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 * tracing_alloc_snapshot - allocate snapshot buffer.
600 * This only allocates the snapshot buffer if it isn't already
601 * allocated - it doesn't also take a snapshot.
603 * This is meant to be used in cases where the snapshot buffer needs
604 * to be set up for events that can't sleep but need to be able to
605 * trigger a snapshot.
607 int tracing_alloc_snapshot(void)
609 struct trace_array *tr = &global_trace;
612 ret = alloc_snapshot(tr);
617 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
620 * trace_snapshot_alloc - allocate and take a snapshot of the current buffer.
622 * This is similar to trace_snapshot(), but it will allocate the
623 * snapshot buffer if it isn't already allocated. Use this only
624 * where it is safe to sleep, as the allocation may sleep.
626 * This causes a swap between the snapshot buffer and the current live
627 * tracing buffer. You can use this to take snapshots of the live
628 * trace when some condition is triggered, but continue to trace.
630 void tracing_snapshot_alloc(void)
634 ret = tracing_alloc_snapshot();
640 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
642 void tracing_snapshot(void)
644 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
646 EXPORT_SYMBOL_GPL(tracing_snapshot);
647 int tracing_alloc_snapshot(void)
649 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
652 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
653 void tracing_snapshot_alloc(void)
658 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
659 #endif /* CONFIG_TRACER_SNAPSHOT */
661 static void tracer_tracing_off(struct trace_array *tr)
663 if (tr->trace_buffer.buffer)
664 ring_buffer_record_off(tr->trace_buffer.buffer);
666 * This flag is looked at when buffers haven't been allocated
667 * yet, or by some tracers (like irqsoff), that just want to
668 * know if the ring buffer has been disabled, but it can handle
669 * races of where it gets disabled but we still do a record.
670 * As the check is in the fast path of the tracers, it is more
671 * important to be fast than accurate.
673 tr->buffer_disabled = 1;
674 /* Make the flag seen by readers */
679 * tracing_off - turn off tracing buffers
681 * This function stops the tracing buffers from recording data.
682 * It does not disable any overhead the tracers themselves may
683 * be causing. This function simply causes all recording to
684 * the ring buffers to fail.
686 void tracing_off(void)
688 tracer_tracing_off(&global_trace);
690 EXPORT_SYMBOL_GPL(tracing_off);
692 void disable_trace_on_warning(void)
694 if (__disable_trace_on_warning)
699 * tracer_tracing_is_on - show real state of ring buffer enabled
700 * @tr : the trace array to know if ring buffer is enabled
702 * Shows real state of the ring buffer if it is enabled or not.
704 static int tracer_tracing_is_on(struct trace_array *tr)
706 if (tr->trace_buffer.buffer)
707 return ring_buffer_record_is_on(tr->trace_buffer.buffer);
708 return !tr->buffer_disabled;
712 * tracing_is_on - show state of ring buffers enabled
714 int tracing_is_on(void)
716 return tracer_tracing_is_on(&global_trace);
718 EXPORT_SYMBOL_GPL(tracing_is_on);
720 static int __init set_buf_size(char *str)
722 unsigned long buf_size;
726 buf_size = memparse(str, &str);
727 /* nr_entries can not be zero */
730 trace_buf_size = buf_size;
733 __setup("trace_buf_size=", set_buf_size);
735 static int __init set_tracing_thresh(char *str)
737 unsigned long threshold;
742 ret = kstrtoul(str, 0, &threshold);
745 tracing_thresh = threshold * 1000;
748 __setup("tracing_thresh=", set_tracing_thresh);
750 unsigned long nsecs_to_usecs(unsigned long nsecs)
755 /* These must match the bit postions in trace_iterator_flags */
756 static const char *trace_options[] = {
789 int in_ns; /* is this clock in nanoseconds? */
791 { trace_clock_local, "local", 1 },
792 { trace_clock_global, "global", 1 },
793 { trace_clock_counter, "counter", 0 },
794 { trace_clock_jiffies, "uptime", 1 },
795 { trace_clock, "perf", 1 },
800 * trace_parser_get_init - gets the buffer for trace parser
802 int trace_parser_get_init(struct trace_parser *parser, int size)
804 memset(parser, 0, sizeof(*parser));
806 parser->buffer = kmalloc(size, GFP_KERNEL);
815 * trace_parser_put - frees the buffer for trace parser
817 void trace_parser_put(struct trace_parser *parser)
819 kfree(parser->buffer);
823 * trace_get_user - reads the user input string separated by space
824 * (matched by isspace(ch))
826 * For each string found the 'struct trace_parser' is updated,
827 * and the function returns.
829 * Returns number of bytes read.
831 * See kernel/trace/trace.h for 'struct trace_parser' details.
833 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
834 size_t cnt, loff_t *ppos)
841 trace_parser_clear(parser);
843 ret = get_user(ch, ubuf++);
851 * The parser is not finished with the last write,
852 * continue reading the user input without skipping spaces.
855 /* skip white space */
856 while (cnt && isspace(ch)) {
857 ret = get_user(ch, ubuf++);
864 /* only spaces were written */
874 /* read the non-space input */
875 while (cnt && !isspace(ch)) {
876 if (parser->idx < parser->size - 1)
877 parser->buffer[parser->idx++] = ch;
882 ret = get_user(ch, ubuf++);
889 /* We either got finished input or we have to wait for another call. */
891 parser->buffer[parser->idx] = 0;
892 parser->cont = false;
893 } else if (parser->idx < parser->size - 1) {
895 parser->buffer[parser->idx++] = ch;
908 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
916 if (s->len <= s->readpos)
919 len = s->len - s->readpos;
922 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
932 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
936 if (s->len <= s->readpos)
939 len = s->len - s->readpos;
942 memcpy(buf, s->buffer + s->readpos, cnt);
949 * ftrace_max_lock is used to protect the swapping of buffers
950 * when taking a max snapshot. The buffers themselves are
951 * protected by per_cpu spinlocks. But the action of the swap
952 * needs its own lock.
954 * This is defined as a arch_spinlock_t in order to help
955 * with performance when lockdep debugging is enabled.
957 * It is also used in other places outside the update_max_tr
958 * so it needs to be defined outside of the
959 * CONFIG_TRACER_MAX_TRACE.
961 static arch_spinlock_t ftrace_max_lock =
962 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
964 unsigned long __read_mostly tracing_thresh;
966 #ifdef CONFIG_TRACER_MAX_TRACE
967 unsigned long __read_mostly tracing_max_latency;
970 * Copy the new maximum trace into the separate maximum-trace
971 * structure. (this way the maximum trace is permanently saved,
972 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
975 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
977 struct trace_buffer *trace_buf = &tr->trace_buffer;
978 struct trace_buffer *max_buf = &tr->max_buffer;
979 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
980 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
983 max_buf->time_start = data->preempt_timestamp;
985 max_data->saved_latency = tracing_max_latency;
986 max_data->critical_start = data->critical_start;
987 max_data->critical_end = data->critical_end;
989 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
990 max_data->pid = tsk->pid;
992 * If tsk == current, then use current_uid(), as that does not use
993 * RCU. The irq tracer can be called out of RCU scope.
996 max_data->uid = current_uid();
998 max_data->uid = task_uid(tsk);
1000 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1001 max_data->policy = tsk->policy;
1002 max_data->rt_priority = tsk->rt_priority;
1004 /* record this tasks comm */
1005 tracing_record_cmdline(tsk);
1009 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1011 * @tsk: the task with the latency
1012 * @cpu: The cpu that initiated the trace.
1014 * Flip the buffers between the @tr and the max_tr and record information
1015 * about which task was the cause of this latency.
1018 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1020 struct ring_buffer *buf;
1025 WARN_ON_ONCE(!irqs_disabled());
1027 if (!tr->allocated_snapshot) {
1028 /* Only the nop tracer should hit this when disabling */
1029 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1033 arch_spin_lock(&ftrace_max_lock);
1035 buf = tr->trace_buffer.buffer;
1036 tr->trace_buffer.buffer = tr->max_buffer.buffer;
1037 tr->max_buffer.buffer = buf;
1039 __update_max_tr(tr, tsk, cpu);
1040 arch_spin_unlock(&ftrace_max_lock);
1044 * update_max_tr_single - only copy one trace over, and reset the rest
1046 * @tsk - task with the latency
1047 * @cpu - the cpu of the buffer to copy.
1049 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1052 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1059 WARN_ON_ONCE(!irqs_disabled());
1060 if (!tr->allocated_snapshot) {
1061 /* Only the nop tracer should hit this when disabling */
1062 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1066 arch_spin_lock(&ftrace_max_lock);
1068 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->trace_buffer.buffer, cpu);
1070 if (ret == -EBUSY) {
1072 * We failed to swap the buffer due to a commit taking
1073 * place on this CPU. We fail to record, but we reset
1074 * the max trace buffer (no one writes directly to it)
1075 * and flag that it failed.
1077 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1078 "Failed to swap buffers due to commit in progress\n");
1081 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1083 __update_max_tr(tr, tsk, cpu);
1084 arch_spin_unlock(&ftrace_max_lock);
1086 #endif /* CONFIG_TRACER_MAX_TRACE */
1088 static void default_wait_pipe(struct trace_iterator *iter)
1090 /* Iterators are static, they should be filled or empty */
1091 if (trace_buffer_iter(iter, iter->cpu_file))
1094 ring_buffer_wait(iter->trace_buffer->buffer, iter->cpu_file);
1097 #ifdef CONFIG_FTRACE_STARTUP_TEST
1098 static int run_tracer_selftest(struct tracer *type)
1100 struct trace_array *tr = &global_trace;
1101 struct tracer *saved_tracer = tr->current_trace;
1104 if (!type->selftest || tracing_selftest_disabled)
1108 * Run a selftest on this tracer.
1109 * Here we reset the trace buffer, and set the current
1110 * tracer to be this tracer. The tracer can then run some
1111 * internal tracing to verify that everything is in order.
1112 * If we fail, we do not register this tracer.
1114 tracing_reset_online_cpus(&tr->trace_buffer);
1116 tr->current_trace = type;
1118 #ifdef CONFIG_TRACER_MAX_TRACE
1119 if (type->use_max_tr) {
1120 /* If we expanded the buffers, make sure the max is expanded too */
1121 if (ring_buffer_expanded)
1122 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1123 RING_BUFFER_ALL_CPUS);
1124 tr->allocated_snapshot = true;
1128 /* the test is responsible for initializing and enabling */
1129 pr_info("Testing tracer %s: ", type->name);
1130 ret = type->selftest(type, tr);
1131 /* the test is responsible for resetting too */
1132 tr->current_trace = saved_tracer;
1134 printk(KERN_CONT "FAILED!\n");
1135 /* Add the warning after printing 'FAILED' */
1139 /* Only reset on passing, to avoid touching corrupted buffers */
1140 tracing_reset_online_cpus(&tr->trace_buffer);
1142 #ifdef CONFIG_TRACER_MAX_TRACE
1143 if (type->use_max_tr) {
1144 tr->allocated_snapshot = false;
1146 /* Shrink the max buffer again */
1147 if (ring_buffer_expanded)
1148 ring_buffer_resize(tr->max_buffer.buffer, 1,
1149 RING_BUFFER_ALL_CPUS);
1153 printk(KERN_CONT "PASSED\n");
1157 static inline int run_tracer_selftest(struct tracer *type)
1161 #endif /* CONFIG_FTRACE_STARTUP_TEST */
1164 * register_tracer - register a tracer with the ftrace system.
1165 * @type - the plugin for the tracer
1167 * Register a new plugin tracer.
1169 int register_tracer(struct tracer *type)
1175 pr_info("Tracer must have a name\n");
1179 if (strlen(type->name) >= MAX_TRACER_SIZE) {
1180 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
1184 mutex_lock(&trace_types_lock);
1186 tracing_selftest_running = true;
1188 for (t = trace_types; t; t = t->next) {
1189 if (strcmp(type->name, t->name) == 0) {
1191 pr_info("Tracer %s already registered\n",
1198 if (!type->set_flag)
1199 type->set_flag = &dummy_set_flag;
1201 type->flags = &dummy_tracer_flags;
1203 if (!type->flags->opts)
1204 type->flags->opts = dummy_tracer_opt;
1205 if (!type->wait_pipe)
1206 type->wait_pipe = default_wait_pipe;
1208 ret = run_tracer_selftest(type);
1212 type->next = trace_types;
1216 tracing_selftest_running = false;
1217 mutex_unlock(&trace_types_lock);
1219 if (ret || !default_bootup_tracer)
1222 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
1225 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
1226 /* Do we want this tracer to start on bootup? */
1227 tracing_set_tracer(type->name);
1228 default_bootup_tracer = NULL;
1229 /* disable other selftests, since this will break it. */
1230 tracing_selftest_disabled = true;
1231 #ifdef CONFIG_FTRACE_STARTUP_TEST
1232 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
1240 void tracing_reset(struct trace_buffer *buf, int cpu)
1242 struct ring_buffer *buffer = buf->buffer;
1247 ring_buffer_record_disable(buffer);
1249 /* Make sure all commits have finished */
1250 synchronize_sched();
1251 ring_buffer_reset_cpu(buffer, cpu);
1253 ring_buffer_record_enable(buffer);
1256 void tracing_reset_online_cpus(struct trace_buffer *buf)
1258 struct ring_buffer *buffer = buf->buffer;
1264 ring_buffer_record_disable(buffer);
1266 /* Make sure all commits have finished */
1267 synchronize_sched();
1269 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
1271 for_each_online_cpu(cpu)
1272 ring_buffer_reset_cpu(buffer, cpu);
1274 ring_buffer_record_enable(buffer);
1277 /* Must have trace_types_lock held */
1278 void tracing_reset_all_online_cpus(void)
1280 struct trace_array *tr;
1282 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1283 tracing_reset_online_cpus(&tr->trace_buffer);
1284 #ifdef CONFIG_TRACER_MAX_TRACE
1285 tracing_reset_online_cpus(&tr->max_buffer);
1290 #define SAVED_CMDLINES 128
1291 #define NO_CMDLINE_MAP UINT_MAX
1292 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
1293 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
1294 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
1295 static int cmdline_idx;
1296 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1298 /* temporary disable recording */
1299 static atomic_t trace_record_cmdline_disabled __read_mostly;
1301 static void trace_init_cmdlines(void)
1303 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
1304 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
1308 int is_tracing_stopped(void)
1310 return global_trace.stop_count;
1314 * tracing_start - quick start of the tracer
1316 * If tracing is enabled but was stopped by tracing_stop,
1317 * this will start the tracer back up.
