(iter->cpu_file);
4333 /* Fill as many pages as possible. */
4334 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
4335 spd.pages[i] = alloc_page(GFP_KERNEL);
4339 rem = tracing_fill_pipe_page(rem, iter);
4341 /* Copy the data into the page, so we can start over. */
4342 ret = trace_seq_to_buffer(&iter->seq,
4343 page_address(spd.pages[i]),
4346 __free_page(spd.pages[i]);
4349 spd.partial[i].offset = 0;
4350 spd.partial[i].len = iter->seq.len;
4352 trace_seq_init(&iter->seq);
4355 trace_access_unlock(iter->cpu_file);
4356 trace_event_read_unlock();
4357 mutex_unlock(&iter->mutex);
4361 ret = splice_to_pipe(pipe, &spd);
4363 splice_shrink_spd(&spd);
4367 mutex_unlock(&iter->mutex);
4372 tracing_entries_read(struct file *filp, char __user *ubuf,
4373 size_t cnt, loff_t *ppos)
4375 struct inode *inode = file_inode(filp);
4376 struct trace_array *tr = inode->i_private;
4377 int cpu = tracing_get_cpu(inode);
4382 mutex_lock(&trace_types_lock);
4384 if (cpu == RING_BUFFER_ALL_CPUS) {
4385 int cpu, buf_size_same;
4390 /* check if all cpu sizes are same */
4391 for_each_tracing_cpu(cpu) {
4392 /* fill in the size from first enabled cpu */
4394 size = per_cpu_ptr(tr->trace_buffer.data, cpu)->entries;
4395 if (size != per_cpu_ptr(tr->trace_buffer.data, cpu)->entries) {
4401 if (buf_size_same) {
4402 if (!ring_buffer_expanded)
4403 r = sprintf(buf, "%lu (expanded: %lu)\n",
4405 trace_buf_size >> 10);
4407 r = sprintf(buf, "%lu\n", size >> 10);
4409 r = sprintf(buf, "X\n");
4411 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10);
4413 mutex_unlock(&trace_types_lock);
4415 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4420 tracing_entries_write(struct file *filp, const char __user *ubuf,
4421 size_t cnt, loff_t *ppos)
4423 struct inode *inode = file_inode(filp);
4424 struct trace_array *tr = inode->i_private;
4428 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4432 /* must have at least 1 entry */
4436 /* value is in KB */
4438 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
4448 tracing_total_entries_read(struct file *filp, char __user *ubuf,
4449 size_t cnt, loff_t *ppos)
4451 struct trace_array *tr = filp->private_data;
4454 unsigned long size = 0, expanded_size = 0;
4456 mutex_lock(&trace_types_lock);
4457 for_each_tracing_cpu(cpu) {
4458 size += per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10;
4459 if (!ring_buffer_expanded)
4460 expanded_size += trace_buf_size >> 10;
4462 if (ring_buffer_expanded)
4463 r = sprintf(buf, "%lu\n", size);
4465 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
4466 mutex_unlock(&trace_types_lock);
4468 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4472 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
4473 size_t cnt, loff_t *ppos)
4476 * There is no need to read what the user has written, this function
4477 * is just to make sure that there is no error when "echo" is used
4486 tracing_free_buffer_release(struct inode *inode, struct file *filp)
4488 struct trace_array *tr = inode->i_private;
4490 /* disable tracing ? */
4491 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
4492 tracer_tracing_off(tr);
4493 /* resize the ring buffer to 0 */
4494 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
4496 trace_array_put(tr);
4502 tracing_mark_write(struct file *filp, const char __user *ubuf,
4503 size_t cnt, loff_t *fpos)
4505 unsigned long addr = (unsigned long)ubuf;
4506 struct trace_array *tr = filp->private_data;
4507 struct ring_buffer_event *event;
4508 struct ring_buffer *buffer;
4509 struct print_entry *entry;
4510 unsigned long irq_flags;
4511 struct page *pages[2];
4521 if (tracing_disabled)
4524 if (!(trace_flags & TRACE_ITER_MARKERS))
4527 if (cnt > TRACE_BUF_SIZE)
4528 cnt = TRACE_BUF_SIZE;
4531 * Userspace is injecting traces into the kernel trace buffer.
4532 * We want to be as non intrusive as possible.
4533 * To do so, we do not want to allocate any special buffers
4534 * or take any locks, but instead write the userspace data
4535 * straight into the ring buffer.
4537 * First we need to pin the userspace buffer into memory,
4538 * which, most likely it is, because it just referenced it.
4539 * But there's no guarantee that it is. By using get_user_pages_fast()
4540 * and kmap_atomic/kunmap_atomic() we can get access to the
4541 * pages directly. We then write the data directly into the
4544 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
4546 /* check if we cross pages */
4547 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
4550 offset = addr & (PAGE_SIZE - 1);
4553 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
4554 if (ret < nr_pages) {
4556 put_page(pages[ret]);
4561 for (i = 0; i < nr_pages; i++)
4562 map_page[i] = kmap_atomic(pages[i]);
4564 local_save_flags(irq_flags);
4565 size = sizeof(*entry) + cnt + 2; /* possible \n added */
4566 buffer = tr->trace_buffer.buffer;
4567 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
4568 irq_flags, preempt_count());
4570 /* Ring buffer disabled, return as if not open for write */
4575 entry = ring_buffer_event_data(event);
4576 entry->ip = _THIS_IP_;
4578 if (nr_pages == 2) {
4579 len = PAGE_SIZE - offset;
4580 memcpy(&entry->buf, map_page[0] + offset, len);
4581 memcpy(&entry->buf[len], map_page[1], cnt - len);
4583 memcpy(&entry->buf, map_page[0] + offset, cnt);
4585 if (entry->buf[cnt - 1] != '\n') {
4586 entry->buf[cnt] = '\n';
4587 entry->buf[cnt + 1] = '\0';
4589 entry->buf[cnt] = '\0';
4591 __buffer_unlock_commit(buffer, event);
4598 for (i = 0; i < nr_pages; i++){
4599 kunmap_atomic(map_page[i]);
4606 static int tracing_clock_show(struct seq_file *m, void *v)
4608 struct trace_array *tr = m->private;
4611 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
4613 "%s%s%s%s", i ? " " : "",
4614 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
4615 i == tr->clock_id ? "]" : "");
4621 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
4622 size_t cnt, loff_t *fpos)
4624 struct seq_file *m = filp->private_data;
4625 struct trace_array *tr = m->private;
4627 const char *clockstr;
4630 if (cnt >= sizeof(buf))
4633 if (copy_from_user(&buf, ubuf, cnt))
4638 clockstr = strstrip(buf);
4640 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
4641 if (strcmp(trace_clocks[i].name, clockstr) == 0)
4644 if (i == ARRAY_SIZE(trace_clocks))
4647 mutex_lock(&trace_types_lock);
4651 ring_buffer_set_clock(tr->trace_buffer.buffer, trace_clocks[i].func);
4654 * New clock may not be consistent with the previous clock.
