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_current_check_discard(struct ring_buffer *buffer,
239 struct ftrace_event_call *call, void *rec,
240 struct ring_buffer_event *event)
242 return filter_check_discard(call, rec, buffer, event);
244 EXPORT_SYMBOL_GPL(filter_current_check_discard);
246 cycle_t ftrace_now(int cpu)
250 /* Early boot up does not have a buffer yet */
251 if (!global_trace.trace_buffer.buffer)
252 return trace_clock_local();
254 ts = ring_buffer_time_stamp(global_trace.trace_buffer.buffer, cpu);
255 ring_buffer_normalize_time_stamp(global_trace.trace_buffer.buffer, cpu, &ts);
261 * tracing_is_enabled - Show if global_trace has been disabled
263 * Shows if the global trace has been enabled or not. It uses the
264 * mirror flag "buffer_disabled" to be used in fast paths such as for
265 * the irqsoff tracer. But it may be inaccurate due to races. If you
266 * need to know the accurate state, use tracing_is_on() which is a little
267 * slower, but accurate.
269 int tracing_is_enabled(void)
272 * For quick access (irqsoff uses this in fast path), just
273 * return the mirror variable of the state of the ring buffer.
274 * It's a little racy, but we don't really care.
277 return !global_trace.buffer_disabled;
281 * trace_buf_size is the size in bytes that is allocated
282 * for a buffer. Note, the number of bytes is always rounded
285 * This number is purposely set to a low number of 16384.
286 * If the dump on oops happens, it will be much appreciated
287 * to not have to wait for all that output. Anyway this can be
288 * boot time and run time configurable.
290 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
292 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
294 /* trace_types holds a link list of available tracers. */
295 static struct tracer *trace_types __read_mostly;
298 * trace_types_lock is used to protect the trace_types list.
300 DEFINE_MUTEX(trace_types_lock);
303 * serialize the access of the ring buffer
305 * ring buffer serializes readers, but it is low level protection.
306 * The validity of the events (which returns by ring_buffer_peek() ..etc)
307 * are not protected by ring buffer.
309 * The content of events may become garbage if we allow other process consumes
310 * these events concurrently:
311 * A) the page of the consumed events may become a normal page
312 * (not reader page) in ring buffer, and this page will be rewrited
313 * by events producer.
314 * B) The page of the consumed events may become a page for splice_read,
315 * and this page will be returned to system.
317 * These primitives allow multi process access to different cpu ring buffer
320 * These primitives don't distinguish read-only and read-consume access.
321 * Multi read-only access are also serialized.
325 static DECLARE_RWSEM(all_cpu_access_lock);
326 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
328 static inline void trace_access_lock(int cpu)
330 if (cpu == RING_BUFFER_ALL_CPUS) {
331 /* gain it for accessing the whole ring buffer. */
332 down_write(&all_cpu_access_lock);
334 /* gain it for accessing a cpu ring buffer. */
336 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
337 down_read(&all_cpu_access_lock);
339 /* Secondly block other access to this @cpu ring buffer. */
340 mutex_lock(&per_cpu(cpu_access_lock, cpu));
344 static inline void trace_access_unlock(int cpu)
346 if (cpu == RING_BUFFER_ALL_CPUS) {
347 up_write(&all_cpu_access_lock);
349 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
350 up_read(&all_cpu_access_lock);
354 static inline void trace_access_lock_init(void)
358 for_each_possible_cpu(cpu)
359 mutex_init(&per_cpu(cpu_access_lock, cpu));
364 static DEFINE_MUTEX(access_lock);
366 static inline void trace_access_lock(int cpu)
369 mutex_lock(&access_lock);
372 static inline void trace_access_unlock(int cpu)
375 mutex_unlock(&access_lock);
378 static inline void trace_access_lock_init(void)
384 /* trace_flags holds trace_options default values */
385 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
386 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
387 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE |
388 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | TRACE_ITER_FUNCTION;
390 static void tracer_tracing_on(struct trace_array *tr)
392 if (tr->trace_buffer.buffer)
393 ring_buffer_record_on(tr->trace_buffer.buffer);
395 * This flag is looked at when buffers haven't been allocated
396 * yet, or by some tracers (like irqsoff), that just want to
397 * know if the ring buffer has been disabled, but it can handle
398 * races of where it gets disabled but we still do a record.
399 * As the check is in the fast path of the tracers, it is more
400 * important to be fast than accurate.
402 tr->buffer_disabled = 0;
403 /* Make the flag seen by readers */
408 * tracing_on - enable tracing buffers
410 * This function enables tracing buffers that may have been
411 * disabled with tracing_off.
413 void tracing_on(void)
415 tracer_tracing_on(&global_trace);
417 EXPORT_SYMBOL_GPL(tracing_on);
420 * __trace_puts - write a constant string into the trace buffer.
421 * @ip: The address of the caller
422 * @str: The constant string to write
423 * @size: The size of the string.
425 int __trace_puts(unsigned long ip, const char *str, int size)
427 struct ring_buffer_event *event;
428 struct ring_buffer *buffer;
429 struct print_entry *entry;
430 unsigned long irq_flags;
433 alloc = sizeof(*entry) + size + 2; /* possible \n added */
435 local_save_flags(irq_flags);
436 buffer = global_trace.trace_buffer.buffer;
437 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
438 irq_flags, preempt_count());
442 entry = ring_buffer_event_data(event);
445 memcpy(&entry->buf, str, size);
447 /* Add a newline if necessary */
448 if (entry->buf[size - 1] != '\n') {
449 entry->buf[size] = '\n';
450 entry->buf[size + 1] = '\0';
452 entry->buf[size] = '\0';
454 __buffer_unlock_commit(buffer, event);
458 EXPORT_SYMBOL_GPL(__trace_puts);
461 * __trace_bputs - write the pointer to a constant string into trace buffer
462 * @ip: The address of the caller
463 * @str: The constant string to write to the buffer to
465 int __trace_bputs(unsigned long ip, const char *str)
467 struct ring_buffer_event *event;
468 struct ring_buffer *buffer;
469 struct bputs_entry *entry;
470 unsigned long irq_flags;
471 int size = sizeof(struct bputs_entry);
473 local_save_flags(irq_flags);
474 buffer = global_trace.trace_buffer.buffer;
475 event = trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
476 irq_flags, preempt_count());
480 entry = ring_buffer_event_data(event);
484 __buffer_unlock_commit(buffer, event);
488 EXPORT_SYMBOL_GPL(__trace_bputs);
490 #ifdef CONFIG_TRACER_SNAPSHOT
492 * trace_snapshot - take a snapshot of the current buffer.
494 * This causes a swap between the snapshot buffer and the current live
495 * tracing buffer. You can use this to take snapshots of the live
496 * trace when some condition is triggered, but continue to trace.
498 * Note, make sure to allocate the snapshot with either
499 * a tracing_snapshot_alloc(), or by doing it manually
500 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
502 * If the snapshot buffer is not allocated, it will stop tracing.
503 * Basically making a permanent snapshot.
505 void tracing_snapshot(void)
507 struct trace_array *tr = &global_trace;
508 struct tracer *tracer = tr->current_trace;
512 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
513 internal_trace_puts("*** snapshot is being ignored ***\n");
517 if (!tr->allocated_snapshot) {
518 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
519 internal_trace_puts("*** stopping trace here! ***\n");
524 /* Note, snapshot can not be used when the tracer uses it */
525 if (tracer->use_max_tr) {
526 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
527 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
531 local_irq_save(flags);
532 update_max_tr(tr, current, smp_processor_id());
533 local_irq_restore(flags);
535 EXPORT_SYMBOL_GPL(tracing_snapshot);
537 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
538 struct trace_buffer *size_buf, int cpu_id);
539 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val);
541 static int alloc_snapshot(struct trace_array *tr)
545 if (!tr->allocated_snapshot) {
547 /* allocate spare buffer */
548 ret = resize_buffer_duplicate_size(&tr->max_buffer,
549 &tr->trace_buffer, RING_BUFFER_ALL_CPUS);
553 tr->allocated_snapshot = true;
559 void free_snapshot(struct trace_array *tr)
562 * We don't free the ring buffer. instead, resize it because
563 * The max_tr ring buffer has some state (e.g. ring->clock) and
564 * we want preserve it.
566 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
567 set_buffer_entries(&tr->max_buffer, 1);
568 tracing_reset_online_cpus(&tr->max_buffer);
569 tr->allocated_snapshot = false;
573 * trace_snapshot_alloc - allocate and take a snapshot of the current buffer.
575 * This is similar to trace_snapshot(), but it will allocate the
576 * snapshot buffer if it isn't already allocated. Use this only
577 * where it is safe to sleep, as the allocation may sleep.
579 * This causes a swap between the snapshot buffer and the current live
580 * tracing buffer. You can use this to take snapshots of the live
581 * trace when some condition is triggered, but continue to trace.
583 void tracing_snapshot_alloc(void)
585 struct trace_array *tr = &global_trace;
588 ret = alloc_snapshot(tr);
589 if (WARN_ON(ret < 0))
594 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
596 void tracing_snapshot(void)
598 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
600 EXPORT_SYMBOL_GPL(tracing_snapshot);
601 void tracing_snapshot_alloc(void)
606 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
607 #endif /* CONFIG_TRACER_SNAPSHOT */
609 static void tracer_tracing_off(struct trace_array *tr)
611 if (tr->trace_buffer.buffer)
612 ring_buffer_record_off(tr->trace_buffer.buffer);
614 * This flag is looked at when buffers haven't been allocated
615 * yet, or by some tracers (like irqsoff), that just want to
616 * know if the ring buffer has been disabled, but it can handle
617 * races of where it gets disabled but we still do a record.
618 * As the check is in the fast path of the tracers, it is more
619 * important to be fast than accurate.
621 tr->buffer_disabled = 1;
622 /* Make the flag seen by readers */
627 * tracing_off - turn off tracing buffers
629 * This function stops the tracing buffers from recording data.
630 * It does not disable any overhead the tracers themselves may
631 * be causing. This function simply causes all recording to
632 * the ring buffers to fail.
634 void tracing_off(void)
636 tracer_tracing_off(&global_trace);
638 EXPORT_SYMBOL_GPL(tracing_off);
640 void disable_trace_on_warning(void)
642 if (__disable_trace_on_warning)
647 * tracer_tracing_is_on - show real state of ring buffer enabled
648 * @tr : the trace array to know if ring buffer is enabled
650 * Shows real state of the ring buffer if it is enabled or not.
652 static int tracer_tracing_is_on(struct trace_array *tr)
654 if (tr->trace_buffer.buffer)
655 return ring_buffer_record_is_on(tr->trace_buffer.buffer);
656 return !tr->buffer_disabled;
660 * tracing_is_on - show state of ring buffers enabled
662 int tracing_is_on(void)
664 return tracer_tracing_is_on(&global_trace);
666 EXPORT_SYMBOL_GPL(tracing_is_on);
668 static int __init set_buf_size(char *str)
670 unsigned long buf_size;
674 buf_size = memparse(str, &str);
675 /* nr_entries can not be zero */
678 trace_buf_size = buf_size;
681 __setup("trace_buf_size=", set_buf_size);
683 static int __init set_tracing_thresh(char *str)
685 unsigned long threshold;
690 ret = kstrtoul(str, 0, &threshold);
693 tracing_thresh = threshold * 1000;
696 __setup("tracing_thresh=", set_tracing_thresh);
698 unsigned long nsecs_to_usecs(unsigned long nsecs)
703 /* These must match the bit postions in trace_iterator_flags */
704 static const char *trace_options[] = {
737 int in_ns; /* is this clock in nanoseconds? */
739 { trace_clock_local, "local", 1 },
740 { trace_clock_global, "global", 1 },
741 { trace_clock_counter, "counter", 0 },
742 { trace_clock_jiffies, "uptime", 1 },
743 { trace_clock, "perf", 1 },
750 * trace_parser_get_init - gets the buffer for trace parser
752 int trace_parser_get_init(struct trace_parser *parser, int size)
754 memset(parser, 0, sizeof(*parser));
756 parser->buffer = kmalloc(size, GFP_KERNEL);
765 * trace_parser_put - frees the buffer for trace parser
767 void trace_parser_put(struct trace_parser *parser)
769 kfree(parser->buffer);
773 * trace_get_user - reads the user input string separated by space
774 * (matched by isspace(ch))
776 * For each string found the 'struct trace_parser' is updated,
777 * and the function returns.
779 * Returns number of bytes read.
781 * See kernel/trace/trace.h for 'struct trace_parser' details.
783 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
784 size_t cnt, loff_t *ppos)
791 trace_parser_clear(parser);
793 ret = get_user(ch, ubuf++);
801 * The parser is not finished with the last write,
802 * continue reading the user input without skipping spaces.
805 /* skip white space */
806 while (cnt && isspace(ch)) {
807 ret = get_user(ch, ubuf++);
814 /* only spaces were written */
824 /* read the non-space input */
825 while (cnt && !isspace(ch)) {
826 if (parser->idx < parser->size - 1)
827 parser->buffer[parser->idx++] = ch;
832 ret = get_user(ch, ubuf++);
839 /* We either got finished input or we have to wait for another call. */
841 parser->buffer[parser->idx] = 0;
842 parser->cont = false;
845 parser->buffer[parser->idx++] = ch;
855 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
863 if (s->len <= s->readpos)
866 len = s->len - s->readpos;
869 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
879 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
883 if (s->len <= s->readpos)
886 len = s->len - s->readpos;
889 memcpy(buf, s->buffer + s->readpos, cnt);
896 * ftrace_max_lock is used to protect the swapping of buffers
897 * when taking a max snapshot. The buffers themselves are
898 * protected by per_cpu spinlocks. But the action of the swap
899 * needs its own lock.
901 * This is defined as a arch_spinlock_t in order to help
902 * with performance when lockdep debugging is enabled.
904 * It is also used in other places outside the update_max_tr
905 * so it needs to be defined outside of the
906 * CONFIG_TRACER_MAX_TRACE.
908 static arch_spinlock_t ftrace_max_lock =
909 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
911 unsigned long __read_mostly tracing_thresh;
913 #ifdef CONFIG_TRACER_MAX_TRACE
914 unsigned long __read_mostly tracing_max_latency;
917 * Copy the new maximum trace into the separate maximum-trace
918 * structure. (this way the maximum trace is permanently saved,
919 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
922 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
924 struct trace_buffer *trace_buf = &tr->trace_buffer;
925 struct trace_buffer *max_buf = &tr->max_buffer;
926 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
927 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
930 max_buf->time_start = data->preempt_timestamp;
932 max_data->saved_latency = tracing_max_latency;
933 max_data->critical_start = data->critical_start;
934 max_data->critical_end = data->critical_end;
936 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
937 max_data->pid = tsk->pid;
939 * If tsk == current, then use current_uid(), as that does not use
940 * RCU. The irq tracer can be called out of RCU scope.
943 max_data->uid = current_uid();
945 max_data->uid = task_uid(tsk);
947 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
948 max_data->policy = tsk->policy;
949 max_data->rt_priority = tsk->rt_priority;
951 /* record this tasks comm */
952 tracing_record_cmdline(tsk);
956 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
958 * @tsk: the task with the latency
959 * @cpu: The cpu that initiated the trace.
961 * Flip the buffers between the @tr and the max_tr and record information
962 * about which task was the cause of this latency.
965 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
967 struct ring_buffer *buf;
972 WARN_ON_ONCE(!irqs_disabled());
974 if (!tr->allocated_snapshot) {
975 /* Only the nop tracer should hit this when disabling */
976 WARN_ON_ONCE(tr->current_trace != &nop_trace);
980 arch_spin_lock(&ftrace_max_lock);
982 buf = tr->trace_buffer.buffer;
983 tr->trace_buffer.buffer = tr->max_buffer.buffer;
984 tr->max_buffer.buffer = buf;
986 __update_max_tr(tr, tsk, cpu);
987 arch_spin_unlock(&ftrace_max_lock);
991 * update_max_tr_single - only copy one trace over, and reset the rest
993 * @tsk - task with the latency
994 * @cpu - the cpu of the buffer to copy.
996 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
999 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1006 WARN_ON_ONCE(!irqs_disabled());
1007 if (!tr->allocated_snapshot) {
1008 /* Only the nop tracer should hit this when disabling */
1009 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1013 arch_spin_lock(&ftrace_max_lock);
1015 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->trace_buffer.buffer, cpu);
1017 if (ret == -EBUSY) {
1019 * We failed to swap the buffer due to a commit taking
1020 * place on this CPU. We fail to record, but we reset
1021 * the max trace buffer (no one writes directly to it)
1022 * and flag that it failed.