1319 void tracing_start(void)
1321 struct ring_buffer *buffer;
1322 unsigned long flags;
1324 if (tracing_disabled)
1327 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1328 if (--global_trace.stop_count) {
1329 if (global_trace.stop_count < 0) {
1330 /* Someone screwed up their debugging */
1332 global_trace.stop_count = 0;
1337 /* Prevent the buffers from switching */
1338 arch_spin_lock(&ftrace_max_lock);
1340 buffer = global_trace.trace_buffer.buffer;
1342 ring_buffer_record_enable(buffer);
1344 #ifdef CONFIG_TRACER_MAX_TRACE
1345 buffer = global_trace.max_buffer.buffer;
1347 ring_buffer_record_enable(buffer);
1350 arch_spin_unlock(&ftrace_max_lock);
1354 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1357 static void tracing_start_tr(struct trace_array *tr)
1359 struct ring_buffer *buffer;
1360 unsigned long flags;
1362 if (tracing_disabled)
1365 /* If global, we need to also start the max tracer */
1366 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1367 return tracing_start();
1369 raw_spin_lock_irqsave(&tr->start_lock, flags);
1371 if (--tr->stop_count) {
1372 if (tr->stop_count < 0) {
1373 /* Someone screwed up their debugging */
1380 buffer = tr->trace_buffer.buffer;
1382 ring_buffer_record_enable(buffer);
1385 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1389 * tracing_stop - quick stop of the tracer
1391 * Light weight way to stop tracing. Use in conjunction with
1394 void tracing_stop(void)
1396 struct ring_buffer *buffer;
1397 unsigned long flags;
1400 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1401 if (global_trace.stop_count++)
1404 /* Prevent the buffers from switching */
1405 arch_spin_lock(&ftrace_max_lock);
1407 buffer = global_trace.trace_buffer.buffer;
1409 ring_buffer_record_disable(buffer);
1411 #ifdef CONFIG_TRACER_MAX_TRACE
1412 buffer = global_trace.max_buffer.buffer;
1414 ring_buffer_record_disable(buffer);
1417 arch_spin_unlock(&ftrace_max_lock);
1420 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1423 static void tracing_stop_tr(struct trace_array *tr)
1425 struct ring_buffer *buffer;
1426 unsigned long flags;
1428 /* If global, we need to also stop the max tracer */
1429 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1430 return tracing_stop();
1432 raw_spin_lock_irqsave(&tr->start_lock, flags);
1433 if (tr->stop_count++)
1436 buffer = tr->trace_buffer.buffer;
1438 ring_buffer_record_disable(buffer);
1441 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1444 void trace_stop_cmdline_recording(void);
1446 static void trace_save_cmdline(struct task_struct *tsk)
1450 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1454 * It's not the end of the world if we don't get
1455 * the lock, but we also don't want to spin
1456 * nor do we want to disable interrupts,
1457 * so if we miss here, then better luck next time.
1459 if (!arch_spin_trylock(&trace_cmdline_lock))
1462 idx = map_pid_to_cmdline[tsk->pid];
1463 if (idx == NO_CMDLINE_MAP) {
1464 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1467 * Check whether the cmdline buffer at idx has a pid
1468 * mapped. We are going to overwrite that entry so we
1469 * need to clear the map_pid_to_cmdline. Otherwise we
1470 * would read the new comm for the old pid.
1472 pid = map_cmdline_to_pid[idx];
1473 if (pid != NO_CMDLINE_MAP)
1474 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1476 map_cmdline_to_pid[idx] = tsk->pid;
1477 map_pid_to_cmdline[tsk->pid] = idx;
1482 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1484 arch_spin_unlock(&trace_cmdline_lock);
1487 void trace_find_cmdline(int pid, char comm[])
1492 strcpy(comm, "<idle>");
1496 if (WARN_ON_ONCE(pid < 0)) {
1497 strcpy(comm, "<XXX>");
1501 if (pid > PID_MAX_DEFAULT) {
1502 strcpy(comm, "<...>");
1507 arch_spin_lock(&trace_cmdline_lock);
1508 map = map_pid_to_cmdline[pid];
1509 if (map != NO_CMDLINE_MAP)
1510 strcpy(comm, saved_cmdlines[map]);
1512 strcpy(comm, "<...>");
1514 arch_spin_unlock(&trace_cmdline_lock);
1518 void tracing_record_cmdline(struct task_struct *tsk)
1520 if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on())
1523 if (!__this_cpu_read(trace_cmdline_save))
1526 __this_cpu_write(trace_cmdline_save, false);
1528 trace_save_cmdline(tsk);
1532 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1535 struct task_struct *tsk = current;
1537 entry->preempt_count = pc & 0xff;
1538 entry->pid = (tsk) ? tsk->pid : 0;
1540 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1541 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1543 TRACE_FLAG_IRQS_NOSUPPORT |
1545 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1546 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1547 (tif_need_resched() ? TRACE_FLAG_NEED_RESCHED : 0) |
1548 (test_preempt_need_resched() ? TRACE_FLAG_PREEMPT_RESCHED : 0);
1550 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1552 struct ring_buffer_event *
1553 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1556 unsigned long flags, int pc)
1558 struct ring_buffer_event *event;
1560 event = ring_buffer_lock_reserve(buffer, len);
1561 if (event != NULL) {
1562 struct trace_entry *ent = ring_buffer_event_data(event);
1564 tracing_generic_entry_update(ent, flags, pc);
1572 __buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event)
1574 __this_cpu_write(trace_cmdline_save, true);
1575 ring_buffer_unlock_commit(buffer, event);
1579 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1580 struct ring_buffer_event *event,
1581 unsigned long flags, int pc)
1583 __buffer_unlock_commit(buffer, event);
1585 ftrace_trace_stack(buffer, flags, 6, pc);
1586 ftrace_trace_userstack(buffer, flags, pc);
1589 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1590 struct ring_buffer_event *event,
1591 unsigned long flags, int pc)
1593 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1595 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit);
1597 struct ring_buffer_event *
1598 trace_event_buffer_lock_reserve(struct ring_buffer **current_rb,
1599 struct ftrace_event_file *ftrace_file,
1600 int type, unsigned long len,
1601 unsigned long flags, int pc)
1603 *current_rb = ftrace_file->tr->trace_buffer.buffer;
1604 return trace_buffer_lock_reserve(*current_rb,
1605 type, len, flags, pc);
1607 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
1609 struct ring_buffer_event *
1610 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1611 int type, unsigned long len,
1612 unsigned long flags, int pc)
1614 *current_rb = global_trace.trace_buffer.buffer;
1615 return trace_buffer_lock_reserve(*current_rb,
1616 type, len, flags, pc);
1618 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1620 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1621 struct ring_buffer_event *event,
1622 unsigned long flags, int pc)
1624 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1626 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1628 void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1629 struct ring_buffer_event *event,
1630 unsigned long flags, int pc,
1631 struct pt_regs *regs)
1633 __buffer_unlock_commit(buffer, event);
1635 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1636 ftrace_trace_userstack(buffer, flags, pc);
1638 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs);
1640 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1641 struct ring_buffer_event *event)
1643 ring_buffer_discard_commit(buffer, event);
1645 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1648 trace_function(struct trace_array *tr,
1649 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1652 struct ftrace_event_call *call = &event_function;
1653 struct ring_buffer *buffer = tr->trace_buffer.buffer;
1654 struct ring_buffer_event *event;
1655 struct ftrace_entry *entry;
1657 /* If we are reading the ring buffer, don't trace */
1658 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1661 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1665 entry = ring_buffer_event_data(event);
1667 entry->parent_ip = parent_ip;
1669 if (!call_filter_check_discard(call, entry, buffer, event))
1670 __buffer_unlock_commit(buffer, event);
1673 #ifdef CONFIG_STACKTRACE
1675 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1676 struct ftrace_stack {
1677 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1680 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1681 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1683 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1684 unsigned long flags,
1685 int skip, int pc, struct pt_regs *regs)
1687 struct ftrace_event_call *call = &event_kernel_stack;
1688 struct ring_buffer_event *event;
1689 struct stack_entry *entry;
1690 struct stack_trace trace;
1692 int size = FTRACE_STACK_ENTRIES;
1694 trace.nr_entries = 0;
1698 * Since events can happen in NMIs there's no safe way to
1699 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1700 * or NMI comes in, it will just have to use the default
1701 * FTRACE_STACK_SIZE.
1703 preempt_disable_notrace();
1705 use_stack = __this_cpu_inc_return(ftrace_stack_reserve);
1707 * We don't need any atomic variables, just a barrier.
1708 * If an interrupt comes in, we don't care, because it would
1709 * have exited and put the counter back to what we want.
1710 * We just need a barrier to keep gcc from moving things
1714 if (use_stack == 1) {
1715 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1716 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1719 save_stack_trace_regs(regs, &trace);
1721 save_stack_trace(&trace);
1723 if (trace.nr_entries > size)
1724 size = trace.nr_entries;
1726 /* From now on, use_stack is a boolean */
1729 size *= sizeof(unsigned long);
1731 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1732 sizeof(*entry) + size, flags, pc);
1735 entry = ring_buffer_event_data(event);
1737 memset(&entry->caller, 0, size);
1740 memcpy(&entry->caller, trace.entries,
1741 trace.nr_entries * sizeof(unsigned long));
1743 trace.max_entries = FTRACE_STACK_ENTRIES;
1744 trace.entries = entry->caller;
1746 save_stack_trace_regs(regs, &trace);
1748 save_stack_trace(&trace);
1751 entry->size = trace.nr_entries;
1753 if (!call_filter_check_discard(call, entry, buffer, event))
1754 __buffer_unlock_commit(buffer, event);
1757 /* Again, don't let gcc optimize things here */
1759 __this_cpu_dec(ftrace_stack_reserve);
1760 preempt_enable_notrace();
1764 void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1765 int skip, int pc, struct pt_regs *regs)
1767 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1770 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1773 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1776 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1779 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1782 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1785 __ftrace_trace_stack(tr->trace_buffer.buffer, flags, skip, pc, NULL);
1789 * trace_dump_stack - record a stack back trace in the trace buffer
1790 * @skip: Number of functions to skip (helper handlers)
1792 void trace_dump_stack(int skip)
1794 unsigned long flags;
1796 if (tracing_disabled || tracing_selftest_running)
1799 local_save_flags(flags);
1802 * Skip 3 more, seems to get us at the caller of
1806 __ftrace_trace_stack(global_trace.trace_buffer.buffer,
1807 flags, skip, preempt_count(), NULL);
1810 static DEFINE_PER_CPU(int, user_stack_count);
1813 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1815 struct ftrace_event_call *call = &event_user_stack;
1816 struct ring_buffer_event *event;
1817 struct userstack_entry *entry;
1818 struct stack_trace trace;
1820 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1824 * NMIs can not handle page faults, even with fix ups.
1825 * The save user stack can (and often does) fault.
1827 if (unlikely(in_nmi()))
1831 * prevent recursion, since the user stack tracing may
1832 * trigger other kernel events.
1835 if (__this_cpu_read(user_stack_count))
1838 __this_cpu_inc(user_stack_count);
1840 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1841 sizeof(*entry), flags, pc);
1843 goto out_drop_count;
1844 entry = ring_buffer_event_data(event);
1846 entry->tgid = current->tgid;
1847 memset(&entry->caller, 0, sizeof(entry->caller));
1849 trace.nr_entries = 0;
1850 trace.max_entries = FTRACE_STACK_ENTRIES;
1852 trace.entries = entry->caller;
1854 save_stack_trace_user(&trace);
1855 if (!call_filter_check_discard(call, entry, buffer, event))
1856 __buffer_unlock_commit(buffer, event);
1859 __this_cpu_dec(user_stack_count);
1865 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1867 ftrace_trace_userstack(tr, flags, preempt_count());
1871 #endif /* CONFIG_STACKTRACE */
1873 /* created for use with alloc_percpu */
1874 struct trace_buffer_struct {
1875 char buffer[TRACE_BUF_SIZE];
1878 static struct trace_buffer_struct *trace_percpu_buffer;
1879 static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1880 static struct trace_buffer_struct *trace_percpu_irq_buffer;
1881 static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1884 * The buffer used is dependent on the context. There is a per cpu
1885 * buffer for normal context, softirq contex, hard irq context and
1886 * for NMI context. Thise allows for lockless recording.
1888 * Note, if the buffers failed to be allocated, then this returns NULL
1890 static char *get_trace_buf(void)
1892 struct trace_buffer_struct *percpu_buffer;
1895 * If we have allocated per cpu buffers, then we do not
1896 * need to do any locking.
1899 percpu_buffer = trace_percpu_nmi_buffer;
1901 percpu_buffer = trace_percpu_irq_buffer;
1902 else if (in_softirq())
1903 percpu_buffer = trace_percpu_sirq_buffer;
1905 percpu_buffer = trace_percpu_buffer;
1910 return this_cpu_ptr(&percpu_buffer->buffer[0]);
1913 static int alloc_percpu_trace_buffer(void)
1915 struct trace_buffer_struct *buffers;
1916 struct trace_buffer_struct *sirq_buffers;
1917 struct trace_buffer_struct *irq_buffers;
1918 struct trace_buffer_struct *nmi_buffers;
1920 buffers = alloc_percpu(struct trace_buffer_struct);
1924 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
1928 irq_buffers = alloc_percpu(struct trace_buffer_struct);
1932 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
1936 trace_percpu_buffer = buffers;
1937 trace_percpu_sirq_buffer = sirq_buffers;
1938 trace_percpu_irq_buffer = irq_buffers;
1939 trace_percpu_nmi_buffer = nmi_buffers;
1944 free_percpu(irq_buffers);
1946 free_percpu(sirq_buffers);
1948 free_percpu(buffers);
1950 WARN(1, "Could not allocate percpu trace_printk buffer");
1954 static int buffers_allocated;
1956 void trace_printk_init_buffers(void)
1958 if (buffers_allocated)
1961 if (alloc_percpu_trace_buffer())
1964 pr_info("ftrace: Allocated trace_printk buffers\n");
1966 /* Expand the buffers to set size */
1967 tracing_update_buffers();
1969 buffers_allocated = 1;
1972 * trace_printk_init_buffers() can be called by modules.
1973 * If that happens, then we need to start cmdline recording
1974 * directly here. If the global_trace.buffer is already
1975 * allocated here, then this was called by module code.
1977 if (global_trace.trace_buffer.buffer)
1978 tracing_start_cmdline_record();
1981 void trace_printk_start_comm(void)
1983 /* Start tracing comms if trace printk is set */
1984 if (!buffers_allocated)
1986 tracing_start_cmdline_record();
1989 static void trace_printk_start_stop_comm(int enabled)
1991 if (!buffers_allocated)
1995 tracing_start_cmdline_record();
1997 tracing_stop_cmdline_record();
2001 * trace_vbprintk - write binary msg to tracing buffer
2004 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
2006 struct ftrace_event_call *call = &event_bprint;
2007 struct ring_buffer_event *event;
2008 struct ring_buffer *buffer;
2009 struct trace_array *tr = &global_trace;
2010 struct bprint_entry *entry;
2011 unsigned long flags;
2013 int len = 0, size, pc;
2015 if (unlikely(tracing_selftest_running || tracing_disabled))
2018 /* Don't pollute graph traces with trace_vprintk internals */
2019 pause_graph_tracing();
2021 pc = preempt_count();
2022 preempt_disable_notrace();
2024 tbuffer = get_trace_buf();
2030 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
2032 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
2035 local_save_flags(flags);
2036 size = sizeof(*entry) + sizeof(u32) * len;
2037 buffer = tr->trace_buffer.buffer;
2038 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
2042 entry = ring_buffer_event_data(event);
2046 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
2047 if (!call_filter_check_discard(call, entry, buffer, event)) {
2048 __buffer_unlock_commit(buffer, event);
2049 ftrace_trace_stack(buffer, flags, 6, pc);
2053 preempt_enable_notrace();
2054 unpause_graph_tracing();
2058 EXPORT_SYMBOL_GPL(trace_vbprintk);
2061 __trace_array_vprintk(struct ring_buffer *buffer,
2062 unsigned long ip, const char *fmt, va_list args)
2064 struct ftrace_event_call *call = &event_print;
2065 struct ring_buffer_event *event;
2066 int len = 0, size, pc;
2067 struct print_entry *entry;
2068 unsigned long flags;
2071 if (tracing_disabled || tracing_selftest_running)
2074 /* Don't pollute graph traces with trace_vprintk internals */
2075 pause_graph_tracing();
2077 pc = preempt_count();
2078 preempt_disable_notrace();
2081 tbuffer = get_trace_buf();
2087 len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
2088 if (len > TRACE_BUF_SIZE)
2091 local_save_flags(flags);
2092 size = sizeof(*entry) + len + 1;
2093 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
2097 entry = ring_buffer_event_data(event);
2100 memcpy(&entry->buf, tbuffer, len);
2101 entry->buf[len] = '\0';
2102 if (!call_filter_check_discard(call, entry, buffer, event)) {
2103 __buffer_unlock_commit(buffer, event);
2104 ftrace_trace_stack(buffer, flags, 6, pc);
2107 preempt_enable_notrace();
2108 unpause_graph_tracing();
2113 int trace_array_vprintk(struct trace_array *tr,
2114 unsigned long ip, const char *fmt, va_list args)
2116 return __trace_array_vprintk(tr->trace_buffer.buffer, ip, fmt, args);
2119 int trace_array_printk(struct trace_array *tr,
2120 unsigned long ip, const char *fmt, ...)