4655 * Reset the buffer so that it doesn't have incomparable timestamps.
4657 tracing_reset_online_cpus(&tr->trace_buffer);
4659 #ifdef CONFIG_TRACER_MAX_TRACE
4660 if (tr->flags & TRACE_ARRAY_FL_GLOBAL && tr->max_buffer.buffer)
4661 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
4662 tracing_reset_online_cpus(&tr->max_buffer);
4665 mutex_unlock(&trace_types_lock);
4672 static int tracing_clock_open(struct inode *inode, struct file *file)
4674 struct trace_array *tr = inode->i_private;
4677 if (tracing_disabled)
4680 if (trace_array_get(tr))
4683 ret = single_open(file, tracing_clock_show, inode->i_private);
4685 trace_array_put(tr);
4690 struct ftrace_buffer_info {
4691 struct trace_iterator iter;
4696 #ifdef CONFIG_TRACER_SNAPSHOT
4697 static int tracing_snapshot_open(struct inode *inode, struct file *file)
4699 struct trace_array *tr = inode->i_private;
4700 struct trace_iterator *iter;
4704 if (trace_array_get(tr) < 0)
4707 if (file->f_mode & FMODE_READ) {
4708 iter = __tracing_open(inode, file, true);
4710 ret = PTR_ERR(iter);
4712 /* Writes still need the seq_file to hold the private data */
4714 m = kzalloc(sizeof(*m), GFP_KERNEL);
4717 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4725 iter->trace_buffer = &tr->max_buffer;
4726 iter->cpu_file = tracing_get_cpu(inode);
4728 file->private_data = m;
4732 trace_array_put(tr);
4738 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
4741 struct seq_file *m = filp->private_data;
4742 struct trace_iterator *iter = m->private;
4743 struct trace_array *tr = iter->tr;
4747 ret = tracing_update_buffers();
4751 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4755 mutex_lock(&trace_types_lock);
4757 if (tr->current_trace->use_max_tr) {
4764 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4768 if (tr->allocated_snapshot)
4772 /* Only allow per-cpu swap if the ring buffer supports it */
4773 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
4774 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4779 if (!tr->allocated_snapshot) {
4780 ret = alloc_snapshot(tr);
4784 local_irq_disable();
4785 /* Now, we're going to swap */
4786 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4787 update_max_tr(tr, current, smp_processor_id());
4789 update_max_tr_single(tr, current, iter->cpu_file);
4793 if (tr->allocated_snapshot) {
4794 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4795 tracing_reset_online_cpus(&tr->max_buffer);
4797 tracing_reset(&tr->max_buffer, iter->cpu_file);
4807 mutex_unlock(&trace_types_lock);
4811 static int tracing_snapshot_release(struct inode *inode, struct file *file)
4813 struct seq_file *m = file->private_data;
4816 ret = tracing_release(inode, file);
4818 if (file->f_mode & FMODE_READ)
4821 /* If write only, the seq_file is just a stub */
4829 static int tracing_buffers_open(struct inode *inode, struct file *filp);
4830 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
4831 size_t count, loff_t *ppos);
4832 static int tracing_buffers_release(struct inode *inode, struct file *file);
4833 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4834 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
4836 static int snapshot_raw_open(struct inode *inode, struct file *filp)
4838 struct ftrace_buffer_info *info;
4841 ret = tracing_buffers_open(inode, filp);
4845 info = filp->private_data;
4847 if (info->iter.trace->use_max_tr) {
4848 tracing_buffers_release(inode, filp);
4852 info->iter.snapshot = true;
4853 info->iter.trace_buffer = &info->iter.tr->max_buffer;
4858 #endif /* CONFIG_TRACER_SNAPSHOT */
4861 static const struct file_operations tracing_max_lat_fops = {
4862 .open = tracing_open_generic,
4863 .read = tracing_max_lat_read,
4864 .write = tracing_max_lat_write,
4865 .llseek = generic_file_llseek,
4868 static const struct file_operations set_tracer_fops = {
4869 .open = tracing_open_generic,
4870 .read = tracing_set_trace_read,
4871 .write = tracing_set_trace_write,
4872 .llseek = generic_file_llseek,
4875 static const struct file_operations tracing_pipe_fops = {
4876 .open = tracing_open_pipe,
4877 .poll = tracing_poll_pipe,
4878 .read = tracing_read_pipe,
4879 .splice_read = tracing_splice_read_pipe,
4880 .release = tracing_release_pipe,
4881 .llseek = no_llseek,
4884 static const struct file_operations tracing_entries_fops = {
4885 .open = tracing_open_generic_tr,
4886 .read = tracing_entries_read,
4887 .write = tracing_entries_write,
4888 .llseek = generic_file_llseek,
4889 .release = tracing_release_generic_tr,
