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 * tracing_start - quick start of the tracer
1289 * If tracing is enabled but was stopped by tracing_stop,
1290 * this will start the tracer back up.
1292 void tracing_start(void)
1294 struct ring_buffer *buffer;
1295 unsigned long flags;
1297 if (tracing_disabled)
1300 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1301 if (--global_trace.stop_count) {
1302 if (global_trace.stop_count < 0) {
1303 /* Someone screwed up their debugging */
1305 global_trace.stop_count = 0;
1310 /* Prevent the buffers from switching */
1311 arch_spin_lock(&ftrace_max_lock);
1313 buffer = global_trace.trace_buffer.buffer;
1315 ring_buffer_record_enable(buffer);
1317 #ifdef CONFIG_TRACER_MAX_TRACE
1318 buffer = global_trace.max_buffer.buffer;
1320 ring_buffer_record_enable(buffer);
1323 arch_spin_unlock(&ftrace_max_lock);
1327 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1330 static void tracing_start_tr(struct trace_array *tr)
1332 struct ring_buffer *buffer;
1333 unsigned long flags;
1335 if (tracing_disabled)
1338 /* If global, we need to also start the max tracer */
1339 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1340 return tracing_start();
1342 raw_spin_lock_irqsave(&tr->start_lock, flags);
1344 if (--tr->stop_count) {
1345 if (tr->stop_count < 0) {
1346 /* Someone screwed up their debugging */
1353 buffer = tr->trace_buffer.buffer;
1355 ring_buffer_record_enable(buffer);
1358 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1362 * tracing_stop - quick stop of the tracer
1364 * Light weight way to stop tracing. Use in conjunction with
1367 void tracing_stop(void)
1369 struct ring_buffer *buffer;
1370 unsigned long flags;
1373 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1374 if (global_trace.stop_count++)
1377 /* Prevent the buffers from switching */
1378 arch_spin_lock(&ftrace_max_lock);
1380 buffer = global_trace.trace_buffer.buffer;
1382 ring_buffer_record_disable(buffer);
1384 #ifdef CONFIG_TRACER_MAX_TRACE
1385 buffer = global_trace.max_buffer.buffer;
1387 ring_buffer_record_disable(buffer);
1390 arch_spin_unlock(&ftrace_max_lock);
1393 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1396 static void tracing_stop_tr(struct trace_array *tr)
1398 struct ring_buffer *buffer;
1399 unsigned long flags;
1401 /* If global, we need to also stop the max tracer */
1402 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1403 return tracing_stop();
1405 raw_spin_lock_irqsave(&tr->start_lock, flags);
1406 if (tr->stop_count++)
1409 buffer = tr->trace_buffer.buffer;
1411 ring_buffer_record_disable(buffer);
1414 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1417 void trace_stop_cmdline_recording(void);
1419 static void trace_save_cmdline(struct task_struct *tsk)
1423 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1427 * It's not the end of the world if we don't get
1428 * the lock, but we also don't want to spin
1429 * nor do we want to disable interrupts,
1430 * so if we miss here, then better luck next time.
1432 if (!arch_spin_trylock(&trace_cmdline_lock))
1435 idx = map_pid_to_cmdline[tsk->pid];
1436 if (idx == NO_CMDLINE_MAP) {
1437 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1440 * Check whether the cmdline buffer at idx has a pid
1441 * mapped. We are going to overwrite that entry so we
1442 * need to clear the map_pid_to_cmdline. Otherwise we
1443 * would read the new comm for the old pid.
1445 pid = map_cmdline_to_pid[idx];
1446 if (pid != NO_CMDLINE_MAP)
1447 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1449 map_cmdline_to_pid[idx] = tsk->pid;
1450 map_pid_to_cmdline[tsk->pid] = idx;
1455 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1457 arch_spin_unlock(&trace_cmdline_lock);
1460 void trace_find_cmdline(int pid, char comm[])
1465 strcpy(comm, "<idle>");
1469 if (WARN_ON_ONCE(pid < 0)) {
1470 strcpy(comm, "<XXX>");
1474 if (pid > PID_MAX_DEFAULT) {
1475 strcpy(comm, "<...>");
1480 arch_spin_lock(&trace_cmdline_lock);
1481 map = map_pid_to_cmdline[pid];
1482 if (map != NO_CMDLINE_MAP)
1483 strcpy(comm, saved_cmdlines[map]);
1485 strcpy(comm, "<...>");
1487 arch_spin_unlock(&trace_cmdline_lock);
1491 void tracing_record_cmdline(struct task_struct *tsk)
1493 if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on())
1496 if (!__this_cpu_read(trace_cmdline_save))
1499 __this_cpu_write(trace_cmdline_save, false);
1501 trace_save_cmdline(tsk);
1505 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1508 struct task_struct *tsk = current;
1510 entry->preempt_count = pc & 0xff;
1511 entry->pid = (tsk) ? tsk->pid : 0;
1513 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1514 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1516 TRACE_FLAG_IRQS_NOSUPPORT |
1518 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1519 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1520 (tif_need_resched() ? TRACE_FLAG_NEED_RESCHED : 0) |
1521 (test_preempt_need_resched() ? TRACE_FLAG_PREEMPT_RESCHED : 0);
1523 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1525 struct ring_buffer_event *
1526 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1529 unsigned long flags, int pc)
1531 struct ring_buffer_event *event;
1533 event = ring_buffer_lock_reserve(buffer, len);
1534 if (event != NULL) {
1535 struct trace_entry *ent = ring_buffer_event_data(event);
1537 tracing_generic_entry_update(ent, flags, pc);
1545 __buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event)
1547 __this_cpu_write(trace_cmdline_save, true);
1548 ring_buffer_unlock_commit(buffer, event);
1552 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1553 struct ring_buffer_event *event,
1554 unsigned long flags, int pc)
1556 __buffer_unlock_commit(buffer, event);
1558 ftrace_trace_stack(buffer, flags, 6, pc);
1559 ftrace_trace_userstack(buffer, flags, pc);
1562 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1563 struct ring_buffer_event *event,
1564 unsigned long flags, int pc)
1566 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1568 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit);
1570 struct ring_buffer_event *
1571 trace_event_buffer_lock_reserve(struct ring_buffer **current_rb,
1572 struct ftrace_event_file *ftrace_file,
1573 int type, unsigned long len,
1574 unsigned long flags, int pc)
1576 *current_rb = ftrace_file->tr->trace_buffer.buffer;
1577 return trace_buffer_lock_reserve(*current_rb,
1578 type, len, flags, pc);
1580 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
1582 struct ring_buffer_event *
1583 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1584 int type, unsigned long len,
1585 unsigned long flags, int pc)
1587 *current_rb = global_trace.trace_buffer.buffer;
1588 return trace_buffer_lock_reserve(*current_rb,
1589 type, len, flags, pc);
1591 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1593 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1594 struct ring_buffer_event *event,
1595 unsigned long flags, int pc)
1597 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1599 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1601 void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1602 struct ring_buffer_event *event,
1603 unsigned long flags, int pc,
1604 struct pt_regs *regs)
1606 __buffer_unlock_commit(buffer, event);
1608 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1609 ftrace_trace_userstack(buffer, flags, pc);
1611 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs);
1613 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1614 struct ring_buffer_event *event)
1616 ring_buffer_discard_commit(buffer, event);
1618 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1621 trace_function(struct trace_array *tr,
1622 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1625 struct ftrace_event_call *call = &event_function;
1626 struct ring_buffer *buffer = tr->trace_buffer.buffer;
1627 struct ring_buffer_event *event;
1628 struct ftrace_entry *entry;
1630 /* If we are reading the ring buffer, don't trace */
1631 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1634 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1638 entry = ring_buffer_event_data(event);
1640 entry->parent_ip = parent_ip;
1642 if (!call_filter_check_discard(call, entry, buffer, event))
1643 __buffer_unlock_commit(buffer, event);
1646 #ifdef CONFIG_STACKTRACE
1648 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1649 struct ftrace_stack {
1650 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1653 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1654 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1656 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1657 unsigned long flags,
1658 int skip, int pc, struct pt_regs *regs)
1660 struct ftrace_event_call *call = &event_kernel_stack;
1661 struct ring_buffer_event *event;
1662 struct stack_entry *entry;
1663 struct stack_trace trace;
1665 int size = FTRACE_STACK_ENTRIES;
1667 trace.nr_entries = 0;
1671 * Since events can happen in NMIs there's no safe way to
1672 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1673 * or NMI comes in, it will just have to use the default
1674 * FTRACE_STACK_SIZE.
1676 preempt_disable_notrace();
1678 use_stack = __this_cpu_inc_return(ftrace_stack_reserve);
1680 * We don't need any atomic variables, just a barrier.
1681 * If an interrupt comes in, we don't care, because it would
1682 * have exited and put the counter back to what we want.
1683 * We just need a barrier to keep gcc from moving things
1687 if (use_stack == 1) {
1688 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1689 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1692 save_stack_trace_regs(regs, &trace);
1694 save_stack_trace(&trace);
1696 if (trace.nr_entries > size)
1697 size = trace.nr_entries;
1699 /* From now on, use_stack is a boolean */
1702 size *= sizeof(unsigned long);
1704 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1705 sizeof(*entry) + size, flags, pc);
1708 entry = ring_buffer_event_data(event);
1710 memset(&entry->caller, 0, size);
1713 memcpy(&entry->caller, trace.entries,
1714 trace.nr_entries * sizeof(unsigned long));
1716 trace.max_entries = FTRACE_STACK_ENTRIES;
1717 trace.entries = entry->caller;
1719 save_stack_trace_regs(regs, &trace);
1721 save_stack_trace(&trace);
1724 entry->size = trace.nr_entries;
1726 if (!call_filter_check_discard(call, entry, buffer, event))
1727 __buffer_unlock_commit(buffer, event);
1730 /* Again, don't let gcc optimize things here */
1732 __this_cpu_dec(ftrace_stack_reserve);
1733 preempt_enable_notrace();
1737 void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1738 int skip, int pc, struct pt_regs *regs)
1740 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1743 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1746 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1749 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1752 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1755 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1758 __ftrace_trace_stack(tr->trace_buffer.buffer, flags, skip, pc, NULL);
1762 * trace_dump_stack - record a stack back trace in the trace buffer
1763 * @skip: Number of functions to skip (helper handlers)
1765 void trace_dump_stack(int skip)
1767 unsigned long flags;
1769 if (tracing_disabled || tracing_selftest_running)
1772 local_save_flags(flags);
1775 * Skip 3 more, seems to get us at the caller of
1779 __ftrace_trace_stack(global_trace.trace_buffer.buffer,
1780 flags, skip, preempt_count(), NULL);
1783 static DEFINE_PER_CPU(int, user_stack_count);
1786 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1788 struct ftrace_event_call *call = &event_user_stack;
1789 struct ring_buffer_event *event;
1790 struct userstack_entry *entry;
1791 struct stack_trace trace;
1793 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1797 * NMIs can not handle page faults, even with fix ups.
1798 * The save user stack can (and often does) fault.
1800 if (unlikely(in_nmi()))
1804 * prevent recursion, since the user stack tracing may
1805 * trigger other kernel events.
1808 if (__this_cpu_read(user_stack_count))
1811 __this_cpu_inc(user_stack_count);
1813 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1814 sizeof(*entry), flags, pc);
1816 goto out_drop_count;
1817 entry = ring_buffer_event_data(event);
1819 entry->tgid = current->tgid;
1820 memset(&entry->caller, 0, sizeof(entry->caller));
1822 trace.nr_entries = 0;
1823 trace.max_entries = FTRACE_STACK_ENTRIES;
1825 trace.entries = entry->caller;
1827 save_stack_trace_user(&trace);
1828 if (!call_filter_check_discard(call, entry, buffer, event))
1829 __buffer_unlock_commit(buffer, event);
1832 __this_cpu_dec(user_stack_count);
1838 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1840 ftrace_trace_userstack(tr, flags, preempt_count());
1844 #endif /* CONFIG_STACKTRACE */
1846 /* created for use with alloc_percpu */
1847 struct trace_buffer_struct {
1848 char buffer[TRACE_BUF_SIZE];
1851 static struct trace_buffer_struct *trace_percpu_buffer;
1852 static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1853 static struct trace_buffer_struct *trace_percpu_irq_buffer;
1854 static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1857 * The buffer used is dependent on the context. There is a per cpu
1858 * buffer for normal context, softirq contex, hard irq context and
1859 * for NMI context. Thise allows for lockless recording.
1861 * Note, if the buffers failed to be allocated, then this returns NULL
1863 static char *get_trace_buf(void)
1865 struct trace_buffer_struct *percpu_buffer;
1868 * If we have allocated per cpu buffers, then we do not
1869 * need to do any locking.
1872 percpu_buffer = trace_percpu_nmi_buffer;
1874 percpu_buffer = trace_percpu_irq_buffer;
1875 else if (in_softirq())
1876 percpu_buffer = trace_percpu_sirq_buffer;
1878 percpu_buffer = trace_percpu_buffer;
1883 return this_cpu_ptr(&percpu_buffer->buffer[0]);
1886 static int alloc_percpu_trace_buffer(void)
1888 struct trace_buffer_struct *buffers;
1889 struct trace_buffer_struct *sirq_buffers;
1890 struct trace_buffer_struct *irq_buffers;
1891 struct trace_buffer_struct *nmi_buffers;
1893 buffers = alloc_percpu(struct trace_buffer_struct);
1897 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
1901 irq_buffers = alloc_percpu(struct trace_buffer_struct);
1905 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
1909 trace_percpu_buffer = buffers;
1910 trace_percpu_sirq_buffer = sirq_buffers;
1911 trace_percpu_irq_buffer = irq_buffers;
1912 trace_percpu_nmi_buffer = nmi_buffers;
1917 free_percpu(irq_buffers);
1919 free_percpu(sirq_buffers);
1921 free_percpu(buffers);
1923 WARN(1, "Could not allocate percpu trace_printk buffer");
1927 static int buffers_allocated;
1929 void trace_printk_init_buffers(void)
1931 if (buffers_allocated)
1934 if (alloc_percpu_trace_buffer())
1937 pr_info("ftrace: Allocated trace_printk buffers\n");
1939 /* Expand the buffers to set size */
1940 tracing_update_buffers();
1942 buffers_allocated = 1;
1945 * trace_printk_init_buffers() can be called by modules.