2125 if (!(trace_flags & TRACE_ITER_PRINTK))
2129 ret = trace_array_vprintk(tr, ip, fmt, ap);
2134 int trace_array_printk_buf(struct ring_buffer *buffer,
2135 unsigned long ip, const char *fmt, ...)
2140 if (!(trace_flags & TRACE_ITER_PRINTK))
2144 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
2149 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
2151 return trace_array_vprintk(&global_trace, ip, fmt, args);
2153 EXPORT_SYMBOL_GPL(trace_vprintk);
2155 static void trace_iterator_increment(struct trace_iterator *iter)
2157 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
2161 ring_buffer_read(buf_iter, NULL);
2164 static struct trace_entry *
2165 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
2166 unsigned long *lost_events)
2168 struct ring_buffer_event *event;
2169 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
2172 event = ring_buffer_iter_peek(buf_iter, ts);
2174 event = ring_buffer_peek(iter->trace_buffer->buffer, cpu, ts,
2178 iter->ent_size = ring_buffer_event_length(event);
2179 return ring_buffer_event_data(event);
2185 static struct trace_entry *
2186 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
2187 unsigned long *missing_events, u64 *ent_ts)
2189 struct ring_buffer *buffer = iter->trace_buffer->buffer;
2190 struct trace_entry *ent, *next = NULL;
2191 unsigned long lost_events = 0, next_lost = 0;
2192 int cpu_file = iter->cpu_file;
2193 u64 next_ts = 0, ts;
2199 * If we are in a per_cpu trace file, don't bother by iterating over
2200 * all cpu and peek directly.
2202 if (cpu_file > RING_BUFFER_ALL_CPUS) {
2203 if (ring_buffer_empty_cpu(buffer, cpu_file))
2205 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
2207 *ent_cpu = cpu_file;
2212 for_each_tracing_cpu(cpu) {
2214 if (ring_buffer_empty_cpu(buffer, cpu))
2217 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
2220 * Pick the entry with the smallest timestamp:
2222 if (ent && (!next || ts < next_ts)) {
2226 next_lost = lost_events;
2227 next_size = iter->ent_size;
2231 iter->ent_size = next_size;
2234 *ent_cpu = next_cpu;
2240 *missing_events = next_lost;
2245 /* Find the next real entry, without updating the iterator itself */
2246 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
2247 int *ent_cpu, u64 *ent_ts)
2249 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
2252 /* Find the next real entry, and increment the iterator to the next entry */
2253 void *trace_find_next_entry_inc(struct trace_iterator *iter)
2255 iter->ent = __find_next_entry(iter, &iter->cpu,
2256 &iter->lost_events, &iter->ts);
2259 trace_iterator_increment(iter);
2261 return iter->ent ? iter : NULL;
2264 static void trace_consume(struct trace_iterator *iter)
2266 ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu, &iter->ts,
2267 &iter->lost_events);
2270 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
2272 struct trace_iterator *iter = m->private;
2276 WARN_ON_ONCE(iter->leftover);
2280 /* can't go backwards */
2285 ent = trace_find_next_entry_inc(iter);
2289 while (ent && iter->idx < i)
2290 ent = trace_find_next_entry_inc(iter);
2297 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
2299 struct ring_buffer_event *event;
2300 struct ring_buffer_iter *buf_iter;
2301 unsigned long entries = 0;
2304 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = 0;
2306 buf_iter = trace_buffer_iter(iter, cpu);
2310 ring_buffer_iter_reset(buf_iter);
2313 * We could have the case with the max latency tracers
2314 * that a reset never took place on a cpu. This is evident
2315 * by the timestamp being before the start of the buffer.
2317 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
2318 if (ts >= iter->trace_buffer->time_start)
2321 ring_buffer_read(buf_iter, NULL);
2324 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = entries;
2328 * The current tracer is copied to avoid a global locking
2331 static void *s_start(struct seq_file *m, loff_t *pos)
2333 struct trace_iterator *iter = m->private;
2334 struct trace_array *tr = iter->tr;
2335 int cpu_file = iter->cpu_file;
2341 * copy the tracer to avoid using a global lock all around.
2342 * iter->trace is a copy of current_trace, the pointer to the
2343 * name may be used instead of a strcmp(), as iter->trace->name
2344 * will point to the same string as current_trace->name.
2346 mutex_lock(&trace_types_lock);
2347 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
2348 *iter->trace = *tr->current_trace;
2349 mutex_unlock(&trace_types_lock);
2351 #ifdef CONFIG_TRACER_MAX_TRACE
2352 if (iter->snapshot && iter->trace->use_max_tr)
2353 return ERR_PTR(-EBUSY);
2356 if (!iter->snapshot)
2357 atomic_inc(&trace_record_cmdline_disabled);
2359 if (*pos != iter->pos) {
2364 if (cpu_file == RING_BUFFER_ALL_CPUS) {
2365 for_each_tracing_cpu(cpu)
2366 tracing_iter_reset(iter, cpu);
2368 tracing_iter_reset(iter, cpu_file);
2371 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
2376 * If we overflowed the seq_file before, then we want
2377 * to just reuse the trace_seq buffer again.
2383 p = s_next(m, p, &l);
2387 trace_event_read_lock();
2388 trace_access_lock(cpu_file);
2392 static void s_stop(struct seq_file *m, void *p)
2394 struct trace_iterator *iter = m->private;
2396 #ifdef CONFIG_TRACER_MAX_TRACE
2397 if (iter->snapshot && iter->trace->use_max_tr)
2401 if (!iter->snapshot)
2402 atomic_dec(&trace_record_cmdline_disabled);
2404 trace_access_unlock(iter->cpu_file);
2405 trace_event_read_unlock();
2409 get_total_entries(struct trace_buffer *buf,
2410 unsigned long *total, unsigned long *entries)
2412 unsigned long count;
2418 for_each_tracing_cpu(cpu) {
2419 count = ring_buffer_entries_cpu(buf->buffer, cpu);
2421 * If this buffer has skipped entries, then we hold all
2422 * entries for the trace and we need to ignore the
2423 * ones before the time stamp.
2425 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
2426 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
2427 /* total is the same as the entries */
2431 ring_buffer_overrun_cpu(buf->buffer, cpu);
2436 static void print_lat_help_header(struct seq_file *m)
2438 seq_puts(m, "# _------=> CPU# \n");
2439 seq_puts(m, "# / _-----=> irqs-off \n");
2440 seq_puts(m, "# | / _----=> need-resched \n");
2441 seq_puts(m, "# || / _---=> hardirq/softirq \n");
2442 seq_puts(m, "# ||| / _--=> preempt-depth \n");
2443 seq_puts(m, "# |||| / delay \n");
2444 seq_puts(m, "# cmd pid ||||| time | caller \n");
2445 seq_puts(m, "# \\ / ||||| \\ | / \n");
2448 static void print_event_info(struct trace_buffer *buf, struct seq_file *m)
2450 unsigned long total;
2451 unsigned long entries;
2453 get_total_entries(buf, &total, &entries);
2454 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2455 entries, total, num_online_cpus());
2459 static void print_func_help_header(struct trace_buffer *buf, struct seq_file *m)
2461 print_event_info(buf, m);
2462 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2463 seq_puts(m, "# | | | | |\n");
2466 static void print_func_help_header_irq(struct trace_buffer *buf, struct seq_file *m)
2468 print_event_info(buf, m);
2469 seq_puts(m, "# _-----=> irqs-off\n");
2470 seq_puts(m, "# / _----=> need-resched\n");
2471 seq_puts(m, "# | / _---=> hardirq/softirq\n");
2472 seq_puts(m, "# || / _--=> preempt-depth\n");
2473 seq_puts(m, "# ||| / delay\n");
2474 seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2475 seq_puts(m, "# | | | |||| | |\n");
2479 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2481 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2482 struct trace_buffer *buf = iter->trace_buffer;
2483 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
2484 struct tracer *type = iter->trace;
2485 unsigned long entries;
2486 unsigned long total;
2487 const char *name = "preemption";
2491 get_total_entries(buf, &total, &entries);
2493 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2495 seq_puts(m, "# -----------------------------------"
2496 "---------------------------------\n");
2497 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2498 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2499 nsecs_to_usecs(data->saved_latency),
2503 #if defined(CONFIG_PREEMPT_NONE)
2505 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2507 #elif defined(CONFIG_PREEMPT)
2512 /* These are reserved for later use */
2515 seq_printf(m, " #P:%d)\n", num_online_cpus());
2519 seq_puts(m, "# -----------------\n");
2520 seq_printf(m, "# | task: %.16s-%d "
2521 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2522 data->comm, data->pid,
2523 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
2524 data->policy, data->rt_priority);
2525 seq_puts(m, "# -----------------\n");
2527 if (data->critical_start) {
2528 seq_puts(m, "# => started at: ");
2529 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2530 trace_print_seq(m, &iter->seq);
2531 seq_puts(m, "\n# => ended at: ");
2532 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2533 trace_print_seq(m, &iter->seq);
2534 seq_puts(m, "\n#\n");
2540 static void test_cpu_buff_start(struct trace_iterator *iter)
2542 struct trace_seq *s = &iter->seq;
2544 if (!(trace_flags & TRACE_ITER_ANNOTATE))
2547 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2550 if (cpumask_test_cpu(iter->cpu, iter->started))
2553 if (per_cpu_ptr(iter->trace_buffer->data, iter->cpu)->skipped_entries)
2556 cpumask_set_cpu(iter->cpu, iter->started);
2558 /* Don't print started cpu buffer for the first entry of the trace */
2560 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2564 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2566 struct trace_seq *s = &iter->seq;
2567 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2568 struct trace_entry *entry;
2569 struct trace_event *event;
2573 test_cpu_buff_start(iter);
2575 event = ftrace_find_event(entry->type);
2577 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2578 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2579 if (!trace_print_lat_context(iter))
2582 if (!trace_print_context(iter))
2588 return event->funcs->trace(iter, sym_flags, event);
2590 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2593 return TRACE_TYPE_HANDLED;
2595 return TRACE_TYPE_PARTIAL_LINE;
2598 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2600 struct trace_seq *s = &iter->seq;
2601 struct trace_entry *entry;
2602 struct trace_event *event;
2606 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2607 if (!trace_seq_printf(s, "%d %d %llu ",
2608 entry->pid, iter->cpu, iter->ts))
2612 event = ftrace_find_event(entry->type);
2614 return event->funcs->raw(iter, 0, event);
2616 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2619 return TRACE_TYPE_HANDLED;
2621 return TRACE_TYPE_PARTIAL_LINE;
2624 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2626 struct trace_seq *s = &iter->seq;
2627 unsigned char newline = '\n';
2628 struct trace_entry *entry;
2629 struct trace_event *event;
2633 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2634 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2635 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2636 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2639 event = ftrace_find_event(entry->type);
2641 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2642 if (ret != TRACE_TYPE_HANDLED)
2646 SEQ_PUT_FIELD_RET(s, newline);
2648 return TRACE_TYPE_HANDLED;
2651 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2653 struct trace_seq *s = &iter->seq;
2654 struct trace_entry *entry;
2655 struct trace_event *event;
2659 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2660 SEQ_PUT_FIELD_RET(s, entry->pid);
2661 SEQ_PUT_FIELD_RET(s, iter->cpu);
2662 SEQ_PUT_FIELD_RET(s, iter->ts);
2665 event = ftrace_find_event(entry->type);
2666 return event ? event->funcs->binary(iter, 0, event) :
2670 int trace_empty(struct trace_iterator *iter)
2672 struct ring_buffer_iter *buf_iter;
2675 /* If we are looking at one CPU buffer, only check that one */
2676 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
2677 cpu = iter->cpu_file;
2678 buf_iter = trace_buffer_iter(iter, cpu);
2680 if (!ring_buffer_iter_empty(buf_iter))
2683 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2689 for_each_tracing_cpu(cpu) {
2690 buf_iter = trace_buffer_iter(iter, cpu);
2692 if (!ring_buffer_iter_empty(buf_iter))
2695 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2703 /* Called with trace_event_read_lock() held. */
2704 enum print_line_t print_trace_line(struct trace_iterator *iter)
2706 enum print_line_t ret;
2708 if (iter->lost_events &&
2709 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2710 iter->cpu, iter->lost_events))
2711 return TRACE_TYPE_PARTIAL_LINE;
2713 if (iter->trace && iter->trace->print_line) {
2714 ret = iter->trace->print_line(iter);
2715 if (ret != TRACE_TYPE_UNHANDLED)
2719 if (iter->ent->type == TRACE_BPUTS &&
2720 trace_flags & TRACE_ITER_PRINTK &&
2721 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2722 return trace_print_bputs_msg_only(iter);
2724 if (iter->ent->type == TRACE_BPRINT &&
2725 trace_flags & TRACE_ITER_PRINTK &&
2726 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2727 return trace_print_bprintk_msg_only(iter);
2729 if (iter->ent->type == TRACE_PRINT &&
2730 trace_flags & TRACE_ITER_PRINTK &&
2731 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2732 return trace_print_printk_msg_only(iter);
2734 if (trace_flags & TRACE_ITER_BIN)
2735 return print_bin_fmt(iter);
2737 if (trace_flags & TRACE_ITER_HEX)
2738 return print_hex_fmt(iter);
2740 if (trace_flags & TRACE_ITER_RAW)
2741 return print_raw_fmt(iter);
2743 return print_trace_fmt(iter);
2746 void trace_latency_header(struct seq_file *m)
2748 struct trace_iterator *iter = m->private;
2750 /* print nothing if the buffers are empty */
2751 if (trace_empty(iter))
2754 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2755 print_trace_header(m, iter);
2757 if (!(trace_flags & TRACE_ITER_VERBOSE))
2758 print_lat_help_header(m);
2761 void trace_default_header(struct seq_file *m)
2763 struct trace_iterator *iter = m->private;
2765 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2768 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2769 /* print nothing if the buffers are empty */
2770 if (trace_empty(iter))
2772 print_trace_header(m, iter);
2773 if (!(trace_flags & TRACE_ITER_VERBOSE))
2774 print_lat_help_header(m);
2776 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2777 if (trace_flags & TRACE_ITER_IRQ_INFO)
2778 print_func_help_header_irq(iter->trace_buffer, m);
2780 print_func_help_header(iter->trace_buffer, m);
2785 static void test_ftrace_alive(struct seq_file *m)
2787 if (!ftrace_is_dead())
2789 seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2790 seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2793 #ifdef CONFIG_TRACER_MAX_TRACE
2794 static void show_snapshot_main_help(struct seq_file *m)
2796 seq_printf(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n");
2797 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2798 seq_printf(m, "# Takes a snapshot of the main buffer.\n");
2799 seq_printf(m, "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n");
2800 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2801 seq_printf(m, "# is not a '0' or '1')\n");
2804 static void show_snapshot_percpu_help(struct seq_file *m)
2806 seq_printf(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
2807 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
2808 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2809 seq_printf(m, "# Takes a snapshot of the main buffer for this cpu.\n");
2811 seq_printf(m, "# echo 1 > snapshot : Not supported with this kernel.\n");
2812 seq_printf(m, "# Must use main snapshot file to allocate.\n");
2814 seq_printf(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n");
2815 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2816 seq_printf(m, "# is not a '0' or '1')\n");
2819 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
2821 if (iter->tr->allocated_snapshot)
2822 seq_printf(m, "#\n# * Snapshot is allocated *\n#\n");
2824 seq_printf(m, "#\n# * Snapshot is freed *\n#\n");
2826 seq_printf(m, "# Snapshot commands:\n");
2827 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
2828 show_snapshot_main_help(m);
2830 show_snapshot_percpu_help(m);
2833 /* Should never be called */
2834 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
2837 static int s_show(struct seq_file *m, void *v)
2839 struct trace_iterator *iter = v;
2842 if (iter->ent == NULL) {
2844 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2846 test_ftrace_alive(m);
2848 if (iter->snapshot && trace_empty(iter))
2849 print_snapshot_help(m, iter);
2850 else if (iter->trace && iter->trace->print_header)
2851 iter->trace->print_header(m);
2853 trace_default_header(m);
2855 } else if (iter->leftover) {
2857 * If we filled the seq_file buffer earlier, we
2858 * want to just show it now.