1024 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1025 "Failed to swap buffers due to commit in progress\n");
1028 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1030 __update_max_tr(tr, tsk, cpu);
1031 arch_spin_unlock(&ftrace_max_lock);
1033 #endif /* CONFIG_TRACER_MAX_TRACE */
1035 static void default_wait_pipe(struct trace_iterator *iter)
1037 /* Iterators are static, they should be filled or empty */
1038 if (trace_buffer_iter(iter, iter->cpu_file))
1041 ring_buffer_wait(iter->trace_buffer->buffer, iter->cpu_file);
1044 #ifdef CONFIG_FTRACE_STARTUP_TEST
1045 static int run_tracer_selftest(struct tracer *type)
1047 struct trace_array *tr = &global_trace;
1048 struct tracer *saved_tracer = tr->current_trace;
1051 if (!type->selftest || tracing_selftest_disabled)
1055 * Run a selftest on this tracer.
1056 * Here we reset the trace buffer, and set the current
1057 * tracer to be this tracer. The tracer can then run some
1058 * internal tracing to verify that everything is in order.
1059 * If we fail, we do not register this tracer.
1061 tracing_reset_online_cpus(&tr->trace_buffer);
1063 tr->current_trace = type;
1065 #ifdef CONFIG_TRACER_MAX_TRACE
1066 if (type->use_max_tr) {
1067 /* If we expanded the buffers, make sure the max is expanded too */
1068 if (ring_buffer_expanded)
1069 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1070 RING_BUFFER_ALL_CPUS);
1071 tr->allocated_snapshot = true;
1075 /* the test is responsible for initializing and enabling */
1076 pr_info("Testing tracer %s: ", type->name);
1077 ret = type->selftest(type, tr);
1078 /* the test is responsible for resetting too */
1079 tr->current_trace = saved_tracer;
1081 printk(KERN_CONT "FAILED!\n");
1082 /* Add the warning after printing 'FAILED' */
1086 /* Only reset on passing, to avoid touching corrupted buffers */
1087 tracing_reset_online_cpus(&tr->trace_buffer);
1089 #ifdef CONFIG_TRACER_MAX_TRACE
1090 if (type->use_max_tr) {
1091 tr->allocated_snapshot = false;
1093 /* Shrink the max buffer again */
1094 if (ring_buffer_expanded)
1095 ring_buffer_resize(tr->max_buffer.buffer, 1,
1096 RING_BUFFER_ALL_CPUS);
1100 printk(KERN_CONT "PASSED\n");
1104 static inline int run_tracer_selftest(struct tracer *type)
1108 #endif /* CONFIG_FTRACE_STARTUP_TEST */
1111 * register_tracer - register a tracer with the ftrace system.
1112 * @type - the plugin for the tracer
1114 * Register a new plugin tracer.
1116 int register_tracer(struct tracer *type)
1122 pr_info("Tracer must have a name\n");
1126 if (strlen(type->name) >= MAX_TRACER_SIZE) {
1127 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
1131 mutex_lock(&trace_types_lock);
1133 tracing_selftest_running = true;
1135 for (t = trace_types; t; t = t->next) {
1136 if (strcmp(type->name, t->name) == 0) {
1138 pr_info("Tracer %s already registered\n",
1145 if (!type->set_flag)
1146 type->set_flag = &dummy_set_flag;
1148 type->flags = &dummy_tracer_flags;
1150 if (!type->flags->opts)
1151 type->flags->opts = dummy_tracer_opt;
1152 if (!type->wait_pipe)
1153 type->wait_pipe = default_wait_pipe;
1155 ret = run_tracer_selftest(type);
1159 type->next = trace_types;
1163 tracing_selftest_running = false;
1164 mutex_unlock(&trace_types_lock);
1166 if (ret || !default_bootup_tracer)
1169 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
1172 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
1173 /* Do we want this tracer to start on bootup? */
1174 tracing_set_tracer(type->name);
1175 default_bootup_tracer = NULL;
1176 /* disable other selftests, since this will break it. */
1177 tracing_selftest_disabled = true;
1178 #ifdef CONFIG_FTRACE_STARTUP_TEST
1179 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
1187 void tracing_reset(struct trace_buffer *buf, int cpu)
1189 struct ring_buffer *buffer = buf->buffer;
1194 ring_buffer_record_disable(buffer);
1196 /* Make sure all commits have finished */
1197 synchronize_sched();
1198 ring_buffer_reset_cpu(buffer, cpu);
1200 ring_buffer_record_enable(buffer);
1203 void tracing_reset_online_cpus(struct trace_buffer *buf)
1205 struct ring_buffer *buffer = buf->buffer;
1211 ring_buffer_record_disable(buffer);
1213 /* Make sure all commits have finished */
1214 synchronize_sched();
1216 buf->time_start = ftrace_now(buf->cpu);
1218 for_each_online_cpu(cpu)
1219 ring_buffer_reset_cpu(buffer, cpu);
1221 ring_buffer_record_enable(buffer);
1224 void tracing_reset_current(int cpu)
1226 tracing_reset(&global_trace.trace_buffer, cpu);
1229 void tracing_reset_all_online_cpus(void)
1231 struct trace_array *tr;
1233 mutex_lock(&trace_types_lock);
1234 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1235 tracing_reset_online_cpus(&tr->trace_buffer);
1236 #ifdef CONFIG_TRACER_MAX_TRACE
1237 tracing_reset_online_cpus(&tr->max_buffer);
1240 mutex_unlock(&trace_types_lock);
1243 #define SAVED_CMDLINES 128
1244 #define NO_CMDLINE_MAP UINT_MAX
1245 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
1246 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
1247 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
1248 static int cmdline_idx;
1249 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1251 /* temporary disable recording */
1252 static atomic_t trace_record_cmdline_disabled __read_mostly;
1254 static void trace_init_cmdlines(void)
1256 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
1257 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
1261 int is_tracing_stopped(void)
1263 return global_trace.stop_count;
1267 * ftrace_off_permanent - disable all ftrace code permanently
1269 * This should only be called when a serious anomally has
1270 * been detected. This will turn off the function tracing,
1271 * ring buffers, and other tracing utilites. It takes no
1272 * locks and can be called from any context.
1274 void ftrace_off_permanent(void)
1276 tracing_disabled = 1;
1278 tracing_off_permanent();
1282 * tracing_start - quick start of the tracer
1284 * If tracing is enabled but was stopped by tracing_stop,
1285 * this will start the tracer back up.
1287 void tracing_start(void)
1289 struct ring_buffer *buffer;
1290 unsigned long flags;
1292 if (tracing_disabled)
1295 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1296 if (--global_trace.stop_count) {
1297 if (global_trace.stop_count < 0) {
1298 /* Someone screwed up their debugging */
1300 global_trace.stop_count = 0;
1305 /* Prevent the buffers from switching */
1306 arch_spin_lock(&ftrace_max_lock);
1308 buffer = global_trace.trace_buffer.buffer;
1310 ring_buffer_record_enable(buffer);
1312 #ifdef CONFIG_TRACER_MAX_TRACE
1313 buffer = global_trace.max_buffer.buffer;
1315 ring_buffer_record_enable(buffer);
1318 arch_spin_unlock(&ftrace_max_lock);
1322 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1325 static void tracing_start_tr(struct trace_array *tr)
1327 struct ring_buffer *buffer;
1328 unsigned long flags;
1330 if (tracing_disabled)
1333 /* If global, we need to also start the max tracer */
1334 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1335 return tracing_start();
1337 raw_spin_lock_irqsave(&tr->start_lock, flags);
1339 if (--tr->stop_count) {
1340 if (tr->stop_count < 0) {
1341 /* Someone screwed up their debugging */
1348 buffer = tr->trace_buffer.buffer;
1350 ring_buffer_record_enable(buffer);
1353 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1357 * tracing_stop - quick stop of the tracer
1359 * Light weight way to stop tracing. Use in conjunction with
1362 void tracing_stop(void)
1364 struct ring_buffer *buffer;
1365 unsigned long flags;
1368 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1369 if (global_trace.stop_count++)
1372 /* Prevent the buffers from switching */
1373 arch_spin_lock(&ftrace_max_lock);
1375 buffer = global_trace.trace_buffer.buffer;
1377 ring_buffer_record_disable(buffer);
1379 #ifdef CONFIG_TRACER_MAX_TRACE
1380 buffer = global_trace.max_buffer.buffer;
1382 ring_buffer_record_disable(buffer);
1385 arch_spin_unlock(&ftrace_max_lock);
1388 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1391 static void tracing_stop_tr(struct trace_array *tr)
1393 struct ring_buffer *buffer;
1394 unsigned long flags;
1396 /* If global, we need to also stop the max tracer */
1397 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1398 return tracing_stop();
1400 raw_spin_lock_irqsave(&tr->start_lock, flags);
1401 if (tr->stop_count++)
1404 buffer = tr->trace_buffer.buffer;
1406 ring_buffer_record_disable(buffer);
1409 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1412 void trace_stop_cmdline_recording(void);
1414 static void trace_save_cmdline(struct task_struct *tsk)
1418 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1422 * It's not the end of the world if we don't get
1423 * the lock, but we also don't want to spin
1424 * nor do we want to disable interrupts,
1425 * so if we miss here, then better luck next time.
1427 if (!arch_spin_trylock(&trace_cmdline_lock))
1430 idx = map_pid_to_cmdline[tsk->pid];
1431 if (idx == NO_CMDLINE_MAP) {
1432 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1435 * Check whether the cmdline buffer at idx has a pid
1436 * mapped. We are going to overwrite that entry so we
1437 * need to clear the map_pid_to_cmdline. Otherwise we
1438 * would read the new comm for the old pid.
1440 pid = map_cmdline_to_pid[idx];
1441 if (pid != NO_CMDLINE_MAP)
1442 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1444 map_cmdline_to_pid[idx] = tsk->pid;
1445 map_pid_to_cmdline[tsk->pid] = idx;
1450 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1452 arch_spin_unlock(&trace_cmdline_lock);
1455 void trace_find_cmdline(int pid, char comm[])
1460 strcpy(comm, "<idle>");
1464 if (WARN_ON_ONCE(pid < 0)) {
1465 strcpy(comm, "<XXX>");
1469 if (pid > PID_MAX_DEFAULT) {
1470 strcpy(comm, "<...>");
1475 arch_spin_lock(&trace_cmdline_lock);
1476 map = map_pid_to_cmdline[pid];
1477 if (map != NO_CMDLINE_MAP)
1478 strcpy(comm, saved_cmdlines[map]);
1480 strcpy(comm, "<...>");
1482 arch_spin_unlock(&trace_cmdline_lock);
1486 void tracing_record_cmdline(struct task_struct *tsk)
1488 if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on())
1491 if (!__this_cpu_read(trace_cmdline_save))
1494 __this_cpu_write(trace_cmdline_save, false);
1496 trace_save_cmdline(tsk);
1500 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1503 struct task_struct *tsk = current;
1505 entry->preempt_count = pc & 0xff;
1506 entry->pid = (tsk) ? tsk->pid : 0;
1508 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1509 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1511 TRACE_FLAG_IRQS_NOSUPPORT |
1513 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1514 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1515 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1517 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1519 struct ring_buffer_event *
1520 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1523 unsigned long flags, int pc)
1525 struct ring_buffer_event *event;
1527 event = ring_buffer_lock_reserve(buffer, len);
1528 if (event != NULL) {
1529 struct trace_entry *ent = ring_buffer_event_data(event);
1531 tracing_generic_entry_update(ent, flags, pc);
1539 __buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event)
1541 __this_cpu_write(trace_cmdline_save, true);
1542 ring_buffer_unlock_commit(buffer, event);
1546 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1547 struct ring_buffer_event *event,
1548 unsigned long flags, int pc)
1550 __buffer_unlock_commit(buffer, event);
1552 ftrace_trace_stack(buffer, flags, 6, pc);
1553 ftrace_trace_userstack(buffer, flags, pc);
1556 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1557 struct ring_buffer_event *event,
1558 unsigned long flags, int pc)
1560 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1562 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit);
1564 struct ring_buffer_event *
1565 trace_event_buffer_lock_reserve(struct ring_buffer **current_rb,
1566 struct ftrace_event_file *ftrace_file,
1567 int type, unsigned long len,
1568 unsigned long flags, int pc)
1570 *current_rb = ftrace_file->tr->trace_buffer.buffer;
1571 return trace_buffer_lock_reserve(*current_rb,
1572 type, len, flags, pc);
1574 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
1576 struct ring_buffer_event *
1577 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1578 int type, unsigned long len,
1579 unsigned long flags, int pc)
1581 *current_rb = global_trace.trace_buffer.buffer;
1582 return trace_buffer_lock_reserve(*current_rb,
1583 type, len, flags, pc);
1585 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1587 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1588 struct ring_buffer_event *event,
1589 unsigned long flags, int pc)
1591 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1593 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1595 void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1596 struct ring_buffer_event *event,
1597 unsigned long flags, int pc,
1598 struct pt_regs *regs)
1600 __buffer_unlock_commit(buffer, event);
1602 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1603 ftrace_trace_userstack(buffer, flags, pc);
1605 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs);
1607 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1608 struct ring_buffer_event *event)
1610 ring_buffer_discard_commit(buffer, event);
1612 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1615 trace_function(struct trace_array *tr,
1616 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1619 struct ftrace_event_call *call = &event_function;
1620 struct ring_buffer *buffer = tr->trace_buffer.buffer;
1621 struct ring_buffer_event *event;
1622 struct ftrace_entry *entry;
1624 /* If we are reading the ring buffer, don't trace */
1625 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1628 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1632 entry = ring_buffer_event_data(event);
1634 entry->parent_ip = parent_ip;
1636 if (!filter_check_discard(call, entry, buffer, event))
1637 __buffer_unlock_commit(buffer, event);
1641 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1642 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1645 if (likely(!atomic_read(&data->disabled)))
1646 trace_function(tr, ip, parent_ip, flags, pc);
1649 #ifdef CONFIG_STACKTRACE
1651 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1652 struct ftrace_stack {
1653 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1656 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1657 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1659 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1660 unsigned long flags,
1661 int skip, int pc, struct pt_regs *regs)
1663 struct ftrace_event_call *call = &event_kernel_stack;
1664 struct ring_buffer_event *event;
1665 struct stack_entry *entry;
1666 struct stack_trace trace;
1668 int size = FTRACE_STACK_ENTRIES;
1670 trace.nr_entries = 0;
1674 * Since events can happen in NMIs there's no safe way to
1675 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1676 * or NMI comes in, it will just have to use the default
1677 * FTRACE_STACK_SIZE.
1679 preempt_disable_notrace();
1681 use_stack = __this_cpu_inc_return(ftrace_stack_reserve);
1683 * We don't need any atomic variables, just a barrier.
1684 * If an interrupt comes in, we don't care, because it would
1685 * have exited and put the counter back to what we want.
1686 * We just need a barrier to keep gcc from moving things
1690 if (use_stack == 1) {
1691 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1692 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1695 save_stack_trace_regs(regs, &trace);
1697 save_stack_trace(&trace);
1699 if (trace.nr_entries > size)
1700 size = trace.nr_entries;
1702 /* From now on, use_stack is a boolean */
1705 size *= sizeof(unsigned long);
1707 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1708 sizeof(*entry) + size, flags, pc);
1711 entry = ring_buffer_event_data(event);
1713 memset(&entry->caller, 0, size);
1716 memcpy(&entry->caller, trace.entries,
1717 trace.nr_entries * sizeof(unsigned long));
1719 trace.max_entries = FTRACE_STACK_ENTRIES;
1720 trace.entries = entry->caller;
1722 save_stack_trace_regs(regs, &trace);
1724 save_stack_trace(&trace);
1727 entry->size = trace.nr_entries;
1729 if (!filter_check_discard(call, entry, buffer, event))
1730 __buffer_unlock_commit(buffer, event);
1733 /* Again, don't let gcc optimize things here */
1735 __this_cpu_dec(ftrace_stack_reserve);
1736 preempt_enable_notrace();
1740 void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1741 int skip, int pc, struct pt_regs *regs)
1743 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1746 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1749 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1752 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1755 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1758 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1761 __ftrace_trace_stack(tr->trace_buffer.buffer, flags, skip, pc, NULL);
1765 * trace_dump_stack - record a stack back trace in the trace buffer
1766 * @skip: Number of functions to skip (helper handlers)
1768 void trace_dump_stack(int skip)
1770 unsigned long flags;
1772 if (tracing_disabled || tracing_selftest_running)
1775 local_save_flags(flags);
1778 * Skip 3 more, seems to get us at the caller of
1782 __ftrace_trace_stack(global_trace.trace_buffer.buffer,
1783 flags, skip, preempt_count(), NULL);
1786 static DEFINE_PER_CPU(int, user_stack_count);
1789 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1791 struct ftrace_event_call *call = &event_user_stack;
1792 struct ring_buffer_event *event;
1793 struct userstack_entry *entry;
1794 struct stack_trace trace;
1796 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1800 * NMIs can not handle page faults, even with fix ups.
1801 * The save user stack can (and often does) fault.
1803 if (unlikely(in_nmi()))
1807 * prevent recursion, since the user stack tracing may
1808 * trigger other kernel events.