4892 static const struct file_operations tracing_total_entries_fops = {
4893 .open = tracing_open_generic_tr,
4894 .read = tracing_total_entries_read,
4895 .llseek = generic_file_llseek,
4896 .release = tracing_release_generic_tr,
4899 static const struct file_operations tracing_free_buffer_fops = {
4900 .open = tracing_open_generic_tr,
4901 .write = tracing_free_buffer_write,
4902 .release = tracing_free_buffer_release,
4905 static const struct file_operations tracing_mark_fops = {
4906 .open = tracing_open_generic_tr,
4907 .write = tracing_mark_write,
4908 .llseek = generic_file_llseek,
4909 .release = tracing_release_generic_tr,
4912 static const struct file_operations trace_clock_fops = {
4913 .open = tracing_clock_open,
4915 .llseek = seq_lseek,
4916 .release = tracing_single_release_tr,
4917 .write = tracing_clock_write,
4920 #ifdef CONFIG_TRACER_SNAPSHOT
4921 static const struct file_operations snapshot_fops = {
4922 .open = tracing_snapshot_open,
4924 .write = tracing_snapshot_write,
4925 .llseek = tracing_seek,
4926 .release = tracing_snapshot_release,
4929 static const struct file_operations snapshot_raw_fops = {
4930 .open = snapshot_raw_open,
4931 .read = tracing_buffers_read,
4932 .release = tracing_buffers_release,
4933 .splice_read = tracing_buffers_splice_read,
4934 .llseek = no_llseek,
4937 #endif /* CONFIG_TRACER_SNAPSHOT */
4939 static int tracing_buffers_open(struct inode *inode, struct file *filp)
4941 struct trace_array *tr = inode->i_private;
4942 struct ftrace_buffer_info *info;
4945 if (tracing_disabled)
4948 if (trace_array_get(tr) < 0)
4951 info = kzalloc(sizeof(*info), GFP_KERNEL);
4953 trace_array_put(tr);
4957 mutex_lock(&trace_types_lock);
4960 info->iter.cpu_file = tracing_get_cpu(inode);
4961 info->iter.trace = tr->current_trace;
4962 info->iter.trace_buffer = &tr->trace_buffer;
4964 /* Force reading ring buffer for first read */
4965 info->read = (unsigned int)-1;
4967 filp->private_data = info;
4969 mutex_unlock(&trace_types_lock);
4971 ret = nonseekable_open(inode, filp);
4973 trace_array_put(tr);
4979 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
4981 struct ftrace_buffer_info *info = filp->private_data;
4982 struct trace_iterator *iter = &info->iter;
4984 return trace_poll(iter, filp, poll_table);
4988 tracing_buffers_read(struct file *filp, char __user *ubuf,
4989 size_t count, loff_t *ppos)
4991 struct ftrace_buffer_info *info = filp->private_data;
4992 struct trace_iterator *iter = &info->iter;
4999 mutex_lock(&trace_types_lock);
5001 #ifdef CONFIG_TRACER_MAX_TRACE
5002 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
5009 info->spare = ring_buffer_alloc_read_page(iter->trace_buffer->buffer,
5015 /* Do we have previous read data to read? */
5016 if (info->read < PAGE_SIZE)
5020 trace_access_lock(iter->cpu_file);
5021 ret = ring_buffer_read_page(iter->trace_buffer->buffer,
5025 trace_access_unlock(iter->cpu_file);
5028 if (trace_empty(iter)) {
5029 if ((filp->f_flags & O_NONBLOCK)) {
5033 mutex_unlock(&trace_types_lock);
5034 iter->trace->wait_pipe(iter);
5035 mutex_lock(&trace_types_lock);
5036 if (signal_pending(current)) {
5048 size = PAGE_SIZE - info->read;
5052 ret = copy_to_user(ubuf, info->spare + info->read, size);
5063 mutex_unlock(&trace_types_lock);
5068 static int tracing_buffers_release(struct inode *inode, struct file *file)
5070 struct ftrace_buffer_info *info = file->private_data;
5071 struct trace_iterator *iter = &info->iter;
5073 mutex_lock(&trace_types_lock);
5075 __trace_array_put(iter->tr);
5078 ring_buffer_free_read_page(iter->trace_buffer->buffer, info->spare);
5081 mutex_unlock(&trace_types_lock);
5087 struct ring_buffer *buffer;
5092 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
5093 struct pipe_buffer *buf)
5095 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5100 ring_buffer_free_read_page(ref->buffer, ref->page);
5105 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
5106 struct pipe_buffer *buf)
5108 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5113 /* Pipe buffer operations for a buffer. */
5114 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
5116 .map = generic_pipe_buf_map,
5117 .unmap = generic_pipe_buf_unmap,
5118 .confirm = generic_pipe_buf_confirm,
5119 .release = buffer_pipe_buf_release,
5120 .steal = generic_pipe_buf_steal,
5121 .get = buffer_pipe_buf_get,
5125 * Callback from splice_to_pipe(), if we need to release some pages
5126 * at the end of the spd in case we error'ed out in filling the pipe.