2860 ret = trace_print_seq(m, &iter->seq);
2862 /* ret should this time be zero, but you never know */
2863 iter->leftover = ret;
2866 print_trace_line(iter);
2867 ret = trace_print_seq(m, &iter->seq);
2869 * If we overflow the seq_file buffer, then it will
2870 * ask us for this data again at start up.
2872 * ret is 0 if seq_file write succeeded.
2875 iter->leftover = ret;
2882 * Should be used after trace_array_get(), trace_types_lock
2883 * ensures that i_cdev was already initialized.
2885 static inline int tracing_get_cpu(struct inode *inode)
2887 if (inode->i_cdev) /* See trace_create_cpu_file() */
2888 return (long)inode->i_cdev - 1;
2889 return RING_BUFFER_ALL_CPUS;
2892 static const struct seq_operations tracer_seq_ops = {
2899 static struct trace_iterator *
2900 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
2902 struct trace_array *tr = inode->i_private;
2903 struct trace_iterator *iter;
2906 if (tracing_disabled)
2907 return ERR_PTR(-ENODEV);
2909 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
2911 return ERR_PTR(-ENOMEM);
2913 iter->buffer_iter = kzalloc(sizeof(*iter->buffer_iter) * num_possible_cpus(),
2915 if (!iter->buffer_iter)
2919 * We make a copy of the current tracer to avoid concurrent
2920 * changes on it while we are reading.
2922 mutex_lock(&trace_types_lock);
2923 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2927 *iter->trace = *tr->current_trace;
2929 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2934 #ifdef CONFIG_TRACER_MAX_TRACE
2935 /* Currently only the top directory has a snapshot */
2936 if (tr->current_trace->print_max || snapshot)
2937 iter->trace_buffer = &tr->max_buffer;
2940 iter->trace_buffer = &tr->trace_buffer;
2941 iter->snapshot = snapshot;
2943 iter->cpu_file = tracing_get_cpu(inode);
2944 mutex_init(&iter->mutex);
2946 /* Notify the tracer early; before we stop tracing. */
2947 if (iter->trace && iter->trace->open)
2948 iter->trace->open(iter);
2950 /* Annotate start of buffers if we had overruns */
2951 if (ring_buffer_overruns(iter->trace_buffer->buffer))
2952 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2954 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
2955 if (trace_clocks[tr->clock_id].in_ns)
2956 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
2958 /* stop the trace while dumping if we are not opening "snapshot" */
2959 if (!iter->snapshot)
2960 tracing_stop_tr(tr);
2962 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
2963 for_each_tracing_cpu(cpu) {
2964 iter->buffer_iter[cpu] =
2965 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
2967 ring_buffer_read_prepare_sync();
2968 for_each_tracing_cpu(cpu) {
2969 ring_buffer_read_start(iter->buffer_iter[cpu]);
2970 tracing_iter_reset(iter, cpu);
2973 cpu = iter->cpu_file;
2974 iter->buffer_iter[cpu] =
2975 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
2976 ring_buffer_read_prepare_sync();
2977 ring_buffer_read_start(iter->buffer_iter[cpu]);
2978 tracing_iter_reset(iter, cpu);
2981 mutex_unlock(&trace_types_lock);
2986 mutex_unlock(&trace_types_lock);
2988 kfree(iter->buffer_iter);
2990 seq_release_private(inode, file);
2991 return ERR_PTR(-ENOMEM);
2994 int tracing_open_generic(struct inode *inode, struct file *filp)
2996 if (tracing_disabled)
2999 filp->private_data = inode->i_private;
3003 bool tracing_is_disabled(void)
3005 return (tracing_disabled) ? true: false;
3009 * Open and update trace_array ref count.
3010 * Must have the current trace_array passed to it.
3012 static int tracing_open_generic_tr(struct inode *inode, struct file *filp)
3014 struct trace_array *tr = inode->i_private;
3016 if (tracing_disabled)
3019 if (trace_array_get(tr) < 0)
3022 filp->private_data = inode->i_private;
3027 static int tracing_release(struct inode *inode, struct file *file)
3029 struct trace_array *tr = inode->i_private;
3030 struct seq_file *m = file->private_data;
3031 struct trace_iterator *iter;
3034 if (!(file->f_mode & FMODE_READ)) {
3035 trace_array_put(tr);
3039 /* Writes do not use seq_file */
3041 mutex_lock(&trace_types_lock);
3043 for_each_tracing_cpu(cpu) {
3044 if (iter->buffer_iter[cpu])
3045 ring_buffer_read_finish(iter->buffer_iter[cpu]);
3048 if (iter->trace && iter->trace->close)
3049 iter->trace->close(iter);
3051 if (!iter->snapshot)
3052 /* reenable tracing if it was previously enabled */
3053 tracing_start_tr(tr);
3055 __trace_array_put(tr);
3057 mutex_unlock(&trace_types_lock);
3059 mutex_destroy(&iter->mutex);
3060 free_cpumask_var(iter->started);
3062 kfree(iter->buffer_iter);
3063 seq_release_private(inode, file);
3068 static int tracing_release_generic_tr(struct inode *inode, struct file *file)
3070 struct trace_array *tr = inode->i_private;
3072 trace_array_put(tr);
3076 static int tracing_single_release_tr(struct inode *inode, struct file *file)
3078 struct trace_array *tr = inode->i_private;
3080 trace_array_put(tr);
3082 return single_release(inode, file);
3085 static int tracing_open(struct inode *inode, struct file *file)
3087 struct trace_array *tr = inode->i_private;
3088 struct trace_iterator *iter;
3091 if (trace_array_get(tr) < 0)
3094 /* If this file was open for write, then erase contents */
3095 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
3096 int cpu = tracing_get_cpu(inode);
3098 if (cpu == RING_BUFFER_ALL_CPUS)
3099 tracing_reset_online_cpus(&tr->trace_buffer);
3101 tracing_reset(&tr->trace_buffer, cpu);
3104 if (file->f_mode & FMODE_READ) {
3105 iter = __tracing_open(inode, file, false);
3107 ret = PTR_ERR(iter);
3108 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
3109 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3113 trace_array_put(tr);
3119 t_next(struct seq_file *m, void *v, loff_t *pos)
3121 struct tracer *t = v;
3131 static void *t_start(struct seq_file *m, loff_t *pos)
3136 mutex_lock(&trace_types_lock);
3137 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
3143 static void t_stop(struct seq_file *m, void *p)
3145 mutex_unlock(&trace_types_lock);
3148 static int t_show(struct seq_file *m, void *v)
3150 struct tracer *t = v;
3155 seq_printf(m, "%s", t->name);
3164 static const struct seq_operations show_traces_seq_ops = {
3171 static int show_traces_open(struct inode *inode, struct file *file)
3173 if (tracing_disabled)
3176 return seq_open(file, &show_traces_seq_ops);
3180 tracing_write_stub(struct file *filp, const char __user *ubuf,
3181 size_t count, loff_t *ppos)
3186 loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
3190 if (file->f_mode & FMODE_READ)
3191 ret = seq_lseek(file, offset, whence);
3193 file->f_pos = ret = 0;
3198 static const struct file_operations tracing_fops = {
3199 .open = tracing_open,
3201 .write = tracing_write_stub,
3202 .llseek = tracing_lseek,
3203 .release = tracing_release,
3206 static const struct file_operations show_traces_fops = {
3207 .open = show_traces_open,
3209 .release = seq_release,
3210 .llseek = seq_lseek,
3214 * The tracer itself will not take this lock, but still we want
3215 * to provide a consistent cpumask to user-space:
3217 static DEFINE_MUTEX(tracing_cpumask_update_lock);
3220 * Temporary storage for the character representation of the
3221 * CPU bitmask (and one more byte for the newline):
3223 static char mask_str[NR_CPUS + 1];
3226 tracing_cpumask_read(struct file *filp, char __user *ubuf,
3227 size_t count, loff_t *ppos)
3229 struct trace_array *tr = file_inode(filp)->i_private;
3232 mutex_lock(&tracing_cpumask_update_lock);
3234 len = cpumask_scnprintf(mask_str, count, tr->tracing_cpumask);
3235 if (count - len < 2) {
3239 len += sprintf(mask_str + len, "\n");
3240 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
3243 mutex_unlock(&tracing_cpumask_update_lock);
3249 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
3250 size_t count, loff_t *ppos)
3252 struct trace_array *tr = file_inode(filp)->i_private;
3253 cpumask_var_t tracing_cpumask_new;
3256 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
3259 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
3263 mutex_lock(&tracing_cpumask_update_lock);
3265 local_irq_disable();
3266 arch_spin_lock(&ftrace_max_lock);
3267 for_each_tracing_cpu(cpu) {
3269 * Increase/decrease the disabled counter if we are
3270 * about to flip a bit in the cpumask:
3272 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
3273 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3274 atomic_inc(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3275 ring_buffer_record_disable_cpu(tr->trace_buffer.buffer, cpu);
3277 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
3278 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3279 atomic_dec(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3280 ring_buffer_record_enable_cpu(tr->trace_buffer.buffer, cpu);
3283 arch_spin_unlock(&ftrace_max_lock);
3286 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
3288 mutex_unlock(&tracing_cpumask_update_lock);
3289 free_cpumask_var(tracing_cpumask_new);
3294 free_cpumask_var(tracing_cpumask_new);
3299 static const struct file_operations tracing_cpumask_fops = {
3300 .open = tracing_open_generic_tr,
3301 .read = tracing_cpumask_read,
3302 .write = tracing_cpumask_write,
3303 .release = tracing_release_generic_tr,
3304 .llseek = generic_file_llseek,
3307 static int tracing_trace_options_show(struct seq_file *m, void *v)
3309 struct tracer_opt *trace_opts;
3310 struct trace_array *tr = m->private;
3314 mutex_lock(&trace_types_lock);
3315 tracer_flags = tr->current_trace->flags->val;
3316 trace_opts = tr->current_trace->flags->opts;
3318 for (i = 0; trace_options[i]; i++) {
3319 if (trace_flags & (1 << i))
3320 seq_printf(m, "%s\n", trace_options[i]);
3322 seq_printf(m, "no%s\n", trace_options[i]);
3325 for (i = 0; trace_opts[i].name; i++) {
3326 if (tracer_flags & trace_opts[i].bit)
3327 seq_printf(m, "%s\n", trace_opts[i].name);
3329 seq_printf(m, "no%s\n", trace_opts[i].name);
3331 mutex_unlock(&trace_types_lock);
3336 static int __set_tracer_option(struct tracer *trace,
3337 struct tracer_flags *tracer_flags,
3338 struct tracer_opt *opts, int neg)
3342 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
3347 tracer_flags->val &= ~opts->bit;
3349 tracer_flags->val |= opts->bit;
3353 /* Try to assign a tracer specific option */
3354 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
3356 struct tracer_flags *tracer_flags = trace->flags;
3357 struct tracer_opt *opts = NULL;
3360 for (i = 0; tracer_flags->opts[i].name; i++) {
3361 opts = &tracer_flags->opts[i];
3363 if (strcmp(cmp, opts->name) == 0)
3364 return __set_tracer_option(trace, trace->flags,
3371 /* Some tracers require overwrite to stay enabled */
3372 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
3374 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
3380 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
3382 /* do nothing if flag is already set */
3383 if (!!(trace_flags & mask) == !!enabled)
3386 /* Give the tracer a chance to approve the change */
3387 if (tr->current_trace->flag_changed)
3388 if (tr->current_trace->flag_changed(tr->current_trace, mask, !!enabled))
3392 trace_flags |= mask;
3394 trace_flags &= ~mask;
3396 if (mask == TRACE_ITER_RECORD_CMD)
3397 trace_event_enable_cmd_record(enabled);
3399 if (mask == TRACE_ITER_OVERWRITE) {
3400 ring_buffer_change_overwrite(tr->trace_buffer.buffer, enabled);
3401 #ifdef CONFIG_TRACER_MAX_TRACE
3402 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
3406 if (mask == TRACE_ITER_PRINTK)
3407 trace_printk_start_stop_comm(enabled);
3412 static int trace_set_options(struct trace_array *tr, char *option)
3419 cmp = strstrip(option);
3421 if (strncmp(cmp, "no", 2) == 0) {
3426 mutex_lock(&trace_types_lock);
3428 for (i = 0; trace_options[i]; i++) {
3429 if (strcmp(cmp, trace_options[i]) == 0) {
3430 ret = set_tracer_flag(tr, 1 << i, !neg);
3435 /* If no option could be set, test the specific tracer options */
3436 if (!trace_options[i])
3437 ret = set_tracer_option(tr->current_trace, cmp, neg);
3439 mutex_unlock(&trace_types_lock);
3445 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
3446 size_t cnt, loff_t *ppos)
3448 struct seq_file *m = filp->private_data;
3449 struct trace_array *tr = m->private;
3453 if (cnt >= sizeof(buf))
3456 if (copy_from_user(&buf, ubuf, cnt))
3461 ret = trace_set_options(tr, buf);
3470 static int tracing_trace_options_open(struct inode *inode, struct file *file)
3472 struct trace_array *tr = inode->i_private;
3475 if (tracing_disabled)
3478 if (trace_array_get(tr) < 0)
3481 ret = single_open(file, tracing_trace_options_show, inode->i_private);
3483 trace_array_put(tr);
3488 static const struct file_operations tracing_iter_fops = {
3489 .open = tracing_trace_options_open,
3491 .llseek = seq_lseek,
3492 .release = tracing_single_release_tr,
3493 .write = tracing_trace_options_write,
3496 static const char readme_msg[] =
3497 "tracing mini-HOWTO:\n\n"
3498 "# echo 0 > tracing_on : quick way to disable tracing\n"
3499 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
3500 " Important files:\n"
3501 " trace\t\t\t- The static contents of the buffer\n"
3502 "\t\t\t To clear the buffer write into this file: echo > trace\n"
3503 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
3504 " current_tracer\t- function and latency tracers\n"
3505 " available_tracers\t- list of configured tracers for current_tracer\n"
3506 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
3507 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
3508 " trace_clock\t\t-change the clock used to order events\n"
3509 " local: Per cpu clock but may not be synced across CPUs\n"
3510 " global: Synced across CPUs but slows tracing down.\n"
3511 " counter: Not a clock, but just an increment\n"
3512 " uptime: Jiffy counter from time of boot\n"
3513 " perf: Same clock that perf events use\n"
3514 #ifdef CONFIG_X86_64
3515 " x86-tsc: TSC cycle counter\n"
3517 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
3518 " tracing_cpumask\t- Limit which CPUs to trace\n"
3519 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
3520 "\t\t\t Remove sub-buffer with rmdir\n"
3521 " trace_options\t\t- Set format or modify how tracing happens\n"
3522 "\t\t\t Disable an option by adding a suffix 'no' to the option name\n"
3523 #ifdef CONFIG_DYNAMIC_FTRACE
3524 "\n available_filter_functions - list of functions that can be filtered on\n"
3525 " set_ftrace_filter\t- echo function name in here to only trace these functions\n"
3526 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3527 " modules: Can select a group via module\n"
3528 " Format: :mod:<module-name>\n"
3529 " example: echo :mod:ext3 > set_ftrace_filter\n"
3530 " triggers: a command to perform when function is hit\n"
3531 " Format: <function>:<trigger>[:count]\n"
3532 " trigger: traceon, traceoff\n"
3533 " enable_event:<system>:<event>\n"
3534 " disable_event:<system>:<event>\n"
3535 #ifdef CONFIG_STACKTRACE
3538 #ifdef CONFIG_TRACER_SNAPSHOT
3541 " example: echo do_fault:traceoff > set_ftrace_filter\n"
3542 " echo do_trap:traceoff:3 > set_ftrace_filter\n"
3543 " The first one will disable tracing every time do_fault is hit\n"
3544 " The second will disable tracing at most 3 times when do_trap is hit\n"
3545 " The first time do trap is hit and it disables tracing, the counter\n"
3546 " will decrement to 2. If tracing is already disabled, the counter\n"
3547 " will not decrement. It only decrements when the trigger did work\n"
3548 " To remove trigger without count:\n"
3549 " echo '!<function>:<trigger> > set_ftrace_filter\n"
3550 " To remove trigger with a count:\n"
3551 " echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
3552 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
3553 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3554 " modules: Can select a group via module command :mod:\n"
3555 " Does not accept triggers\n"
3556 #endif /* CONFIG_DYNAMIC_FTRACE */
3557 #ifdef CONFIG_FUNCTION_TRACER
3558 " set_ftrace_pid\t- Write pid(s) to only function trace those pids (function)\n"
3560 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3561 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
3562 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
3564 #ifdef CONFIG_TRACER_SNAPSHOT
3565 "\n snapshot\t\t- Like 'trace' but shows the content of the static snapshot buffer\n"
3566 "\t\t\t Read the contents for more information\n"
3568 #ifdef CONFIG_STACK_TRACER
3569 " stack_trace\t\t- Shows the max stack trace when active\n"
3570 " stack_max_size\t- Shows current max stack size that was traced\n"
3571 "\t\t\t Write into this file to reset the max size (trigger a new trace)\n"
3572 #ifdef CONFIG_DYNAMIC_FTRACE
3573 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace traces\n"
3575 #endif /* CONFIG_STACK_TRACER */
3579 tracing_readme_read(struct file *filp, char __user *ubuf,
3580 size_t cnt, loff_t *ppos)
3582 return simple_read_from_buffer(ubuf, cnt, ppos,
3583 readme_msg, strlen(readme_msg));
3586 static const struct file_operations tracing_readme_fops = {
3587 .open = tracing_open_generic,
3588 .read = tracing_readme_read,
3589 .llseek = generic_file_llseek,
3593 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
3594 size_t cnt, loff_t *ppos)
3603 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
3607 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
3615 for (i = 0; i < SAVED_CMDLINES; i++) {
3618 pid = map_cmdline_to_pid[i];
3619 if (pid == -1 || pid == NO_CMDLINE_MAP)
3622 trace_find_cmdline(pid, buf_comm);
3623 r = sprintf(buf, "%d %s\n", pid, buf_comm);
3628 len = simple_read_from_buffer(ubuf, cnt, ppos,
3637 static const struct file_operations tracing_saved_cmdlines_fops = {
3638 .open = tracing_open_generic,
3639 .read = tracing_saved_cmdlines_read,
3640 .llseek = generic_file_llseek,
3644 tracing_set_trace_read(struct file *filp, char __user *ubuf,
3645 size_t cnt, loff_t *ppos)
3647 struct trace_array *tr = filp->private_data;
3648 char buf[MAX_TRACER_SIZE+2];
3651 mutex_lock(&trace_types_lock);
3652 r = sprintf(buf, "%s\n", tr->current_trace->name);
3653 mutex_unlock(&trace_types_lock);
3655 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3658 int tracer_init(struct tracer *t, struct trace_array *tr)
3660 tracing_reset_online_cpus(&tr->trace_buffer);
3664 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val)
3668 for_each_tracing_cpu(cpu)
3669 per_cpu_ptr(buf->data, cpu)->entries = val;
3672 #ifdef CONFIG_TRACER_MAX_TRACE
3673 /* resize @tr's buffer to the size of @size_tr's entries */
3674 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
3675 struct trace_buffer *size_buf, int cpu_id)
3679 if (cpu_id == RING_BUFFER_ALL_CPUS) {
3680 for_each_tracing_cpu(cpu) {
3681 ret = ring_buffer_resize(trace_buf->buffer,
3682 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
3685 per_cpu_ptr(trace_buf->data, cpu)->entries =
3686 per_cpu_ptr(size_buf->data, cpu)->entries;
3689 ret = ring_buffer_resize(trace_buf->buffer,
3690 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
3692 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
3693 per_cpu_ptr(size_buf->data, cpu_id)->entries;
3698 #endif /* CONFIG_TRACER_MAX_TRACE */
3700 static int __tracing_resize_ring_buffer(struct trace_array *tr,
3701 unsigned long size, int cpu)
3706 * If kernel or user changes the size of the ring buffer
3707 * we use the size that was given, and we can forget about
3708 * expanding it later.