1946 * If that happens, then we need to start cmdline recording
1947 * directly here. If the global_trace.buffer is already
1948 * allocated here, then this was called by module code.
1950 if (global_trace.trace_buffer.buffer)
1951 tracing_start_cmdline_record();
1954 void trace_printk_start_comm(void)
1956 /* Start tracing comms if trace printk is set */
1957 if (!buffers_allocated)
1959 tracing_start_cmdline_record();
1962 static void trace_printk_start_stop_comm(int enabled)
1964 if (!buffers_allocated)
1968 tracing_start_cmdline_record();
1970 tracing_stop_cmdline_record();
1974 * trace_vbprintk - write binary msg to tracing buffer
1977 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1979 struct ftrace_event_call *call = &event_bprint;
1980 struct ring_buffer_event *event;
1981 struct ring_buffer *buffer;
1982 struct trace_array *tr = &global_trace;
1983 struct bprint_entry *entry;
1984 unsigned long flags;
1986 int len = 0, size, pc;
1988 if (unlikely(tracing_selftest_running || tracing_disabled))
1991 /* Don't pollute graph traces with trace_vprintk internals */
1992 pause_graph_tracing();
1994 pc = preempt_count();
1995 preempt_disable_notrace();
1997 tbuffer = get_trace_buf();
2003 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
2005 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
2008 local_save_flags(flags);
2009 size = sizeof(*entry) + sizeof(u32) * len;
2010 buffer = tr->trace_buffer.buffer;
2011 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
2015 entry = ring_buffer_event_data(event);
2019 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
2020 if (!call_filter_check_discard(call, entry, buffer, event)) {
2021 __buffer_unlock_commit(buffer, event);
2022 ftrace_trace_stack(buffer, flags, 6, pc);
2026 preempt_enable_notrace();
2027 unpause_graph_tracing();
2031 EXPORT_SYMBOL_GPL(trace_vbprintk);
2034 __trace_array_vprintk(struct ring_buffer *buffer,
2035 unsigned long ip, const char *fmt, va_list args)
2037 struct ftrace_event_call *call = &event_print;
2038 struct ring_buffer_event *event;
2039 int len = 0, size, pc;
2040 struct print_entry *entry;
2041 unsigned long flags;
2044 if (tracing_disabled || tracing_selftest_running)
2047 /* Don't pollute graph traces with trace_vprintk internals */
2048 pause_graph_tracing();
2050 pc = preempt_count();
2051 preempt_disable_notrace();
2054 tbuffer = get_trace_buf();
2060 len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
2061 if (len > TRACE_BUF_SIZE)
2064 local_save_flags(flags);
2065 size = sizeof(*entry) + len + 1;
2066 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
2070 entry = ring_buffer_event_data(event);
2073 memcpy(&entry->buf, tbuffer, len);
2074 entry->buf[len] = '\0';
2075 if (!call_filter_check_discard(call, entry, buffer, event)) {
2076 __buffer_unlock_commit(buffer, event);
2077 ftrace_trace_stack(buffer, flags, 6, pc);
2080 preempt_enable_notrace();
2081 unpause_graph_tracing();
2086 int trace_array_vprintk(struct trace_array *tr,
2087 unsigned long ip, const char *fmt, va_list args)
2089 return __trace_array_vprintk(tr->trace_buffer.buffer, ip, fmt, args);
2092 int trace_array_printk(struct trace_array *tr,
2093 unsigned long ip, const char *fmt, ...)
2098 if (!(trace_flags & TRACE_ITER_PRINTK))
2102 ret = trace_array_vprintk(tr, ip, fmt, ap);
2107 int trace_array_printk_buf(struct ring_buffer *buffer,
2108 unsigned long ip, const char *fmt, ...)
2113 if (!(trace_flags & TRACE_ITER_PRINTK))
2117 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
2122 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
2124 return trace_array_vprintk(&global_trace, ip, fmt, args);
2126 EXPORT_SYMBOL_GPL(trace_vprintk);
2128 static void trace_iterator_increment(struct trace_iterator *iter)
2130 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
2134 ring_buffer_read(buf_iter, NULL);
2137 static struct trace_entry *
2138 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
2139 unsigned long *lost_events)
2141 struct ring_buffer_event *event;
2142 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
2145 event = ring_buffer_iter_peek(buf_iter, ts);
2147 event = ring_buffer_peek(iter->trace_buffer->buffer, cpu, ts,
2151 iter->ent_size = ring_buffer_event_length(event);
2152 return ring_buffer_event_data(event);
2158 static struct trace_entry *
2159 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
2160 unsigned long *missing_events, u64 *ent_ts)
2162 struct ring_buffer *buffer = iter->trace_buffer->buffer;
2163 struct trace_entry *ent, *next = NULL;
2164 unsigned long lost_events = 0, next_lost = 0;
2165 int cpu_file = iter->cpu_file;
2166 u64 next_ts = 0, ts;
2172 * If we are in a per_cpu trace file, don't bother by iterating over
2173 * all cpu and peek directly.
2175 if (cpu_file > RING_BUFFER_ALL_CPUS) {
2176 if (ring_buffer_empty_cpu(buffer, cpu_file))
2178 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
2180 *ent_cpu = cpu_file;
2185 for_each_tracing_cpu(cpu) {
2187 if (ring_buffer_empty_cpu(buffer, cpu))
2190 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
2193 * Pick the entry with the smallest timestamp:
2195 if (ent && (!next || ts < next_ts)) {
2199 next_lost = lost_events;
2200 next_size = iter->ent_size;
2204 iter->ent_size = next_size;
2207 *ent_cpu = next_cpu;
2213 *missing_events = next_lost;
2218 /* Find the next real entry, without updating the iterator itself */
2219 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
2220 int *ent_cpu, u64 *ent_ts)
2222 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
2225 /* Find the next real entry, and increment the iterator to the next entry */
2226 void *trace_find_next_entry_inc(struct trace_iterator *iter)
2228 iter->ent = __find_next_entry(iter, &iter->cpu,
2229 &iter->lost_events, &iter->ts);
2232 trace_iterator_increment(iter);
2234 return iter->ent ? iter : NULL;
2237 static void trace_consume(struct trace_iterator *iter)
2239 ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu, &iter->ts,
2240 &iter->lost_events);
2243 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
2245 struct trace_iterator *iter = m->private;
2249 WARN_ON_ONCE(iter->leftover);
2253 /* can't go backwards */
2258 ent = trace_find_next_entry_inc(iter);
2262 while (ent && iter->idx < i)
2263 ent = trace_find_next_entry_inc(iter);
2270 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
2272 struct ring_buffer_event *event;
2273 struct ring_buffer_iter *buf_iter;
2274 unsigned long entries = 0;
2277 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = 0;
2279 buf_iter = trace_buffer_iter(iter, cpu);
2283 ring_buffer_iter_reset(buf_iter);
2286 * We could have the case with the max latency tracers
2287 * that a reset never took place on a cpu. This is evident
2288 * by the timestamp being before the start of the buffer.
2290 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
2291 if (ts >= iter->trace_buffer->time_start)
2294 ring_buffer_read(buf_iter, NULL);
2297 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = entries;
2301 * The current tracer is copied to avoid a global locking
2304 static void *s_start(struct seq_file *m, loff_t *pos)
2306 struct trace_iterator *iter = m->private;
2307 struct trace_array *tr = iter->tr;
2308 int cpu_file = iter->cpu_file;
2314 * copy the tracer to avoid using a global lock all around.
2315 * iter->trace is a copy of current_trace, the pointer to the
2316 * name may be used instead of a strcmp(), as iter->trace->name
2317 * will point to the same string as current_trace->name.
2319 mutex_lock(&trace_types_lock);
2320 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
2321 *iter->trace = *tr->current_trace;
2322 mutex_unlock(&trace_types_lock);
2324 #ifdef CONFIG_TRACER_MAX_TRACE
2325 if (iter->snapshot && iter->trace->use_max_tr)
2326 return ERR_PTR(-EBUSY);
2329 if (!iter->snapshot)
2330 atomic_inc(&trace_record_cmdline_disabled);
2332 if (*pos != iter->pos) {
2337 if (cpu_file == RING_BUFFER_ALL_CPUS) {
2338 for_each_tracing_cpu(cpu)
2339 tracing_iter_reset(iter, cpu);
2341 tracing_iter_reset(iter, cpu_file);
2344 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
2349 * If we overflowed the seq_file before, then we want
2350 * to just reuse the trace_seq buffer again.
2356 p = s_next(m, p, &l);
2360 trace_event_read_lock();
2361 trace_access_lock(cpu_file);
2365 static void s_stop(struct seq_file *m, void *p)
2367 struct trace_iterator *iter = m->private;
2369 #ifdef CONFIG_TRACER_MAX_TRACE
2370 if (iter->snapshot && iter->trace->use_max_tr)
2374 if (!iter->snapshot)
2375 atomic_dec(&trace_record_cmdline_disabled);
2377 trace_access_unlock(iter->cpu_file);
2378 trace_event_read_unlock();
2382 get_total_entries(struct trace_buffer *buf,
2383 unsigned long *total, unsigned long *entries)
2385 unsigned long count;
2391 for_each_tracing_cpu(cpu) {
2392 count = ring_buffer_entries_cpu(buf->buffer, cpu);
2394 * If this buffer has skipped entries, then we hold all
2395 * entries for the trace and we need to ignore the
2396 * ones before the time stamp.
2398 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
2399 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
2400 /* total is the same as the entries */
2404 ring_buffer_overrun_cpu(buf->buffer, cpu);
2409 static void print_lat_help_header(struct seq_file *m)
2411 seq_puts(m, "# _------=> CPU# \n");
2412 seq_puts(m, "# / _-----=> irqs-off \n");
2413 seq_puts(m, "# | / _----=> need-resched \n");
2414 seq_puts(m, "# || / _---=> hardirq/softirq \n");
2415 seq_puts(m, "# ||| / _--=> preempt-depth \n");
2416 seq_puts(m, "# |||| / delay \n");
2417 seq_puts(m, "# cmd pid ||||| time | caller \n");
2418 seq_puts(m, "# \\ / ||||| \\ | / \n");
2421 static void print_event_info(struct trace_buffer *buf, struct seq_file *m)
2423 unsigned long total;
2424 unsigned long entries;
2426 get_total_entries(buf, &total, &entries);
2427 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2428 entries, total, num_online_cpus());
2432 static void print_func_help_header(struct trace_buffer *buf, struct seq_file *m)
2434 print_event_info(buf, m);
2435 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2436 seq_puts(m, "# | | | | |\n");
2439 static void print_func_help_header_irq(struct trace_buffer *buf, struct seq_file *m)
2441 print_event_info(buf, m);
2442 seq_puts(m, "# _-----=> irqs-off\n");
2443 seq_puts(m, "# / _----=> need-resched\n");
2444 seq_puts(m, "# | / _---=> hardirq/softirq\n");
2445 seq_puts(m, "# || / _--=> preempt-depth\n");
2446 seq_puts(m, "# ||| / delay\n");
2447 seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2448 seq_puts(m, "# | | | |||| | |\n");
2452 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2454 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2455 struct trace_buffer *buf = iter->trace_buffer;
2456 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
2457 struct tracer *type = iter->trace;
2458 unsigned long entries;
2459 unsigned long total;
2460 const char *name = "preemption";
2464 get_total_entries(buf, &total, &entries);
2466 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2468 seq_puts(m, "# -----------------------------------"
2469 "---------------------------------\n");
2470 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2471 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2472 nsecs_to_usecs(data->saved_latency),
2476 #if defined(CONFIG_PREEMPT_NONE)
2478 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2480 #elif defined(CONFIG_PREEMPT)
2485 /* These are reserved for later use */
2488 seq_printf(m, " #P:%d)\n", num_online_cpus());
2492 seq_puts(m, "# -----------------\n");
2493 seq_printf(m, "# | task: %.16s-%d "
2494 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2495 data->comm, data->pid,
2496 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
2497 data->policy, data->rt_priority);
2498 seq_puts(m, "# -----------------\n");
2500 if (data->critical_start) {
2501 seq_puts(m, "# => started at: ");
2502 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2503 trace_print_seq(m, &iter->seq);
2504 seq_puts(m, "\n# => ended at: ");
2505 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2506 trace_print_seq(m, &iter->seq);
2507 seq_puts(m, "\n#\n");
2513 static void test_cpu_buff_start(struct trace_iterator *iter)
2515 struct trace_seq *s = &iter->seq;
2517 if (!(trace_flags & TRACE_ITER_ANNOTATE))
2520 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2523 if (cpumask_test_cpu(iter->cpu, iter->started))
2526 if (per_cpu_ptr(iter->trace_buffer->data, iter->cpu)->skipped_entries)
2529 cpumask_set_cpu(iter->cpu, iter->started);
2531 /* Don't print started cpu buffer for the first entry of the trace */
2533 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2537 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2539 struct trace_seq *s = &iter->seq;
2540 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2541 struct trace_entry *entry;
2542 struct trace_event *event;
2546 test_cpu_buff_start(iter);
2548 event = ftrace_find_event(entry->type);
2550 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2551 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2552 if (!trace_print_lat_context(iter))
2555 if (!trace_print_context(iter))
2561 return event->funcs->trace(iter, sym_flags, event);
2563 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2566 return TRACE_TYPE_HANDLED;
2568 return TRACE_TYPE_PARTIAL_LINE;
2571 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2573 struct trace_seq *s = &iter->seq;
2574 struct trace_entry *entry;
2575 struct trace_event *event;
2579 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2580 if (!trace_seq_printf(s, "%d %d %llu ",
2581 entry->pid, iter->cpu, iter->ts))
2585 event = ftrace_find_event(entry->type);
2587 return event->funcs->raw(iter, 0, event);
2589 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2592 return TRACE_TYPE_HANDLED;