5128 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
5130 struct buffer_ref *ref =
5131 (struct buffer_ref *)spd->partial[i].private;
5136 ring_buffer_free_read_page(ref->buffer, ref->page);
5138 spd->partial[i].private = 0;
5142 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
5143 struct pipe_inode_info *pipe, size_t len,
5146 struct ftrace_buffer_info *info = file->private_data;
5147 struct trace_iterator *iter = &info->iter;
5148 struct partial_page partial_def[PIPE_DEF_BUFFERS];
5149 struct page *pages_def[PIPE_DEF_BUFFERS];
5150 struct splice_pipe_desc spd = {
5152 .partial = partial_def,
5153 .nr_pages_max = PIPE_DEF_BUFFERS,
5155 .ops = &buffer_pipe_buf_ops,
5156 .spd_release = buffer_spd_release,
5158 struct buffer_ref *ref;
5159 int entries, size, i;
5162 mutex_lock(&trace_types_lock);
5164 #ifdef CONFIG_TRACER_MAX_TRACE
5165 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
5171 if (splice_grow_spd(pipe, &spd)) {
5176 if (*ppos & (PAGE_SIZE - 1)) {
5181 if (len & (PAGE_SIZE - 1)) {
5182 if (len < PAGE_SIZE) {
5190 trace_access_lock(iter->cpu_file);
5191 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5193 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
5197 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
5202 ref->buffer = iter->trace_buffer->buffer;
5203 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
5209 r = ring_buffer_read_page(ref->buffer, &ref->page,
5210 len, iter->cpu_file, 1);
5212 ring_buffer_free_read_page(ref->buffer, ref->page);
5218 * zero out any left over data, this is going to
5221 size = ring_buffer_page_len(ref->page);
5222 if (size < PAGE_SIZE)
5223 memset(ref->page + size, 0, PAGE_SIZE - size);
5225 page = virt_to_page(ref->page);
5227 spd.pages[i] = page;
5228 spd.partial[i].len = PAGE_SIZE;
5229 spd.partial[i].offset = 0;
5230 spd.partial[i].private = (unsigned long)ref;
5234 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5237 trace_access_unlock(iter->cpu_file);
5240 /* did we read anything? */
5241 if (!spd.nr_pages) {
5242 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK)) {
5246 mutex_unlock(&trace_types_lock);
5247 iter->trace->wait_pipe(iter);
5248 mutex_lock(&trace_types_lock);
5249 if (signal_pending(current)) {
5256 ret = splice_to_pipe(pipe, &spd);
5257 splice_shrink_spd(&spd);
5259 mutex_unlock(&trace_types_lock);
5264 static const struct file_operations tracing_buffers_fops = {
5265 .open = tracing_buffers_open,
5266 .read = tracing_buffers_read,
5267 .poll = tracing_buffers_poll,
5268 .release = tracing_buffers_release,
5269 .splice_read = tracing_buffers_splice_read,
5270 .llseek = no_llseek,
5274 tracing_stats_read(struct file *filp, char __user *ubuf,
5275 size_t count, loff_t *ppos)
5277 struct inode *inode = file_inode(filp);
5278 struct trace_array *tr = inode->i_private;
5279 struct trace_buffer *trace_buf = &tr->trace_buffer;
5280 int cpu = tracing_get_cpu(inode);
5281 struct trace_seq *s;
5283 unsigned long long t;
5284 unsigned long usec_rem;
5286 s = kmalloc(sizeof(*s), GFP_KERNEL);
5292 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
5293 trace_seq_printf(s, "entries: %ld\n", cnt);
5295 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
5296 trace_seq_printf(s, "overrun: %ld\n", cnt);
5298 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
5299 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
5301 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
5302 trace_seq_printf(s, "bytes: %ld\n", cnt);
5304 if (trace_clocks[tr->clock_id].in_ns) {
5305 /* local or global for trace_clock */
5306 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5307 usec_rem = do_div(t, USEC_PER_SEC);
5308 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
5311 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
5312 usec_rem = do_div(t, USEC_PER_SEC);
5313 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
5315 /* counter or tsc mode for trace_clock */
5316 trace_seq_printf(s, "oldest event ts: %llu\n",
5317 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5319 trace_seq_printf(s, "now ts: %llu\n",
5320 ring_buffer_time_stamp(trace_buf->buffer, cpu));
5323 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
5324 trace_seq_printf(s, "dropped events: %ld\n", cnt);
5326 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
5327 trace_seq_printf(s, "read events: %ld\n", cnt);
5329 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
5336 static const struct file_operations tracing_stats_fops = {
5337 .open = tracing_open_generic_tr,
5338 .read = tracing_stats_read,
5339 .llseek = generic_file_llseek,
5340 .release = tracing_release_generic_tr,
5343 #ifdef CONFIG_DYNAMIC_FTRACE
5345 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
5351 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
5352 size_t cnt, loff_t *ppos)
5354 static char ftrace_dyn_info_buffer[1024];
5355 static DEFINE_MUTEX(dyn_info_mutex);
5356 unsigned long *p = filp->private_data;
5357 char *buf = ftrace_dyn_info_buffer;
5358 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
5361 mutex_lock(&dyn_info_mutex);
5362 r = sprintf(buf, "%ld ", *p);
5364 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
5367 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5369 mutex_unlock(&dyn_info_mutex);
5374 static const struct file_operations tracing_dyn_info_fops = {
5375 .open = tracing_open_generic,
5376 .read = tracing_read_dyn_info,
5377 .llseek = generic_file_llseek,
5379 #endif /* CONFIG_DYNAMIC_FTRACE */