3710 ring_buffer_expanded = true;
3712 /* May be called before buffers are initialized */
3713 if (!tr->trace_buffer.buffer)
3716 ret = ring_buffer_resize(tr->trace_buffer.buffer, size, cpu);
3720 #ifdef CONFIG_TRACER_MAX_TRACE
3721 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
3722 !tr->current_trace->use_max_tr)
3725 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
3727 int r = resize_buffer_duplicate_size(&tr->trace_buffer,
3728 &tr->trace_buffer, cpu);
3731 * AARGH! We are left with different
3732 * size max buffer!!!!
3733 * The max buffer is our "snapshot" buffer.
3734 * When a tracer needs a snapshot (one of the
3735 * latency tracers), it swaps the max buffer
3736 * with the saved snap shot. We succeeded to
3737 * update the size of the main buffer, but failed to
3738 * update the size of the max buffer. But when we tried
3739 * to reset the main buffer to the original size, we
3740 * failed there too. This is very unlikely to
3741 * happen, but if it does, warn and kill all
3745 tracing_disabled = 1;
3750 if (cpu == RING_BUFFER_ALL_CPUS)
3751 set_buffer_entries(&tr->max_buffer, size);
3753 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
3756 #endif /* CONFIG_TRACER_MAX_TRACE */
3758 if (cpu == RING_BUFFER_ALL_CPUS)
3759 set_buffer_entries(&tr->trace_buffer, size);
3761 per_cpu_ptr(tr->trace_buffer.data, cpu)->entries = size;
3766 static ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
3767 unsigned long size, int cpu_id)
3771 mutex_lock(&trace_types_lock);
3773 if (cpu_id != RING_BUFFER_ALL_CPUS) {
3774 /* make sure, this cpu is enabled in the mask */
3775 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
3781 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
3786 mutex_unlock(&trace_types_lock);
3793 * tracing_update_buffers - used by tracing facility to expand ring buffers
3795 * To save on memory when the tracing is never used on a system with it
3796 * configured in. The ring buffers are set to a minimum size. But once
3797 * a user starts to use the tracing facility, then they need to grow
3798 * to their default size.
3800 * This function is to be called when a tracer is about to be used.
3802 int tracing_update_buffers(void)
3806 mutex_lock(&trace_types_lock);
3807 if (!ring_buffer_expanded)
3808 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
3809 RING_BUFFER_ALL_CPUS);
3810 mutex_unlock(&trace_types_lock);
3815 struct trace_option_dentry;
3817 static struct trace_option_dentry *
3818 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
3821 destroy_trace_option_files(struct trace_option_dentry *topts);
3823 static int tracing_set_tracer(const char *buf)
3825 static struct trace_option_dentry *topts;
3826 struct trace_array *tr = &global_trace;
3828 #ifdef CONFIG_TRACER_MAX_TRACE
3833 mutex_lock(&trace_types_lock);
3835 if (!ring_buffer_expanded) {
3836 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
3837 RING_BUFFER_ALL_CPUS);
3843 for (t = trace_types; t; t = t->next) {
3844 if (strcmp(t->name, buf) == 0)
3851 if (t == tr->current_trace)
3854 trace_branch_disable();
3856 tr->current_trace->enabled = false;
3858 if (tr->current_trace->reset)
3859 tr->current_trace->reset(tr);
3861 /* Current trace needs to be nop_trace before synchronize_sched */
3862 tr->current_trace = &nop_trace;
3864 #ifdef CONFIG_TRACER_MAX_TRACE
3865 had_max_tr = tr->allocated_snapshot;
3867 if (had_max_tr && !t->use_max_tr) {
3869 * We need to make sure that the update_max_tr sees that
3870 * current_trace changed to nop_trace to keep it from
3871 * swapping the buffers after we resize it.
3872 * The update_max_tr is called from interrupts disabled
3873 * so a synchronized_sched() is sufficient.
3875 synchronize_sched();
3879 destroy_trace_option_files(topts);
3881 topts = create_trace_option_files(tr, t);
3883 #ifdef CONFIG_TRACER_MAX_TRACE
3884 if (t->use_max_tr && !had_max_tr) {
3885 ret = alloc_snapshot(tr);
3892 ret = tracer_init(t, tr);
3897 tr->current_trace = t;
3898 tr->current_trace->enabled = true;
3899 trace_branch_enable(tr);
3901 mutex_unlock(&trace_types_lock);
3907 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3908 size_t cnt, loff_t *ppos)
3910 char buf[MAX_TRACER_SIZE+1];
3917 if (cnt > MAX_TRACER_SIZE)
3918 cnt = MAX_TRACER_SIZE;
3920 if (copy_from_user(&buf, ubuf, cnt))
3925 /* strip ending whitespace. */
3926 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3929 err = tracing_set_tracer(buf);
3939 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3940 size_t cnt, loff_t *ppos)
3942 unsigned long *ptr = filp->private_data;
3946 r = snprintf(buf, sizeof(buf), "%ld\n",
3947 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3948 if (r > sizeof(buf))
3950 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3954 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3955 size_t cnt, loff_t *ppos)
3957 unsigned long *ptr = filp->private_data;
3961 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3970 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3972 struct trace_array *tr = inode->i_private;
3973 struct trace_iterator *iter;
3976 if (tracing_disabled)
3979 if (trace_array_get(tr) < 0)
3982 mutex_lock(&trace_types_lock);
3984 /* create a buffer to store the information to pass to userspace */
3985 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3988 __trace_array_put(tr);
3993 * We make a copy of the current tracer to avoid concurrent
3994 * changes on it while we are reading.
3996 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
4001 *iter->trace = *tr->current_trace;
4003 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
4008 /* trace pipe does not show start of buffer */
4009 cpumask_setall(iter->started);
4011 if (trace_flags & TRACE_ITER_LATENCY_FMT)
4012 iter->iter_flags |= TRACE_FILE_LAT_FMT;
4014 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4015 if (trace_clocks[tr->clock_id].in_ns)
4016 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4019 iter->trace_buffer = &tr->trace_buffer;
4020 iter->cpu_file = tracing_get_cpu(inode);
4021 mutex_init(&iter->mutex);
4022 filp->private_data = iter;
4024 if (iter->trace->pipe_open)
4025 iter->trace->pipe_open(iter);
4027 nonseekable_open(inode, filp);
4029 mutex_unlock(&trace_types_lock);
4035 __trace_array_put(tr);
4036 mutex_unlock(&trace_types_lock);
4040 static int tracing_release_pipe(struct inode *inode, struct file *file)
4042 struct trace_iterator *iter = file->private_data;
4043 struct trace_array *tr = inode->i_private;
4045 mutex_lock(&trace_types_lock);
4047 if (iter->trace->pipe_close)
4048 iter->trace->pipe_close(iter);
4050 mutex_unlock(&trace_types_lock);
4052 free_cpumask_var(iter->started);
4053 mutex_destroy(&iter->mutex);
4057 trace_array_put(tr);
4063 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
4065 /* Iterators are static, they should be filled or empty */
4066 if (trace_buffer_iter(iter, iter->cpu_file))
4067 return POLLIN | POLLRDNORM;
4069 if (trace_flags & TRACE_ITER_BLOCK)
4071 * Always select as readable when in blocking mode
4073 return POLLIN | POLLRDNORM;
4075 return ring_buffer_poll_wait(iter->trace_buffer->buffer, iter->cpu_file,
4080 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
4082 struct trace_iterator *iter = filp->private_data;
4084 return trace_poll(iter, filp, poll_table);
4088 * This is a make-shift waitqueue.
4089 * A tracer might use this callback on some rare cases:
4091 * 1) the current tracer might hold the runqueue lock when it wakes up
4092 * a reader, hence a deadlock (sched, function, and function graph tracers)
4093 * 2) the function tracers, trace all functions, we don't want
4094 * the overhead of calling wake_up and friends
4095 * (and tracing them too)
4097 * Anyway, this is really very primitive wakeup.
4099 void poll_wait_pipe(struct trace_iterator *iter)
4101 set_current_state(TASK_INTERRUPTIBLE);
4102 /* sleep for 100 msecs, and try again. */
4103 schedule_timeout(HZ / 10);
4106 /* Must be called with trace_types_lock mutex held. */
4107 static int tracing_wait_pipe(struct file *filp)
4109 struct trace_iterator *iter = filp->private_data;
4111 while (trace_empty(iter)) {
4113 if ((filp->f_flags & O_NONBLOCK)) {
4117 mutex_unlock(&iter->mutex);
4119 iter->trace->wait_pipe(iter);
4121 mutex_lock(&iter->mutex);
4123 if (signal_pending(current))
4127 * We block until we read something and tracing is disabled.
4128 * We still block if tracing is disabled, but we have never
4129 * read anything. This allows a user to cat this file, and
4130 * then enable tracing. But after we have read something,
4131 * we give an EOF when tracing is again disabled.
4133 * iter->pos will be 0 if we haven't read anything.
4135 if (!tracing_is_on() && iter->pos)
4146 tracing_read_pipe(struct file *filp, char __user *ubuf,
4147 size_t cnt, loff_t *ppos)
4149 struct trace_iterator *iter = filp->private_data;
4150 struct trace_array *tr = iter->tr;
4153 /* return any leftover data */
4154 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4158 trace_seq_init(&iter->seq);
4160 /* copy the tracer to avoid using a global lock all around */
4161 mutex_lock(&trace_types_lock);
4162 if (unlikely(iter->trace->name != tr->current_trace->name))
4163 *iter->trace = *tr->current_trace;
4164 mutex_unlock(&trace_types_lock);
4167 * Avoid more than one consumer on a single file descriptor
4168 * This is just a matter of traces coherency, the ring buffer itself
4171 mutex_lock(&iter->mutex);
4172 if (iter->trace->read) {
4173 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
4179 sret = tracing_wait_pipe(filp);
4183 /* stop when tracing is finished */
4184 if (trace_empty(iter)) {
4189 if (cnt >= PAGE_SIZE)
4190 cnt = PAGE_SIZE - 1;
4192 /* reset all but tr, trace, and overruns */
4193 memset(&iter->seq, 0,
4194 sizeof(struct trace_iterator) -
4195 offsetof(struct trace_iterator, seq));
4196 cpumask_clear(iter->started);
4199 trace_event_read_lock();
4200 trace_access_lock(iter->cpu_file);
4201 while (trace_find_next_entry_inc(iter) != NULL) {
4202 enum print_line_t ret;
4203 int len = iter->seq.len;
4205 ret = print_trace_line(iter);
4206 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4207 /* don't print partial lines */
4208 iter->seq.len = len;
4211 if (ret != TRACE_TYPE_NO_CONSUME)
4212 trace_consume(iter);
4214 if (iter->seq.len >= cnt)
4218 * Setting the full flag means we reached the trace_seq buffer
4219 * size and we should leave by partial output condition above.
4220 * One of the trace_seq_* functions is not used properly.
4222 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
4225 trace_access_unlock(iter->cpu_file);
4226 trace_event_read_unlock();
4228 /* Now copy what we have to the user */
4229 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4230 if (iter->seq.readpos >= iter->seq.len)
4231 trace_seq_init(&iter->seq);
4234 * If there was nothing to send to user, in spite of consuming trace
4235 * entries, go back to wait for more entries.