1811 if (__this_cpu_read(user_stack_count))
1814 __this_cpu_inc(user_stack_count);
1816 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1817 sizeof(*entry), flags, pc);
1819 goto out_drop_count;
1820 entry = ring_buffer_event_data(event);
1822 entry->tgid = current->tgid;
1823 memset(&entry->caller, 0, sizeof(entry->caller));
1825 trace.nr_entries = 0;
1826 trace.max_entries = FTRACE_STACK_ENTRIES;
1828 trace.entries = entry->caller;
1830 save_stack_trace_user(&trace);
1831 if (!filter_check_discard(call, entry, buffer, event))
1832 __buffer_unlock_commit(buffer, event);
1835 __this_cpu_dec(user_stack_count);
1841 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1843 ftrace_trace_userstack(tr, flags, preempt_count());
1847 #endif /* CONFIG_STACKTRACE */
1849 /* created for use with alloc_percpu */
1850 struct trace_buffer_struct {
1851 char buffer[TRACE_BUF_SIZE];
1854 static struct trace_buffer_struct *trace_percpu_buffer;
1855 static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1856 static struct trace_buffer_struct *trace_percpu_irq_buffer;
1857 static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1860 * The buffer used is dependent on the context. There is a per cpu
1861 * buffer for normal context, softirq contex, hard irq context and
1862 * for NMI context. Thise allows for lockless recording.
1864 * Note, if the buffers failed to be allocated, then this returns NULL
1866 static char *get_trace_buf(void)
1868 struct trace_buffer_struct *percpu_buffer;
1871 * If we have allocated per cpu buffers, then we do not
1872 * need to do any locking.
1875 percpu_buffer = trace_percpu_nmi_buffer;
1877 percpu_buffer = trace_percpu_irq_buffer;
1878 else if (in_softirq())
1879 percpu_buffer = trace_percpu_sirq_buffer;
1881 percpu_buffer = trace_percpu_buffer;
1886 return this_cpu_ptr(&percpu_buffer->buffer[0]);
1889 static int alloc_percpu_trace_buffer(void)
1891 struct trace_buffer_struct *buffers;
1892 struct trace_buffer_struct *sirq_buffers;
1893 struct trace_buffer_struct *irq_buffers;
1894 struct trace_buffer_struct *nmi_buffers;
1896 buffers = alloc_percpu(struct trace_buffer_struct);
1900 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
1904 irq_buffers = alloc_percpu(struct trace_buffer_struct);
1908 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
1912 trace_percpu_buffer = buffers;
1913 trace_percpu_sirq_buffer = sirq_buffers;
1914 trace_percpu_irq_buffer = irq_buffers;
1915 trace_percpu_nmi_buffer = nmi_buffers;
1920 free_percpu(irq_buffers);
1922 free_percpu(sirq_buffers);
1924 free_percpu(buffers);
1926 WARN(1, "Could not allocate percpu trace_printk buffer");
1930 static int buffers_allocated;
1932 void trace_printk_init_buffers(void)
1934 if (buffers_allocated)
1937 if (alloc_percpu_trace_buffer())
1940 pr_info("ftrace: Allocated trace_printk buffers\n");
1942 /* Expand the buffers to set size */
1943 tracing_update_buffers();
1945 buffers_allocated = 1;
1948 * trace_printk_init_buffers() can be called by modules.
1949 * If that happens, then we need to start cmdline recording
1950 * directly here. If the global_trace.buffer is already
1951 * allocated here, then this was called by module code.
1953 if (global_trace.trace_buffer.buffer)
1954 tracing_start_cmdline_record();
1957 void trace_printk_start_comm(void)
1959 /* Start tracing comms if trace printk is set */
1960 if (!buffers_allocated)
1962 tracing_start_cmdline_record();
1965 static void trace_printk_start_stop_comm(int enabled)
1967 if (!buffers_allocated)
1971 tracing_start_cmdline_record();
1973 tracing_stop_cmdline_record();
1977 * trace_vbprintk - write binary msg to tracing buffer
1980 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1982 struct ftrace_event_call *call = &event_bprint;
1983 struct ring_buffer_event *event;
1984 struct ring_buffer *buffer;
1985 struct trace_array *tr = &global_trace;
1986 struct bprint_entry *entry;
1987 unsigned long flags;
1989 int len = 0, size, pc;
1991 if (unlikely(tracing_selftest_running || tracing_disabled))
1994 /* Don't pollute graph traces with trace_vprintk internals */
1995 pause_graph_tracing();
1997 pc = preempt_count();
1998 preempt_disable_notrace();
2000 tbuffer = get_trace_buf();
2006 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
2008 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
2011 local_save_flags(flags);
2012 size = sizeof(*entry) + sizeof(u32) * len;
2013 buffer = tr->trace_buffer.buffer;
2014 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
2018 entry = ring_buffer_event_data(event);
2022 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
2023 if (!filter_check_discard(call, entry, buffer, event)) {
2024 __buffer_unlock_commit(buffer, event);
2025 ftrace_trace_stack(buffer, flags, 6, pc);
2029 preempt_enable_notrace();
2030 unpause_graph_tracing();
2034 EXPORT_SYMBOL_GPL(trace_vbprintk);
2037 __trace_array_vprintk(struct ring_buffer *buffer,
2038 unsigned long ip, const char *fmt, va_list args)
2040 struct ftrace_event_call *call = &event_print;
2041 struct ring_buffer_event *event;
2042 int len = 0, size, pc;
2043 struct print_entry *entry;
2044 unsigned long flags;
2047 if (tracing_disabled || tracing_selftest_running)
2050 /* Don't pollute graph traces with trace_vprintk internals */
2051 pause_graph_tracing();
2053 pc = preempt_count();
2054 preempt_disable_notrace();
2057 tbuffer = get_trace_buf();
2063 len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
2064 if (len > TRACE_BUF_SIZE)
2067 local_save_flags(flags);
2068 size = sizeof(*entry) + len + 1;
2069 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
2073 entry = ring_buffer_event_data(event);
2076 memcpy(&entry->buf, tbuffer, len);
2077 entry->buf[len] = '\0';
2078 if (!filter_check_discard(call, entry, buffer, event)) {
2079 __buffer_unlock_commit(buffer, event);
2080 ftrace_trace_stack(buffer, flags, 6, pc);
2083 preempt_enable_notrace();
2084 unpause_graph_tracing();
2089 int trace_array_vprintk(struct trace_array *tr,
2090 unsigned long ip, const char *fmt, va_list args)
2092 return __trace_array_vprintk(tr->trace_buffer.buffer, ip, fmt, args);
2095 int trace_array_printk(struct trace_array *tr,
2096 unsigned long ip, const char *fmt, ...)
2101 if (!(trace_flags & TRACE_ITER_PRINTK))
2105 ret = trace_array_vprintk(tr, ip, fmt, ap);
2110 int trace_array_printk_buf(struct ring_buffer *buffer,
2111 unsigned long ip, const char *fmt, ...)
2116 if (!(trace_flags & TRACE_ITER_PRINTK))
2120 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
2125 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
2127 return trace_array_vprintk(&global_trace, ip, fmt, args);
2129 EXPORT_SYMBOL_GPL(trace_vprintk);
2131 static void trace_iterator_increment(struct trace_iterator *iter)
2133 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
2137 ring_buffer_read(buf_iter, NULL);
2140 static struct trace_entry *
2141 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
2142 unsigned long *lost_events)
2144 struct ring_buffer_event *event;
2145 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
2148 event = ring_buffer_iter_peek(buf_iter, ts);
2150 event = ring_buffer_peek(iter->trace_buffer->buffer, cpu, ts,
2154 iter->ent_size = ring_buffer_event_length(event);
2155 return ring_buffer_event_data(event);
2161 static struct trace_entry *
2162 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
2163 unsigned long *missing_events, u64 *ent_ts)
2165 struct ring_buffer *buffer = iter->trace_buffer->buffer;
2166 struct trace_entry *ent, *next = NULL;
2167 unsigned long lost_events = 0, next_lost = 0;
2168 int cpu_file = iter->cpu_file;
2169 u64 next_ts = 0, ts;
2175 * If we are in a per_cpu trace file, don't bother by iterating over
2176 * all cpu and peek directly.
2178 if (cpu_file > RING_BUFFER_ALL_CPUS) {
2179 if (ring_buffer_empty_cpu(buffer, cpu_file))
2181 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
2183 *ent_cpu = cpu_file;
2188 for_each_tracing_cpu(cpu) {
2190 if (ring_buffer_empty_cpu(buffer, cpu))
2193 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
2196 * Pick the entry with the smallest timestamp:
2198 if (ent && (!next || ts < next_ts)) {
2202 next_lost = lost_events;
2203 next_size = iter->ent_size;
2207 iter->ent_size = next_size;
2210 *ent_cpu = next_cpu;
2216 *missing_events = next_lost;
2221 /* Find the next real entry, without updating the iterator itself */
2222 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
2223 int *ent_cpu, u64 *ent_ts)
2225 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
2228 /* Find the next real entry, and increment the iterator to the next entry */
2229 void *trace_find_next_entry_inc(struct trace_iterator *iter)
2231 iter->ent = __find_next_entry(iter, &iter->cpu,
2232 &iter->lost_events, &iter->ts);
2235 trace_iterator_increment(iter);
2237 return iter->ent ? iter : NULL;
2240 static void trace_consume(struct trace_iterator *iter)
2242 ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu, &iter->ts,
2243 &iter->lost_events);
2246 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
2248 struct trace_iterator *iter = m->private;
2252 WARN_ON_ONCE(iter->leftover);
2256 /* can't go backwards */
2261 ent = trace_find_next_entry_inc(iter);
2265 while (ent && iter->idx < i)
2266 ent = trace_find_next_entry_inc(iter);
2273 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
2275 struct ring_buffer_event *event;
2276 struct ring_buffer_iter *buf_iter;
2277 unsigned long entries = 0;
2280 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = 0;
2282 buf_iter = trace_buffer_iter(iter, cpu);
2286 ring_buffer_iter_reset(buf_iter);
2289 * We could have the case with the max latency tracers
2290 * that a reset never took place on a cpu. This is evident
2291 * by the timestamp being before the start of the buffer.
2293 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
2294 if (ts >= iter->trace_buffer->time_start)
2297 ring_buffer_read(buf_iter, NULL);
2300 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = entries;
2304 * The current tracer is copied to avoid a global locking
2307 static void *s_start(struct seq_file *m, loff_t *pos)
2309 struct trace_iterator *iter = m->private;
2310 struct trace_array *tr = iter->tr;
2311 int cpu_file = iter->cpu_file;
2317 * copy the tracer to avoid using a global lock all around.
2318 * iter->trace is a copy of current_trace, the pointer to the
2319 * name may be used instead of a strcmp(), as iter->trace->name
2320 * will point to the same string as current_trace->name.
2322 mutex_lock(&trace_types_lock);
2323 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
2324 *iter->trace = *tr->current_trace;
2325 mutex_unlock(&trace_types_lock);
2327 #ifdef CONFIG_TRACER_MAX_TRACE
2328 if (iter->snapshot && iter->trace->use_max_tr)
2329 return ERR_PTR(-EBUSY);
2332 if (!iter->snapshot)
2333 atomic_inc(&trace_record_cmdline_disabled);
2335 if (*pos != iter->pos) {
2340 if (cpu_file == RING_BUFFER_ALL_CPUS) {
2341 for_each_tracing_cpu(cpu)
2342 tracing_iter_reset(iter, cpu);
2344 tracing_iter_reset(iter, cpu_file);
2347 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
2352 * If we overflowed the seq_file before, then we want
2353 * to just reuse the trace_seq buffer again.
2359 p = s_next(m, p, &l);
2363 trace_event_read_lock();
2364 trace_access_lock(cpu_file);
2368 static void s_stop(struct seq_file *m, void *p)
2370 struct trace_iterator *iter = m->private;
2372 #ifdef CONFIG_TRACER_MAX_TRACE
2373 if (iter->snapshot && iter->trace->use_max_tr)
2377 if (!iter->snapshot)
2378 atomic_dec(&trace_record_cmdline_disabled);
2380 trace_access_unlock(iter->cpu_file);
2381 trace_event_read_unlock();
2385 get_total_entries(struct trace_buffer *buf,
2386 unsigned long *total, unsigned long *entries)
2388 unsigned long count;
2394 for_each_tracing_cpu(cpu) {
2395 count = ring_buffer_entries_cpu(buf->buffer, cpu);
2397 * If this buffer has skipped entries, then we hold all
2398 * entries for the trace and we need to ignore the
2399 * ones before the time stamp.