5381 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
5383 ftrace_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5389 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5391 unsigned long *count = (long *)data;
5403 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
5404 struct ftrace_probe_ops *ops, void *data)
5406 long count = (long)data;
5408 seq_printf(m, "%ps:", (void *)ip);
5410 seq_printf(m, "snapshot");
5413 seq_printf(m, ":unlimited\n");
5415 seq_printf(m, ":count=%ld\n", count);
5420 static struct ftrace_probe_ops snapshot_probe_ops = {
5421 .func = ftrace_snapshot,
5422 .print = ftrace_snapshot_print,
5425 static struct ftrace_probe_ops snapshot_count_probe_ops = {
5426 .func = ftrace_count_snapshot,
5427 .print = ftrace_snapshot_print,
5431 ftrace_trace_snapshot_callback(struct ftrace_hash *hash,
5432 char *glob, char *cmd, char *param, int enable)
5434 struct ftrace_probe_ops *ops;
5435 void *count = (void *)-1;
5439 /* hash funcs only work with set_ftrace_filter */
5443 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
5445 if (glob[0] == '!') {
5446 unregister_ftrace_function_probe_func(glob+1, ops);
5453 number = strsep(¶m, ":");
5455 if (!strlen(number))
5459 * We use the callback data field (which is a pointer)
5462 ret = kstrtoul(number, 0, (unsigned long *)&count);
5467 ret = register_ftrace_function_probe(glob, ops, count);
5470 alloc_snapshot(&global_trace);
5472 return ret < 0 ? ret : 0;
5475 static struct ftrace_func_command ftrace_snapshot_cmd = {
5477 .func = ftrace_trace_snapshot_callback,
5480 static int register_snapshot_cmd(void)
5482 return register_ftrace_command(&ftrace_snapshot_cmd);
5485 static inline int register_snapshot_cmd(void) { return 0; }
5486 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
5488 struct dentry *tracing_init_dentry_tr(struct trace_array *tr)
5493 if (!debugfs_initialized())
5496 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
5497 tr->dir = debugfs_create_dir("tracing", NULL);
5500 pr_warn_once("Could not create debugfs directory 'tracing'\n");
5505 struct dentry *tracing_init_dentry(void)
5507 return tracing_init_dentry_tr(&global_trace);
5510 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
5512 struct dentry *d_tracer;
5515 return tr->percpu_dir;
5517 d_tracer = tracing_init_dentry_tr(tr);
5521 tr->percpu_dir = debugfs_create_dir("per_cpu", d_tracer);
5523 WARN_ONCE(!tr->percpu_dir,
5524 "Could not create debugfs directory 'per_cpu/%d'\n", cpu);
5526 return tr->percpu_dir;
5529 static struct dentry *
5530 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
5531 void *data, long cpu, const struct file_operations *fops)
5533 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
5535 if (ret) /* See tracing_get_cpu() */
5536 ret->d_inode->i_cdev = (void *)(cpu + 1);
5541 tracing_init_debugfs_percpu(struct trace_array *tr, long cpu)
5543 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
5544 struct dentry *d_cpu;
5545 char cpu_dir[30]; /* 30 characters should be more than enough */
5550 snprintf(cpu_dir, 30, "cpu%ld", cpu);
5551 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
5553 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
5557 /* per cpu trace_pipe */
5558 trace_create_cpu_file("trace_pipe", 0444, d_cpu,
5559 tr, cpu, &tracing_pipe_fops);
5562 trace_create_cpu_file("trace", 0644, d_cpu,
5563 tr, cpu, &tracing_fops);
5565 trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu,
5566 tr, cpu, &tracing_buffers_fops);
5568 trace_create_cpu_file("stats", 0444, d_cpu,
5569 tr, cpu, &tracing_stats_fops);
5571 trace_create_cpu_file("buffer_size_kb", 0444, d_cpu,
5572 tr, cpu, &tracing_entries_fops);
5574 #ifdef CONFIG_TRACER_SNAPSHOT
5575 trace_create_cpu_file("snapshot", 0644, d_cpu,
5576 tr, cpu, &snapshot_fops);
5578 trace_create_cpu_file("snapshot_raw", 0444, d_cpu,
5579 tr, cpu, &snapshot_raw_fops);
5583 #ifdef CONFIG_FTRACE_SELFTEST
5584 /* Let selftest have access to static functions in this file */
5585 #include "trace_selftest.c"
5588 struct trace_option_dentry {
5589 struct tracer_opt *opt;
5590 struct tracer_flags *flags;
5591 struct trace_array *tr;
5592 struct dentry *entry;
5596 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
5599 struct trace_option_dentry *topt = filp->private_data;
5602 if (topt->flags->val & topt->opt->bit)
5607 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5611 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
5614 struct trace_option_dentry *topt = filp->private_data;
5618 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5622 if (val != 0 && val != 1)
5625 if (!!(topt->flags->val & topt->opt->bit) != val) {
5626 mutex_lock(&trace_types_lock);
5627 ret = __set_tracer_option(topt->tr->current_trace, topt->flags,
5629 mutex_unlock(&trace_types_lock);
5640 static const struct file_operations trace_options_fops = {
5641 .open = tracing_open_generic,
5642 .read = trace_options_read,
5643 .write = trace_options_write,
5644 .llseek = generic_file_llseek,
5648 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
5651 long index = (long)filp->private_data;
5654 if (trace_flags & (1 << index))
5659 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5663 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
5666 struct trace_array *tr = &global_trace;
5667 long index = (long)filp->private_data;
5671 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5675 if (val != 0 && val != 1)
5678 mutex_lock(&trace_types_lock);