4241 mutex_unlock(&iter->mutex);
4246 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
4249 __free_page(spd->pages[idx]);
4252 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
4254 .map = generic_pipe_buf_map,
4255 .unmap = generic_pipe_buf_unmap,
4256 .confirm = generic_pipe_buf_confirm,
4257 .release = generic_pipe_buf_release,
4258 .steal = generic_pipe_buf_steal,
4259 .get = generic_pipe_buf_get,
4263 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
4268 /* Seq buffer is page-sized, exactly what we need. */
4270 count = iter->seq.len;
4271 ret = print_trace_line(iter);
4272 count = iter->seq.len - count;
4275 iter->seq.len -= count;
4278 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4279 iter->seq.len -= count;
4283 if (ret != TRACE_TYPE_NO_CONSUME)
4284 trace_consume(iter);
4286 if (!trace_find_next_entry_inc(iter)) {
4296 static ssize_t tracing_splice_read_pipe(struct file *filp,
4298 struct pipe_inode_info *pipe,
4302 struct page *pages_def[PIPE_DEF_BUFFERS];
4303 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4304 struct trace_iterator *iter = filp->private_data;
4305 struct splice_pipe_desc spd = {
4307 .partial = partial_def,
4308 .nr_pages = 0, /* This gets updated below. */
4309 .nr_pages_max = PIPE_DEF_BUFFERS,
4311 .ops = &tracing_pipe_buf_ops,
4312 .spd_release = tracing_spd_release_pipe,
4314 struct trace_array *tr = iter->tr;
4319 if (splice_grow_spd(pipe, &spd))
4322 /* copy the tracer to avoid using a global lock all around */
4323 mutex_lock(&trace_types_lock);
4324 if (unlikely(iter->trace->name != tr->current_trace->name))
4325 *iter->trace = *tr->current_trace;
4326 mutex_unlock(&trace_types_lock);
4328 mutex_lock(&iter->mutex);
4330 if (iter->trace->splice_read) {
4331 ret = iter->trace->splice_read(iter, filp,
4332 ppos, pipe, len, flags);
4337 ret = tracing_wait_pipe(filp);
4341 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
4346 trace_event_read_lock();
4347 trace_access_lock(iter->cpu_file);
4349 /* Fill as many pages as possible. */
4350 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
4351 spd.pages[i] = alloc_page(GFP_KERNEL);
4355 rem = tracing_fill_pipe_page(rem, iter);
4357 /* Copy the data into the page, so we can start over. */
4358 ret = trace_seq_to_buffer(&iter->seq,
4359 page_address(spd.pages[i]),
4362 __free_page(spd.pages[i]);
4365 spd.partial[i].offset = 0;
4366 spd.partial[i].len = iter->seq.len;
4368 trace_seq_init(&iter->seq);
4371 trace_access_unlock(iter->cpu_file);
4372 trace_event_read_unlock();
4373 mutex_unlock(&iter->mutex);
4377 ret = splice_to_pipe(pipe, &spd);
4379 splice_shrink_spd(&spd);
4383 mutex_unlock(&iter->mutex);
4388 tracing_entries_read(struct file *filp, char __user *ubuf,
4389 size_t cnt, loff_t *ppos)
4391 struct inode *inode = file_inode(filp);
4392 struct trace_array *tr = inode->i_private;
4393 int cpu = tracing_get_cpu(inode);
4398 mutex_lock(&trace_types_lock);
4400 if (cpu == RING_BUFFER_ALL_CPUS) {
4401 int cpu, buf_size_same;
4406 /* check if all cpu sizes are same */
4407 for_each_tracing_cpu(cpu) {
4408 /* fill in the size from first enabled cpu */
4410 size = per_cpu_ptr(tr->trace_buffer.data, cpu)->entries;
4411 if (size != per_cpu_ptr(tr->trace_buffer.data, cpu)->entries) {
4417 if (buf_size_same) {
4418 if (!ring_buffer_expanded)
4419 r = sprintf(buf, "%lu (expanded: %lu)\n",
4421 trace_buf_size >> 10);
4423 r = sprintf(buf, "%lu\n", size >> 10);
4425 r = sprintf(buf, "X\n");
4427 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10);
4429 mutex_unlock(&trace_types_lock);
4431 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4436 tracing_entries_write(struct file *filp, const char __user *ubuf,
4437 size_t cnt, loff_t *ppos)
4439 struct inode *inode = file_inode(filp);
4440 struct trace_array *tr = inode->i_private;
4444 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4448 /* must have at least 1 entry */
4452 /* value is in KB */
4454 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
4464 tracing_total_entries_read(struct file *filp, char __user *ubuf,
4465 size_t cnt, loff_t *ppos)
4467 struct trace_array *tr = filp->private_data;
4470 unsigned long size = 0, expanded_size = 0;
4472 mutex_lock(&trace_types_lock);
4473 for_each_tracing_cpu(cpu) {
4474 size += per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10;
4475 if (!ring_buffer_expanded)
4476 expanded_size += trace_buf_size >> 10;
4478 if (ring_buffer_expanded)
4479 r = sprintf(buf, "%lu\n", size);
4481 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
4482 mutex_unlock(&trace_types_lock);
4484 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4488 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
4489 size_t cnt, loff_t *ppos)
4492 * There is no need to read what the user has written, this function
4493 * is just to make sure that there is no error when "echo" is used
4502 tracing_free_buffer_release(struct inode *inode, struct file *filp)
4504 struct trace_array *tr = inode->i_private;
4506 /* disable tracing ? */
4507 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
4508 tracer_tracing_off(tr);
4509 /* resize the ring buffer to 0 */
4510 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
4512 trace_array_put(tr);
4518 tracing_mark_write(struct file *filp, const char __user *ubuf,
4519 size_t cnt, loff_t *fpos)
4521 unsigned long addr = (unsigned long)ubuf;
4522 struct trace_array *tr = filp->private_data;
4523 struct ring_buffer_event *event;
4524 struct ring_buffer *buffer;
4525 struct print_entry *entry;
4526 unsigned long irq_flags;
4527 struct page *pages[2];
4537 if (tracing_disabled)
4540 if (!(trace_flags & TRACE_ITER_MARKERS))
4543 if (cnt > TRACE_BUF_SIZE)
4544 cnt = TRACE_BUF_SIZE;
4547 * Userspace is injecting traces into the kernel trace buffer.
4548 * We want to be as non intrusive as possible.
4549 * To do so, we do not want to allocate any special buffers
4550 * or take any locks, but instead write the userspace data
4551 * straight into the ring buffer.
4553 * First we need to pin the userspace buffer into memory,
4554 * which, most likely it is, because it just referenced it.
4555 * But there's no guarantee that it is. By using get_user_pages_fast()
4556 * and kmap_atomic/kunmap_atomic() we can get access to the
4557 * pages directly. We then write the data directly into the
4560 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
4562 /* check if we cross pages */
4563 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
4566 offset = addr & (PAGE_SIZE - 1);
4569 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
4570 if (ret < nr_pages) {
4572 put_page(pages[ret]);
4577 for (i = 0; i < nr_pages; i++)
4578 map_page[i] = kmap_atomic(pages[i]);
4580 local_save_flags(irq_flags);
4581 size = sizeof(*entry) + cnt + 2; /* possible \n added */
4582 buffer = tr->trace_buffer.buffer;
4583 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
4584 irq_flags, preempt_count());
4586 /* Ring buffer disabled, return as if not open for write */
4591 entry = ring_buffer_event_data(event);
4592 entry->ip = _THIS_IP_;
4594 if (nr_pages == 2) {
4595 len = PAGE_SIZE - offset;
4596 memcpy(&entry->buf, map_page[0] + offset, len);
4597 memcpy(&entry->buf[len], map_page[1], cnt - len);
4599 memcpy(&entry->buf, map_page[0] + offset, cnt);
4601 if (entry->buf[cnt - 1] != '\n') {
4602 entry->buf[cnt] = '\n';
4603 entry->buf[cnt + 1] = '\0';
4605 entry->buf[cnt] = '\0';
4607 __buffer_unlock_commit(buffer, event);
4614 for (i = 0; i < nr_pages; i++){
4615 kunmap_atomic(map_page[i]);
4622 static int tracing_clock_show(struct seq_file *m, void *v)
4624 struct trace_array *tr = m->private;
4627 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
4629 "%s%s%s%s", i ? " " : "",
4630 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
4631 i == tr->clock_id ? "]" : "");
4637 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
4638 size_t cnt, loff_t *fpos)
4640 struct seq_file *m = filp->private_data;
4641 struct trace_array *tr = m->private;
4643 const char *clockstr;
4646 if (cnt >= sizeof(buf))
4649 if (copy_from_user(&buf, ubuf, cnt))
4654 clockstr = strstrip(buf);
4656 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
4657 if (strcmp(trace_clocks[i].name, clockstr) == 0)
4660 if (i == ARRAY_SIZE(trace_clocks))
4663 mutex_lock(&trace_types_lock);
4667 ring_buffer_set_clock(tr->trace_buffer.buffer, trace_clocks[i].func);
4670 * New clock may not be consistent with the previous clock.
4671 * Reset the buffer so that it doesn't have incomparable timestamps.
4673 tracing_reset_online_cpus(&tr->trace_buffer);
4675 #ifdef CONFIG_TRACER_MAX_TRACE
4676 if (tr->flags & TRACE_ARRAY_FL_GLOBAL && tr->max_buffer.buffer)
4677 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
4678 tracing_reset_online_cpus(&tr->max_buffer);
4681 mutex_unlock(&trace_types_lock);
4688 static int tracing_clock_open(struct inode *inode, struct file *file)
4690 struct trace_array *tr = inode->i_private;
4693 if (tracing_disabled)
4696 if (trace_array_get(tr))
4699 ret = single_open(file, tracing_clock_show, inode->i_private);
4701 trace_array_put(tr);
4706 struct ftrace_buffer_info {
4707 struct trace_iterator iter;
4712 #ifdef CONFIG_TRACER_SNAPSHOT
4713 static int tracing_snapshot_open(struct inode *inode, struct file *file)
4715 struct trace_array *tr = inode->i_private;
4716 struct trace_iterator *iter;
4720 if (trace_array_get(tr) < 0)
4723 if (file->f_mode & FMODE_READ) {
4724 iter = __tracing_open(inode, file, true);
4726 ret = PTR_ERR(iter);
4728 /* Writes still need the seq_file to hold the private data */
4730 m = kzalloc(sizeof(*m), GFP_KERNEL);
4733 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4741 iter->trace_buffer = &tr->max_buffer;
4742 iter->cpu_file = tracing_get_cpu(inode);
4744 file->private_data = m;
4748 trace_array_put(tr);
4754 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
4757 struct seq_file *m = filp->private_data;
4758 struct trace_iterator *iter = m->private;
4759 struct trace_array *tr = iter->tr;
4763 ret = tracing_update_buffers();
4767 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4771 mutex_lock(&trace_types_lock);
4773 if (tr->current_trace->use_max_tr) {
4780 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4784 if (tr->allocated_snapshot)
4788 /* Only allow per-cpu swap if the ring buffer supports it */
4789 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
4790 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4795 if (!tr->allocated_snapshot) {
4796 ret = alloc_snapshot(tr);
4800 local_irq_disable();
4801 /* Now, we're going to swap */
4802 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4803 update_max_tr(tr, current, smp_processor_id());
4805 update_max_tr_single(tr, current, iter->cpu_file);
4809 if (tr->allocated_snapshot) {
4810 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4811 tracing_reset_online_cpus(&tr->max_buffer);
4813 tracing_reset(&tr->max_buffer, iter->cpu_file);
4823 mutex_unlock(&trace_types_lock);
4827 static int tracing_snapshot_release(struct inode *inode, struct file *file)
4829 struct seq_file *m = file->private_data;
4832 ret = tracing_release(inode, file);
4834 if (file->f_mode & FMODE_READ)
4837 /* If write only, the seq_file is just a stub */
4845 static int tracing_buffers_open(struct inode *inode, struct file *filp);
4846 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
4847 size_t count, loff_t *ppos);
4848 static int tracing_buffers_release(struct inode *inode, struct file *file);
4849 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4850 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
4852 static int snapshot_raw_open(struct inode *inode, struct file *filp)
4854 struct ftrace_buffer_info *info;
4857 ret = tracing_buffers_open(inode, filp);
4861 info = filp->private_data;
4863 if (info->iter.trace->use_max_tr) {
4864 tracing_buffers_release(inode, filp);
4868 info->iter.snapshot = true;
4869 info->iter.trace_buffer = &info->iter.tr->max_buffer;
4874 #endif /* CONFIG_TRACER_SNAPSHOT */
4877 static const struct file_operations tracing_max_lat_fops = {
4878 .open = tracing_open_generic,
4879 .read = tracing_max_lat_read,
4880 .write = tracing_max_lat_write,
4881 .llseek = generic_file_llseek,
4884 static const struct file_operations set_tracer_fops = {
4885 .open = tracing_open_generic,
4886 .read = tracing_set_trace_read,
4887 .write = tracing_set_trace_write,
4888 .llseek = generic_file_llseek,
4891 static const struct file_operations tracing_pipe_fops = {
4892 .open = tracing_open_pipe,
4893 .poll = tracing_poll_pipe,
4894 .read = tracing_read_pipe,
4895 .splice_read = tracing_splice_read_pipe,
4896 .release = tracing_release_pipe,
4897 .llseek = no_llseek,
4900 static const struct file_operations tracing_entries_fops = {
4901 .open = tracing_open_generic_tr,
4902 .read = tracing_entries_read,
4903 .write = tracing_entries_write,
4904 .llseek = generic_file_llseek,
4905 .release = tracing_release_generic_tr,
4908 static const struct file_operations tracing_total_entries_fops = {
4909 .open = tracing_open_generic_tr,
4910 .read = tracing_total_entries_read,
4911 .llseek = generic_file_llseek,
4912 .release = tracing_release_generic_tr,
4915 static const struct file_operations tracing_free_buffer_fops = {
4916 .open = tracing_open_generic_tr,
4917 .write = tracing_free_buffer_write,
4918 .release = tracing_free_buffer_release,
4921 static const struct file_operations tracing_mark_fops = {
4922 .open = tracing_open_generic_tr,
4923 .write = tracing_mark_write,
4924 .llseek = generic_file_llseek,
4925 .release = tracing_release_generic_tr,
4928 static const struct file_operations trace_clock_fops = {
4929 .open = tracing_clock_open,
4931 .llseek = seq_lseek,
4932 .release = tracing_single_release_tr,
4933 .write = tracing_clock_write,
4936 #ifdef CONFIG_TRACER_SNAPSHOT
4937 static const struct file_operations snapshot_fops = {
4938 .open = tracing_snapshot_open,
4940 .write = tracing_snapshot_write,
4941 .llseek = tracing_lseek,
4942 .release = tracing_snapshot_release,
4945 static const struct file_operations snapshot_raw_fops = {
4946 .open = snapshot_raw_open,
4947 .read = tracing_buffers_read,
4948 .release = tracing_buffers_release,
4949 .splice_read = tracing_buffers_splice_read,
4950 .llseek = no_llseek,
4953 #endif /* CONFIG_TRACER_SNAPSHOT */
4955 static int tracing_buffers_open(struct inode *inode, struct file *filp)
4957 struct trace_array *tr = inode->i_private;
4958 struct ftrace_buffer_info *info;
4961 if (tracing_disabled)
4964 if (trace_array_get(tr) < 0)
4967 info = kzalloc(sizeof(*info), GFP_KERNEL);
4969 trace_array_put(tr);
4973 mutex_lock(&trace_types_lock);
4976 info->iter.cpu_file = tracing_get_cpu(inode);
4977 info->iter.trace = tr->current_trace;
4978 info->iter.trace_buffer = &tr->trace_buffer;
4980 /* Force reading ring buffer for first read */
4981 info->read = (unsigned int)-1;
4983 filp->private_data = info;
4985 mutex_unlock(&trace_types_lock);
4987 ret = nonseekable_open(inode, filp);
4989 trace_array_put(tr);
4995 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
4997 struct ftrace_buffer_info *info = filp->private_data;
4998 struct trace_iterator *iter = &info->iter;
5000 return trace_poll(iter, filp, poll_table);
5004 tracing_buffers_read(struct file *filp, char __user *ubuf,
5005 size_t count, loff_t *ppos)
5007 struct ftrace_buffer_info *info = filp->private_data;
5008 struct trace_iterator *iter = &info->iter;
5015 mutex_lock(&trace_types_lock);
5017 #ifdef CONFIG_TRACER_MAX_TRACE
5018 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
5025 info->spare = ring_buffer_alloc_read_page(iter->trace_buffer->buffer,
5031 /* Do we have previous read data to read? */
5032 if (info->read < PAGE_SIZE)
5036 trace_access_lock(iter->cpu_file);
5037 ret = ring_buffer_read_page(iter->trace_buffer->buffer,
5041 trace_access_unlock(iter->cpu_file);
5044 if (trace_empty(iter)) {
5045 if ((filp->f_flags & O_NONBLOCK)) {
5049 mutex_unlock(&trace_types_lock);
5050 iter->trace->wait_pipe(iter);
5051 mutex_lock(&trace_types_lock);
5052 if (signal_pending(current)) {
5064 size = PAGE_SIZE - info->read;
5068 ret = copy_to_user(ubuf, info->spare + info->read, size);
5079 mutex_unlock(&trace_types_lock);
5084 static int tracing_buffers_release(struct inode *inode, struct file *file)
5086 struct ftrace_buffer_info *info = file->private_data;
5087 struct trace_iterator *iter = &info->iter;
5089 mutex_lock(&trace_types_lock);
5091 __trace_array_put(iter->tr);
5094 ring_buffer_free_read_page(iter->trace_buffer->buffer, info->spare);
5097 mutex_unlock(&trace_types_lock);
5103 struct ring_buffer *buffer;
5108 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
5109 struct pipe_buffer *buf)
5111 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5116 ring_buffer_free_read_page(ref->buffer, ref->page);
5121 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
5122 struct pipe_buffer *buf)
5124 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5129 /* Pipe buffer operations for a buffer. */
5130 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
5132 .map = generic_pipe_buf_map,
5133 .unmap = generic_pipe_buf_unmap,
5134 .confirm = generic_pipe_buf_confirm,
5135 .release = buffer_pipe_buf_release,
5136 .steal = generic_pipe_buf_steal,
5137 .get = buffer_pipe_buf_get,
5141 * Callback from splice_to_pipe(), if we need to release some pages
5142 * at the end of the spd in case we error'ed out in filling the pipe.