2594 return TRACE_TYPE_PARTIAL_LINE;
2597 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2599 struct trace_seq *s = &iter->seq;
2600 unsigned char newline = '\n';
2601 struct trace_entry *entry;
2602 struct trace_event *event;
2606 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2607 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2608 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2609 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2612 event = ftrace_find_event(entry->type);
2614 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2615 if (ret != TRACE_TYPE_HANDLED)
2619 SEQ_PUT_FIELD_RET(s, newline);
2621 return TRACE_TYPE_HANDLED;
2624 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2626 struct trace_seq *s = &iter->seq;
2627 struct trace_entry *entry;
2628 struct trace_event *event;
2632 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2633 SEQ_PUT_FIELD_RET(s, entry->pid);
2634 SEQ_PUT_FIELD_RET(s, iter->cpu);
2635 SEQ_PUT_FIELD_RET(s, iter->ts);
2638 event = ftrace_find_event(entry->type);
2639 return event ? event->funcs->binary(iter, 0, event) :
2643 int trace_empty(struct trace_iterator *iter)
2645 struct ring_buffer_iter *buf_iter;
2648 /* If we are looking at one CPU buffer, only check that one */
2649 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
2650 cpu = iter->cpu_file;
2651 buf_iter = trace_buffer_iter(iter, cpu);
2653 if (!ring_buffer_iter_empty(buf_iter))
2656 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2662 for_each_tracing_cpu(cpu) {
2663 buf_iter = trace_buffer_iter(iter, cpu);
2665 if (!ring_buffer_iter_empty(buf_iter))
2668 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2676 /* Called with trace_event_read_lock() held. */
2677 enum print_line_t print_trace_line(struct trace_iterator *iter)
2679 enum print_line_t ret;
2681 if (iter->lost_events &&
2682 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2683 iter->cpu, iter->lost_events))
2684 return TRACE_TYPE_PARTIAL_LINE;
2686 if (iter->trace && iter->trace->print_line) {
2687 ret = iter->trace->print_line(iter);
2688 if (ret != TRACE_TYPE_UNHANDLED)
2692 if (iter->ent->type == TRACE_BPUTS &&
2693 trace_flags & TRACE_ITER_PRINTK &&
2694 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2695 return trace_print_bputs_msg_only(iter);
2697 if (iter->ent->type == TRACE_BPRINT &&
2698 trace_flags & TRACE_ITER_PRINTK &&
2699 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2700 return trace_print_bprintk_msg_only(iter);
2702 if (iter->ent->type == TRACE_PRINT &&
2703 trace_flags & TRACE_ITER_PRINTK &&
2704 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2705 return trace_print_printk_msg_only(iter);
2707 if (trace_flags & TRACE_ITER_BIN)
2708 return print_bin_fmt(iter);
2710 if (trace_flags & TRACE_ITER_HEX)
2711 return print_hex_fmt(iter);
2713 if (trace_flags & TRACE_ITER_RAW)
2714 return print_raw_fmt(iter);
2716 return print_trace_fmt(iter);
2719 void trace_latency_header(struct seq_file *m)
2721 struct trace_iterator *iter = m->private;
2723 /* print nothing if the buffers are empty */
2724 if (trace_empty(iter))
2727 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2728 print_trace_header(m, iter);
2730 if (!(trace_flags & TRACE_ITER_VERBOSE))
2731 print_lat_help_header(m);
2734 void trace_default_header(struct seq_file *m)
2736 struct trace_iterator *iter = m->private;
2738 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2741 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2742 /* print nothing if the buffers are empty */
2743 if (trace_empty(iter))
2745 print_trace_header(m, iter);
2746 if (!(trace_flags & TRACE_ITER_VERBOSE))
2747 print_lat_help_header(m);
2749 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2750 if (trace_flags & TRACE_ITER_IRQ_INFO)
2751 print_func_help_header_irq(iter->trace_buffer, m);
2753 print_func_help_header(iter->trace_buffer, m);
2758 static void test_ftrace_alive(struct seq_file *m)
2760 if (!ftrace_is_dead())
2762 seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2763 seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2766 #ifdef CONFIG_TRACER_MAX_TRACE
2767 static void show_snapshot_main_help(struct seq_file *m)
2769 seq_printf(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n");
2770 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2771 seq_printf(m, "# Takes a snapshot of the main buffer.\n");
2772 seq_printf(m, "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n");
2773 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2774 seq_printf(m, "# is not a '0' or '1')\n");
2777 static void show_snapshot_percpu_help(struct seq_file *m)
2779 seq_printf(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
2780 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
2781 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2782 seq_printf(m, "# Takes a snapshot of the main buffer for this cpu.\n");
2784 seq_printf(m, "# echo 1 > snapshot : Not supported with this kernel.\n");
2785 seq_printf(m, "# Must use main snapshot file to allocate.\n");
2787 seq_printf(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n");
2788 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2789 seq_printf(m, "# is not a '0' or '1')\n");
2792 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
2794 if (iter->tr->allocated_snapshot)
2795 seq_printf(m, "#\n# * Snapshot is allocated *\n#\n");
2797 seq_printf(m, "#\n# * Snapshot is freed *\n#\n");
2799 seq_printf(m, "# Snapshot commands:\n");
2800 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
2801 show_snapshot_main_help(m);
2803 show_snapshot_percpu_help(m);
2806 /* Should never be called */
2807 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
2810 static int s_show(struct seq_file *m, void *v)
2812 struct trace_iterator *iter = v;
2815 if (iter->ent == NULL) {
2817 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2819 test_ftrace_alive(m);
2821 if (iter->snapshot && trace_empty(iter))
2822 print_snapshot_help(m, iter);
2823 else if (iter->trace && iter->trace->print_header)
2824 iter->trace->print_header(m);
2826 trace_default_header(m);
2828 } else if (iter->leftover) {
2830 * If we filled the seq_file buffer earlier, we
2831 * want to just show it now.
2833 ret = trace_print_seq(m, &iter->seq);
2835 /* ret should this time be zero, but you never know */
2836 iter->leftover = ret;
2839 print_trace_line(iter);
2840 ret = trace_print_seq(m, &iter->seq);
2842 * If we overflow the seq_file buffer, then it will
2843 * ask us for this data again at start up.
2845 * ret is 0 if seq_file write succeeded.
2848 iter->leftover = ret;
2855 * Should be used after trace_array_get(), trace_types_lock
2856 * ensures that i_cdev was already initialized.
2858 static inline int tracing_get_cpu(struct inode *inode)
2860 if (inode->i_cdev) /* See trace_create_cpu_file() */
2861 return (long)inode->i_cdev - 1;
2862 return RING_BUFFER_ALL_CPUS;
2865 static const struct seq_operations tracer_seq_ops = {
2872 static struct trace_iterator *
2873 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
2875 struct trace_array *tr = inode->i_private;
2876 struct trace_iterator *iter;
2879 if (tracing_disabled)
2880 return ERR_PTR(-ENODEV);
2882 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
2884 return ERR_PTR(-ENOMEM);
2886 iter->buffer_iter = kzalloc(sizeof(*iter->buffer_iter) * num_possible_cpus(),
2888 if (!iter->buffer_iter)
2892 * We make a copy of the current tracer to avoid concurrent
2893 * changes on it while we are reading.
2895 mutex_lock(&trace_types_lock);
2896 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2900 *iter->trace = *tr->current_trace;
2902 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2907 #ifdef CONFIG_TRACER_MAX_TRACE
2908 /* Currently only the top directory has a snapshot */
2909 if (tr->current_trace->print_max || snapshot)
2910 iter->trace_buffer = &tr->max_buffer;
2913 iter->trace_buffer = &tr->trace_buffer;
2914 iter->snapshot = snapshot;
2916 iter->cpu_file = tracing_get_cpu(inode);
2917 mutex_init(&iter->mutex);
2919 /* Notify the tracer early; before we stop tracing. */
2920 if (iter->trace && iter->trace->open)
2921 iter->trace->open(iter);
2923 /* Annotate start of buffers if we had overruns */
2924 if (ring_buffer_overruns(iter->trace_buffer->buffer))
2925 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2927 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
2928 if (trace_clocks[tr->clock_id].in_ns)
2929 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
2931 /* stop the trace while dumping if we are not opening "snapshot" */
2932 if (!iter->snapshot)
2933 tracing_stop_tr(tr);
2935 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
2936 for_each_tracing_cpu(cpu) {
2937 iter->buffer_iter[cpu] =
2938 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
2940 ring_buffer_read_prepare_sync();
2941 for_each_tracing_cpu(cpu) {
2942 ring_buffer_read_start(iter->buffer_iter[cpu]);
2943 tracing_iter_reset(iter, cpu);
2946 cpu = iter->cpu_file;
2947 iter->buffer_iter[cpu] =
2948 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
2949 ring_buffer_read_prepare_sync();
2950 ring_buffer_read_start(iter->buffer_iter[cpu]);
2951 tracing_iter_reset(iter, cpu);
2954 mutex_unlock(&trace_types_lock);
2959 mutex_unlock(&trace_types_lock);
2961 kfree(iter->buffer_iter);
2963 seq_release_private(inode, file);
2964 return ERR_PTR(-ENOMEM);
2967 int tracing_open_generic(struct inode *inode, struct file *filp)
2969 if (tracing_disabled)
2972 filp->private_data = inode->i_private;
2976 bool tracing_is_disabled(void)
2978 return (tracing_disabled) ? true: false;
2982 * Open and update trace_array ref count.
2983 * Must have the current trace_array passed to it.
2985 static int tracing_open_generic_tr(struct inode *inode, struct file *filp)
2987 struct trace_array *tr = inode->i_private;
2989 if (tracing_disabled)
2992 if (trace_array_get(tr) < 0)
2995 filp->private_data = inode->i_private;
3000 static int tracing_release(struct inode *inode, struct file *file)
3002 struct trace_array *tr = inode->i_private;
3003 struct seq_file *m = file->private_data;
3004 struct trace_iterator *iter;
3007 if (!(file->f_mode & FMODE_READ)) {
3008 trace_array_put(tr);
3012 /* Writes do not use seq_file */
3014 mutex_lock(&trace_types_lock);
3016 for_each_tracing_cpu(cpu) {
3017 if (iter->buffer_iter[cpu])
3018 ring_buffer_read_finish(iter->buffer_iter[cpu]);
3021 if (iter->trace && iter->trace->close)
3022 iter->trace->close(iter);
3024 if (!iter->snapshot)
3025 /* reenable tracing if it was previously enabled */
3026 tracing_start_tr(tr);
3028 __trace_array_put(tr);
3030 mutex_unlock(&trace_types_lock);
3032 mutex_destroy(&iter->mutex);
3033 free_cpumask_var(iter->started);
3035 kfree(iter->buffer_iter);
3036 seq_release_private(inode, file);
3041 static int tracing_release_generic_tr(struct inode *inode, struct file *file)
3043 struct trace_array *tr = inode->i_private;
3045 trace_array_put(tr);
3049 static int tracing_single_release_tr(struct inode *inode, struct file *file)
3051 struct trace_array *tr = inode->i_private;
3053 trace_array_put(tr);
3055 return single_release(inode, file);
3058 static int tracing_open(struct inode *inode, struct file *file)
3060 struct trace_array *tr = inode->i_private;
3061 struct trace_iterator *iter;
3064 if (trace_array_get(tr) < 0)
3067 /* If this file was open for write, then erase contents */
3068 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
3069 int cpu = tracing_get_cpu(inode);
3071 if (cpu == RING_BUFFER_ALL_CPUS)
3072 tracing_reset_online_cpus(&tr->trace_buffer);
3074 tracing_reset(&tr->trace_buffer, cpu);
3077 if (file->f_mode & FMODE_READ) {
3078 iter = __tracing_open(inode, file, false);
3080 ret = PTR_ERR(iter);
3081 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
3082 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3086 trace_array_put(tr);
3092 t_next(struct seq_file *m, void *v, loff_t *pos)
3094 struct tracer *t = v;
3104 static void *t_start(struct seq_file *m, loff_t *pos)
3109 mutex_lock(&trace_types_lock);
3110 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
3116 static void t_stop(struct seq_file *m, void *p)
3118 mutex_unlock(&trace_types_lock);
3121 static int t_show(struct seq_file *m, void *v)
3123 struct tracer *t = v;
3128 seq_printf(m, "%s", t->name);
3137 static const struct seq_operations show_traces_seq_ops = {
3144 static int show_traces_open(struct inode *inode, struct file *file)
3146 if (tracing_disabled)
3149 return seq_open(file, &show_traces_seq_ops);
3153 tracing_write_stub(struct file *filp, const char __user *ubuf,
3154 size_t count, loff_t *ppos)
3159 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
3161 if (file->f_mode & FMODE_READ)
3162 return seq_lseek(file, offset, origin);
3167 static const struct file_operations tracing_fops = {
3168 .open = tracing_open,
3170 .write = tracing_write_stub,
3171 .llseek = tracing_seek,
3172 .release = tracing_release,
3175 static const struct file_operations show_traces_fops = {
3176 .open = show_traces_open,
3178 .release = seq_release,
3179 .llseek = seq_lseek,
3183 * The tracer itself will not take this lock, but still we want
3184 * to provide a consistent cpumask to user-space:
3186 static DEFINE_MUTEX(tracing_cpumask_update_lock);
3189 * Temporary storage for the character representation of the
3190 * CPU bitmask (and one more byte for the newline):
3192 static char mask_str[NR_CPUS + 1];
3195 tracing_cpumask_read(struct file *filp, char __user *ubuf,
3196 size_t count, loff_t *ppos)
3198 struct trace_array *tr = file_inode(filp)->i_private;
3201 mutex_lock(&tracing_cpumask_update_lock);
3203 len = cpumask_scnprintf(mask_str, count, tr->tracing_cpumask);
3204 if (count - len < 2) {
3208 len += sprintf(mask_str + len, "\n");
3209 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
3212 mutex_unlock(&tracing_cpumask_update_lock);