2401 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
2402 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
2403 /* total is the same as the entries */
2407 ring_buffer_overrun_cpu(buf->buffer, cpu);
2412 static void print_lat_help_header(struct seq_file *m)
2414 seq_puts(m, "# _------=> CPU# \n");
2415 seq_puts(m, "# / _-----=> irqs-off \n");
2416 seq_puts(m, "# | / _----=> need-resched \n");
2417 seq_puts(m, "# || / _---=> hardirq/softirq \n");
2418 seq_puts(m, "# ||| / _--=> preempt-depth \n");
2419 seq_puts(m, "# |||| / delay \n");
2420 seq_puts(m, "# cmd pid ||||| time | caller \n");
2421 seq_puts(m, "# \\ / ||||| \\ | / \n");
2424 static void print_event_info(struct trace_buffer *buf, struct seq_file *m)
2426 unsigned long total;
2427 unsigned long entries;
2429 get_total_entries(buf, &total, &entries);
2430 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2431 entries, total, num_online_cpus());
2435 static void print_func_help_header(struct trace_buffer *buf, struct seq_file *m)
2437 print_event_info(buf, m);
2438 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2439 seq_puts(m, "# | | | | |\n");
2442 static void print_func_help_header_irq(struct trace_buffer *buf, struct seq_file *m)
2444 print_event_info(buf, m);
2445 seq_puts(m, "# _-----=> irqs-off\n");
2446 seq_puts(m, "# / _----=> need-resched\n");
2447 seq_puts(m, "# | / _---=> hardirq/softirq\n");
2448 seq_puts(m, "# || / _--=> preempt-depth\n");
2449 seq_puts(m, "# ||| / delay\n");
2450 seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2451 seq_puts(m, "# | | | |||| | |\n");
2455 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2457 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2458 struct trace_buffer *buf = iter->trace_buffer;
2459 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
2460 struct tracer *type = iter->trace;
2461 unsigned long entries;
2462 unsigned long total;
2463 const char *name = "preemption";
2467 get_total_entries(buf, &total, &entries);
2469 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2471 seq_puts(m, "# -----------------------------------"
2472 "---------------------------------\n");
2473 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2474 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2475 nsecs_to_usecs(data->saved_latency),
2479 #if defined(CONFIG_PREEMPT_NONE)
2481 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2483 #elif defined(CONFIG_PREEMPT)
2488 /* These are reserved for later use */
2491 seq_printf(m, " #P:%d)\n", num_online_cpus());
2495 seq_puts(m, "# -----------------\n");
2496 seq_printf(m, "# | task: %.16s-%d "
2497 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2498 data->comm, data->pid,
2499 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
2500 data->policy, data->rt_priority);
2501 seq_puts(m, "# -----------------\n");
2503 if (data->critical_start) {
2504 seq_puts(m, "# => started at: ");
2505 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2506 trace_print_seq(m, &iter->seq);
2507 seq_puts(m, "\n# => ended at: ");
2508 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2509 trace_print_seq(m, &iter->seq);
2510 seq_puts(m, "\n#\n");
2516 static void test_cpu_buff_start(struct trace_iterator *iter)
2518 struct trace_seq *s = &iter->seq;
2520 if (!(trace_flags & TRACE_ITER_ANNOTATE))
2523 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2526 if (cpumask_test_cpu(iter->cpu, iter->started))
2529 if (per_cpu_ptr(iter->trace_buffer->data, iter->cpu)->skipped_entries)
2532 cpumask_set_cpu(iter->cpu, iter->started);
2534 /* Don't print started cpu buffer for the first entry of the trace */
2536 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2540 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2542 struct trace_seq *s = &iter->seq;
2543 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2544 struct trace_entry *entry;
2545 struct trace_event *event;
2549 test_cpu_buff_start(iter);
2551 event = ftrace_find_event(entry->type);
2553 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2554 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2555 if (!trace_print_lat_context(iter))
2558 if (!trace_print_context(iter))
2564 return event->funcs->trace(iter, sym_flags, event);
2566 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2569 return TRACE_TYPE_HANDLED;
2571 return TRACE_TYPE_PARTIAL_LINE;
2574 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2576 struct trace_seq *s = &iter->seq;
2577 struct trace_entry *entry;
2578 struct trace_event *event;
2582 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2583 if (!trace_seq_printf(s, "%d %d %llu ",
2584 entry->pid, iter->cpu, iter->ts))
2588 event = ftrace_find_event(entry->type);
2590 return event->funcs->raw(iter, 0, event);
2592 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2595 return TRACE_TYPE_HANDLED;
2597 return TRACE_TYPE_PARTIAL_LINE;
2600 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2602 struct trace_seq *s = &iter->seq;
2603 unsigned char newline = '\n';
2604 struct trace_entry *entry;
2605 struct trace_event *event;
2609 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2610 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2611 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2612 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2615 event = ftrace_find_event(entry->type);
2617 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2618 if (ret != TRACE_TYPE_HANDLED)
2622 SEQ_PUT_FIELD_RET(s, newline);
2624 return TRACE_TYPE_HANDLED;
2627 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2629 struct trace_seq *s = &iter->seq;
2630 struct trace_entry *entry;
2631 struct trace_event *event;
2635 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2636 SEQ_PUT_FIELD_RET(s, entry->pid);
2637 SEQ_PUT_FIELD_RET(s, iter->cpu);
2638 SEQ_PUT_FIELD_RET(s, iter->ts);
2641 event = ftrace_find_event(entry->type);
2642 return event ? event->funcs->binary(iter, 0, event) :
2646 int trace_empty(struct trace_iterator *iter)
2648 struct ring_buffer_iter *buf_iter;
2651 /* If we are looking at one CPU buffer, only check that one */
2652 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
2653 cpu = iter->cpu_file;
2654 buf_iter = trace_buffer_iter(iter, cpu);
2656 if (!ring_buffer_iter_empty(buf_iter))
2659 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2665 for_each_tracing_cpu(cpu) {
2666 buf_iter = trace_buffer_iter(iter, cpu);
2668 if (!ring_buffer_iter_empty(buf_iter))
2671 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2679 /* Called with trace_event_read_lock() held. */
2680 enum print_line_t print_trace_line(struct trace_iterator *iter)
2682 enum print_line_t ret;
2684 if (iter->lost_events &&
2685 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2686 iter->cpu, iter->lost_events))
2687 return TRACE_TYPE_PARTIAL_LINE;
2689 if (iter->trace && iter->trace->print_line) {
2690 ret = iter->trace->print_line(iter);
2691 if (ret != TRACE_TYPE_UNHANDLED)
2695 if (iter->ent->type == TRACE_BPUTS &&
2696 trace_flags & TRACE_ITER_PRINTK &&
2697 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2698 return trace_print_bputs_msg_only(iter);
2700 if (iter->ent->type == TRACE_BPRINT &&
2701 trace_flags & TRACE_ITER_PRINTK &&
2702 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2703 return trace_print_bprintk_msg_only(iter);
2705 if (iter->ent->type == TRACE_PRINT &&
2706 trace_flags & TRACE_ITER_PRINTK &&
2707 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2708 return trace_print_printk_msg_only(iter);
2710 if (trace_flags & TRACE_ITER_BIN)
2711 return print_bin_fmt(iter);
2713 if (trace_flags & TRACE_ITER_HEX)
2714 return print_hex_fmt(iter);
2716 if (trace_flags & TRACE_ITER_RAW)
2717 return print_raw_fmt(iter);
2719 return print_trace_fmt(iter);
2722 void trace_latency_header(struct seq_file *m)
2724 struct trace_iterator *iter = m->private;
2726 /* print nothing if the buffers are empty */
2727 if (trace_empty(iter))
2730 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2731 print_trace_header(m, iter);
2733 if (!(trace_flags & TRACE_ITER_VERBOSE))
2734 print_lat_help_header(m);
2737 void trace_default_header(struct seq_file *m)
2739 struct trace_iterator *iter = m->private;
2741 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2744 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2745 /* print nothing if the buffers are empty */
2746 if (trace_empty(iter))
2748 print_trace_header(m, iter);
2749 if (!(trace_flags & TRACE_ITER_VERBOSE))
2750 print_lat_help_header(m);
2752 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2753 if (trace_flags & TRACE_ITER_IRQ_INFO)
2754 print_func_help_header_irq(iter->trace_buffer, m);
2756 print_func_help_header(iter->trace_buffer, m);
2761 static void test_ftrace_alive(struct seq_file *m)
2763 if (!ftrace_is_dead())
2765 seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2766 seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2769 #ifdef CONFIG_TRACER_MAX_TRACE
2770 static void show_snapshot_main_help(struct seq_file *m)
2772 seq_printf(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n");
2773 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2774 seq_printf(m, "# Takes a snapshot of the main buffer.\n");
2775 seq_printf(m, "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate)\n");
2776 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2777 seq_printf(m, "# is not a '0' or '1')\n");
2780 static void show_snapshot_percpu_help(struct seq_file *m)
2782 seq_printf(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
2783 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
2784 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2785 seq_printf(m, "# Takes a snapshot of the main buffer for this cpu.\n");
2787 seq_printf(m, "# echo 1 > snapshot : Not supported with this kernel.\n");
2788 seq_printf(m, "# Must use main snapshot file to allocate.\n");
2790 seq_printf(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n");
2791 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2792 seq_printf(m, "# is not a '0' or '1')\n");
2795 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
2797 if (iter->tr->allocated_snapshot)
2798 seq_printf(m, "#\n# * Snapshot is allocated *\n#\n");
2800 seq_printf(m, "#\n# * Snapshot is freed *\n#\n");
2802 seq_printf(m, "# Snapshot commands:\n");
2803 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
2804 show_snapshot_main_help(m);
2806 show_snapshot_percpu_help(m);
2809 /* Should never be called */
2810 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
2813 static int s_show(struct seq_file *m, void *v)
2815 struct trace_iterator *iter = v;
2818 if (iter->ent == NULL) {
2820 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2822 test_ftrace_alive(m);
2824 if (iter->snapshot && trace_empty(iter))
2825 print_snapshot_help(m, iter);
2826 else if (iter->trace && iter->trace->print_header)
2827 iter->trace->print_header(m);
2829 trace_default_header(m);
2831 } else if (iter->leftover) {
2833 * If we filled the seq_file buffer earlier, we
2834 * want to just show it now.
2836 ret = trace_print_seq(m, &iter->seq);
2838 /* ret should this time be zero, but you never know */
2839 iter->leftover = ret;
2842 print_trace_line(iter);
2843 ret = trace_print_seq(m, &iter->seq);
2845 * If we overflow the seq_file buffer, then it will
2846 * ask us for this data again at start up.
2848 * ret is 0 if seq_file write succeeded.
2851 iter->leftover = ret;
2857 static const struct seq_operations tracer_seq_ops = {
2864 static struct trace_iterator *
2865 __tracing_open(struct trace_array *tr, struct trace_cpu *tc,
2866 struct inode *inode, struct file *file, bool snapshot)
2868 struct trace_iterator *iter;
2871 if (tracing_disabled)
2872 return ERR_PTR(-ENODEV);
2874 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
2876 return ERR_PTR(-ENOMEM);
2878 iter->buffer_iter = kzalloc(sizeof(*iter->buffer_iter) * num_possible_cpus(),
2880 if (!iter->buffer_iter)
2884 * We make a copy of the current tracer to avoid concurrent
2885 * changes on it while we are reading.
2887 mutex_lock(&trace_types_lock);
2888 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2892 *iter->trace = *tr->current_trace;
2894 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2899 #ifdef CONFIG_TRACER_MAX_TRACE
2900 /* Currently only the top directory has a snapshot */
2901 if (tr->current_trace->print_max || snapshot)
2902 iter->trace_buffer = &tr->max_buffer;
2905 iter->trace_buffer = &tr->trace_buffer;
2906 iter->snapshot = snapshot;
2908 mutex_init(&iter->mutex);
2909 iter->cpu_file = tc->cpu;
2911 /* Notify the tracer early; before we stop tracing. */
2912 if (iter->trace && iter->trace->open)
2913 iter->trace->open(iter);
2915 /* Annotate start of buffers if we had overruns */
2916 if (ring_buffer_overruns(iter->trace_buffer->buffer))
2917 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2919 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
2920 if (trace_clocks[trace_clock_id].in_ns)
2921 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
2923 /* stop the trace while dumping if we are not opening "snapshot" */
2924 if (!iter->snapshot)
2925 tracing_stop_tr(tr);
2927 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
2928 for_each_tracing_cpu(cpu) {
2929 iter->buffer_iter[cpu] =
2930 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
2932 ring_buffer_read_prepare_sync();
2933 for_each_tracing_cpu(cpu) {
2934 ring_buffer_read_start(iter->buffer_iter[cpu]);
2935 tracing_iter_reset(iter, cpu);
2938 cpu = iter->cpu_file;
2939 iter->buffer_iter[cpu] =
2940 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
2941 ring_buffer_read_prepare_sync();
2942 ring_buffer_read_start(iter->buffer_iter[cpu]);
2943 tracing_iter_reset(iter, cpu);
2946 mutex_unlock(&trace_types_lock);
2951 mutex_unlock(&trace_types_lock);
2953 kfree(iter->buffer_iter);
2955 seq_release_private(inode, file);
2956 return ERR_PTR(-ENOMEM);
2959 int tracing_open_generic(struct inode *inode, struct file *filp)
2961 if (tracing_disabled)
2964 filp->private_data = inode->i_private;
2969 * Open and update trace_array ref count.
2970 * Must have the current trace_array passed to it.
2972 int tracing_open_generic_tr(struct inode *inode, struct file *filp)
2974 struct trace_array *tr = inode->i_private;
2976 if (tracing_disabled)
2979 if (trace_array_get(tr) < 0)
2982 filp->private_data = inode->i_private;
2988 int tracing_open_generic_tc(struct inode *inode, struct file *filp)
2990 struct trace_cpu *tc = inode->i_private;
2991 struct trace_array *tr = tc->tr;
2993 if (tracing_disabled)
2996 if (trace_array_get(tr) < 0)
2999 filp->private_data = inode->i_private;
3005 static int tracing_release(struct inode *inode, struct file *file)
3007 struct seq_file *m = file->private_data;
3008 struct trace_iterator *iter;
3009 struct trace_array *tr;
3012 /* Writes do not use seq_file, need to grab tr from inode */
3013 if (!(file->f_mode & FMODE_READ)) {
3014 struct trace_cpu *tc = inode->i_private;
3016 trace_array_put(tc->tr);
3022 trace_array_put(tr);
3024 mutex_lock(&trace_types_lock);
3026 for_each_tracing_cpu(cpu) {
3027 if (iter->buffer_iter[cpu])
3028 ring_buffer_read_finish(iter->buffer_iter[cpu]);
3031 if (iter->trace && iter->trace->close)
3032 iter->trace->close(iter);
3034 if (!iter->snapshot)
3035 /* reenable tracing if it was previously enabled */
3036 tracing_start_tr(tr);
3037 mutex_unlock(&trace_types_lock);
3039 mutex_destroy(&iter->mutex);
3040 free_cpumask_var(iter->started);
3042 kfree(iter->buffer_iter);
3043 seq_release_private(inode, file);
3048 static int tracing_release_generic_tr(struct inode *inode, struct file *file)
3050 struct trace_array *tr = inode->i_private;
3052 trace_array_put(tr);
3056 static int tracing_release_generic_tc(struct inode *inode, struct file *file)
3058 struct trace_cpu *tc = inode->i_private;
3059 struct trace_array *tr = tc->tr;
3061 trace_array_put(tr);
3065 static int tracing_single_release_tr(struct inode *inode, struct file *file)
3067 struct trace_array *tr = inode->i_private;
3069 trace_array_put(tr);
3071 return single_release(inode, file);
3074 static int tracing_open(struct inode *inode, struct file *file)
3076 struct trace_cpu *tc = inode->i_private;
3077 struct trace_array *tr = tc->tr;
3078 struct trace_iterator *iter;
3081 if (trace_array_get(tr) < 0)
3084 /* If this file was open for write, then erase contents */
3085 if ((file->f_mode & FMODE_WRITE) &&
3086 (file->f_flags & O_TRUNC)) {
3087 if (tc->cpu == RING_BUFFER_ALL_CPUS)
3088 tracing_reset_online_cpus(&tr->trace_buffer);
3090 tracing_reset(&tr->trace_buffer, tc->cpu);
3093 if (file->f_mode & FMODE_READ) {
3094 iter = __tracing_open(tr, tc, inode, file, false);
3096 ret = PTR_ERR(iter);
3097 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
3098 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3102 trace_array_put(tr);
3108 t_next(struct seq_file *m, void *v, loff_t *pos)
3110 struct tracer *t = v;
3120 static void *t_start(struct seq_file *m, loff_t *pos)
3125 mutex_lock(&trace_types_lock);
3126 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
3132 static void t_stop(struct seq_file *m, void *p)
3134 mutex_unlock(&trace_types_lock);
3137 static int t_show(struct seq_file *m, void *v)
3139 struct tracer *t = v;
3144 seq_printf(m, "%s", t->name);
3153 static const struct seq_operations show_traces_seq_ops = {
3160 static int show_traces_open(struct inode *inode, struct file *file)
3162 if (tracing_disabled)
3165 return seq_open(file, &show_traces_seq_ops);
3169 tracing_write_stub(struct file *filp, const char __user *ubuf,
3170 size_t count, loff_t *ppos)
3175 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
3177 if (file->f_mode & FMODE_READ)
3178 return seq_lseek(file, offset, origin);
3183 static const struct file_operations tracing_fops = {
3184 .open = tracing_open,
3186 .write = tracing_write_stub,
3187 .llseek = tracing_seek,
3188 .release = tracing_release,
3191 static const struct file_operations show_traces_fops = {
3192 .open = show_traces_open,
3194 .release = seq_release,
3195 .llseek = seq_lseek,
3199 * Only trace on a CPU if the bitmask is set:
3201 static cpumask_var_t tracing_cpumask;
3204 * The tracer itself will not take this lock, but still we want
3205 * to provide a consistent cpumask to user-space:
3207 static DEFINE_MUTEX(tracing_cpumask_update_lock);
3210 * Temporary storage for the character representation of the
3211 * CPU bitmask (and one more byte for the newline):
3213 static char mask_str[NR_CPUS + 1];
3216 tracing_cpumask_read(struct file *filp, char __user *ubuf,
3217 size_t count, loff_t *ppos)
3221 mutex_lock(&tracing_cpumask_update_lock);
3223 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
3224 if (count - len < 2) {
3228 len += sprintf(mask_str + len, "\n");
3229 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
3232 mutex_unlock(&tracing_cpumask_update_lock);
3238 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
3239 size_t count, loff_t *ppos)
3241 struct trace_array *tr = filp->private_data;
3242 cpumask_var_t tracing_cpumask_new;
3245 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
3248 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
3252 mutex_lock(&tracing_cpumask_update_lock);
3254 local_irq_disable();
3255 arch_spin_lock(&ftrace_max_lock);
3256 for_each_tracing_cpu(cpu) {
3258 * Increase/decrease the disabled counter if we are
3259 * about to flip a bit in the cpumask:
3261 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
3262 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3263 atomic_inc(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3264 ring_buffer_record_disable_cpu(tr->trace_buffer.buffer, cpu);
3266 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
3267 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3268 atomic_dec(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3269 ring_buffer_record_enable_cpu(tr->trace_buffer.buffer, cpu);
3272 arch_spin_unlock(&ftrace_max_lock);
3275 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
3277 mutex_unlock(&tracing_cpumask_update_lock);
3278 free_cpumask_var(tracing_cpumask_new);
3283 free_cpumask_var(tracing_cpumask_new);
3288 static const struct file_operations tracing_cpumask_fops = {
3289 .open = tracing_open_generic,
3290 .read = tracing_cpumask_read,
3291 .write = tracing_cpumask_write,
3292 .llseek = generic_file_llseek,
3295 static int tracing_trace_options_show(struct seq_file *m, void *v)
3297 struct tracer_opt *trace_opts;
3298 struct trace_array *tr = m->private;
3302 mutex_lock(&trace_types_lock);
3303 tracer_flags = tr->current_trace->flags->val;
3304 trace_opts = tr->current_trace->flags->opts;
3306 for (i = 0; trace_options[i]; i++) {
3307 if (trace_flags & (1 << i))
3308 seq_printf(m, "%s\n", trace_options[i]);
3310 seq_printf(m, "no%s\n", trace_options[i]);
3313 for (i = 0; trace_opts[i].name; i++) {
3314 if (tracer_flags & trace_opts[i].bit)
3315 seq_printf(m, "%s\n", trace_opts[i].name);
3317 seq_printf(m, "no%s\n", trace_opts[i].name);
3319 mutex_unlock(&trace_types_lock);
3324 static int __set_tracer_option(struct tracer *trace,
3325 struct tracer_flags *tracer_flags,
3326 struct tracer_opt *opts, int neg)
3330 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
3335 tracer_flags->val &= ~opts->bit;
3337 tracer_flags->val |= opts->bit;
3341 /* Try to assign a tracer specific option */
3342 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
3344 struct tracer_flags *tracer_flags = trace->flags;
3345 struct tracer_opt *opts = NULL;
3348 for (i = 0; tracer_flags->opts[i].name; i++) {
3349 opts = &tracer_flags->opts[i];
3351 if (strcmp(cmp, opts->name) == 0)
3352 return __set_tracer_option(trace, trace->flags,
3359 /* Some tracers require overwrite to stay enabled */
3360 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
3362 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
3368 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
3370 /* do nothing if flag is already set */
3371 if (!!(trace_flags & mask) == !!enabled)
3374 /* Give the tracer a chance to approve the change */
3375 if (tr->current_trace->flag_changed)
3376 if (tr->current_trace->flag_changed(tr->current_trace, mask, !!enabled))
3380 trace_flags |= mask;
3382 trace_flags &= ~mask;
3384 if (mask == TRACE_ITER_RECORD_CMD)
3385 trace_event_enable_cmd_record(enabled);
3387 if (mask == TRACE_ITER_OVERWRITE) {
3388 ring_buffer_change_overwrite(tr->trace_buffer.buffer, enabled);
3389 #ifdef CONFIG_TRACER_MAX_TRACE
3390 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
3394 if (mask == TRACE_ITER_PRINTK)
3395 trace_printk_start_stop_comm(enabled);
3400 static int trace_set_options(struct trace_array *tr, char *option)
3407 cmp = strstrip(option);
3409 if (strncmp(cmp, "no", 2) == 0) {
3414 mutex_lock(&trace_types_lock);
3416 for (i = 0; trace_options[i]; i++) {
3417 if (strcmp(cmp, trace_options[i]) == 0) {
3418 ret = set_tracer_flag(tr, 1 << i, !neg);
3423 /* If no option could be set, test the specific tracer options */
3424 if (!trace_options[i])
3425 ret = set_tracer_option(tr->current_trace, cmp, neg);
3427 mutex_unlock(&trace_types_lock);
3433 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
3434 size_t cnt, loff_t *ppos)
3436 struct seq_file *m = filp->private_data;
3437 struct trace_array *tr = m->private;
3441 if (cnt >= sizeof(buf))
3444 if (copy_from_user(&buf, ubuf, cnt))
3449 ret = trace_set_options(tr, buf);
3458 static int tracing_trace_options_open(struct inode *inode, struct file *file)
3460 struct trace_array *tr = inode->i_private;
3462 if (tracing_disabled)
3465 if (trace_array_get(tr) < 0)
3468 return single_open(file, tracing_trace_options_show, inode->i_private);
3471 static const struct file_operations tracing_iter_fops = {
3472 .open = tracing_trace_options_open,
3474 .llseek = seq_lseek,
3475 .release = tracing_single_release_tr,
3476 .write = tracing_trace_options_write,
3479 static const char readme_msg[] =
3480 "tracing mini-HOWTO:\n\n"
3481 "# echo 0 > tracing_on : quick way to disable tracing\n"
3482 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
3483 " Important files:\n"
3484 " trace\t\t\t- The static contents of the buffer\n"
3485 "\t\t\t To clear the buffer write into this file: echo > trace\n"
3486 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
3487 " current_tracer\t- function and latency tracers\n"
3488 " available_tracers\t- list of configured tracers for current_tracer\n"
3489 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
3490 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
3491 " trace_clock\t\t-change the clock used to order events\n"
3492 " local: Per cpu clock but may not be synced across CPUs\n"
3493 " global: Synced across CPUs but slows tracing down.\n"
3494 " counter: Not a clock, but just an increment\n"
3495 " uptime: Jiffy counter from time of boot\n"
3496 " perf: Same clock that perf events use\n"
3497 #ifdef CONFIG_X86_64
3498 " x86-tsc: TSC cycle counter\n"
3500 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
3501 " tracing_cpumask\t- Limit which CPUs to trace\n"
3502 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
3503 "\t\t\t Remove sub-buffer with rmdir\n"
3504 " trace_options\t\t- Set format or modify how tracing happens\n"
3505 "\t\t\t Disable an option by adding a suffix 'no' to the option name\n"
3506 #ifdef CONFIG_DYNAMIC_FTRACE
3507 "\n available_filter_functions - list of functions that can be filtered on\n"
3508 " set_ftrace_filter\t- echo function name in here to only trace these functions\n"
3509 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3510 " modules: Can select a group via module\n"
3511 " Format: :mod:<module-name>\n"
3512 " example: echo :mod:ext3 > set_ftrace_filter\n"
3513 " triggers: a command to perform when function is hit\n"
3514 " Format: <function>:<trigger>[:count]\n"
3515 " trigger: traceon, traceoff\n"
3516 " enable_event:<system>:<event>\n"
3517 " disable_event:<system>:<event>\n"
3518 #ifdef CONFIG_STACKTRACE
3521 #ifdef CONFIG_TRACER_SNAPSHOT
3524 " example: echo do_fault:traceoff > set_ftrace_filter\n"
3525 " echo do_trap:traceoff:3 > set_ftrace_filter\n"
3526 " The first one will disable tracing every time do_fault is hit\n"
3527 " The second will disable tracing at most 3 times when do_trap is hit\n"
3528 " The first time do trap is hit and it disables tracing, the counter\n"
3529 " will decrement to 2. If tracing is already disabled, the counter\n"
3530 " will not decrement. It only decrements when the trigger did work\n"
3531 " To remove trigger without count:\n"
3532 " echo '!<function>:<trigger> > set_ftrace_filter\n"
3533 " To remove trigger with a count:\n"
3534 " echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
3535 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
3536 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3537 " modules: Can select a group via module command :mod:\n"
3538 " Does not accept triggers\n"
3539 #endif /* CONFIG_DYNAMIC_FTRACE */
3540 #ifdef CONFIG_FUNCTION_TRACER
3541 " set_ftrace_pid\t- Write pid(s) to only function trace those pids (function)\n"
3543 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3544 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
3545 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
3547 #ifdef CONFIG_TRACER_SNAPSHOT
3548 "\n snapshot\t\t- Like 'trace' but shows the content of the static snapshot buffer\n"
3549 "\t\t\t Read the contents for more information\n"
3551 #ifdef CONFIG_STACKTRACE
3552 " stack_trace\t\t- Shows the max stack trace when active\n"
3553 " stack_max_size\t- Shows current max stack size that was traced\n"
3554 "\t\t\t Write into this file to reset the max size (trigger a new trace)\n"
3555 #ifdef CONFIG_DYNAMIC_FTRACE
3556 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace traces\n"
3558 #endif /* CONFIG_STACKTRACE */
3562 tracing_readme_read(struct file *filp, char __user *ubuf,
3563 size_t cnt, loff_t *ppos)
3565 return simple_read_from_buffer(ubuf, cnt, ppos,
3566 readme_msg, strlen(readme_msg));
3569 static const struct file_operations tracing_readme_fops = {
3570 .open = tracing_open_generic,
3571 .read = tracing_readme_read,
3572 .llseek = generic_file_llseek,
3576 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
3577 size_t cnt, loff_t *ppos)
3586 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
3590 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
3598 for (i = 0; i < SAVED_CMDLINES; i++) {
3601 pid = map_cmdline_to_pid[i];
3602 if (pid == -1 || pid == NO_CMDLINE_MAP)
3605 trace_find_cmdline(pid, buf_comm);
3606 r = sprintf(buf, "%d %s\n", pid, buf_comm);
3611 len = simple_read_from_buffer(ubuf, cnt, ppos,
3620 static const struct file_operations tracing_saved_cmdlines_fops = {
3621 .open = tracing_open_generic,
3622 .read = tracing_saved_cmdlines_read,
3623 .llseek = generic_file_llseek,
3627 tracing_set_trace_read(struct file *filp, char __user *ubuf,
3628 size_t cnt, loff_t *ppos)
3630 struct trace_array *tr = filp->private_data;
3631 char buf[MAX_TRACER_SIZE+2];
3634 mutex_lock(&trace_types_lock);
3635 r = sprintf(buf, "%s\n", tr->current_trace->name);
3636 mutex_unlock(&trace_types_lock);
3638 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3641 int tracer_init(struct tracer *t, struct trace_array *tr)
3643 tracing_reset_online_cpus(&tr->trace_buffer);
3647 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val)
3651 for_each_tracing_cpu(cpu)
3652 per_cpu_ptr(buf->data, cpu)->entries = val;
3655 #ifdef CONFIG_TRACER_MAX_TRACE
3656 /* resize @tr's buffer to the size of @size_tr's entries */
3657 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
3658 struct trace_buffer *size_buf, int cpu_id)
3662 if (cpu_id == RING_BUFFER_ALL_CPUS) {
3663 for_each_tracing_cpu(cpu) {
3664 ret = ring_buffer_resize(trace_buf->buffer,
3665 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
3668 per_cpu_ptr(trace_buf->data, cpu)->entries =
3669 per_cpu_ptr(size_buf->data, cpu)->entries;
3672 ret = ring_buffer_resize(trace_buf->buffer,
3673 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
3675 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
3676 per_cpu_ptr(size_buf->data, cpu_id)->entries;
3681 #endif /* CONFIG_TRACER_MAX_TRACE */
3683 static int __tracing_resize_ring_buffer(struct trace_array *tr,
3684 unsigned long size, int cpu)
3689 * If kernel or user changes the size of the ring buffer
3690 * we use the size that was given, and we can forget about
3691 * expanding it later.
3693 ring_buffer_expanded = true;
3695 /* May be called before buffers are initialized */
3696 if (!tr->trace_buffer.buffer)
3699 ret = ring_buffer_resize(tr->trace_buffer.buffer, size, cpu);
3703 #ifdef CONFIG_TRACER_MAX_TRACE
3704 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
3705 !tr->current_trace->use_max_tr)
3708 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
3710 int r = resize_buffer_duplicate_size(&tr->trace_buffer,
3711 &tr->trace_buffer, cpu);
3714 * AARGH! We are left with different
3715 * size max buffer!!!!
3716 * The max buffer is our "snapshot" buffer.
3717 * When a tracer needs a snapshot (one of the
3718 * latency tracers), it swaps the max buffer
3719 * with the saved snap shot. We succeeded to
3720 * update the size of the main buffer, but failed to
3721 * update the size of the max buffer. But when we tried
3722 * to reset the main buffer to the original size, we
3723 * failed there too. This is very unlikely to
3724 * happen, but if it does, warn and kill all
3728 tracing_disabled = 1;
3733 if (cpu == RING_BUFFER_ALL_CPUS)
3734 set_buffer_entries(&tr->max_buffer, size);
3736 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
3739 #endif /* CONFIG_TRACER_MAX_TRACE */
3741 if (cpu == RING_BUFFER_ALL_CPUS)
3742 set_buffer_entries(&tr->trace_buffer, size);
3744 per_cpu_ptr(tr->trace_buffer.data, cpu)->entries = size;
3749 static ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
3750 unsigned long size, int cpu_id)
3754 mutex_lock(&trace_types_lock);
3756 if (cpu_id != RING_BUFFER_ALL_CPUS) {
3757 /* make sure, this cpu is enabled in the mask */
3758 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
3764 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
3769 mutex_unlock(&trace_types_lock);
3776 * tracing_update_buffers - used by tracing facility to expand ring buffers
3778 * To save on memory when the tracing is never used on a system with it
3779 * configured in. The ring buffers are set to a minimum size. But once
3780 * a user starts to use the tracing facility, then they need to grow
3781 * to their default size.
3783 * This function is to be called when a tracer is about to be used.
3785 int tracing_update_buffers(void)
3789 mutex_lock(&trace_types_lock);
3790 if (!ring_buffer_expanded)
3791 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
3792 RING_BUFFER_ALL_CPUS);
3793 mutex_unlock(&trace_types_lock);
3798 struct trace_option_dentry;
3800 static struct trace_option_dentry *
3801 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
3804 destroy_trace_option_files(struct trace_option_dentry *topts);
3806 static int tracing_set_tracer(const char *buf)
3808 static struct trace_option_dentry *topts;
3809 struct trace_array *tr = &global_trace;
3811 #ifdef CONFIG_TRACER_MAX_TRACE
3816 mutex_lock(&trace_types_lock);
3818 if (!ring_buffer_expanded) {
3819 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
3820 RING_BUFFER_ALL_CPUS);
3826 for (t = trace_types; t; t = t->next) {
3827 if (strcmp(t->name, buf) == 0)
3834 if (t == tr->current_trace)
3837 trace_branch_disable();
3839 tr->current_trace->enabled = false;
3841 if (tr->current_trace->reset)
3842 tr->current_trace->reset(tr);
3844 /* Current trace needs to be nop_trace before synchronize_sched */
3845 tr->current_trace = &nop_trace;
3847 #ifdef CONFIG_TRACER_MAX_TRACE
3848 had_max_tr = tr->allocated_snapshot;
3850 if (had_max_tr && !t->use_max_tr) {
3852 * We need to make sure that the update_max_tr sees that
3853 * current_trace changed to nop_trace to keep it from
3854 * swapping the buffers after we resize it.
3855 * The update_max_tr is called from interrupts disabled
3856 * so a synchronized_sched() is sufficient.
3858 synchronize_sched();
3862 destroy_trace_option_files(topts);
3864 topts = create_trace_option_files(tr, t);
3866 #ifdef CONFIG_TRACER_MAX_TRACE
3867 if (t->use_max_tr && !had_max_tr) {
3868 ret = alloc_snapshot(tr);
3875 ret = tracer_init(t, tr);
3880 tr->current_trace = t;
3881 tr->current_trace->enabled = true;
3882 trace_branch_enable(tr);
3884 mutex_unlock(&trace_types_lock);
3890 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3891 size_t cnt, loff_t *ppos)
3893 char buf[MAX_TRACER_SIZE+1];
3900 if (cnt > MAX_TRACER_SIZE)
3901 cnt = MAX_TRACER_SIZE;
3903 if (copy_from_user(&buf, ubuf, cnt))
3908 /* strip ending whitespace. */
3909 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3912 err = tracing_set_tracer(buf);
3922 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3923 size_t cnt, loff_t *ppos)
3925 unsigned long *ptr = filp->private_data;
3929 r = snprintf(buf, sizeof(buf), "%ld\n",
3930 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3931 if (r > sizeof(buf))
3933 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3937 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3938 size_t cnt, loff_t *ppos)
3940 unsigned long *ptr = filp->private_data;
3944 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3953 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3955 struct trace_cpu *tc = inode->i_private;
3956 struct trace_array *tr = tc->tr;
3957 struct trace_iterator *iter;
3960 if (tracing_disabled)
3963 if (trace_array_get(tr) < 0)
3966 mutex_lock(&trace_types_lock);
3968 /* create a buffer to store the information to pass to userspace */
3969 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3976 * We make a copy of the current tracer to avoid concurrent
3977 * changes on it while we are reading.
3979 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3984 *iter->trace = *tr->current_trace;
3986 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3991 /* trace pipe does not show start of buffer */
3992 cpumask_setall(iter->started);
3994 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3995 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3997 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3998 if (trace_clocks[trace_clock_id].in_ns)
3999 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4001 iter->cpu_file = tc->cpu;
4003 iter->trace_buffer = &tc->tr->trace_buffer;
4004 mutex_init(&iter->mutex);
4005 filp->private_data = iter;
4007 if (iter->trace->pipe_open)
4008 iter->trace->pipe_open(iter);
4010 nonseekable_open(inode, filp);
4012 mutex_unlock(&trace_types_lock);
4018 __trace_array_put(tr);
4019 mutex_unlock(&trace_types_lock);
4023 static int tracing_release_pipe(struct inode *inode, struct file *file)
4025 struct trace_iterator *iter = file->private_data;
4026 struct trace_cpu *tc = inode->i_private;
4027 struct trace_array *tr = tc->tr;
4029 mutex_lock(&trace_types_lock);
4031 if (iter->trace->pipe_close)
4032 iter->trace->pipe_close(iter);
4034 mutex_unlock(&trace_types_lock);
4036 free_cpumask_var(iter->started);
4037 mutex_destroy(&iter->mutex);
4041 trace_array_put(tr);
4047 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
4049 /* Iterators are static, they should be filled or empty */
4050 if (trace_buffer_iter(iter, iter->cpu_file))
4051 return POLLIN | POLLRDNORM;
4053 if (trace_flags & TRACE_ITER_BLOCK)
4055 * Always select as readable when in blocking mode
4057 return POLLIN | POLLRDNORM;
4059 return ring_buffer_poll_wait(iter->trace_buffer->buffer, iter->cpu_file,
4064 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
4066 struct trace_iterator *iter = filp->private_data;
4068 return trace_poll(iter, filp, poll_table);
4072 * This is a make-shift waitqueue.
4073 * A tracer might use this callback on some rare cases:
4075 * 1) the current tracer might hold the runqueue lock when it wakes up
4076 * a reader, hence a deadlock (sched, function, and function graph tracers)
4077 * 2) the function tracers, trace all functions, we don't want
4078 * the overhead of calling wake_up and friends
4079 * (and tracing them too)
4081 * Anyway, this is really very primitive wakeup.
4083 void poll_wait_pipe(struct trace_iterator *iter)
4085 set_current_state(TASK_INTERRUPTIBLE);
4086 /* sleep for 100 msecs, and try again. */
4087 schedule_timeout(HZ / 10);
4090 /* Must be called with trace_types_lock mutex held. */
4091 static int tracing_wait_pipe(struct file *filp)
4093 struct trace_iterator *iter = filp->private_data;
4095 while (trace_empty(iter)) {
4097 if ((filp->f_flags & O_NONBLOCK)) {
4101 mutex_unlock(&iter->mutex);
4103 iter->trace->wait_pipe(iter);
4105 mutex_lock(&iter->mutex);
4107 if (signal_pending(current))
4111 * We block until we read something and tracing is disabled.