5144 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
5146 struct buffer_ref *ref =
5147 (struct buffer_ref *)spd->partial[i].private;
5152 ring_buffer_free_read_page(ref->buffer, ref->page);
5154 spd->partial[i].private = 0;
5158 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
5159 struct pipe_inode_info *pipe, size_t len,
5162 struct ftrace_buffer_info *info = file->private_data;
5163 struct trace_iterator *iter = &info->iter;
5164 struct partial_page partial_def[PIPE_DEF_BUFFERS];
5165 struct page *pages_def[PIPE_DEF_BUFFERS];
5166 struct splice_pipe_desc spd = {
5168 .partial = partial_def,
5169 .nr_pages_max = PIPE_DEF_BUFFERS,
5171 .ops = &buffer_pipe_buf_ops,
5172 .spd_release = buffer_spd_release,
5174 struct buffer_ref *ref;
5175 int entries, size, i;
5178 mutex_lock(&trace_types_lock);
5180 #ifdef CONFIG_TRACER_MAX_TRACE
5181 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
5187 if (splice_grow_spd(pipe, &spd)) {
5192 if (*ppos & (PAGE_SIZE - 1)) {
5197 if (len & (PAGE_SIZE - 1)) {
5198 if (len < PAGE_SIZE) {
5206 trace_access_lock(iter->cpu_file);
5207 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5209 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
5213 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
5218 ref->buffer = iter->trace_buffer->buffer;
5219 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
5225 r = ring_buffer_read_page(ref->buffer, &ref->page,
5226 len, iter->cpu_file, 1);
5228 ring_buffer_free_read_page(ref->buffer, ref->page);
5234 * zero out any left over data, this is going to
5237 size = ring_buffer_page_len(ref->page);
5238 if (size < PAGE_SIZE)
5239 memset(ref->page + size, 0, PAGE_SIZE - size);
5241 page = virt_to_page(ref->page);
5243 spd.pages[i] = page;
5244 spd.partial[i].len = PAGE_SIZE;
5245 spd.partial[i].offset = 0;
5246 spd.partial[i].private = (unsigned long)ref;
5250 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5253 trace_access_unlock(iter->cpu_file);
5256 /* did we read anything? */
5257 if (!spd.nr_pages) {
5258 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK)) {
5262 mutex_unlock(&trace_types_lock);
5263 iter->trace->wait_pipe(iter);
5264 mutex_lock(&trace_types_lock);
5265 if (signal_pending(current)) {
5272 ret = splice_to_pipe(pipe, &spd);
5273 splice_shrink_spd(&spd);
5275 mutex_unlock(&trace_types_lock);
5280 static const struct file_operations tracing_buffers_fops = {
5281 .open = tracing_buffers_open,
5282 .read = tracing_buffers_read,
5283 .poll = tracing_buffers_poll,
5284 .release = tracing_buffers_release,
5285 .splice_read = tracing_buffers_splice_read,
5286 .llseek = no_llseek,
5290 tracing_stats_read(struct file *filp, char __user *ubuf,
5291 size_t count, loff_t *ppos)
5293 struct inode *inode = file_inode(filp);
5294 struct trace_array *tr = inode->i_private;
5295 struct trace_buffer *trace_buf = &tr->trace_buffer;
5296 int cpu = tracing_get_cpu(inode);
5297 struct trace_seq *s;
5299 unsigned long long t;
5300 unsigned long usec_rem;
5302 s = kmalloc(sizeof(*s), GFP_KERNEL);
5308 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
5309 trace_seq_printf(s, "entries: %ld\n", cnt);
5311 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
5312 trace_seq_printf(s, "overrun: %ld\n", cnt);
5314 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
5315 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
5317 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
5318 trace_seq_printf(s, "bytes: %ld\n", cnt);
5320 if (trace_clocks[tr->clock_id].in_ns) {
5321 /* local or global for trace_clock */
5322 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5323 usec_rem = do_div(t, USEC_PER_SEC);
5324 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
5327 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
5328 usec_rem = do_div(t, USEC_PER_SEC);
5329 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
5331 /* counter or tsc mode for trace_clock */
5332 trace_seq_printf(s, "oldest event ts: %llu\n",
5333 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5335 trace_seq_printf(s, "now ts: %llu\n",
5336 ring_buffer_time_stamp(trace_buf->buffer, cpu));
5339 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
5340 trace_seq_printf(s, "dropped events: %ld\n", cnt);
5342 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
5343 trace_seq_printf(s, "read events: %ld\n", cnt);
5345 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
5352 static const struct file_operations tracing_stats_fops = {
5353 .open = tracing_open_generic_tr,
5354 .read = tracing_stats_read,
5355 .llseek = generic_file_llseek,
5356 .release = tracing_release_generic_tr,
5359 #ifdef CONFIG_DYNAMIC_FTRACE
5361 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
5367 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
5368 size_t cnt, loff_t *ppos)
5370 static char ftrace_dyn_info_buffer[1024];
5371 static DEFINE_MUTEX(dyn_info_mutex);
5372 unsigned long *p = filp->private_data;
5373 char *buf = ftrace_dyn_info_buffer;
5374 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
5377 mutex_lock(&dyn_info_mutex);
5378 r = sprintf(buf, "%ld ", *p);
5380 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
5383 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5385 mutex_unlock(&dyn_info_mutex);
5390 static const struct file_operations tracing_dyn_info_fops = {
5391 .open = tracing_open_generic,
5392 .read = tracing_read_dyn_info,
5393 .llseek = generic_file_llseek,
5395 #endif /* CONFIG_DYNAMIC_FTRACE */
5397 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
5399 ftrace_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5405 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5407 unsigned long *count = (long *)data;
5419 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
5420 struct ftrace_probe_ops *ops, void *data)
5422 long count = (long)data;
5424 seq_printf(m, "%ps:", (void *)ip);
5426 seq_printf(m, "snapshot");
5429 seq_printf(m, ":unlimited\n");
5431 seq_printf(m, ":count=%ld\n", count);
5436 static struct ftrace_probe_ops snapshot_probe_ops = {
5437 .func = ftrace_snapshot,
5438 .print = ftrace_snapshot_print,
5441 static struct ftrace_probe_ops snapshot_count_probe_ops = {
5442 .func = ftrace_count_snapshot,
5443 .print = ftrace_snapshot_print,
5447 ftrace_trace_snapshot_callback(struct ftrace_hash *hash,
5448 char *glob, char *cmd, char *param, int enable)
5450 struct ftrace_probe_ops *ops;
5451 void *count = (void *)-1;
5455 /* hash funcs only work with set_ftrace_filter */
5459 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
5461 if (glob[0] == '!') {
5462 unregister_ftrace_function_probe_func(glob+1, ops);
5469 number = strsep(¶m, ":");
5471 if (!strlen(number))
5475 * We use the callback data field (which is a pointer)
5478 ret = kstrtoul(number, 0, (unsigned long *)&count);
5483 ret = register_ftrace_function_probe(glob, ops, count);
5486 alloc_snapshot(&global_trace);
5488 return ret < 0 ? ret : 0;
5491 static struct ftrace_func_command ftrace_snapshot_cmd = {
5493 .func = ftrace_trace_snapshot_callback,
5496 static __init int register_snapshot_cmd(void)
5498 return register_ftrace_command(&ftrace_snapshot_cmd);
5501 static inline __init int register_snapshot_cmd(void) { return 0; }
5502 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
5504 struct dentry *tracing_init_dentry_tr(struct trace_array *tr)
5509 if (!debugfs_initialized())
5512 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
5513 tr->dir = debugfs_create_dir("tracing", NULL);
5516 pr_warn_once("Could not create debugfs directory 'tracing'\n");
5521 struct dentry *tracing_init_dentry(void)
5523 return tracing_init_dentry_tr(&global_trace);
5526 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
5528 struct dentry *d_tracer;
5531 return tr->percpu_dir;
5533 d_tracer = tracing_init_dentry_tr(tr);
5537 tr->percpu_dir = debugfs_create_dir("per_cpu", d_tracer);
5539 WARN_ONCE(!tr->percpu_dir,
5540 "Could not create debugfs directory 'per_cpu/%d'\n", cpu);
5542 return tr->percpu_dir;
5545 static struct dentry *
5546 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
5547 void *data, long cpu, const struct file_operations *fops)
5549 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
5551 if (ret) /* See tracing_get_cpu() */
5552 ret->d_inode->i_cdev = (void *)(cpu + 1);
5557 tracing_init_debugfs_percpu(struct trace_array *tr, long cpu)
5559 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
5560 struct dentry *d_cpu;
5561 char cpu_dir[30]; /* 30 characters should be more than enough */
5566 snprintf(cpu_dir, 30, "cpu%ld", cpu);
5567 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
5569 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
5573 /* per cpu trace_pipe */
5574 trace_create_cpu_file("trace_pipe", 0444, d_cpu,
5575 tr, cpu, &tracing_pipe_fops);
5578 trace_create_cpu_file("trace", 0644, d_cpu,
5579 tr, cpu, &tracing_fops);
5581 trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu,
5582 tr, cpu, &tracing_buffers_fops);
5584 trace_create_cpu_file("stats", 0444, d_cpu,
5585 tr, cpu, &tracing_stats_fops);
5587 trace_create_cpu_file("buffer_size_kb", 0444, d_cpu,
5588 tr, cpu, &tracing_entries_fops);
5590 #ifdef CONFIG_TRACER_SNAPSHOT
5591 trace_create_cpu_file("snapshot", 0644, d_cpu,
5592 tr, cpu, &snapshot_fops);
5594 trace_create_cpu_file("snapshot_raw", 0444, d_cpu,
5595 tr, cpu, &snapshot_raw_fops);
5599 #ifdef CONFIG_FTRACE_SELFTEST
5600 /* Let selftest have access to static functions in this file */
5601 #include "trace_selftest.c"
5604 struct trace_option_dentry {
5605 struct tracer_opt *opt;
5606 struct tracer_flags *flags;
5607 struct trace_array *tr;
5608 struct dentry *entry;
5612 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
5615 struct trace_option_dentry *topt = filp->private_data;
5618 if (topt->flags->val & topt->opt->bit)
5623 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5627 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
5630 struct trace_option_dentry *topt = filp->private_data;
5634 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5638 if (val != 0 && val != 1)
5641 if (!!(topt->flags->val & topt->opt->bit) != val) {
5642 mutex_lock(&trace_types_lock);
5643 ret = __set_tracer_option(topt->tr->current_trace, topt->flags,
5645 mutex_unlock(&trace_types_lock);
5656 static const struct file_operations trace_options_fops = {
5657 .open = tracing_open_generic,
5658 .read = trace_options_read,
5659 .write = trace_options_write,
5660 .llseek = generic_file_llseek,
5664 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
5667 long index = (long)filp->private_data;
5670 if (trace_flags & (1 << index))
5675 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5679 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
5682 struct trace_array *tr = &global_trace;
5683 long index = (long)filp->private_data;
5687 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5691 if (val != 0 && val != 1)
5694 mutex_lock(&trace_types_lock);
5695 ret = set_tracer_flag(tr, 1 << index, val);
5696 mutex_unlock(&trace_types_lock);
5706 static const struct file_operations trace_options_core_fops = {
5707 .open = tracing_open_generic,
5708 .read = trace_options_core_read,
5709 .write = trace_options_core_write,
5710 .llseek = generic_file_llseek,
5713 struct dentry *trace_create_file(const char *name,
5715 struct dentry *parent,
5717 const struct file_operations *fops)
5721 ret = debugfs_create_file(name, mode, parent, data, fops);
5723 pr_warning("Could not create debugfs '%s' entry\n", name);
5729 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
5731 struct dentry *d_tracer;
5736 d_tracer = tracing_init_dentry_tr(tr);
5740 tr->options = debugfs_create_dir("options", d_tracer);
5742 pr_warning("Could not create debugfs directory 'options'\n");
5750 create_trace_option_file(struct trace_array *tr,
5751 struct trace_option_dentry *topt,
5752 struct tracer_flags *flags,
5753 struct tracer_opt *opt)
5755 struct dentry *t_options;
5757 t_options = trace_options_init_dentry(tr);
5761 topt->flags = flags;
5765 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
5766 &trace_options_fops);
5770 static struct trace_option_dentry *
5771 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
5773 struct trace_option_dentry *topts;
5774 struct tracer_flags *flags;
5775 struct tracer_opt *opts;
5781 flags = tracer->flags;
5783 if (!flags || !flags->opts)
5788 for (cnt = 0; opts[cnt].name; cnt++)
5791 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
5795 for (cnt = 0; opts[cnt].name; cnt++)
5796 create_trace_option_file(tr, &topts[cnt], flags,
5803 destroy_trace_option_files(struct trace_option_dentry *topts)
5810 for (cnt = 0; topts[cnt].opt; cnt++) {
5811 if (topts[cnt].entry)
5812 debugfs_remove(topts[cnt].entry);
5818 static struct dentry *
5819 create_trace_option_core_file(struct trace_array *tr,
5820 const char *option, long index)
5822 struct dentry *t_options;
5824 t_options = trace_options_init_dentry(tr);
5828 return trace_create_file(option, 0644, t_options, (void *)index,
5829 &trace_options_core_fops);
5832 static __init void create_trace_options_dir(struct trace_array *tr)
5834 struct dentry *t_options;
5837 t_options = trace_options_init_dentry(tr);
5841 for (i = 0; trace_options[i]; i++)
5842 create_trace_option_core_file(tr, trace_options[i], i);
5846 rb_simple_read(struct file *filp, char __user *ubuf,
5847 size_t cnt, loff_t *ppos)
5849 struct trace_array *tr = filp->private_data;
5853 r = tracer_tracing_is_on(tr);
5854 r = sprintf(buf, "%d\n", r);
5856 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5860 rb_simple_write(struct file *filp, const char __user *ubuf,
5861 size_t cnt, loff_t *ppos)
5863 struct trace_array *tr = filp->private_data;
5864 struct ring_buffer *buffer = tr->trace_buffer.buffer;
5868 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5873 mutex_lock(&trace_types_lock);
5875 tracer_tracing_on(tr);
5876 if (tr->current_trace->start)
5877 tr->current_trace->start(tr);
5879 tracer_tracing_off(tr);
5880 if (tr->current_trace->stop)
5881 tr->current_trace->stop(tr);
5883 mutex_unlock(&trace_types_lock);
5891 static const struct file_operations rb_simple_fops = {
5892 .open = tracing_open_generic_tr,
5893 .read = rb_simple_read,
5894 .write = rb_simple_write,
5895 .release = tracing_release_generic_tr,
5896 .llseek = default_llseek,
5899 struct dentry *trace_instance_dir;
5902 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer);
5905 allocate_trace_buffer(struct trace_array *tr, struct trace_buffer *buf, int size)
5907 enum ring_buffer_flags rb_flags;
5909 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
5913 buf->buffer = ring_buffer_alloc(size, rb_flags);
5917 buf->data = alloc_percpu(struct trace_array_cpu);
5919 ring_buffer_free(buf->buffer);
5923 /* Allocate the first page for all buffers */
5924 set_buffer_entries(&tr->trace_buffer,
5925 ring_buffer_size(tr->trace_buffer.buffer, 0));
5930 static int allocate_trace_buffers(struct trace_array *tr, int size)
5934 ret = allocate_trace_buffer(tr, &tr->trace_buffer, size);
5938 #ifdef CONFIG_TRACER_MAX_TRACE
5939 ret = allocate_trace_buffer(tr, &tr->max_buffer,
5940 allocate_snapshot ? size : 1);
5942 ring_buffer_free(tr->trace_buffer.buffer);
5943 free_percpu(tr->trace_buffer.data);
5946 tr->allocated_snapshot = allocate_snapshot;
5949 * Only the top level trace array gets its snapshot allocated
5950 * from the kernel command line.