3218 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
3219 size_t count, loff_t *ppos)
3221 struct trace_array *tr = file_inode(filp)->i_private;
3222 cpumask_var_t tracing_cpumask_new;
3225 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
3228 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
3232 mutex_lock(&tracing_cpumask_update_lock);
3234 local_irq_disable();
3235 arch_spin_lock(&ftrace_max_lock);
3236 for_each_tracing_cpu(cpu) {
3238 * Increase/decrease the disabled counter if we are
3239 * about to flip a bit in the cpumask:
3241 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
3242 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3243 atomic_inc(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3244 ring_buffer_record_disable_cpu(tr->trace_buffer.buffer, cpu);
3246 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
3247 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3248 atomic_dec(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3249 ring_buffer_record_enable_cpu(tr->trace_buffer.buffer, cpu);
3252 arch_spin_unlock(&ftrace_max_lock);
3255 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
3257 mutex_unlock(&tracing_cpumask_update_lock);
3258 free_cpumask_var(tracing_cpumask_new);
3263 free_cpumask_var(tracing_cpumask_new);
3268 static const struct file_operations tracing_cpumask_fops = {
3269 .open = tracing_open_generic_tr,
3270 .read = tracing_cpumask_read,
3271 .write = tracing_cpumask_write,
3272 .release = tracing_release_generic_tr,
3273 .llseek = generic_file_llseek,
3276 static int tracing_trace_options_show(struct seq_file *m, void *v)
3278 struct tracer_opt *trace_opts;
3279 struct trace_array *tr = m->private;
3283 mutex_lock(&trace_types_lock);
3284 tracer_flags = tr->current_trace->flags->val;
3285 trace_opts = tr->current_trace->flags->opts;
3287 for (i = 0; trace_options[i]; i++) {
3288 if (trace_flags & (1 << i))
3289 seq_printf(m, "%s\n", trace_options[i]);
3291 seq_printf(m, "no%s\n", trace_options[i]);
3294 for (i = 0; trace_opts[i].name; i++) {
3295 if (tracer_flags & trace_opts[i].bit)
3296 seq_printf(m, "%s\n", trace_opts[i].name);
3298 seq_printf(m, "no%s\n", trace_opts[i].name);
3300 mutex_unlock(&trace_types_lock);
3305 static int __set_tracer_option(struct tracer *trace,
3306 struct tracer_flags *tracer_flags,
3307 struct tracer_opt *opts, int neg)
3311 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
3316 tracer_flags->val &= ~opts->bit;
3318 tracer_flags->val |= opts->bit;
3322 /* Try to assign a tracer specific option */
3323 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
3325 struct tracer_flags *tracer_flags = trace->flags;
3326 struct tracer_opt *opts = NULL;
3329 for (i = 0; tracer_flags->opts[i].name; i++) {
3330 opts = &tracer_flags->opts[i];
3332 if (strcmp(cmp, opts->name) == 0)
3333 return __set_tracer_option(trace, trace->flags,
3340 /* Some tracers require overwrite to stay enabled */
3341 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
3343 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
3349 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
3351 /* do nothing if flag is already set */
3352 if (!!(trace_flags & mask) == !!enabled)
3355 /* Give the tracer a chance to approve the change */
3356 if (tr->current_trace->flag_changed)
3357 if (tr->current_trace->flag_changed(tr->current_trace, mask, !!enabled))
3361 trace_flags |= mask;
3363 trace_flags &= ~mask;
3365 if (mask == TRACE_ITER_RECORD_CMD)
3366 trace_event_enable_cmd_record(enabled);
3368 if (mask == TRACE_ITER_OVERWRITE) {
3369 ring_buffer_change_overwrite(tr->trace_buffer.buffer, enabled);
3370 #ifdef CONFIG_TRACER_MAX_TRACE
3371 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
3375 if (mask == TRACE_ITER_PRINTK)
3376 trace_printk_start_stop_comm(enabled);
3381 static int trace_set_options(struct trace_array *tr, char *option)
3388 cmp = strstrip(option);
3390 if (strncmp(cmp, "no", 2) == 0) {
3395 mutex_lock(&trace_types_lock);
3397 for (i = 0; trace_options[i]; i++) {
3398 if (strcmp(cmp, trace_options[i]) == 0) {
3399 ret = set_tracer_flag(tr, 1 << i, !neg);
3404 /* If no option could be set, test the specific tracer options */
3405 if (!trace_options[i])
3406 ret = set_tracer_option(tr->current_trace, cmp, neg);
3408 mutex_unlock(&trace_types_lock);
3414 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
3415 size_t cnt, loff_t *ppos)
3417 struct seq_file *m = filp->private_data;
3418 struct trace_array *tr = m->private;
3422 if (cnt >= sizeof(buf))
3425 if (copy_from_user(&buf, ubuf, cnt))
3430 ret = trace_set_options(tr, buf);
3439 static int tracing_trace_options_open(struct inode *inode, struct file *file)
3441 struct trace_array *tr = inode->i_private;
3444 if (tracing_disabled)
3447 if (trace_array_get(tr) < 0)
3450 ret = single_open(file, tracing_trace_options_show, inode->i_private);
3452 trace_array_put(tr);
3457 static const struct file_operations tracing_iter_fops = {
3458 .open = tracing_trace_options_open,
3460 .llseek = seq_lseek,
3461 .release = tracing_single_release_tr,
3462 .write = tracing_trace_options_write,
3465 static const char readme_msg[] =
3466 "tracing mini-HOWTO:\n\n"
3467 "# echo 0 > tracing_on : quick way to disable tracing\n"
3468 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
3469 " Important files:\n"
3470 " trace\t\t\t- The static contents of the buffer\n"
3471 "\t\t\t To clear the buffer write into this file: echo > trace\n"
3472 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
3473 " current_tracer\t- function and latency tracers\n"
3474 " available_tracers\t- list of configured tracers for current_tracer\n"
3475 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
3476 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
3477 " trace_clock\t\t-change the clock used to order events\n"
3478 " local: Per cpu clock but may not be synced across CPUs\n"
3479 " global: Synced across CPUs but slows tracing down.\n"
3480 " counter: Not a clock, but just an increment\n"
3481 " uptime: Jiffy counter from time of boot\n"
3482 " perf: Same clock that perf events use\n"
3483 #ifdef CONFIG_X86_64
3484 " x86-tsc: TSC cycle counter\n"
3486 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
3487 " tracing_cpumask\t- Limit which CPUs to trace\n"
3488 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
3489 "\t\t\t Remove sub-buffer with rmdir\n"
3490 " trace_options\t\t- Set format or modify how tracing happens\n"
3491 "\t\t\t Disable an option by adding a suffix 'no' to the option name\n"
3492 #ifdef CONFIG_DYNAMIC_FTRACE
3493 "\n available_filter_functions - list of functions that can be filtered on\n"
3494 " set_ftrace_filter\t- echo function name in here to only trace these functions\n"
3495 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3496 " modules: Can select a group via module\n"
3497 " Format: :mod:<module-name>\n"
3498 " example: echo :mod:ext3 > set_ftrace_filter\n"
3499 " triggers: a command to perform when function is hit\n"
3500 " Format: <function>:<trigger>[:count]\n"
3501 " trigger: traceon, traceoff\n"
3502 " enable_event:<system>:<event>\n"
3503 " disable_event:<system>:<event>\n"
3504 #ifdef CONFIG_STACKTRACE
3507 #ifdef CONFIG_TRACER_SNAPSHOT
3510 " example: echo do_fault:traceoff > set_ftrace_filter\n"
3511 " echo do_trap:traceoff:3 > set_ftrace_filter\n"
3512 " The first one will disable tracing every time do_fault is hit\n"
3513 " The second will disable tracing at most 3 times when do_trap is hit\n"
3514 " The first time do trap is hit and it disables tracing, the counter\n"
3515 " will decrement to 2. If tracing is already disabled, the counter\n"
3516 " will not decrement. It only decrements when the trigger did work\n"
3517 " To remove trigger without count:\n"
3518 " echo '!<function>:<trigger> > set_ftrace_filter\n"
3519 " To remove trigger with a count:\n"
3520 " echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
3521 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
3522 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3523 " modules: Can select a group via module command :mod:\n"
3524 " Does not accept triggers\n"
3525 #endif /* CONFIG_DYNAMIC_FTRACE */
3526 #ifdef CONFIG_FUNCTION_TRACER
3527 " set_ftrace_pid\t- Write pid(s) to only function trace those pids (function)\n"
3529 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3530 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
3531 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
3533 #ifdef CONFIG_TRACER_SNAPSHOT
3534 "\n snapshot\t\t- Like 'trace' but shows the content of the static snapshot buffer\n"
3535 "\t\t\t Read the contents for more information\n"
3537 #ifdef CONFIG_STACK_TRACER
3538 " stack_trace\t\t- Shows the max stack trace when active\n"
3539 " stack_max_size\t- Shows current max stack size that was traced\n"
3540 "\t\t\t Write into this file to reset the max size (trigger a new trace)\n"
3541 #ifdef CONFIG_DYNAMIC_FTRACE
3542 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace traces\n"
3544 #endif /* CONFIG_STACK_TRACER */
3548 tracing_readme_read(struct file *filp, char __user *ubuf,
3549 size_t cnt, loff_t *ppos)
3551 return simple_read_from_buffer(ubuf, cnt, ppos,
3552 readme_msg, strlen(readme_msg));
3555 static const struct file_operations tracing_readme_fops = {
3556 .open = tracing_open_generic,
3557 .read = tracing_readme_read,
3558 .llseek = generic_file_llseek,
3562 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
3563 size_t cnt, loff_t *ppos)
3572 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
3576 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
3584 for (i = 0; i < SAVED_CMDLINES; i++) {
3587 pid = map_cmdline_to_pid[i];
3588 if (pid == -1 || pid == NO_CMDLINE_MAP)
3591 trace_find_cmdline(pid, buf_comm);
3592 r = sprintf(buf, "%d %s\n", pid, buf_comm);
3597 len = simple_read_from_buffer(ubuf, cnt, ppos,
3606 static const struct file_operations tracing_saved_cmdlines_fops = {
3607 .open = tracing_open_generic,
3608 .read = tracing_saved_cmdlines_read,
3609 .llseek = generic_file_llseek,
3613 tracing_set_trace_read(struct file *filp, char __user *ubuf,
3614 size_t cnt, loff_t *ppos)
3616 struct trace_array *tr = filp->private_data;
3617 char buf[MAX_TRACER_SIZE+2];
3620 mutex_lock(&trace_types_lock);
3621 r = sprintf(buf, "%s\n", tr->current_trace->name);
3622 mutex_unlock(&trace_types_lock);
3624 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3627 int tracer_init(struct tracer *t, struct trace_array *tr)
3629 tracing_reset_online_cpus(&tr->trace_buffer);
3633 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val)
3637 for_each_tracing_cpu(cpu)
3638 per_cpu_ptr(buf->data, cpu)->entries = val;
3641 #ifdef CONFIG_TRACER_MAX_TRACE
3642 /* resize @tr's buffer to the size of @size_tr's entries */
3643 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
3644 struct trace_buffer *size_buf, int cpu_id)
3648 if (cpu_id == RING_BUFFER_ALL_CPUS) {
3649 for_each_tracing_cpu(cpu) {
3650 ret = ring_buffer_resize(trace_buf->buffer,
3651 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
3654 per_cpu_ptr(trace_buf->data, cpu)->entries =
3655 per_cpu_ptr(size_buf->data, cpu)->entries;
3658 ret = ring_buffer_resize(trace_buf->buffer,
3659 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
3661 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
3662 per_cpu_ptr(size_buf->data, cpu_id)->entries;
3667 #endif /* CONFIG_TRACER_MAX_TRACE */
3669 static int __tracing_resize_ring_buffer(struct trace_array *tr,
3670 unsigned long size, int cpu)
3675 * If kernel or user changes the size of the ring buffer
3676 * we use the size that was given, and we can forget about
3677 * expanding it later.
3679 ring_buffer_expanded = true;
3681 /* May be called before buffers are initialized */
3682 if (!tr->trace_buffer.buffer)
3685 ret = ring_buffer_resize(tr->trace_buffer.buffer, size, cpu);
3689 #ifdef CONFIG_TRACER_MAX_TRACE
3690 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
3691 !tr->current_trace->use_max_tr)
3694 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
3696 int r = resize_buffer_duplicate_size(&tr->trace_buffer,
3697 &tr->trace_buffer, cpu);
3700 * AARGH! We are left with different
3701 * size max buffer!!!!
3702 * The max buffer is our "snapshot" buffer.
3703 * When a tracer needs a snapshot (one of the
3704 * latency tracers), it swaps the max buffer
3705 * with the saved snap shot. We succeeded to
3706 * update the size of the main buffer, but failed to
3707 * update the size of the max buffer. But when we tried
3708 * to reset the main buffer to the original size, we
3709 * failed there too. This is very unlikely to
3710 * happen, but if it does, warn and kill all
3714 tracing_disabled = 1;
3719 if (cpu == RING_BUFFER_ALL_CPUS)
3720 set_buffer_entries(&tr->max_buffer, size);
3722 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
3725 #endif /* CONFIG_TRACER_MAX_TRACE */
3727 if (cpu == RING_BUFFER_ALL_CPUS)
3728 set_buffer_entries(&tr->trace_buffer, size);
3730 per_cpu_ptr(tr->trace_buffer.data, cpu)->entries = size;
3735 static ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
3736 unsigned long size, int cpu_id)
3740 mutex_lock(&trace_types_lock);
3742 if (cpu_id != RING_BUFFER_ALL_CPUS) {
3743 /* make sure, this cpu is enabled in the mask */
3744 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
3750 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
3755 mutex_unlock(&trace_types_lock);
3762 * tracing_update_buffers - used by tracing facility to expand ring buffers
3764 * To save on memory when the tracing is never used on a system with it
3765 * configured in. The ring buffers are set to a minimum size. But once
3766 * a user starts to use the tracing facility, then they need to grow
3767 * to their default size.
3769 * This function is to be called when a tracer is about to be used.