4112 * We still block if tracing is disabled, but we have never
4113 * read anything. This allows a user to cat this file, and
4114 * then enable tracing. But after we have read something,
4115 * we give an EOF when tracing is again disabled.
4117 * iter->pos will be 0 if we haven't read anything.
4119 if (!tracing_is_on() && iter->pos)
4130 tracing_read_pipe(struct file *filp, char __user *ubuf,
4131 size_t cnt, loff_t *ppos)
4133 struct trace_iterator *iter = filp->private_data;
4134 struct trace_array *tr = iter->tr;
4137 /* return any leftover data */
4138 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4142 trace_seq_init(&iter->seq);
4144 /* copy the tracer to avoid using a global lock all around */
4145 mutex_lock(&trace_types_lock);
4146 if (unlikely(iter->trace->name != tr->current_trace->name))
4147 *iter->trace = *tr->current_trace;
4148 mutex_unlock(&trace_types_lock);
4151 * Avoid more than one consumer on a single file descriptor
4152 * This is just a matter of traces coherency, the ring buffer itself
4155 mutex_lock(&iter->mutex);
4156 if (iter->trace->read) {
4157 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
4163 sret = tracing_wait_pipe(filp);
4167 /* stop when tracing is finished */
4168 if (trace_empty(iter)) {
4173 if (cnt >= PAGE_SIZE)
4174 cnt = PAGE_SIZE - 1;
4176 /* reset all but tr, trace, and overruns */
4177 memset(&iter->seq, 0,
4178 sizeof(struct trace_iterator) -
4179 offsetof(struct trace_iterator, seq));
4182 trace_event_read_lock();
4183 trace_access_lock(iter->cpu_file);
4184 while (trace_find_next_entry_inc(iter) != NULL) {
4185 enum print_line_t ret;
4186 int len = iter->seq.len;
4188 ret = print_trace_line(iter);
4189 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4190 /* don't print partial lines */
4191 iter->seq.len = len;
4194 if (ret != TRACE_TYPE_NO_CONSUME)
4195 trace_consume(iter);
4197 if (iter->seq.len >= cnt)
4201 * Setting the full flag means we reached the trace_seq buffer
4202 * size and we should leave by partial output condition above.
4203 * One of the trace_seq_* functions is not used properly.
4205 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
4208 trace_access_unlock(iter->cpu_file);
4209 trace_event_read_unlock();
4211 /* Now copy what we have to the user */
4212 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4213 if (iter->seq.readpos >= iter->seq.len)
4214 trace_seq_init(&iter->seq);
4217 * If there was nothing to send to user, in spite of consuming trace
4218 * entries, go back to wait for more entries.
4224 mutex_unlock(&iter->mutex);
4229 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
4230 struct pipe_buffer *buf)
4232 __free_page(buf->page);
4235 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
4238 __free_page(spd->pages[idx]);
4241 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
4243 .map = generic_pipe_buf_map,
4244 .unmap = generic_pipe_buf_unmap,
4245 .confirm = generic_pipe_buf_confirm,
4246 .release = tracing_pipe_buf_release,
4247 .steal = generic_pipe_buf_steal,
4248 .get = generic_pipe_buf_get,
4252 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
4257 /* Seq buffer is page-sized, exactly what we need. */
4259 count = iter->seq.len;
4260 ret = print_trace_line(iter);
4261 count = iter->seq.len - count;
4264 iter->seq.len -= count;
4267 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4268 iter->seq.len -= count;
4272 if (ret != TRACE_TYPE_NO_CONSUME)
4273 trace_consume(iter);
4275 if (!trace_find_next_entry_inc(iter)) {
4285 static ssize_t tracing_splice_read_pipe(struct file *filp,
4287 struct pipe_inode_info *pipe,
4291 struct page *pages_def[PIPE_DEF_BUFFERS];
4292 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4293 struct trace_iterator *iter = filp->private_data;
4294 struct splice_pipe_desc spd = {
4296 .partial = partial_def,
4297 .nr_pages = 0, /* This gets updated below. */
4298 .nr_pages_max = PIPE_DEF_BUFFERS,
4300 .ops = &tracing_pipe_buf_ops,
4301 .spd_release = tracing_spd_release_pipe,
4303 struct trace_array *tr = iter->tr;
4308 if (splice_grow_spd(pipe, &spd))
4311 /* copy the tracer to avoid using a global lock all around */
4312 mutex_lock(&trace_types_lock);
4313 if (unlikely(iter->trace->name != tr->current_trace->name))
4314 *iter->trace = *tr->current_trace;
4315 mutex_unlock(&trace_types_lock);
4317 mutex_lock(&iter->mutex);
4319 if (iter->trace->splice_read) {
4320 ret = iter->trace->splice_read(iter, filp,
4321 ppos, pipe, len, flags);
4326 ret = tracing_wait_pipe(filp);
4330 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
4335 trace_event_read_lock();
4336 trace_access_lock(iter->cpu_file);
4338 /* Fill as many pages as possible. */
4339 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
4340 spd.pages[i] = alloc_page(GFP_KERNEL);
4344 rem = tracing_fill_pipe_page(rem, iter);
4346 /* Copy the data into the page, so we can start over. */
4347 ret = trace_seq_to_buffer(&iter->seq,
4348 page_address(spd.pages[i]),
4351 __free_page(spd.pages[i]);
4354 spd.partial[i].offset = 0;
4355 spd.partial[i].len = iter->seq.len;
4357 trace_seq_init(&iter->seq);
4360 trace_access_unlock(iter->cpu_file);
4361 trace_event_read_unlock();
4362 mutex_unlock(&iter->mutex);
4366 ret = splice_to_pipe(pipe, &spd);
4368 splice_shrink_spd(&spd);
4372 mutex_unlock(&iter->mutex);
4377 tracing_entries_read(struct file *filp, char __user *ubuf,
4378 size_t cnt, loff_t *ppos)
4380 struct trace_cpu *tc = filp->private_data;
4381 struct trace_array *tr = tc->tr;
4386 mutex_lock(&trace_types_lock);
4388 if (tc->cpu == RING_BUFFER_ALL_CPUS) {
4389 int cpu, buf_size_same;
4394 /* check if all cpu sizes are same */
4395 for_each_tracing_cpu(cpu) {
4396 /* fill in the size from first enabled cpu */
4398 size = per_cpu_ptr(tr->trace_buffer.data, cpu)->entries;
4399 if (size != per_cpu_ptr(tr->trace_buffer.data, cpu)->entries) {
4405 if (buf_size_same) {
4406 if (!ring_buffer_expanded)
4407 r = sprintf(buf, "%lu (expanded: %lu)\n",
4409 trace_buf_size >> 10);
4411 r = sprintf(buf, "%lu\n", size >> 10);
4413 r = sprintf(buf, "X\n");
4415 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->trace_buffer.data, tc->cpu)->entries >> 10);
4417 mutex_unlock(&trace_types_lock);
4419 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4424 tracing_entries_write(struct file *filp, const char __user *ubuf,
4425 size_t cnt, loff_t *ppos)
4427 struct trace_cpu *tc = filp->private_data;
4431 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4435 /* must have at least 1 entry */
4439 /* value is in KB */
4442 ret = tracing_resize_ring_buffer(tc->tr, val, tc->cpu);
4452 tracing_total_entries_read(struct file *filp, char __user *ubuf,
4453 size_t cnt, loff_t *ppos)
4455 struct trace_array *tr = filp->private_data;
4458 unsigned long size = 0, expanded_size = 0;
4460 mutex_lock(&trace_types_lock);
4461 for_each_tracing_cpu(cpu) {
4462 size += per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10;
4463 if (!ring_buffer_expanded)
4464 expanded_size += trace_buf_size >> 10;
4466 if (ring_buffer_expanded)
4467 r = sprintf(buf, "%lu\n", size);
4469 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
4470 mutex_unlock(&trace_types_lock);
4472 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4476 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
4477 size_t cnt, loff_t *ppos)
4480 * There is no need to read what the user has written, this function
4481 * is just to make sure that there is no error when "echo" is used
4490 tracing_free_buffer_release(struct inode *inode, struct file *filp)
4492 struct trace_array *tr = inode->i_private;
4494 /* disable tracing ? */
4495 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
4497 /* resize the ring buffer to 0 */
4498 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
4500 trace_array_put(tr);
4506 tracing_mark_write(struct file *filp, const char __user *ubuf,
4507 size_t cnt, loff_t *fpos)
4509 unsigned long addr = (unsigned long)ubuf;
4510 struct trace_array *tr = filp->private_data;
4511 struct ring_buffer_event *event;
4512 struct ring_buffer *buffer;
4513 struct print_entry *entry;
4514 unsigned long irq_flags;
4515 struct page *pages[2];
4525 if (tracing_disabled)
4528 if (!(trace_flags & TRACE_ITER_MARKERS))
4531 if (cnt > TRACE_BUF_SIZE)
4532 cnt = TRACE_BUF_SIZE;
4535 * Userspace is injecting traces into the kernel trace buffer.
4536 * We want to be as non intrusive as possible.
4537 * To do so, we do not want to allocate any special buffers
4538 * or take any locks, but instead write the userspace data
4539 * straight into the ring buffer.
4541 * First we need to pin the userspace buffer into memory,
4542 * which, most likely it is, because it just referenced it.
4543 * But there's no guarantee that it is. By using get_user_pages_fast()
4544 * and kmap_atomic/kunmap_atomic() we can get access to the
4545 * pages directly. We then write the data directly into the
4548 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
4550 /* check if we cross pages */
4551 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
4554 offset = addr & (PAGE_SIZE - 1);
4557 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
4558 if (ret < nr_pages) {
4560 put_page(pages[ret]);
4565 for (i = 0; i < nr_pages; i++)
4566 map_page[i] = kmap_atomic(pages[i]);
4568 local_save_flags(irq_flags);
4569 size = sizeof(*entry) + cnt + 2; /* possible \n added */
4570 buffer = tr->trace_buffer.buffer;
4571 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
4572 irq_flags, preempt_count());
4574 /* Ring buffer disabled, return as if not open for write */
4579 entry = ring_buffer_event_data(event);
4580 entry->ip = _THIS_IP_;
4582 if (nr_pages == 2) {
4583 len = PAGE_SIZE - offset;
4584 memcpy(&entry->buf, map_page[0] + offset, len);
4585 memcpy(&entry->buf[len], map_page[1], cnt - len);
4587 memcpy(&entry->buf, map_page[0] + offset, cnt);
4589 if (entry->buf[cnt - 1] != '\n') {
4590 entry->buf[cnt] = '\n';
4591 entry->buf[cnt + 1] = '\0';
4593 entry->buf[cnt] = '\0';
4595 __buffer_unlock_commit(buffer, event);
4602 for (i = 0; i < nr_pages; i++){
4603 kunmap_atomic(map_page[i]);
4610 static int tracing_clock_show(struct seq_file *m, void *v)
4612 struct trace_array *tr = m->private;
4615 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
4617 "%s%s%s%s", i ? " " : "",
4618 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
4619 i == tr->clock_id ? "]" : "");
4625 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
4626 size_t cnt, loff_t *fpos)
4628 struct seq_file *m = filp->private_data;
4629 struct trace_array *tr = m->private;
4631 const char *clockstr;
4634 if (cnt >= sizeof(buf))
4637 if (copy_from_user(&buf, ubuf, cnt))
4642 clockstr = strstrip(buf);
4644 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
4645 if (strcmp(trace_clocks[i].name, clockstr) == 0)
4648 if (i == ARRAY_SIZE(trace_clocks))
4651 mutex_lock(&trace_types_lock);
4655 ring_buffer_set_clock(tr->trace_buffer.buffer, trace_clocks[i].func);
4658 * New clock may not be consistent with the previous clock.
4659 * Reset the buffer so that it doesn't have incomparable timestamps.
4661 tracing_reset_online_cpus(&global_trace.trace_buffer);
4663 #ifdef CONFIG_TRACER_MAX_TRACE
4664 if (tr->flags & TRACE_ARRAY_FL_GLOBAL && tr->max_buffer.buffer)
4665 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
4666 tracing_reset_online_cpus(&global_trace.max_buffer);
4669 mutex_unlock(&trace_types_lock);
4676 static int tracing_clock_open(struct inode *inode, struct file *file)
4678 struct trace_array *tr = inode->i_private;
4681 if (tracing_disabled)
4684 if (trace_array_get(tr))
4687 ret = single_open(file, tracing_clock_show, inode->i_private);
4689 trace_array_put(tr);
4694 struct ftrace_buffer_info {
4695 struct trace_iterator iter;
4700 #ifdef CONFIG_TRACER_SNAPSHOT
4701 static int tracing_snapshot_open(struct inode *inode, struct file *file)
4703 struct trace_cpu *tc = inode->i_private;
4704 struct trace_array *tr = tc->tr;
4705 struct trace_iterator *iter;
4709 if (trace_array_get(tr) < 0)
4712 if (file->f_mode & FMODE_READ) {
4713 iter = __tracing_open(tr, tc, inode, file, true);
4715 ret = PTR_ERR(iter);
4717 /* Writes still need the seq_file to hold the private data */
4718 m = kzalloc(sizeof(*m), GFP_KERNEL);
4721 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4727 iter->trace_buffer = &tc->tr->max_buffer;
4728 iter->cpu_file = tc->cpu;
4730 file->private_data = m;
4734 trace_array_put(tr);
4740 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
4743 struct seq_file *m = filp->private_data;
4744 struct trace_iterator *iter = m->private;
4745 struct trace_array *tr = iter->tr;
4749 ret = tracing_update_buffers();
4753 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4757 mutex_lock(&trace_types_lock);
4759 if (tr->current_trace->use_max_tr) {
4766 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4770 if (tr->allocated_snapshot)
4774 /* Only allow per-cpu swap if the ring buffer supports it */
4775 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
4776 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4781 if (!tr->allocated_snapshot) {
4782 ret = alloc_snapshot(tr);
4786 local_irq_disable();
4787 /* Now, we're going to swap */
4788 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4789 update_max_tr(tr, current, smp_processor_id());
4791 update_max_tr_single(tr, current, iter->cpu_file);
4795 if (tr->allocated_snapshot) {
4796 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4797 tracing_reset_online_cpus(&tr->max_buffer);
4799 tracing_reset(&tr->max_buffer, iter->cpu_file);
4809 mutex_unlock(&trace_types_lock);
4813 static int tracing_snapshot_release(struct inode *inode, struct file *file)
4815 struct seq_file *m = file->private_data;
4818 ret = tracing_release(inode, file);
4820 if (file->f_mode & FMODE_READ)
4823 /* If write only, the seq_file is just a stub */
4831 static int tracing_buffers_open(struct inode *inode, struct file *filp);
4832 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
4833 size_t count, loff_t *ppos);
4834 static int tracing_buffers_release(struct inode *inode, struct file *file);
4835 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4836 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
4838 static int snapshot_raw_open(struct inode *inode, struct file *filp)
4840 struct ftrace_buffer_info *info;
4843 ret = tracing_buffers_open(inode, filp);
4847 info = filp->private_data;
4849 if (info->iter.trace->use_max_tr) {
4850 tracing_buffers_release(inode, filp);
4854 info->iter.snapshot = true;
4855 info->iter.trace_buffer = &info->iter.tr->max_buffer;
4860 #endif /* CONFIG_TRACER_SNAPSHOT */
4863 static const struct file_operations tracing_max_lat_fops = {
4864 .open = tracing_open_generic,
4865 .read = tracing_max_lat_read,
4866 .write = tracing_max_lat_write,
4867 .llseek = generic_file_llseek,
4870 static const struct file_operations set_tracer_fops = {
4871 .open = tracing_open_generic,
4872 .read = tracing_set_trace_read,
4873 .write = tracing_set_trace_write,
4874 .llseek = generic_file_llseek,
4877 static const struct file_operations tracing_pipe_fops = {
4878 .open = tracing_open_pipe,
4879 .poll = tracing_poll_pipe,
4880 .read = tracing_read_pipe,
4881 .splice_read = tracing_splice_read_pipe,
4882 .release = tracing_release_pipe,
4883 .llseek = no_llseek,
4886 static const struct file_operations tracing_entries_fops = {
4887 .open = tracing_open_generic_tc,
4888 .read = tracing_entries_read,
4889 .write = tracing_entries_write,
4890 .llseek = generic_file_llseek,
4891 .release = tracing_release_generic_tc,
4894 static const struct file_operations tracing_total_entries_fops = {
4895 .open = tracing_open_generic_tr,
4896 .read = tracing_total_entries_read,
4897 .llseek = generic_file_llseek,
4898 .release = tracing_release_generic_tr,
4901 static const struct file_operations tracing_free_buffer_fops = {
4902 .open = tracing_open_generic_tr,
4903 .write = tracing_free_buffer_write,
4904 .release = tracing_free_buffer_release,
4907 static const struct file_operations tracing_mark_fops = {
4908 .open = tracing_open_generic_tr,
4909 .write = tracing_mark_write,
4910 .llseek = generic_file_llseek,
4911 .release = tracing_release_generic_tr,
4914 static const struct file_operations trace_clock_fops = {
4915 .open = tracing_clock_open,
4917 .llseek = seq_lseek,
4918 .release = tracing_single_release_tr,
4919 .write = tracing_clock_write,
4922 #ifdef CONFIG_TRACER_SNAPSHOT
4923 static const struct file_operations snapshot_fops = {
4924 .open = tracing_snapshot_open,
4926 .write = tracing_snapshot_write,
4927 .llseek = tracing_seek,
4928 .release = tracing_snapshot_release,
4931 static const struct file_operations snapshot_raw_fops = {
4932 .open = snapshot_raw_open,
4933 .read = tracing_buffers_read,
4934 .release = tracing_buffers_release,
4935 .splice_read = tracing_buffers_splice_read,
4936 .llseek = no_llseek,
4939 #endif /* CONFIG_TRACER_SNAPSHOT */
4941 static int tracing_buffers_open(struct inode *inode, struct file *filp)
4943 struct trace_cpu *tc = inode->i_private;
4944 struct trace_array *tr = tc->tr;
4945 struct ftrace_buffer_info *info;
4948 if (tracing_disabled)
4951 if (trace_array_get(tr) < 0)
4954 info = kzalloc(sizeof(*info), GFP_KERNEL);
4956 trace_array_put(tr);
4960 mutex_lock(&trace_types_lock);
4965 info->iter.cpu_file = tc->cpu;
4966 info->iter.trace = tr->current_trace;
4967 info->iter.trace_buffer = &tr->trace_buffer;
4969 /* Force reading ring buffer for first read */
4970 info->read = (unsigned int)-1;
4972 filp->private_data = info;
4974 mutex_unlock(&trace_types_lock);
4976 ret = nonseekable_open(inode, filp);
4978 trace_array_put(tr);
4984 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
4986 struct ftrace_buffer_info *info = filp->private_data;
4987 struct trace_iterator *iter = &info->iter;
4989 return trace_poll(iter, filp, poll_table);
4993 tracing_buffers_read(struct file *filp, char __user *ubuf,
4994 size_t count, loff_t *ppos)
4996 struct ftrace_buffer_info *info = filp->private_data;
4997 struct trace_iterator *iter = &info->iter;
5004 mutex_lock(&trace_types_lock);
5006 #ifdef CONFIG_TRACER_MAX_TRACE
5007 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
5014 info->spare = ring_buffer_alloc_read_page(iter->trace_buffer->buffer,
5020 /* Do we have previous read data to read? */
5021 if (info->read < PAGE_SIZE)
5025 trace_access_lock(iter->cpu_file);
5026 ret = ring_buffer_read_page(iter->trace_buffer->buffer,
5030 trace_access_unlock(iter->cpu_file);
5033 if (trace_empty(iter)) {
5034 if ((filp->f_flags & O_NONBLOCK)) {
5038 mutex_unlock(&trace_types_lock);
5039 iter->trace->wait_pipe(iter);
5040 mutex_lock(&trace_types_lock);
5041 if (signal_pending(current)) {
5053 size = PAGE_SIZE - info->read;
5057 ret = copy_to_user(ubuf, info->spare + info->read, size);
5068 mutex_unlock(&trace_types_lock);
5073 static int tracing_buffers_release(struct inode *inode, struct file *file)
5075 struct ftrace_buffer_info *info = file->private_data;
5076 struct trace_iterator *iter = &info->iter;
5078 mutex_lock(&trace_types_lock);
5080 __trace_array_put(iter->tr);
5083 ring_buffer_free_read_page(iter->trace_buffer->buffer, info->spare);
5086 mutex_unlock(&trace_types_lock);
5092 struct ring_buffer *buffer;
5097 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
5098 struct pipe_buffer *buf)
5100 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5105 ring_buffer_free_read_page(ref->buffer, ref->page);
5110 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
5111 struct pipe_buffer *buf)
5113 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5118 /* Pipe buffer operations for a buffer. */
5119 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
5121 .map = generic_pipe_buf_map,
5122 .unmap = generic_pipe_buf_unmap,
5123 .confirm = generic_pipe_buf_confirm,
5124 .release = buffer_pipe_buf_release,
5125 .steal = generic_pipe_buf_steal,
5126 .get = buffer_pipe_buf_get,
5130 * Callback from splice_to_pipe(), if we need to release some pages
5131 * at the end of the spd in case we error'ed out in filling the pipe.