5679 ret = set_tracer_flag(tr, 1 << index, val);
5680 mutex_unlock(&trace_types_lock);
5690 static const struct file_operations trace_options_core_fops = {
5691 .open = tracing_open_generic,
5692 .read = trace_options_core_read,
5693 .write = trace_options_core_write,
5694 .llseek = generic_file_llseek,
5697 struct dentry *trace_create_file(const char *name,
5699 struct dentry *parent,
5701 const struct file_operations *fops)
5705 ret = debugfs_create_file(name, mode, parent, data, fops);
5707 pr_warning("Could not create debugfs '%s' entry\n", name);
5713 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
5715 struct dentry *d_tracer;
5720 d_tracer = tracing_init_dentry_tr(tr);
5724 tr->options = debugfs_create_dir("options", d_tracer);
5726 pr_warning("Could not create debugfs directory 'options'\n");
5734 create_trace_option_file(struct trace_array *tr,
5735 struct trace_option_dentry *topt,
5736 struct tracer_flags *flags,
5737 struct tracer_opt *opt)
5739 struct dentry *t_options;
5741 t_options = trace_options_init_dentry(tr);
5745 topt->flags = flags;
5749 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
5750 &trace_options_fops);
5754 static struct trace_option_dentry *
5755 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
5757 struct trace_option_dentry *topts;
5758 struct tracer_flags *flags;
5759 struct tracer_opt *opts;
5765 flags = tracer->flags;
5767 if (!flags || !flags->opts)
5772 for (cnt = 0; opts[cnt].name; cnt++)
5775 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
5779 for (cnt = 0; opts[cnt].name; cnt++)
5780 create_trace_option_file(tr, &topts[cnt], flags,
5787 destroy_trace_option_files(struct trace_option_dentry *topts)
5794 for (cnt = 0; topts[cnt].opt; cnt++) {
5795 if (topts[cnt].entry)
5796 debugfs_remove(topts[cnt].entry);
5802 static struct dentry *
5803 create_trace_option_core_file(struct trace_array *tr,
5804 const char *option, long index)
5806 struct dentry *t_options;
5808 t_options = trace_options_init_dentry(tr);
5812 return trace_create_file(option, 0644, t_options, (void *)index,
5813 &trace_options_core_fops);
5816 static __init void create_trace_options_dir(struct trace_array *tr)
5818 struct dentry *t_options;
5821 t_options = trace_options_init_dentry(tr);
5825 for (i = 0; trace_options[i]; i++)
5826 create_trace_option_core_file(tr, trace_options[i], i);
5830 rb_simple_read(struct file *filp, char __user *ubuf,
5831 size_t cnt, loff_t *ppos)
5833 struct trace_array *tr = filp->private_data;
5837 r = tracer_tracing_is_on(tr);
5838 r = sprintf(buf, "%d\n", r);
5840 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5844 rb_simple_write(struct file *filp, const char __user *ubuf,
5845 size_t cnt, loff_t *ppos)
5847 struct trace_array *tr = filp->private_data;
5848 struct ring_buffer *buffer = tr->trace_buffer.buffer;
5852 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5857 mutex_lock(&trace_types_lock);
5859 tracer_tracing_on(tr);
5860 if (tr->current_trace->start)
5861 tr->current_trace->start(tr);
5863 tracer_tracing_off(tr);
5864 if (tr->current_trace->stop)
5865 tr->current_trace->stop(tr);
5867 mutex_unlock(&trace_types_lock);
5875 static const struct file_operations rb_simple_fops = {
5876 .open = tracing_open_generic_tr,
5877 .read = rb_simple_read,
5878 .write = rb_simple_write,
5879 .release = tracing_release_generic_tr,
5880 .llseek = default_llseek,
5883 struct dentry *trace_instance_dir;
5886 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer);
5889 allocate_trace_buffer(struct trace_array *tr, struct trace_buffer *buf, int size)
5891 enum ring_buffer_flags rb_flags;
5893 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
5895 buf->buffer = ring_buffer_alloc(size, rb_flags);
5899 buf->data = alloc_percpu(struct trace_array_cpu);
5901 ring_buffer_free(buf->buffer);
5905 /* Allocate the first page for all buffers */
5906 set_buffer_entries(&tr->trace_buffer,
5907 ring_buffer_size(tr->trace_buffer.buffer, 0));
5912 static int allocate_trace_buffers(struct trace_array *tr, int size)
5916 ret = allocate_trace_buffer(tr, &tr->trace_buffer, size);
5920 #ifdef CONFIG_TRACER_MAX_TRACE
5921 ret = allocate_trace_buffer(tr, &tr->max_buffer,
5922 allocate_snapshot ? size : 1);
5924 ring_buffer_free(tr->trace_buffer.buffer);
5925 free_percpu(tr->trace_buffer.data);
5928 tr->allocated_snapshot = allocate_snapshot;
5931 * Only the top level trace array gets its snapshot allocated
5932 * from the kernel command line.
5934 allocate_snapshot = false;
5939 static int new_instance_create(const char *name)
5941 struct trace_array *tr;
5944 mutex_lock(&trace_types_lock);
5947 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5948 if (tr->name && strcmp(tr->name, name) == 0)
5953 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
5957 tr->name = kstrdup(name, GFP_KERNEL);
5961 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
5964 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
5966 raw_spin_lock_init(&tr->start_lock);
5968 tr->current_trace = &nop_trace;
5970 INIT_LIST_HEAD(&tr->systems);
5971 INIT_LIST_HEAD(&tr->events);
5973 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
5976 tr->dir = debugfs_create_dir(name, trace_instance_dir);
5980 ret = event_trace_add_tracer(tr->dir, tr);
5982 debugfs_remove_recursive(tr->dir);
5986 init_tracer_debugfs(tr, tr->dir);
5988 list_add(&tr->list, &ftrace_trace_arrays);
5990 mutex_unlock(&trace_types_lock);
5995 if (tr->trace_buffer.buffer)
5996 ring_buffer_free(tr->trace_buffer.buffer);