3771 int tracing_update_buffers(void)
3775 mutex_lock(&trace_types_lock);
3776 if (!ring_buffer_expanded)
3777 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
3778 RING_BUFFER_ALL_CPUS);
3779 mutex_unlock(&trace_types_lock);
3784 struct trace_option_dentry;
3786 static struct trace_option_dentry *
3787 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
3790 destroy_trace_option_files(struct trace_option_dentry *topts);
3792 static int tracing_set_tracer(const char *buf)
3794 static struct trace_option_dentry *topts;
3795 struct trace_array *tr = &global_trace;
3797 #ifdef CONFIG_TRACER_MAX_TRACE
3802 mutex_lock(&trace_types_lock);
3804 if (!ring_buffer_expanded) {
3805 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
3806 RING_BUFFER_ALL_CPUS);
3812 for (t = trace_types; t; t = t->next) {
3813 if (strcmp(t->name, buf) == 0)
3820 if (t == tr->current_trace)
3823 trace_branch_disable();
3825 tr->current_trace->enabled = false;
3827 if (tr->current_trace->reset)
3828 tr->current_trace->reset(tr);
3830 /* Current trace needs to be nop_trace before synchronize_sched */
3831 tr->current_trace = &nop_trace;
3833 #ifdef CONFIG_TRACER_MAX_TRACE
3834 had_max_tr = tr->allocated_snapshot;
3836 if (had_max_tr && !t->use_max_tr) {
3838 * We need to make sure that the update_max_tr sees that
3839 * current_trace changed to nop_trace to keep it from
3840 * swapping the buffers after we resize it.
3841 * The update_max_tr is called from interrupts disabled
3842 * so a synchronized_sched() is sufficient.
3844 synchronize_sched();
3848 destroy_trace_option_files(topts);
3850 topts = create_trace_option_files(tr, t);
3852 #ifdef CONFIG_TRACER_MAX_TRACE
3853 if (t->use_max_tr && !had_max_tr) {
3854 ret = alloc_snapshot(tr);
3861 ret = tracer_init(t, tr);
3866 tr->current_trace = t;
3867 tr->current_trace->enabled = true;
3868 trace_branch_enable(tr);
3870 mutex_unlock(&trace_types_lock);
3876 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3877 size_t cnt, loff_t *ppos)
3879 char buf[MAX_TRACER_SIZE+1];
3886 if (cnt > MAX_TRACER_SIZE)
3887 cnt = MAX_TRACER_SIZE;
3889 if (copy_from_user(&buf, ubuf, cnt))
3894 /* strip ending whitespace. */
3895 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3898 err = tracing_set_tracer(buf);
3908 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3909 size_t cnt, loff_t *ppos)
3911 unsigned long *ptr = filp->private_data;
3915 r = snprintf(buf, sizeof(buf), "%ld\n",
3916 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3917 if (r > sizeof(buf))
3919 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3923 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3924 size_t cnt, loff_t *ppos)
3926 unsigned long *ptr = filp->private_data;
3930 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3939 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3941 struct trace_array *tr = inode->i_private;
3942 struct trace_iterator *iter;
3945 if (tracing_disabled)
3948 if (trace_array_get(tr) < 0)
3951 mutex_lock(&trace_types_lock);
3953 /* create a buffer to store the information to pass to userspace */
3954 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3957 __trace_array_put(tr);
3962 * We make a copy of the current tracer to avoid concurrent
3963 * changes on it while we are reading.
3965 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3970 *iter->trace = *tr->current_trace;
3972 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3977 /* trace pipe does not show start of buffer */
3978 cpumask_setall(iter->started);
3980 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3981 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3983 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3984 if (trace_clocks[tr->clock_id].in_ns)
3985 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
3988 iter->trace_buffer = &tr->trace_buffer;
3989 iter->cpu_file = tracing_get_cpu(inode);
3990 mutex_init(&iter->mutex);
3991 filp->private_data = iter;
3993 if (iter->trace->pipe_open)
3994 iter->trace->pipe_open(iter);
3996 nonseekable_open(inode, filp);
3998 mutex_unlock(&trace_types_lock);
4004 __trace_array_put(tr);
4005 mutex_unlock(&trace_types_lock);
4009 static int tracing_release_pipe(struct inode *inode, struct file *file)
4011 struct trace_iterator *iter = file->private_data;
4012 struct trace_array *tr = inode->i_private;
4014 mutex_lock(&trace_types_lock);
4016 if (iter->trace->pipe_close)
4017 iter->trace->pipe_close(iter);
4019 mutex_unlock(&trace_types_lock);
4021 free_cpumask_var(iter->started);
4022 mutex_destroy(&iter->mutex);
4026 trace_array_put(tr);
4032 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
4034 /* Iterators are static, they should be filled or empty */
4035 if (trace_buffer_iter(iter, iter->cpu_file))
4036 return POLLIN | POLLRDNORM;
4038 if (trace_flags & TRACE_ITER_BLOCK)
4040 * Always select as readable when in blocking mode
4042 return POLLIN | POLLRDNORM;
4044 return ring_buffer_poll_wait(iter->trace_buffer->buffer, iter->cpu_file,
4049 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
4051 struct trace_iterator *iter = filp->private_data;
4053 return trace_poll(iter, filp, poll_table);
4057 * This is a make-shift waitqueue.
4058 * A tracer might use this callback on some rare cases:
4060 * 1) the current tracer might hold the runqueue lock when it wakes up
4061 * a reader, hence a deadlock (sched, function, and function graph tracers)
4062 * 2) the function tracers, trace all functions, we don't want
4063 * the overhead of calling wake_up and friends
4064 * (and tracing them too)
4066 * Anyway, this is really very primitive wakeup.
4068 void poll_wait_pipe(struct trace_iterator *iter)
4070 set_current_state(TASK_INTERRUPTIBLE);
4071 /* sleep for 100 msecs, and try again. */
4072 schedule_timeout(HZ / 10);
4075 /* Must be called with trace_types_lock mutex held. */
4076 static int tracing_wait_pipe(struct file *filp)
4078 struct trace_iterator *iter = filp->private_data;
4080 while (trace_empty(iter)) {
4082 if ((filp->f_flags & O_NONBLOCK)) {
4086 mutex_unlock(&iter->mutex);
4088 iter->trace->wait_pipe(iter);
4090 mutex_lock(&iter->mutex);
4092 if (signal_pending(current))
4096 * We block until we read something and tracing is disabled.
4097 * We still block if tracing is disabled, but we have never
4098 * read anything. This allows a user to cat this file, and
4099 * then enable tracing. But after we have read something,
4100 * we give an EOF when tracing is again disabled.
4102 * iter->pos will be 0 if we haven't read anything.
4104 if (!tracing_is_on() && iter->pos)
4115 tracing_read_pipe(struct file *filp, char __user *ubuf,
4116 size_t cnt, loff_t *ppos)
4118 struct trace_iterator *iter = filp->private_data;
4119 struct trace_array *tr = iter->tr;
4122 /* return any leftover data */
4123 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4127 trace_seq_init(&iter->seq);
4129 /* copy the tracer to avoid using a global lock all around */
4130 mutex_lock(&trace_types_lock);
4131 if (unlikely(iter->trace->name != tr->current_trace->name))
4132 *iter->trace = *tr->current_trace;
4133 mutex_unlock(&trace_types_lock);
4136 * Avoid more than one consumer on a single file descriptor
4137 * This is just a matter of traces coherency, the ring buffer itself
4140 mutex_lock(&iter->mutex);
4141 if (iter->trace->read) {
4142 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
4148 sret = tracing_wait_pipe(filp);
4152 /* stop when tracing is finished */
4153 if (trace_empty(iter)) {
4158 if (cnt >= PAGE_SIZE)
4159 cnt = PAGE_SIZE - 1;
4161 /* reset all but tr, trace, and overruns */
4162 memset(&iter->seq, 0,
4163 sizeof(struct trace_iterator) -
4164 offsetof(struct trace_iterator, seq));
4165 cpumask_clear(iter->started);
4168 trace_event_read_lock();
4169 trace_access_lock(iter->cpu_file);
4170 while (trace_find_next_entry_inc(iter) != NULL) {
4171 enum print_line_t ret;
4172 int len = iter->seq.len;
4174 ret = print_trace_line(iter);
4175 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4176 /* don't print partial lines */
4177 iter->seq.len = len;
4180 if (ret != TRACE_TYPE_NO_CONSUME)
4181 trace_consume(iter);
4183 if (iter->seq.len >= cnt)
4187 * Setting the full flag means we reached the trace_seq buffer
4188 * size and we should leave by partial output condition above.
4189 * One of the trace_seq_* functions is not used properly.
4191 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
4194 trace_access_unlock(iter->cpu_file);
4195 trace_event_read_unlock();
4197 /* Now copy what we have to the user */
4198 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4199 if (iter->seq.readpos >= iter->seq.len)
4200 trace_seq_init(&iter->seq);
4203 * If there was nothing to send to user, in spite of consuming trace
4204 * entries, go back to wait for more entries.
4210 mutex_unlock(&iter->mutex);
4215 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
4216 struct pipe_buffer *buf)
4218 __free_page(buf->page);
4221 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
4224 __free_page(spd->pages[idx]);
4227 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
4229 .map = generic_pipe_buf_map,
4230 .unmap = generic_pipe_buf_unmap,
4231 .confirm = generic_pipe_buf_confirm,
4232 .release = tracing_pipe_buf_release,
4233 .steal = generic_pipe_buf_steal,
4234 .get = generic_pipe_buf_get,
4238 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
4243 /* Seq buffer is page-sized, exactly what we need. */
4245 count = iter->seq.len;
4246 ret = print_trace_line(iter);
4247 count = iter->seq.len - count;
4250 iter->seq.len -= count;
4253 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4254 iter->seq.len -= count;
4258 if (ret != TRACE_TYPE_NO_CONSUME)
4259 trace_consume(iter);
4261 if (!trace_find_next_entry_inc(iter)) {
4271 static ssize_t tracing_splice_read_pipe(struct file *filp,
4273 struct pipe_inode_info *pipe,
4277 struct page *pages_def[PIPE_DEF_BUFFERS];
4278 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4279 struct trace_iterator *iter = filp->private_data;
4280 struct splice_pipe_desc spd = {
4282 .partial = partial_def,
4283 .nr_pages = 0, /* This gets updated below. */
4284 .nr_pages_max = PIPE_DEF_BUFFERS,
4286 .ops = &tracing_pipe_buf_ops,
4287 .spd_release = tracing_spd_release_pipe,
4289 struct trace_array *tr = iter->tr;
4294 if (splice_grow_spd(pipe, &spd))
4297 /* copy the tracer to avoid using a global lock all around */
4298 mutex_lock(&trace_types_lock);
4299 if (unlikely(iter->trace->name != tr->current_trace->name))
4300 *iter->trace = *tr->current_trace;
4301 mutex_unlock(&trace_types_lock);
4303 mutex_lock(&iter->mutex);
4305 if (iter->trace->splice_read) {
4306 ret = iter->trace->splice_read(iter, filp,
4307 ppos, pipe, len, flags);
4312 ret = tracing_wait_pipe(filp);
4316 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
4321 trace_event_read_lock();
4322 trace_access_lock(iter->cpu_file);
4324 /* Fill as many pages as possible. */
4325 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
4326 spd.pages[i] = alloc_page(GFP_KERNEL);
4330 rem = tracing_fill_pipe_page(rem, iter);
4332 /* Copy the data into the page, so we can start over. */
4333 ret = trace_seq_to_buffer(&iter->seq,
4334 page_address(spd.pages[i]),
4337 __free_page(spd.pages[i]);
4340 spd.partial[i].offset = 0;
4341 spd.partial[i].len = iter->seq.len;
4343 trace_seq_init(&iter->seq);
4346 trace_access_unlock(iter->cpu_file);
4347 trace_event_read_unlock();
4348 mutex_unlock(&iter->mutex);
4352 ret = splice_to_pipe(pipe, &spd);
4354 splice_shrink_spd(&spd);
4358 mutex_unlock(&iter->mutex);
4363 tracing_entries_read(struct file *filp, char __user *ubuf,
4364 size_t cnt, loff_t *ppos)
4366 struct inode *inode = file_inode(filp);
4367 struct trace_array *tr = inode->i_private;
4368 int cpu = tracing_get_cpu(inode);
4373 mutex_lock(&trace_types_lock);
4375 if (cpu == RING_BUFFER_ALL_CPUS) {
4376 int cpu, buf_size_same;
4381 /* check if all cpu sizes are same */
4382 for_each_tracing_cpu(cpu) {
4383 /* fill in the size from first enabled cpu */
4385 size = per_cpu_ptr(tr->trace_buffer.data, cpu)->entries;
4386 if (size != per_cpu_ptr(tr->trace_buffer.data, cpu)->entries) {
4392 if (buf_size_same) {
4393 if (!ring_buffer_expanded)
4394 r = sprintf(buf, "%lu (expanded: %lu)\n",
4396 trace_buf_size >> 10);
4398 r = sprintf(buf, "%lu\n", size >> 10);
4400 r = sprintf(buf, "X\n");
4402 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10);
4404 mutex_unlock(&trace_types_lock);
4406 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4411 tracing_entries_write(struct file *filp, const char __user *ubuf,
4412 size_t cnt, loff_t *ppos)
4414 struct inode *inode = file_inode(filp);
4415 struct trace_array *tr = inode->i_private;
4419 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4423 /* must have at least 1 entry */
4427 /* value is in KB */
4429 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
4439 tracing_total_entries_read(struct file *filp, char __user *ubuf,
4440 size_t cnt, loff_t *ppos)
4442 struct trace_array *tr = filp->private_data;
4445 unsigned long size = 0, expanded_size = 0;
4447 mutex_lock(&trace_types_lock);
4448 for_each_tracing_cpu(cpu) {
4449 size += per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10;
4450 if (!ring_buffer_expanded)
4451 expanded_size += trace_buf_size >> 10;
4453 if (ring_buffer_expanded)
4454 r = sprintf(buf, "%lu\n", size);
4456 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
4457 mutex_unlock(&trace_types_lock);
4459 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4463 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
4464 size_t cnt, loff_t *ppos)
4467 * There is no need to read what the user has written, this function
4468 * is just to make sure that there is no error when "echo" is used
4477 tracing_free_buffer_release(struct inode *inode, struct file *filp)
4479 struct trace_array *tr = inode->i_private;
4481 /* disable tracing ? */
4482 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
4483 tracer_tracing_off(tr);
4484 /* resize the ring buffer to 0 */
4485 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
4487 trace_array_put(tr);
4493 tracing_mark_write(struct file *filp, const char __user *ubuf,
4494 size_t cnt, loff_t *fpos)
4496 unsigned long addr = (unsigned long)ubuf;
4497 struct trace_array *tr = filp->private_data;
4498 struct ring_buffer_event *event;
4499 struct ring_buffer *buffer;
4500 struct print_entry *entry;
4501 unsigned long irq_flags;
4502 struct page *pages[2];
4512 if (tracing_disabled)
4515 if (!(trace_flags & TRACE_ITER_MARKERS))
4518 if (cnt > TRACE_BUF_SIZE)
4519 cnt = TRACE_BUF_SIZE;
4522 * Userspace is injecting traces into the kernel trace buffer.