5133 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
5135 struct buffer_ref *ref =
5136 (struct buffer_ref *)spd->partial[i].private;
5141 ring_buffer_free_read_page(ref->buffer, ref->page);
5143 spd->partial[i].private = 0;
5147 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
5148 struct pipe_inode_info *pipe, size_t len,
5151 struct ftrace_buffer_info *info = file->private_data;
5152 struct trace_iterator *iter = &info->iter;
5153 struct partial_page partial_def[PIPE_DEF_BUFFERS];
5154 struct page *pages_def[PIPE_DEF_BUFFERS];
5155 struct splice_pipe_desc spd = {
5157 .partial = partial_def,
5158 .nr_pages_max = PIPE_DEF_BUFFERS,
5160 .ops = &buffer_pipe_buf_ops,
5161 .spd_release = buffer_spd_release,
5163 struct buffer_ref *ref;
5164 int entries, size, i;
5167 mutex_lock(&trace_types_lock);
5169 #ifdef CONFIG_TRACER_MAX_TRACE
5170 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
5176 if (splice_grow_spd(pipe, &spd)) {
5181 if (*ppos & (PAGE_SIZE - 1)) {
5186 if (len & (PAGE_SIZE - 1)) {
5187 if (len < PAGE_SIZE) {
5195 trace_access_lock(iter->cpu_file);
5196 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5198 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
5202 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
5207 ref->buffer = iter->trace_buffer->buffer;
5208 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
5214 r = ring_buffer_read_page(ref->buffer, &ref->page,
5215 len, iter->cpu_file, 1);
5217 ring_buffer_free_read_page(ref->buffer, ref->page);
5223 * zero out any left over data, this is going to
5226 size = ring_buffer_page_len(ref->page);
5227 if (size < PAGE_SIZE)
5228 memset(ref->page + size, 0, PAGE_SIZE - size);
5230 page = virt_to_page(ref->page);
5232 spd.pages[i] = page;
5233 spd.partial[i].len = PAGE_SIZE;
5234 spd.partial[i].offset = 0;
5235 spd.partial[i].private = (unsigned long)ref;
5239 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5242 trace_access_unlock(iter->cpu_file);
5245 /* did we read anything? */
5246 if (!spd.nr_pages) {
5247 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK)) {
5251 mutex_unlock(&trace_types_lock);
5252 iter->trace->wait_pipe(iter);
5253 mutex_lock(&trace_types_lock);
5254 if (signal_pending(current)) {
5261 ret = splice_to_pipe(pipe, &spd);
5262 splice_shrink_spd(&spd);
5264 mutex_unlock(&trace_types_lock);
5269 static const struct file_operations tracing_buffers_fops = {
5270 .open = tracing_buffers_open,
5271 .read = tracing_buffers_read,
5272 .poll = tracing_buffers_poll,
5273 .release = tracing_buffers_release,
5274 .splice_read = tracing_buffers_splice_read,
5275 .llseek = no_llseek,
5279 tracing_stats_read(struct file *filp, char __user *ubuf,
5280 size_t count, loff_t *ppos)
5282 struct trace_cpu *tc = filp->private_data;
5283 struct trace_array *tr = tc->tr;
5284 struct trace_buffer *trace_buf = &tr->trace_buffer;
5285 struct trace_seq *s;
5287 unsigned long long t;
5288 unsigned long usec_rem;
5291 s = kmalloc(sizeof(*s), GFP_KERNEL);
5297 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
5298 trace_seq_printf(s, "entries: %ld\n", cnt);
5300 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
5301 trace_seq_printf(s, "overrun: %ld\n", cnt);
5303 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
5304 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
5306 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
5307 trace_seq_printf(s, "bytes: %ld\n", cnt);
5309 if (trace_clocks[trace_clock_id].in_ns) {
5310 /* local or global for trace_clock */
5311 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5312 usec_rem = do_div(t, USEC_PER_SEC);
5313 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
5316 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
5317 usec_rem = do_div(t, USEC_PER_SEC);
5318 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
5320 /* counter or tsc mode for trace_clock */
5321 trace_seq_printf(s, "oldest event ts: %llu\n",
5322 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5324 trace_seq_printf(s, "now ts: %llu\n",
5325 ring_buffer_time_stamp(trace_buf->buffer, cpu));
5328 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
5329 trace_seq_printf(s, "dropped events: %ld\n", cnt);
5331 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
5332 trace_seq_printf(s, "read events: %ld\n", cnt);
5334 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
5341 static const struct file_operations tracing_stats_fops = {
5342 .open = tracing_open_generic,
5343 .read = tracing_stats_read,
5344 .llseek = generic_file_llseek,
5347 #ifdef CONFIG_DYNAMIC_FTRACE
5349 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
5355 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
5356 size_t cnt, loff_t *ppos)
5358 static char ftrace_dyn_info_buffer[1024];
5359 static DEFINE_MUTEX(dyn_info_mutex);
5360 unsigned long *p = filp->private_data;
5361 char *buf = ftrace_dyn_info_buffer;
5362 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
5365 mutex_lock(&dyn_info_mutex);
5366 r = sprintf(buf, "%ld ", *p);
5368 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
5371 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5373 mutex_unlock(&dyn_info_mutex);
5378 static const struct file_operations tracing_dyn_info_fops = {
5379 .open = tracing_open_generic,
5380 .read = tracing_read_dyn_info,
5381 .llseek = generic_file_llseek,
5383 #endif /* CONFIG_DYNAMIC_FTRACE */
5385 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
5387 ftrace_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5393 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5395 unsigned long *count = (long *)data;
5407 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
5408 struct ftrace_probe_ops *ops, void *data)
5410 long count = (long)data;
5412 seq_printf(m, "%ps:", (void *)ip);
5414 seq_printf(m, "snapshot");
5417 seq_printf(m, ":unlimited\n");
5419 seq_printf(m, ":count=%ld\n", count);
5424 static struct ftrace_probe_ops snapshot_probe_ops = {
5425 .func = ftrace_snapshot,
5426 .print = ftrace_snapshot_print,
5429 static struct ftrace_probe_ops snapshot_count_probe_ops = {
5430 .func = ftrace_count_snapshot,
5431 .print = ftrace_snapshot_print,
5435 ftrace_trace_snapshot_callback(struct ftrace_hash *hash,
5436 char *glob, char *cmd, char *param, int enable)
5438 struct ftrace_probe_ops *ops;
5439 void *count = (void *)-1;
5443 /* hash funcs only work with set_ftrace_filter */
5447 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
5449 if (glob[0] == '!') {
5450 unregister_ftrace_function_probe_func(glob+1, ops);
5457 number = strsep(¶m, ":");
5459 if (!strlen(number))
5463 * We use the callback data field (which is a pointer)
5466 ret = kstrtoul(number, 0, (unsigned long *)&count);
5471 ret = register_ftrace_function_probe(glob, ops, count);
5474 alloc_snapshot(&global_trace);
5476 return ret < 0 ? ret : 0;
5479 static struct ftrace_func_command ftrace_snapshot_cmd = {
5481 .func = ftrace_trace_snapshot_callback,
5484 static int register_snapshot_cmd(void)
5486 return register_ftrace_command(&ftrace_snapshot_cmd);
5489 static inline int register_snapshot_cmd(void) { return 0; }
5490 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
5492 struct dentry *tracing_init_dentry_tr(struct trace_array *tr)
5497 if (!debugfs_initialized())
5500 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
5501 tr->dir = debugfs_create_dir("tracing", NULL);
5504 pr_warn_once("Could not create debugfs directory 'tracing'\n");
5509 struct dentry *tracing_init_dentry(void)
5511 return tracing_init_dentry_tr(&global_trace);
5514 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
5516 struct dentry *d_tracer;
5519 return tr->percpu_dir;
5521 d_tracer = tracing_init_dentry_tr(tr);
5525 tr->percpu_dir = debugfs_create_dir("per_cpu", d_tracer);
5527 WARN_ONCE(!tr->percpu_dir,
5528 "Could not create debugfs directory 'per_cpu/%d'\n", cpu);
5530 return tr->percpu_dir;
5534 tracing_init_debugfs_percpu(struct trace_array *tr, long cpu)
5536 struct trace_array_cpu *data = per_cpu_ptr(tr->trace_buffer.data, cpu);
5537 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
5538 struct dentry *d_cpu;
5539 char cpu_dir[30]; /* 30 characters should be more than enough */
5544 snprintf(cpu_dir, 30, "cpu%ld", cpu);
5545 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
5547 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
5551 /* per cpu trace_pipe */
5552 trace_create_file("trace_pipe", 0444, d_cpu,
5553 (void *)&data->trace_cpu, &tracing_pipe_fops);
5556 trace_create_file("trace", 0644, d_cpu,
5557 (void *)&data->trace_cpu, &tracing_fops);
5559 trace_create_file("trace_pipe_raw", 0444, d_cpu,
5560 (void *)&data->trace_cpu, &tracing_buffers_fops);
5562 trace_create_file("stats", 0444, d_cpu,
5563 (void *)&data->trace_cpu, &tracing_stats_fops);
5565 trace_create_file("buffer_size_kb", 0444, d_cpu,
5566 (void *)&data->trace_cpu, &tracing_entries_fops);
5568 #ifdef CONFIG_TRACER_SNAPSHOT
5569 trace_create_file("snapshot", 0644, d_cpu,
5570 (void *)&data->trace_cpu, &snapshot_fops);
5572 trace_create_file("snapshot_raw", 0444, d_cpu,
5573 (void *)&data->trace_cpu, &snapshot_raw_fops);
5577 #ifdef CONFIG_FTRACE_SELFTEST
5578 /* Let selftest have access to static functions in this file */
5579 #include "trace_selftest.c"
5582 struct trace_option_dentry {
5583 struct tracer_opt *opt;
5584 struct tracer_flags *flags;
5585 struct trace_array *tr;
5586 struct dentry *entry;
5590 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
5593 struct trace_option_dentry *topt = filp->private_data;
5596 if (topt->flags->val & topt->opt->bit)
5601 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5605 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
5608 struct trace_option_dentry *topt = filp->private_data;
5612 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5616 if (val != 0 && val != 1)
5619 if (!!(topt->flags->val & topt->opt->bit) != val) {
5620 mutex_lock(&trace_types_lock);
5621 ret = __set_tracer_option(topt->tr->current_trace, topt->flags,
5623 mutex_unlock(&trace_types_lock);
5634 static const struct file_operations trace_options_fops = {
5635 .open = tracing_open_generic,
5636 .read = trace_options_read,
5637 .write = trace_options_write,
5638 .llseek = generic_file_llseek,
5642 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
5645 long index = (long)filp->private_data;
5648 if (trace_flags & (1 << index))
5653 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5657 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
5660 struct trace_array *tr = &global_trace;
5661 long index = (long)filp->private_data;
5665 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5669 if (val != 0 && val != 1)
5672 mutex_lock(&trace_types_lock);
5673 ret = set_tracer_flag(tr, 1 << index, val);
5674 mutex_unlock(&trace_types_lock);
5684 static const struct file_operations trace_options_core_fops = {
5685 .open = tracing_open_generic,
5686 .read = trace_options_core_read,
5687 .write = trace_options_core_write,
5688 .llseek = generic_file_llseek,
5691 struct dentry *trace_create_file(const char *name,
5693 struct dentry *parent,
5695 const struct file_operations *fops)
5699 ret = debugfs_create_file(name, mode, parent, data, fops);
5701 pr_warning("Could not create debugfs '%s' entry\n", name);
5707 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
5709 struct dentry *d_tracer;
5714 d_tracer = tracing_init_dentry_tr(tr);
5718 tr->options = debugfs_create_dir("options", d_tracer);
5720 pr_warning("Could not create debugfs directory 'options'\n");
5728 create_trace_option_file(struct trace_array *tr,
5729 struct trace_option_dentry *topt,
5730 struct tracer_flags *flags,
5731 struct tracer_opt *opt)
5733 struct dentry *t_options;
5735 t_options = trace_options_init_dentry(tr);
5739 topt->flags = flags;
5743 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
5744 &trace_options_fops);
5748 static struct trace_option_dentry *
5749 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
5751 struct trace_option_dentry *topts;
5752 struct tracer_flags *flags;
5753 struct tracer_opt *opts;
5759 flags = tracer->flags;
5761 if (!flags || !flags->opts)
5766 for (cnt = 0; opts[cnt].name; cnt++)
5769 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
5773 for (cnt = 0; opts[cnt].name; cnt++)
5774 create_trace_option_file(tr, &topts[cnt], flags,
5781 destroy_trace_option_files(struct trace_option_dentry *topts)
5788 for (cnt = 0; topts[cnt].opt; cnt++) {
5789 if (topts[cnt].entry)
5790 debugfs_remove(topts[cnt].entry);
5796 static struct dentry *
5797 create_trace_option_core_file(struct trace_array *tr,
5798 const char *option, long index)
5800 struct dentry *t_options;
5802 t_options = trace_options_init_dentry(tr);
5806 return trace_create_file(option, 0644, t_options, (void *)index,
5807 &trace_options_core_fops);
5810 static __init void create_trace_options_dir(struct trace_array *tr)
5812 struct dentry *t_options;
5815 t_options = trace_options_init_dentry(tr);
5819 for (i = 0; trace_options[i]; i++)
5820 create_trace_option_core_file(tr, trace_options[i], i);
5824 rb_simple_read(struct file *filp, char __user *ubuf,
5825 size_t cnt, loff_t *ppos)
5827 struct trace_array *tr = filp->private_data;
5831 r = tracer_tracing_is_on(tr);
5832 r = sprintf(buf, "%d\n", r);
5834 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5838 rb_simple_write(struct file *filp, const char __user *ubuf,
5839 size_t cnt, loff_t *ppos)
5841 struct trace_array *tr = filp->private_data;
5842 struct ring_buffer *buffer = tr->trace_buffer.buffer;
5846 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5851 mutex_lock(&trace_types_lock);
5853 tracer_tracing_on(tr);
5854 if (tr->current_trace->start)
5855 tr->current_trace->start(tr);
5857 tracer_tracing_off(tr);
5858 if (tr->current_trace->stop)
5859 tr->current_trace->stop(tr);
5861 mutex_unlock(&trace_types_lock);
5869 static const struct file_operations rb_simple_fops = {
5870 .open = tracing_open_generic_tr,
5871 .read = rb_simple_read,
5872 .write = rb_simple_write,
5873 .release = tracing_release_generic_tr,
5874 .llseek = default_llseek,
5877 struct dentry *trace_instance_dir;
5880 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer);
5882 static void init_trace_buffers(struct trace_array *tr, struct trace_buffer *buf)
5886 for_each_tracing_cpu(cpu) {
5887 memset(per_cpu_ptr(buf->data, cpu), 0, sizeof(struct trace_array_cpu));
5888 per_cpu_ptr(buf->data, cpu)->trace_cpu.cpu = cpu;
5889 per_cpu_ptr(buf->data, cpu)->trace_cpu.tr = tr;
5894 allocate_trace_buffer(struct trace_array *tr, struct trace_buffer *buf, int size)
5896 enum ring_buffer_flags rb_flags;
5898 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
5900 buf->buffer = ring_buffer_alloc(size, rb_flags);
5904 buf->data = alloc_percpu(struct trace_array_cpu);
5906 ring_buffer_free(buf->buffer);
5910 init_trace_buffers(tr, buf);
5912 /* Allocate the first page for all buffers */
5913 set_buffer_entries(&tr->trace_buffer,
5914 ring_buffer_size(tr->trace_buffer.buffer, 0));
5919 static int allocate_trace_buffers(struct trace_array *tr, int size)
5923 ret = allocate_trace_buffer(tr, &tr->trace_buffer, size);
5927 #ifdef CONFIG_TRACER_MAX_TRACE
5928 ret = allocate_trace_buffer(tr, &tr->max_buffer,
5929 allocate_snapshot ? size : 1);
5931 ring_buffer_free(tr->trace_buffer.buffer);
5932 free_percpu(tr->trace_buffer.data);
5935 tr->allocated_snapshot = allocate_snapshot;
5938 * Only the top level trace array gets its snapshot allocated
5939 * from the kernel command line.