5997 free_cpumask_var(tr->tracing_cpumask);
6002 mutex_unlock(&trace_types_lock);
6008 static int instance_delete(const char *name)
6010 struct trace_array *tr;
6014 mutex_lock(&trace_types_lock);
6017 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
6018 if (tr->name && strcmp(tr->name, name) == 0) {
6030 list_del(&tr->list);
6032 event_trace_del_tracer(tr);
6033 debugfs_remove_recursive(tr->dir);
6034 free_percpu(tr->trace_buffer.data);
6035 ring_buffer_free(tr->trace_buffer.buffer);
6043 mutex_unlock(&trace_types_lock);
6048 static int instance_mkdir (struct inode *inode, struct dentry *dentry, umode_t mode)
6050 struct dentry *parent;
6053 /* Paranoid: Make sure the parent is the "instances" directory */
6054 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
6055 if (WARN_ON_ONCE(parent != trace_instance_dir))
6059 * The inode mutex is locked, but debugfs_create_dir() will also
6060 * take the mutex. As the instances directory can not be destroyed
6061 * or changed in any other way, it is safe to unlock it, and
6062 * let the dentry try. If two users try to make the same dir at
6063 * the same time, then the new_instance_create() will determine the
6066 mutex_unlock(&inode->i_mutex);
6068 ret = new_instance_create(dentry->d_iname);
6070 mutex_lock(&inode->i_mutex);
6075 static int instance_rmdir(struct inode *inode, struct dentry *dentry)
6077 struct dentry *parent;
6080 /* Paranoid: Make sure the parent is the "instances" directory */
6081 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
6082 if (WARN_ON_ONCE(parent != trace_instance_dir))
6085 /* The caller did a dget() on dentry */
6086 mutex_unlock(&dentry->d_inode->i_mutex);
6089 * The inode mutex is locked, but debugfs_create_dir() will also
6090 * take the mutex. As the instances directory can not be destroyed
6091 * or changed in any other way, it is safe to unlock it, and
6092 * let the dentry try. If two users try to make the same dir at
6093 * the same time, then the instance_delete() will determine the
6096 mutex_unlock(&inode->i_mutex);
6098 ret = instance_delete(dentry->d_iname);
6100 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
6101 mutex_lock(&dentry->d_inode->i_mutex);
6106 static const struct inode_operations instance_dir_inode_operations = {
6107 .lookup = simple_lookup,
6108 .mkdir = instance_mkdir,
6109 .rmdir = instance_rmdir,
6112 static __init void create_trace_instances(struct dentry *d_tracer)
6114 trace_instance_dir = debugfs_create_dir("instances", d_tracer);
6115 if (WARN_ON(!trace_instance_dir))
6118 /* Hijack the dir inode operations, to allow mkdir */
6119 trace_instance_dir->d_inode->i_op = &instance_dir_inode_operations;
6123 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer)
6127 trace_create_file("tracing_cpumask", 0644, d_tracer,
6128 tr, &tracing_cpumask_fops);
6130 trace_create_file("trace_options", 0644, d_tracer,
6131 tr, &tracing_iter_fops);
6133 trace_create_file("trace", 0644, d_tracer,
6136 trace_create_file("trace_pipe", 0444, d_tracer,
6137 tr, &tracing_pipe_fops);
6139 trace_create_file("buffer_size_kb", 0644, d_tracer,
6140 tr, &tracing_entries_fops);
6142 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
6143 tr, &tracing_total_entries_fops);
6145 trace_create_file("free_buffer", 0200, d_tracer,
6146 tr, &tracing_free_buffer_fops);
6148 trace_create_file("trace_marker", 0220, d_tracer,
6149 tr, &tracing_mark_fops);
6151 trace_create_file("trace_clock", 0644, d_tracer, tr,
6154 trace_create_file("tracing_on", 0644, d_tracer,
6155 tr, &rb_simple_fops);
6157 #ifdef CONFIG_TRACER_SNAPSHOT
6158 trace_create_file("snapshot", 0644, d_tracer,
6159 tr, &snapshot_fops);
6162 for_each_tracing_cpu(cpu)
6163 tracing_init_debugfs_percpu(tr, cpu);
6167 static __init int tracer_init_debugfs(void)
6169 struct dentry *d_tracer;
6171 trace_access_lock_init();
6173 d_tracer = tracing_init_dentry();
6177 init_tracer_debugfs(&global_trace, d_tracer);
6179 trace_create_file("available_tracers", 0444, d_tracer,
6180 &global_trace, &show_traces_fops);
6182 trace_create_file("current_tracer", 0644, d_tracer,
6183 &global_trace, &set_tracer_fops);
6185 #ifdef CONFIG_TRACER_MAX_TRACE
6186 trace_create_file("tracing_max_latency", 0644, d_tracer,
6187 &tracing_max_latency, &tracing_max_lat_fops);
6190 trace_create_file("tracing_thresh", 0644, d_tracer,
6191 &tracing_thresh, &tracing_max_lat_fops);
6193 trace_create_file("README", 0444, d_tracer,
6194 NULL, &tracing_readme_fops);
6196 trace_create_file("saved_cmdlines", 0444, d_tracer,
6197 NULL, &tracing_saved_cmdlines_fops);
6199 #ifdef CONFIG_DYNAMIC_FTRACE
6200 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
6201 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
6204 create_trace_instances(d_tracer);
6206 create_trace_options_dir(&global_trace);
6211 static int trace_panic_handler(struct notifier_block *this,
6212 unsigned long event, void *unused)
6214 if (ftrace_dump_on_oops)
6215 ftrace_dump(ftrace_dump_on_oops);
6219 static struct notifier_block trace_panic_notifier = {
6220 .notifier_call = trace_panic_handler,
6222 .priority = 150 /* priority: INT_MAX >= x >= 0 */
6225 static int trace_die_handler(struct notifier_block *self,
6231 if (ftrace_dump_on_oops)
6232 ftrace_dump(ftrace_dump_on_oops);
6240 static struct notifier_block trace_die_notifier = {
6241 .notifier_call = trace_die_handler,
6246 * printk is set to max of 1024, we really don't need it that big.
6247 * Nothing should be printing 1000 characters anyway.