4523 * We want to be as non intrusive as possible.
4524 * To do so, we do not want to allocate any special buffers
4525 * or take any locks, but instead write the userspace data
4526 * straight into the ring buffer.
4528 * First we need to pin the userspace buffer into memory,
4529 * which, most likely it is, because it just referenced it.
4530 * But there's no guarantee that it is. By using get_user_pages_fast()
4531 * and kmap_atomic/kunmap_atomic() we can get access to the
4532 * pages directly. We then write the data directly into the
4535 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
4537 /* check if we cross pages */
4538 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
4541 offset = addr & (PAGE_SIZE - 1);
4544 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
4545 if (ret < nr_pages) {
4547 put_page(pages[ret]);
4552 for (i = 0; i < nr_pages; i++)
4553 map_page[i] = kmap_atomic(pages[i]);
4555 local_save_flags(irq_flags);
4556 size = sizeof(*entry) + cnt + 2; /* possible \n added */
4557 buffer = tr->trace_buffer.buffer;
4558 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
4559 irq_flags, preempt_count());
4561 /* Ring buffer disabled, return as if not open for write */
4566 entry = ring_buffer_event_data(event);
4567 entry->ip = _THIS_IP_;
4569 if (nr_pages == 2) {
4570 len = PAGE_SIZE - offset;
4571 memcpy(&entry->buf, map_page[0] + offset, len);
4572 memcpy(&entry->buf[len], map_page[1], cnt - len);
4574 memcpy(&entry->buf, map_page[0] + offset, cnt);
4576 if (entry->buf[cnt - 1] != '\n') {
4577 entry->buf[cnt] = '\n';
4578 entry->buf[cnt + 1] = '\0';
4580 entry->buf[cnt] = '\0';
4582 __buffer_unlock_commit(buffer, event);
4589 for (i = 0; i < nr_pages; i++){
4590 kunmap_atomic(map_page[i]);
4597 static int tracing_clock_show(struct seq_file *m, void *v)
4599 struct trace_array *tr = m->private;
4602 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
4604 "%s%s%s%s", i ? " " : "",
4605 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
4606 i == tr->clock_id ? "]" : "");
4612 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
4613 size_t cnt, loff_t *fpos)
4615 struct seq_file *m = filp->private_data;
4616 struct trace_array *tr = m->private;
4618 const char *clockstr;
4621 if (cnt >= sizeof(buf))
4624 if (copy_from_user(&buf, ubuf, cnt))
4629 clockstr = strstrip(buf);
4631 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
4632 if (strcmp(trace_clocks[i].name, clockstr) == 0)
4635 if (i == ARRAY_SIZE(trace_clocks))
4638 mutex_lock(&trace_types_lock);
4642 ring_buffer_set_clock(tr->trace_buffer.buffer, trace_clocks[i].func);
4645 * New clock may not be consistent with the previous clock.
4646 * Reset the buffer so that it doesn't have incomparable timestamps.
4648 tracing_reset_online_cpus(&tr->trace_buffer);
4650 #ifdef CONFIG_TRACER_MAX_TRACE
4651 if (tr->flags & TRACE_ARRAY_FL_GLOBAL && tr->max_buffer.buffer)
4652 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
4653 tracing_reset_online_cpus(&tr->max_buffer);
4656 mutex_unlock(&trace_types_lock);
4663 static int tracing_clock_open(struct inode *inode, struct file *file)
4665 struct trace_array *tr = inode->i_private;
4668 if (tracing_disabled)
4671 if (trace_array_get(tr))
4674 ret = single_open(file, tracing_clock_show, inode->i_private);
4676 trace_array_put(tr);
4681 struct ftrace_buffer_info {
4682 struct trace_iterator iter;
4687 #ifdef CONFIG_TRACER_SNAPSHOT
4688 static int tracing_snapshot_open(struct inode *inode, struct file *file)
4690 struct trace_array *tr = inode->i_private;
4691 struct trace_iterator *iter;
4695 if (trace_array_get(tr) < 0)
4698 if (file->f_mode & FMODE_READ) {
4699 iter = __tracing_open(inode, file, true);
4701 ret = PTR_ERR(iter);
4703 /* Writes still need the seq_file to hold the private data */
4705 m = kzalloc(sizeof(*m), GFP_KERNEL);
4708 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4716 iter->trace_buffer = &tr->max_buffer;
4717 iter->cpu_file = tracing_get_cpu(inode);
4719 file->private_data = m;
4723 trace_array_put(tr);
4729 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
4732 struct seq_file *m = filp->private_data;
4733 struct trace_iterator *iter = m->private;
4734 struct trace_array *tr = iter->tr;
4738 ret = tracing_update_buffers();
4742 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4746 mutex_lock(&trace_types_lock);
4748 if (tr->current_trace->use_max_tr) {
4755 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4759 if (tr->allocated_snapshot)
4763 /* Only allow per-cpu swap if the ring buffer supports it */
4764 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
4765 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4770 if (!tr->allocated_snapshot) {
4771 ret = alloc_snapshot(tr);
4775 local_irq_disable();
4776 /* Now, we're going to swap */
4777 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4778 update_max_tr(tr, current, smp_processor_id());
4780 update_max_tr_single(tr, current, iter->cpu_file);
4784 if (tr->allocated_snapshot) {
4785 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4786 tracing_reset_online_cpus(&tr->max_buffer);
4788 tracing_reset(&tr->max_buffer, iter->cpu_file);
4798 mutex_unlock(&trace_types_lock);
4802 static int tracing_snapshot_release(struct inode *inode, struct file *file)
4804 struct seq_file *m = file->private_data;
4807 ret = tracing_release(inode, file);
4809 if (file->f_mode & FMODE_READ)
4812 /* If write only, the seq_file is just a stub */
4820 static int tracing_buffers_open(struct inode *inode, struct file *filp);
4821 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
4822 size_t count, loff_t *ppos);
4823 static int tracing_buffers_release(struct inode *inode, struct file *file);
4824 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4825 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
4827 static int snapshot_raw_open(struct inode *inode, struct file *filp)
4829 struct ftrace_buffer_info *info;
4832 ret = tracing_buffers_open(inode, filp);
4836 info = filp->private_data;
4838 if (info->iter.trace->use_max_tr) {
4839 tracing_buffers_release(inode, filp);
4843 info->iter.snapshot = true;
4844 info->iter.trace_buffer = &info->iter.tr->max_buffer;
4849 #endif /* CONFIG_TRACER_SNAPSHOT */
4852 static const struct file_operations tracing_max_lat_fops = {
4853 .open = tracing_open_generic,
4854 .read = tracing_max_lat_read,
4855 .write = tracing_max_lat_write,
4856 .llseek = generic_file_llseek,
4859 static const struct file_operations set_tracer_fops = {
4860 .open = tracing_open_generic,
4861 .read = tracing_set_trace_read,
4862 .write = tracing_set_trace_write,
4863 .llseek = generic_file_llseek,
4866 static const struct file_operations tracing_pipe_fops = {
4867 .open = tracing_open_pipe,
4868 .poll = tracing_poll_pipe,
4869 .read = tracing_read_pipe,
4870 .splice_read = tracing_splice_read_pipe,
4871 .release = tracing_release_pipe,
4872 .llseek = no_llseek,
4875 static const struct file_operations tracing_entries_fops = {
4876 .open = tracing_open_generic_tr,
4877 .read = tracing_entries_read,
4878 .write = tracing_entries_write,
4879 .llseek = generic_file_llseek,
4880 .release = tracing_release_generic_tr,
4883 static const struct file_operations tracing_total_entries_fops = {
4884 .open = tracing_open_generic_tr,
4885 .read = tracing_total_entries_read,
4886 .llseek = generic_file_llseek,
4887 .release = tracing_release_generic_tr,
4890 static const struct file_operations tracing_free_buffer_fops = {
4891 .open = tracing_open_generic_tr,
4892 .write = tracing_free_buffer_write,
4893 .release = tracing_free_buffer_release,
4896 static const struct file_operations tracing_mark_fops = {
4897 .open = tracing_open_generic_tr,
4898 .write = tracing_mark_write,
4899 .llseek = generic_file_llseek,
4900 .release = tracing_release_generic_tr,
4903 static const struct file_operations trace_clock_fops = {
4904 .open = tracing_clock_open,
4906 .llseek = seq_lseek,
4907 .release = tracing_single_release_tr,
4908 .write = tracing_clock_write,
4911 #ifdef CONFIG_TRACER_SNAPSHOT
4912 static const struct file_operations snapshot_fops = {
4913 .open = tracing_snapshot_open,
4915 .write = tracing_snapshot_write,
4916 .llseek = tracing_seek,
4917 .release = tracing_snapshot_release,
4920 static const struct file_operations snapshot_raw_fops = {
4921 .open = snapshot_raw_open,
4922 .read = tracing_buffers_read,
4923 .release = tracing_buffers_release,
4924 .splice_read = tracing_buffers_splice_read,
4925 .llseek = no_llseek,
4928 #endif /* CONFIG_TRACER_SNAPSHOT */
4930 static int tracing_buffers_open(struct inode *inode, struct file *filp)
4932 struct trace_array *tr = inode->i_private;
4933 struct ftrace_buffer_info *info;
4936 if (tracing_disabled)
4939 if (trace_array_get(tr) < 0)
4942 info = kzalloc(sizeof(*info), GFP_KERNEL);
4944 trace_array_put(tr);
4948 mutex_lock(&trace_types_lock);
4951 info->iter.cpu_file = tracing_get_cpu(inode);
4952 info->iter.trace = tr->current_trace;
4953 info->iter.trace_buffer = &tr->trace_buffer;
4955 /* Force reading ring buffer for first read */
4956 info->read = (unsigned int)-1;
4958 filp->private_data = info;
4960 mutex_unlock(&trace_types_lock);
4962 ret = nonseekable_open(inode, filp);
4964 trace_array_put(tr);
4970 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
4972 struct ftrace_buffer_info *info = filp->private_data;
4973 struct trace_iterator *iter = &info->iter;
4975 return trace_poll(iter, filp, poll_table);
4979 tracing_buffers_read(struct file *filp, char __user *ubuf,
4980 size_t count, loff_t *ppos)
4982 struct ftrace_buffer_info *info = filp->private_data;
4983 struct trace_iterator *iter = &info->iter;
4990 mutex_lock(&trace_types_lock);
4992 #ifdef CONFIG_TRACER_MAX_TRACE
4993 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
5000 info->spare = ring_buffer_alloc_read_page(iter->trace_buffer->buffer,
5006 /* Do we have previous read data to read? */
5007 if (info->read < PAGE_SIZE)
5011 trace_access_lock(iter->cpu_file);
5012 ret = ring_buffer_read_page(iter->trace_buffer->buffer,
5016 trace_access_unlock(iter->cpu_file);
5019 if (trace_empty(iter)) {
5020 if ((filp->f_flags & O_NONBLOCK)) {
5024 mutex_unlock(&trace_types_lock);
5025 iter->trace->wait_pipe(iter);
5026 mutex_lock(&trace_types_lock);
5027 if (signal_pending(current)) {
5039 size = PAGE_SIZE - info->read;
5043 ret = copy_to_user(ubuf, info->spare + info->read, size);
5054 mutex_unlock(&trace_types_lock);
5059 static int tracing_buffers_release(struct inode *inode, struct file *file)
5061 struct ftrace_buffer_info *info = file->private_data;
5062 struct trace_iterator *iter = &info->iter;
5064 mutex_lock(&trace_types_lock);
5066 __trace_array_put(iter->tr);
5069 ring_buffer_free_read_page(iter->trace_buffer->buffer, info->spare);
5072 mutex_unlock(&trace_types_lock);
5078 struct ring_buffer *buffer;
5083 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
5084 struct pipe_buffer *buf)
5086 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5091 ring_buffer_free_read_page(ref->buffer, ref->page);
5096 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
5097 struct pipe_buffer *buf)
5099 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5104 /* Pipe buffer operations for a buffer. */
5105 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
5107 .map = generic_pipe_buf_map,
5108 .unmap = generic_pipe_buf_unmap,
5109 .confirm = generic_pipe_buf_confirm,
5110 .release = buffer_pipe_buf_release,
5111 .steal = generic_pipe_buf_steal,
5112 .get = buffer_pipe_buf_get,
5116 * Callback from splice_to_pipe(), if we need to release some pages
5117 * at the end of the spd in case we error'ed out in filling the pipe.