5952 allocate_snapshot = false;
5957 static int new_instance_create(const char *name)
5959 struct trace_array *tr;
5962 mutex_lock(&trace_types_lock);
5965 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5966 if (tr->name && strcmp(tr->name, name) == 0)
5971 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
5975 tr->name = kstrdup(name, GFP_KERNEL);
5979 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
5982 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
5984 raw_spin_lock_init(&tr->start_lock);
5986 tr->current_trace = &nop_trace;
5988 INIT_LIST_HEAD(&tr->systems);
5989 INIT_LIST_HEAD(&tr->events);
5991 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
5994 tr->dir = debugfs_create_dir(name, trace_instance_dir);
5998 ret = event_trace_add_tracer(tr->dir, tr);
6000 debugfs_remove_recursive(tr->dir);
6004 init_tracer_debugfs(tr, tr->dir);
6006 list_add(&tr->list, &ftrace_trace_arrays);
6008 mutex_unlock(&trace_types_lock);
6013 if (tr->trace_buffer.buffer)
6014 ring_buffer_free(tr->trace_buffer.buffer);
6015 free_cpumask_var(tr->tracing_cpumask);
6020 mutex_unlock(&trace_types_lock);
6026 static int instance_delete(const char *name)
6028 struct trace_array *tr;
6032 mutex_lock(&trace_types_lock);
6035 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
6036 if (tr->name && strcmp(tr->name, name) == 0) {
6048 list_del(&tr->list);
6050 event_trace_del_tracer(tr);
6051 debugfs_remove_recursive(tr->dir);
6052 free_percpu(tr->trace_buffer.data);
6053 ring_buffer_free(tr->trace_buffer.buffer);
6061 mutex_unlock(&trace_types_lock);
6066 static int instance_mkdir (struct inode *inode, struct dentry *dentry, umode_t mode)
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))
6077 * The inode mutex is locked, but debugfs_create_dir() will also
6078 * take the mutex. As the instances directory can not be destroyed
6079 * or changed in any other way, it is safe to unlock it, and
6080 * let the dentry try. If two users try to make the same dir at
6081 * the same time, then the new_instance_create() will determine the
6084 mutex_unlock(&inode->i_mutex);
6086 ret = new_instance_create(dentry->d_iname);
6088 mutex_lock(&inode->i_mutex);
6093 static int instance_rmdir(struct inode *inode, struct dentry *dentry)
6095 struct dentry *parent;
6098 /* Paranoid: Make sure the parent is the "instances" directory */
6099 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
6100 if (WARN_ON_ONCE(parent != trace_instance_dir))
6103 /* The caller did a dget() on dentry */
6104 mutex_unlock(&dentry->d_inode->i_mutex);
6107 * The inode mutex is locked, but debugfs_create_dir() will also
6108 * take the mutex. As the instances directory can not be destroyed
6109 * or changed in any other way, it is safe to unlock it, and
6110 * let the dentry try. If two users try to make the same dir at
6111 * the same time, then the instance_delete() will determine the
6114 mutex_unlock(&inode->i_mutex);
6116 ret = instance_delete(dentry->d_iname);
6118 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
6119 mutex_lock(&dentry->d_inode->i_mutex);
6124 static const struct inode_operations instance_dir_inode_operations = {
6125 .lookup = simple_lookup,
6126 .mkdir = instance_mkdir,
6127 .rmdir = instance_rmdir,
6130 static __init void create_trace_instances(struct dentry *d_tracer)
6132 trace_instance_dir = debugfs_create_dir("instances", d_tracer);
6133 if (WARN_ON(!trace_instance_dir))
6136 /* Hijack the dir inode operations, to allow mkdir */
6137 trace_instance_dir->d_inode->i_op = &instance_dir_inode_operations;
6141 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer)
6145 trace_create_file("tracing_cpumask", 0644, d_tracer,
6146 tr, &tracing_cpumask_fops);
6148 trace_create_file("trace_options", 0644, d_tracer,
6149 tr, &tracing_iter_fops);
6151 trace_create_file("trace", 0644, d_tracer,
6154 trace_create_file("trace_pipe", 0444, d_tracer,
6155 tr, &tracing_pipe_fops);
6157 trace_create_file("buffer_size_kb", 0644, d_tracer,
6158 tr, &tracing_entries_fops);
6160 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
6161 tr, &tracing_total_entries_fops);
6163 trace_create_file("free_buffer", 0200, d_tracer,
6164 tr, &tracing_free_buffer_fops);
6166 trace_create_file("trace_marker", 0220, d_tracer,
6167 tr, &tracing_mark_fops);
6169 trace_create_file("trace_clock", 0644, d_tracer, tr,
6172 trace_create_file("tracing_on", 0644, d_tracer,
6173 tr, &rb_simple_fops);
6175 #ifdef CONFIG_TRACER_SNAPSHOT
6176 trace_create_file("snapshot", 0644, d_tracer,
6177 tr, &snapshot_fops);
6180 for_each_tracing_cpu(cpu)
6181 tracing_init_debugfs_percpu(tr, cpu);
6185 static __init int tracer_init_debugfs(void)
6187 struct dentry *d_tracer;
6189 trace_access_lock_init();
6191 d_tracer = tracing_init_dentry();
6195 init_tracer_debugfs(&global_trace, d_tracer);
6197 trace_create_file("available_tracers", 0444, d_tracer,
6198 &global_trace, &show_traces_fops);
6200 trace_create_file("current_tracer", 0644, d_tracer,
6201 &global_trace, &set_tracer_fops);
6203 #ifdef CONFIG_TRACER_MAX_TRACE
6204 trace_create_file("tracing_max_latency", 0644, d_tracer,
6205 &tracing_max_latency, &tracing_max_lat_fops);
6208 trace_create_file("tracing_thresh", 0644, d_tracer,
6209 &tracing_thresh, &tracing_max_lat_fops);
6211 trace_create_file("README", 0444, d_tracer,
6212 NULL, &tracing_readme_fops);
6214 trace_create_file("saved_cmdlines", 0444, d_tracer,
6215 NULL, &tracing_saved_cmdlines_fops);
6217 #ifdef CONFIG_DYNAMIC_FTRACE
6218 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
6219 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
6222 create_trace_instances(d_tracer);
6224 create_trace_options_dir(&global_trace);
6229 static int trace_panic_handler(struct notifier_block *this,
6230 unsigned long event, void *unused)
6232 if (ftrace_dump_on_oops)
6233 ftrace_dump(ftrace_dump_on_oops);
6237 static struct notifier_block trace_panic_notifier = {
6238 .notifier_call = trace_panic_handler,
6240 .priority = 150 /* priority: INT_MAX >= x >= 0 */
6243 static int trace_die_handler(struct notifier_block *self,
6249 if (ftrace_dump_on_oops)
6250 ftrace_dump(ftrace_dump_on_oops);
6258 static struct notifier_block trace_die_notifier = {
6259 .notifier_call = trace_die_handler,
6264 * printk is set to max of 1024, we really don't need it that big.
6265 * Nothing should be printing 1000 characters anyway.
6267 #define TRACE_MAX_PRINT 1000
6270 * Define here KERN_TRACE so that we have one place to modify
6271 * it if we decide to change what log level the ftrace dump
6274 #define KERN_TRACE KERN_EMERG
6277 trace_printk_seq(struct trace_seq *s)
6279 /* Probably should print a warning here. */
6280 if (s->len >= TRACE_MAX_PRINT)
6281 s->len = TRACE_MAX_PRINT;
6283 /* should be zero ended, but we are paranoid. */
6284 s->buffer[s->len] = 0;
6286 printk(KERN_TRACE "%s", s->buffer);
6291 void trace_init_global_iter(struct trace_iterator *iter)
6293 iter->tr = &global_trace;
6294 iter->trace = iter->tr->current_trace;
6295 iter->cpu_file = RING_BUFFER_ALL_CPUS;
6296 iter->trace_buffer = &global_trace.trace_buffer;
6298 if (iter->trace && iter->trace->open)
6299 iter->trace->open(iter);
6301 /* Annotate start of buffers if we had overruns */
6302 if (ring_buffer_overruns(iter->trace_buffer->buffer))
6303 iter->iter_flags |= TRACE_FILE_ANNOTATE;
6305 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6306 if (trace_clocks[iter->tr->clock_id].in_ns)
6307 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6310 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
6312 /* use static because iter can be a bit big for the stack */
6313 static struct trace_iterator iter;
6314 static atomic_t dump_running;
6315 unsigned int old_userobj;
6316 unsigned long flags;
6319 /* Only allow one dump user at a time. */
6320 if (atomic_inc_return(&dump_running) != 1) {
6321 atomic_dec(&dump_running);
6326 * Always turn off tracing when we dump.
6327 * We don't need to show trace output of what happens
6328 * between multiple crashes.
6330 * If the user does a sysrq-z, then they can re-enable
6331 * tracing with echo 1 > tracing_on.
6335 local_irq_save(flags);
6337 /* Simulate the iterator */
6338 trace_init_global_iter(&iter);
6340 for_each_tracing_cpu(cpu) {
6341 atomic_inc(&per_cpu_ptr(iter.tr->trace_buffer.data, cpu)->disabled);
6344 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
6346 /* don't look at user memory in panic mode */
6347 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
6349 switch (oops_dump_mode) {
6351 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6354 iter.cpu_file = raw_smp_processor_id();
6359 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
6360 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6363 printk(KERN_TRACE "Dumping ftrace buffer:\n");
6365 /* Did function tracer already get disabled? */
6366 if (ftrace_is_dead()) {
6367 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
6368 printk("# MAY BE MISSING FUNCTION EVENTS\n");
6372 * We need to stop all tracing on all CPUS to read the
6373 * the next buffer. This is a bit expensive, but is
6374 * not done often. We fill all what we can read,
6375 * and then release the locks again.
6378 while (!trace_empty(&iter)) {
6381 printk(KERN_TRACE "---------------------------------\n");
6385 /* reset all but tr, trace, and overruns */
6386 memset(&iter.seq, 0,
6387 sizeof(struct trace_iterator) -
6388 offsetof(struct trace_iterator, seq));
6389 iter.iter_flags |= TRACE_FILE_LAT_FMT;
6392 if (trace_find_next_entry_inc(&iter) != NULL) {
6395 ret = print_trace_line(&iter);
6396 if (ret != TRACE_TYPE_NO_CONSUME)
6397 trace_consume(&iter);
6399 touch_nmi_watchdog();
6401 trace_printk_seq(&iter.seq);
6405 printk(KERN_TRACE " (ftrace buffer empty)\n");
6407 printk(KERN_TRACE "---------------------------------\n");
6410 trace_flags |= old_userobj;
6412 for_each_tracing_cpu(cpu) {
6413 atomic_dec(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);
6415 atomic_dec(&dump_running);
6416 local_irq_restore(flags);
6418 EXPORT_SYMBOL_GPL(ftrace_dump);
6420 __init static int tracer_alloc_buffers(void)
6426 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
6429 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
6430 goto out_free_buffer_mask;
6432 /* Only allocate trace_printk buffers if a trace_printk exists */
6433 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
6434 /* Must be called before global_trace.buffer is allocated */
6435 trace_printk_init_buffers();
6437 /* To save memory, keep the ring buffer size to its minimum */
6438 if (ring_buffer_expanded)
6439 ring_buf_size = trace_buf_size;
6443 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
6444 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
6446 raw_spin_lock_init(&global_trace.start_lock);
6448 /* TODO: make the number of buffers hot pluggable with CPUS */
6449 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
6450 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
6452 goto out_free_cpumask;
6455 if (global_trace.buffer_disabled)
6458 trace_init_cmdlines();
6461 * register_tracer() might reference current_trace, so it
6462 * needs to be set before we register anything. This is
6463 * just a bootstrap of current_trace anyway.
6465 global_trace.current_trace = &nop_trace;
6467 register_tracer(&nop_trace);
6469 /* All seems OK, enable tracing */
6470 tracing_disabled = 0;
6472 atomic_notifier_chain_register(&panic_notifier_list,
6473 &trace_panic_notifier);
6475 register_die_notifier(&trace_die_notifier);
6477 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
6479 INIT_LIST_HEAD(&global_trace.systems);
6480 INIT_LIST_HEAD(&global_trace.events);
6481 list_add(&global_trace.list, &ftrace_trace_arrays);
6483 while (trace_boot_options) {
6486 option = strsep(&trace_boot_options, ",");
6487 trace_set_options(&global_trace, option);
6490 register_snapshot_cmd();
6495 free_percpu(global_trace.trace_buffer.data);
6496 #ifdef CONFIG_TRACER_MAX_TRACE
6497 free_percpu(global_trace.max_buffer.data);
6499 free_cpumask_var(global_trace.tracing_cpumask);
6500 out_free_buffer_mask:
6501 free_cpumask_var(tracing_buffer_mask);
6506 __init static int clear_boot_tracer(void)
6509 * The default tracer at boot buffer is an init section.
6510 * This function is called in lateinit. If we did not
6511 * find the boot tracer, then clear it out, to prevent
6512 * later registration from accessing the buffer that is
6513 * about to be freed.
6515 if (!default_bootup_tracer)
6518 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
6519 default_bootup_tracer);
6520 default_bootup_tracer = NULL;
6525 early_initcall(tracer_alloc_buffers);
6526 fs_initcall(tracer_init_debugfs);
6527 late_initcall(clear_boot_tracer);