6249 #define TRACE_MAX_PRINT 1000
6252 * Define here KERN_TRACE so that we have one place to modify
6253 * it if we decide to change what log level the ftrace dump
6256 #define KERN_TRACE KERN_EMERG
6259 trace_printk_seq(struct trace_seq *s)
6261 /* Probably should print a warning here. */
6262 if (s->len >= TRACE_MAX_PRINT)
6263 s->len = TRACE_MAX_PRINT;
6265 /* should be zero ended, but we are paranoid. */
6266 s->buffer[s->len] = 0;
6268 printk(KERN_TRACE "%s", s->buffer);
6273 void trace_init_global_iter(struct trace_iterator *iter)
6275 iter->tr = &global_trace;
6276 iter->trace = iter->tr->current_trace;
6277 iter->cpu_file = RING_BUFFER_ALL_CPUS;
6278 iter->trace_buffer = &global_trace.trace_buffer;
6281 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
6283 /* use static because iter can be a bit big for the stack */
6284 static struct trace_iterator iter;
6285 static atomic_t dump_running;
6286 unsigned int old_userobj;
6287 unsigned long flags;
6290 /* Only allow one dump user at a time. */
6291 if (atomic_inc_return(&dump_running) != 1) {
6292 atomic_dec(&dump_running);
6297 * Always turn off tracing when we dump.
6298 * We don't need to show trace output of what happens
6299 * between multiple crashes.
6301 * If the user does a sysrq-z, then they can re-enable
6302 * tracing with echo 1 > tracing_on.
6306 local_irq_save(flags);
6308 /* Simulate the iterator */
6309 trace_init_global_iter(&iter);
6311 for_each_tracing_cpu(cpu) {
6312 atomic_inc(&per_cpu_ptr(iter.tr->trace_buffer.data, cpu)->disabled);
6315 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
6317 /* don't look at user memory in panic mode */
6318 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
6320 switch (oops_dump_mode) {
6322 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6325 iter.cpu_file = raw_smp_processor_id();
6330 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
6331 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6334 printk(KERN_TRACE "Dumping ftrace buffer:\n");
6336 /* Did function tracer already get disabled? */
6337 if (ftrace_is_dead()) {
6338 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
6339 printk("# MAY BE MISSING FUNCTION EVENTS\n");
6343 * We need to stop all tracing on all CPUS to read the
6344 * the next buffer. This is a bit expensive, but is
6345 * not done often. We fill all what we can read,
6346 * and then release the locks again.
6349 while (!trace_empty(&iter)) {
6352 printk(KERN_TRACE "---------------------------------\n");
6356 /* reset all but tr, trace, and overruns */
6357 memset(&iter.seq, 0,
6358 sizeof(struct trace_iterator) -
6359 offsetof(struct trace_iterator, seq));
6360 iter.iter_flags |= TRACE_FILE_LAT_FMT;
6363 if (trace_find_next_entry_inc(&iter) != NULL) {
6366 ret = print_trace_line(&iter);
6367 if (ret != TRACE_TYPE_NO_CONSUME)
6368 trace_consume(&iter);
6370 touch_nmi_watchdog();
6372 trace_printk_seq(&iter.seq);
6376 printk(KERN_TRACE " (ftrace buffer empty)\n");
6378 printk(KERN_TRACE "---------------------------------\n");
6381 trace_flags |= old_userobj;
6383 for_each_tracing_cpu(cpu) {
6384 atomic_dec(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);
6386 atomic_dec(&dump_running);
6387 local_irq_restore(flags);
6389 EXPORT_SYMBOL_GPL(ftrace_dump);
6391 __init static int tracer_alloc_buffers(void)
6397 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
6400 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
6401 goto out_free_buffer_mask;
6403 /* Only allocate trace_printk buffers if a trace_printk exists */
6404 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
6405 /* Must be called before global_trace.buffer is allocated */
6406 trace_printk_init_buffers();
6408 /* To save memory, keep the ring buffer size to its minimum */
6409 if (ring_buffer_expanded)
6410 ring_buf_size = trace_buf_size;
6414 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
6415 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
6417 raw_spin_lock_init(&global_trace.start_lock);
6419 /* TODO: make the number of buffers hot pluggable with CPUS */
6420 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
6421 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
6423 goto out_free_cpumask;
6426 if (global_trace.buffer_disabled)
6429 trace_init_cmdlines();
6432 * register_tracer() might reference current_trace, so it
6433 * needs to be set before we register anything. This is
6434 * just a bootstrap of current_trace anyway.
6436 global_trace.current_trace = &nop_trace;
6438 register_tracer(&nop_trace);
6440 /* All seems OK, enable tracing */
6441 tracing_disabled = 0;
6443 atomic_notifier_chain_register(&panic_notifier_list,
6444 &trace_panic_notifier);
6446 register_die_notifier(&trace_die_notifier);
6448 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
6450 INIT_LIST_HEAD(&global_trace.systems);
6451 INIT_LIST_HEAD(&global_trace.events);
6452 list_add(&global_trace.list, &ftrace_trace_arrays);
6454 while (trace_boot_options) {
6457 option = strsep(&trace_boot_options, ",");
6458 trace_set_options(&global_trace, option);
6461 register_snapshot_cmd();
6466 free_percpu(global_trace.trace_buffer.data);
6467 #ifdef CONFIG_TRACER_MAX_TRACE
6468 free_percpu(global_trace.max_buffer.data);
6470 free_cpumask_var(global_trace.tracing_cpumask);
6471 out_free_buffer_mask:
6472 free_cpumask_var(tracing_buffer_mask);
6477 __init static int clear_boot_tracer(void)
6480 * The default tracer at boot buffer is an init section.
6481 * This function is called in lateinit. If we did not
6482 * find the boot tracer, then clear it out, to prevent
6483 * later registration from accessing the buffer that is
6484 * about to be freed.
6486 if (!default_bootup_tracer)
6489 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
6490 default_bootup_tracer);
6491 default_bootup_tracer = NULL;
6496 early_initcall(tracer_alloc_buffers);
6497 fs_initcall(tracer_init_debugfs);
6498 late_initcall(clear_boot_tracer);