5119 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
5121 struct buffer_ref *ref =
5122 (struct buffer_ref *)spd->partial[i].private;
5127 ring_buffer_free_read_page(ref->buffer, ref->page);
5129 spd->partial[i].private = 0;
5133 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
5134 struct pipe_inode_info *pipe, size_t len,
5137 struct ftrace_buffer_info *info = file->private_data;
5138 struct trace_iterator *iter = &info->iter;
5139 struct partial_page partial_def[PIPE_DEF_BUFFERS];
5140 struct page *pages_def[PIPE_DEF_BUFFERS];
5141 struct splice_pipe_desc spd = {
5143 .partial = partial_def,
5144 .nr_pages_max = PIPE_DEF_BUFFERS,
5146 .ops = &buffer_pipe_buf_ops,
5147 .spd_release = buffer_spd_release,
5149 struct buffer_ref *ref;
5150 int entries, size, i;
5153 mutex_lock(&trace_types_lock);
5155 #ifdef CONFIG_TRACER_MAX_TRACE
5156 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
5162 if (splice_grow_spd(pipe, &spd)) {
5167 if (*ppos & (PAGE_SIZE - 1)) {
5172 if (len & (PAGE_SIZE - 1)) {
5173 if (len < PAGE_SIZE) {
5181 trace_access_lock(iter->cpu_file);
5182 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5184 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
5188 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
5193 ref->buffer = iter->trace_buffer->buffer;
5194 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
5200 r = ring_buffer_read_page(ref->buffer, &ref->page,
5201 len, iter->cpu_file, 1);
5203 ring_buffer_free_read_page(ref->buffer, ref->page);
5209 * zero out any left over data, this is going to
5212 size = ring_buffer_page_len(ref->page);
5213 if (size < PAGE_SIZE)
5214 memset(ref->page + size, 0, PAGE_SIZE - size);
5216 page = virt_to_page(ref->page);
5218 spd.pages[i] = page;
5219 spd.partial[i].len = PAGE_SIZE;
5220 spd.partial[i].offset = 0;
5221 spd.partial[i].private = (unsigned long)ref;
5225 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5228 trace_access_unlock(iter->cpu_file);
5231 /* did we read anything? */
5232 if (!spd.nr_pages) {
5233 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK)) {
5237 mutex_unlock(&trace_types_lock);
5238 iter->trace->wait_pipe(iter);
5239 mutex_lock(&trace_types_lock);
5240 if (signal_pending(current)) {
5247 ret = splice_to_pipe(pipe, &spd);
5248 splice_shrink_spd(&spd);
5250 mutex_unlock(&trace_types_lock);
5255 static const struct file_operations tracing_buffers_fops = {
5256 .open = tracing_buffers_open,
5257 .read = tracing_buffers_read,
5258 .poll = tracing_buffers_poll,
5259 .release = tracing_buffers_release,
5260 .splice_read = tracing_buffers_splice_read,
5261 .llseek = no_llseek,
5265 tracing_stats_read(struct file *filp, char __user *ubuf,
5266 size_t count, loff_t *ppos)
5268 struct inode *inode = file_inode(filp);
5269 struct trace_array *tr = inode->i_private;
5270 struct trace_buffer *trace_buf = &tr->trace_buffer;
5271 int cpu = tracing_get_cpu(inode);
5272 struct trace_seq *s;
5274 unsigned long long t;
5275 unsigned long usec_rem;
5277 s = kmalloc(sizeof(*s), GFP_KERNEL);
5283 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
5284 trace_seq_printf(s, "entries: %ld\n", cnt);
5286 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
5287 trace_seq_printf(s, "overrun: %ld\n", cnt);
5289 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
5290 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
5292 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
5293 trace_seq_printf(s, "bytes: %ld\n", cnt);
5295 if (trace_clocks[tr->clock_id].in_ns) {
5296 /* local or global for trace_clock */
5297 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5298 usec_rem = do_div(t, USEC_PER_SEC);
5299 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
5302 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
5303 usec_rem = do_div(t, USEC_PER_SEC);
5304 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
5306 /* counter or tsc mode for trace_clock */
5307 trace_seq_printf(s, "oldest event ts: %llu\n",
5308 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5310 trace_seq_printf(s, "now ts: %llu\n",
5311 ring_buffer_time_stamp(trace_buf->buffer, cpu));
5314 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
5315 trace_seq_printf(s, "dropped events: %ld\n", cnt);
5317 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
5318 trace_seq_printf(s, "read events: %ld\n", cnt);
5320 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
5327 static const struct file_operations tracing_stats_fops = {
5328 .open = tracing_open_generic_tr,
5329 .read = tracing_stats_read,
5330 .llseek = generic_file_llseek,
5331 .release = tracing_release_generic_tr,
5334 #ifdef CONFIG_DYNAMIC_FTRACE
5336 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
5342 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
5343 size_t cnt, loff_t *ppos)
5345 static char ftrace_dyn_info_buffer[1024];
5346 static DEFINE_MUTEX(dyn_info_mutex);
5347 unsigned long *p = filp->private_data;
5348 char *buf = ftrace_dyn_info_buffer;
5349 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
5352 mutex_lock(&dyn_info_mutex);
5353 r = sprintf(buf, "%ld ", *p);
5355 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
5358 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5360 mutex_unlock(&dyn_info_mutex);
5365 static const struct file_operations tracing_dyn_info_fops = {
5366 .open = tracing_open_generic,
5367 .read = tracing_read_dyn_info,
5368 .llseek = generic_file_llseek,
5370 #endif /* CONFIG_DYNAMIC_FTRACE */
5372 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
5374 ftrace_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5380 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5382 unsigned long *count = (long *)data;
5394 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
5395 struct ftrace_probe_ops *ops, void *data)
5397 long count = (long)data;
5399 seq_printf(m, "%ps:", (void *)ip);
5401 seq_printf(m, "snapshot");
5404 seq_printf(m, ":unlimited\n");
5406 seq_printf(m, ":count=%ld\n", count);
5411 static struct ftrace_probe_ops snapshot_probe_ops = {
5412 .func = ftrace_snapshot,
5413 .print = ftrace_snapshot_print,
5416 static struct ftrace_probe_ops snapshot_count_probe_ops = {
5417 .func = ftrace_count_snapshot,
5418 .print = ftrace_snapshot_print,
5422 ftrace_trace_snapshot_callback(struct ftrace_hash *hash,
5423 char *glob, char *cmd, char *param, int enable)
5425 struct ftrace_probe_ops *ops;
5426 void *count = (void *)-1;
5430 /* hash funcs only work with set_ftrace_filter */
5434 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
5436 if (glob[0] == '!') {
5437 unregister_ftrace_function_probe_func(glob+1, ops);
5444 number = strsep(¶m, ":");
5446 if (!strlen(number))
5450 * We use the callback data field (which is a pointer)
5453 ret = kstrtoul(number, 0, (unsigned long *)&count);
5458 ret = register_ftrace_function_probe(glob, ops, count);
5461 alloc_snapshot(&global_trace);
5463 return ret < 0 ? ret : 0;
5466 static struct ftrace_func_command ftrace_snapshot_cmd = {
5468 .func = ftrace_trace_snapshot_callback,
5471 static __init int register_snapshot_cmd(void)
5473 return register_ftrace_command(&ftrace_snapshot_cmd);
5476 static inline __init int register_snapshot_cmd(void) { return 0; }
5477 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
5479 struct dentry *tracing_init_dentry_tr(struct trace_array *tr)
5484 if (!debugfs_initialized())
5487 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
5488 tr->dir = debugfs_create_dir("tracing", NULL);
5491 pr_warn_once("Could not create debugfs directory 'tracing'\n");
5496 struct dentry *tracing_init_dentry(void)
5498 return tracing_init_dentry_tr(&global_trace);
5501 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
5503 struct dentry *d_tracer;
5506 return tr->percpu_dir;
5508 d_tracer = tracing_init_dentry_tr(tr);
5512 tr->percpu_dir = debugfs_create_dir("per_cpu", d_tracer);
5514 WARN_ONCE(!tr->percpu_dir,
5515 "Could not create debugfs directory 'per_cpu/%d'\n", cpu);
5517 return tr->percpu_dir;
5520 static struct dentry *
5521 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
5522 void *data, long cpu, const struct file_operations *fops)
5524 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
5526 if (ret) /* See tracing_get_cpu() */
5527 ret->d_inode->i_cdev = (void *)(cpu + 1);
5532 tracing_init_debugfs_percpu(struct trace_array *tr, long cpu)
5534 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
5535 struct dentry *d_cpu;
5536 char cpu_dir[30]; /* 30 characters should be more than enough */
5541 snprintf(cpu_dir, 30, "cpu%ld", cpu);
5542 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
5544 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
5548 /* per cpu trace_pipe */
5549 trace_create_cpu_file("trace_pipe", 0444, d_cpu,
5550 tr, cpu, &tracing_pipe_fops);
5553 trace_create_cpu_file("trace", 0644, d_cpu,
5554 tr, cpu, &tracing_fops);
5556 trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu,
5557 tr, cpu, &tracing_buffers_fops);
5559 trace_create_cpu_file("stats", 0444, d_cpu,
5560 tr, cpu, &tracing_stats_fops);
5562 trace_create_cpu_file("buffer_size_kb", 0444, d_cpu,
5563 tr, cpu, &tracing_entries_fops);
5565 #ifdef CONFIG_TRACER_SNAPSHOT
5566 trace_create_cpu_file("snapshot", 0644, d_cpu,
5567 tr, cpu, &snapshot_fops);
5569 trace_create_cpu_file("snapshot_raw", 0444, d_cpu,
5570 tr, cpu, &snapshot_raw_fops);
5574 #ifdef CONFIG_FTRACE_SELFTEST
5575 /* Let selftest have access to static functions in this file */
5576 #include "trace_selftest.c"
5579 struct trace_option_dentry {
5580 struct tracer_opt *opt;
5581 struct tracer_flags *flags;
5582 struct trace_array *tr;
5583 struct dentry *entry;
5587 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
5590 struct trace_option_dentry *topt = filp->private_data;
5593 if (topt->flags->val & topt->opt->bit)
5598 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5602 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
5605 struct trace_option_dentry *topt = filp->private_data;
5609 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5613 if (val != 0 && val != 1)
5616 if (!!(topt->flags->val & topt->opt->bit) != val) {
5617 mutex_lock(&trace_types_lock);
5618 ret = __set_tracer_option(topt->tr->current_trace, topt->flags,
5620 mutex_unlock(&trace_types_lock);
5631 static const struct file_operations trace_options_fops = {
5632 .open = tracing_open_generic,
5633 .read = trace_options_read,
5634 .write = trace_options_write,
5635 .llseek = generic_file_llseek,
5639 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
5642 long index = (long)filp->private_data;
5645 if (trace_flags & (1 << index))
5650 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5654 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
5657 struct trace_array *tr = &global_trace;
5658 long index = (long)filp->private_data;
5662 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5666 if (val != 0 && val != 1)
5669 mutex_lock(&trace_types_lock);
5670 ret = set_tracer_flag(tr, 1 << index, val);
5671 mutex_unlock(&trace_types_lock);
5681 static const struct file_operations trace_options_core_fops = {
5682 .open = tracing_open_generic,
5683 .read = trace_options_core_read,
5684 .write = trace_options_core_write,
5685 .llseek = generic_file_llseek,
5688 struct dentry *trace_create_file(const char *name,
5690 struct dentry *parent,
5692 const struct file_operations *fops)
5696 ret = debugfs_create_file(name, mode, parent, data, fops);
5698 pr_warning("Could not create debugfs '%s' entry\n", name);
5704 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
5706 struct dentry *d_tracer;
5711 d_tracer = tracing_init_dentry_tr(tr);
5715 tr->options = debugfs_create_dir("options", d_tracer);
5717 pr_warning("Could not create debugfs directory 'options'\n");
5725 create_trace_option_file(struct trace_array *tr,
5726 struct trace_option_dentry *topt,
5727 struct tracer_flags *flags,
5728 struct tracer_opt *opt)
5730 struct dentry *t_options;
5732 t_options = trace_options_init_dentry(tr);
5736 topt->flags = flags;
5740 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
5741 &trace_options_fops);
5745 static struct trace_option_dentry *
5746 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
5748 struct trace_option_dentry *topts;
5749 struct tracer_flags *flags;
5750 struct tracer_opt *opts;
5756 flags = tracer->flags;
5758 if (!flags || !flags->opts)
5763 for (cnt = 0; opts[cnt].name; cnt++)
5766 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
5770 for (cnt = 0; opts[cnt].name; cnt++)
5771 create_trace_option_file(tr, &topts[cnt], flags,
5778 destroy_trace_option_files(struct trace_option_dentry *topts)
5785 for (cnt = 0; topts[cnt].opt; cnt++) {
5786 if (topts[cnt].entry)
5787 debugfs_remove(topts[cnt].entry);
5793 static struct dentry *
5794 create_trace_option_core_file(struct trace_array *tr,
5795 const char *option, long index)
5797 struct dentry *t_options;
5799 t_options = trace_options_init_dentry(tr);
5803 return trace_create_file(option, 0644, t_options, (void *)index,
5804 &trace_options_core_fops);
5807 static __init void create_trace_options_dir(struct trace_array *tr)
5809 struct dentry *t_options;
5812 t_options = trace_options_init_dentry(tr);
5816 for (i = 0; trace_options[i]; i++)
5817 create_trace_option_core_file(tr, trace_options[i], i);
5821 rb_simple_read(struct file *filp, char __user *ubuf,
5822 size_t cnt, loff_t *ppos)
5824 struct trace_array *tr = filp->private_data;
5828 r = tracer_tracing_is_on(tr);
5829 r = sprintf(buf, "%d\n", r);
5831 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5835 rb_simple_write(struct file *filp, const char __user *ubuf,
5836 size_t cnt, loff_t *ppos)
5838 struct trace_array *tr = filp->private_data;
5839 struct ring_buffer *buffer = tr->trace_buffer.buffer;
5843 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5848 mutex_lock(&trace_types_lock);
5850 tracer_tracing_on(tr);
5851 if (tr->current_trace->start)
5852 tr->current_trace->start(tr);
5854 tracer_tracing_off(tr);
5855 if (tr->current_trace->stop)
5856 tr->current_trace->stop(tr);
5858 mutex_unlock(&trace_types_lock);
5866 static const struct file_operations rb_simple_fops = {
5867 .open = tracing_open_generic_tr,
5868 .read = rb_simple_read,
5869 .write = rb_simple_write,
5870 .release = tracing_release_generic_tr,
5871 .llseek = default_llseek,
5874 struct dentry *trace_instance_dir;
5877 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer);
5880 allocate_trace_buffer(struct trace_array *tr, struct trace_buffer *buf, int size)
5882 enum ring_buffer_flags rb_flags;
5884 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
5886 buf->buffer = ring_buffer_alloc(size, rb_flags);
5890 buf->data = alloc_percpu(struct trace_array_cpu);
5892 ring_buffer_free(buf->buffer);
5896 /* Allocate the first page for all buffers */
5897 set_buffer_entries(&tr->trace_buffer,
5898 ring_buffer_size(tr->trace_buffer.buffer, 0));
5903 static int allocate_trace_buffers(struct trace_array *tr, int size)
5907 ret = allocate_trace_buffer(tr, &tr->trace_buffer, size);
5911 #ifdef CONFIG_TRACER_MAX_TRACE
5912 ret = allocate_trace_buffer(tr, &tr->max_buffer,
5913 allocate_snapshot ? size : 1);
5915 ring_buffer_free(tr->trace_buffer.buffer);
5916 free_percpu(tr->trace_buffer.data);
5919 tr->allocated_snapshot = allocate_snapshot;
5922 * Only the top level trace array gets its snapshot allocated
5923 * from the kernel command line.