5941 allocate_snapshot = false;
5946 static int new_instance_create(const char *name)
5948 struct trace_array *tr;
5951 mutex_lock(&trace_types_lock);
5954 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5955 if (tr->name && strcmp(tr->name, name) == 0)
5960 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
5964 tr->name = kstrdup(name, GFP_KERNEL);
5968 raw_spin_lock_init(&tr->start_lock);
5970 tr->current_trace = &nop_trace;
5972 INIT_LIST_HEAD(&tr->systems);
5973 INIT_LIST_HEAD(&tr->events);
5975 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
5978 /* Holder for file callbacks */
5979 tr->trace_cpu.cpu = RING_BUFFER_ALL_CPUS;
5980 tr->trace_cpu.tr = tr;
5982 tr->dir = debugfs_create_dir(name, trace_instance_dir);
5986 ret = event_trace_add_tracer(tr->dir, tr);
5990 init_tracer_debugfs(tr, tr->dir);
5992 list_add(&tr->list, &ftrace_trace_arrays);
5994 mutex_unlock(&trace_types_lock);
5999 if (tr->trace_buffer.buffer)
6000 ring_buffer_free(tr->trace_buffer.buffer);
6005 mutex_unlock(&trace_types_lock);
6011 static int instance_delete(const char *name)
6013 struct trace_array *tr;
6017 mutex_lock(&trace_types_lock);
6020 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
6021 if (tr->name && strcmp(tr->name, name) == 0) {
6033 list_del(&tr->list);
6035 event_trace_del_tracer(tr);
6036 debugfs_remove_recursive(tr->dir);
6037 free_percpu(tr->trace_buffer.data);
6038 ring_buffer_free(tr->trace_buffer.buffer);
6046 mutex_unlock(&trace_types_lock);
6051 static int instance_mkdir (struct inode *inode, struct dentry *dentry, umode_t mode)
6053 struct dentry *parent;
6056 /* Paranoid: Make sure the parent is the "instances" directory */
6057 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
6058 if (WARN_ON_ONCE(parent != trace_instance_dir))
6062 * The inode mutex is locked, but debugfs_create_dir() will also
6063 * take the mutex. As the instances directory can not be destroyed
6064 * or changed in any other way, it is safe to unlock it, and
6065 * let the dentry try. If two users try to make the same dir at
6066 * the same time, then the new_instance_create() will determine the
6069 mutex_unlock(&inode->i_mutex);
6071 ret = new_instance_create(dentry->d_iname);
6073 mutex_lock(&inode->i_mutex);
6078 static int instance_rmdir(struct inode *inode, struct dentry *dentry)
6080 struct dentry *parent;
6083 /* Paranoid: Make sure the parent is the "instances" directory */
6084 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
6085 if (WARN_ON_ONCE(parent != trace_instance_dir))
6088 /* The caller did a dget() on dentry */
6089 mutex_unlock(&dentry->d_inode->i_mutex);
6092 * The inode mutex is locked, but debugfs_create_dir() will also
6093 * take the mutex. As the instances directory can not be destroyed
6094 * or changed in any other way, it is safe to unlock it, and
6095 * let the dentry try. If two users try to make the same dir at
6096 * the same time, then the instance_delete() will determine the
6099 mutex_unlock(&inode->i_mutex);
6101 ret = instance_delete(dentry->d_iname);
6103 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
6104 mutex_lock(&dentry->d_inode->i_mutex);
6109 static const struct inode_operations instance_dir_inode_operations = {
6110 .lookup = simple_lookup,
6111 .mkdir = instance_mkdir,
6112 .rmdir = instance_rmdir,
6115 static __init void create_trace_instances(struct dentry *d_tracer)
6117 trace_instance_dir = debugfs_create_dir("instances", d_tracer);
6118 if (WARN_ON(!trace_instance_dir))
6121 /* Hijack the dir inode operations, to allow mkdir */
6122 trace_instance_dir->d_inode->i_op = &instance_dir_inode_operations;
6126 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer)
6130 trace_create_file("trace_options", 0644, d_tracer,
6131 tr, &tracing_iter_fops);
6133 trace_create_file("trace", 0644, d_tracer,
6134 (void *)&tr->trace_cpu, &tracing_fops);
6136 trace_create_file("trace_pipe", 0444, d_tracer,
6137 (void *)&tr->trace_cpu, &tracing_pipe_fops);
6139 trace_create_file("buffer_size_kb", 0644, d_tracer,
6140 (void *)&tr->trace_cpu, &tracing_entries_fops);
6142 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
6143 tr, &tracing_total_entries_fops);
6145 trace_create_file("free_buffer", 0200, d_tracer,
6146 tr, &tracing_free_buffer_fops);
6148 trace_create_file("trace_marker", 0220, d_tracer,
6149 tr, &tracing_mark_fops);
6151 trace_create_file("trace_clock", 0644, d_tracer, tr,
6154 trace_create_file("tracing_on", 0644, d_tracer,
6155 tr, &rb_simple_fops);
6157 #ifdef CONFIG_TRACER_SNAPSHOT
6158 trace_create_file("snapshot", 0644, d_tracer,
6159 (void *)&tr->trace_cpu, &snapshot_fops);
6162 for_each_tracing_cpu(cpu)
6163 tracing_init_debugfs_percpu(tr, cpu);
6167 static __init int tracer_init_debugfs(void)
6169 struct dentry *d_tracer;
6171 trace_access_lock_init();
6173 d_tracer = tracing_init_dentry();
6177 init_tracer_debugfs(&global_trace, d_tracer);
6179 trace_create_file("tracing_cpumask", 0644, d_tracer,
6180 &global_trace, &tracing_cpumask_fops);
6182 trace_create_file("available_tracers", 0444, d_tracer,
6183 &global_trace, &show_traces_fops);
6185 trace_create_file("current_tracer", 0644, d_tracer,
6186 &global_trace, &set_tracer_fops);
6188 #ifdef CONFIG_TRACER_MAX_TRACE
6189 trace_create_file("tracing_max_latency", 0644, d_tracer,
6190 &tracing_max_latency, &tracing_max_lat_fops);
6193 trace_create_file("tracing_thresh", 0644, d_tracer,
6194 &tracing_thresh, &tracing_max_lat_fops);
6196 trace_create_file("README", 0444, d_tracer,
6197 NULL, &tracing_readme_fops);
6199 trace_create_file("saved_cmdlines", 0444, d_tracer,
6200 NULL, &tracing_saved_cmdlines_fops);
6202 #ifdef CONFIG_DYNAMIC_FTRACE
6203 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
6204 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
6207 create_trace_instances(d_tracer);
6209 create_trace_options_dir(&global_trace);
6214 static int trace_panic_handler(struct notifier_block *this,
6215 unsigned long event, void *unused)
6217 if (ftrace_dump_on_oops)
6218 ftrace_dump(ftrace_dump_on_oops);
6222 static struct notifier_block trace_panic_notifier = {
6223 .notifier_call = trace_panic_handler,
6225 .priority = 150 /* priority: INT_MAX >= x >= 0 */
6228 static int trace_die_handler(struct notifier_block *self,
6234 if (ftrace_dump_on_oops)
6235 ftrace_dump(ftrace_dump_on_oops);
6243 static struct notifier_block trace_die_notifier = {
6244 .notifier_call = trace_die_handler,
6249 * printk is set to max of 1024, we really don't need it that big.
6250 * Nothing should be printing 1000 characters anyway.
6252 #define TRACE_MAX_PRINT 1000
6255 * Define here KERN_TRACE so that we have one place to modify
6256 * it if we decide to change what log level the ftrace dump
6259 #define KERN_TRACE KERN_EMERG
6262 trace_printk_seq(struct trace_seq *s)
6264 /* Probably should print a warning here. */
6265 if (s->len >= TRACE_MAX_PRINT)
6266 s->len = TRACE_MAX_PRINT;
6268 /* should be zero ended, but we are paranoid. */
6269 s->buffer[s->len] = 0;
6271 printk(KERN_TRACE "%s", s->buffer);
6276 void trace_init_global_iter(struct trace_iterator *iter)
6278 iter->tr = &global_trace;
6279 iter->trace = iter->tr->current_trace;
6280 iter->cpu_file = RING_BUFFER_ALL_CPUS;
6281 iter->trace_buffer = &global_trace.trace_buffer;
6284 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
6286 /* use static because iter can be a bit big for the stack */
6287 static struct trace_iterator iter;
6288 static atomic_t dump_running;
6289 unsigned int old_userobj;
6290 unsigned long flags;
6293 /* Only allow one dump user at a time. */
6294 if (atomic_inc_return(&dump_running) != 1) {
6295 atomic_dec(&dump_running);
6300 * Always turn off tracing when we dump.
6301 * We don't need to show trace output of what happens
6302 * between multiple crashes.
6304 * If the user does a sysrq-z, then they can re-enable
6305 * tracing with echo 1 > tracing_on.
6309 local_irq_save(flags);
6311 /* Simulate the iterator */
6312 trace_init_global_iter(&iter);
6314 for_each_tracing_cpu(cpu) {
6315 atomic_inc(&per_cpu_ptr(iter.tr->trace_buffer.data, cpu)->disabled);
6318 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
6320 /* don't look at user memory in panic mode */
6321 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
6323 switch (oops_dump_mode) {
6325 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6328 iter.cpu_file = raw_smp_processor_id();
6333 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
6334 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6337 printk(KERN_TRACE "Dumping ftrace buffer:\n");
6339 /* Did function tracer already get disabled? */
6340 if (ftrace_is_dead()) {
6341 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
6342 printk("# MAY BE MISSING FUNCTION EVENTS\n");
6346 * We need to stop all tracing on all CPUS to read the
6347 * the next buffer. This is a bit expensive, but is
6348 * not done often. We fill all what we can read,
6349 * and then release the locks again.
6352 while (!trace_empty(&iter)) {
6355 printk(KERN_TRACE "---------------------------------\n");
6359 /* reset all but tr, trace, and overruns */
6360 memset(&iter.seq, 0,
6361 sizeof(struct trace_iterator) -
6362 offsetof(struct trace_iterator, seq));
6363 iter.iter_flags |= TRACE_FILE_LAT_FMT;
6366 if (trace_find_next_entry_inc(&iter) != NULL) {
6369 ret = print_trace_line(&iter);
6370 if (ret != TRACE_TYPE_NO_CONSUME)
6371 trace_consume(&iter);
6373 touch_nmi_watchdog();
6375 trace_printk_seq(&iter.seq);
6379 printk(KERN_TRACE " (ftrace buffer empty)\n");
6381 printk(KERN_TRACE "---------------------------------\n");
6384 trace_flags |= old_userobj;
6386 for_each_tracing_cpu(cpu) {
6387 atomic_dec(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);
6389 atomic_dec(&dump_running);
6390 local_irq_restore(flags);
6392 EXPORT_SYMBOL_GPL(ftrace_dump);
6394 __init static int tracer_alloc_buffers(void)
6400 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
6403 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
6404 goto out_free_buffer_mask;
6406 /* Only allocate trace_printk buffers if a trace_printk exists */
6407 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
6408 /* Must be called before global_trace.buffer is allocated */
6409 trace_printk_init_buffers();
6411 /* To save memory, keep the ring buffer size to its minimum */
6412 if (ring_buffer_expanded)
6413 ring_buf_size = trace_buf_size;
6417 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
6418 cpumask_copy(tracing_cpumask, cpu_all_mask);
6420 raw_spin_lock_init(&global_trace.start_lock);
6422 /* TODO: make the number of buffers hot pluggable with CPUS */
6423 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
6424 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
6426 goto out_free_cpumask;
6429 if (global_trace.buffer_disabled)
6432 trace_init_cmdlines();
6435 * register_tracer() might reference current_trace, so it
6436 * needs to be set before we register anything. This is
6437 * just a bootstrap of current_trace anyway.
6439 global_trace.current_trace = &nop_trace;
6441 register_tracer(&nop_trace);
6443 /* All seems OK, enable tracing */
6444 tracing_disabled = 0;
6446 atomic_notifier_chain_register(&panic_notifier_list,
6447 &trace_panic_notifier);
6449 register_die_notifier(&trace_die_notifier);
6451 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
6453 /* Holder for file callbacks */
6454 global_trace.trace_cpu.cpu = RING_BUFFER_ALL_CPUS;
6455 global_trace.trace_cpu.tr = &global_trace;
6457 INIT_LIST_HEAD(&global_trace.systems);
6458 INIT_LIST_HEAD(&global_trace.events);
6459 list_add(&global_trace.list, &ftrace_trace_arrays);
6461 while (trace_boot_options) {
6464 option = strsep(&trace_boot_options, ",");
6465 trace_set_options(&global_trace, option);
6468 register_snapshot_cmd();
6473 free_percpu(global_trace.trace_buffer.data);
6474 #ifdef CONFIG_TRACER_MAX_TRACE
6475 free_percpu(global_trace.max_buffer.data);
6477 free_cpumask_var(tracing_cpumask);
6478 out_free_buffer_mask:
6479 free_cpumask_var(tracing_buffer_mask);
6484 __init static int clear_boot_tracer(void)
6487 * The default tracer at boot buffer is an init section.
6488 * This function is called in lateinit. If we did not
6489 * find the boot tracer, then clear it out, to prevent
6490 * later registration from accessing the buffer that is
6491 * about to be freed.
6493 if (!default_bootup_tracer)
6496 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
6497 default_bootup_tracer);
6498 default_bootup_tracer = NULL;
6503 early_initcall(tracer_alloc_buffers);
6504 fs_initcall(tracer_init_debugfs);
6505 late_initcall(clear_boot_tracer);