5925 allocate_snapshot = false;
5930 static int new_instance_create(const char *name)
5932 struct trace_array *tr;
5935 mutex_lock(&trace_types_lock);
5938 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5939 if (tr->name && strcmp(tr->name, name) == 0)
5944 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
5948 tr->name = kstrdup(name, GFP_KERNEL);
5952 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
5955 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
5957 raw_spin_lock_init(&tr->start_lock);
5959 tr->current_trace = &nop_trace;
5961 INIT_LIST_HEAD(&tr->systems);
5962 INIT_LIST_HEAD(&tr->events);
5964 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
5967 tr->dir = debugfs_create_dir(name, trace_instance_dir);
5971 ret = event_trace_add_tracer(tr->dir, tr);
5973 debugfs_remove_recursive(tr->dir);
5977 init_tracer_debugfs(tr, tr->dir);
5979 list_add(&tr->list, &ftrace_trace_arrays);
5981 mutex_unlock(&trace_types_lock);
5986 if (tr->trace_buffer.buffer)
5987 ring_buffer_free(tr->trace_buffer.buffer);
5988 free_cpumask_var(tr->tracing_cpumask);
5993 mutex_unlock(&trace_types_lock);
5999 static int instance_delete(const char *name)
6001 struct trace_array *tr;
6005 mutex_lock(&trace_types_lock);
6008 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
6009 if (tr->name && strcmp(tr->name, name) == 0) {
6021 list_del(&tr->list);
6023 event_trace_del_tracer(tr);
6024 debugfs_remove_recursive(tr->dir);
6025 free_percpu(tr->trace_buffer.data);
6026 ring_buffer_free(tr->trace_buffer.buffer);
6034 mutex_unlock(&trace_types_lock);
6039 static int instance_mkdir (struct inode *inode, struct dentry *dentry, umode_t mode)
6041 struct dentry *parent;
6044 /* Paranoid: Make sure the parent is the "instances" directory */
6045 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
6046 if (WARN_ON_ONCE(parent != trace_instance_dir))
6050 * The inode mutex is locked, but debugfs_create_dir() will also
6051 * take the mutex. As the instances directory can not be destroyed
6052 * or changed in any other way, it is safe to unlock it, and
6053 * let the dentry try. If two users try to make the same dir at
6054 * the same time, then the new_instance_create() will determine the
6057 mutex_unlock(&inode->i_mutex);
6059 ret = new_instance_create(dentry->d_iname);
6061 mutex_lock(&inode->i_mutex);
6066 static int instance_rmdir(struct inode *inode, struct dentry *dentry)
6068 struct dentry *parent;
6071 /* Paranoid: Make sure the parent is the "instances" directory */
6072 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
6073 if (WARN_ON_ONCE(parent != trace_instance_dir))
6076 /* The caller did a dget() on dentry */
6077 mutex_unlock(&dentry->d_inode->i_mutex);
6080 * The inode mutex is locked, but debugfs_create_dir() will also
6081 * take the mutex. As the instances directory can not be destroyed
6082 * or changed in any other way, it is safe to unlock it, and
6083 * let the dentry try. If two users try to make the same dir at
6084 * the same time, then the instance_delete() will determine the
6087 mutex_unlock(&inode->i_mutex);
6089 ret = instance_delete(dentry->d_iname);
6091 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
6092 mutex_lock(&dentry->d_inode->i_mutex);
6097 static const struct inode_operations instance_dir_inode_operations = {
6098 .lookup = simple_lookup,
6099 .mkdir = instance_mkdir,
6100 .rmdir = instance_rmdir,
6103 static __init void create_trace_instances(struct dentry *d_tracer)
6105 trace_instance_dir = debugfs_create_dir("instances", d_tracer);
6106 if (WARN_ON(!trace_instance_dir))
6109 /* Hijack the dir inode operations, to allow mkdir */
6110 trace_instance_dir->d_inode->i_op = &instance_dir_inode_operations;
6114 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer)
6118 trace_create_file("tracing_cpumask", 0644, d_tracer,
6119 tr, &tracing_cpumask_fops);
6121 trace_create_file("trace_options", 0644, d_tracer,
6122 tr, &tracing_iter_fops);
6124 trace_create_file("trace", 0644, d_tracer,
6127 trace_create_file("trace_pipe", 0444, d_tracer,
6128 tr, &tracing_pipe_fops);
6130 trace_create_file("buffer_size_kb", 0644, d_tracer,
6131 tr, &tracing_entries_fops);
6133 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
6134 tr, &tracing_total_entries_fops);
6136 trace_create_file("free_buffer", 0200, d_tracer,
6137 tr, &tracing_free_buffer_fops);
6139 trace_create_file("trace_marker", 0220, d_tracer,
6140 tr, &tracing_mark_fops);
6142 trace_create_file("trace_clock", 0644, d_tracer, tr,
6145 trace_create_file("tracing_on", 0644, d_tracer,
6146 tr, &rb_simple_fops);
6148 #ifdef CONFIG_TRACER_SNAPSHOT
6149 trace_create_file("snapshot", 0644, d_tracer,
6150 tr, &snapshot_fops);
6153 for_each_tracing_cpu(cpu)
6154 tracing_init_debugfs_percpu(tr, cpu);
6158 static __init int tracer_init_debugfs(void)
6160 struct dentry *d_tracer;
6162 trace_access_lock_init();
6164 d_tracer = tracing_init_dentry();
6168 init_tracer_debugfs(&global_trace, d_tracer);
6170 trace_create_file("available_tracers", 0444, d_tracer,
6171 &global_trace, &show_traces_fops);
6173 trace_create_file("current_tracer", 0644, d_tracer,
6174 &global_trace, &set_tracer_fops);
6176 #ifdef CONFIG_TRACER_MAX_TRACE
6177 trace_create_file("tracing_max_latency", 0644, d_tracer,
6178 &tracing_max_latency, &tracing_max_lat_fops);
6181 trace_create_file("tracing_thresh", 0644, d_tracer,
6182 &tracing_thresh, &tracing_max_lat_fops);
6184 trace_create_file("README", 0444, d_tracer,
6185 NULL, &tracing_readme_fops);
6187 trace_create_file("saved_cmdlines", 0444, d_tracer,
6188 NULL, &tracing_saved_cmdlines_fops);
6190 #ifdef CONFIG_DYNAMIC_FTRACE
6191 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
6192 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
6195 create_trace_instances(d_tracer);
6197 create_trace_options_dir(&global_trace);
6202 static int trace_panic_handler(struct notifier_block *this,
6203 unsigned long event, void *unused)
6205 if (ftrace_dump_on_oops)
6206 ftrace_dump(ftrace_dump_on_oops);
6210 static struct notifier_block trace_panic_notifier = {
6211 .notifier_call = trace_panic_handler,
6213 .priority = 150 /* priority: INT_MAX >= x >= 0 */
6216 static int trace_die_handler(struct notifier_block *self,
6222 if (ftrace_dump_on_oops)
6223 ftrace_dump(ftrace_dump_on_oops);
6231 static struct notifier_block trace_die_notifier = {
6232 .notifier_call = trace_die_handler,
6237 * printk is set to max of 1024, we really don't need it that big.
6238 * Nothing should be printing 1000 characters anyway.
6240 #define TRACE_MAX_PRINT 1000
6243 * Define here KERN_TRACE so that we have one place to modify
6244 * it if we decide to change what log level the ftrace dump
6247 #define KERN_TRACE KERN_EMERG
6250 trace_printk_seq(struct trace_seq *s)
6252 /* Probably should print a warning here. */
6253 if (s->len >= TRACE_MAX_PRINT)
6254 s->len = TRACE_MAX_PRINT;
6256 /* should be zero ended, but we are paranoid. */
6257 s->buffer[s->len] = 0;
6259 printk(KERN_TRACE "%s", s->buffer);
6264 void trace_init_global_iter(struct trace_iterator *iter)
6266 iter->tr = &global_trace;
6267 iter->trace = iter->tr->current_trace;
6268 iter->cpu_file = RING_BUFFER_ALL_CPUS;
6269 iter->trace_buffer = &global_trace.trace_buffer;
6271 if (iter->trace && iter->trace->open)
6272 iter->trace->open(iter);
6274 /* Annotate start of buffers if we had overruns */
6275 if (ring_buffer_overruns(iter->trace_buffer->buffer))
6276 iter->iter_flags |= TRACE_FILE_ANNOTATE;
6278 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6279 if (trace_clocks[iter->tr->clock_id].in_ns)
6280 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6283 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
6285 /* use static because iter can be a bit big for the stack */
6286 static struct trace_iterator iter;
6287 static atomic_t dump_running;
6288 unsigned int old_userobj;
6289 unsigned long flags;
6292 /* Only allow one dump user at a time. */
6293 if (atomic_inc_return(&dump_running) != 1) {
6294 atomic_dec(&dump_running);
6299 * Always turn off tracing when we dump.
6300 * We don't need to show trace output of what happens
6301 * between multiple crashes.
6303 * If the user does a sysrq-z, then they can re-enable
6304 * tracing with echo 1 > tracing_on.
6308 local_irq_save(flags);
6310 /* Simulate the iterator */
6311 trace_init_global_iter(&iter);
6313 for_each_tracing_cpu(cpu) {
6314 atomic_inc(&per_cpu_ptr(iter.tr->trace_buffer.data, cpu)->disabled);
6317 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
6319 /* don't look at user memory in panic mode */
6320 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
6322 switch (oops_dump_mode) {
6324 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6327 iter.cpu_file = raw_smp_processor_id();
6332 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
6333 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6336 printk(KERN_TRACE "Dumping ftrace buffer:\n");
6338 /* Did function tracer already get disabled? */
6339 if (ftrace_is_dead()) {
6340 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
6341 printk("# MAY BE MISSING FUNCTION EVENTS\n");
6345 * We need to stop all tracing on all CPUS to read the
6346 * the next buffer. This is a bit expensive, but is
6347 * not done often. We fill all what we can read,
6348 * and then release the locks again.
6351 while (!trace_empty(&iter)) {
6354 printk(KERN_TRACE "---------------------------------\n");
6358 /* reset all but tr, trace, and overruns */
6359 memset(&iter.seq, 0,
6360 sizeof(struct trace_iterator) -
6361 offsetof(struct trace_iterator, seq));
6362 iter.iter_flags |= TRACE_FILE_LAT_FMT;
6365 if (trace_find_next_entry_inc(&iter) != NULL) {
6368 ret = print_trace_line(&iter);
6369 if (ret != TRACE_TYPE_NO_CONSUME)
6370 trace_consume(&iter);
6372 touch_nmi_watchdog();
6374 trace_printk_seq(&iter.seq);
6378 printk(KERN_TRACE " (ftrace buffer empty)\n");
6380 printk(KERN_TRACE "---------------------------------\n");
6383 trace_flags |= old_userobj;
6385 for_each_tracing_cpu(cpu) {
6386 atomic_dec(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);
6388 atomic_dec(&dump_running);
6389 local_irq_restore(flags);
6391 EXPORT_SYMBOL_GPL(ftrace_dump);
6393 __init static int tracer_alloc_buffers(void)
6399 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
6402 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
6403 goto out_free_buffer_mask;
6405 /* Only allocate trace_printk buffers if a trace_printk exists */
6406 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
6407 /* Must be called before global_trace.buffer is allocated */
6408 trace_printk_init_buffers();
6410 /* To save memory, keep the ring buffer size to its minimum */
6411 if (ring_buffer_expanded)
6412 ring_buf_size = trace_buf_size;
6416 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
6417 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
6419 raw_spin_lock_init(&global_trace.start_lock);
6421 /* TODO: make the number of buffers hot pluggable with CPUS */
6422 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
6423 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
6425 goto out_free_cpumask;
6428 if (global_trace.buffer_disabled)
6431 trace_init_cmdlines();
6434 * register_tracer() might reference current_trace, so it
6435 * needs to be set before we register anything. This is
6436 * just a bootstrap of current_trace anyway.
6438 global_trace.current_trace = &nop_trace;
6440 register_tracer(&nop_trace);
6442 /* All seems OK, enable tracing */
6443 tracing_disabled = 0;
6445 atomic_notifier_chain_register(&panic_notifier_list,
6446 &trace_panic_notifier);
6448 register_die_notifier(&trace_die_notifier);
6450 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
6452 INIT_LIST_HEAD(&global_trace.systems);
6453 INIT_LIST_HEAD(&global_trace.events);
6454 list_add(&global_trace.list, &ftrace_trace_arrays);
6456 while (trace_boot_options) {
6459 option = strsep(&trace_boot_options, ",");
6460 trace_set_options(&global_trace, option);
6463 register_snapshot_cmd();
6468 free_percpu(global_trace.trace_buffer.data);
6469 #ifdef CONFIG_TRACER_MAX_TRACE
6470 free_percpu(global_trace.max_buffer.data);
6472 free_cpumask_var(global_trace.tracing_cpumask);
6473 out_free_buffer_mask:
6474 free_cpumask_var(tracing_buffer_mask);
6479 __init static int clear_boot_tracer(void)
6482 * The default tracer at boot buffer is an init section.
6483 * This function is called in lateinit. If we did not
6484 * find the boot tracer, then clear it out, to prevent
6485 * later registration from accessing the buffer that is
6486 * about to be freed.
6488 if (!default_bootup_tracer)
6491 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
6492 default_bootup_tracer);
6493 default_bootup_tracer = NULL;
6498 early_initcall(tracer_alloc_buffers);
6499 fs_initcall(tracer_init_debugfs);
6500 late_initcall(clear_boot_tracer);